HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES QUANTUM LOGIC
This page intentionally left blank
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES QUANTUM LOGIC
Edited by
KURT ENGESSER Universität Konstanz, Konstanz, Germany
DOV M. GABBAY King's College London, Strand, London, UK
DANIEL LEHMANN The Hebrew University of Jerusalem, Jerusalem, Israel
Amsterdam • Boston • Heidelberg • London • New York • Oxford Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo NorthHolland is an imprint of Elsevier
NorthHolland is an imprint of Elsevier Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK First edition 2009 Copyright © 2009 Elsevier B.V. All rights reserved No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+ 44) (0) 1865 843830; fax (+ 44) (0) 1865 853333; email:
[email protected] Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library Library of Congress CataloginginPublication Data A catalog record for this book is available from the Library of Congress ISBN: 9780444528698 For information on all NorthHolland publications visit our website at elsevierdirect.com
Printed and bound in Hungary 09 10 11 12
10 9 8 7 6 5 4 3 2 1
CONTENTS Foreword Anatolij Dvureˇ censkij
vii
Editorial Preface Kurt Engesser, Dov Gabbay and Daniel Lehmann
ix
The Birkhoﬀ–von Neumann Concept of Quantum Logic os Redei ´ e Mikl´
1
Is Quantum Logic a Logic? Mladen Paviˇ ci´ c and Norman D. Megill
23
Is Logic Empirical? Guido Bacciagaluppi
49
Quantum Axiomatics Diederik Aerts
79
Quantum Logic and Nonclassical Logics 127 Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli Gentzen Methods in Quantum Logic Hirokazu Nishimura
227
Categorical Quantum Mechanics Samson Abramsky and Bob Coecke
261
Extending Classical Logic for Reasoning about Quantum Systems Rohit Chadha, Paulo Mateus, Am´ılcar Sernadas and Cristina Sernadas
325
Sol`er’s Theorem Alexander Prestel
373
Operational Quantum Logic: A Survey and Analysis David J. Moore and Frank Valckenborgh
389
vi
Contents
Test Spaces Alexander Wilce
443
Contexts in Quantum, Classical and Partition Logic Karl Svozil
551
Nonmonotonicity and Holicity in Quantum Logic Kurt Engesser, Dov Gabbay and Daniel Lehmann
587
A Quantum Logic of Down Below Peter D. Bruza, Dominic Widdows and John H. Woods
625
A Completeness Theorem of Quantum Set Theory Satoko Titani
661
Index
703
vii
FOREWORD
More than a century ago Hilbert posed his unsolved (and now famous), 23 problems of mathematics. When browsing through the Internet recently I found that Hilbert termed his Sixth Problem nonmathematical. How could Hilbert call a problem of mathematics nonmathematical? And what does this problem say? In 1900, Hilbert, inspired by Euclid’s axiomatic system of geometry, formulated his Sixth Problem as follows: To ﬁnd a few physical axioms that, similar to the axioms of geometry, can describe a theory for a class of physical events that is as large as possible. The twenties and thirties of the last century were truly exciting times. On the one hand there emerged the new physics which we call quantum physics today. On the other hand, in 1933, N. A. Kolmogorov presented a new axiomatic system which provided a solid basis for modern probability theory. These milestones marked the entrance into a new epoch in that quantum mechanics and modern probability theory opened new gates, not just for science, but for human thinking in general. Heisenberg’s Uncertainty Principle showed, however, that the micro world is governed by a new kind of probability laws which diﬀer from the Kolmogorovian ones. This was a great challenge to mathematicians as well as to physicists and logicians. One of the responses to this situation was the, now famous, 1936 paper by Garret Birkhoﬀ and John von Neumann entitled “The logic of quantum mechanics”, in which they suggested a new logical model which was based on the Hilbert space formalism of quantum mechanics and which we, today, call a quantum logic. G. Mackey asked the question whether every state on the lattice of projections of a Hilbert space could be described by a density operator; and his young student A. Gleason gave a positive answer to this question. Although this was not part of Gleason’s special ﬁeld of interest, his theorem, now known as Gleason’s theorem, had a profound impact and is rightfully considered one of the most important results about quantum logics and structures. Gleason’s proof was nontrivial. When John Bell became familiar with it, he said he would leave this ﬁeld of research unless there would be a simpler proof of Gleason’s theorem. Fortunately, Bell did ﬁnd a relatively simple proof of the partial result that there exists no twovalued measure on a threedimensional Hilbert space. An elementary proof of Gleason’s theorem was presented by R. Cooke, M. Keane and W. Moran in 1985. In the eighties and nineties it was the American school that greatly enriched the theory of quantum structures. For me personally Varadarajan’s paper and subsequently his book were the primary sources of inspiration for my work together with
viii
Foreword
Gleason’s theorem. The theory of quantum logics and quantum structures inspired many mathematicians, physicists, logicians, experts on information theory as well as philosophers of science. I am proud that in my small country, Czechoslovakia and now Slovakia, research on quantum structures is a thriving ﬁeld of scientiﬁc activity. The achievements characteristic of the eighties and nineties are the fuzzy approaches which provided a new way of looking at quantum structures. A whole hierarchy of quantum structures emerged, and many surprising connections with other branches of mathemtics and other sciences were discovered. Today we can relate phenomena ﬁrst observed in quantum mechanics to other branches of science such as complex computer systems and investigations on the functioning of the human brain, etc. In the early nineties, a new organisation called International Quantum Structures Association (IQSA) was founded. IQSA gathers experts on quantum logic and quantum structures from all over the world under its umbrella. It organises regular biannual meetings: Castiglioncello 1992, Prague 1994, Berlin 1996, Liptovsky Mikulas 1998, Cesenatico 2001, Vienna 2002, Denver 2004, Malta 2006. In spring 2005, Dov Gabbay, Kurt Engesser, Daniel Lehmann and Jane Spurr had an excellent idea — to ask experts on quantum logic and quantum structures to write long chapters for the Handbook of Quantum Logic and Quantum Structures. It was a gigantic task to collect and coordinate these contributions by leading experts from all over the world. We are grateful to all four for preparing this monumental opus and to Elsevier for publishing it. When browsing through this Handbook, in my mind I am wandering back to Hilbert’s Sixth Problem. I am happy that this problem is in fact not a genuinely mathematical one which, once it is solved, brings things to a close. Rather it has led to a new development of scientiﬁc thought which deeply enriched mathematics, the understanding of the foundations of quantum mechanics, logic and the philosophy of science. The present Handbook is a testimony to this fact. Those who bear witness to it are Dov, Kurt, Daniel, Jane and the numerous authors. Thanks to everybody who helped bring it into existence. Anatolij Dvureˇcenskij, President of IQSA July 2006
ix
EDITORIAL PREFACE
There is a wide spread slogan saying that Quantum Mechanics is the most successful physical theory ever. And, in fact, there is hardly a physicist who does not agree with this. However, there is a reverse of the medal. Not only is Quantum mechanics unprecedently successful but it also raises fundamental problems which are equally unprecedented not only in the history of physics but in the history of science in general. The most fundamental problems that Quantum Mechanics raises are conceptual in nature. What is the proper interpretation of Quantum Mechanics? This is a question touching on most fundamental issues, and it is, at this stage, safe to say that there is no answer to this question yet on which physicists and philosophers of science could agree. It is, moreover, no exaggeration to say that the problem of the conceptual understanding of Quantum Mechanics constitutes one of the great intellectual puzzles of our time. The topic of the present Handbook is, though related to this gigantic issue, more modest in nature. It can, brieﬂy, be described as follows. Quantum Mechanics owes is tremendous success to a mathematical formalism. It is the mathematical and logical investigation of the various aspects of this formalism that constitutes the topic of the present Handbook. This formalism the core of which is the mathematical structure of a Hilbert space received its ﬁnal elegant shape in John von Neumann’s classic 1932 book “Mathematical Foundations of Quantum Mechanics”. In 1936 John von Neumann published, jointly with the Harvard matthematician Garret Birkhoﬀ, a paper entitled “The logic of quantum mechanics”. In the Introduction the authors say: ”The object of the present paper is to discover what logical structure we may hope to ﬁnd in physical theories which, like quantum mechanics, do not conform to classical logic”. The idea of the paper, which was as ingenious as it was revolutionary, was that the Hilbert space formalism of Quantum Mechanics displayed a logical structure that could prove useful to the understanding of Quantum Mechanics. Birkoﬀ and von Neumann were the ﬁrst to put forward the idea that there is a link between logic and (the formalism of) Quantum Mechanics, and their now famous paper marked the birth of a ﬁeld of research which has become known as Quantum Logic. The Birkhoﬀvon Neumann paper triggered, after some time of dormancy admittedly, a rapid development of quantum logical research. Various schools of thought emerged. Let us, in this Introduction, highlight just a few of the milestones in this development. In his famous essay “Is logic empirical?” Putnam put forward the view that the role played by logic in Quantum Mechanics is similar to that played by geometry in
x
Editorial Preface
the theory of relativity. On this view logic is as empirical as geometry. Putnam’s revolutionary thesis triggered a discussion which was highly fruitful not only for Quantum Mechanics but for our views on the nature of logic in general. The reader will ﬁnd a discussion of Putnam’s thesis in this Handbook. Another school of thought was initiated by Piron’s “Axiomatique quantique”. This school, which has become known as the Geneva school, aimed at reconstructing the formalism of Quantum Mechanics from ﬁrst principles. It was Piron’s student Diederik Aerts who continued this in Brussels. The achievements of the GenevaBrussels school are reﬂected in various chapters witten by Aerts and former students of his. In Italy it was Enrico Beltrametti and Maria Luisa Dalla Chiara, just to mention two names, who founded another highly inﬂuential school which is well represented in this Handbook. Highly sophisticated eﬀorts resulted in linking the logic of Quantum Mechanics to mainstream logic. Just to give a ﬂavour of this, let us mention that Nishimura studied Gentzen type systems in the context of Quantum Logic. Abramsky and Coecke in Oxford and Sernadas in Lisbon as well as others established the connection with Categorial Logic and Linear Logic, and the connection with NonMonotonic Logic was made by the editors. Prior to this, another direction of research had focused on the lattice structures relevant to Quantum Logic. Essentially, this ﬁeld of research was brought to fruition in the USA by the pioneering work of Foulis and Greechie on orthomodular lattices. Moreover, we have to mention the CzechSlovak school which was highly inﬂuential in establishing the vast ﬁeld of research dealing with the various abstract Quantum Structures which constitute the topic of a whole volume of this Handbook. Let us in this context just mention the names of Anatolij Dvurecenskij and Sylvia Pulmannova in Bratislava and Pavel Ptak in Prague. The editors are happy and grateful to have succeeded in bringing together the most eminent scholars in the ﬁeld of Quantum logic and Quantum Structures for the sake of the present Handbook. We cordially thank all the authors for their contributions and their cooperation during the preparation of this work. Most of these authors are members of the Internatonal Quantum Structures Association (IQSA). We would like to express our deep gratitude to IQSA and in particular to its President, Professor Anatolij Dvurecenskij, for cooperating so closely with us and supporting us so generously during the preparation of this Handbook. The present Handbook is an impressive document of the intellectual achievements which have been made in the study of the logical and mathematical structures arising from Quantum Mechanics. We hope that it will turn out to be a milestone on the path that will ultimately lead to the solution of one of the great intellectual puzzles of our time, namely the understanding of Quantum Mechanics. The Editors: Kurt Engesser, Dov Gabbay and Daniel Lehmann Germany, London, and Israel May 2008
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
1
THE BIRKHOFF–VON NEUMANN CONCEPT OF QUANTUM LOGIC Mikl´os R´edei
1
INTRODUCTION
Quantum logic was born with the following conclusion of Garrett Birkhoﬀ and John von Neumann in their joint paper (henceforth “Birkhoﬀvon Neumann paper”) published in 1936: Hence we conclude that the propositional calculus of quantum mechanics has the same structure as an abstract projective geometry. [Birkhoﬀ and von Neumann, 1936] (Emphasis in the original) This was a striking conclusion in 1936 for two reasons, one ground breaking and one conservative: ground breaking because it opened the way for the development of algebraic logic in the direction of nonclassical algebraic structures that have much weaker properties than Boolean algebras. Conservative because an abstract projective geometry is an orthocomplemeneted, (nondistributive), modular lattice; however, the nonBoolean algebra that seemed in 1936 to be the most natural candidate for quantum logic was the nonmodular, orthomodular lattice of all projections on an inﬁnite dimensional complex Hilbert space. Indeed, subsequently quantum logic was (and typically still is) taken to be an orthocomplemented, nonmodular, orthomodular lattice. Hence, the concept of quantum logic proposed by Birkhoﬀ and von Neumann in their seminal paper diﬀers markedly from the notion that became later the standard view – it is more conservative than one would expect on the basis of later developments. There are not many historical investigations in the enormous quantum logic literature [Pavicic, 1992] that aim at scrutinizing the Birkhoﬀvon Neumann notion of quantum logic, and especially the discrepancy between the standard notion and the Birkhoﬀvon Neumann concept: [Bub, 1981b], [Bub, 1981a], [R´edei, 1996], Chapter 7. in [R´edei, 1998], [R´edei, 2001], [DallaChiara et al., 2007] (see also [Popper, 1968] and [Scheibe, 1974]). The recent discovery and publication in [R´edei, 2005] of von Neumann’s letters to Birkhoﬀ during the preparatory phase (in 1935) of their 1936 paper have made it possible to reconstruct in great detail [R´edei, 2007] the conceptual considerations that culminated in the 1936 paper’s main conclusion cited above. As a result of these historical studies we now understand quite well why Birkhoﬀ and von Neumann postulated the “quantum propositional calculus”
2
Mikl´ os R´ edei
to be a modular lattice and rejected explicitly the idea that quantum propositional calculus can be identiﬁed with the nonmodular, orthomodular lattice of all closed linear subspaces of an inﬁnite dimensional complex Hilbert space (Hilbert lattice). The aim of this review is to recall the Birkhoﬀvon Neumann concept of quantum logic together with the pertinent mathematical notions necessary to understand the development of their ideas. Special emphasis will be put on the analysis of the diﬀerence between their views and the subsequent standard notion of quantum logic. The structure of the review is the following. For reference, and in order to place the Birkhoﬀvon Neumann concept in appropriate context, section 2 recalls brieﬂy the standard notion of quantum logic in terms of algebraic semantics. Based on excerpts from the recently discovered and published letters by von Neumann to Birkhoﬀ, section 3 reconstructs the main steps of the thought process that led Birkhoﬀ and von Neumann to abandon Hilbert lattice as quantum logic and to propose in their published paper an abstract projective geometry as the quantum propositional system. Section 4 argues that von Neumann was not satisﬁed with their idea after 1936: He would have liked to see quantum logic worked out in much greater detail – he himself tried to achieve this but did not succeed. Section 4 also attempts to discern the conceptual obstacles standing in the way of elaborating quantum logic along the lines von Neumann envisaged. The concluding section 5 summarizes the main points and makes some further comments on the signiﬁcance of the Birkhoﬀvon Neumann concept. 2
QUANTUM LOGIC: LOGICIZATION OF NONBOOLEAN ALGEBRAIC STRUCTURES. THE STANDARD VIEW.
It is well known that both the syntactic and the semantic aspects of classical propositional logic can be described completely in terms of Boolean algebras: The TarskiLindenbaum algebra A of classical propositions is a Boolean algebra and a deductive system formulated in a classical propositional logic can be identiﬁed with a ﬁlter in A. The notions of syntactic consistency and completeness correspond to the ﬁlter being proper and being prime (equivalently: maximal), respectively. The notion of interpretation turns out to be a Boolean algebra homomorphism from A into the two element Boolean algebra, and all the semantic notions are deﬁned in terms of these homomorphisms. All this is expressed metaphorically by Halmos’ famous characterization of the (classical) logician as the dual of a (Boolean) algebraist [Halmos, 1962, p. 22], a characterization which has been recently “dualized” by Dunn and Hardegree: “By duality we obtain that the algebraist is the dual of the logician.” [Dunn and Hardegree, 2001, p. 6]. The problem of quantum logic can be formulated as the question of whether the duality alluded to above also obtains if Boolean algebras are replaced by other, typically weaker algebraic structures arising from the mathematical formalism of quantum mechanics. It turns out that formal logicization is possible for a large class of nonBoolean structures. Following Hardegree [Hardegree, 1981b], [Hardegree, 1981a] the standard (sometimes called “orthodox”) concept is described
The Birkhoﬀ–von Neumann Concept of Quantum Logic
3
below, and it is this concept with which the Birkhoﬀ von Neumann concept will be contrasted. Standard quantum logic comes in two forms: abstract (also called orthomodular) quantum logic and concrete (also called Hilbert) quantum logic. The semantics is similar in both cases, but the latter determines a stronger logic. Let K = {P, &, , ∼} be a zeroth order formal language with the set P of sentence variables p, q . . ., two place connectives & (and), (or), negation sign ∼, parentheses (,), and let F be the set of well formed formulas in K deﬁned in the standard way by induction from P : F is the smallest set for which the following two conditions hold: P if φ, ψ ∈ F then
⊂
F
(1)
(φ&ψ), (φ ψ), (∼ φ) ∈ F
(2)
Let (L, ∨, ∧, ⊥) be an orthomodular lattice. Orthomodularity of L means that the following condition holds: (3) orthomodularity: If A ≤ B and A⊥ ≤ C, then A∨(B∧C) = (A∨B)∧(A∨C) Orthomodularity is a weakening of the modularity law: (4) modularity:
If A ≤ B, then A ∨ (B ∧ C) = (A ∨ B) ∧ (A ∨ C)
which itself is a weakening of the distributivity law: (5) distributivity:
A ∨ (B ∧ C) = (A ∨ B) ∧ (A ∨ C)
for all A,B,C
The set GQ ⊆ L in an orthomodular lattice is called a generalized ﬁlter if ⊥
if A ∈ GQ and A ∨ (A ∧ B) ∈ GQ then
I ∈ GQ B ∈ GQ
(6) (7)
Given a pair (L, GQ ) the map i : F → L is called an (L, GQ )interpretation if i(φ&ψ) = i(φ) ∧ i(ψ) i(φ ψ) = i(φ) ∨ i(ψ) i(∼ φ) = i(φ)⊥ Each interpretation i determines ⎧ ⎨ 1 (true) 0 (f alse) (11) vi (φ) = ⎩ undetermined
(8) (9) (10)
a (L, GQ )valuation vi by if if otherwise
i(φ) ∈ GQ i(∼ φ) ∈ GQ
If V (L) denotes the set of all (L, GQ )valuations and V is the class of valuations determined by the class of orthomodular lattices, then φ ∈ F is called valid if v(φ) = 1 for every v ∈ V , and a class of formulas Γ is deﬁned to entail φ if v(ψ) = 1 for all ψ ∈ Γ implies v(φ) = 1. One can deﬁne the quantum analog →Q of the classical conditional by
4
Mikl´ os R´ edei
(12) φ →Q ψ =∼ φ (φ&ψ) and one can formulate a deduction system in K using →Q such that one can prove soundness and completeness theorems for the resulting quantum logical system (see [Hardegree, 1981b],[Hardegree, 1981a]). A speciﬁc class of orthomodular lattices is the category of Hilbert lattices: A Hilbert lattice P(H) with lattice operations ∧, ∨, ⊥ is the set P(H) of all projections (equivalently: closed linear subspaces) of a complex, possibly inﬁnite dimensional Hilbert space H, where the lattice operations ∧, ∨ and ⊥ are the set theoretical intersection, closure of the sum and orthogonal complement, respectively. Note that the Hilbert lattice P(H) is not only nondistributive but it also is nonmodular if the dimension of the Hilbert space is inﬁnite [R´edei, 1998]. It is important that while all the deﬁnitions and stipulations made above in connection with orthomodular lattices are meaningful for Hilbert lattices, no completeness results are presently known for the semantics determined by Hilbert lattices: The deduction system that works in the case of (abstract) orthomodular lattices is not strong enough to yield all statements that are valid in Hilbert lattices: the “orthoarguesian law”, which is valid in Hilbert lattices, does not hold in every orthomodular lattice (see [Kalmbach, 1981], [DallaChiara and Giuntini, 2002]).
3 THE BIRKHOFFVON NEUMANN CONCEPT OF QUANTUM LOGIC The Birkhoﬀvon Neumann paper can be viewed as the ﬁrst paper in which the suggestion to logicize a nonBoolean lattice appears. There are however several types of nonBoolean lattice. Which one is supposed to be logicized?
3.1 Which nonBoolean lattice to logicize? At the time of the birth of quantum logic the notion of an abstract orthomodular lattice did not yet exist; however, the canonical example of nondistributive, orthomodular lattices, the Hilbert lattice P(H), was known already, and, since this structure emerges naturally from the Hilbert space formalism of quantum mechanics, P(H) was the most natural candidate in 1935 for the propositional system of quantum logic. Indeed, Birkhoﬀ and von Neumann did consider P(H) as a possible propositional system of quantum logic; yet, this lattice was not their choice: The ﬁrst indication that P(H) may not be a suitable candidate for a quantum propositional system is in von Neumann’s letter of (January 19, 1935). Von Neumann writes:
The Birkhoﬀ–von Neumann Concept of Quantum Logic
Using the operatordescription, a ∨ b, a ∧ b can be formed, if the physically signiﬁcant operators form a ring. This, I think should be assumed anyhow, even if one does not require that all operators are phys.[ically] signiﬁcant.
5
(← I believe this).
(← but I am rather doubting lately this.)
But we need probably not insist on this point too much. (von Neumann to Birkhoﬀ, January 19, 1935 ?1 ), [R´edei, 2005, p. 51]
3.2
Von Neumann algebras
A “ring of operators” von Neumann is referring to in the above quotation is known today as a von Neumann algebra: a set N of bounded operators on a Hilbert space H is a von Neumann algebra if it contains the unit, is closed with respect to the adjoint operation and is closed in the strong operator topology. The latter requirement means that if Qn is a sequence of operators from M such that for all ξ ∈ H we have Qn ξ → Qξ for some bounded operator Q on H, then Q ∈ M (see [Takesaki, 1979] for the theory of von Neumann algebras). If S is any set of bounded operators on H, then its (ﬁrst) commutant S is the set of bounded operators that commute with every element in S i.e.: S ≡ {Q : QX = XQ, for all X ∈ S} The operation of taking the commutant can be iterated: S ≡ (S ) , and it is clear that S is contained in the second commutant, so the second commutant S is an extension of S. How much of an extension? The answer to this question, von Neumann’s double commutant theorem, is the most fundamental result in the theory of von Neumann algebras: PROPOSITION 1 Double commutant theorem. S is strongly dense in S . Proposition 1 implies that a ∗algebra of bounded operators on a Hilbert space that contains the unit is a von Neumann algebra if and only if it coincides with its second commutant. A von Neumann algebra N is called a factor if the only elements in N that commute with every other element in N are the constant multiples of the identity, i.e. if (13) N ∩ N = {λI : λ a complex number} An immediate corollary of the double commutant theorem is the characteristic property of von Neumann algebras that they contain many projections; in fact, they contain enough projections for the set of projections P(M) to determine the von Neumann algebra completely in the sense 1 Von Neumann’s letters are not always properly dated: the year is occasionally missing. If this is the case, we put a question mark after the year – the context makes it clear that the year of writing is 1935 in cases of all the letters quoted here.
6
Mikl´ os R´ edei
(14) M = (P(M)) Moreover, we have PROPOSITION 2. The set P(M) of projections of a von Neumann algebra M is a complete, orthomodular lattice (called von Neumann lattice) with respect to the lattice operations inherited from P(H). In particular the Hilbert lattice P(H) of all projections on a Hilbert space is a complete orthomodular lattice, since the set B(H) of all bounded operators on H is a von Neumann algebra. Speciﬁcally the spectral projections of the set of all (not necessarily bounded) selfadjoint operators coincides with the set of all projections. It is very important that while all von Neumann lattices are orthomodular, some have the stronger property of modularity. There is a subtle connection between the type of a von Neumann algebra (in the sense of the Murrayvon Neumann classiﬁcation theory) and the modularity of its projection lattice. We shall return to this issue later.
3.3 Nonmodularity of Hilbert lattice While in January 1935 von Neumann did not intend to insist on restricting the set of physical quantities to a proper subset of all possible operators, by November 1935 he changed his mind: I am somewhat scared to consider all physical quantities = bounded selfadjoint operators as a lattice. (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 53] The reason why he changed his mind was the realization that the Hilbert lattice P(H) is not modular if the Hilbert space is inﬁnite dimensional (note that “Blattice” means modular lattice in the next quotation): In any linear space H the linear subspaces K, L, M, . . . form a Blattice L with the “meet” K ∩ L: intersection of K and L in the sense of set theory “join” K ∪ L : linear sum of K and L, i.e. set of all f + g, f ∈ K, g ∈ L (Proof obvious.) But in a metriclinear space H the lattice L of all closedlinear subspaces KLM , . . . , for which the “join” is deﬁned as “join” K ∨ K:
closure of the linear sum of K and L, i.e. the set of all condensation points of the f + g, f ∈ K, g ∈ L
The Birkhoﬀ–von Neumann Concept of Quantum Logic
7
while the “meet” is as above, is not necessarily a Blattice. This is in particular the case in Hilbert space. (K ∪ L and K ∨ L are identical if K, L are both closed and orthogonal to each other, but not for any two closed K, L !) In fact, it is possible to ﬁnd three closedlinear subspaces K, L, M of Hilbert space H, for which (15) K M ,
(K ∨ L) ∩ M K ∨ (L ∩ M )
(von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 54], emphasis in original. This letter contains a detailed proof that subspaces K, L, M exist that satisfy (15) (and thereby violate modularity (4)). (The Birkhoﬀvon Neumann paper just states this fact without detailed argument.) Von Neumann’s proof makes use of the theory of unbounded selfadjoint operators, utilizing the fact that one can ﬁnd two unbounded selfadjoint operators X and Y on an inﬁnite dimensional Hilbert space such that the intersection of their domains is empty. Von Neumann emphasizes this feature of his proof: Examples could be constructed which make no use of operator theory, but I think that this example shows more clearly “what it’s all about”: It is the existence of “pathological” operators – like X, Y above – in Hilbert space, which destroys the Blattice character. (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 55] Von Neumann regarded this pathological behavior of the set of all unbounded operators on a Hilbert space a very serious problem because it prohibits adding and composing these operators in general, which entails that these operators do not form an algebra. In von Neumann’s eyes this was a great obstacle to doing computations with those operators, and since the selfadjoint operators are representatives of quantum physical quantities, it appeared unnatural to him that they behave so irregularly that forming an algebra from them was not possible. He pointed out this pathology several times in his published papers (see e.g. paragraph 6. of Introduction in [Murray and von Neumann, 1936]), and the pathological character of the set of all selfadjoint operators was one of the main reasons why he hoped as late as in his famous talk on “Unsolved Problems in Mathematics” in 1954 (see [von Neumann, 2001] and [R´edei, 1999]) that a restricted set of operators, and therefore a speciﬁc von Neumann algebra, the type II1 factor (see below) would be a more suitable mathematical framework for quantum mechanics than Hilbert space theory. In this situation von Neumann saw two options:
8
Mikl´ os R´ edei
(I) Either we deﬁne the “join” by ∪ (as a honest linear sum), then the lattice is B, but we must admitt all (notnecessarilyclosed) linear subspaces, (II) or we deﬁne the “join” by ∨ (closure of the linear sum), then the Bcharacter is lost. (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 55] Since P(H) is not modular and given that von Neumann wished to preserve modularity as a property of the quantum propositional system, one would expect von Neumann to choose option (I). But this is not the case. Von Neumann thinks through the consequences of choosing option (I) ﬁrst: Let us ﬁrst consider the alternative (I). The orthogonal complement K still has the property K ∪ L = (K ∩ L) , but K ∩ L = (K ∪ L) and K = K are lost. We have K ∩K = 0, while K ∪K is everywhere dense, but not necessarily I. There is a funny relationship between K and its “closure” K . (For instance: All probabilities in the state K are equal to those in the state K , but “meets” (K ∩ L and K ∩ L , even for closed L’s) may diﬀer.) The situation is strongly reminiscent of the “excluded middle” troubles, although I did not yet compare all details with those of the classcalculus in “intuitionistic” logics. After all it is so in normal logics, too, that these troubles arise as soon as you pass to inﬁnite systems, although I must admitt, that the diﬃculties there are more “optional” then2 here. It has to be said, ﬁnally, that even in this case (I) complements exist, i.e., that for every K there exists K ∗ ’s for which K ∪K ∗ = I, K ∩K ∗ = 0, but one needs the Hammelbasisconstruction to get them. (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 55] So, while von Neumann evaluates alternative (I) as representing an option which cannot be excluded on the grounds of being either algebraically or logically extremely weird (although it is clear from the above that he did not like the asymmetric failure of De Morgan’s law), he prefers option (II) in spite of its being seemingly counterintuitive. Here is why: Alternative (II) seems to exclude Hilbert space, if one sticks to Blattices.3 Still one may observe this: Consider a ring R of operators in Hilbert space. The idempotents of R form a lattice LR . One sees easily, that LR is irreducible (=no direct sum), if and only if the center of R consists of the αI (α=complex 2 Spelling 3 Recall
error, should be “than”. that Blattice means modular lattice.
The Birkhoﬀ–von Neumann Concept of Quantum Logic
9
number) only, i.e. if R is a ring of the sort which Murray and I considered. (We called them “factors”.) LR contains 0, I and a complement which dualises ∪ and ∩. (Now ∪ corresponds to what I called ∨, case (II).) One may ask: When is LR a Blattice? The answer is (this is not diﬃcult to prove): If and only if the ring R is ﬁnite in the classiﬁcation Murray and I gave. I.e.: R must be isomorphic: 1. either to the full matrixring of a ﬁnitedimensional Euclidean space (say ndimensional, n = 1, 2, . . .), 2. to one of those of our rings, in which each idempotent has a “dimensionality”, the range of which consists of all real numbers ≥ 0, ≤ 1, and which is uniquely determined by its formal properties. We called 1. “Case In ” and 2. “Case II1 ”. Thus for operatorlattices the Blattice axiom a≤b
→
(a ∪ b) ∩ c = a ∪ (b ∩ c)
leads directly to the cases I1 , I2 , . . . and II1 ! (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 56]
3.4
Types of von Neumann algebras
Von Neumann refers here to the Murrayvon Neumann classiﬁcation theory of factors, which was worked out by him in collaboration with F.J Murray precisely at the time (19341935) when he was working with G. Birkhoﬀ on quantum logic [Murray and von Neumann, 1936]. Von Neumann (partly in collaboration with F.J. Murray) published four major papers on the theory of von Neumann algebras [Murray and von Neumann, 1936], [Murray and von Neumann, 1937], [von Neumann, 1940] and [Murray and von Neumann, 1943]. The ﬁrst paper’s main result was a classiﬁcation theory of von Neumann algebras that are irreducible in the sense of not containing nontrivial operators commuting with every other operator in the algebra (i.e. “factors”). The set B(H) of all bounded operators is clearly a factor and it turned out that there are ﬁve classes of factors, the diﬀerent types are denoted by von Neumann as In , I∞ , II1 , II∞ and III∞ . The classiﬁcation of factors was given in terms of a (relative) dimension function d deﬁned on the lattice of projections P(M) of a von Neumann algebra M. The map d from P(M) into the set IR+ ∪ {∞} is a dimension function if (i) d(A) > 0 if and only if A = 0, (ii) d(A) = d(B) if there exists an isometry U ∈ M between ranges of the projections A and B, (iii) d(A) + d(B) = d(A ∨ B) + d(A ∧ B).
10
Mikl´ os R´ edei
The (relative) dimension function d on an arbitrary factor is a generalization of the ordinary linear dimension of the closed linear subspace a projection projects to, and the ordinary dimension takes on the positive integer values 0, 1, 2, . . . , n and 0, 1, 2, . . ., respectively, in the two wellknown cases of the set of all bounded operators on a ﬁnite, ndimensional (respectively inﬁnite) dimensional Hilbert space. In the cases II1 , II∞ and III∞ the ranges of the dimension function are, respectively, the following: the unit interval [0, 1], the set of nonnegative real numbers IR+ and the two element set {0, ∞}. (See [Takesaki, 1979] for a systematic treatment, or [R´edei, 1998], [Petz and R´edei, 1995] for a brief review of the Murrayvon Neumann classiﬁcation theory). The result of the classiﬁcation theory can thus be summarized in the form of the following table:
range of d {0, 1, 2, . . . , n} {0, 1, 2, . . . , ∞} [0, 1] IR+ {0, ∞}
type of factor N In I∞ II1 II∞ III∞
the lattice P(M) modular orthomodular, nonmodular modular orthomodular, nonmodular orthomodular, nonmodular
3.5 The type II1 factor Thus the signiﬁcance of the existence of type II1 factors is that their projection lattices are modular. (Accordingly, the set of all (not necessarily bounded) selfadjoint operators that they determine are free of the pathologies which von Neumann considered undesirable.) Von Neumann’s conclusion: This makes me strongly inclined, therefore, to take the ring of all bounded operators of Hilbert space (“Case I∞ ” in our notation) less seriously, and Case II1 more seriously, when thinking of an ultimate basis of quantum mechanics. (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 56] As can be inferred from von Neumann’s letter to Birkhoﬀ (November 6, 1935 ?), [R´edei, 2005, p. 5964], Birkhoﬀ suggested another idea to save the modularity of the lattice formed by some subspaces of a Hilbert space: by restricting the linear subspaces to the ﬁnite dimensional ones. Von Neumann did not consider this idea in detail, but thought that it was not an attractive one: Many thanks for your letter. Your idea of requiring a ≤ c → a∪(b∩c) = (a∪b)∩c in Hilbert space for the ﬁnite a, b, c only is very interesting, but will it permit to diﬀerentiate between Hilbertspace and other Banachspaces? (von Neumann to Birkhoﬀ, November 13, 1935 ?), [R´edei, 2005, p. 59]
The Birkhoﬀ–von Neumann Concept of Quantum Logic
11
Rather than answering this rhetorical question, von Neumann makes his famous confession (quoted in part by Birkhoﬀ in [Birkhoﬀ, 1961]), reaﬃrming that the operator algebraic results related to classiﬁcation theory of von Neumann algebras reduce the privileged status of Hilbert space quantum mechanics: I would like to make a confession which may seem immoral: I do not believe absolutely in Hilbert space any more. After all Hilbertspace (as far as quantummechanical things are concerned) was obtained by generalizing Euclidean space, footing on the principle of “conserving the validity of all formal rules”. This is very clear, if you consider the axiomaticgeometric deﬁnition of Hilbertspace, where one simply takes Weyl’s axioms for a unitaryEuclideanspace, drops the condition on the existence of a ﬁnite linear basis, and replaces it by a minimum of topological assumptions (completeness + separability). Thus Hilbertspace is the straightforward generalization of Euclidean space, if one considers the vectors as the essential notions. Now we4 begin to believe, that it is not the vectors which matter but the lattice of all linear (closed) subspaces. Because: 1. The vectors ought to represent the physical states, but they do it redundantly, up to a complex factor, only. 2. And besides the states are merely a derived notion, the primitive (phenomenologically given) notion being the qualities, which correspond to the linear closed subspaces. But if we wish to generalize the lattice of all linear closed subspaces from a Euclidean space to inﬁnitely many dimensions, then one does not obtain Hilbert space, but that conﬁguration, which Murray and I called “case II1 .” (The lattice of all linear closed subspaces of Hilbertspace is our “case I∞ ”.) And this is chieﬂy due to the presence of the rule a ≤ c → a ∪ (b ∩ c) = (a ∪ b) ∩ c This “formal rule” would be lost, by passing to Hilbert space! (von Neumann to Birkhoﬀ, November 13, 1935 ?), [R´edei, 2005, p. 59]
3.6
From the type II1 factor to abstract continuous geometry
So it would seem that the modular lattice of the type II1 factor von Neumann algebra emerges as the strongest candidate for logicization, and so one would expect this lattice to be declared in the Birkhoﬀvon Neumann paper to be the propositional system of quantum logic. But this is not the case; in fact, the published paper does not at all refer to the results of the Murrayvon Neumann classiﬁcation theory of von Neumann algebras to support the modularity postulate. 4 With
F.J. Murray, von Neumann’s coauthor.
12
Mikl´ os R´ edei
Why? Von Neumann’s letters to Birkhoﬀ also contain clues for the the answer to this question, and the answer is that von Neumann’s mind moved extremely quickly from the level of abstractness of von Neumann algebras to the level of abstractness represented by continuous geometries — and this move was taking place precisely during the preparation of the quantum logic paper: in his letter to Birkhoﬀ (November 6, 1935 ?) von Neumann writes: Mathematically – and physically, too – it seems to be desirable, to try to make a general theory of dimension in complemented, irreducible Blattices, without requiring “ﬁnite chain conditions”. I am convinced, that by adding a moderate amount of continuityconditions, the existence of a numerical dimensionality could be proved, which 1. is uniquely determined by its formal properties, 2. and after a suitable normalisation has either the range d = 1, 2, . . . , n (n = 1, 2, . . . , ﬁnite!) or d ≥ 0, ≤ 1. I have already obtained some results in this direction, which connect the notion of dimension in a very funny way with the perspectivities and projectivities in projective geometry. It will perhaps amuse you if I give some details of this. Here they are: (von Neumann to Birkhoﬀ, November 6, 1935 ?), [R´edei, 2005, p. 56] And there follows in the letter a three page long exposition of the theory of continuous geometries, which is not reproduced here. In his letter written a week later (November 13, 1935 ?), [R´edei, 2005, p. 5964], von Neumann gives an even more detailed description of his results on continuous geometry, which conﬁrm the two conjectures 1. and 2. above completely: on every projective geometry there exists a dimension function d having the properties 0 ≤ d(A) d(A) + d(B) =
≤1 d(A ∨ B) + d(A ∧ B)
(16) (17)
and having discrete or continuous range.5 These results do not appear in the Birkhoﬀvon Neumann paper on quantum logic, von Neumann published them separately [von Neumann, 1936] (cf. footnote 33 in the Birkhoﬀvon Neumann paper [Birkhoﬀ and von Neumann, 1936]). Thus by the time it came to the ﬁnal version of the quantum logic paper, von Neumann knew that the projection lattice of a type II1 von Neumann algebra is just a special case of more general continuous geometries that admit wellbehaving probability measures, and this explains why the major postulate in the Birkhoﬀvon Neumann paper is formulated in the section entitled “Relation to abstract projective geometries” and reads: 5 The discrepancy between Eq. (16) and the ranges mentioned under 2. above are due to diﬀerent normalizations.
The Birkhoﬀ–von Neumann Concept of Quantum Logic
13
Hence we conclude that the propositional calculus of quantum mechanics has the same structure as an abstract projective geometry. [Birkhoﬀ and von Neumann, 1936] (Emphasis in the original)
3.7
Probability and quantum logic
The ﬁnite dimension function on a projective geometry, in particular the dimension function with the range [0, 1] on the continuous projective geometry, was for von Neumann crucially important in his search for a proper quantum logic: he interpreted the dimension function as an a priori probability measure on the modular lattice of the quantum propositional system. Thus by requiring quantum logic to be an abstract continuous geometry with a dimension function, Birkhoﬀ and von Neumann created an analogy with classical logic and probability theory, where a Boolean algebra is both a propositional system and a random event structure on which probability measures are deﬁned. While there is no detailed discussion of this aspect of the dimension function in the von NeumannBirkhoﬀ correspondence, the Birkhoﬀvon Neumann paper points out that properties (16)(17) of the dimension function describe the formal properties of probability. Von Neumann regarded it as another pathology of the total Hilbert lattice P(H) that there exists no probability measure on it that satisﬁes conditions (16)(17. This is because one has the following theorem: PROPOSITION 3. Let L be a bounded lattice. If there exists a ﬁnite dimension function d on L (i.e. a map d from L into the set of real numbers that has the properties (16)(17)), then the lattice is modular. It is very easy to see that subadditivity (property (17)) is a necessary condition for a measure to be interpreted as probability understood as relative frequency in the sense of von Mises [von Mises, 1919], [von Mises, 1981]: Assume that the probability p(X) (X = A, B, A ∧ B, A ∨ B) is to be interpreted as relative frequency in the following sense: 1. There exists a ﬁxed ensemble E consisting of N events such that 2. for each event X one can decide unambiguously and 3. without changing the ensemble whether X is the case or not; 4. p(X) = case.6
#(X) N
where #(X) is the number of events in E for which X is the
Under the assumptions 1.4. it trivially follows that (17) holds since one can 6 Strictly
#(X)
speaking one should write p(X) = limN →∞ N ; however, the limit is not important from the point of view of the present considerations, so we omit it.
14
Mikl´ os R´ edei
write #(A ∪ B) #(A ∩ B) + N N #((A \ A ∩ B) ∪ (B \ A ∩ B) ∪ A ∩ B)) #(A ∩ B) + N N #(A \ A ∩ B) + #(B \ A ∩ B) + #(A ∩ B) + #(A ∩ B) N #(A) + #(B) N
= = =
which is the subadditivity. Thus, if a map d on a lattice does not have subadditivity (17) then the probabilities d(X) cannot be interpreted as probabilities in the sense of relative frequency formulated above via 1.4.; consequently, the lattice cannot be viewed as representing a collection of random events in the sense of a relative frequency interpreted probability theory speciﬁed by 1.4. (with the understanding that A ∧ B denotes the joint occurrence of events A and B). Since von Neumann embraced the frequency interpretation of probability in the years 19271935, this makes understandable why he considered the subadditivity (17) a key feature of probability and, consequently, modularity an important condition to require. Thus it would seem that within the mathematical framework of continuous geometry, especially within the theory of the type II1 von Neumann algebras, the Birkhoﬀ and von Neumann concept of quantum logic could restore the harmonious classical picture: random events can be identiﬁed with the propositions stating that the event happens, and probabilities can be viewed as relative frequencies of the occurrences of the events. But this restored harmony is illusory for the following reason: von Neumann and Murray showed that a dimension function d on the projection lattice P(N ) of a type II1 algebra N can be extended to a trace τ on N . The deﬁning property of a trace τ is (18) τ (XY ) = τ (Y X)
for all X, Y ∈ N .
That is to say, the trace is exactly the functional which is insensitive (in the sense of (18)) for the noncommutativity of the algebra. On the other hand, it can be shown that a normal state φ on a von Neumann lattice satisﬁes the additivity (17) if (and only if) it is a trace [Petz and Zemanek, 1988]. Thus, the only quantum probability measures that mesh with the relative frequency interpretation via 1.4. are the ones given by the trace. Behind the mathematical fact that only traces satisfy subadditivity lies the conceptual diﬃculty that assumptions 2. and 3. of the frequency interpretation of probability cannot be upheld in interpreting the elements of a von Neumann lattice as random quantum events and the lattice operation A ∧ B as the joint occurrence of A and B: assumption 3. fails if “deciding” means “measuring”, since measuring disturbs the measured system, hence also the ensemble; therefore, there is no single, ﬁxed, welldeﬁned ensemble in which to compute as relative frequencies the probabilities of all projections representing quantum attributes.
The Birkhoﬀ–von Neumann Concept of Quantum Logic
15
Von Neumann was fully aware of this diﬃculty: One of his arguments against hidden variables is essentially the argument that if hidden parameters did exist, then it should be possible to resolve any ensemble into subensembles that are dispersionfree, but this is not possible if “resolving” means selecting subensembles by measurement, since if one selection ensures dispersionfreeness with respect to observable Q1 , the subsequent selection by measurement of this subensemble into a further subensemble in which another observable Q2 has sharp value destroys the result of the ﬁrst step if Q1 and Q2 are incompatible (see [von Neumann, 1932, p. 304]). Yet, in his 1932 book von Neumann thought to be able to maintain an ensemble interpretation of quantum probability by getting around the problem that quantum measurements disturb the ensemble: Even if two or more quantities R, S in a single system are not simultaneously measurable, their probability distributions in a given ensemble [S1 , . . . SN ] can be obtained with arbitrary accuracy if N is suﬃciently large. Indeed, with an ensemble of N elements it suﬃces to carry out the statistical inspections, relative to the distribution of values of the quantity R, not on all N elements [S1 , . . . SN ], but on any subset of M (≤ N ) elements, say [S1 , . . . SM ] – provided that M, N are both large, and that M is very small compared to N . Then only the M/N th part of the ensemble is aﬀected by the changes which result from the measurement. The eﬀect is an arbitrary small one if M/N is chosen small enough – which is possible for suﬃciently large N , even in the case of large M ... In order to measure two (or several) quantities R, S simultaneuosly, we need two subensembles, say [S1 , . . . SM ], [SM +1 , . . . S2M ] (2M ≤ N ), of such a type that the ﬁrst is employed obtaining the statistics of R, and the second in obtaining those of S. The two measurements therefore do not disturb each other, although they are performed in the same ensemble [S1 , . . . SN ] and they can change this ensemble only by an arbitrarily small amount, if 2M/N is suﬃciently small – which is possible for suﬃciently large N even in the case of large M . . . [von Neumann, 1932, p. 300] Implicit in this reasoning is the assumption that the subensembles are representative of the large ensemble in the sense that the relative frequency of every attribute is the same both in the original and in the subensemble. This nontrivial assumption, known in von Mises’ theory as the requirement of “randomness” concerning the ensembles that can serve as ensembles to compute probabilities as frequencies, is crucial in von Mises’ theory, and von Mises takes pains in giving it a precise formulation (see “Forderung II” in [von Mises, 1919], [von Mises, 1981, p. 61]). Von Neumann does not elaborate on the details and signiﬁcance for his interpretation of quantum probability of the randomness requirement; apparently he did not see any problem with taking advantage of this nontrivial (and controversial) feature of von Mises interpretation.
16
Mikl´ os R´ edei
However, even granting that an ensemble interpretation remains meaningful if one relaxes 3. in the speciﬁcation of the frequency interpretation of probability in the way von Neumann does, the problem remains for von Neumann that 2. does not make sense at all in quantum mechanics if one takes the position that (i) A ∧ B represents the joint occurrence of A and B, and (ii) the joint occurrence cannot be checked by measurement at all on whatever ensemble if A and B are not simultaneously measurable. In sum: There are no “properly noncommutative” probability spaces – as long as one insists on the frequency interpretation of probability; hence, if one wants to maintain the idea of noncommutative probability spaces, with a nondistributive lattice taking the place of Boolean algebra, the frequency view has to go. 4
VON NEUMANN’S POST 1936 STRUGGLE WITH QUANTUM LOGIC
It did: von Neumann abandoned the frequency interpretation after 1936. In an unﬁnished manuscript written about 1937 and entitled “Quantum logic (strictand probability logics)” he writes: This view, the socalled ‘frequency theory of probability’ has been very brilliantly upheld and expounded by R. von Mises. This view, however, is not acceptable to us, at least not in the present ‘logical’ context. [von Neumann, 1961] Instead, von Neumann embraces in this unﬁnished note a “logical theory of probability”, which he associates with J. M. Keynes, but which he does not spell out in detail in the unﬁnished note. The only formulation of this idea can be found in his talk delivered years later at the International Congress of mathematicians in Amsterdam 1954: Essentially if a state of a system is given by one vector, the transition probability in another state is the inner product of the two which is the square of the cosine of the angle between them. In other words, probability corresponds precisely to introducing the angles geometrically. Furthermore, there is only one way to introduce it. The more so because in the quantum mechanical machinery the negation of a statement, so the negation of a statement which is represented by a linear set of vectors, corresponds to the orthogonal complement of this linear space. And therefore, as soon as you have introduced into the projective geometry the ordinary machinery of logics, you must have introduced the concept of orthogonality. ... In order to have probability all you need is a concept of all angles, I mean angles other than 90◦ . Now it is perfectly quite true that in geometry, as soon as you can deﬁne the right angle, you can deﬁne all angles. Another way to put it is
The Birkhoﬀ–von Neumann Concept of Quantum Logic
17
that if you take the case of an orthogonal space, those mappings of this space on itself, which leave orthogonality intact, leave all angles intact, in other words, in those systems which can be used as models of the logical background for quantum theory, it is true that as soon as all the ordinary concepts of logic are ﬁxed under some isomorphic transformation, all of probability theory is already ﬁxed. What I now say is not more profound than saying that the concept of a priori probability in quantum mechanics is uniquely given from the start. ... This means, however, that one has a formal mechanism, in which logics and probability theory arise simultaneously and are derived simultaneously. [von Neumann, 2001, p. 244245] Von Neumann was intrigued by the determination of probability by logic in the quantum context but he did not consider this “logical theory” (interpretation) of probability as fully understood: he mentions the need for an axiomatic investigation of this issue in his address to the International Congress of Mathematicians (Amsterdam, 1954) [von Neumann, 2001] as one of the open problems in mathematics: he thought that it would shed ... a great deal of new light on logics and probably alter the whole formal structure of logics considerably, if one succeeds in deriving this system from ﬁrst principles, in other words from a suitable set of axioms. [von Neumann, 2001, p. 245] It seems that he worked on this problem and tried to work out a systematic theory of quantum logic after 1936, but he did not succeed. The recently published series of letters he wrote to F.B. Silsbee in [R´edei, 2005], prove this in a remarkable way. The correspondence between von Neumann and Silsbee, then the president of the Washington Philosophical Society, starts with Silsbee’s letter (October 31, 1944, unpublished, Library of Congress, Washington D.C.) inviting von Neumann to deliver the Fourteenth Joseph Henry Lecture scheduled for March 17, 1945. In his reply to Silsbee (November 3, 1944) [R´edei, 2005, p. 216] von Neumann accepted the invitation, promising at the same time to write up the lecture in a paper. In his second letter to Silsbee (February 14, 1945) [R´edei, 2005, p. 217] von Neumann speciﬁes the problem of relation of logic and probability in quantum mechanics as the topic of the lecture. Von Neumann did deliver his talk as planned; however, as his letter to Silsbee (June 11, 1945) [R´edei, 2005, p. 218] shows, he did not meet the ﬁrst deadline of submitting the manuscript of the planned paper entitled “Logic of quantum mechanics”. He promises at the same time to deliver the paper by July 7, the latest. But he did not meet this second deadline either; in fact, the promised paper was never written, and von Neumann’s subsequent letters to Silsbee (July 2, 1945, October 22, 1945, April 20, 1946 and December 23, 1946, all in [R´edei, 2005]) show von Neumann agonizing over this project. Most revealing
18
Mikl´ os R´ edei
is his letter of July 2, 1945 [R´edei, 2005, p. 218220]. Von Neumann confesses that even if he had not been disrupted by warwork, he might not have been able to write the promised paper on quantum logic: It is with great regret that I am writing these lines to you, but I simply cannot help myself. In spite of very serious attempts to write the article on the “Logics of quantum mechanics” I ﬁnd it completely impossible to do it at this time. As you may know, I wrote a paper on this subject with Garrett Birkhoﬀ in 1936 (“Annals of Mathematics”, vol. 37, pp. 823843), and I have thought a good deal on the subject since. My work on continuous geometries, on which I gave the Amer. Math. Soc. Colloqium lectures in 1937, comes to a considerable extent from this source. Also a good deal concerning the relationship between strict and probability logics (upon which I touched brieﬂy in the Henry Joseph Lecture) and the extension of this “Propositional calculus” work to “logics with quantiﬁers” (which I never so far discussed in public). All these things should be presented as a connected whole (I mean the propositional and the “quantiﬁer” strict logics, the probability logics, plus a short indication of the ideas of “continuous” projective geometry), and I have been mainly interrupted in this (as well as in writing a book on continuous geometries, which I still owe the Amer. Math. Soc. Colloqium Series) by the war. To do it properly would require a good deal of work, since the subjects that have to be correlated are a very heterogenous collection — although I think that I can show how they belong together. When I oﬀered to give the Henry Joseph Lecture on this subject, I thought (and I hope that I was not too far wrong in this) that I could give a reasonable general survey of at least part of the subject in a talk, which might have some interest to the audience. I did not realize the importance nor the diﬃculties of reducing this to writing. I have now learned — after a considerable number of serious but very unsuccessful eﬀorts — that they are exceedingly great. I must, of course, accept a good part of the responsibility for my method of writing — I write rather freely and fast if a subject is “mature” in my mind, but develop the worst traits of pedantism and ineﬃciency if I attempt to give a preliminary account of a subject which I do not have yet in what I can believe to be in its ﬁnal form. I have tried to live up to my promise and to force myself to write this article, and spent much more time on it than on many comparable ones which I wrote with no diﬃculty at all — and it just didn’t work. Perhaps if I were not continually interrupted by journeys and other obligations arising from still surviving war work, I might have been able to do it — although I am not even sure of this. (von Neumann to Silsbee, July 2, 1945), [R´edei, 2005][p. 218219]
The Birkhoﬀ–von Neumann Concept of Quantum Logic
19
5 SUMMARY AND CONCLUDING REMARKS As we have seen, the modularity postulate of the 1936 Birkhoﬀvon Neumann concept of quantum logic was motivated by the need to establish conceptual coherence between diﬀerent things “quantum”: (quantum) logic, (quantum) probability and quantum mechanics. Thus to understand the BirkhoﬀVon Neumann concept of quantum logic one has to keep in mind that while searching for quantum logic, von Neumann did not just want to create a nonclassical logic: he was motivated by the desire of creating the nonclassical (noncommutative) analogy of the classical situation where a Boolean algebra plays two roles at the same time: it represents the propositional algebra of a classical propositional calculus and it also represents the set of random events on which probability measures are deﬁned. Moreover, probabilities given by a classical probability measure can in principle be interpreted as relative frequencies in the sense of von Mises (the wellknown diﬃculties of the frequency interpretation notwithstanding). Von Neumann had realized that the failure of modularity in the Hilbert lattice prohibits the interpretation of the Hilbert lattice as an event structure for a relative frequency interpreted noncommutative probability theory and, consequently, he abandoned the Hilbert lattice as quantum logic — thereby also abandoning Hilbert space quantum mechanics as well. He could do this in 1936 only because by then he was in the position to suggest another mathematical framework for quantum theory, a framework that he hoped would be conceptually more suitable than Hilbert space quantum theory: at the time of cooperating with Birkhoﬀ on the 1936 quantum logic paper, he also was working on the theory of “rings of operators” (“von Neumann algebras”), and in the year of the publication of the Birkhoﬀvon Neumann paper on quantum logic von Neumann also published a joint paper with J. Murray that established the classiﬁcation theory of von Neumann algebras. One result of this classiﬁcation theory was the discovery of a speciﬁc type of a von Neumann algebra, the socalled type “II1 factor”. The projections of this algebra form a modular lattice that von Neumann hoped would be a proper quantum logic. The lattice of a type II1 factor is a special case of a continuous geometry. Thus the 1936 Birkhoﬀvon Neumann concept of quantum logic is related to deep mathematical discoveries in the mid thirties and to conceptual diﬃculties in connection with the frequency interpretation of quantum probability. So the issue is a convoluted one. The complexity of the problem is also reﬂected by the fact that, as we have seen, von Neumann himself was never quite satisﬁed with how he had worked out quantum logic: It turned out that even the modular lattices of type II1 von Neumann algebras are too week algebraically to regard them as event structure for a truly noncommutative probability theory — if the probabilities are to be interpreted as relative frequencies. This was likely the main reason why von Neumann abandoned the frequency interpretation of quantum probability after 1936 in favor of a “logical interpretation” of probability, which von Neumann did not regard as very well developed and understood, however. While the conceptual coherence between quantum logic, quantum probability
20
Mikl´ os R´ edei
and quantum mechanics cannot be achieved in the way Birkhoﬀ and von Neumann seem to have envisaged, and the interpretation of quantum probability remains a much debated issue even today, their challenging the Hilbert space formalism was a very signiﬁcant move — albeit one not widely noticed at the time. Through the type II1 factor and its projection lattice the Birkhoﬀvon Neumann concept of quantum logic got related to the classiﬁcation theory of von Neumann algebras, which opened up the possibility that types other than II1 and the I∞ (= Hilbert space quantum mechanics) might also be utilized in applications to quantum systems. The subsequent development of quantum theory proved that von Neumann algebras of all the types discovered by Murray and von Neumann in 1935 are needed in modelling quantum systems (see [R´edei and Summers, 2007] for a review of the role of type in quantum theory). The projection lattices of such von Neumann algebras all represent nonclassical (quantum) logics, and the speciﬁc typerelated features of these projection lattices are crucial in understanding the behavior of the quantum world. Detailing these latter issues is beyond the scope of the present review however. Nor were they discussed in the Birkhoﬀvon Neumann paper: Birkhoﬀ and von Neumann were fully aware that their work was just the beginning of a rich ﬁeld to be developed further: Your general remarks, I think, are very true: I, too, think, that our paper will not be very exhaustive or conclusive, but that we should not attempt to make it such: The subject is obviously only at the beginning of a development, and we want to suggest the direction of this development much more, than to reach “ﬁnal” results. I, for one, do not even believe, that the right formal frame for quantum mechanics is already found. (von Neumann to Birkhoﬀ, November 21, 1935 ?), [R´edei, 2005, p. 65] ACKNOWLEDGEMENTS This paper was written while I was staying as Donders Professor in the Department of History and Foundations of Mathematics and Science, Utrecht University, The Netherlands. I wish to thank Utrecht University and especially, Professor D. Dieks for their hospitality. Work also supported by OTKA (contract number: T 43642) BIBLIOGRAPHY [Birkhoﬀ and von Neumann, 1936] G.D. Birkhoﬀ and J. von Neumann. The logic of quantum mechanics. Annals of Mathematics, 37:823–843, 1936. Reprinted in [Taub, 1961a], No. 7. [Birkhoﬀ, 1961] G. Birkhoﬀ. Lattices in applied mathematics. In R.P. Dilworth, editor, Lattice Theory: Proceedings of the Second Symposium in Pure Mathematics of the American Mathematical Society, Monterey, April, 1959, pages 155–184. American Mathematical Society, Providence, 1961. [Bub, 1981a] J. Bub. Hidden variables and quantum mechanics — a sceptical review. Erkenntnis, 16:275–293, 1981.
The Birkhoﬀ–von Neumann Concept of Quantum Logic
21
[Bub, 1981b] J. Bub. What does quantum logic explain? In E. Beltrametti and B.C. van Fraassen, editors, Current Issues in Quantum Logic, pages 89–100. Plenum Press, New York, 1981. [DallaChiara and Giuntini, 2002] M.L. DallaChiara and R. Giuntini. Quantum logic. In D.M. Gabbay and F. Guenthner, editors, Handbook of Philosophical Logic, volume 6, pages 129–228. Kluwer Academic Publishers, Dordrecht, 2002. [DallaChiara et al., 2007] M.L. DallaChiara, R. Giuntini, and M. R´edei. The history of quantum logic. In D.M. Gabbay, editor, Handbook of History of Logic. Kluwer Academic Publishers, Dordrecht. Vol. 38 (2007), 390–417. [Dunn and Hardegree, 2001] J. Michael Dunn and Garry M. Hardegree. Algebraic Methods in Philosophical Logic. Claredon Press, Oxford, 2001. [Halmos, 1962] P.R. Halmos. Algebraic Logic. Chelsa Publishing Company, New York, 1962. [Hardegree, 1981a] G.M. Hardegree. An axiom system for orthomodular quantum logic. Studia Logica, 40:1–12, 1981. [Hardegree, 1981b] G.M. Hardegree. Some problems and methods in formal quantum logic. In E. Beltrametti and B.C. van Fraassen, editors, Current Issues in Quantum Logic, pages 209–225. Plenum Press, New York, 1981. [Kalmbach, 1981] G. Kalmbach. Omologic as Hilbert type calculus. In E. Beltrametti and B.C. van Fraassen, editors, Current Issues in Quantum Logic, pages 333–340. Plenum Press, New York, 1981. [Murray and von Neumann, 1936] F.J. Murray and J. von Neumann. On rings of operators. Annals of Mathematics, 37:116–229, 1936. Reprinted in [Taub, 1961b] No. 2. [Murray and von Neumann, 1937] F.J. Murray and J. von Neumann. On rings of operators, II. American Mathematical Society Transactions, 41:208–248, 1937. Reprinted in [Taub, 1961b] No. 3. [Murray and von Neumann, 1943] F.J. Murray and J. von Neumann. On rings of operators, IV. Annals of Mathematics, 44:716–808, 1943. Reprinted in [Taub, 1961b] No. 5. [Pavicic, 1992] M. Pavicic. Bibliography on quantum logic. International Journal of Theoretical Physics, 31, 1992. [Petz and R´ edei, 1995] D. Petz and M. R´ edei. John von Neumann and the theory of operator algebras. In F. Brody and T. V´ amos, editors, The Neumann Compendium, volume I of World Scientiﬁc Series of 20th Century Mathematics, pages 163–181. World Scientiﬁc, Singapore, 1995. [Petz and Zemanek, 1988] D. Petz and J. Zemanek. Characterizations of the trace. Linear Algebra and its Applications, 111:43–52, 1988. [Popper, 1968] K.R. Popper. Bikhoﬀ and von Neumann’s interpretation of quantum mechanics. Nature, 219:682–685, 1968. [R´ edei and Summers, 2007] M. R´ edei and S.J. Summers. Quantum probability theory. Studies in the History and Philosophy of Modern Physics, 2007. forthcoming. [R´ edei, 1996] M. R´ edei. Why John von Neumann did not like the Hilbert space formalism of quantum mechanics (and what he liked instead). Studies in the History and Philosophy of Modern Physics, 27:1309–1321, 1996. [R´ edei, 1998] M. R´ edei. Quantum Logic in Algebraic Approach, volume 91 of Fundamental Theories of Physics. Kluwer Academic Publisher, 1998. [R´ edei, 1999] M. R´ edei. ‘Unsolved problems in mathematics’ J. von Neumann’s address to the International Congress of Mathematicians Amsterdam, September 29, 1954. The Mathematical Intelligencer, 21:7–12, 1999. [R´ edei, 2001] M. R´ edei. Von Neumann’s concept of quantum logic and quantum probability. In M. R´ edei and M. St¨ oltzner, editors, John von Neumann and the Foundations of Quantum Physics, Institute Vienna Circle Yearbook, pages 153–172. Kluwer Academic Publishers, Dordrecht, 2001. [R´ edei, 2005] M. R´ edei, editor. John von Neumann: Selected Letters, volume 27 of History of Mathematics, Rhode Island, 2005. American Mathematical Society and London Mathematical Society. [R´ edei, 2007] M. R´ edei. The birth of quantum logic. History and Philosophy of Logic, 28:107– 122, May 2007. [Scheibe, 1974] E. Scheibe. Popper and quantum logic. The British Journal for the Philosophy of Science, 25:319–342, 1974.
22
Mikl´ os R´ edei
[Takesaki, 1979] M. Takesaki. Theory of Operator Algebras, volume I. Springer Verlag, New York, 1979. [Taub, 1961a] A.H. Taub, editor. John von Neumann: Collected Works, volume IV. Continuous Geometry and Other Topics, New York and Oxford, 1961. Pergamon Press. [Taub, 1961b] A.H. Taub, editor. John von Neumann: Collected Works, volume III. Rings of Operators, New York and Oxford, 1961. Pergamon Press. [von Mises, 1919] R. von Mises. Grundlagen der Wahrscheinlichkeitsrechnung. Mathematische Zeitschrift, 5:52–99, 1919. [von Mises, 1981] R. von Mises. Probability, Statistics and Truth. Dover Publications, New York, 2nd edition, 1981. Originally published as ‘Wahrscheinlichkeit, Statistik und Wahrheit’ (Springer, 1928). [von Neumann, 1932] J. von Neumann. Mathematische Grundlagen der Quantenmechanik. Springer Verlag, Berlin, 1932. English translation: Mathematical Foundations of Quantum Mechanics (Princeton University Press, Princeton) 1955. [von Neumann, 1936] J. von Neumann. Continuous geometry. Proceedings of the National Academy of Sciences, 22:92–100, 1936. Reprinted in [Taub, 1961a] pp. 126134. [von Neumann, 1940] J. von Neumann. On rings of operators, iii. Annals of Mathematics, 41:94–161, 1940. Reprinted in [Taub, 1961b] No. 4. [von Neumann, 1961] J. von Neumann. Quantum logic (strict and probability logics). In A.H. Taub, editor, John von Neumann: Collected Works, volume IV. Continuous Geometry and Other Topics, pages 195–197. Pergamon Press, New York and Oxford, 1961. Unﬁnished manuscript, reviewed by A.H. Taub. [von Neumann, 2001] J. von Neumann. Unsolved problems in mathematics. In M. R´edei and M. St¨ oltzner, editors, John von Neumann and the Foundations of Quantum Physics, Institute Vienna Circle Yearbook, pages 231–245. Kluwer Academic Publishers, Dordrecht, 2001.
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
23
IS QUANTUM LOGIC A LOGIC? Mladen Paviˇci´c and Norman D. Megill
1
INTRODUCTION
Thirtyseven years ago, Richard Greechie and Stanley Gudder wrote a paper entitled Is a Quantum Logic a Logic? [Greechie and Gudder, 1971] in which they strengthen a previous negative result of Josef Jauch and Constantin Piron. [Jauch and Piron, 1970] “Jauch and Piron have considered a possibility that a quantum propositional system is an inﬁnite valued logic. . . and shown that standard propositional systems (that is, ones that are isomorphic to the lattice of all closed subspaces of a Hilbert space) are not conditional and thus cannot be logic in the usual sense.” [Greechie and Gudder, 1971] A conditional lattice is deﬁned as follows. We deﬁne a valuation v[a] as a mapping from an element a of the lattice to the interval [0, 1]. We say that two elements a, b are conditional if there exists a unique c such that v[c] = min{1, 1 − v[a] + v[b]}. We call c the conditional of a and b and write c = a → b. We say that the lattice is conditional if every pair a, b is conditional. Greechie and Gudder then proved that a lattice is conditional if and only if it contains only two elements 0 and 1.1 This implies that [0,1] reduces to {0, 1} and that the lattice reduces to a twovalued Boolean algebra. In eﬀect, this result shows that one cannot apply the same kind of valuation to both quantum and classical logics. It became obvious that if we wanted to arrive at a proper quantum logic, we should take an axiomatically deﬁned set of propositions closed under substitutions and some rules of inference, and apply a modeltheoretic approach to obtain valuations of every axiom and theorem of the logic. So, a valuation should not be a mapping to [0,1] or {0, 1} but to the elements of a model. For classical logic, a model for logic was a complemented distributive lattice, i.e., a Boolean algebra. For quantum logics the most natural candidate for a model was the orthomodular lattice, while the logics themselves were still to be formulated. Here we come to the question of what logic is. We take that logic is about propositions and inferences between them, so as to form an axiomatic deductive system. The system always has some algebras as models, and we always deﬁne valuations that map its propositions to elements of the algebra — we say, the system always has its semantics — but our deﬁnition stops short of taking semantics to be a part of the 1 We
deﬁne 0 and 1 in a lattice in Section 2.
24
Mladen Paviˇci´ c and Norman D. Megill
system itself. Our title refers to such a deﬁnition of logic, and we call quantum logic so deﬁned deductive quantum logic.2 Classical logic is deductive in the same sense. In the early seventies, a number of results and a number of predecessors to deductive quantum logics were formulated. Jauch, Piron, Greechie, and Gudder above assumed the conditional — from now on we will call it implication — to be deﬁned as a →0 b = a ∪b (see Section 2 for notation). However, it was already then known that in an orthomodular lattice3 an implication so deﬁned would not satisfy the condition a → b = 1 ⇔ a ≤ b , which holds in every Boolean algebra and which was considered plausible to hold in an orthomodular lattice too. In 1970, the following implication was found to satisfy this condition: a →1 b = a ∪ (a ∩ b) (the socalled Sasaki hook4 ) by Peter Mittelstaedt [Mittelstaedt, 1970] and Peter Finch [Finch, 1970]. The Sasaki hook becomes equal to a ∪ b when an orthomodular lattice satisﬁes the distributive law, i.e. when it is a Boolean algebra. The Sasaki implication ﬁrst served several authors simply to reformulate the orthomodular lattice in a logiclike way and call it “quantum logic.” [Finch, 1970; Clark, 1973; Piziak, 1974] In 1974 Gudrun Kalmbach proved that in addition to the Sasaki hook, there are exactly four other “quantum implications” that satisfy the above plausible condition and that all reduce to a ∪ b in a Boolean algebra. In the very same year, four genuine (i.e. propositional) deductive quantum logics — using three diﬀerent implications and none at all, respectively — were formulated by Gudrun Kalmbach [Kalmbach, 1974] (a standard propositional logic based on the Kalmbach implication,5 ) Hermann Dishkant [Dishkant, 1974] (a ﬁrstorder predicate logic based on the Dishkant implication6 ), Peter Mittelstaedt [Mittelstaedt, 1974] (a dialog logic based on the Sasaki hook), and Robert Goldblatt [Goldblatt, 1974] (a binary logic with no implication — the binary inference ‘’ represented the lattice ‘≤’). Several other quantum logics were later formulated by Maria Luisa Dalla Chiara [Dalla Chiara, 1977] (ﬁrstorder quantum logic), Jay Zeman [Zeman, 1978] (normal logic), Hirokazu Nishimura [Nishimura, 1980] (Gentzen sequent logic), George Georgacarakos [Georgacarakos, 1980] (orthomodular logics based on relevance,7 Sasaki, and Dishkant implications), Michael Dunn [Dunn, 1981] (predicate binary logic), ErnstWalter Stachow [Stachow, 1976] (tableaux calculus, a Gentzenlike calculus of sequents, and a Brouwerlike logic), Gary Hardegree [Hardegree, 1981] (orthomodular calculus), John Bell [Bell, 1986] (quantum 2 Note that many authors understand quantum logic as simply a lattice [Jauch, 1968] or a poset [Varadarajan, 1968; Pt´ ak and Pulmannov´ a, 1991]. Quantum logics so deﬁned do not have the aforementioned valuation and are not deductive quantum logics. Such a deﬁnition stems from an operationalist approach, which started with the idea that quantum logic might be empirical. It was argued that propositions might be measured and that properties such as orthomodularity for quantum systems or distributivity for classical ones can be experimentally veriﬁed. [Jauch, 1968] 3 The lattice of all closed subspaces of a Hilbert space is an orthomodular lattice. See Section 2. 4 The Sasaki hook is an orthocomplement to the Sasaki projection [Sasaki, 1964]. 5 Kalmbach implication is deﬁned as a → b = (a ∩ b) ∪ (a ∩ b ) ∪ (a ∩ (a ∪ b)). 3 6 Dishkant implication is deﬁned as a → b = b → a . 2 1 7 Relevance implication is deﬁned as a → b = (a ∩ b) ∪ (a ∩ b) ∪ (a ∩ b ). 5
Is Quantum Logic a Logic?
25
“attribute” logic), Mladen Paviˇci´c [Paviˇci´c, 1987] (binary quantum logics with merged implications8 ), Mladen Paviˇci´c [Paviˇci´c, 1989] (unary quantum logic with merged implications),9 Mladen Paviˇci´c and Norman Megill [Paviˇci´c and Megill, 1999] (unary quantum logics with merged equivalences10 ), etc. Logics with the v(a) = 1 lattice valuation corresponding to a we call unary logics and logics with the v(a) ≤ v(b) lattice valuation corresponding to a b we call binary logics. Still, the parallels with classical logic were a major concern of the researchers at the time. “I would argue that a ‘logic’ without an implication . . . is radically incomplete, and indeed, hardly qualiﬁes as a theory of deduction” (Jay Zeman, 1978). [Zeman, 1978] So, an extensive search was undertaken in the seventies and eighties to single out a “proper quantum implication” from the ﬁve possible ones on purely logical grounds,11 but none of the attempts proved successful. In 1987 Mladen Paviˇci´c [Paviˇci´c, 1987; Paviˇci´c, 1989] proved that there is no “proper quantum implication” since any one of the conditions a →i b = 1 ⇔ a ≤ b, i = 1, . . . , 512 is the very orthomodularity which, when satisﬁed by an orthocomplemented lattice (the socalled ortholattice), makes it orthomodular. In terms of a logic, the corresponding logical rules of inference turn any orthologic or minimal quantum logic into a quantum logic. He also proved that when we add the condition a →0 b = 1 ⇔ a ≤ b is saytisﬁed by an ortholattice, the lattice becomes a Boolean algebra.13 A corresponding logical rule of inference turns any orthologic into a classical logic. This ﬁnding was soon complemented by a proof given by Jacek Malinowski in 1990 that “no logic determined by any class of orthomodular lattices admits the deduction theorem,” [Malinowski, 1990] where the deduction theorem says that if we can derive b from S {a} then we can derive a → b from S.14 He also proved that no extension of quantum logic, i.e., no logic between the quantum and the classical one, satisﬁes the deduction theorem. [Mortensen, 1991] The conclusion was: “Since orthomodular logic is algebraically well behaved, this perhaps shows that implication is not such a desirable operation to have.” [Mortensen, 1991] The conjecture was conﬁrmed by Mladen Paviˇci´c in 1993 [Paviˇci´c, 1993]. The above orthomodularity condition does not require implications. One can also have 8 Under merged implications all six implications are meant; a → b, i = 0, 1, 2, 3, 5 are deﬁned i above; a →4 b = b →3 a is called nontollens implication. In these logics of Paviˇci´ c, axioms of identical form hold for each of the implications yielding ﬁve quantum logics and one classical (for i = 0). 9 Again, axioms of identical form hold for all implications. 10 Merged equivalences, a ≡ b, i = 0, . . . , 5, are explicit expressions (by means of ∪, ∩, ) of i (a →i b) ∩ (b →j a), i = 0, . . . , 5, j = 0, . . . , 5, in any orthomodular lattice as given by Table 1 of Ref. [Paviˇ ci´ c and Megill, 1999]. In these logics, axioms of identical form hold for all equivalences. 11 An excellent contemporary review of the state of the art was written in 1979 by Gary Hardegree [Hardegree, 1979]. 12 a → b, i = 1, . . . , 5 are deﬁned above. See footnotes Nos. 8 and 9. i 13 In any Boolean algebra all six implications merge. 14 It should be stressed here that the deduction theorem is not essential for classical logic either. It was ﬁrst proved by Jaques Herbrand in 1930. [Herbrand, 1931] All classical logic systems before 1930, e.g., the ones by Whitehead and Russell, Hilbert, Ackermann, Post, Skolem, L ukasiewicz, Tarski, etc., were formulated without it.
26
Mladen Paviˇci´ c and Norman D. Megill
it with an essentially weaker equivalence operation: a ≡ b = 1 ⇔ a = b, where a ≡ b = (a ∩ b) ∪ (a ∩ b ); we say a and b are equivalent. [Paviˇci´c, 1993; Paviˇci´c and Megill, 1999] As above, when this condition is satisﬁed by an ortholattice it makes it orthomodular.15 Moreover in any orthomodular lattice a ≡ b = (a →i b) ∩ (b →i a), i = 1, . . . , 5. The analogous classical condition a ≡0 b = 1 ⇔ a = b, where a ≡0 b = (a ∪ b) ∩ (a ∪ b ), amounts to distributivity: when satisﬁed by an ortholattice, it makes it a Boolean algebra. [Paviˇci´c, 1998; Paviˇci´c and Megill, 1999] On the other hand, it turned out that everything in orthomodular lattices is sixfold deﬁned: binary operations, unary operation, variables and even unities and zeros. They all collapse to standard Boolean operations, variables and 0,1 when we add distributivity. For example, as proved by Norman Megill and Mladen Paviˇci´c [Megill and Paviˇci´c, 2001] 01(a,b) = a ∩ (a ∪ b) ∩ (a ∪ b ),. . .,05(a,b) = (a∪b)∩(a∪b )∩(a ∪b)∩(a ∪b ); a ≡3 b = (a ∪b)∩(a∪(a ∩b )); etc. [Megill and Paviˇci´c, 2002] Moreover, we can express any of such expressions by means of every appropriate other in a huge although deﬁnite number of equivalence classes. [Megill and Paviˇci´c, 2002] For example, a shortest expression for ∪ expressed by means of quantum implications is a ∪ b = (a →i b) →i (((a →i b) →i (b →i a)) →i a), i = 1, . . . , 5. [Megill and Paviˇci´c, 2001; Megill and Paviˇci´c, 2002; Paviˇci´c and Megill, 1998a; Megill and Paviˇci´c, 2003] For such a “weird” model, the question emerged as to whether it is possible to formulate a proper deductive quantum logic as a general theory of inference and how independent of its model this logic can be. In other words, can such a logic be more general than its orthomodular model? The answer turned out to be aﬃrmative. In 1998 Mladen Paviˇci´c and Norman Megill showed that the deductive quantum logic is not only more general but also very diﬀerent from their models. [Paviˇci´c and Megill, 1998b; Paviˇci´c and Megill, 1999] They proved that • Deductive quantum logic is not orthomodular. • Deductive quantum logic has models that are ortholattices that are not orthomodular. • Deductive quantum logic is sound and complete under these models. This shows that quantum logic is not much diﬀerent from the classical one since they also proved that [Paviˇci´c and Megill, 1999] • Classical logic is not distributive.16 15 The same holds for a ≡ b, i = 1, . . . , 5 from footnote No. 10, as well. [Paviˇ ci´ c and Megill, i 1999] 16 Don’t be alarmed. This is not in contradiction with anything in the literature. The classical logic still stands intact, and the fact that it is not distributive is just a feature of classical logic that — due to Boole’s heritage — simply has not occurred to anyone as possible and which therefore has not been discovered before. See the proof of Theorem 30, Theorem 45, Lemma 50, and the discussion in Section 10.
Is Quantum Logic a Logic?
27
• Classical logic has models that are ortholattices that are not orthomodular and therefore also not distributive. • Classical logic is sound and complete under these models. These remarkably similar results reveal that quantum logic is a logic in the very same way in which classical logic is a logic. In the present chapter, we present the results in some detail. The chapter is organized as follows. In Section 2, we deﬁne the ortholattice, orthomodular lattice, complemented distributive lattice (Boolean algebra), weakly orthomodular lattice WOML (which is not necessarily orthomodular), weakly distributive lattice WDOL (which is not necessarily either distributive or orthomodular), and some results that connect the lattices. In Section 3, we deﬁne quantum and classical logics. In Sections 4 and 5, we prove the soundness of quantum logic for WOML and of classical logic for WDOL, respectively. In Sections 6 and 7, we prove the completeness of the logics for WOML and WDOL, respectively. In Sections 8 and 9, we prove the completeness of the logics for OML and Boolean algebra, respectively, and show that the latter proofs of completeness introduce hidden axioms of orthomodularity and distributivity in the respective Lindenbaum algebras of the logics. In Section 10, we discuss the obtained results. 2 LATTICES In this section, we introduce two models for deductive quantum logic, orthomodular lattice and WOML, and two models for classical logic, Boolean algebra and WDOL. They are gradually deﬁned as follows. There are two equivalent ways to deﬁne a lattice: as a partially ordered set (poset)17 [Maeda and Maeda, 1970] or as an algebra [Birkhoﬀ, 1948, II.3. Lattices as Abstract Algebras ]. We shall adopt the latter approach. DEFINITION 1. An ortholattice, OL, is an algebra OL0 , , ∪, ∩ such that the following conditions are satisﬁed for any a, b, c ∈ OL0 [Megill and Paviˇci´c, 2002]: a∪b = b∪a
(1)
(a ∪ b) ∪ c = a ∪ (b ∪ c) a = a
(2) (3)
a ∪ (b ∪ b ) = b ∪ b a ∪ (a ∩ b) = a a ∩ b = (a ∪ b )
(4) (5) (6)
In addition, since a ∪ a = b ∪ b for any a, b ∈ OL0 , we deﬁne: 1 = a ∪ a , def
0 = a ∩ a def
(7)
17 Any two elements a and b of the poset have a least upper bound a ∪ b — called join — and a greatest lower bound a ∩ b — called meet.
28
Mladen Paviˇci´ c and Norman D. Megill
and def
⇐⇒
a≤b
a∩b=a
⇐⇒
a∪b=b
(8)
Connectives →1 (quantum implication, Sasaki hook), →0 (classical implication), ≡ (quantum equivalence), and ≡0 (classical equivalence) are deﬁned as follows: def
DEFINITION 3. 18 DEFINITION 4.
def
a →1 b = a ∪ (a ∩ b),
DEFINITION 2.
a →0 b = a ∪ b.
def
a ≡ b = (a ∩ b) ∪ (a ∩ b ). def
a ≡0 b = (a →0 b) ∩ (b →0 a).
Connectives bind from weakest to strongest in the order →1 (→0 ), ≡ (≡0 ), ∪, ∩, and . DEFINITION 5. (Paviˇci´c and Megill [Paviˇci´c and Megill, 1999]) An ortholattice that satisﬁes the following condition: a≡b=1
⇒
(a ∪ c) ≡ (b ∪ c) = 1
(9)
is called a weakly orthomodular ortholattice, WOML. DEFINITION 6. (Paviˇci´c [Paviˇci´c, 1993]) An ortholattice that satiﬁes the following condition: a≡b=1
⇒
a = b,
(10)
is called an orthomodular lattice, OML. Equivalently: DEFINITION 7. (Foulis [Foulis, 1962], Kalmbach [Kalmbach, 1974]) An ortholattice that satisﬁes either of the following conditions: a ∪ (a ∩ (a ∪ b)) = a ∪ b a C b & a C c ⇒ a ∩ (b ∪ c) = (a ∩ b) ∪ (a ∩ c)
(11) (12)
def
where a C b ⇐⇒ a = (a∩b)∪(a∩b ) (a commutes with b), is called an orthomodular lattice, OML. DEFINITION 8. (Paviˇci´c and Megill [Paviˇci´c and Megill, 1999]) An ortholattice that satisﬁes the following condition:19 (a ≡ b) ∪ (a ≡ b ) = (a ∩ b) ∪ (a ∩ b ) ∪ (a ∩ b) ∪ (a ∩ b ) = 1
(13)
is called a weakly distributive ortholattice, WDOL. every orthomodular lattice a ≡ b = (a →1 b) ∩ (b →1 a), but not in every ortholattice. condition is known as commensurability. [Mittelstaedt, 1970, Deﬁnition (2.13), p. 32] Commensurability is a weaker form of the commutativity from Deﬁnition 7. Actually, a metaimplication from commensurability to commutativity is yet another way to express orthomodularity. They coincide in any OML. 18 In
19 This
Is Quantum Logic a Logic?
29
DEFINITION 9. (Paviˇci´c [Paviˇci´c, 1998]) An ortholattice that satisﬁes the following condition: a ≡0 b = 1
⇒
a=b
(14)
is called a Boolean algebra. Equivalently: DEFINITION 10. (Schr¨ oder [Schr¨ oder, 1890]) An ortholattice to which the following condition is added: a ∩ (b ∪ c) = (a ∩ b) ∪ (a ∩ c)
(15)
is called a Boolean algebra. The opposite directions in Eqs. (10) and (14) hold in any OL. Any ﬁnite lattice can be represented by a Hasse diagram that consists of points (vertices) and lines (edges). Each point represents an element of the lattice, and positioning element a above element b and connecting them with a line means a ≤ b. For example, in Figure 1 we have 0 ≤ x ≤ y ≤ 1. We also see that in this lattice, e.g., x does not have a relation with either x or y . Deﬁnition 11 and Theorems 12 and 14 will turn out to be crucial for the completeness proofs of both quantum and classical logics in Sections 6 and 7. DEFINITION 11. We deﬁne O6 as the lattice shown in Figure 1, with the meaning 0 < x < y < 1 and 0 < y < x < 1,
y r
1r @ @r x
x r @ @r 0
r y
Figure 1. Ortholattice O6, also called benzene ring and hexagon. THEOREM 12. An ortholattice is orthomodular if only if it does not include a subalgebra isomorphic to the lattice O6. Proof. Samuel Holland [Holland, 1970]. See also Gudrun Kalmbach [Kalmbach, 1983, p. 22]. COROLLARY 13. O6 violates the distributive law. Proof. Distributivity implies orthomodularity. We can also easily verify on the diagram: y ∩ (x ∪ x ) = y ∩ 1 = y, but (y ∩ x) ∪ (y ∩ x ) = x ∪ 0 = x.
30
Mladen Paviˇci´ c and Norman D. Megill
THEOREM 14. All conditions of WOML and WDOL hold in O6. Proof. As given by Mladen Paviˇci´c and Norman Megill. [Paviˇci´c and Megill, 1998b; Paviˇci´c and Megill, 1999] It boils down to the fact that O6 violates none of the conditions given by Eqs. (16), (9), and (13) THEOREM 15. There exist WDOL lattices that are not orthomodular and therefore not distributive, WOML lattices that are not orthomodular, ortholattices that are neither WOML nor WDOL, and there are WOML lattices that are not WDOL. Proof. As given by Mladen Paviˇci´c and Norman Megill. [Paviˇci´c and Megill, 1998b; Paviˇci´c and Megill, 1999]. On the one hand, the equations that hold in OML and Boolean algebra properly include those that hold in WOML and WDOL, since WOML and WDOL are strictly more general classes of algebras. But on the other hand, there is also a sense in which the equations of WOML and WDOL can be considered to properly include those of OML and Boolean algebra, via mappings that the next theorems describe. THEOREM 16. The equational theory of OMLs can be simulated by a proper subset of the equational theory of WOMLs. Proof. The equational theory of OML consists of equality conditions, Eqs. (1)–(6) together with the orthomodularity condition Eq. (11) (or Eq. (10) or Eq. (12)). We construct a mapping from these conditions to WOML conditions as follows. We map each of the OML conditions, which is an equation in the form t = s (where t and s are terms), to the equation t ≡ s = 1, which holds in WOML. Any equational proof in OML can then be simulated in WOML by replacing each axiom reference in the OML proof with its corresponding WOML mapping [Paviˇci´c and Megill, 2008]. Such a mapped proof will use only a proper subset of the equations that hold in WOML: any equation whose righthand side does not equal 1, such as a = a, will never be used. COROLLARY 17. No set of equations of the form t ≡ s = 1, where t and s are terms in OML and where t = s holds in OML, determines OML when added to the conditions for ortholattices. Proof. Theorem 16 shows that all equations of this form hold in a WOML and none of WOML conditions given by Eqs. (16,9) is violated by O6. Hence, Theorem 12 completes the proof. THEOREM 18. The equational theory of Boolean algebras can be simulated by a proper subset of the equational theory of WDOLs.
Is Quantum Logic a Logic?
31
Proof. The equational theory of Boolean algebras consists of equality conditions Eqs. (1)–(6) together with the distributivity condition Eq. (15). We construct a mapping from these conditions into WDOL as follows. We map each of the Boolean algebra conditions, which is an equation in the form t = s (where t and s are terms), to the equation t ≡0 s = 1, which holds in WDOL. Any equational proof in a Boolean algebra can then be simulated in WDOL by replacing each condition reference in the Boolean algebra proof with its corresponding WDOL mapping. [Paviˇci´c and Megill, 2008] Such a mapped proof will use only a proper subset of the equations that hold in WDOL: any equation whose righthand side does not equal 1, such as a = a, will never be used. COROLLARY 19. No set of equations of the form t ≡0 s = 1, where t and s are terms in any Boolean algebra and where t = s holds in the algebra, determines a Boolean algebra when added to an ortholattice. Proof. Theorem 18 shows that all equations of this form hold in a WDOL and none of WDOL conditions given by Eqs. (16,8) is violated by O6. Hence, Corollary 13 completes the proof. 3
LOGICS
Logic, L, is a language consisting of propositions and a set of conditions and rules imposed on them called axioms and rules of inference. The propositions we use are wellformed formulas (wﬀs), deﬁned as follows. We denote elementary, or primitive, propositions by p0 , p1 , p2 , . . ., and have the following primitive connectives: ¬ (negation) and ∨ (disjunction). The set of wﬀs is deﬁned recursively as follows: pj is a wﬀ for j = 0, 1, 2, . . . ¬A is a wﬀ if A is a wﬀ. A ∨ B is a wﬀ if A and B are wﬀs. We introduce conjunction with the following deﬁnition: def
DEFINITION 20. A ∧ B = ¬(¬A ∨ ¬B). The statement calculus of our metalanguage consists of axioms and rules from the object language as elementary metapropositions and of compound metapropositions built up by means of the following metaconnectives: ∼ (not), & (and), ∨ (or), ⇒ (if. . . , then), and ⇔ (iﬀ), with the usual classical meaning. Our metalanguage statement calculus is actually the very same classical logic we deal with in this chapter, only with the {0,1} valuation. We extend the statement calculus of the metalanguage with ﬁrstorder predicate calculus — with quantiﬁers ∀ (for all) and ∃ (exists) — and informal set theory in the usual way.
32
Mladen Paviˇci´ c and Norman D. Megill
The operations of implication are the following ones (classical, Sasaki, and Kalmbach) [Paviˇci´c, 1987]: def
DEFINITION 21.
A →0 B = ¬A ∨ B.
DEFINITION 22.
A →1 B = ¬A ∨ (A ∧ B).
DEFINITION 23.
A →3 B = (¬A ∧ B) ∨ (¬A ∧ ¬B) ∨ (A ∧ (¬A ∨ B)).
def def
We also deﬁne the equivalence operations as follows: def
DEFINITION 24.
A ≡ B = (A ∧ B) ∨ (¬A ∧ ¬B).
DEFINITION 25.
A ≡0 B = (A →0 B) ∧ (B →0 A).
def
Connectives bind from weakest to strongest in the order →, ≡, ∨, ∧, ¬. Let F ◦ be the set of all propositions, i.e., of all wﬀs. Of the above connectives, ∨ and ¬ are primitive ones. Wﬀs containing ∨ and ¬ within logic L are used to build an algebra F = F ◦ , ¬, ∨. In L, a set of axioms and rules of inference are imposed on F. From a set of axioms by means of rules of inference, we get other expressions which we call theorems. Axioms themselves are also theorems. A special symbol is used to denote the set of theorems. Hence A ∈ iﬀ A is a theorem. The statement A ∈ is usually written as A. We read this: “A is provable” since if A is a theorem, then there is a proof for it. We present the axiom systems of our propositional logics in schemata form (so that we dispense with the rule of substitution).
3.1 Quantum Logic All unary quantum logics we mentioned in the Introduction are equivalent. Here we present Kalmbach’s quantum logic because it is the system which has been investigated in the greatest detail in her book [Kalmbach, 1983] and elsewhere [Kalmbach, 1974; Paviˇci´c and Megill, 1998b]. Quantum logic, QL, is deﬁned as a language consisting of propositions and connectives (operations) as introduced above, and the following axioms and a rule of inference. We will use QL to denote provability from the axioms and rule of QL and omit the subscript when it is clear from context (such as in the list of axioms that follow). Axioms A1 A2
A≡A A ≡ B →0 (B ≡ C →0 A ≡ C)
(16) (17)
A3 A4
A ≡ B →0 ¬A ≡ ¬B A ≡ B →0 A ∧ C ≡ B ∧ C
(18) (19)
A5 A6
A∧B ≡B∧A A ∧ (B ∧ C) ≡ (A ∧ B) ∧ C
(20) (21)
A7
A ∧ (A ∨ B) ≡ A
(22)
Is Quantum Logic a Logic?
33
¬A ∧ A ≡ (¬A ∧ A) ∧ B A ≡ ¬¬A
(23) (24)
A10 A11
¬(A ∨ B) ≡ ¬A ∧ ¬B A ∨ (¬A ∧ (A ∨ B)) ≡ A ∨ B
(25) (26)
A12 A13
(A ≡ B) ≡ (B ≡ A) A ≡ B →0 (A →0 B)
(27) (28)
A14 A15
(A →0 B) →3 (A →3 (A →3 B)) (A →3 B) →0 (A →0 B)
(29) (30)
A8 A9
Rule of Inference (Modus Ponens) R1
A &
A →3 B
⇒
B
(31)
In Kalmbach’s presentation, the connectives ∨, ∧, and ¬ are primitive. In the base set of any model (such as an OML or WOML model) that belongs to OL, ∩ can be deﬁned in terms of ∪ and , as justiﬁed by DeMorgan’s laws, and thus the corresponding ∧ can be deﬁned in terms of ∨ and ¬ (Deﬁnition 20). We shall do this for simplicity. Regardless of whether we consider ∧ primitive or deﬁned, we can drop axioms A1, A11, and A15 because it has been proved that they are redundant, i.e., can be derived from the other axioms. [Paviˇci´c and Megill, 1998b] Note that A11 is what we would expect to be the orthomodularity20 — see Eq. (37) and the discussion following the equation. DEFINITION 26. For Γ ⊆ F ◦ we say A is derivable from Γ and write Γ QL A or just Γ A if there is a sequence of formulas ending with A, each of which is either one of the axioms of QL or is a member of Γ or is obtained from its precursors with the help of a rule of inference of the logic.
3.2
Classical Logic
We make use of the PM classical logical system CL (Whitehead and Russell’s Principia Mathematica axiomatization in Hilbert and Ackermann’s presentation [Hilbert and Ackermann, 1950] but in schemata form so that we dispense with their rule of substitution). In this system, the connectives ∨ and ¬ are primitive, and the →0 connective shown in the axioms is implicitly understood to be expanded according to its deﬁnition. We will use CL to denote provability from the axioms and rule of CL, omitting the subscript when it is clear from context. Axioms
20 Cf.
A1
A ∨ A →0 A
(32)
A2 A3
A →0 A ∨ B A ∨ B →0 B ∨ A
(33) (34)
A4
(A →0 B) →0 (C ∨ A →0 C ∨ B)
(35)
Deﬁnition (7), Eq. (11)
34
Mladen Paviˇci´ c and Norman D. Megill
Rule of Inference (Modus Ponens) R1
A
&
A →0 B
⇒
B
(36)
We assume that the only legitimate way of inferring theorems in CL is by means of these axioms and the Modus Ponens rule. We make no assumption about valuations of the primitive propositions from which wﬀs are built, but instead are interested in wﬀs that are valid in the underlying models. Soundness and completeness will show that those theorems that can be inferred from the axioms and the rule of inference are exactly those that are valid. We deﬁne derivability in CL, Γ CL A or just Γ A, in the same way as we do for system QL. 4 THE SOUNDNESS OF QL: ORTHOMODULARITY LOST In this section we show that the syntax of QL does not correspond to the syntax of an orthomodular lattice. We do this by proving the soundness of QL for WOML. To prove soundness means to prove that all axioms as well as the rules of inference (and therefore all theorems) of QL hold in its models. Since by Theorem 16 WOML properly includes OML, proving the soundness of QL for OML would not tell us anything new, and we can dispense with it. DEFINITION 27. We call M = L, h a model if L is an algebra and h : F ◦ −→ L, called a valuation, is a morphism of formulas F ◦ into L, preserving the operations ¬, ∨ while turning them into , ∪. Whenever the base set L of a model belongs to WOML (or another class of algebras), we say (informally) that the model belongs to WOML (or the other class). In particular, if we say “for all models in WOML” or “for all WOML models,” we mean for all base sets in WOML and for all valuations on each base set. The term “model” may refer either to a speciﬁc pair L, h or to all possible such pairs with the base set L, depending on context. DEFINITION 28. We call a formula A ∈ F ◦ valid in the model M, and write M A, if h(A) = 1 for all valuations h on the model, i.e. for all h associated with the base set L of the model. We call a formula A ∈ F ◦ a consequence of Γ ⊆ F ◦ in the model M and write Γ M A if h(X) = 1 for all X in Γ implies h(A) = 1, for all valuations h. For brevity, whenever we do not make it explicit, the notations M A and Γ M A will always be implicitly quantiﬁed over all models of the appropriate type, in this section for all WOML models M. Similarly, when we say “valid” without qualiﬁcation, we will mean valid in all models of that type. We now prove the soundness of quantum logic by means of WOML, i.e., that if A is a theorem in QL, then A is valid in any WOML model. THEOREM 29. [Soundness]
ΓA
⇒
Γ M A
Is Quantum Logic a Logic?
35
Proof. We must show that any axiom A1–A15, given by Eqs. (16–30), is valid in any WOML model M, and that any set of formulas that are consequences of Γ in the model are closed under the rule of inference R1, Eq. (31). Let us put a = h(A), b = h(B), . . . By Theorem 16, we can prove that WOML is equal to OL restricted to all orthomodular lattice conditions of the form t ≡ s = 1, where t and s are terms (polynomials) built from the ortholattice operations and t = s is an equation that holds in all OMLs. Hence, mappings of QL axioms and its rule of inference can be easily proved to hold in WOML. Moreover, mappings of A1,A3,A5–A13,A15 and R1 hold in any ortholattice. In particular, the A11 mapping :
(a ∪ (a ∩ (a ∪ b))) ≡ (a ∪ b) = 1
(37)
holds in every ortholattice and A11 itself is redundant, i.e., can be be inferred from other axioms. Notice that by Corollary 17, a ≡ b = 1 does not imply a = b. In particular, Eq. (37) does not imply (a ∪ (a ∩ (a ∪ b))) = (a ∪ b) 5 THE SOUNDNESS OF CL: DISTRIBUTIVITY LOST In this section we show that the syntax of CL does not correspond to the syntax of a Boolean algebra. In a way analogous to the QL soundness proof, we prove the soundness of CL only by means of WDOL. Recall Deﬁnitions 27 and 28 for “model,” “valid,” and “consequence.” We now prove the soundness of classical logic by means of WDOL, i.e., that if A is a theorem in CL, then A is valid in any WDOL model. THEOREM 30. [Soundness]
ΓA
⇒
Γ M A
Proof. We must show that any axiom A1–A4, given by Eqs. (32–35), is valid in any WDOL model M, and that any set of formulas that are consequences of Γ in the model are closed under the rule of inference R1, Eq. (36). Let us put a = h(A), b = h(B), . . . By Theorem 18, we can prove that WDOL is equal to OL restricted to all Boolean algebra conditions of the form t ≡0 s = 1, where t and s are terms and t = s is an equation that holds in all Boolean algebras. Notice that according to Corollary 19, t ≡0 s = 1 is not generally equivalent to t = s in WDOL. For example, the mappings of A1–A3 and R1 hold in every ortholattice, and the ortholattice mapping of A4 does not make the ortholattice even orthomodular let alone distributive. In other words, (a ∩ (b ∪ c)) ≡0 ((a ∩ b) ∪ (a ∩ c)) = 1
(38)
does not imply (a ∩ (b ∪ c)) = ((a ∩ b) ∪ (a ∩ c)), and therefore we cannot speak of distributivity within CL.
36
Mladen Paviˇci´ c and Norman D. Megill
6
THE COMPLETENESS OF QL FOR WOML MODELS: NONORTHOMODULARITY CONFIRMED
Our main task in proving the soundness of QL in the previous section was to show that all axioms as well as the rules of inference (and therefore all theorems) from QL hold in WOML. The task of proving the completeness of QL is the opposite one: we have to impose the structure of WOML on the set F ◦ of formulas of QL. We start with a relation of congruence, i.e., a relation of equivalence compatible with the operations in QL. We make use of an equivalence relation to establish a correspondence between formulas of QL and formulas of WOML. The resulting equivalence classes stand for elements of a WOML and enable the completeness proof of QL by means of this WOML. Our deﬁnition of congruence involves a special set of valuations on lattice O6 (shown in Figure 1 in Section 2) called O6 and deﬁned as follows. Its deﬁnition is the same for both the quantum logic completeness proof in this section and the classical logic completeness proof in Section 7. DEFINITION 31. Letting O6 represent the lattice from Deﬁnition 11, we deﬁne O6 as the set of all mappings o : F ◦ −→ O6 such that for A, B ∈ F ◦ , o(¬A) = o(A) , and o(A ∨ B) = o(A) ∪ o(B). The purpose of O6 is to let us reﬁne the equivalence classes used for the completeness proof, so that the Lindenbaum algebra will be a proper WOML, i.e. one that is not orthomodular. This is accomplished by conjoining the term (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)] to the equivalence relation deﬁnition, meaning that for equivalence we require also that (whenever the valuations o of the wﬀs in Γ are all 1) the valuations of wﬀs A and B map to the same point in the lattice O6. For example, the two wﬀs A∨B and A∨(¬A∧(A∨B)) will become members of two separate equivalence classes by Theorem 37 below. Without the conjoined term, these two wﬀs would belong to the same equivalence class. The point of doing this is to provide a completeness proof that is not dependent in any way on the orthomodular law, to show that completeness does not require that the underlying models be OMLs. THEOREM 32. The relation of equivalence ≈Γ,QL or just ≈, deﬁned as A≈B
(39)
def
= Γ A ≡ B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)],
is a relation of congruence in the algebra F, where Γ ⊆ F ◦ Proof. Let us ﬁrst prove that ≈ is an equivalence relation. A ≈ A follows from A1 [Eq. (16)] of system QL and the identity law of equality. If Γ A ≡ B, we can detach the lefthand side of A12 to conclude Γ B ≡ A, through the use of A13 and repeated uses of A14 and R1. From this and commutativity of equality, we conclude A ≈ B ⇒ B ≈ A. (For brevity we will not usually mention further
Is Quantum Logic a Logic?
37
uses of A12, A13, A14, and R1 in what follows.) The proof of transitivity runs as follows. A≈B
& B≈C ⇒ ΓA≡B
(40) &
ΓB≡C
& (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)] & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(B) = o(C)] ⇒ ΓA≡C & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B) & o(B) = o(C)]. In the last line above, Γ A ≡ C follows (see Sec. 3.1) using A2, A14 twice, and R1 six times, and the last metaconjunction reduces to o(A) = o(C) by transitivity of equality. Hence the conclusion A ≈ C by deﬁnition. In order to be a relation of congruence, the relation of equivalence must be compatible with the operations ¬ and ∨. These proofs run as follows. A≈B
(41) ⇒ΓA≡B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)] ⇒ Γ ¬A ≡ ¬B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B) ] ⇒ Γ ¬A ≡ ¬B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(¬A) = o(¬B)] ⇒ ¬A ≈ ¬B
A≈B
(42) ⇒ΓA≡B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)] ⇒ Γ (A ∨ C) ≡ (B ∨ C)
& (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) ∪ o(C) = o(B) ∪ o(C)] ⇒ (A ∨ C) ≈ (B ∨ C) In the second step of Eq. 41, we used A3. In the second step of Eq. 42, we used A4 and A10. For the quantiﬁed part of these expressions, we applied the deﬁnition of O6. DEFINITION 33. The equivalence class for wﬀ A under the relation of equivalence ≈ is deﬁned as A = {B ∈ F ◦ : A ≈ B}, and we denote F ◦ / ≈ = {A : A ∈ F ◦ }. The equivalence classes deﬁne the natural morphism f : F ◦ −→ F ◦ / ≈, which gives f (A) =def A. We write a = f (A), b = f (B), etc.
38
Mladen Paviˇci´ c and Norman D. Megill
LEMMA 34. The relation a = b on F ◦ / ≈ is given by: A = B
⇔
A≈B
(43)
LEMMA 35. The Lindenbaum algebra A = F ◦ / ≈, ¬/ ≈, ∨/ ≈ is a WOML, i.e., Eqs. (1)–(6) and Eq. (9) hold for ¬/ ≈ and ∨/ ≈ as and ∪ respectively [where — for simplicity — we use the same symbols ( and ∪) as for O6, since there are no ambiguous expressions in which the origin of the operations would not be clear from the context]. Proof. For the Γ A ≡ B part of the A ≈ B deﬁnition, the proofs of the ortholattice conditions, Eqs. (1)–(6), follow from A5, A6, A9, the dual of A8, the dual of A7, and DeMorgan’s laws respectively. (The duals follow from DeMorgan’s laws, derived from A10, A9, and A3.) A11 gives us an analog of the OML law for the Γ A ≡ B part, and the WOML law Eq. (9) follows from the OML law in an ortholattice. For the quantiﬁed part of the A ≈ B deﬁnition, lattice O6 is a WOML by Theorem 14. LEMMA 36. In the Lindenbaum algebra A, if f (X) = 1 for all X in Γ implies f (A) = 1, then Γ A. Proof. Let us assume that f (X) = 1 for all X in Γ implies f (A) = 1 i.e. A = 1 = A ∪ A = A ∨ ¬A, where the ﬁrst equality is from Deﬁnition 33, the second equality follows from Eq. (7) (the deﬁnition of 1 in an ortholattice), and the third from the fact that ≈ is a congruence. Thus A ≈ (A∨¬A), which by deﬁnition means Γ A ≡ (A∨¬A) & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o((A∨¬A))]. This implies, in particular, Γ A ≡ (A ∨ ¬A). In any ortholattice, a ≡ (a ∪ a ) = a holds. By analogy, we can prove Γ (A ≡ (A ∨ ¬A)) ≡ A from QL axioms A1– A15. Detaching the lefthand side (using A12, A13, A14, and R1), we conclude Γ A. THEOREM 37. The orthomodular law does not hold in A. Proof. This is Theorem 3.27 from [Paviˇci´c and Megill, 1999], and the proof provided there runs as follows. We assume F ◦ contains at least two elementary (primitive) propositions p0 , p1 , . . .. We pick a valuation o that maps two of them, A and B, to distinct nodes o(A) and o(B) of O6 that are neither 0 nor 1 such that o(A) ≤ o(B) [i.e. o(A) and o(B) are on the same side of hexagon O6 in Figure 1 in Section 2]. From the structure of O6, we obtain o(A) ∪ o(B) = o(B) and o(A)∪(o(A) ∩(o(A)∪o(B))) = o(A)∪(o(A) ∩o(B)) = o(A)∪0 = o(A). Therefore o(A) ∪ o(B) = o(A) ∪ (o(A) ∩ (o(A) ∪ o(B)), i.e., o(A ∨ B) = o(A ∨ (¬A ∧ (A ∨ B))). This falsiﬁes (A ∨ B) ≈ (A ∨ (¬A ∧ (A ∨ B)). Therefore a ∪ b = a ∪ (a ∩ (a ∪ b)), providing a counterexample to the orthomodular law for F ◦ / ≈. LEMMA 38. M = F/ ≈, f is a WOML model.
Is Quantum Logic a Logic?
39
Proof. Follows from Lemma 35.
Now we are able to prove the completeness of QL, i.e., that if a formula A is a consequence of a set of wﬀs Γ in all WOML models, then Γ A. In particular, when Γ = ∅, all valid formulas are provable in QL. (Recall from the note below Deﬁnition 28 that the lefthand side of the metaimplication below is implicitly quantiﬁed over all WOML models M.) THEOREM 39. [Completeness]
Γ M A
⇒
Γ A.
Proof. Γ M A means that in all WOML models M, if f (X) = 1 for all X in Γ, then f (A) = 1 holds. In particular, it holds for M = F/ ≈, f , which is a WOML model by Lemma 38. Therefore, in the Lindenbaum algebra A, if f (X) = 1 for all X in Γ, then f (A) = 1 holds. By Lemma 36, it follows that Γ A. 7 THE COMPLETENESS OF CL FOR WDOL MODELS: NONDISTRIBUTIVITY CONFIRMED In this section we will prove the completeness of CL, i.e., we will impose the structure of WDOL on the set F ◦ of formulas of CL. We start with a relation of congruence, i.e., a relation of equivalence compatible with the operations in CL. We have to make use of an equivalence relation to establish a correspondence between formulas from CL and formulas from WDOL. The resulting equivalence classes stand for elements of a WDOL and enable the completeness proof of CL. THEOREM 40. The relation of equivalence ≈Γ,CL or just ≈, deﬁned as A≈B
(44)
def
= Γ A ≡0 B & (∀o ∈ O6)[(∀X ∈ Γ)(o(X) = 1) ⇒ o(A) = o(B)],
is a relation of congruence in the algebra F. Proof. The axioms and rules of QL, A1–A15 and R1, i.e., Eqs. (16)–(31), are theorems of CL, A1–A4 and R1, i.e. Eqs. (32)–(36). To verify this we refer the reader to Principia Mathematica by Alfred Whitehead and Bertrand Russell [Whitehead and Russell, 1910], where the QL axioms either will be found as theorems or can easily be derived from them. For example, axiom A1 of QL is given as Theorem *4.2 [Whitehead and Russell, 1910, p. 116] after using Theorem *5.23 to convert from ≡0 to ≡. This will let us take advantage of parts of the completeness proof for QL, implicitly using Theorem *5.23 [Whitehead and Russell, 1910, p. 124] in either direction as required. With this in mind, the proof that ≈ is an equivalence and congruence relation becomes exactly the proof of Theorem 32.
40
Mladen Paviˇci´ c and Norman D. Megill
DEFINITION 41. The equivalence class for wﬀ A under the relation of equivalence ≈ is deﬁned as A = {B ∈ F ◦ : A ≈ B}, and we denote F ◦ / ≈ = {A ∈ F ◦ }. The equivalence classes deﬁne the natural morphism f : F ◦ −→ F ◦ / ≈, which gives f (A) =def A. We write a = f (A), b = f (B), etc. LEMMA 42. The relation a = b on F ◦ / ≈ is given as: A = B
⇔
A≈B
(45)
LEMMA 43. The Lindenbaum algebra A = F ◦ / ≈, ¬/ ≈, ∨/ ≈, ∧/ ≈ is a WDOL, i.e., Eqs. (1)–(6) and Eq. (13), hold for ¬/ ≈ and ∨/ ≈ as and ∪ respectively. Proof. For the Γ A ≡0 B part of the A ≈ B deﬁnition, the proofs of the ortholattice axioms are identical to those in the proof of Lemma 35 (after using using Theorem *5.23 on p. 124 of Ref. [Whitehead and Russell, 1910] to convert between ≡0 and ≡). The WDOL law Eq. (13) for the Γ A ≡0 B part can be derived using Theorems *5.24, *4.21, *5.17, *3.2, *2.11, and *5.1 [Whitehead and Russell, 1910, pp. 101–124]. For the quantiﬁed part of the A ≈ B deﬁnition, lattice O6 is a WDOL by Theorem 14. LEMMA 44. In the Lindenbaum algebra A, if f (X) = 1 for all X in Γ implies f (A) = 1, then Γ A.
Proof. Identical to the proof of Lemma 36. THEOREM 45. Distributivity does not hold in A.
Proof. (a ∩ (b ∪ c)) = ((a ∩ b) ∪ (a ∩ c)) fails in O6. Cf. the proof of Theorem 37. LEMMA 46. M = F/ ≈, f is a WDOL model.
Proof. Follows Lemma 43.
Now we are able to prove the completeness of CL, i.e., that if a formula A is a consequence of a set of wﬀs Γ in all WDOL models, then Γ A. In particular, when Γ = ∅, all valid formulas are provable in QL. THEOREM 47. [Completeness]
Γ M A
⇒
ΓA
Proof. Analogous to the proof of Theorem 39.
8 THE COMPLETENESS OF QL FOR OML MODELS: ORTHOMODULARITY REGAINED Completeness proofs for QL carried out in the literature so far — with the exception of Paviˇci´c and Megill [Paviˇci´c and Megill, 1999] — do not invoke Deﬁnition 11 and Theorem 14, and instead of Theorem 32 one invokes the following one:
Is Quantum Logic a Logic?
41
THEOREM 48. Relation ≈ deﬁned as A≈B
def
ΓA≡B
=
(46)
is a relation of congruence in the algebra F. Instead of Deﬁnition 33 one has: DEFINITION 49. The equivalence class under the relation of equivalence is deﬁned as A = {B ∈ F ◦ : A ≈ B}, and we denote F ◦ / ≈ = {A ∈ F ◦ } The equivalence classes deﬁne the natural morphism f : F ◦ −→ F ◦ / ≈, which gives f (A) =def A. We write a = f (A), b = f (A), etc. And instead of Lemma 34 one is able to obtain: LEMMA 50. The relation a = b on F ◦ / ≈ is given as: a=b
⇔
A = B
⇔
A≈B
⇔
ΓA≡B
(47)
Hence, from the following easily provable theorem in QL: (A ≡ B) ≡ (C ∨ ¬C)
⇒
A≡B
(48)
one is also able to get: a≡b=1
⇒
a=b
(49)
in the Lindenbaum algebra A, which is the orthomodularity as given by Deﬁnition 6. [Paviˇci´c, 1998] The point here is that Eq. (49) has nothing to do with any axiom or rule of inference from QL — it is nothing but a consequence of the deﬁnition of the relation of equivalence from Theorem 48. Hence, the very deﬁnition of the standard relation of equivalence introduces a hidden axiom — the orthomodularity — into the Lindenbaum algebra A, thus turning it into an orthomodular lattice. Without this hidden axiom, the Lindenbaum algebra stays WOML as required by the QL syntax. With it the Lindenbaum algebra turns into OML as follows. LEMMA 51. In the Lindenbaum algebra A, if f (X) = 1 for all X in Γ implies f (A) = 1, then Γ A. Proof. In complete analogy to the proof of Theorem 36.
THEOREM 52. The orthomodular law holds in A. Proof. a ∪ (a ∩ (a ∪ b)) = a ∪ b follows from A11, Eq. (26) and Eq. (49).
LEMMA 53. M = F/ ≈, f is an OML model.
Proof. Follows from Lemma 51.
Now we are able to prove the completeness of QL, i.e., that if a formula A is a consequence of a set of wﬀs Γ in all OML models, then Γ A. THEOREM 54. [Completeness]
Γ M A
Proof. Analogous to the proof of Theorem 39.
⇒
ΓA
42
Mladen Paviˇci´ c and Norman D. Megill
9 THE COMPLETENESS OF CL FOR BOOLEAN ALGEBRA MODELS: DISTRIBUTIVITY REGAINED The completeness proof carried out in almost all logic books and textbooks do not invoke Deﬁnition 11, Theorem 14, and Theorem 40. An exception is the Classical and Nonclassical Logics by Eric Schechter [Schechter, 2005, p. 272] who adopted them from Paviˇci´c and Megill [Paviˇci´c and Megill, 1999] and presented in a reduced approach which he called the hexagon interpretation. Other books, though, are based on: THEOREM 55. Relation ≈ deﬁned as A≈B
def
Γ A ≡0 B
=
(50)
is a relation of congruence in the algebra F. Instead of Deﬁnition 41 one has: DEFINITION 56. The equivalence class under the relation of equivalence is deﬁned as A = {B ∈ F ◦ : A ≈ B}, and we denote F ◦ / ≈ = {A ∈ F ◦ } The equivalence classes deﬁne the natural morphism f : F ◦ −→ F ◦ / ≈, which gives f (A) =def A. We write a = f (A), b = f (A), etc. And instead of Lemma 42 one is able to obtain: LEMMA 57. The relation a = b on F ◦ / ≈ is given as: ⇔
a=b
A = B
⇔
A≈B
⇔
Γ A ≡0 B
(51)
Hence, from the following easily provable theorem in CL: (A ≡0 B) ≡0 (C ∨ ¬C)
⇒
A ≡0 B
(52)
one is also able to get: a ≡0 b = 1
⇒
a=b
(53)
in the Lindenbaum algebra A, which is the distributivity as given by Deﬁnition 9. [Paviˇci´c, 1998] The point here is that Eq. (53) has nothing to do with any axiom or rule of inference from CL — it is nothing but a consequence of the deﬁnition of the relation of equivalence from Theorem 55. Hence, the very deﬁnition of the standard relation of equivalence introduces the distributivity as a hidden axiom into the Lindenbaum algebra A and turns it into a Boolean algebra. THEOREM 58. [Completeness]
Γ M A
Proof. Analogous to the proof of Theorem 47.
⇒
ΓA
Is Quantum Logic a Logic?
10
43
DISCUSSION
In the above sections, we reviewed the historical results that we considered relevant to decide whether quantum logic can be considered a logic or not. In the Introduction, we showed that many authors in the past thirty years tried to decide on this question by starting with particular models and their syntax — the orthomodular lattice for quantum logic and Boolean algebra for classical. They compared the models and often came to a conclusion that since they are so diﬀerent, quantum logic should not be considered a logic. This was, however, in obvious conﬂict with the growing number of wellformulated quantum logic systems over the same period. We mentioned some of them in the Introduction. Orthomodular lattices and Boolean algebras are very diﬀerent. As reviewed in the Introduction, in any orthomodular lattice all operations, variables, and constants are sixfold deﬁned (ﬁve quantum and one classical), and in a Boolean algebra they all merge to classical operations, variables, and constants (0,1). Both an orthomodular lattice and a Boolean algebra can be formulated as equational systems — as reviewed in Section 2. Such equational systems can mimic both quantum and classical logics and show that one can formulate the Deduction Theorem in a special orthomodular lattice — a distributive one, i.e., a Boolean algebra — but cannot in a general one. As a consequence, the operation of implication — which the Deduction Theorem21 is based on — plays a special unique role in classical logic and does not in quantum logic. Also, the Boolean algebra used as a model for classical logic is almost always twovalued, i.e., it consists of only two elements 0 and 1, and an orthomodular lattice, according to the KochenSpecker theorem, cannot be given a {0, 1} valuation.22 So, recently research was carried out on whether a logic could have more than one model of the same type, e.g., an ortholattice, with the idea of freeing logics from any semantics and valuation. The result was aﬃrmative, and a consequence was that quantum logic can be considered a logic in the same sense in which classical logic can be considered a logic. The details are given in Sections 3–9, where we chose Kalmbach’s system to represent quantum logic in Section 3.1 and Hilbert and Ackermann’s presentation of Principa Mathematica to represent classical logic in Section 3.2 (although we could have chosen any other system mentioned in the Introduction or from the literature).23 In Sections 4 and 6, we then proved the soundness and completeness, respectively, of quantum logic QL for a nonorthomodular model WOML and in Sections 21 See
footnote No. 14. 2004 Mladen Paviˇci´ c, JeanPierre Merlet, Brendan McKay, and Norman Megill gave exhaustive algorithms for generation of KochenSpecker vector systems with arbitrary number of vectors in Hilbert spaces of arbitrary dimension. [Paviˇ ci´ c et al., 2004; Paviˇ ci´ c et al, 2005; Paviˇ ci´ c, 2005] The algorithms use MMP (McKayMegillPaviˇci´ c) diagrams for which in 3dim Hilbert space a direct correspondence to Greechie and Hasse diagrams can be established. Thus, we also have a constructive proof within the lattice itself. 23 Quantum logics given by Mladen Paviˇ ci´ c [Paviˇ ci´ c, 1989] and by Mladen Paviˇci´ c and Norman Megill [Paviˇ ci´ c and Megill, 1999] are particulary instructive since they contain only axioms designed so as to directly map into WOML conditions. 22 In
44
Mladen Paviˇci´ c and Norman D. Megill
5 and 7 the soundness and completeness, respectively, of classical logic CL for a nondistributive model WDOL. Hence, with respect to these models, quantum logic QL cannot be called orthomodular and classical logic CL cannot be called distributive or Boolean. Also, neither QL nor CL can have a numerical valuation in general, since the truth table method is inapplicable within their OML, WOML, and WDOL models. One might be tempted to “explain” these results in the following way. “It is true that WOML and WDOL obviously contain lattices that violate the orthomodularity law, for example the O6 hexagon (shown in Figure 1 in Section 2) itself, but most probably they also must contain lattices that pass the law and that would, with reference to Theorem 16, explain why we were able to prove the completeness of quantum and classical logic for WOML and WDOL.” This is, however, not the case. We can prove the soundness and completeness of quantum and classical logics using a class of WOML lattices none of which pass the orthomodularity law [Paviˇci´c and Megill, 2008]. Moreover, Eric Schechter has simpliﬁed the results of Paviˇci´c and Megill [Paviˇci´c and Megill, 1999] to the point of proving the soundness and completeness of classical logic for nothing but O6 itself. [Schechter, 2005, p. 272] One of the conclusions Eric Schechter has drawn from the unexpected nondistributivity of the WDOL models, especially when reduced to the O6 lattice alone, is that all the axioms that one can prove by means of {0, 1} truth tables, one can also prove by any Boolean algebra, and by O6. So, logics are, ﬁrst of all, axiomatic deductive systems. Semantics are a next layer that concern models and valuations. Quantum and classical logics can be considered to be two such deductive systems. There are no grounds for considering any of the two logics more “proper” than the other. As we have shown above, semantics of the logics that consider their models show bigger diﬀerences between the two aforementioned classical models than between two corresponding quantum and classical models. Whether we will ever use O6 semantics of classical logic or WOML semantics of quantum logic remains an open question, but these semantics certainly enrich our understanding of the role of logics in applications to mathematics and physics. We cannot make use of bare axiomatics of logic without specifying semantics (models and valuations) for the purpose. By making such a choice we commit ourselves to a particular model and disregard the original logical axioms and their syntax. Thus we do not use quantum logic itself in quantum mechanics and in quantum computers but instead an orthomodular lattice, and we do not use classical logic in our computers today but instead a twovalued Boolean algebra (we even hardly ever use more complicated Boolean algebras). We certainly cannot use O6 semantics to build a computer or an arithmetic; however, one day we might come forward with signiﬁcant applications of these alternative semantics, and then it might prove important to have a common formal denominator for all the models — logics they are semantics of. We can also implement an alternative scenario — searching for diﬀerent ortholattice semantics of the same logics [Paviˇci´c and Megill, 2008].
Is Quantum Logic a Logic?
45
Whatever strategy we choose to apply, we should always bear in mind that the syntaxes of the logics correspond to WOML, WDOL, and O6 semantics (models) while OML and Boolean algebra semantics (models) are imposed on the logics with the help of “hidden” axioms, Eqs. (49) and (53), that emerge from the standard way of deﬁning the relation of equivalence in the completeness proofs, Theorems 48 and 55, of the logics for the latter models. ACKNOWLEDGEMENT We acknowledge the support of the Ministry of Science, Education and Sport of Croatia through teh project No. 08209825623160. BIBLIOGRAPHY [Bell, 1986] J. L. Bell, A New Approach to Quantum Logic, Brit. J. Phil. Sci. 37, 83–99 (1986). [Birkhoﬀ, 1948] G. Birkhoﬀ, Lattice Theory, volume XXV of American Mathematical Society Colloqium Publications, American Mathematical Society, New York, 2nd (revised) edition, 1948. [Clark, 1973] I. D. Clark, An Axiomatization of Quantum Logic, J. Symb. Logic 38, 389–392 (1973). [Dalla Chiara, 1977] M. L. Dalla Chiara, Quantum Logic and Physical Modalities, J. Phil. Logic 6, 391–404 (1977). [Dishkant, 1974] H. Dishkant, The First Order Predicate Calculus Based on the Logic of Quantum Mechanics, Rep. Math. Logic 3, 9–18 (1974). [Dunn, 1981] J. M. Dunn, Quantum mathematics, in PSA 1980: Proceedings of the 1980 Biennial Meeting of the Philosophy of Science Association, edited by P. D. Asquith and R. N. Giere, volume 2, pages 512–531, Philosophy of Science Association, East Lansing, Michigan, 1981. [Finch, 1970] P. D. Finch, Quantum Logic as an Implication Algebra, Bull. Austral. Math. Soc. 2, 101–106 (1970). [Foulis, 1962] D. J. Foulis, A Note on Orthomodular Lattice, Portugal. Math. 21, 65–72 (1962). [Georgacarakos, 1980] G. N. Georgacarakos, Equationally Deﬁnable Implication Algebras for Orthomodular Lattices, Studia Logica 39, 5–18 (1980). [Goldblatt, 1974] R. I. Goldblatt, Semantic Analysis of Orthologic, J. Phil. Logic 3, 19–35 (1974). [Greechie and Gudder, 1971] R. J. Greechie and S. P. Gudder, Is a Quantum Logic a Logic?, Helv. Phys. Acta 44, 238–240 (1971). [Hardegree, 1979] G. M. Hardegree, The Conditional in Abstract and Concrete Quantum Logic, in The LogicoAlgebraic Approach to Quantum Mechanics, edited by C. A. Hooker, volume II. Contemporary Consolidation, pages 49–108, D. Reidel, Dordrecht, 1979. [Hardegree, 1981] G. M. Hardegree, An Axiomatic System for Orthomodular Quantum Logic, Studia Logica 40, 1–12 (1981). [Herbrand, 1931] J. Herbrand, Recherches sur la th´eorie de la d´emonstration, Travaux de la Soci´ et´ e des Sciences et des Lettres de Varsovie, Classe III sci. math. et phys. 24, 12–56 (1931), Translated into English in Jean van Heijenoort (ed.), From Frege to G¨ odel: A Source Book in Mathematical Logic, 18791931, Harvard University Press, Cambridge, Mass. 1967, under the title: “Investigations in Proof Theory: The Properties of True Propositions,” pp. 525–581. [Hilbert and Ackermann, 1950] D. Hilbert and W. Ackermann, Principles of Mathematical Logic, Chelsea, New York, 1950. [Holland, 1970] S. S. Holland, JR., The Current Interest in Orthomodular Lattices, in Trends in Lattice Theory, edited by J. C. Abbot, pages 41–126, Van Nostrand Reinhold, New York, 1970.
46
Mladen Paviˇci´ c and Norman D. Megill
[Jauch, 1968] J. M. Jauch, Foundations of Quantum Mechanics, AddisonWesley, Reading, Massachusetts, 1968. [Jauch and Piron, 1970] J. M. Jauch and C. Piron, What is ‘Quantum Logic’ ?, in Quanta, edited by P. G. O. Freund, C. J. Goebel, and Y. Nambu, pages 166–181, The University of Chicago Press, Chicago and London, 1970. [Kalmbach, 1974] G. Kalmbach, Orthomodular Logic, Z. math. Logik Grundl. Math. 20, 395– 406 (1974). [Kalmbach, 1983] G. Kalmbach, Orthomodular Lattices, Academic Press, London, 1983. [Maeda and Maeda, 1970] F. Maeda and S. Maeda, Theory of Symmetric Lattices, SpringerVerlag, New York, 1970. [Malinowski, 1990] J. Malinowski, The Deduction Theorem for Quantum Logic — Some Negative Results, J. Symb. Logic 55, 615–625 (1990). [Megill and Paviˇci´ c, 2001] N. D. Megill and M. Paviˇci´ c, Orthomodular Lattices and a Quantum Algebra, Int. J. Theor. Phys. 40, 1387–1410 (2001). [Megill and Paviˇci´ c, 2002] N. D. Megill and M. Paviˇci´ c, Deduction, Ordering, and Operations in Quantum Logic, Found. Phys. 32, 357–378 (2002). [Megill and Paviˇci´ c, 2003] N. D. Megill and M. Paviˇci´ c, Quantum Implication Algebras, Int. J. Theor. Phys. 42, 2825–2840 (2003). [Mittelstaedt, 1970] P. Mittelstaedt, Quantenlogische Interpretation orthokomplement¨ arer quasimodularer Verb¨ ande, Z. Naturforsch. 25, 1773–1778 (1970). [Mittelstaedt, 1974] P. Mittelstaedt, Quantum Logic, in PSA 1974, Proceedings of the 1974 Biennial Meeting of the Philosophy of Science Association, edited by R. S. Cohen, C. A. Hooker, A. C. Michalos, and J. W. van Evra, volume 101 of Synthese Library, pages 501–514, D. Reidel, Dordrecht–Holland, 1974. [Mortensen, 1991] C. Mortensen, “Malinowski, Jacek, The Deduction Theorem for Quantum Logic — Some Negative Results,” Mathematical Review 1056375, 91g:03124 (1991). [Nishimura, 1980] H. Nishimura, Sequential method in quantum logic, J. Symb. Logic 45, 339–352 (1980). [Paviˇ ci´ c, 1987] M. Paviˇ ci´ c, Minimal Quantum Logic with Merged Implications, Int. J. Theor. Phys. 26, 845–852 (1987). [Paviˇ ci´ c, 1989] M. Paviˇ ci´ c, Uniﬁed Quantum Logic, Found. Phys. 19, 999–1016 (1989). [Paviˇ ci´ c, 1993] M. Paviˇ ci´ c, Nonordered Quantum Logic and Its YES–NO Representation, Int. J. Theor. Phys. 32, 1481–1505 (1993). [Paviˇ ci´ c, 1998] M. Paviˇ ci´ c, Identity Rule for Classical and Quantum Theories, Int. J. Theor. Phys. 37, 2099–2103 (1998). [Paviˇ ci´ c, 2005] M. Paviˇ ci´ c, Quantum Computation and Quantum Communication: Theory and Experiments, Springer, New York, 2005. [Paviˇ ci´ c and Megill, 1998a] M. Paviˇ ci´ c and N. D. Megill, Quantum and Classical Implication Algebras with Primitive Implications, Int. J. Theor. Phys. 37, 2091–2098 (1998). [Paviˇ ci´ c and Megill, 1998b] M. Paviˇ ci´ c and N. D. Megill, Binary Orthologic with Modus Ponens Is either Orthomodular or Distributive, Helv. Phys. Acta 71, 610–628 (1998). [Paviˇ ci´ c and Megill, 1999] M. Paviˇ ci´ c and N. D. Megill, NonOrthomodular Models for Both Standard Quantum Logic and Standard Classical Logic: Repercussions for Quantum Computers, Helv. Phys. Acta 72, 189–210 (1999). [Paviˇ ci´ c and Megill, 2008] M. Paviˇ ci´ c and N. D. Megill, Standard Logics are ValuationMonotonic, to appear in the Journal of Logic and Computation, (2008). [Paviˇ ci´ c et al., 2004] M. Paviˇ ci´ c, J.P. Merlet, and N. D. Megill, Exhaustive Enumeration of Kochen–Specker Vector Systems, The French National Institute for Research in Computer Science and Control Research Reports RR5388 (2004). [Paviˇ ci´ c et al, 2005] M. Paviˇ ci´ c, J.P. Merlet, B. D. McKay, and N. D. Megill, Kochen–Specker Vectors, J. Phys. A 38, 497–503 (2005), Corrigendum, J. Phys. A 38, 3709 (2005). [Piziak, 1974] R. Piziak, Orthomodular Lattices as Implication Algebras, J. Phil. Logic 3, 413–438 (1974). [Pt´ ak and Pulmannov´ a, 1991] P. Pt´ ak and S. Pulmannov´ a, Orthomodular Structures as Quantum Logics, Kluwer, Dordrecht, 1991. [Sasaki, 1964] U. Sasaki, Orthocomplemented Lattices Satisfying the Exchange Axiom, J. Sci. Hiroshima Univ. A 17, 293–302 (1964). [Schechter, 2005] E. Schechter, Classical and Nonclassical Logics: An Introduction to the Mathematics of Propositions, Princeton University Press, Princeton, 2005.
Is Quantum Logic a Logic?
47
[Schr¨ oder, 1890] E. Schr¨ oder, Vorlesungen u ¨ber die Algebra der Logik (exacte Logik), volume 1, Leipzig, 1890. [Stachow, 1976] E. Stachow, Quantum Logical Calculi and Lattice Structures, J. Phil. Logic 6, 347–386 (1976). [Varadarajan, 1968] V. S. Varadarajan, Geometry of Quantum Theory, Vols. 1 & 2, John Wiley & Sons, NewYork, 1968, 1970. [Whitehead and Russell, 1910] A. N. Whitehead and B. Russell, Principia Mathematica, Cambridge University Press, Cambridge, 1910. [Zeman, 1978] J. J. Zeman, Generalized Normal Logic, J. Phil. Logic 7, 225–243 (1978).
This page intentionally left blank
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
49
IS LOGIC EMPIRICAL? Guido Bacciagaluppi
1
INTRODUCTION
In 1968 Hilary Putnam published a wellknown paper on the question ‘Is logic empirical?’ (Putnam 1968), which gave rise to much controversy in the 1970s and 1980s. The main claims of Putnam’s paper (repeated in Putnam 1974) can be paraphrased as follows: (a) Quantum mechanics prompts us to revise our classical logical notions in favour of ‘quantum logical’ ones. This is explained by analogy to geometry, in the sense that also general relativity prompts us to revise our Euclidean (or rather Minkowskian) geometrical notions in favour of Riemannian (or rather pseudoRiemannian) geometrical notions. (b) This revision of logic is not merely local, i. e. not merely an instance of a logical system especially suited to a particular subject matter, but it is truly global. Quantum logic is the ‘true’ logic (just as the ‘true’ geometry of spacetime is nonEuclidean). Indeed, we have so far failed to recognise that our usual logical connectives are the connectives of quantum logic. (c) Recognising that logic is thus quantum solves the standard paradoxes of quantum mechanics, such as the measurement problem or Schr¨odinger’s cat. Of these truly ambitious and indeed exciting claims, the third claim (c) in particular was discussed extensively, and an almost universal consensus was reached (now shared by Putnam, 1994) that a move to quantum logic, even were it otherwise justiﬁed, would not resolve the puzzles of quantum mechanics. There have been notable reactions also to Putnam’s ﬁrst two claims. Yet, with few exceptions (one needs only to recall the masterly paper by Michael Dummett, 1976), the topic seems to be riddled with misunderstandings. Indeed, very few philosophers appear to still consider seriously the possibility that quantum mechanics might have something to say about the ‘true’ logic (I know of only one recent attempt to resurrect this idea, namely by Michael Dickson, 2001, on whose views more below). This chapter aims at clearing such misunderstandings, and at providing a muchneeded overall assessment of Putnam’s claims, by updating the debate in the light of the current state of the art in the foundations of quantum mechanics. As regards Putnam’s claim (a), I take it that it is indeed justiﬁed, at least provided one takes ‘quantum logic’ as a local logic, suitable to describing a class of propositions in the context of quantum mechanical experiments (or the corresponding class of propositions about properties of quantum mechanical systems).
50
Guido Bacciagaluppi
This claim is analogous to the claim that intuitionistic logic is indeed suitable to describing a class of propositions dealing with mathematical constructions. This is distinct from the claim that intuitionistic logic is in fact the logic that underlies all rigorous human thought (and is thus the ‘true’ logic). Claim (a) understood in this sense, I should think, is relatively uncontroversial, and shall be taken as such for purposes of further discussion. The explanation that quantum logic, suitably deﬁned, has all the main formal properties required of a ‘good’ logic will also fall into this part of the discussion. Claim (b) is the most controversial one, and its assessment will therefore need the most care. There are two points at issue (both well emphasised already by Dummett). The ﬁrst point is that motivating a revision of logic does not only require motivating the introduction of some nonclassical connectives. An advocate of a revision of logic must show why these connectives do not merely sit alongside the classical connectives, but actually replace them. The second point is a Quinean one: such a revision of logic means that, as part of the various revisions to our network of beliefs prompted by the empirical consideration of quantum phenomena, it is possible to choose to make some revisions in our conception of logic. But it is clear that empirical considerations alone cannot force us to revise our logic: a distinctly philosophical component will be needed in order to justify whether a revision of logic, as opposed to a revision somewhere else in our network of beliefs, might be desirable. (In the case of geometry, this is the same situation we have known ever since Poincar´e. And indeed, we shall note in section 7 that in the interpretation of quantum mechanics one can ﬁnd a rather close analogy to issues about conventionalism in physical geometry.) An aspect of claim (b) that is of special importance is the subsidiary claim that the quantum logical connectives are not new connectives that can be deﬁned in terms of the classical ones (and of some additional physical concepts), but that the classical connectives are in fact the quantum logical ones in disguise. We shall therefore have to discuss in depth whether there is a sense in which the classical connectives might be reducible to the quantum logical ones, either in some strict formal sense, or in some physical limit. In this context, as we shall see, questions of interpretation in quantum mechanics play an important role. Indeed, most discussions of quantum logic as the ‘true’ logic have taken place, at least implicitly, in the context of the socalled standard interpretation of quantum mechanics. This, however, is the interpretation that is riddled with the usual paradoxes. As we shall see, which alternative approach to the foundations of quantum mechanics one accepts might inﬂuence the assessment of whether a global revision of logic is acceptable. Conversely, one might add, one’s views on whether a global revision of logic is acceptable might inﬂuence the assessment of which approach to the foundations of quantum mechanics is most appealing. We shall proceed as follows. In section 2, we sketch a few basic elements of quantum mechanics that will be needed later. In section 3, we introduce quantum logic (in its latticetheoretic form) as a local logic of certain experimental propositions; we further discuss the formal properties of such a logic, and mention a few
Is Logic Empirical?
51
alternative forms of quantum logic. Section 4 introduces the socalled standard interpretation of quantum mechanics, and section 5 assesses Putnam’s claims in the context of this interpretation. The claims are found hard to defend, but the standard interpretation itself is not a believable interpretation because it gives rise to the usual paradoxes, at least if one applies it to standard quantum mechanics. Putnam’s claims are thus reassessed ﬁrst, in section 6, in the context of more general quantumlike theories (based on von Neumann algebras), where the classical connectives seem indeed to be reducible to the quantum ones. Then, in section 7, we shall reassess Putnam’s claims in the context of the main current approaches to the foundations of quantum mechanics that explicitly address the paradoxes of the standard theory. Our conclusions will be that, while in the case of the approaches known as de BroglieBohm theory and as spontaneous collapse theories quantum logic at most can be introduced alongside classical logic, and thus in no way can be construed as replacing it, in the case of the Everett (or manyworlds) approach a case can indeed be made that the classical connectives emerge from the quantum ones. Before proceeding, I should emphasise that although the title of this chapter may suggest a general treatment of the question of whether logic is empirical, it will deal only with the question of whether considerations related to quantum mechanics may provide an argument for the general claim. (Putnam’s original paper, 1968, does the same.1 ) Of course, if quantum logic provides us with an intelligible global alternative to classical logic, the case for logic being empirical will be strengthened. However, I believe that a comprehensive assessment of the question of whether empirical considerations might prompt us to revise our logic will depend less on the details of the physics and more on the largely conceptual question of whether the notion of logical consequence is a priori or is an abstraction from what appear to be valid inferences in our practical use of language.2 Indeed, unless one tends towards the latter position, i. e. unless one thinks that classical logic is already an abstraction from the classical empirical world around us (thus already conceding that ‘logic is empirical’), one will be disinclined to take discoveries in microphysics to be relevant at all to the revisability of logic. We shall not attempt to address this more general question. 2 QUANTUM MECHANICS IN A NUTSHELL In the interest of a selfcontained presentation, I summarise a few essentials about quantum mechanics that will be needed below. (This section will be rather abstract but elementary.) In classical mechanics, the state of a system can be represented by a point 1 As a matter of fact, Putnam’s paper was later reprinted with the modiﬁed title ‘The logic of quantum mechanics’. 2 For a recent discussion of the apriorism issue in logic, see e. g. Bueno and Colyvan (2004). Note also that one could very well conceive adopting an apriorist position with regard to quantum logic rather than classical logic.
52
Guido Bacciagaluppi
in (or a subset of, or a probability distribution over) a set called phase space, encoding the positions and momenta of all the particles forming the system. In quantum mechanics, instead, the state of a system is represented by an element in a complex Hilbert space (which is a vector space, equipped with a scalar product, that is complete in the norm induced by the scalar product). In particular, this means that for any two states (e.g., for a spin1/2 system, the states of spinup and spindown in direction x), an arbitrary linear combination (or ‘superposition’) is also a possible state: (1) ϕ = α+x + β−x . Note that the same vector can always be expressed as an appropriate linear combination of vectors in any other basis: (2) ϕ = γ+y + δ−y . In quantum mechanics, overall scalar factors do not count, i. e. the vectors ϕ and εϕ for arbitrary complex ε represent the same state, and by convention all states are usually normalised, i. e. have length 1. Therefore, if the basis vectors are normalised and orthogonal, as in the example above, one has α2 + β2 = γ2 + δ2 = 1. A second crucial distinction between classical and quantum mechanics is that, when describing composite systems in quantum mechanics, instead of taking the Cartesian product of the given phase spaces as in classical mechanics, one has to take the tensor product of the given Hilbert spaces. For instance, for two spin1/2 subsystems with Hilbert spaces generated (spanned) by (3) {+1x , −1x },
{+2x , −2x } ,
one takes the Hilbert space generated by a basis given by the products of the basis vectors: (4) {+1x ⊗ +2x , +1x ⊗ −2x , −1x ⊗ +2x , −1x ⊗ −2x } (this construction is independent of the bases chosen for the subsystems). The fundamental consequence of taking tensor products to describe composite systems is that some states of the composite are not product states, e. g. the socalled singlet state of two spin1/2 systems: 1 (5) √ +1x ⊗ −2x − −1x ⊗ +2x . 2 Such nonfactorisable states are called entangled (the property of being entangled is also independent of the bases chosen for representation in the component systems3 ). If the state of a composite system is entangled, then the subsystems are evidently not described separately by vectors in their respective Hilbert spaces. 3 Incidentally, it is not independent of the choice of the subsystems into which the system is decomposed.
Is Logic Empirical?
53
This is a characteristic trait of quantum mechanics (Schr¨ odinger, 1935, p. 555, called it ‘not .... one but rather the characteristic trait of quantum mechanics’), and it is related to the EinsteinPodolskyRosen paradox, the Bell inequalities, quantum nonlocality et cetera. How can this be? The key, and the third crucial ingredient in our brief summary of quantum mechanics besides Hilbert spaces and tensor products, is the phenomenology of measurement. In classical mechanics one can idealise measurements as testing whether a system lies in a certain subset of its phase space. This can be done in principle without disturbing the system, and the result of the test is in principle fully determined by the state of the system. In quantum mechanics we are empirically confronted with the following situation. (i) Measurements can be idealised as testing whether the system lies in a certain (normclosed) subspace of its Hilbert space — a subset which, in particular, is closed under linear combinations. (ii) A measurement in general disturbs a system: unless the state of the system is either contained in or orthogonal to the tested subspace, the state is projected (‘collapsed’) onto either the tested subspace or its orthogonal complement. (c) This collapse process is indeterministic, and the relevant probabilities are given by the squared norms of the projections of the state on the given subspace and its orthogonal complement, respectively. For example, take the initial state (1) and test for spinup in direction x (test for the subspace P+x ): the ﬁnal state will be +x with probability α2 , or −x with probability β2 . Now take the singlet state (5) as the intial state and test for P+1 x ⊗ P+2 x : the test will come out negative with probability 1, and the state will be undisturbed, since it lies in a subspace orthogonal to the tested one. (The same will be the case if one tests for P−1 x ⊗ P−2 x .) Test instead for P+1 x ⊗ P−2 x (or for P−1 x ⊗ P+2 x ): the result (in both cases) will now be +1x ⊗ −2x or −1x ⊗ +2x , each with probability 1/2. Weaker correlations will be observed if spin is measured along two diﬀerent directions on the two subsystems. Entanglement thus introduces what appear to be irreducible correlations between results of measurements ( even carried out at a distance), and this for a generic pair of tests. The last two elements of quantum mechanics that we shall also refer to are the Schr¨ odinger equation and the notion of (selfajoint) operator as an observable quantity. The Schr¨ odinger equation describes the time evolution of quantum state vectors. It is a linear and unitary equation, i. e. it maps linear combinations into linear combinations, and it preserves the norm (length) of vectors. In its most familiar form, it is a diﬀerential equation for the quantum states represented as complex (squareintegrable) functions on conﬁguration space (the space of positions of all particles), the socalled Schr¨ odinger waves or wave functions. Operators, speciﬁcally selfadjoint operators (which by the spectral theorem can be decomposed uniquely — in the simplest case — into a real linear combination of projectors onto a family of mutually orthogonal subspaces) play two roles in quantum mechanics. On the one hand they mathematically generate Schr¨ odingertype evolutions, on the other hand they can be conveniently used to classify simulta
54
Guido Bacciagaluppi
neous experimental tests of families of mutually orthogonal subspaces. A system will test positively to only one of these tests, and to this test will be associated the measured value of the corresponding observable. Instead of being understood as specifying probabilities for results of tests, quantum states can be thus equivalently understood as specifying expectation values for observables. 3
QUANTUM LOGIC IN A NUTSHELL
3.1 Quantum logic as a logic of experimental propositions The easiest way to introduce the concepts of quantum logic is in terms of a logic of ‘experimental propositions’. That is, one can deﬁne explicitly some nonclassical connectives for a certain special class of propositions, relating to idealised quantum mechanical tests. These connectives will be arguably well suited for the limited subject matter at hand. If as a result one obtains a logical system satisfying certain formal requirements, we shall say that one has introduced a local nonclassical logic. This is meant to be uncontroversial. Indeed, it should be relatively uncontroversial that (provided the formal requirements are indeed met) such a procedure is legitimate, although there may still be scope for disagreement as to how useful the introduction of such a logic is. In order to go on to assess Putnam’s further claims it is essential, at least for the sake of argument, that one accept that in this sense diﬀerent logics may be better adapted to diﬀerent subject matters.4 The prime example for such a procedure is Kolmogoroﬀ’s (1931) interpretation of intuitionistic logic as a calculus of mathematical tasks (Aufgabenrechnung). In this framework, each mathematical proposition p stands for solving the corresponding mathematical task. The classical negation of p (not solving the task) is not itself a mathematical task, so the chosen set of propositions is not closed under classical negation. Instead, showing that a task is impossible to solve is again a mathematical task. This justiﬁes introducing a strong negation, for which the law of excluded middle p ∨ ¬p breaks down. On this basis, one can set up a logical system, which is just the system of intuitionistic logic. More radical claims are not engaged with at this stage. (Indeed, one can argue that this is the correct and only way of interpreting intuitionistic logic, thus safeguarding the primacy of classical logic.)5 In the quantum context, let us deﬁne experimental propositions as (suitable equivalence classes of) statements of the form: ‘The system passes a certain test with probability 1’. From the discussion in the previous section, we recognise that these propositions are in bijective correspondence to closed subspaces of the 4 What we sketch here is quantum logic as descriptive of the empirical behaviour of certain experiments (albeit idealised ones). One can of course also introduce quantum logic abstractly and axiomatically based on the notion of a ‘yesno’ test. This is the approach of the socalled ‘Geneva school’ of quantum logic (see e. g. Jauch and Piron 1969). 5 There are further analogies between intuitionistic logic and quantum logic that could be brought to bear on the issue of the revision of logic. Both logics, for instance, allow for classical modal translations (see, respectively, G¨ odel 1933, and Dalla Chiara 1981).
Is Logic Empirical?
55
Hilbert space of the system. The classical negation of such a proposition is not an experimental proposition in this sense. Instead, the proposition stating that the system passes with probability 1 the test corresponding to the orthogonal complement of the given subspace is an experimental proposition. This, again, can be taken to deﬁne a strong negation. In the quantum case, however, we go further than in the case of Kolmogoroﬀ’s task logic. Indeed, even the classical disjunction of two experimental propositions p and q, corresponding to the settheoretic union of the two subspaces, is not itself a subspace in general, thus it is not an experimental proposition. Instead, the proposition corresponding to the (closed) span of the two subspaces P and Q (the smallest closed subspace containing both the subspaces P and Q) is an experimental proposition, and we can introduce a corresponding ‘quantum logical’ disjunction. This proposition corresponds to the most stringent test that will be passed with probability 1 if the tests corresponding to P and Q will. The classical conjunction of p and q, corresponding to the intersection of the two subspaces P and Q, is itself an experimental proposition, so in this sense there is no need to introduce a separate quantum logical conjunction. The closed subspaces of a Hilbert space are ordered by inclusion and form a lattice (i. e. suprema and inﬁma are pairwise always well deﬁned), which is further orthocomplemented under the orthogonal complement deﬁned via the scalar product. The quantum logical connectives correspond to the supremum, inﬁmum and orthocomplement in this lattice. As a consequence of the introduction of the quantum logical connectives, it is not the law of excluded middle that fails, but (one half of) the distributive law : the proposition p∧(q ∨r) is generally weaker than the proposition (p∧q)∨(p∧r). This can be trivially seen by taking the subspaces Q and R to be two rays spanning a plane, and P to be a ray lying in the same plane but noncollinear with either Q or R. In that case, p ∧ (q ∨ r) corresponds to the same subspace as p, but both p ∧ q and p ∧ r correspond to the zero subspace, and so does their quantum logical disjunction. Propositions that thus engender violations of distributivity are called incompatible; more precisely, two propositions p and q are called compatible iﬀ the sublattice generated by (the subspaces corresponding to) p, ¬p, q and ¬q is distributive.6
3.2
Formal properties of the logic
So far, what we have described is a semiinterpreted language (Van Fraassen, 1970). We have taken a propositional language, and we have ﬁxed a class of structures that are intended as models of the language, namely the class of lattices of subspaces of Hilbert spaces (henceforth: Hilbert lattices). A model in this sense will be a mapping of the propositions onto the subspaces of some Hilbert space, such that (syntactic) conjunctions shall be mapped to intersections, disjunctions to (closed) 6 There is more than one deﬁnition of compatibility in the literature, but this is immaterial for the purposes of this paper. Furthermore, they all coincide in the important case of orthomodular lattices.
56
Guido Bacciagaluppi
spans and negations to orthogonal complements of the corresponding subspaces. In order to say that we are introducing a logic in the formal sense (even a local one), we must have at least also a notion of logical consequence and of logical validity, and presumably other formal properties as well, such as soundness and completeness results for some appropriate logical calculus. With this in mind, let us return to the classical case. Also in the classical case, we could deﬁne a semiinterpreted language by deﬁning a model of the language in terms of subsets of some set, and mapping the logical connectives to the corresponding set unions, set intersections, and complements within the set (these are the lattice operations for the subset ordering relation). Every such lattice of subsets is a distributive lattice (also called a Boolean lattice or Boolean algebra), and conversely every distributive lattice is representable as the lattice of subsets of some set. One can turn this semiinterpreted language into a logic by deﬁning truth valuations as (orthocomplementedlattice) homomorphisms from an arbitrary Boolean algebra onto the twoelement algebra {0, 1}, and deﬁning the notion of logical validity by taking the class of all Boolean algebras as reference class. That is, a sentence in the language will be a logical truth, iﬀ it is true under every truth valuation of every model. The logic characterised by this notion of logical validity can be axiomatised, is sound and complete, and is of course the usual classical logic. In order to extend this treatment to quantum logic, we need to extend the notion of a truth valuation to nondistributive lattices. Homomorphisms of the entire lattice onto {0, 1} will not do, because in general there are no such total homomorphisms (Jauch and Piron 1963). More precisely, Jauch and Piron show that any socalled orthomodular lattice (in particular any Hilbert lattice) admits total homomorphisms onto {0, 1} iﬀ it is distributive.7 Note that this means that any form of quantum logic must give up bivalence. Thus, to insist that every proposition is indeed always true or false (as a matter of logic!) would be questionbegging, and, at least for the sake of argument, the failure of bivalence must not be taken as a reason for rejecting the whole framework out of hand. Instead, one can deﬁne workable truth valuations as partial homomorphisms onto {0, 1}, i. e. homomorphisms q deﬁned on some proper (orthocomplemented) sublattice Q of a given lattice L, provided one requires also that such a partial homomorphism be ﬁltered, i. e. for all a ∈ Q and b ∈ L, (6) a < b, q(a) = 1
⇒
b ∈ Q and q(b) = 1 ,
and maximal, i. e. have no proper extensions. The intuition behind these properties is that as many propositions as possible should be true or false under a truth valuation (maximality) and, in particular, a proposition that is weaker than a true proposition should also be true (ﬁltering). Both properties are of course trivial for total homomorphisms on Boolean lattices. Note also that any partial homomorphism has both a canonical ﬁltered extension and — by an application 7A
quick proof for the special case of Hilbert lattices is given in Bell (1987, pp. 5–6).
Is Logic Empirical?
57
of Zorn’s lemma — a maximal extension. A maximal partial homomorphism is always ﬁltered. A useful characterisation of truth valuations is the following. For any partial homomorphism q, let S denote the subset of all s ∈ Q such that q(s) = 1 (the set of all true propositions). The set S is a nonempty proper subset of Q, closed under conjunctions. Together with property (6), this means that it is a socalled ﬁlter ; and maximality of q means that S is a maximal ﬁlter, socalled ultraﬁlter. Truth valuations q are thus in bijective correspondence with ultraﬁlters S on the lattice. Note that S ⊥ is the set of all false propositions, and Q = S ∪ S ⊥ . Given the above deﬁnition of truth valuation, one can now proceed with quantum logic as with classical logic and deﬁne a notion of logical validity and logical consequence by ﬁxing a suitable reference class of nondistributive lattices. Quantum mechanics (if assumed to be strictly true) tells us that the world is one speciﬁc (only partially known) Hilbert lattice, but the corresponding notion of logic will need to be general enough to cover all possible Hilbert lattices.8 Admittedly, the choice of reference class is not as obvious as in the case of Boolean algebras, and there is some tradeoﬀ involved in the choice. One could choose the class of all Hilbert lattices, but it is unclear to date whether the resulting logic is axiomatisable. On the other hand, one can choose more general classes of lattices as reference class, for instance the class of all orthocomplemented lattices or the more restrictive class of all orthomodular lattices. These yield axiomatisable logics that are both sound and complete (see e. g. Dalla Chiara and Giuntini 2002, section 6). Note that the logic of all Hilbert lattices, the logic of all orthomodular lattices and the logic of all orthocomplemented lattices are indeed all distinct, i. e. they have diﬀerent sets of logical truths. Choosing the logic of all Hilbert lattices would more properly characterise the ‘logic of quantum mechanics’. On the other hand, even if one takes a reference class more general than that of all Hilbert lattices, one can still argue that quantum phenomena have prompted the adoption (at least locally) of a nonclassical logic. (Also, as mentioned in section 6, quantum theories of systems with inﬁnitely many degrees of freedom seem to require a larger reference class.) The choice of orthomodular lattices seems particularly attractive, since in an orthomodular lattice there is a unique conditional reducing to the standard conditional for compatible propositions; the resulting connective has some unusual features, but these can be explained in analogy to counterfactual connectives, as is reasonably intuitive in a logic that gives up bivalence (Hardegree 1975). In any case, the resulting logic is strictly weaker than classical logic, since the reference class that deﬁnes logical validity is extended beyond the class of Boolean algebras. Irrespectively of the details of the choice, we shall take it that such a notion of quantum logic provides us with a basis for discussing Putnam’s claims, the interest of which after all lies primarily in the idea that empirical considerations 8 Similarly, general relativity (if assumed to be strictly true) tells us that the world is one speciﬁc (only partially known) Lorentzian manifold, but the corresponding notion of geometry will cover all possible Lorentzian or pseudoRiemannian manifolds.
58
Guido Bacciagaluppi
might force us to give up classical logic, and not (or only in the second place) in the details of which logic should replace it.
3.3 Alternative frameworks As an aside, let us remark that we have presented above merely one possible framework for introducing a quantum logic, and that others have been proposed. We should mention two in particular. First, one could choose a diﬀerent idealisation for quantum mechanical experiments, in order to include more realistic measurements (described technically by positiveoperatorvalued measures rather than projectionvalued measures). This leads one to consider, instead of the lattice of projections (equivalent to the lattice of subspaces), the poset (partially ordered set) of positive operators. This in turn prompts the introduction of fuzzy quantum logics and other quantum logics that generalise the latticetheoretic approach (see e. g. Dalla Chiara and Giuntini 2002, sections 11–16). More general posettheoretical structures arise also as the logics associated with theories of quantum probability, as in the test space approach of Foulis and Randall (1981). Second, one can focus on a diﬀerent general aspect of quantum mechanical experiments, namely their incompatibility; and instead of introducing apparently new logical connectives, one can restrict the use of the usual connectives to pairs of compatible propositions. This is the partial Boolean algebra approach to quantum logic (Kochen and Specker 1965a, b, 1967), which also gives rise to logical systems with nice formal properties. The partial Boolean algebra approach and the posettheoretical approach overlap, unsurprisingly, in that socalled transitive partial Boolean algebras are canonically equivalent to socalled coherent orthomodular posets (Finch 1969, Gudder 1972), so that the corresponding logics are the same. Note that the partial Boolean algebra approach may present advantages to the advocate of a global revision of logic, because the implied revision of logic appears to be more modest (although in a sense equivalent), and because it is easier to argue that the meaning of the logical connectives has remained the same. One does not construct new connectives that must somehow turn out to be the usual ones in disguise. One merely needs to argue that our usual connectives can be applied only to propositions that are compatible, and that it is an empirical matter, settled in the negative by quantum mechanics, whether all propositions are indeed so. We shall not attempt to develop here this line of argument, merely note that Putnam himself switched to using at least the formalism of partial Boolean algebras in some later publications (notably Friedman and Putnam 1978). We shall keep to talking of quantum logic in the latticetheoretic approach, because most of the discussion about Putnam’s suggested revision of logic has been in the context of this approach and of the corresponding failure of distributivity.
Is Logic Empirical?
59
4 STANDARD INTERPRETATION AND MEASUREMENT PROBLEM OF QUANTUM MECHANICS It is certainly an empirical fact that, if one deﬁnes experimental propositions as in the previous section, the resulting lattice fails to be distributive, and a fact that is characteristic of quantum mechanics. If all physics were classical, then the lattice of experimental propositions deﬁned in this way would be distributive. It may also be reasonable to want to deﬁne a local nondistributive logic for dealing with such experimental propositions. However, it is not clear at this stage why this logic should be even a candidate for a revised global logic. If one takes a ‘naive’ instrumentalist position, then quantum mechanics just provides us with the means of calculating the probabilities for the results of our experiments. The resulting procedure is certainly diﬀerent from that in any classical framework, but there seems to be little need to revise anything but our algorithmic procedures for predicting experimental results. If one adopts a subtler Bohrian position, then the language of classical physics becomes a prerequisite for the description of quantum experiments, so that the very formulation of quantum mechanics would seem to require classical logic. Clearly, more than empirical considerations are needed in order to mount a case for the revision of logic at the global level. In particular, a strong opposition to the instrumentalist or Bohrian position is necessary in order to reject the overall package that includes classical logic and an instrumentalist or Bohrian reading of quantum mechanics. In this section we shall sketch the ‘naive’ realist interpretation of quantum mechanics. This interpretation, variously referred to as ‘standard’ or ‘orthodox’ or ‘von NeumannDirac’, is problematic, because it gives rise to the usual paradoxes, but it is usually taken as the starting point for further discussion and elaboration of other subtler approaches to quantum mechanics. It is thus, so to speak, the default realist position in the foundations of quantum mechanics. And in fact, it is the interpretation of choice (at least implicitly) also for discussions of Putnam’s claims on the revision of logic. (Other realist approaches, and their implications for Putnam’s claims, will be discussed in section 7.) The standard interpretation consists in the following assignment of (intrinsic) properties to quantum systems. A quantum system has a certain property iﬀ it passes with probability 1 a corresponding experimental test (in the sense of the previous section). Properties assigned in this way are thus in bijective correspondence to the closed subspaces of Hilbert space. What can it mean to assign such properties to a physical system? The case of onedimensional subspaces is relatively straightforward: a onedimensional subspace (ray) is the set of all scalar multiples of a given vector, and these all describe the same quantum state. So, saying that a quantum system has a certain onedimensional property corresponds to saying that its state is a certain vector in the Hilbert space. The case of multidimensional properties is more diﬃcult, but it is also quite crucial. In this case, one should think of entangled systems, where there is a
60
Guido Bacciagaluppi
vector describing the composite system but no vector describing each subsystem separately. The composite system will thus be assigned a onedimensional property, but not the subsystems. Nevertheless, if two systems are entangled there are always multidimensional tests (in general nontrivial) on the individual subsystems, for which the subsystems will test positively with probability 1. Therefore, according to the standard interpretation, the subsystems are assigned the corresponding multidimensional properties. Unless one accepts some form of holism, in which only the composite system is assigned intrinsic properties, one is forced to generalise the notion of properties for the individual subsystems to include also multidimensional ones. The motivation behind the standard interpretation can be phrased in the language of dispositions. Quantum mechanical systems exhibit a range of dispositional properties in the context of experimental tests, some of which are sureﬁre dispositions. The standard interpretation suggests that these sureﬁre dispositions (whether onedimensional or multidimensional) support an inference to real, objective properties of the quantum system, which are revealed by idealised tests. But now, enter the paradoxes, speciﬁcally the measurement problem and Schr¨ odinger’s cat (which we shall take here as two examples of the same problem). If we take quantum mechanics to be correct and universally valid, then one can easily construct examples in which the dynamics of the theory, the Schr¨ odinger equation, will lead to entanglement between microscopic and macroscopic systems, e. g. between a quantum system being measured and the corresponding measuring apparatus, or between a microscopic system and a cat, in such a way that macroscopically distinguishable states of the apparatus (diﬀerent readings) or the cat (alive or dead) are correlated with diﬀerent states of the microscopic system. In such cases, on the standard interpretation, only the multidimensional subspaces spanned by the macroscopically distinguishable states correspond to properties of the macroscopic system, and these do not correspond to the macroscopic states we witness (the diﬀerent readings, the cat alive or dead).9 One could say, paradoxically, that the cat is neither alive nor dead, but this formulation trades on an ambiguity: this statement would be paradoxical if ‘dead’ were understood as ‘not alive’ in the classical sense, but if it is understood as ‘not alive’ in the sense of the strong negation of quantum logic (assuming that the live and dead states span the Hilbert space of the cat), then the statement makes perfect sense, since in this case ‘dead’ is strictly stronger than the classical ‘not alive’. 9 Furthermore, this problem cannot be lifted by modelling the states of the apparatus as statistical distributions over microscopic states. If the dynamics is the unitary Schr¨ odinger dynamics, one cannot reproduce the correct measurement statistics for all initial states, unless the state of the apparatus depends on the state of the system to be measured. This result was known already to von Neumann — indeed it prefaces his discussion of measurement in quantum mechanics (von Neumann 1932, section VI.3). For a modern, more general discussion, see e. g. Brown (1986) and references therein.
Is Logic Empirical?
5
5.1
61
QUANTUM LOGIC AND THE STANDARD INTERPRETATION OF QUANTUM MECHANICS
Quantum logic as a logic of properties
The obvious interest of the standard interpretation, from the point of view of quantum logic, is that it allows one to apply the quantum logical structures introduced for experimental propositions also to propositions about intrinsic properties of a quantum system. Thus one speaks of the property lattice of the system, or of the lattice of ‘testable’ propositions about the system. This move from experimental propositions to properties of a system is explicitly made for instance by Jauch and Piron (1969), who further propose that a quantum state should be understood as the set (in fact the ultraﬁlter) of true properties about the system. They thus propose, in eﬀect, that a quantum state should be understood as a truth valuation on the lattice of properties of a quantum system. And, indeed, quantum states in the sense of rays in the Hilbert space are in bijective correspondence to the ultraﬁlters of true propositions they generate (by assigning them probability 1 upon measurement). Thus, truth valuations on the Hilbert lattice of quantum propositions encode all the information about quantum mechanical expectation values. Abstractly, the introduction of quantum logic for testable propositions is possible simply because there is a bijective correspondence between the experimental propositions and the testable propositions (both being in bijective correspondence with the closed subspaces of the Hilbert space). A closed subspace of the Hilbert space will now represent a proposition about an intrinsic property of the relevant quantum system, and the closed span, intersection and orthogonal complement of such subspaces will correspond to the quantum logical disjunction, conjunction and negation of the respective propositions. Concretely, the standard interpretation introduces properties corresponding to onedimensional subspaces P , Q etc., and properties corresponding to multidimensional subspaces such as the span of P and Q. The novelty of these properties lies in the fact that under all possible truth valuations, whenever P obtains or whenever Q obtains, also the property corresponding to their span obtains. The interpretation of this property as a quantum logical disjunction p ∨ q allows one to interpret such relations between propositions as relations of logical consequence. By considering quantum logic at the level of intrinsic properties of physical objects, we make a further step in the direction of Putnam’s proposals. Indeed, at least as regards the more modest claim (a), the fact that this logic in general is a nondistributive lattice is clearly an empirical fact. The fact that it is best understood as a semiinterpreted language, and the fact that this language has a number of properties that justify calling it a logic in the formal sense, have been discussed above. In this sense thus, it should be relatively uncontroversial that quantum phenomena give us empirical grounds for introducing a logic adapted to the world of physics that is nondistributive and hence nonclassical. I take it
62
Guido Bacciagaluppi
that what we have called above Putnam’s claim (a) is thus both intelligible and justiﬁable. The claim that is controversial is claim (b), that this gives us further reasons to revise logic tout court, i. e. that this logic of testable quantum mechanical propositions, or logic of quantum mechanics, is in fact the ‘true’ logic and that we have failed to recognise so far that our usual, apparently classical connectives are in fact the connectives of quantum logic. We turn at last to this controversial point, for the time being in the context of the standard interpretation of quantum mechanics.
5.2 The revision of logic If we have successfully introduced the quantum logical connectives in the context of propositions about material properties of physical objects, is this not ipso facto saying that ‘the logic’ of the world is quantum? Surely quantum mechanics is a theory that applies to all material objects, so that the resulting quantum logic is not a local but a global logic? As mentioned in the introduction, the fact that one may justify the introduction of nonclassical connectives does not yet mean that logic has been revised. The crucial point is whether these connectives have been introduced alongside the classical connectives, or whether they replace them (in an appropriate sense). As we shall see now, the standard interpretation is neutral with regard to this question. Indeed, one has a choice between two opposing views. On the one hand, it is perfectly possible to interpret the properties assigned to systems in the standard interpretation as elementary properties in the sense of classical logic. Indeed, a quantum logical disjunction p ∨ q classically must be an elementary proposition: it is not a classical disjunction of terms that include p and q, although one might be tempted to think that it is the disjunction of all onedimensional subspaces contained in the span of P and Q. As a matter of fact, this is not true: if it were, in the case of entangled systems there would be a quantum state that describes an individual subsystem, contrary to what quantum mechanics says. On top of the elementary propositions, however, one can perfectly well consider complex ones, constructing them by applying the classical connectives to this new quantum set of elementary propositions, e. g. the disjunction of all onedimensional subspaces contained in the span of P and Q can be considered alongside with the quantum disjunction p ∨ q itself. The quantum aspect is physical and lies in the determination of the elementary properties, while the logic remains classical. This position has admittedly some disadvantages. Complex propositions in general are not directly testable, i. e. veriﬁable with probability 1. This is simply because neither the set union of P and Q nor the set complement of P are subspaces of the Hilbert space. This would thus be an empirical limitation characteristic of quantum mechanics. More importantly, perhaps, the above relation between p or q and the quantum logical p ∨ q cannot be analysed as logical consequence. That is, there are triples of propositions p, q, r such that whenever p or q hold, also r holds; but since r is elementary, this relation of consequence cannot be analysed
Is Logic Empirical?
63
as logical consequence. This, however, may seem a small price to pay in order to refrain from revising our logic. And in fact, we shall see in section 7 that this is arguably the position most naturally associated with the approaches to quantum mechanics known as spontaneous collapse theories. The opposite position consists in maintaining that the properties assigned in the standard interpretation are all the possible properties of a physical system: there is no property corresponding to the classical disjunction of p and q, or to the classical negation of p. An equivalent way of saying this is that if p and q are the propositions that some physical quantities take certain values, say ‘A = 4’ and ‘B = 9’, then there is no meaningful physical quantity that can encode the classical disjunction ‘A = 4 or B = 9’. Note that there is no quantum mechanical observable that encodes it. Indeed, in disanalogy to classical physics, the operator (A − 4)(B − 9) in general does not represent a quantum mechanical quantity, because in general the operators A and B do not commute (so that (A − 4)(B − 9) is not selfadjoint). But if there is no meaningful physical quantity whatsoever that represents a classical disjunction, insisting that the properties of the standard interpretation are elementary would mean that the vast majority of complex propositions constructed from elementary propositions about quantum systems are meaningless. If one drops altogether the possibility of using the classical connectives to form complex propositions, one can instead interpret some of the testable propositions as complex propositions in the sense of quantum logic. In so doing one removes the mismatch between logical propositions and physical propositions (indeed, all propositions are testable propositions), and one ensures that the consequence relation described above between testable propositions and their quantum disjunctions is indeed a relation of logical consequence. This is presumably the best case that can be made for a revision of logic in the context of the standard interpretation of quantum mechanics. It is not made explicitly by Putnam, although some of it must be implicit in his discussion; it is present more or less in Dickson (2001), who explicitly denies the ‘empirical significance’ of classical disjunctions and negations. Still, it appears that if one follows this line of argument, the quantum logical connectives have supplanted the meaningless classical ones. What about the claim that the quantum logical connectives are the same as the classical connectives? Indeed, since every physical system is a quantum system, we seem to have arrived at the conclusion that an ‘everyday’ disjunctive proposition about any physical object whatsoever is meaningless. But Putnam’s claim that the classical and quantum connectives are the same is surely meant in the sense that we should be able to gain a better understanding of our usual everyday classical connectives by realising that they are indeed quantum logical. What is missing from the above is an explanation of why classical logic appears to have been so eﬀective until now. One needs to explain how, if the true logic is nondistributive, it is still possible for the connectives to behave truthfunctionally
64
Guido Bacciagaluppi
in special cases. This would give rise to the possibility of abstracting classical logic (empirically!) from our everyday use, and of applying it in the appropriate circumstances (as Putnam undoubtedly did in the act of writing his famous paper). Putnam’s claim (c), that adopting quantum logic will solve the paradoxes of quantum mechanics, can be understood as an attempt to ﬁll this gap: the quantum logical point of view does indeed explain, according to Putnam, why the world appears classical to us. Indeed, for Putnam the main advantage of a revision of logic is precisely that it will solve the paradoxes of quantum mechanics. We shall brieﬂy discuss now how Putnam argues for this point and why his arguments are justly regarded as ﬂawed. One other author at least, namely Dickson, attempts to argue that, although the quantum connectives are the true connectives, they behave classically when applied to the everyday, macroscopic realm. As we shall also see, his attempt appears to fail on ultimately similar grounds. If this is so, then we are left with the following situation. There is a coherent, perhaps even a reasonably convincing case to be made that a nonclassical logic is well adapted to a world in which quantum mechanics under the standard interpretation is true. But this world is hugely diﬀerent from our own. This is precisely what the measurement problem and Schr¨ odinger’s cat highlight. Indeed, in such a world it would not seem possible for any intelligent beings to develop at all, let alone beings capable of formulating any kind of logic (let alone quantum mechanics). If the argument does not apply to our world (or at least to a possible world similar to ours), then it loses most of its interest.
5.3 Putnam and the paradoxes The seemingly logical paradox of Schr¨ odinger’s cat, that the cat is neither alive nor dead, trades as we have mentioned on the ambiguity between classical and quantum logical terms. Putnam’s way of resolving the paradox is to choose a strictly quantum logical reading: ‘dead’ is interpreted as ‘not alive’ in the quantum logical sense of orthocomplementation in the lattice, and the cat is then indeed alive or dead, but in the sense of the quantum logical disjunction. Putnam, however, seems to want to go further, namely he claims that, since the cat is alive or dead (quantum logically), there is a matter of fact about the biological state of the cat. To make the point clearer, let us take an example adapted from Putnam himself (1968, pp. 184–185). Consider an ndimensional Hilbert space and take an orthonormal basis x1 , . . . , xn in the Hilbert space, which one can associate with a family of tests, or equivalently with some observable X. Denoting the propositions corresponding to the onedimensional projectors onto the basis vectors as x1 , . . . , xn , the following is a true proposition under all truth valuations: (7) x1 ∨ . . . ∨ xn . Its truth, however, is understood by Putnam as meaning that the observable X has indeed a value corresponding to one of the xi . As the reasoning is independent of the particular choice of basis, Putnam concludes that the system possesses values
Is Logic Empirical?
65
for all such observables. He then interprets measurements as simply revealing those preexisting values, thus proposing that the measurement problem of quantum mechanics is solved by a move to quantum logic. This is rather bewildering, since, as we have seen in section 3, quantum logic comes equipped with a welldeﬁned semantics, which underlies the quantum logical notion of consequence. And we have seen that truth valuations in this semantics are such that the proposition x1 ∨ . . . ∨ xn can be true without any of the xi being true. Any quantum state that is a nontrivial linear combination of the basis vectors will deﬁne such a truth valuation; and in the case of entangled systems, we have seen that a quantum logical proposition can be true without any of the onedimensional projections spanning it being true. To be fair, at the time of Putnam’s 1968 paper, the semantics of quantum logic was not fully developed as yet, but the reasoning implied in the paper seems to be technically in error, since he appears to be using a diﬀerent semantics from that required in quantum logic. A more charitable reading (perhaps more in line with his later papers, e. g. Putnam 1981), takes Putnam as distinguishing between a quantum level, obeying quantum logic, and a ‘hidden’ level obeying classical logic. It has in particular been suggested that Putnam’s proposals can be analysed in terms of a socalled noncontextual hidden variables theory (Friedman and Glymour 1972), which however confronts them with the standard problems facing such approaches, notably the nogo theorem by Kochen and Specker (1967). Perhaps more plausibly, it has also been suggested to analyse Putnam’s proposals in terms of a socalled contextual hidden variables theory (Bacciagaluppi 1993), which however confronts them with the proofs of nonlocality for this kind of approaches, speciﬁcally those by Heywood and Redhead (1983) and by Stairs (1983). In either case, however, Putnam would seem to be backing away from the proposal that quantum logic is the global logic. (For Putnam’s most recent views on the subject, see Putnam 1994.) In section 7, we shall return to the issue of quantum logic in hidden variables approaches, namely in the context of the most successful of these, pilotwave or de BroglieBohm theory. Dickson’s (2001) attempt to explain how classical logic is eﬀective despite quantum logic supposedly being the global logic, proceeds along slightly diﬀerent lines. Dickson points out that in the macroscopic realm, when talking about measurement results or cats, we apply logic always to a distributive sublattice of all (quantum logical) propositions. As it stands, however, this argument is inconclusive. The sublattice generated by the propositions x1 , . . . , xn in Putnam’s example is distributive, but this fact does not guarantee that the logical connectives will behave truthfunctionally, and that is what is at stake. Again to be fair, Dickson suggests that the proper framework for discussing Putnam’s claims is that of the more general quantumlike theories based on the formalism of von Neumann algebras. And we shall see in the next section that in that framework the connectives can indeed be shown to behave truthfunctionally in certain cases. As far as the claim concerns the usual formalism of quantum mechanics, however, it may be that Dickson falls prey to a common fallacy. Admittedly, it is
66
Guido Bacciagaluppi
a fact that we cannot construct in practice an experiment that would test for a macroscopically entangled state (in particular because of the phenomenon called decoherence), and that at the macroscopic level the only tests we have available are all compatible (so that the corresponding experimental propositions form a distributive lattice). And this fact has often been trumpeted as showing that the measurement problem does not arise. But this practical impossibility is irrelevant to the point that macroscopically entangled states (under the standard interpretation) are incompatible with macroscopic objects having the properties they appear to have, nor does it show that such states do not arise in practice. It thus seems that Dickson’s argument fails to improve on Putnam’s attempt. 6 QUANTUM LOGIC AND CLASSICAL PROPOSITIONS Before proceeding further and enquiring into the status of quantum logic in realist approaches to quantum mechanics other than the standard interpretation, let us dwell in more detail on the question of what it could mean for the quantum logical connectives to be the same as the classical connectives. There is an interesting way of making the case that the meaning of the connectives is indeed the same in classical and quantum logic, namely to argue that it is always given in terms of the supremum, inﬁmum and orthocomplement of the lattice: the conjunction of two propositions is the weakest proposition that implies both propositions, their disjunction is the strongest proposition that is implied by either, and the negation of a proposition is its orthocomplement in the lattice. Until empirical evidence for quantum mechanics was obtained, we used to believe that all lattices of propositions we could ever consider would be distributive. We used to believe that the universe of sets was the correct framework for abstract semantics, because we believed it was rich enough to describe the physical world. But, so the argument goes, it has turned out that it is only the ‘universe of Hilbert spaces’ that is rich enough for that purpose. (This line of thought presupposes of course that one has already accepted that the logic should be read oﬀ the structure of the lattice of empirical propositions.) The trouble with this suggestion is that, although at this more general level the quantum and classical connectives can thus be said to be the same, still, if the actual lattice of properties is a Hilbert lattice (of dimension greater than 1), the connectives will just not behave truthfunctionally, so that the quantum connectives do not seem to reduce to the classical ones in everyday macroscopic situations. This is precisely the problem facing Dickson: can one have the (unique) logical connectives behave truthfunctionally when applied to some propositions in the lattice but not to others? We shall see in the present section that, if, as is quite standardly done, one deﬁnes logical consequence through a reference class of lattices that is larger than the class of all Hilbert lattices (which, as noted above, is not known to lead to an axiomatisable logic), in particular if one considers quantum logic to be the logic of all orthocomplemented lattices or of all orthomodular lattices, then there is a rigorous sense in which the connectives
Is Logic Empirical?
67
interpreted in these nondistributive lattices (i. e. the standard quantum logical connectives) can behave truthfunctionally in certain cases. Thus, at least in this more abstract setting, there are situations in which one could arguably ‘mistake’ the logic to be classical. Recall that two propositions p and q are compatible iﬀ the lattice generated by p, ¬p, q and ¬q is distributive. For a subset A of an orthocomplemented lattice L, denote by Ac the set of propositions compatible with all propositions in A. If one considers lattices L more general than Hilbert lattices, the set Lc (the socalled centre of the lattice) may be nontrivial, i. e. there may exist propositions (other than the trivially true and false propositions) that are compatible with all propositions in the lattice. Such propositions are called classical propositions. Now, it is a theorem that under any truth valuation on L, a classical proposition is always true or false. Indeed, let q be a truth valuation from L onto {0, 1}, deﬁned on an orthosublattice Q = S ∪ S ⊥ of L, where S is the ultraﬁlter of propositions made true by q. For any ultraﬁlter S in L, (8) (S ∪ S ⊥ )c ⊂ (S ∪ S ⊥ ) (Raggio 1981, Appendix 5, Proposition 3). A classical proposition, being compatible with any a ∈ L, is obviously contained in (S ∪ S ⊥ )c for any set S. Therefore, for any truth valuation q, c ∈ Q, i. e. q(c) = 1 or q(c) = 0. QED. It now follows, just as in the classical case, that if a lattice contains classical propositions, the latticetheoretical connectives applied to the classical propositions will behave truthfunctionally, in particular for any two classical propositions a and b, and any truth valuation q that makes a ∨ b true, q will make a true or b true. Indeed, let q be any truth valuation with q(a ∨ b) = 1. Since a and b are classical, by the above they are both either true or false under q. But if q(a) = 0 and q(b) = 0, then, since q is homomorphic, q(a ∨ b) = 0, contrary to assumption. Therefore, if a and b are classical, (9) q(a ∨ b) = 1
⇒
q(a) = 1 or q(b) = 1 ,
for any truth valuation q. QED. Note that the fact that a certain proposition a is classical depends on the lattice L chosen as a model of the logic. Speciﬁcally, it depends on the relation of a with all the other propositions in the chosen model. It thus depends on the meaning of a. We see that the quantum logical connectives can indeed behave truthfunctionally in certain models, but depending on the meaning of the propositions involved. Classical logic appears to be valid in special cases, but the additional inferences one can make in these cases are not logical inferences: they are not based on the propositional form of the statements involved, they are based instead on the fact that the statements have a classical content. If the lattice of properties in our world is the lattice of projections of some Hilbert space, our world does not contain classical propositions (pace Putnam
68
Guido Bacciagaluppi
and Dickson). On the other hand, at least some lattices that are more general than Hilbert lattices appear to be physically motivated. Indeed, generalisations of quantum mechanics that allow in general for classical propositions exist, and are required to treat systems with inﬁnitely many degrees of freedom, such as in quantum ﬁeld theory or in quantum statistical mechanics (when taking thermodynamic limits). Mathematically, these theories are based on more abstract algebras of observables than the algebra of (selfadjoint) operators on a Hilbert space. For the purposes of quantum logic, the most interesting class of such algebras is that of socalled von Neumann (or W ∗ ) algebras, which can be represented as certain subalgebras of operators on Hilbert space. Von Neumann algebras can be generated by their projections, so that one can again reduce all statements about observables to statements about projections (i. e. to yesno tests).10 The lattices of projections of von Neumann algebras are always orthomodular lattices. (Indeed, historically, the study of orthomodular lattices developed out of the study of von Neumann algebras.) Therefore, unless one insists on characterising quantum logic by the class of all Hilbert lattices, lattices of projections of von Neumann algebras are already included in the models of the most usual varieties of quantum logic, and they are thus a bona ﬁde source of examples for the behaviour of the usual quantum logical connectives. Incidentally, we note that J. von Neumann is also associated both with the standard interpretation of quantum mechanics (rightly or wrongly), through his book Mathematische Grundlagen der Quantenmechanik (von Neumann 1932), and with the ﬁrst proposal that quantum mechanics should be interpreted in terms of a nondistributive logic, clearly stated in his paper with Birkhoﬀ four years later (Birkhoﬀ and von Neumann 1936). When we say that general lattices of projections of von Neumann algebras include classical propositions, the intuition behind it is that there is a breakdown in the linear structure of the state space of a physical system. Indeed, deﬁning a classical observable as an observable C such that propositions of the form ‘C has value α’ are classical, superpositions of states in which a classical observable has diﬀerent values simply do not exist (one says that such states are separated by a superselection rule). The framework of von Neumann algebras is general enough to include both quantum and classical physics, and intermediate theories besides. For instance, one can build algebras that are tensor products of a standard quantum system and a purely classical system, and for which there are no states entangling the quantum system and the classical system (Raggio 1988, see also Baez 1987). Although it is generally believed that such theories would be rather ad hoc, they do allow one to describe a world in which the measurement problem of quantum mechanics does not arise, a world in which all measuring apparatuses (as well as cats) are made 10 Indeed, Raggio (1981) has proved that if L is the projection lattice of a W ∗ algebra M, there is a bijective correspondence between truth valuations on L and pure normal states on M, in the sense of normalised positive linear functionals. That is, truth valuations indeed encode all the information about expectation values of observables in the algebra.
Is Logic Empirical?
69
out of classical observables. It is instructive to see explicitly how the truthfunctionality of the quantum logical connectives would apply to a measurement scenario if the ‘pointer’ observable of a measuring apparatus were assumed to be a classical observable C. Suppose the apparatus measures a nonclassical observable B taking, say, the two values ±1. Now assume that at the end of an (ideal) measurement the following proposition is true (in obvious notation): (10) (B = 1 ∧ C = 1) ∨ (B = −1 ∧ C = −1) , where ∧ and ∨ denote the inﬁmum and supremum in the lattice. We can now show from the fact that C is classical that the disjunction in (10) is truthfunctional, i. e. under any truth valuation q that makes (10) true, (B = 1 ∧ C = 1) or (B = −1 ∧ C = −1) are also true. Let q be such a truth valuation, i. e. (11) q (B = 1 ∧ C = 1) ∨ (B = −1 ∧ C = −1) = 1 . We need to show that (12) q(B = 1 ∧ C = 1) = 1
or q(B = −1 ∧ C = −1) = 1 .
Because q is ﬁltered, we have (13) q(C = 1 ∨ C = −1) = 1 . Since C is classical, (14) q(C = 1) = 1 or q(C = −1) = 1 , by (9). Suppose for instance that q(C = 1) = 1. Since q is a homomorphism, we have that (15) q (B = 1 ∧ C = 1) ∨ (B = −1 ∧ C = −1) ∧ C = 1 = 1 . But now, the propositions B = ±1, C = ±1 are all mutually compatible, so that we can distribute over ∨ in (15), yielding (16) q(B = 1 ∧ C = 1) = 1 . Analogously, if q(C = −1) = 1 we obtain q(B = −1 ∧ C = −1) = 1. QED. Note in particular that the truthfunctionality has spread to propositions that include nonclassical terms. (This appears to be related to the results by Bub and Clifton (1996) on maximal truthvalue assignments in a Hilbert lattice compatible with a certain ‘preferred’ observable being assigned deﬁnite values.) If such examples do not describe the actual physics, however, what have we gained in showing that the connectives can sometimes behave truthfunctionally? We should perhaps distinguish two questions: (i) Can we envisage worlds, perhaps
70
Guido Bacciagaluppi
merely inspired by quantum mechanics and suﬃciently close to our own, in which we would consider revising our logic? (ii) Is our world such a world? In a world as the above, one could indeed maintain that the only meaningful propositions are the propositions in the lattice, since the lattice is general enough to include propositions for which classical logic holds, and a generalised quantum mechanics together with the standard interpretation could arguably meet the objections detailed in the previous section against a revision of logic. We can thus make a case that logic is empirical because there is a possible world in which we might be prompted by empirical considerations to revise our logic (question (i)). This is diﬀerent from establishing that in our world we may have good reasons for a revision of logic (question (ii)). Note that while Putnam’s ultimate aim was to show that logic is indeed empirical, his actual claim was that we have reasons to revise our logic in this world. In order to proceed further with question (ii), and thus address the revision of logic in Putnam’s own terms, we shall have to return to standard quantum mechanics. However, we shall have to consider approaches to the foundations of quantum mechanics other than the standard interpretation, in particular approaches that have some credible claim to providing solutions to the standard puzzles. 7
QUANTUM LOGIC IN OTHER APPROACHES TO QUANTUM MECHANICS
We now leave the standard interpretation of quantum mechanics. While reverting to the standard formalism of quantum mechanics in this section, we shall discuss the status of the claims about quantum logic in the context of other approaches to the foundations of quantum mechanics, approaches that do propose solutions to the puzzles presented to us by quantum mechanics and in particular propose to explain why classical logic is eﬀective (whether or not it be the true logic) in a world in which quantum mechanics is indeed true. The approaches we shall discuss in turn are (i) the pilotwave theory of de Broglie and Bohm, (ii) spontaneous collapse theories, and (iii) the Everett or manyworlds interpretation. (The presentation of these approaches will necessarily be rather condensed.)
7.1 de BroglieBohm theory The pilotwave theory of de Broglie and Bohm is a very wellknown and wellunderstood approach to the foundations of quantum mechanics. The theory, as presented by Louis de Broglie at the ﬁfth Solvay conference in October 1927 (de Broglie 1928), is a new dynamics for nparticle systems, described in conﬁguration space (which encodes only the positions of the particles) rather than in phase space. The motion of the particles is determined by a ﬁeld of velocities deﬁned by the phase S of the complex wave function. At least as regards particle detections, the theory can clearly predict both interference and diﬀraction phenomena: around the zeros of the wave function, the phase S will behave very
Is Logic Empirical?
71
irregularly, so one can at least qualitatively expect that the particles will be driven away from regions of conﬁguration space where the wave function is small (as is indeed the case). In fact, it was the qualitative prediction of electron diﬀraction and its experimental detection that established the signiﬁcance of de Broglie’s matter waves even before his detailed theory of 1927. The measurement theory for observables other than functions of position was worked out in general by David Bohm, who rediscovered the theory a quarter of a century later (Bohm 1952). Indeed, it is at ﬁrst puzzling how such a theory of particles in motion may eﬀectively reproduce the collapse process and the rest of the full phenomenology of quantum mechanics. In modern terminology, what Bohm showed in general is that in situations such as measurements, the wave function of the total system decoheres, that is diﬀerent components of the wave function eﬀectively cease to interfere, because they are in fact separated in conﬁguration space by regions with very small wave function. This has as a consequence that the conﬁguration of the system is eﬀectively trapped inside one of the components. This component alone will be relevant at later times for the dynamics of the system, so that the particles behave as if the wave function had collapsed. Quantitatively, the statistical predictions of quantum mechanics are reproduced if the positions of the particles are distributed according to the usual quantum distribution. As was known to de Broglie, the velocity ﬁeld preserves the form of the particle distribution if at any time it is given by the quantum distribution. Intuitively, this is some kind of timedependent equilibrium distribution, and there is indeed a deep analogy between the statistical aspects of de BroglieBohm theory and classical statistical mechanics. Under the assumption of nonequilibrium distributions, the theory instead yields novel predictions as compared to quantum mechanics. Furthermore, pilotwave theory is explicitly nonlocal, as any hidden variables theory must be in order to recover the quantum mechanical violations of the Bell inequalities. Finally, the theory can be easily modiﬁed to include spin; and various generalisations aiming to cover quantum electrodynamics and other ﬁeld theories have been proposed. Incidentally, J. S. Bell contributed decisively to the theory’s current revival (Bell 1987, passim). As regards quantum logic, it is obvious that since de BroglieBohm theory reproduces the phenomenology of quantum mechanics, physical systems have the same dispositions to elicit measurement results in pilotwave theory as in standard quantum mechanics. The introduction of the connectives at the level of the experimental propositions therefore goes through unaltered. At the level of the intrinsic properties of a system, however, it should be clear that conﬁgurationspace properties obey classical logic no less than phasespace properties in classical physics. Indeed, de BroglieBohm theory can be viewed as a theory that is entirely classical at the level of kinematics (particles moving in space and time), and which is quantum only as regards its ‘new dynamics’ (as in the title of de Broglie’s paper). Thus, the way de BroglieBohm theory explains the eﬀectivenesss of classical logic at the macroscopic level is that it is already the logic that is operative at the hidden (‘untestable’) level of the particles.
72
Guido Bacciagaluppi
Indeed, the emergence of the classical world around us in de BroglieBohm theory happens as follows. At the level of the wave function, a process of decoherence ensures that macroscopically diﬀerent components will develop that will not generally reinterfere, e. g. it ensures that the ‘live’ and ‘dead’ components of the state of the cat do not reinterfere. What turns these diﬀerent components into diﬀerent classical alternatives, however, is the fact that the conﬁguration of the system is located only in one of these diﬀerent components, and this is already a matter of classical logic. The cat is (classically) either alive or dead, because the particles that compose it are (classically) either in the live component or the dead component of the quantum state. (Decoherence further ensures that they will stay there over time, but this is irrelevant to the point at hand.) Thus, if one takes the pilotwave approach to quantum mechanics, although quantum logic may be introduced as a local logic at the level of experimental propositions, it cannot be taken as the basis for justifying the everyday use of classical logic, and thus cannot aspire to replace classical logic as the ‘true’ logic.11 Incidentally, de BroglieBohm theory lends itself to discussing issues of conventionalism by analogy to the case of geometry (i. e. the choice between revising physics or geometry), as mentioned in section 1. In the case of general relativity, one can take the metric of spacetime to be Einstein’s gμν and the geometry to be curved, or one can take the metric to be the ﬂat Lorentzian ημν and write (17) gμν = ημν + hμν , where hμν is a new physical ﬁeld, and the geometry is still the ﬂat geometry of special relativity. This treatment (which lies at the basis of some attempts to quantise gravity) is observationally equivalent to standard general relativity only given some constraints on the topology of the phenomenological spacetime (i. e. as described by gμν ), but given those constraints, the ﬂat geometry underlying the split (17) is unobservable, and thus the choice between the two descriptions is conventional. Similarly in the case of quantum mechanics: if one accepts some form of Putnam’s argument (say, in the context of an Everett interpretation — see below), then keeping quantum mechanics as it is might indeed prompt us to revise our logic. But one can always retain classical logic, and have de BroglieBohm theory give a story of why this classical level is ‘hidden’. The theory will be observationally equivalent to standard quantum mechanics only given some constraints on the position distribution of the particles (‘equilibrium’), but given these constraints the postulated classical level is indeed not directly observable, and, always given 11 If one wishes, one can choose a dual ontology for the theory, in which both the conﬁgurations and the wave function are properties of the system. In this case, one can argue that quantum logic is applicable also in pilotwave theory to describe those intrinsic properties of a system that are encoded in its quantum state (which above we have called testable properties). However, one rejects the completeness of the standard interpretation, and at the additional level of the hidden variables one retains classical logic. It is the classical logic of the hidden variables that explains the eﬀectiveness of classical logic at the macroscopic level.
Is Logic Empirical?
73
the constraints, one could again argue that the choice between the two descriptions is conventional.
7.2
Spontaneous collapse theories
Spontaneous collapse theories are variants of quantum mechanics in which the Schr¨ odinger evolution is modiﬁed in order to reproduce the phenomenology of collapse. Such theories are generally stochastic, and the bestknown ones are on the one hand theories of the GhirardiRiminiWeber (GRW) type, in which the Schr¨ odinger equation is supplemented by certain discontinuous random transformations (Ghirardi, Rimini and Weber 1986); on the other hand the theory of continuous spontaneous localisation (CSL) and related ones, in which the wave function obeys a stochastic diﬀerential equation of a certain type (Pearle 1989). For simplicity, we shall focus on the former. The original version of the GRW theory consists of the following stochastic evolution of the wave function. For one particle, the Schr¨ odinger equation is supplemented at random times (with a ﬁxed average frequency 1/τ ) by a transformation known as a ‘hit’, namely a multiplication with a threedimensional Gaussian αλ (q − x), with a ﬁxed width λ, centred at some random position x, for which the theory speciﬁes a probability density. For n particles there are n independent such threedimensional hitting processes supplementing the Schr¨ odinger equation, which greatly increases the frequency of any such hit. The form of this evolution is the same as that used to describe some of the more general collapses observed in less idealised experiments, in this case an ‘unsharp’ measurement of position. The novelty with respect to the usual notion of collapse, however, is that this evolution is spontaneous, i. e. takes place irrespectively of the presence or absence of a measuring apparatus or any other system. Indeed, the theory is speciﬁcally intended to provide an approach to quantum mechanics that makes it universally applicable, in particular both to the microscopic and the macroscopic domains, as well as to their interaction. The way the GRW theory proposes to solve the measurement problem relies on the entangled form of the wave function in typical measurement situations, with components of the wave function corresponding (in the standard interpretation) to macroscopically diﬀerent states of the apparatus. For such states, even a single hit will trigger collapse on a macroscopic scale. Thus, at least prima facie, spontaneous collapse theories such as GRW embrace the standard interpretation of quantum mechanics but change the dynamics given by the Schr¨ odinger equation, so that the states that do not correspond to our everyday experience are very eﬃciently suppressed. This might seem to suggest that in spontaneous collapse approaches one can indeed maintain that quantum logic is the true logic, since (as discussed in section 5) the standard interpretation is compatible with this claim, and since spontaneous collapse removes the objections to the standard interpretation. However, spontaneous collapse theories solve the measurement problem by introducing classical
74
Guido Bacciagaluppi
alternatives in the possible evolution of the state, through the stochastic element of the dynamics. Thus, spontaneous collapse theories should be understood according to the alternative position that takes the properties assigned by the standard interpretation to be elementary properties, and complex ones as built from these using classical logic. Each of the alternatives in the evolution of the state will correspond to diﬀerent elementary properties of the quantum system, but the overall state of the system is a classical disjunction of these alternative states. Thus, also in spontaneous collapse theories (as in pilotwave theory), the quantum connectives do not provide the basis for the eﬀectiveness of the classical connectives. There is no story explaining that the cat is dead or alive classically because it is dead or alive quantum logically. The cat is ﬁrst ﬂeetingly (if at all) dead or alive quantum logically, then the dynamics intervenes and ensures the cat is dead or alive classically. Either a hit on the dead component takes place or one on the alive component does. One could say that in spontaneous collapse theories quantum logic is naturally adapted to describing the kinematics of the theory, but that the dynamics of the theory requires classical logic. Thus again, although in spontaneous collapse theories it may be natural to introduce quantum logic as a (local) logic at the level of the testable propositions, it cannot be taken as the basis for justifying the everyday use of classical logic. It might claim a place alongside classical logic, but it cannot replace it.12
7.3 Everett’s manyworlds The views associated with Everett’s manyworlds are indeed many, and some versions, such as the idea that the material universe literally multiplies whenever a measurement occurs, are out of favour for good reason. The modern version we shall adopt here has been developed over the last ﬁfteen years, mainly through the work of Saunders (1993), Wallace (2003, 2005) and Greaves (2004), and nowadays enjoys a broad though admittedly not universal consensus among philosophers of physics. The Everett approach is an interpretation of quantum mechanics in the strict sense, in that it takes the theory without any additions or modiﬁcations. It takes the ontology of quantum mechanics to be given completely by the wave function (of the universe), but instead of adopting a ‘God’s eye’ perspective on the wave function, it asks what would be an internal perspective on such a universe. The key insight of the interpretation is that through the mechanism of decoherence, the wave function develops components that have a stable identity over extremely long periods of time and that are dynamically independent of each other. It thus 12 More recently, a diﬀerent way of interpreting spontaneous collapse theories has been proposed, in terms of matter density (Ghirardi, Grassi and Benatti 1995). Analogously to the case of de BroglieBohm theory, regardless of whether this matter density is taken as the sole ontology of the theory or as a hidden variable additional to an ontological wave function, the eﬀectiveness of classical logic on the macroscopic scale will then derive from the applicability of classical logic to this matter density.
Is Logic Empirical?
75
makes sense to identify these components as quasiseparate ‘worlds’, and to deﬁne an internal perspective as centred on each such world. When a measurement occurs, each observer develops into generally many successors, indexed by their diﬀerent measurement results. So, which measurement result obtains is a matter of perspective: from the perspective of the live cat, the atom has not decayed and thereby triggered the smashing of the phial of cyanide; from the perspective of the dead cat, it has. Further recent work pioneered by Deutsch (1999) and perfected in particular by Wallace (2007) has sought to justify the use of the usual quantum probabilities on the basis of rational decision theory as adapted to such a ‘splitting’ agent. If one accepts Lewis’s Principal Principle as the deﬁnition of objective chances, the DeutschWallace results imply that the quantum probabilities are indeed objective in each world. What about logic? Note ﬁrst of all that, from the perspective of each world, the standard interpretation of quantum mechanics can be applied, taking the relevant component of the universal wave function to be the quantum state for that world. Note also that, although the description of a world given by the relevant component of the wave function is perspectival, it is no less objective than the description of the universe as given by the total wave function. Thus, again from the perspective of each world (which is the only perspective that makes sense empirically), quantum logic is well adapted to describe the intrinsic properties of physical systems. The question, as we know by now, is whether classical logic is required separately to make sense of the eﬀectiveness of classical logic on the macroscopic scale, or whether there is a sense in which quantum logic can explain how classical logic can be eﬀective in everyday cases, and therefore how we may have arrived to our classical conception of logic by abstraction from the everyday world. In the case of the Everett interpretation it now seems that this challenge is met. Indeed, while in general a quantum disjunction does not behave truthfunctionally (because the diﬀerent components of the wave function do not decohere, thus all belong to the same world), there are cases in which it does (because the diﬀerent components do decohere and thus belong to diﬀerent worlds). In such cases, from the perspective of each world, the disjuncts behave like classical alternatives, one of which is actual, the others counterfactual. Although in every world the properties of all physical systems are in bijective correspondence with subspaces of the Hilbert space, de facto unobserved macroscopic superpositions are not the kind of properties that appear in a typical world, unlike the case of the standard interpretation. And this is because interference between diﬀerent components becomes negligible, and an eﬀective superselection rule arises between the diﬀerent noninterfering components, thus mimicking the case of von Neumann algebras discussed in section 6. The relation between the quantum and the classical connectives is not a formal relation as we had in the case of von Neumann algebras, but the connectives behave classically in a suitable physical limit. Thus, while the structure of the intrinsic properties of physical systems supports
76
Guido Bacciagaluppi
a nondistributive logic at the fundamental level (even in the individual worlds), one can claim that, unlike the case of pilotwave theory or spontaneous collapse, the perspectival element characteristic of the Everett interpretation introduces a genuine emergence of the classical connectives from the quantum connectives. In this sense, it is only the Everett interpretation, among the major approaches to quantum mechanics, that is compatible with a revision of logic. One is not forced to accept the overall package, but, while perhaps not entirely as Putnam had articulated it, there is an intelligible sense in which (standard) quantum mechanics may suggest that logic be revised. 8 CONCLUSION We hope to have clariﬁed in what sense empirical considerations of quantum phenomena may have a bearing on the issue of the ‘true’ logic. Some of Putnam’s (1968) claims in this regard can be justiﬁed, but with qualiﬁcations. What can be said about the status of quantum logic in our world, assuming current approaches to the foundations of quantum mechanics, depends on the details of the chosen approach. In particular, one might justify a revision of logic at most if one chooses an Everett interpretation. Indeed, it is a general lesson in the philosophy of physics, conﬁrmed in the present case, that bold philosophical claims made on the basis of quantum mechanics turn out to be highly dependent on the interpretational approach one adopts towards the theory. The scenario in which consideration of quantum or quantumlike phenomena might make a revision of logic most appealing is possibly that of von Neumann algebras — thus perhaps vindicating Dickson’s (2001) intuition —, where there is a rigorous sense in which the quantum and classical connectives can be said to be the same and to behave truthfunctionally or not according to the meaning of the propositions involved. This possibility is presumably not realised in our world, but whether it is or not is itself an empirical issue, thus lending at least some support to the idea that logic is indeed empirical. ACKNOWLEDGEMENTS The ideas contained in this chapter were developed over many years, and it is impossible for me to remember all those who provided signiﬁcant input, feedback or encouragement. Hans Primas and Ernst Specker certainly exerted an early and durable inﬂuence on my ideas. While at Berkeley, I had the opportunity to discuss aspects of this project with a number of colleagues and students, in particular John MacFarlane, Russell O’Connor, Chris Pincock and Zoe SachsArellano. More recently, Huw Price and Ofer Gal gave me opportunities to present and discuss this material, and I am indebted for comments and discussions to Mark Colyvan, Stephen Gaukroger, Jason Grossman and, most particularly, to Sungho Choi. Finally, I wish to thank Kurt Engesser both for his encouragement and his patience as editor, without either of which this chapter would not have been written.
Is Logic Empirical?
77
BIBLIOGRAPHY [Bacciagaluppi, 1993] G. Bacciagaluppi. Critique of Putnam’s quantum logic, International Journal of Theoretical Physics 32, 1835–1846, 1993. [Baez, 1987] J. Baez. Bell’s inequality for C ∗ algebras, Letters in Mathematical Physics 13, 135–136, 1987. [Bell, 1987] J. S. Bell. Speakable and Unspeakable in Quantum Mechanics (Cambridge: Cambridge University Press), 1987. [Beltrametti and van Fraassen, 1981] E. Beltrametti and B. C. van Fraassen, eds. Current Issues in Quantum Logic (New York: Plenum Press), 1981. [Birkhoﬀ and von Neumann, 1936] G. Birkhoﬀ and J. von Neumann. The logic of quantum mechanics, Annals of Mathematics 37, 823–843, 1936. Reprinted in Hooker (1975), pp. 1–26. [Bohm, 1952] D. Bohm. A suggested interpretation of the quantum theory in terms of “hidden” variables, I and II, Physical Review 85, 166–179 and 180–193, 1952. ´ [de Broglie, 1928] L. de Broglie. La nouvelle dynamique des quanta, in Electrons et Photons: Rapports et Discussions du Cinqui` eme Conseil de Physique Solvay (Paris: GauthierVillars), pp. 105–132, 1928. Translated in G. Bacciagaluppi and A. Valentini, Quantum Theory at the Crossroads: Reconsidering the 1927 Solvay Conference (Cambridge: Cambridge University Press, forthcoming), pp. 375–399. [Brown, 1986] H. Brown. The insolubility proof of the quantum measurement problem, Foundations of Physics 16, 857–870, 1986. [Bub and Clifton, 1996] J. Bub and R. Clifton. A uniqueness theorem for “no collapse” interpretations of quantum mechanics, Studies in History and Philosophy of Modern Physics 27B, 181–219, 1996. [Bueno and Colyvan, 2004] O. Bueno and M. Colyvan. Logical nonapriorism and the “law” of noncontradiction, in G. Priest, J. C. Beall and B. ArmourGarb (eds.), The Law of NonContradiction: New Philosophical Essays (Oxford: Oxford University Press), pp. 156–175, 2004. [Dalla Chiara, 1981] M. L. Dalla Chiara. Some metalogical pathologies of quantum logic, in Beltrametti and Van Fraassen (1981), pp. 147–159. [Dalla Chiara and Giuntini, 2002] M. L. Dalla Chiara and R. Giuntini. Quantum logics, in D. Gabbay and F. Guenthner (eds.), Handbook of Philosophical Logic, vol. 6 (Dordrecht: Kluwer), pp. 129–228, 2002. [Deutsch, 1999] D. Deutsch. Quantum theory of probability and decisions, Proceedings of the Royal Society of London A 455, 3129–3137, 1999. [Dickson, 2001] M. Dickson. Quantum logic is alive ∧ (it is true ∨ it is false), Philosophy of Science (Proceedings) 68, S274–S287, 2001. [Dummett, 1976] M. Dummett. Is logic empirical?, in H. D. Lewis (ed.), Contemporary British Philosophy, 4th series (London: Allen and Unwin), pp. 45–68, 1976. Reprinted in M. Dummett, Truth and other Enigmas (London: Duckworth, 1978), pp. 269–289. [Finch, 1969] P. D. Finch. On the structure of quantum logic, Journal of Symbolic Logic 34, 275–282, 1969. Reprinted in Hooker (1975), pp. 415–425. [Foulis and Randall, 1981] D. J. Foulis and C. H. Randall. What are quantum logics and what ought they to be?, in Beltrametti and Van Fraassen (1981), pp. 35–52. [Friedman and Glymour, 1972] M. Friedman and C. Glymour. If quanta had logic, Journal of Philosophical Logic 1, 16–28, 1972. [van Fraassen, 1970] B. C. van Fraassen. On the extension of Beth’s semantics of physical theories, Philosophy of Science 37, 325–339, 1970. [Friedman and Putnam, 1978] M. Friedman and H. Putnam. Quantum logic, conditional probability, and interference, Dialectica 32, 305–315, 1978. [Ghirardi et al., 1995] G.C. Ghirardi, R. Grassi, and F. Benatti. Describing the macroscopic world: closing the circle within the dynamical reduction program, Foundations of Physics 25, 5–38, 1995. [Ghirardi et al., 1986] G.C. Ghirardi, A. Rimini, and T. Weber. Uniﬁed dynamics for microscopic and macroscopic systems, Physical Review D 34, 470–491, 1986.
78
Guido Bacciagaluppi
[G¨ odel, 1933] K. G¨ odel. Eine Interpretation des intuitionistischen Aussagenkalk¨ uls, Ergebnisse eines mathematischen Kolloquiums 4, 39–40, 1933. Reprinted and translated in K. G¨ odel, Collected Works, vol. 1, ed. by S. Feferman, S. Kleene, G. Moore, R. Solovay and J. van Heijenoort (Oxford: Oxford University Press, 1986), pp. 300–303. [Greaves, 2004] H. Greaves. Understanding Deutsch’s probability in a deterministic multiverse, Studies in History and Philosophy of Modern Physics 35B, 423–456, 2004. [Gudder, 1972] S. Gudder. Partial algebraic structures associated with orthomodular posets, Paciﬁc Journal of Mathematics 41, 712–730, 1972. [Hardegree, 1975] G. M. Hardegree. Stalnaker conditionals and quantum logic, Journal of Philosophical Logic 4, 399–421, 1975. [Heywood and Redhead, 1983] P. Heywood and M. L. G. Redhead. Nonlocality and the KochenSpecker paradox, Foundations of Physics 13, 481–499, 1983. [Hooker, 1975] C. A. Hooker. The LogicoAlgebraic Approach to Quantum Mechanics, vol. 1 (Dordrecht: Reidel), 1975. [Jauch and Piron, 1963] J. M. Jauch and C. Piron. Can hidden variables be excluded in quantum mechanics?, Helvetica Physica Acta 36, 827–837, 1963. [Jauch and Piron, 1969] J. M. Jauch and C. Piron. On the structure of quantal proposition systems, Helvetica Physica Acta 43, 842–848, 1969. Reprinted in Hooker (1975), pp. 427–436. [Kochen and Specker, 1965a] S. Kochen and E. P. Specker. Logical structures arising in quantum theory, in L. Addison, L. Henkin and A. Tarski (eds.), The Theory of Models (Amsterdam: NorthHolland), pp. 177–189, 1965. Reprinted in Hooker (1975), pp. 263–276. [Kochen and Specker, 1965b] S. Kochen and E. P. Specker. The calculus of partial propositional functions, in Y. BarHillel (ed.), Logic, Methodology, and Philosophy of Science (Amsterdam: NorthHolland), pp. 45–57, 1965. Reprinted in Hooker (1975), pp. 277–292. [Kochen and Specker, 1967] S. Kochen and E. P. Specker. The problem of hidden variables in quantum mechanics, Journal of Mathematics and Mechanics 17, 59–88, 1967. Reprinted in Hooker (1975), pp. 293–328. [Kolmogoroﬀ, 1931] A. Kolmogoroﬀ. Zur Deutung der intuitionistischen Logik, Mathematische Zeitschrift 35, 58–65, 1931. [von Neumann, 1932] J. von Neumann. Mathematische Grundlagen der Quantenmechanik (Berlin: Springer), 1932. [Pearle, 1989] P. Pearle. Combining stochastic dynamical statevector reduction with spontaneous localization, Physical Review A 39, 2277–2289, 1989. [Putnam, 1968] H. Putnam. Is logic empirical?, in R. Cohen and M. Wartofsky (eds.), Boston Studies in the Philosophy of Science, vol. 5 (Dordrecht: Reidel), pp. 216–241, 1968. Reprinted as ‘The logic of quantum mechanics’ in H. Putnam, Mathematics, Matter, and Method. Philosophical Papers, vol. 1 (Cambridge: Cambridge University Press, 1975), pp. 174–197. [Putnam, 1974] H. Putnam. How to think quantumlogically, Synthese, 29, 55–61, 1974. Reprinted in P. Suppes (ed.), Logic and Probability in Quantum Mechanics (Dordrecht: Reidel, 1976) pp. 47–53. [Putnam, 1981] H. Putnam. Quantum mechanics and the observer, Erkenntnis 16, 193–219, 1981. [Putnam, 1994] H. Putnam. Michael Redhead on quantum logic, in P. Clark and R. Hale (eds.), Reading Putnam (Oxford: Basil Blackwell), pp. 265–280, 1994. [Raggio, 1981] G. A. Raggio. States and Composite Systems in W ∗ algebraic Quantum Mechanics, Diss. ETH No. 6824, 1981. [Raggio, 1988] G. A. Raggio. A remark on Bell’s inequality and decomposable normal states, Letters in Mathematical Physics, 15, 27–29, 1988. [Saunders, 1993] S. Saunders. Decoherence, relative states, and evolutionary adaptation, Foundations of Physics 23, 1553–1585, 1993. [Schr¨ odinger, 1935] E. Schr¨ odinger. Discussion of probability relations between separated systems, Proceedings of the Cambridge Philosophical Society 31, 555–563, 1935. [Stairs, 1983] A. Stairs. Quantum logic, realism, and valuedeﬁniteness, Philosophy of Science 50, 578–602, 1983. [Wallace, 2003] D. Wallace. Everett and structure, Studies in History and Philosophy of Modern Physics 34B, 87–105, 2003. [Wallace, 2007] D. Wallace. Quantum probability from subjective likelihood: improving on Deutsch’s proof of the probability rule, Studies in History and Philosophy of Modern Physics, 38, 311–332, 2007.
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
79
QUANTUM AXIOMATICS Diederik Aerts
1
INTRODUCTION
Quantum axiomatics has its roots in the work of John von Neumann, in collaboration with Garett Birkhoﬀ, that is almost as old as the standard formulation of quantum mechanics itself [Birkhoﬀ and von Neumann, 1936]. Indeed already during the beginning years of quantum mechanics, the formalism that is now referred to as standard quantum mechanics [von Neumann, 1932], was thought to be too speciﬁc by the founding fathers themselves. One of the questions that obviously was at the origin of this early dissatisfaction is: ‘Why would a complex Hilbert space deliver the unique mathematical structure for a complete description of the microworld? Would that not be amazing? What is so special about a complex Hilbert space that its mathematical structure would play such a fundamental role?’ Let us turn for a moment to the other great theory of physics, namely general relativity, to raise more suspicion towards the fundamental role of the complex Hilbert space for quantum mechanics. General relativity is founded on the mathematical structure of Riemann geometry. In this case however it is much more plausible that indeed the right fundamental mathematical structure has been taken. Riemann developed his theory as a synthesis of the work of Gauss, Lobatsjevski and Bolyai on nonEuclidean geometry, and his aim was to work out a theory for the description of the geometrical structure of the world in all its generality. Hence Einstein took recourse to the work of Riemann to express his ideas and intuitions on space time and its geometry and this lead to general relativity. General relativity could be called in this respect ‘the geometrization of a part of the world including gravitation’. There is, of course, a deﬁnite reason why von Neumann used the mathematical structure of a complex Hilbert space for the formalization of quantum mechanics, but this reason is much less profound than it is for Riemann geometry and general relativity. The reason is that Heisenberg’s matrix mechanics and Schr¨ odinger’s wave mechanics turned out to be equivalent, the ﬁrst being a formalization of the new mechanics making use of l2 , the set of all square summable complex sequences, and the second making use of L2 (R3 ), the set of all square integrable complex functions of three real variables. The two spaces l2 and L2 (R3 ) are canonical examples of a complex Hilbert space. This means that Heisenberg and Schr¨ odinger were working already in a complex Hilbert space, when they formulated matrix mechanics and wave mechanics, without being aware of it. This made it a straightforward
80
Diederik Aerts
choice for von Neumann to propose a formulation of quantum mechanics in an abstract complex Hilbert space, reducing matrix mechanics and wave mechanics to two speciﬁc cases. One problem with the Hilbert space representation was known from the start. A (pure) state of a quantum entity is represented by a unit vector or ray of the complex Hilbert space, and not by a vector. Indeed vectors contained in the same ray represent the same state or one has to normalize the vector that represents the state after it has been changed in one way or another. It is well known that if rays of a vector space are called points and two dimensional subspaces of this vector space are called lines, the set of points and lines corresponding in this way to a vector space, form a projective geometry. What we just remarked about the unit vector or ray representing the state of the quantum entity means that in some way the projective geometry corresponding to the complex Hilbert space represents more intrinsically the physics of the quantum world as does the Hilbert space itself. This state of aﬀairs is revealed explicitly in the dynamics of quantum entities, that is built by using group representations, and one has to consider projective representations, which are representations in the corresponding projective geometry, and not vector representations [Wigner, 1959]. The title of the article by John von Neumann and Garett Birkhoﬀ [Birkhoﬀ and von Neumann, 1936] that we mentioned as the founding article for quantum axiomatics is ‘The logic of quantum mechanics’. Let us explain shortly what Birkhoﬀ and von Neumann do in this article. First of all they remark that an operational proposition of a quantum entity is represented in the standard quantum formalism by an orthogonal projection operator or by the corresponding closed subspace of the Hilbert space H. Let us denote the set of all closed subspaces of H by L(H). Next Birkhoﬀ and von Neumann show that the structure of L(H) is not that of a Boolean algebra, the archetypical structure of the set of propositions in classical logic. More speciﬁcally it is the distributive law between conjunction and disjunction (a ∨ b) ∧ c = (a ∧ c) ∨ (b ∧ c)
(1)
that is not necessarily valid for the case of quantum propositions a, b, c ∈ L(H). A whole line of research, called quantum logic, was born as a consequence of the Birkhoﬀ and von Neumann article. The underlying philosophical idea is that, in the same manner as general relativity has introduced nonEuclidean geometry into the reality of the physical world, quantum mechanics introduces nonBoolean logic. The quantum paradoxes would be due to the fact that we reason with Boolean logic about situations with quantum entities, while these situations should be reasoned about with nonBoolean logic. Although fascinating as an approach [Mittelstaedt, 1963], it is not this idea that is at the origin of quantum axiomatics. Another aspect of what Birkhoﬀ and von Neumann did in their article is that they shifted the attention on the mathematical structure of the set of operational propositions L(H) instead of the Hilbert space H itself. In this sense it is important to pay attention to the fact that L(H) is
Quantum Axiomatics
81
the set of all operational propositions, i.e. the set of yes/no experiments on a quantum entity. They opened a way to connect abstract mathematical concepts of the quantum formalism, namely the orthogonal projection operators or closed subspaces of the Hilbert space, directly with physical operations in the laboratory, namely the yes/no experiments. George Mackey followed in on this idea when he wrote his book on the mathematical foundations of quantum mechanics [Mackey, 1963]. He starts the other way around and considers as a basis the set L of all operational propositions, meaning propositions being testable by yes/no experiments on a physical entity. Then he introduces as an axiom that this set L has to have a structure isomorphic to the set of all closed subspaces L(H) of a complex Hilbert space in the case of a quantum entity. He states that it would be interesting to invent a set of axioms on L that gradually would make L more and more alike to L(H) to ﬁnally arrive at an isomorphism when all the axioms are satisﬁed. While Mackey wrote his book results as such were underway. A year later Constantin Piron proved a fundamental representation theorem. Starting from the set L of all operational propositions of a physical entity and introducing ﬁve axioms on L he proved that L is isomorphic to the set of closed subspaces L(V ) of a generalized Hilbert space V whenever these ﬁve axioms are satisﬁed [Piron, 1964]. Let us elaborate on some of the aspects of this representation theorem to be able to explain further what quantum axiomatics is about. We mentioned already that Birkhoﬀ and von Neumann had noticed that the set of closed subspaces L(H) of a complex Hilbert space H is not a Boolean algebra, because distributivity between conjunction and disjunction, like expressed in (1), is not satisﬁed. The set of closed subspaces of a complex Hilbert space forms however a lattice, which is a more general mathematical structure than a Boolean algebra, moreover, a lattice where the distributivity rule (1) is satisﬁed is a Boolean algebra, which indicates that the lattice structure is the one to consider for the quantum mechanical situation. As we will see more in detail later, and to make again a reference to general relativity, the lattice structure is indeed to a Boolean algebra what general Riemann geometry is to Euclidean geometry. And moreover, meanwhile we have understood why the structure of operational propositions of the world is not a Boolean algebra but a lattice. This is strictly due to the fact that measurements can have an uncontrollable inﬂuence on the state of the physical entity under consideration. We explain this insight in detail in [Aerts and Aerts, 2004], and mention it here to make clear that the intuition of Birkhoﬀ and von Neumann, and later Mackey, Piron and others, although only mathematical intuition at that time, was correct. When Piron proved his representation theorem in 1964, he concentrated on the lattice structure for the formulation of the ﬁve axioms. Meanwhile much more research has been done, both physically motivated in an attempt to make the approach more operational, as well as mathematically, trying to get axiomatically closer to the complex Hilbert space. In the presentation of quantum axiomatics we give in this article, we integrate the most recent results, and hence deviate
82
Diederik Aerts
for this reason from the original formulation, for example when we explain the representation theorem of Piron. Axiomatic quantum mechanics is more than just an axiomatization of quantum mechanics. Because of the operational nature of the axiomatization, it holds the potential for ‘more general theories than standard quantum mechanics’ which however are ‘quantum like theories’. In this sense, we believe that it is one of the candidates to generate the framework for the new theory to be developed generalizing quantum mechanics and relativity theory [Aerts and Aerts, 2004]. Let us explain why we believe that quantum axiomatics has the potential to deliver such a generalization of relativity theory and quantum mechanics. General relativity is a theory that brings part of the world that in earlier Newtonian mechanics was classiﬁed within dynamics to the geometrical realm of reality, and more specifically confronting us with the prescientiﬁc and naive realistic vision on space, time, matter and gravitation. It teaches us in a deep and new way, compared to Newtonian physics, ‘what are the things that exists and how they exist and are related and how they inﬂuence each other’. But there is one deep lack in relativity theory: it does not take into account the inﬂuence of the observer, the eﬀect that the measuring apparatus has on the thing observed. It does not confront the subjectobject problem and its inﬂuence on how reality is. It cannot do this because its mathematical apparatus is based on the Riemann geometry of timespace, hence prejudicing that timespace is there, ﬁlled up with ﬁelds and matter, that are also there, independent of the observer. There is no fundamental role for the creation of ‘new’ within relativity theory, everything just ‘is’ and we are only there to ‘detect’ how this everything ‘is’. That is also the reason why general relativity can easily be interpreted as delivering a model for the whole universe, whatever this would mean. We know that quantum mechanics takes into account in an essential way the eﬀect of the observer through the measuring apparatus on the state of the physical entity under study. In a theory generalizing quantum mechanics and relativity, such that both appear as special cases, this eﬀect should certainly also appear in a fundamental way. We believe that general relativity has explored to great depth the question ‘how can things be in the world’. Quantum axiomatics explores in great depth the question ‘how can be acted in the world’. And it does explore this question of ‘action in the world’ in a very similar manner as general relativity theory does with its question of ‘being of the world’. This means that operational quantum axiomatics can be seen as the development of a general theory of ‘actions in the world’ in the same manner that Riemann geometry can be seen as a general theory of ‘geometrical forms existing in the world’. Of course Riemann is not equivalent to general relativity, a lot of detailed physics had to be known to apply Riemann resulting in general relativity. This is the same with operational quantum axiomatics, it has the potential to deliver the framework for the theory generalizing quantum mechanics and relativity theory, but a lot of detailed physics will have to be used to ﬁnd out the exact way of doing this. We want to remark that in principle a theory that describes the possible actions
Quantum Axiomatics
83
in the world, and a theory that delivers a model for the whole universe, should not be incompatible. It should even be so that the theory that delivers a model of the whole universe should incorporate the theory of actions in the world, which would mean for the situation that exists now, general relativity should contain quantum mechanics, if it really delivers a model for the whole universe. That is why we believe that Einstein’s attitude, trying to incorporate the other forces and interactions within general relativity, contrary to common believe, was the right one, globally speaking. What Einstein did not know at that time was ‘the reality of nonlocality in the microworld’. Nonlocality means nonspatiality, which means that the reality of the microworld, and hence the reality of the universe as a whole, is not timespace like. Timespace is not the global theatre of reality, but rather a cristallization and structuration of the macroworld. Timespace has come into existence together with the macroscopic material entities, and hence it is ‘their’ time and space, but it is not the theatre of the microscopic quantum entities. This fact is the fundamental reason why general relativity, built on the mathematical geometrical Riemannian structure of timespace, cannot be the canvas for the new theory to be developed. A way to express this technically would be to say that the set of events cannot be identiﬁed with the set of timespace points as is done in relativity theory. Recourse will have to be taken to a theory that describes reality as a kind of pregeometry, and where the geometrical structure arises as a consequence of interactions that collapse into the timespace context. We believe that operational quantum axiomatics, as presented in this article, can deliver the framework as well as the methodology to construct and elaborate such a theory. In the next section we introduce the basic notions of operational quantum axiomatics. Mackey and Piron introduced the set of yes/no experiments but then immediately shifted to an attempt to axiomatize mathematically the lattice of (operational) propositions of a quantum entity, Mackey postulating right away an isomorphism with L(H) and Piron giving ﬁve axioms to come as close as possible to L(H). Also Piron’s axioms are however mostly motivated by mimicking mathematically the structure of L(H). In later work Piron made a stronger attempt to found operationally part of the axioms [Piron, 1976], and this attempt was worked out further in [Aerts, 1981; Aerts, 1982; Aerts, 1983], to arrive at a full operational foundation only recently [Aerts, 1999a; Aerts, 1999b; Aerts et al., 1999; Aerts, 2002]. Also mathematically the circle was closed only recently. At the time when Piron gave his ﬁve axioms that lead to the representation within a generalized Hilbert space, there only existed three examples of generalized Hilbert spaces that ﬁtted all the axioms, namely real, complex and quaternionic Hilbert space, also referred to as the three standard Hilbert spaces.1 Years later Hans Keller constructed 1 There
do exist a lot of ﬁnite dimensional generalized Hilbert spaces that are diﬀerent from the three standard examples. But since a physical entity has to have at least a position observable, it follows that the generalized Hilbert space must be inﬁnite dimensional. At the time of Piron’s representation theorem, the only inﬁnite dimensional cases that were known are the three standard Hilbert spaces, over the real, complex or quaternionic numbers.
84
Diederik Aerts
the ﬁrst counterexample, more speciﬁcally an example of an inﬁnite dimensional generalized Hilbert space that is not isomorphic to one of the three standard Hilbert spaces [Keller, 1980]. The study of generalized Hilbert spaces, nowadays also called orthomodular spaces, developed into a research subject of its own, and recently Maria Pia Sol`er proved a groundbreaking theorem in this ﬁeld. She proved that an inﬁnite dimensional generalized Hilbert space that contains an orthonormal base is isomorphic with one of the three standard Hilbert spaces [Sol`er, 1995]. It has meanwhile also been possible to formulate an operational axiom, called ‘plane transitivity’ on the set of operational propositions that implies Sol`er’s condition [Aerts and van Steirteghem, 2000], which completes the axiomatics for standard quantum mechanics by means of six axioms, the original ﬁve axioms of Piron and plane transitivity as sixth axiom. 2 STATE PROPERTY SPACES In this section we introduce the basic notions and basic axioms for quantum axiomatics. We introduce notions and axioms that are as simple as possible, but each time show how the more traditional axioms of quantum axiomatics are related and/or derived from our set of axioms.
2.1 States and properties With each entity S corresponds a well deﬁned set of states Σ of the entity. These are the modes of being of the entity. This means that at each moment the entity S ‘is’ in a speciﬁc state p ∈ Σ. Historically quantum axiomatics has been elaborated mainly by considering the set of properties2 . With each entity S corresponds a well deﬁned set of properties L. A property a ∈ L is ‘actual’ or is ‘potential’ for the entity S. To be able to present the axiomatisation of the set of states and the set of properties of an entity S in a mathematical way, we introduce some additional notions. Suppose that the entity S is in a speciﬁc state p ∈ Σ. Then some of the properties of S are actual and some are not, hence they are potential. This means that with each state p ∈ Σ corresponds a set of actual properties, subset of L. This deﬁnes a function ξ : Σ → P(L), which makes each state p ∈ Σ correspond to the set ξ(p) of properties that are actual in this state. With the notation P(L) we mean the ‘powerset’ of L, i.e. the set of all subsets of L. From now on we can replace the statement ‘property a ∈ L is actual for the entity S in state p ∈ Σ’ by ‘a ∈ ξ(p)’. Suppose that for the entity S a speciﬁc property a ∈ L is actual. Then this entity is in a certain state p ∈ Σ that makes a actual. With each property a ∈ L 2 In the original paper of Birkhoﬀ and Von Neumann [Birkhoﬀ and von Neumann, 1936], the basic notion is the one of ‘operational proposition’. An operational proposition is not the same as a property [Randall and Foulis, 1983; Foulis et al., 1983], but it points at the same structural part of quantum axiomatics.
Quantum Axiomatics
85
we can associate the set of states that make this property actual, i.e. a subset of Σ. This deﬁnes a function κ : L → P(Σ), which makes each property a ∈ L correspond to the set of states κ(a) that make this property actual. We can replace the statement ‘property a ∈ L is actual if the entity S is in state p ∈ Σ’ by the expression ‘p ∈ κ(a)’. Summarising the foregoing we have: property a ∈ L is actual for the entity S in state p ∈ Σ ⇔ a ∈ ξ(p) ⇔ p ∈ κ(a)
(2)
This expresses a fundamental ‘duality’ between states and properties. We introduce a speciﬁc mathematical structure to represent an entity S, its states and its properties, taking into account this duality. First we remark that if Σ and L are given, and one of the two functions ξ or κ is given, then the other function can be derived. Let us show this explicitly. Hence suppose that Σ, L and ξ are given, and deﬁne κ : L → P(Σ) such that κ(a) = {p  p ∈ Σ, a ∈ ξ(p)}. Similarly, if Σ, L and κ : L → P(Σ) are given, we can derive ξ in an analogous way. This means that to deﬁne the mathematical structure which carries our notions and relations it is enough to introduce Σ, L and one of the two functions ξ or κ. DEFINITION 1 State property space. Consider two sets Σ and L and a function ξ : Σ ← P(L)
p → ξ(p)
(3)
then we say that (Σ, L, ξ) is a state property space. The elements of Σ are interpreted as states and the elements of L as properties of the entity S. For p ∈ Σ we have that ξ(p) is the set of properties of S which are actual if S is in state p. For a state property space (Σ, L, ξ) we deﬁne: κ : L → P(Σ)
a → κ(a) = {p  p ∈ Σ, a ∈ ξ(p)}
(4)
and hence for a ∈ L we have that κ(a) is the set of states of the entity S which make the property a actual. The function κ is called the Cartan map of the state property space (Σ, L, ξ). PROPOSITION 2. Consider a state property space (Σ, L, ξ), and κ deﬁned as in (4). We have: a ∈ ξ(p) ⇔ p ∈ κ(a)
(5)
There are two natural ‘implication relations’ on a state property space. If the situation is such that if ‘a ∈ L is actual for S in state p ∈ Σ’ implies that ‘b ∈ L is actual for S in state p ∈ Σ’ we say that the property a implies the property b. If the situation is such that ‘a ∈ L is actual for S in state q ∈ Σ’ implies that ‘a ∈ L is actual for S in state p ∈ Σ’ we say that the state p implies the state q.
86
Diederik Aerts
DEFINITION 3 Property implication and state implication. Consider a state property space (Σ, L, ξ). For a, b ∈ L we introduce: a ≤ b ⇔ κ(a) ⊆ κ(b)
(6)
and we say that a ‘implies’ b. For p, q ∈ Σ we introduce: p ≤ q ⇔ ξ(q) ⊆ ξ(p)
(7)
and we say that p ‘implies’ q 3 . DEFINITION 4 Equivalent properties and equivalent states. Consider a state property space (Σ, L, ξ). We call properties a, b ∈ L equivalent, and denote a ≈ b iﬀ κ(a) = κ(b). We call states p, q ∈ Σ equivalent and denote p ≈ q iﬀ ξ(p) = ξ(q). Let us give two important examples of state property spaces. First, consider a set Ω and let P(Ω) be the set of all subsets of Ω, and consider the function ξΩ : Ω → P(P(Ω)), such that for p ∈ Ω ξΩ (p) = {A A ∈ P(Ω), p ∈ A}
(8)
The triple (Ω, P(Ω), ξΩ ) is a state property space. For A ∈ P(Ω) we have κΩ (A) = {p p ∈ Ω, A ∈ ξ(p)} = {p p ∈ Ω, p ∈ A} = A. This shows that κΩ : P(Ω) → P(Ω) is the identity. Second, consider a complex Hilbert space H, and let Σ(H) be the set of unit vectors of H and L(H) the set of orthogonal projection operators of H. Consider the function ξH : Σ(H) → P(L(H)), such that for x ∈ Σ(H) ξH (x) = {A  A ∈ L(H), Ax = x}
(9)
The triple (Σ(H), L(H), ξH ) is a state property space. For A ∈ L(H) we have κH (A) = {x x ∈ Σ(H), Ax = x}. The two examples that we propose here are the archetypical physics examples. The ﬁrst example is the state property space of a classical physical system, where Ω corresponds with its state space. The second example is the state property space of a quantum physical system, where H is the complex Hilbert space connected to the quantum system. DEFINITION 5 Preorder relation. Suppose that we have a set Z. We say that ≤ is a preorder relation on Z iﬀ for x, y, z ∈ Z we have: x≤x x ≤ y and y ≤ z ⇒ x ≤ z
(10)
For two elements x, y ∈ Z such that x ≤ y and y ≤ x we denote x ≈ y and we say that x is equivalent to y. 3 The state implication and property implication are not deﬁned in an analogous way. Indeed, then we should for example have written p ≤ q ⇔ ξ(p) ⊆ ξ(q). That we have chosen to deﬁne the state implication the other way around is because historically this is how intuitively is thought about states implying one another.
Quantum Axiomatics
87
It is easy to verify that the implication relations that we have introduced are preorder relations. PROPOSITION 6. Consider a state property space (Σ, L, ξ), then Σ, ≤ and L, ≤ are preordered sets. We can show the following for a state property space: PROPOSITION 7. Consider a state property space (Σ, L, ξ). (1) Suppose that a, b ∈ L and p ∈ Σ. If a ∈ ξ(p) and a ≤ b, then b ∈ ξ(p). (2) Suppose that p, q ∈ Σ and a ∈ L. If q ∈ κ(a) and p ≤ q then p ∈ κ(a). Proof. (1) We have p ∈ κ(a) and κ(a) ⊆ κ(b). This proves that p ∈ κ(b) and hence b ∈ ξ(p). (2) We have a ∈ ξ(q) and ξ(q) ⊆ ξ(p) and hence a ∈ ξ(p). This shows that p ∈ κ(a). Suppose we consider a set of properties (ai )i ⊆ L. It is very well possible that there exist states of the entity S in which all the properties ai are actual. This is in fact always the case if ∩i κ(ai ) = ∅. Indeed, if we consider p ∈ ∩i κ(ai ) and S in state p, then all the properties ai are actual. If it is such that the situation where all properties ai of a set (ai )i and no other are actual is again a property of the entity S, we will denote this new property by ∧i ai , and call it a ‘meet property’ of (ai )i . Clearly we have ∧i ai is actual for S in state p ∈ Σ iﬀ ai is actual for all i for S in state p. This means that we have ∧i ai ∈ ξ(p) iﬀ ai ∈ ξ(p) ∀i. DEFINITION 8 Meet property. Consider a state property space (Σ, L, ξ) and a set (ai )i ⊆ L of properties. If there exists a property, which we denote by ∧i ai , such that κ(∧i ai ) = ∩i κ(ai )
(11)
we call ∧i ai the ‘meet property’ of the set of properties (ai )i . If we have the structure of a preordered set, we can wonder about the existence of meets and joins with respect to this preorder, or conjunctions and disjunctions with respect to the implication related to this preorder. In relation with the meet property we can prove the following PROPOSITION 9. Consider a state property space (Σ, L, ξ) and a set (ai )i ⊆ L of properties. The property ∧i ai , if it exists, is an inﬁmum4 for the preorder relation ≤ on L. Proof. We have κ(∧i ai ) = ∩i κ(ai ) ⊆ κ(aj ) ∀ j, and hence ∧i ai ≤ aj ∀ j. Suppose that x ∈ L is such that x ≤ aj ∀j, then we have κ(x) ⊆ κ(aj )∀ j, and hence κ(x) ⊆ ∩i κ(ai ) = κ(∧i ai ). As a consequence we have x ≤ ∧i ai . This proves that ∧i ai is an inﬁmum. 4 An inﬁmum of a subset (x ) of a preordered set Z is an element of Z that is smaller than i i all the xi and greater than any element that is smaller than all xi .
88
Diederik Aerts
2.2 Tests For the operational foundations of the state property space, we need to make explicit how we test whether for a physical entity a speciﬁc property is actual. A test is an experiment we can perform on the physical entity under investigation with the aim of knowing whether a speciﬁc property of this physical entity is actual or not. We identify for each test two outcomes, one which we call ‘yes’ corresponding to the occurrence of the expected outcome, and another one which we call ‘no’ corresponding to the non occurrence of the expected outcome. However, if for a test the outcome ‘yes’ occurs, this does not mean that the property which is tested is actual. It is only when we can predict with certainty, i.e. with probability equal to 1, that the test would have an outcome ‘yes’, if we would perform it, that the property a is actual. Let us consider the example of an entity which is a piece of wood. We have in mind the property of ‘burning well’. A possible test for this property consists of taking the piece of wood and setting it on ﬁre. In general, when we perform the test on a piece of dry wood, the piece of wood will be destroyed by the test. So the property of ‘burning well’ is a property that the piece of wood eventually has before we make the test. Of course it is after having done a number of tests with pieces of wood and having got always the outcome yes, we decide that the one new piece of wood, prepared under equivalent conditions, whereon we never performed the test, has actually the property of burning well. We will say that the test is ‘true’ if this is the case. DEFINITION 10 Testing a property. Consider a physical entity with corresponding state property space (Σ, L, ξ). α is a test of the property a ∈ L if we have a ∈ ξ(p) ⇔ ‘yes can be predicted with certainty for α when S is in state p (12) Similarly with the preorder relations on the sets of properties we have preorder relations on the sets of tests. DEFINITION 11 Test implication. We say that a test α is stronger than a test β and denote α ≤ β iﬀ whenever the physical entity is in a state such that α is true then also β is true. PROPOSITION 12. Consider a state property space (Σ, L, ξ). If the test α tests property a, and the test β tests property b, we have α≤β⇔a≤b
(13)
Proof. Suppose that α ≤ β, and consider p ∈ Σ such that a ∈ ξ(p). This means that the test α gives with certainty outcome ‘yes’ if the entity is in state p. Hence also β gives with certainty ‘yes’ if the entity is in state p. This means that b ∈ ξ(p). Hence we have proven that a ≤ b. Suppose now that a ≤ b, and suppose that the entity is in a state p such that α gives with certainty outcome ‘yes’. This means that a ∈ ξ(p). Then we have b ∈ ξ(p). Since β tests b we have that β gives with certainty the outcome ‘yes’. Hence we have proven that α ≤ β.
Quantum Axiomatics
89
DEFINITION 13 Equivalent tests. We say that two tests α and β are equivalent, and denote α ≈ β iﬀ α ≤ β and β ≤ α. PROPOSITION 14. Equivalent tests test equivalent properties, and tests that test equivalent properties are equivalent tests. Proof. Consider two equivalent tests α ≈ β testing respectively properties a and b. Since we have α ≤ β and β ≤ α this implies that a ≤ b and b ≤ a, and hence a ≈ b. Consider two equivalent properties a ≈ b being tested respectively by tests α and β. Since we have a ≤ b and b ≤ a this implies that α ≤ β and β ≤ α, and hence α ≈ β.
In general the outcomes of a test of one property are profoundly inﬂuenced by the testing of another property. In most cases it makes even no sense to perform two tests at once or one after the other on the entity. But still it is so that every entity can have several properties which are actual at once. There is indeed a way to construct a test that makes it possible to test the actuality of several properties at once, even if the tests corresponding to the diﬀerent properties disturb each other profoundly. Let us illustrate this by means of the example of the piece of wood. Consider the following two properties of the piece of wood: Property c ‘the piece of wood burns well’ and property d ‘the piece of wood ﬂoats on water’. Suppose that γ is a test of property c which consists of setting the wood on ﬁre and giving the outcome ‘yes’ if it burns well. The test δ consists of putting the wood on water and giving the outcome ‘yes’ if it ﬂoats, hence it is a test of property d. If we perform ﬁrst the test δ, and put the piece of wood on water, we have changed the state of the wood in a state of ‘wet wood’ and as a result the wood will not burn well. On the other hand if we perform the test γ and burn the wood, it will no longer ﬂoat on water. However we all know plenty of pieces of wood for which both properties c and d are actual at once. This means that the way in which we decide both properties to be actual for a speciﬁc piece of wood is not related to performing both tests one after the other. If we analyse carefully this situation we see that we agree for a piece of wood both properties c and d to be actual if which ever of the tests γ or δ is performed, the outcome ‘yes’ can be predicted with certainty for this test. Hence, to state this in a slightly more formal way: ‘If we choose, or if some process external to us produces a choice, between one of the two tests γ or δ, and it is certain to obtain the outcome ‘yes’ no matter what is this choice, then we agree that both properties c and d are actual for the piece of wood’. This leads us to the following. Given two tests γ and δ we deﬁne a new test which we denote γ · δ and call the product test of γ and δ. The performance of γ · δ consists of a choice being made between γ and δ, and then the performance of this chosen test, and the attribution of the outcome obtained in this way. As a consequence, we have γ · δ is true iﬀ γ is true ‘and’ δ is true, which shows that γ · δ tests both properties c and d, or, it tests the conjunction of properties c and
90
Diederik Aerts
d. Remark that for the performance of the test γ · δ only one test γ or δ has to be performed, and hence the deﬁnition of the product test is valid independent of the way in which tests disturb each other. The deﬁnition of product test is valid for any number of tests, which means that we have found a way to test any number of properties at once, or, to test the conjunction of any number of properties. Let us formally introduce the product test for an arbitrary number of properties. Consider a family (ai )i of properties ai and tests αi , such that αi tests property ai . A test which tests the actuality of all the properties ai , and which we denote Πi αi and call the product of the αi is the following: DEFINITION 15 Product test. The performance of Πi αi consists of a choice between one of the tests αi followed by the performance of this chosen test. PROPOSITION 16. For a set of tests (αi )i we have Πi αi ≤ αj ∀ j
(14)
Let us prove that the product test tests the meet of a set of properties. PROPOSITION 17. Consider an entity with corresponding state property space (Σ, L, ξ) and a set of properties (ai )i ⊆ L. Suppose that we have tests (αi )i available for the properties (ai )i , then the product test Πi αi tests a meet property ∧i ai . Proof. Following the deﬁnition of ‘meet property’ given in Deﬁnition 8, to prove that Πi αi tests the meet property ∧i ai of the set of properties (ai )i , where αi tests ai , we need to show that ‘yes can be predicted with certainty for Πi α the entity being in state p’ is equivalent to ‘ai ∈ ξ(p) ∀ i’. This follows from the deﬁnition of the product test. Indeed ‘yes can be predicted with certainty for Πi αi the entity being in state p’ is equivalent to ‘yes can be predicted with certainty for αi ∀ i the entity being in state p’.
2.3 Orthogonality Let us investigate the operational foundation of orthogonality. DEFINITION 18 Inverse test. For a test α we consider the test that consists of performing the same experiment and changing the role of ‘yes’ and ‘no’. We denote this new test by α
, and call it the inverse test of α. PROPOSITION 19. Consider a test α and a set of tests (αi )i , then we have
α
i αi Π
= α =
Πi α i
(15) (16)
Proof. Obviously if we exchange ‘yes’ and ‘no’ for an experiment corresponding to the test α, and then exchange ‘yes’ and ‘no’ again, we get the same test. The
i αi consists of exchanging ‘yes’ and ‘no’ for the experiment corresponding test Π
Quantum Axiomatics
91
to the test Πi αi . This comes to exchanging ‘yes’ and ‘no’ after the choice of one of the αj is made. The test Πi α i on the contrary consists of exchanging the ‘yes’ and ‘no’ of each of the tests αi , hence before the choice of one of the αj is made. These are the same tests. There is a fundamental problem with the inverse test which is the following. Suppose that α(a) tests the property a, and β(a) also tests the property a, then α
and β in general test completely diﬀerent properties. Let us show this by means of a concrete example. We introduce the test τ , which is the unit test, in the sense that it is a test which gives ‘yes’ as outcome with certainty for each state of the entity. The test τ tests the maximal property I. Obviously τ is a test which gives with certainty outcome ‘no’ for each state of the entity, which means that it tests a property which is never actual. It can be shown that this property which is never actual can only be represented by the minimal property 0, hence τ tests 0. Consider now an arbitrary property a ∈ L, and a test α which tests property a. Let us suppose that α
tests a property b, and that both properties a and b can be actual. Consider the product test α · τ . This is a test which also tests the property a, because indeed α · τ gives with certainty the outcome ‘yes’ iﬀ α gives with certainty the outcome ‘yes’. However α ·τ = α
· τ tests the property 0, and not the property b. Indeed α
· τ gives with certainty the outcome ‘yes’ iﬀ α
gives with certainty ‘yes’ and τ gives with certainty ‘yes’. This is never the case, which proves that it tests the property 0.
DEFINITION 20 Orthogonal states. If p and q are two states of S we will say that p is orthogonal to q, iﬀ there exists a test γ such that γ is true if S is in the state p and γ
is true if S is in the state q. We will denote then p ⊥ q PROPOSITION 21. For p, q, r, s ∈ Σ we have p⊥q p ⊥ q, r ≤ p, s ≤ q p⊥q
q⊥p
(17)
⇒ r⊥s ⇒ p∧q =0
⇒
(18) (19)
Proof. The orthogonality relation is obviously symmetric. If r ≤ p and s ≤ q, and p ⊥ q, and γ is a test such that γ is true if the entity is in state p and γ
is true if the entity is in state q, then we have that γ is true if the entity is in state r and γ
is true if the entity is in state s. This proves that r ⊥ s. DEFINITION 22 Orthogonal properties and states. We say that a state p ∈ Σ is orthogonal to a property a ∈ L iﬀ for every q ∈ Σ such that a ∈ ξ(q) we have p ⊥ q. We denote p ⊥ a. We say that two properties a, b ∈ L are orthogonal iﬀ for every p, q ∈ Σ such that a ∈ ξ(p) and b ∈ ξ(q) we have p ⊥ q. We denote a ⊥ b.
92
Diederik Aerts
PROPOSITION 23. For a, b, c, d ∈ L and p, r ∈ Σ we have p ⊥ a, r ≤ p, c ≤ a ⇒
r⊥c
(20)
a⊥b ⇒ b⊥a a ⊥ b, c ≤ a, d ≤ b ⇒ c ⊥ d
(21) (22)
a⊥b ⇒
3
a∧b=0
(23)
A SET OF AXIOMS
In this section we put forward a set of axioms and derive the consequences for the structure of the state property space of this set of axioms. We make an attempt to introduce all the axioms in a way which is as operational as possible.
3.1 The axiom of property determination The ﬁrst axiom expresses a relation between the states and the properties. We consider two properties a, b ∈ L of the entity S, and suppose that κ(a) = κ(b). This means that each state which make property a actual also makes property b actual, and vice versa. It also means that we cannot distinguish between property a and property b by means of the states of the entity S. Hence, this means that for entity S, property a and property b are equivalent. AXIOM 24 Property determination. We say that the axiom of property determination is satisﬁed for a state property space (Σ, L, ξ) iﬀ for a, b ∈ L we have: κ(a) = κ(b) ⇒ a = b
(24)
DEFINITION 25 Partial order relation. Suppose that we have a set Z. We say that ≤ is a partial order relation on Z iﬀ ≤ is a preorder relation for which equivalent elements are equal. If axiom 24 is satisﬁed for a state property space (Σ, L, ξ), the preorder relation on the set of properties L is then a partial order relation. THEOREM 26. Consider a state property space (Σ, L, ξ) for an entity S for which axiom 24 is satisﬁed. The ‘property implication’ on L is then a partial order relation on L. Proof. Suppose that axiom 24 is satisﬁed for (Σ, L, ξ), and consider a, b ∈ L such that a ≤ b and b ≤ a. Then we have κ(a) ⊆ κ(b) and κ(b) ⊆ κ(a), and hence κ(a) = κ(b). As a consequence, because of axiom 24, we have a = b. This proves that ≤ is a partial order relation on L. The two archetypical examples we have introduced both satisfy the axiom of property determination. Indeed, consider the ﬁrst example of classical mechanics. Since
Quantum Axiomatics
93
κ is the identity, we have for A, B ∈ P(Ω) that κ(A) = κ(B) implies A = B. For the second example of quantum mechanics, consider A, B ∈ L(H), and suppose that κH (A) = κH (B). Consider the vector x ∈ H such that Ax = x. Since κH (A) = κH (B) this implies that Bx = x. This proves that AB = A. In an analoguous way we prove that AB = B, and hence A = B.
3.2
The axiom of completeness
We want to be able to distinguish between properties that are not necessarily of the type that they are meet properties, and between properties which are meet properties. In [Aerts, 1981; Aerts, 1982] we have introduced in this way a subset of properties T ⊆ L, and called it a ‘generating set of properties’ for the state property space (Σ, L, ξ). AXIOM 27 Property completeness. We say that the axiom of property completeness is satisﬁed for a state property space (Σ, L, ξ) iﬀ there exists a subset T ⊆ L such that for each (ai )i ⊆ T there exists a ∈ L such that κ(a) = κ(ai ) (25) i
and, each property a ∈ L is of this form, i.e. for a ∈ L there exists a subset (ai )i ⊆ T such that (25) is satisﬁed. We call T ⊆ L a generating set of properties of the state property space (Σ, L, ξ), and call the property a of (25) a meet of the set of properties (ai )i , and denote it by a= ai (26) i
The following deﬁnition and proposition explain why we have chosen to call axiom 27 the axiom of completeness. DEFINITION 28 Complete preordered set. Suppose that Z, ≤ is a preordered set. We say that Z is a complete preordered set iﬀ for each subset of elements of Z there exists an inﬁmum and a supremum in Z. PROPOSITION 29. Consider a state property space (Σ, L, ξ) for which axiom 27 is satisﬁed. Then L, ≤ is a complete preordered set, and if for a subset (ai )i ⊆ L we denote an inﬁmum of (ai )i by ∧i ai we have κ( ai ) = κ(ai ) (27) i
i
Proof. Consider an arbitrary set (ai )i ⊆ L of properties. We need to prove that there exists an inﬁmum and a supremum in L for this set of properties (ai )i . From axiom 27 we know that for each ai there is a set (biji )ji ⊆ T , such that ai = ∧ji biji , and κ(ai ) = ∩ji κ(biji ). From the same axiom 27 follows that for the subset
94
Diederik Aerts
(biji )iji ⊆ T there exists a property a ∈ L such that κ(a) = ∩i ∩ji κ(biji ) = ∩i κ(ai ). Let us prove that a is an inﬁmum for the set (ai )i ⊆ L. Since κ(a) = ∩i κ(ai ) we have κ(a) ⊆ κ(aj ) ∀j, and hence a ≤ aj ∀ j, which proves that a is a lower bound for (ai )i . Consider x ∈ L such that x ≤ aj ∀j. This implies that κ(x) ⊆ κ(aj ) ∀j, and hence κ(x) ⊆ ∩i κ(ai ) = κ(a). From this follows that x ≤ a, which proves that a is a greatest lower bound or inﬁmum. It is a consequence that for each subset (ai )i ⊆ L, there exists also a supremum in L, let us denote it by ∨i ai . It is given by
ai = x (28) i
x∈L,ai ≤x∀i
This proves that L, ≤ is a complete preordered set.
Remark that the supremum for elements of L, although it exists, has no simple operational meaning. DEFINITION 30 Complete lattice. Suppose that Z, ≤ is a partially ordered set. We say that Z, ≤, ∧, ∨ is a complete lattice iﬀ for each subset (xi )i ⊆ Z of elements of Z there exists an inﬁmum ∧i xi ∈ Z and a supremum ∨i xi ∈ Z in Z. A complete lattice has a minimal element which we denote 0, and which is the inﬁmum of all elements of Z, and a maximal element, which we denote I, and which is the supremum of all elements of Z. THEOREM 31. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed. Then L, ≤, ∧, ∨ is a complete lattice. For I the maximum of L, (ai )i ⊆ L and p ∈ Σ we have: κ(I) = Σ i κ(ai ) = κ( i ai ) i κ(ai ) ⊆ κ( i ai ) ai ∈ ξ(p) ∀i ⇔ i ai ∈ ξ(p)
(29) (30) (31) (32)
Proof. From proposition 26 follows that L, ≤ is a partially ordered set, and from proposition 29 follows that L, ≤, ∧, ∨ is a complete lattice. We have κ(I) ⊆ Σ. For an arbitrary p ∈ Σ consider ξ(p). Since a ≤ I ∀a ∈ ξ(p), we have I ∈ ξ(p), and hence p ∈ κ(I). This proves that Σ ⊆ κ(I). As a consequence we have κ(I) = Σ. From (27) of proposition 29 follows (30). Let us prove (31). Since ∨i ai is a supremum of (ai )i we have aj ≤ ∨i ai ∀ j. Hence κ(aj ) ⊆ κ(∨i ai ) ∀ j. This proves that ∪i κ(ai ) ⊆ κ(∨i ai ). Suppose that ai ∈ ξ(p) ∀i, then p ∈ κ(ai ) ∀i, and hence p ∈ ∩i κ(ai ) = κ(∧i ai ). From this follows that ∧i ai ∈ ξ(p), and hence we have proven one of the implications of (32). Let us prove the other one, and hence suppose that ∧i ai ∈ ξ(p). From this follows that p ∈ κ(∧i ai ) = ∩i κ(ai ). As a consequence we have p ∈ κ(ai ) ∀i, and hence ai ∈ ξ(p) ∀i.
Quantum Axiomatics
95
If axiom 24 and 27 are satisﬁed for a state property space (Σ, L, ξ), and hence the set of properties L is a complete lattice, we can represent the states by means of properties. DEFINITION 32 Property state. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed. For each state p ∈ Σ we deﬁne the ‘property state’ corresponding to p as the property a (33) s(p) = a∈ξ(p)
PROPOSITION 33. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed. For p, q ∈ Σ and a ∈ L we have: s(p) ∈ ξ(p) a ∈ ξ(p) ⇔ s(p) ≤ a p ≤ q ⇔ s(p) ≤ s(q)
(34) (35) (36)
ξ(p) = {a  a ∈ L, s(p) ≤ a} = [s(p), I]
(37)
Proof. That s(p) ∈ ξ(p) follows directly from (32). Suppose a ∈ ξ(p) then ∧a∈ξ(p) a ≤ a and hence s(p) ≤ a. We have that s(p) ∈ ξ(p), and if s(p) ≤ a, from proposition 7 follows then that a ∈ ξ(p). Suppose that p ≤ q. Then we have ξ(q) ⊆ ξ(p). From this follows that s(p) = ∧a∈ξ(p) a ≤ ∧a∈ξ(q) a = s(q). Suppose now that s(p) ≤ s(q). Take a ∈ ξ(q), then we have s(q) ≤ a. Hence also s(p) ≤ a. But this implies that a ∈ ξ(p). Hence this shows that ξ(q) ⊆ ξ(p) and as a consequence we have p ≤ q. Consider b ∈ [s(p), I]. This means that s(p) ≤ b, and hence b ∈ ξ(p). Consider now b ∈ ξ(p). Then s(p) ≤ b and hence b ∈ [s(p), I]. For a state property space satisfying axioms 24 and 27 we can prove that the set of property states is a full set for the complete lattice L. THEOREM 34 Full set of property states. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed. For a ∈ L we have (38) κ(a) = a∈ξ(p) κ(s(p)) (39) a = a∈ξ(p) s(p) Proof. From (35) follows that if a ∈ ξ(p) we have s(p) ≤ a and hence κ(s(p)) ⊆ κ(a). This proves that ∪a∈ξ(p) κ(s(p)) ⊆ κ(a). From (34) follows that for p ∈ Σ we have p ∈ κ(s(p)) and hence {p} ⊆ κ(s(p)). This proves that κ(a) = ∪p∈κ(a) {p} ⊆ ∪p∈κ(a) κ(s(p)) = ∪a∈ξ(p) κ(s(p)). From (35) we have a ∈ ξ(p) then s(p) ≤ a. This proves that ∨a∈ξ(p) s(p) ≤ a. Using (31) we have κ(a) = ∪a∈ξ(p) κ(s(p)) ⊆ κ(∨a∈ξ(p) s(p)). This proves that a ≤ ∨a∈ξ(p) s(p). Hence we have proven that a = ∨a∈ξ(p) s(p).
96
Diederik Aerts
The two archetypical examples of classical mechanics and quantum mechanics satisfy the axiom of completeness. Consider the state property space (Ω, P(Ω), ξΩ ) of a classical mechanical physical system with state space Ω. Consider a set of properties (Ai )i ⊆ P(Ω) of the classical mechanical system. The property A = ∩i Ai makes axiom 27 to be satisﬁed. Indeed, consider an arbitrary state p ∈ Ω. We have ∩i Ai ∈ ξΩ (p) ⇔ p ∈ ∩i Ai ⇔ p ∈ Ai ∀i ⇔ Ai ∈ ξΩ (p) ∀i. From (30) follows that axiom 27 is satisﬁed. Next, consider the state property space (Σ(H), L(H), ξH ) corresponding to a quantum mechanical physical system described by means of a complex Hilbert space H. Consider a set of properties (Ai )i ⊆ L(H) of the quantum mechanical physical system. The property ∩i Ai ∈ L(H) makes axiom 27 to be satisﬁed. Indeed, consider an arbitrary state x ∈ Σ(H). We have ∩i Ai ∈ ξH (P ) ⇔ (∩i Ai )x = x ⇔ Ai x = x ∀i ⇔ Ai ∈ ξH (P ) ∀i. From (30) follows that axiom 27 is satisﬁed.
3.3 Ortho tests We have come to the point where we will introduce the ﬁrst operational element which is speciﬁcally quantum, in the sense that it does not necessarily correspond with our intuition about reality. We will suppose that a special type test exists, which we call an ortho test. DEFINITION 35 Ortho test. A test α is called an ortho test if it is such that if the physical entity is in a state p ⊥ a, where a is a property tested by α, then α
is true, and if the physical entity is in state q ⊥ b, where b is a property tested by α
, then α is true. PROPOSITION 36. Consider a test α. If α is an ortho test then α
is an ortho test. Proof. Follows directly from the deﬁnition.
We can see immediately that ortho tests are special types of test because of the next proposition, where we prove that a product test is never an ortho test, except when it is a trivial product test of equivalent tests. PROPOSITION 37. Consider a set of tests (αi )i . The product test Πi αi is an ortho test iﬀ αj is an ortho test for each j, and αj ≈ αk for each j, k. Proof. Suppose that Πi αi is an ortho test, and let us call ai a property tested by αi , and hence ∧i ai a property tested by Πi αi . Consider an arbitrary αj of the set (αi )i , and a state p such that p ⊥ aj . From (20) follows that p ⊥ ∧i ai . Since
i αi = Πi α
i is Πi αi is an ortho test, we have that whenever p ⊥ ∧i ai the test Π
j is true. Hence, we have true. This means that α
k is true for all k. Hence α
j is true, which is one of the proven that if p is a state orthogonal to aj , then α necessary conditions for αj to be an ortho test. Let us proceed proving the other. Suppose that (bi )i is a set of properties such that each bi is a property tested by
Quantum Axiomatics
97
α
i , and hence ∧i bi is a property tested by Πi α
i . Let us consider α
j which tests bj . Consider a state q such that q ⊥ bj . From (20) follows that q ⊥ ∧i bi , and since
iα
i is an ortho test, we have that Π
i = Παi is true. This implies that αi is Πi α true for all i, and hence αj is true. This proves that αj is an ortho test. Since we had chosen j arbitrary, this proves that all tests αi are ortho tests. Let us next prove that all test are equivalent. Consider αj and suppose that the entity is in state p such that αj is true. From this follows that p ⊥ bj and hence p ⊥ ∧i bi .
i is an ortho test, this implies that Πi αi is true, and hence αk is true for Since Πi α all k. Hence we have proven that αj ≤ αk for all j and k, and as a consequence all the tests are equivalent. Proposition 37 tells us something important. Ortho tests are the test which exist commonly in quantum mechanics. This proposition proves that such an ortho test cannot be generated in a non trivial way by the product test mechanism. This means that if a property can be tested by an ortho test, hence this ortho test is speciﬁc for this property, even if this property is a meet property. It is not a product of other tests, except in a trivial way, when these other tests are also ortho tests testing this same property. But ortho test and the corresponding ortho properties have other unsuspected features. PROPOSITION 38. Consider two tests α, β such that α is an ortho test. We have α ≤ β ⇒ β ≤ α
(40)
Proof. Suppose that α ≤ β and that the state p of the entity is such that β is true. This means that p ⊥ b where b is a property tested by β. Suppose that a is a property tested by α, then we have a ≤ b, and hence p ⊥ a. Since α is an ortho test, it follows that α
is true. Hence we have proven that β ≤ α
.
3.4
The axiom of orthocomplementation
There is a speciﬁc structure, namely the structure of an orthocomplementation, which has been identiﬁed mathematically in the formalism of quantum mechanics, and this structure has played an important role in the mathematical axiomatization, for example the one worked out in [Piron, 1964; Piron, 1976]. With the notion of ortho test we introduce an orthocomplementation in an operational way. Lets ﬁrst deﬁne what an orthocomplementation on a partially ordered set with minimum is. DEFINITION 39 Orthocomplementation. Suppose that we have a set Z with a partialorder ≤ and a smallest element 0. A bijective map ⊥ : Z → Z is an orthocomplementation if for x, y ∈ Z we have (x⊥ )⊥ = x x ≤ y ⇒ y ⊥ ≤ x⊥
(41) (42)
0 is the inﬁmum of x and x⊥
(43)
98
Diederik Aerts
The ortho test satisﬁes (41), (42) and (43). Of course, the set of tests corresponding to a physical entity is not a partially ordered set, but only a preordered set. But the set of properties, if axiom 24 of property determination is satisﬁed, is a partially ordered set. Hence our aim is to make operational steps such that on the set of properties an orthocomplementation arises. This makes us introduce the following deﬁnition for an ortho property. DEFINITION 40 Ortho property. Consider a state property space (Σ, L, ξ). We say that a ∈ L is an ortho property if there exists an ortho test testing a. If α is
. the ortho test testing a, we denote by a⊥ the property tested by α Let us introduce the following deﬁnition. DEFINITION 41 Orthogonal set. For a subset of states A ⊆ Σ we deﬁne the orthogonal A⊥ of this subset A⊥ = {p  p ⊥ q ∀ q ∈ A}
(44)
PROPOSITION 42. Consider a state property space (Σ, L, ξ). If a ∈ L is an ortho property then we have κ(a)⊥ = κ(a⊥ )
(45)
Proof. Suppose that a ∈ L is an ortho property. This means that p ⊥ a ⇔ a⊥ ∈ ξ(p). Hence p ∈ κ(a)⊥ ⇔ p ∈ κ(a⊥ ). And as a consequence we have κ(a)⊥ = κ(a⊥ ). AXIOM 43 Orthocomplementation. Consider a state property space (Σ, L, ξ). For each property a ∈ L there exists an ortho test α testing this property. THEOREM 44. Consider a state property space (Σ, L, ξ) and suppose that axiom 24, axiom 27 and axiom 43 are satisﬁed. For a ∈ L and α an ortho test testing a let us denote the property tested by α
by a⊥ . For a, b ∈ L and p, q ∈ Σ we then have
which proves that
⊥
(a⊥ )⊥ = a a ≤ b ⇒ b⊥ ≤ a⊥
(46) (47)
a ∧ a⊥ = 0
(48)
: L → L is an orthocomplementation.
Proof. First we remark that if α and β are ortho tests testing property a, and
This shows that ⊥ is a function. Consider hence α ≈ β, we have that α
≈ β. now a ∈ L and α an ortho test testing a. Then α
is an ortho test testing a⊥ . We
= α. Hence (a⊥ )⊥ = a. Consider have that (a⊥ )⊥ is the property tested by α a, b ∈ L such that a ≤ b, and α and β ortho tests testing respectively a and b. Hence α ≤ β. From this follows that β ≤ α
, and hence b⊥ ≤ a⊥ . Consider a ∈ L and α an ortho test testing a. Hence α
tests a⊥ . The inﬁmum property a ∧ a⊥ in ⊥ L is tested by α · α
and hence a ∧ a = 0.
Quantum Axiomatics
99
PROPOSITION 45. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. We have for a, b ∈ L, (ai )i ⊆ L and p, q ∈ L ( i ai )⊥ = i a⊥ (49) i⊥ ⊥ ( i ai ) = i ai (50) 0⊥ = I I ⊥ = 0 (51) ⊥ a∨a =I (52) p ⊥ q ⇔ ∃ c ∈ L such that c ∈ ξ(p) and c⊥ ∈ ξ(q)
(53)
Proof. Let us prove (49) and (50). We have ∧i ai ≤ aj ∀j, which implies that ⊥ ⊥ ⊥ ⊥ ⊥ a⊥ j ≤ (∧i ai ) ∀j, and hence ∨i ai ≤ (∧i ai ) . This also implies ∨i (ai ) = ∨i ai ≤ ⊥ ⊥ ⊥ ⊥ ⊥ (∧i a⊥ = ∧i a⊥ i ) . Hence ((∧i ai ) ) i ≤ (∨i ai ) . We also have aj ≤ ∨i ai ∀j, ⊥ ⊥ which implies that (∨i ai ) ≤ aj ∀j. Hence (∨i ai )⊥ ≤ ∧i a⊥ i . This also implies ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ that (∨i a⊥ ) ≤ ∧ (a ) = ∧ a . Hence (∧ a ) ≤ ((∨ a ) ) = ∨i a⊥ i i i i i i i i i i . Consider ⊥ ⊥ ⊥ a ∈ L, then 0 ≤ a , and hence a ≤ 0 . This proves that 0 is a maximal element of L, and hence 0⊥ = I. In a analogous way we prove that I ⊥ = 0. We have I = 0⊥ = (a ∧ a⊥ )⊥ = a⊥ ∨ a which proves (52). To prove (53) we remark that if c ∈ L is such that c ∈ ξ(p) and c⊥ ∈ ξ(q), we have p ∈ κ(c) and q ∈ κ(c⊥ ). Since κ(c⊥ ) = κ(c)⊥ we have p ⊥ q. In foregoing work on quantum axiomatics we have worked most of the time with state property systems [Aerts et al., 1999; Aerts and van Steirteghem, 2000; Aerts et al., 2000; Aerts and Deses, 2002; Aerts and van Valckenborgh, 2002; Aerts and Deses, 2005; Aerts et al., in press; Aerts and Pulmannova, 2006]. DEFINITION 46 State property system. We say that (Σ, L, ξ) is a stateproperty system if (Σ, ≤) is a preordered set, (L, ≤, ∧, ∨) is a complete lattice with the greatest element I and the smallest element 0, and ξ is a function ξ : Σ → P(L)
(54)
such that for p ∈ Σ and (ai )i ⊆ L, we have I ∈ ξ(p),
(55)
0∈ / ξ(p), ai ∈ ξ(p) ∀i ⇔ ∧i ai ∈ ξ(p) (for an arbitrary set of indices)
(56) (57)
and for p, q ∈ Σ and a, b ∈ L we have p ≤ q ⇔ ξ(q) ⊆ ξ(p) a ≤ b ⇔ ∀r ∈ Σ : a ∈ ξ(r) ⇒ b ∈ ξ(r) Elements of Σ are called states, elements of L are called properties.
(58) (59)
100
Diederik Aerts
A state property space for which the three axioms which we have formulated are satisﬁed is a state property system. THEOREM 47. A state property space for which axioms 24, 27 and 43 are satisﬁed is a state property system. Proof. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. From (7) follows that (Σ, ≤) is a preordered set. In theorem 31 we prove that L, ≤, ∧, ∨ is a complete lattice, and from (29) follows that I ∈ ξ(p) ∀ p ∈ Σ. We have a ∧ a⊥ = 0 and hence κ(0) = κ(a) ∩ κ(a)⊥ = ∅. This proves 0 ∈ ξ(p) ∀ p ∈ Σ. From (25) and (26) of axiom 27 follows (57), and (58) and (59) follows respectively from (7) and (6).
4 MORPHISMS We derive the notion of morphism from a covariance situation. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ), describing respectively entities S and S . We will arrive at the notion of morphism by analyzing the situation where the entity S is a subentity of the entity S . In that case, the following three natural requirements should be satisﬁed: i) If the entity S is in a state p then the state m(p ) of S is determined. This deﬁnes a function m from the set of states of S to the set of states of S; ii) If we consider a property a of the entity S, then to a corresponds a property n(a) of the ‘bigger’ entity S . This deﬁnes a function n from the set of properties of S to the set of properties of S ; iii) We want a and n(a) to be two descriptions of the ‘same’ property of S, once considered as an entity on itself, once as a subentity of S . In other words we want a and n(a) to be actual at once. This means that for a state p of S (and a corresponding state m(p ) of S) we want the following ‘covariance principle’ to hold: a ∈ ξ(m(p )) ⇔ n(a) ∈ ξ (p )
(60)
We are now ready to present a formal deﬁnition of a morphism of state property spaces. DEFINITION 48 Morphism. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ). We say that (m, n) : (Σ , L , ξ ) −→ (Σ, L, ξ)
(61)
is a ‘morphism’ (of state property spaces) if m is a function: m : Σ → Σ
(62)
Quantum Axiomatics
101
and n is a function: n : L → L
(63)
such that for a ∈ L and p ∈ Σ the following holds: a ∈ ξ(m(p )) ⇔ n(a) ∈ ξ (p )
(64)
PROPOSITION 49. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ), and functions m : Σ → Σ
n : L → L
(65)
We have that (m, n) : (Σ , L , ξ ) −→ (Σ, L, ξ)
(66)
is a morphism iﬀ for a ∈ L, and p ∈ Σ m(p ) ∈ κ(a) ⇔ p ∈ κ (n(a))
(67)
Proof. Let us prove (64) to show that (m, n) is a morphism. We have a ∈ ξ(m(p )) ⇔ m(p ) ∈ κ(a) ⇔ p ∈ κ (n(a)) ⇔ n(a) ∈ ξ (p ). The next theorem gives some properties of morphisms. THEOREM 50. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ) connected by a morphism (m, n) : (Σ , L , ξ ) → (Σ, L, ξ). For p , q ∈ Σ and a, b ∈ L we have: p ≤ q ⇒ m(p ) ≤ m(q )
(68)
a ≤ b ⇒ n(a) ≤ n(b)
(69)
Proof. Suppose that p ≤ q . We then have ξ (q ) ⊆ ξ (p ). Consider a ∈ ξ(m(q )), then (64) implies that n(a) ∈ ξ (q ), and hence n(a) ∈ ξ (p ), which means that a ∈ ξ(m(p )). As a consequence we have ξ(m(q )) ⊆ ξ(m(p )), whence m(p ) ≤ m(q ). Next consider a ≤ b. We then have κ(a) ⊆ κ(b). Let r ∈ Σ be such that n(a) ∈ ξ (r ). Then we have a ∈ ξ(m(r )) and hence m(r ) ∈ κ(a) ⊆ κ(b). This yields b ∈ ξ(m(r )). From this follows that n(b) ∈ ξ (r ). So we have shown that n(a) ≤ n(b). THEOREM 51. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ) connected by a morphism (m, n) : (Σ , L , ξ ) → (Σ, L, ξ) for which the axioms 24, 27 are satisﬁed. For I and I the maximum of respectively L and L and (ai )i ⊆ L we have: n(I) = I n( i ai ) = i n(ai )
(70) (71)
102
Diederik Aerts
Proof. We clearly have n(I) ≤ I . Hence remains to show that I ≤ n(I). Consider r ∈ Σ = κ (I ), then m(r ) ∈ Σ = κ(I). From (67) follows that r ∈ κ (n(I)). This proves that κ (I ) ⊆ κ (n(I)), and hence I ≤ n(I). Hence we have proven that n(I) = I . From ∧i ai ≤ aj ∀j we obtain n(∧i ai ) ≤ n(aj ) ∀j. This yields n(∧i ai ) ≤ ∧i n(ai ). We still have to show that ∧i n(ai ) ≤ n(∧i ai ). Let r ∈ Σ be such that r ∈ κ (∧i n(ai )). Using (30) we have r ∈ ∩i κ (n(ai )), and hence r ∈ κ (n(ai )) ∀i. From (67) follows that this implies that m(r ) ∈ κ(ai ) ∀i, and hence m(r ) ∈ ∩i κ(ai ) = κ(∧i ai ) using again (30). From (67) this implies that r ∈ κ (n(∧i ai )). Hence we have shown that κ (∧i n(ai )) ⊆ κ (n(∧i ai )), and it follows that ∧i n(ai ) ≤ n(∧i ai ). Hence we have proven that n(∧i ai ) = ∧i n(ai ). DEFINITION 52 Orthomorphism. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ) for which the axioms 24, 27, 43 are satisﬁed. A morphism (m, n) : (Σ , L , ξ ) → (Σ, L, ξ) is an orthomorphism iﬀ m(p ) ∈ κ(a⊥ ) ⇔ p ∈ κ (n(a)⊥ )
(72)
THEOREM 53. Consider two state property spaces (Σ, L, ξ) and (Σ , L , ξ ) for which the axioms 24, 27, 43 are satisﬁed and connected by an orthomorphism (m, n) : (Σ , L , ξ ) → (Σ, L, ξ). For 0 and 0 being the minimal elements of L and L respectively, a ⊆ T and p , q ∈ Σ we have: n(a⊥ ) = n(a)⊥
n(0) = 0 m(p ) ⊥ m(q ) ⇒ p ⊥ q
(73) (74) (75)
Proof.p ∈ κ (n(a)⊥ ) ⇔ m(p ) ∈ κ(a⊥ ) ⇔ p ∈ κ (n(a⊥ )). This proves that κ (n(a)⊥ ) = κ (n(a⊥ )), and hence n(a)⊥ = n(a⊥ ). We have 0 = a ∧ a⊥ , and hence n(0) = n(a ∧ a⊥ ) = n(a) ∧ n(a⊥ ) = n(a) ∧ n(a)⊥ = 0 . Consider p , q ∈ Σ such that m(p ) ⊥ m(q ). This means that there exists a ∈ L such that m(p ) ∈ κ(a) and m(q ) ∈ κ(a⊥ ). Hence p ∈ κ (n(a)) and q ∈ κ (n(a⊥ )) = κ (n(a)⊥ ). This proves that p ⊥ q . 5
DECOMPOSITION OF A STATE PROPERTY SPACE
In this section we introduce the notion of classical test, classical property and classical state. This will lead us to elaborate a decomposition theorem for a state property space into non classical components over a classical state space.
5.1 The classical state property space In this section we identify the classical part of an entity S. We start by introducing the notion of classical test. The basic idea for a classical test is that it is a test
Quantum Axiomatics
103
which does not contain any indeterminism. This means that for such a test the outcome ‘yes’ is certain or the outcome ‘no’ is certain for each state of the physical entity. Hence we put forward the following deﬁnition DEFINITION 54 Classical test. A test α is a classical test if for any arbitrary state p of the physical entity we have α is true or α
is true. For the product of classical tests we can prove the following PROPOSITION 55. If (αi )i is a set of tests, then Πi αi is a classical test iﬀ each of the αj are classical tests and αj ≈ αk for all j, k. Proof. Suppose that Πi αi is a classical test. Consider αj and a state p such that αj is not true if the entity is in state p. Then Πi α is not true and since Πi αi is a
i αi = Π
αi is true. But then α
i is true for all i, and classical test, we have that Π hence α
j is true. This proves that αj is a classical test. Since we had chosen j arbitrary, we can conclude that αi is a classical test for all i. Suppose now that
i is not true. But then, since
j is not true, and hence Π
αi = Πα αj is true. Then α
Παi is a classical test, we have that Πi αi is true, and hence αk is true for all k. Hence we have proven that αj ≤ αk for all k. Hence αj ≈ αk for all j, k. It is easy to see that a classical test is always an ortho test. PROPOSITION 56. If α is a classical test then α is an ortho test. Proof. Suppose that α is a classical test, and consider a state p such that p ⊥ a where a is a property tested by α. This means that a ∈ ξ(p), and hence α
is true for the physical entity in state p. In an analogous way we prove that for q ⊥ a⊥ when a⊥ is a property tested by α
and the physical entity in state q we have α is true. This proves that α is an ortho test. DEFINITION 57 Classical property. A classical property a ∈ L is a property such that there exists a set (αi )i of classical tests αi such that Πi αi tests this property. We denote C the set of all classical properties. A basic classical property a ∈ L is a property such that there exists a classical test α testing this property. We denote K the set of basic classical properties. DEFINITION 58 Classical elements. Consider a state property space for which axioms 24, 27 and 43 are satisﬁed. For p ∈ Σ, we introduce a (76) ω(p) = a∈ξ(p)∩C
(77) and call ω(p) the classical state of the entity S whenever S is in a state p ∈ Σ. The set of all classical states is denoted by Ω. We introduce ξc : Ω
→
C
ω(p) → ξ(p) ∩ C
κc : C
→
P(Ω)
a → {ω(p) a ∈ ξ(p)}
(78) (79)
104
Diederik Aerts
and call κc the classical Cartan map of the state property space (Σ, L, ξ). PROPOSITION 59. Consider a state property space for which axioms 24 and 27 are satisﬁed. For classical states ω(p), ω(q) ∈ Ω, classical property a ∈ C, and states p, q ∈ Σ we have a ∈ ξ(p) ⇔ ω(p) ≤ a ω(p) ≤ ω(q) ⇔ ξc (q) ⊆ ξc (p) p≤q
⇒
ω(p) ≤ ω(q)
(80) (81) (82)
Proof. Suppose that a ∈ C and a ∈ ξ(p). Since ω(p) = ∧a∈ξ(p)∩C a we have ω(p) ≤ a. Suppose now that ω(p) ≤ a. Since ω(p) ∈ ξ(p) we have a ∈ ξ(p). Consider a ∈ ξc (q) = ξ(q) ∩ C and ω(p) ≤ ω(q). This implies that ω(q) ≤ a and hence ω(p) ≤ a. From this follows that a ∈ ξ(p) and hence a ∈ ξ(p) ∩ C = ξc (p). Hence we have proven that ξc (q) ⊆ ξc (p). Suppose now that ξc (q) ⊆ ξc (p), then ω(p) = ∧a∈ξc (p) ≤ ∧a∈ξc (q) = ω(q). Suppose that p ≤ q and hence ξ(q) ⊆ ξ(p). We then have ξc (q) = ξ(q) ∩ C ⊆ ξ(p) ∩ C = ξc (p), and hence ω(p) ≤ ω(q). Let us consider our two physics examples, and see what the notion of classical property and classical state means in these cases. Consider ﬁrst the state property space (Ω, P(Ω), ξΩ ) for a classical physical system, and consider a property A ∈ P(Ω). Take p ∈ Ω, then we have p ∈ A or p ∈ AC . This proves that any arbitrary property A is a classical property for the state property system (Ω, P(Ω), ξΩ ) corresponding to a classical physical system. Clearly, for such a state property system the states coincide with the classical states, which proves that any state is a classical state. Consider now the state property system (Σ(H), L(H), ξH ) corresponding to a quantum physical system, and consider a property A ∈ L(H) such that A = 0 and A = H. In this case we have Aorth = H and Aorth = 0. Take x ∈ A, x = 0 and y ∈ Aorth , y = 0, and consider the vector z = x + y. Then z ∈ A and z ∈ Aorth , and as a consequence z¯ ∈ A and z¯ ∈ Aorth , which proves that A is not a classical property. We have proven that for the state property system corresponding to a quantum physical system the only classical properties are the minimal property and the maximal property. Moreover, the only classical state of the state property system corresponding to a quantum physical system is the classical state corresponding to H itself. This is the state describing the situation ‘the entity is present’. DEFINITION 60 Classical orthogonality relation. Consider a state property space describing a physical entity S for which axioms 24 and 27 are satisﬁed, and classical states ω(p), ω(q) ∈ Ω of this physical entity. We say that ω(p) ⊥c ω(q) if there exists a classical test γ such that γ is true if ω(p) is actual, hence if the
Quantum Axiomatics
105
entity is in classical state ω(p), and γ
is true if ω(q) is actual, hence if the entity is in classical state ω(q). DEFINITION 61 Classical ortho test. Consider a state property space describing a physical entity S for which axioms 24 and 27 are satisﬁed. A classical test α is a classical ortho test if it is such that if the physical entity is in classical state
is true, and if the physical ω(p) ⊥c a, where a is the property tested by α, then α
, then α is true. entity is in state ω(q) ⊥c b, where b is the property tested by α PROPOSITION 62. Consider a state property space for which axioms 24 and 27 are satisﬁed, and classical states ω(p), ω(q) ∈ Ω of this physical entity. We have ω(p) = ω(q) ⇔ ω(p) ⊥c ω(q) ⇔ ω(p) ⊥ ω(q) ⇔ ω(p) ∈ ξ(q) ⇔ q ⊥ ω(p)
(83)
Proof. If ω(p) ⊥c ω(q) then obviously ω(p) = ω(q). Suppose now that ω(p) = ω(q). Since ω(p) and ω(q) as classical states are also both classical properties there exist (αi )i and (βj )j where αi and βj are classical tests for all i, j and such that Πi αi tests ω(p) and Πj βj tests ω(q). If ω(p) = ω(q) this can mean that ω(p) ≤ ω(q) or that ω(q) ≤ ω(p). Suppose we have that ω(p) ≤ ω(q), and suppose that ω(p) is actual. Since in this case ω(q) is not actual there is at least one βj which is not true. But then β j is true. If ω(q) is actual we have that βj is true. Hence we have proven that ω(p) ⊥c ω(q). Analogously we prove that ω(p) ⊥c ω(q) if ω(q) ≤ ω(p). Suppose that ω(p) ⊥c ω(q) and let γ be the test which is true if ω(p) is actual such that γ
is true if ω(q) is actual. Consider states r, s ∈ Σ such that ω(p) ∈ ξ(r) and ω(q) ∈ ξ(s). If c is the property tested by γ and d the property tested by γ
we have ω(p) ≤ c and ω(q) ≤ d. Hence c ∈ ξ(r) and d ∈ ξ(s). This proves that r ⊥ s, and hence ω(p) ⊥ ω(q). Suppose now that ω(p) ⊥ ω(q). Then certainly ω(p) = ω(q) and hence ω(p) ⊥c ω(q). Suppose that ω(p) ∈ ξ(q). Then ω(q) ≤ ω(p), and as a consequence we have ω(q) = ω(p). Hence ω(p) ⊥ ω(q). If ω(p) ⊥ ω(q) then ω(q) ≤ ω(p) and hence ω(p) ∈ ξ(q). PROPOSITION 63. Consider a state property space for which axioms 24 and 27 are satisﬁed. A classical test α is a classical ortho test. Proof. Consider a classical test α such that a is the property tested by α and b is the property tested by α
, and suppose we have ω(p) ⊥c a. Consider q ∈ Σ such that a ∈ ξ(q). Then we have ω(q) ≤ a, and hence ω(p) ⊥c ω(q). As a consequence we have ω(p) ⊥ q and hence p ⊥ q. This implies that b ∈ ξ(p), and hence ω(p) ≤ b. In an analogous way we show that ω(q) ≤ a if ω(q) ⊥c b. This proves that α is a classical ortho test. Suppose we consider a state property space (Σ, L, ξ) describing a physical entity S for which axioms 24, 27 and 43 are satisﬁed. We wonder whether (Ω, C, ξc ) is a state property space satisfying 24, 27 and 43. If this is the case we can consider
106
Diederik Aerts
(Ω, C, ξc ) as the state property space describing the classical aspects of the physical entity S. THEOREM 64. Consider a state property space (Σ, L, ξ) describing a physical entity S for which axioms 24, 27 and 43 are satisﬁed. Consider ω(p), ω(q) ∈ Ω and a, b ∈ C. We have a ∈ ξ(p) ⇔ ω(p) ≤ a ω(p) ≤ ω(q) ⇔ ξc (ω(q)) ⊆ ξc (ω(p))
(84) (85)
ω(p) ∈ κc (a) ⇔ a ∈ ξc (ω(p)) a ≤ b ⇔ κc (a) ⊆ κc (b)
(86) (87)
κc (a) = κc (b) ⇒ a = b κc (∧i ai ) = ∩i κc (ai )
(88) (89)
κ(a⊥ ) = Σ \ κ(a) κc (a⊥c ) = Ω \ κc (a) There exists a classical test α testing a, hence C = K
(90) (91) (92)
and (Ω, C, ξc ) is a state property space satisfying axioms 24, 27 and 43. Proof. Suppose that a ∈ ξ(p). Since ω(p) = ∧a∈ξ(p)∩C a we have ω(p) ≤ a. Suppose now that ω(p) ≤ a. Since ω(p) ∈ ξ(p) we have a ∈ ξ(p). This proves (84). Suppose that ω(p) ≤ ω(q) and consider a ∈ ξc (ω(q)) and hence ω(q) ≤ a. Then we have ω(p) ≤ a and hence a ∈ ξc (ω(p)). This proves that ξc (ω(q)) ⊆ ξc (ω(p)). Suppose now that ξc (ω(q)) ⊆ ξc (ω(p)), and hence ξ(q) ∩ C ⊆ ξ(p) ∩ C. Then we have ω(p) = ∧a∈ξ(p)∩C a ≤ ∧a∈ξ(q)∩C a = ω(q). This proves (85). Suppose that ω(p) ∈ κc (a), then a ∈ ξ(p) and hence ω(p) ≤ a which shows that a ∈ ξc (ω(p)). Contrary, suppose that a ∈ ξc (ω(p)) = ξ(p) ∩ C. Then we have a ∈ ξ(p), and hence ω(p) ∈ κc (a). This proves (86). Suppose that a ≤ b and consider ω(p) ∈ κc (a). We then have a ∈ ξ(p) and hence b ∈ ξ(p). From this follows that ω(p) ∈ κc (b). Hence we have proven that κc (a) ⊆ κc (b). Suppose now that κc (a) ⊆ κc (b) and suppose we have p ∈ Σ such that a ∈ ξ(p). This means that ω(p) ∈ κc (a), and hence ω(p) ∈ κc (b), which implies that b ∈ ξ(p). Hence we have proven that a ≤ b. This proves (87). Suppose that κc (a) = κc (b) and consider p ∈ κ(a). Then we have a ∈ ξ(p) and hence ω(p) ∈ κc (a). From this follows that ω(p) ∈ κc (b), and hence b ∈ ξ(p). As a consequence we have p ∈ κ(b). This means that we have proven that κ(a) ⊆ κ(b). Analogously we prove that κ(b) ⊆ κ(a). Since axiom 24 is satisﬁed for the state property space (Σ, L, ξ) we have a = b. This proves (88). Consider (ai )i ⊆ K ⊂ T . Since axiom 27 is satisﬁed for (Σ, L, ξ) we have a property ∧i ai ∈ L such that κ(∧i ai ) = ∧i κ(ai ). We have ∧i ai ∈ C since ai ∈ C ∀ i. This means that we have the following ω(p) ∈ κc (∧i ai ) ⇔ p ∈ κ(∧i ai ) ⇔ p ∈ ∩i κ(ai ) ⇔ p ∈ κ(ai ) ∀ i ⇔ ω(p) ∈ κc (ai ) ∀ i ⇔ ω(p) ∈ ∩i κc (ai ). This means that κc (∧i ai ) = ∩i κc (ai ), and hence we have proven (89). Consider p ∈ Σ and p ∈ κ(a). Then we have ω(p) ≤ a and hence ω(p) = ω(q) ∀ ω(q) ≤ a. From this follows that ω(p) ⊥c ω(q) ∀ ω(q) ≤ a, and hence ω(p) ⊥ ω(q) ∀ ω(q) ≤ a.
Quantum Axiomatics
107
As a consequence we have that p ⊥ q ∀ q ∈ κ(a), and hence p ∈ κ(a⊥ ). This proves that Σ \ κ(a) ⊆ κ(a⊥ ). We obviously have that κ(a⊥ ) ⊆ Σ \ κ(a). Hence we have proven (90). Consider ω(p) ∈ Ω and ω(p) ∈ κc (a). This means that ω(p) = ω(q) ∀ ω(q) ≤ κc (a), and hence ω(p) ⊥c ω(q) ∀ ω(q) ≤ κc (a). As a consequence we have ω(p) ≤ κc (a)⊥c . This proves that Ω \ κc (a) ⊆ κc (a)⊥c . We also have that κc (a)⊥c ⊆ Ω \ κc (a). Hence we have proven (91). Remark that (85) shows that ξc deﬁnes the preorder relation on the set of states Ω in a way which is necessary for (Ω, C, ξc ) to be a state property space, while (86) shows that the classical Cartan map is indeed the Galois inverse of the function ξc which has to be the case if (Ω, C, ξc ) is a state property space. With (87) we prove that the classical Cartan map indeed deﬁnes the preorder relation on the set of properties. Hence we have proven that (Ω, C, ξc ) is a state property space. That axiom 24 is satisﬁed for the state property space (Ω, C, ξc ) is proven by (88). Consider now an arbitrary a ∈ C. From the deﬁnition of C follows that there exists (ai )i ⊆ K such that for an arbitrary p ∈ Σ we have ai ∈ ξ(p) ∀ i ⇔ a ∈ ξ(p). Hence p ∈ κ(ai ) ∀ i ⇔ p ∈ κ(a). This means that κ(a) = ∩i κ(ai ). This proves that axiom 27 is satisﬁed for the state property space (Ω, C, ξc ). Let us prove now that axiom 43 is satisﬁed for (Ω, C, ξc ). Consider a ∈ C. From (90) we know that κ(a⊥ ) = Σ \ κ(a). Since axiom 43 is satisﬁed for (Σ, L, ξ) there exists an ortho test α, such that α tests a and α
tests a⊥ . Since for p ∈ Σ we have p ∈ κ(a), and then α is true, or p ∈ κ(a⊥ ) and then α
is true, it follows that α is a classical test. This proves (92). Using proposition 63 it follows that α is a classical ortho test, which proves that axiom 43 is satisﬁed for (Ω, C, ξc ). PROPOSITION 65. Suppose that (Σ, L, ξ) is a state property space describing an entity S for which axioms 24, 27 and 43 are satisﬁed. For p ∈ Σ and a ∈ L we have a ∈ C ⇔ κ(a) ∪ κ(a⊥ ) = Σ
(93)
Proof. Suppose that a is a classical property. From (92) follows that a is a basic classical property, and hence there exists a classical test α such that α tests a.
Consider the ortho test β that tests a. We have that α ≈ β and hence α
≈ β. This means that α
tests a⊥ . Since α is true or α
is true we have a ∈ ξ(p) or a⊥ ∈ ξ(p) for an arbitrary p ∈ Σ, and hence Σ = κ(a) ∪ κ(a⊥ ). Suppose now that Σ = κ(a) ∪ κ(a⊥ ). This means that for an arbitrary p ∈ Σ we have a ∈ ξ(p) or
tests a⊥ . For this a⊥ ∈ ξ(p), and consider the ortho test α testing a such that α ortho test we have that α is true or α
is true, which proves that α is a classical test. Hence a ∈ K = C. We can prove that the classical state property system (Ω, C, ξc ) is isomorphic to the canonical state property system (Ω, P(Ω), Id). THEOREM 66. κc : C → P(Ω) is an isomorphism.
108
Diederik Aerts
Proof. From (87) it follows that κc is an injective function. Let us prove that κc is a surjective function. Take an arbitrary element A ∈ P(Ω). Consider the property
a=
ω(p)⊥c
(94)
κc (ω(p))⊆Ω\A
We have κc (a)
= κc (
ω(p)⊥c ) =
κc (ω(p))⊆Ω\A
κc (ω(p)⊥c ) =
κc (ω(p))⊆Ω\A
κc (Ω \ ω(p))
(95)
κc (ω(p))⊆Ω\A
=
Ω\
κc (ω(p)) = Ω \ (Ω \ A) = A
(96)
κc (ω(p))⊆Ω\A
Let us consider again our two archetypical examples. For the state property system (Ω, P(Ω), Id) corresponding to a classical physical system, we have that classical state property system coincides with this state property system. This shows that in the case of a classical physical system our construction comes out as it should be, the classical state property system is the state property system of this classical physics system. For the state property system (Σ(H), P(H), ξH ) of a quantum physical system, the classical state property system (Ω, C, κC ) is the following: Ω = {H}, C = {0, H}, κC : C → P(Ω), such that κC (0) = ∅ and κC (H) = {H}. This classical state property system describes the aspect of the quantum physical system which has to do with the properties ‘the system is present’ and ‘the system is not present’, properties that even for quantum systems are classical properties. In the next chapter we decompose an arbitrary state property system into its nonclassical components and its classical state space. This structure shows us how we can describe a general situation.
5.2 The non classical components of a state property space In this section we study the description of a physical entity whenever it is in a classical state. This leads to the existence of a non classical property space for each classical state describing the non classical elements of the entity. Consider a classical state ω ∈ Ω. Then ω ∈ C is also a classical property. Hence there exists a classical test testing ω. DEFINITION 67 ωtest. A classical test testing the classical state ω ∈ Ω is called a ωtest, and we denote it αω . DEFINITION 68 ωinverse. Consider a test α and the product test α · αω . We ω ω
· αω and call α · αω the ωinverse of α · αω . deﬁne α · αω = α
Quantum Axiomatics
109
PROPOSITION 69. We have ω
ω α · αω = α · αω
(97)
such that the operation is an inverse operation on the set of tests of the form α·αω . ω
ω ω
Proof. We have α · αω =α
· αω = α
· αω = α · αω .
Let us explain the physical meaning of this. Suppose we consider a typical classical property ω in standard quantum mechanics, for example the property ‘the neutron is there’, in case the entity we are considering is a neutron. The test αω consists of verifying whether the neutron is there, for example by absorbing it on a detection screen. In general such a veriﬁcation of the presence of the neutron destroys the neutron, which means that if we want to test another property, this time a non classical property of the neutron, we need to make recourse to the product test. And hence indeed, when we test the quantum test α, for example the spin of the neutron, then actually we perform the test α · αω . We test whether the neutron is there ‘and’ whether it has spin in a certain direction, by making sure that which ever of the two tests αω or α we perform, the outcome will be ‘yes’. But we do not have to perform both tests together, it is suﬃcient to perform one ‘or’ the other. We are in a similar situation as the one with the piece of wood tested to burn well ‘and’ ﬂoat on water, by performing one of the both tests. DEFINITION 70 ωorthogonality. Consider two states p, q ∈ Σ such that ω ∈ ξ(p)∩ξ(q) where ω ∈ Ω. We say that p and q are ωorthogonal, and denote p ⊥ω q, ω · αω if there exists a test α such that α · αω is true if the entity is in state p and α is true if the entity is in state q. PROPOSITION 71. For x ∈ L and a ∈ C we have x = (x ∧ a) ∨ (x ∧ a⊥ ) κ(x)
(98) ⊥
= κ(x ∧ a) ∪ κ(x ∧ a )
(99)
Proof. Since x ∧ a ≤ x and x ∧ a⊥ ≤ x we have (x ∧ a) ∨ (x ∧ a⊥ ) ≤ x. Since a ∈ C we have κ(a) ∪ κ(a⊥ ) = Σ. This gives κ(x) = κ(x) ∩ (κ(a) ∪ κ(a⊥ )) = (κ(x) ∩ κ(a)) ∪ (κ(x) ∩ κ(a⊥ )) = κ(x ∧ a) ∪ κ(x ∧ a⊥ ) ⊆ κ((x ∧ a) ∨ (x ∧ a⊥ )). This proves (98) and (99). PROPOSITION 72. For x, y ∈ L and a ∈ C such that x ≤ a and y ≤ a⊥ we have (x ∨ y)⊥ = (x⊥ ∧ a) ∨ (y ⊥ ∧ a⊥ ) (x ∨ y) ∧ a = x
(100) (101)
110
Diederik Aerts
Proof. We have a⊥ ≤ x⊥ and a ≤ y ⊥ . From this it follows that y ⊥ ∧ a⊥ ≤ x⊥ and x⊥ ∧ a ≤ y ⊥ . This implies that x⊥ ∧ y ⊥ ∧ a⊥ = y ⊥ ∧ a⊥ and x⊥ ∧ y ⊥ ∧ a = x⊥ ∧ a. Since a ∈ C we have x⊥ ∧ y ⊥ = (x⊥ ∧ y ⊥ ∧ a) ∨ (x⊥ ∧ y ⊥ ∧ a⊥ ). So x⊥ ∧ y ⊥ = (x⊥ ∧ a) ∨ (y ⊥ ∧ a⊥ ). Hence x ∨ y = (x ∨ a⊥ ) ∧ (y ∨ a). But then (x ∨ y) ∧ a = (x ∨ a⊥ ) ∧ a. We know that x⊥ = (x⊥ ∧ a) ∨ (x⊥ ∧ a⊥ ) = (x⊥ ∧ a) ∨ a⊥ . Hence x = (x ∨ a⊥ ) ∧ a. This proves that (x ∨ y) ∧ a = x. PROPOSITION 73. For x, xi ∈ L and a ∈ C we have a ∧ (∨i xi ) = ∨i (a ∧ xi )
(102)
a = (a ∧ x) ∨ (a ∧ x⊥ )
(103)
Proof. We have a ∧ (∨i xi ) = a ∧ (∨i ((xi ∧ a) ∨ (xi ∧ a⊥ )) = a ∧ (∨i (xi ∧ a) ∨ ∨i (xi ∧ a⊥ )) = ∨i (xi ∧ a). We have a = a ∧ (x ∨ x⊥ ). From (102) it follows that a ∧ (x ∨ x⊥ ) = (a ∧ x) ∨ (a ∧ x⊥ ), which proves (103). PROPOSITION 74. For a ∈ L we have a =
(a ∧ ω)
(104)
ω∈Ω
κ(a)
=
κ(a ∧ ω)
(105)
ω∈Ω
with a ∧ ω ⊥ a ∧ ω
and
κ(a ∧ ω) ∩ κ(a ∧ ω ) = ∅
for
ω = ω
(106)
Proof. We have that a ∧ ω ≤ a ∀ω ∈ Ω, hence κ(a ∧ ω) ⊆ κ(a) ∀ ω ∈ Ω, and as a consequence ∪ω∈Ω κ(a ∧ ω) ⊆ κ(a). Consider p ∈ κ(a). We have p ∈ κ(ω(p)), and hence p ∈ κ(a) ∩ κ(ω(p)) = κ(a ∧ ω(p)) ⊆ ∪ω∈Ω κ(a ∧ ω). So we have shown that κ(a) ⊆ ∪ω∈Ω κ(a ∧ ω). This proves (105), namely κ(a) = ∪ω∈Ω κ(a ∧ ω). We have that a ∧ ω ≤ a ∀ω ∈ Ω, hence ∨ω∈Ω (a ∧ ω) ≤ a. Consider p ∈ κ(a). We have p ∈ ∪ω∈Ω κ(a ∧ ω) ⊆ κ(∨ω∈Ω (a ∧ ω)). So we have shown that κ(a) ⊆ κ(∨ω∈Ω (a ∧ ω)). From this it follows that a ≤ ∨ω∈Ω (a ∧ ω), which proves (104), namely a = ∨ω∈Ω (a ∧ ω). Consider ω = ω , then we have ω ≤ ω ⊥ . As a consequence a ∧ ω ≤ ω ⊥ ≤ a⊥ ∨ ω ⊥ = (a ∧ ω )⊥ , which proves that a ∧ ω ⊥ a ∧ ω . From this it follows that κ(a ∧ ω) ∩ κ(a ∧ ω ) = ∅. COROLLARY 75. We have Σ=
κ(ω)
(107)
ω∈Ω
with κ(ω) ∩ κ(ω ) = ∅
for
ω = ω
(108)
Quantum Axiomatics
PROPOSITION 76. Consider aω such that aω ≤ ω ∀ ω ∈ Ω. We have
κ( aω ) = κ(aω ) ω∈Ω
111
(109)
ω∈Ω
with κ(aω ) ∩ κ(aω ) = ∅
for
ω = ω
(110)
Proof. We have κ(∨ω∈Ω aω ) = ∪ω ∈Ω κ((∨ω∈Ω aω ) ∧ ω ). From (101) it follows that (∨ω∈Ω aω ) ∧ ω = aω . Hence κ(∨ω∈Ω aω ) = ∪ω ∈Ω κ(aω ). This proves (109). Let us now investigate the nonclassical parts of the state property system (Σ, L, κ). DEFINITION 77 Nonclassical components. Suppose that (Σ, L, ξ) is the state property space of an entity satisfying axioms 24 and 27. For ω ∈ Ω we introduce Σω Lω ξω (p)
= {p ω ∈ ξ(p), p ∈ Σ} = {a a ≤ ω, a ∈ L} = ξ(p) ∩ Lω
(111) (112) (113)
and we call (Σω , Lω , ξω ) the nonclassical components of (Σ, L, ξ) corresponding to ω. We also introduce Cartan map corresponding to ω κω (a) = {p  p ∈ Σω , a ∈ ξω (p)}
(114)
THEOREM 78. Consider (Σ, L, ξ) the state property space of an entity satisfying axioms 24, 27 and 43. For a, b ∈ Lω , (ai )i ⊆ Lω and p, q ∈ Σω we have a ∈ ξω (p) p≤q p ∈ κω (a)
⇔ a ∈ ξ(p) ⇔ ξω (q) ⊆ ξω (p) ⇔ p ∈ κ(a)
a≤b ⇔ κω (a) = κω (b) ⇒
(115) (116) (117)
κω (a) ⊆ κω (b) a=b
(118) (119)
κω (∧i ai ) p ⊥ω q
= ⇔
∩i κω (ai ) p⊥q
(120) (121)
κω (a⊥ω )
=
κ(a⊥ ∧ ω) = κω (a)⊥ω
(122)
and (Σω , Lω , ξω ) is a state property space that satisﬁes axioms 24, 27 and 43. Proof. Suppose that a ∈ ξω (p). This means that a ∈ ξ(p) ∩ Lω , and hence a ∈ ξ(p). Suppose now that a ∈ ξ(p). Since a ∈ Lω we have a ∈ ξ(p) ∩ Lω = ξω (p). This proves (115). Suppose that p ≤ q, then we have ξ(q) ⊆ ξ(p), and hence
112
Diederik Aerts
ξω (q) = ξ(q) ∩ Lω ⊆ ξ(p) ∩ Lω = ξω (p). Suppose next that ξω (q) ⊆ ξω (p), and consider a ∈ ξ(q). Applying (115) this gives that a ∈ ξω (q) and hence a ∈ ξω (p). Applying again (115) this gives that a ∈ ξ(p). Hence we have proven that ξ(q) ⊆ ξ(p) and hence p ≤ q. This proves (116). We have p ∈ κω (a) ⇔ a ∈ ξω (a) ⇔ a ∈ ξ(p) ⇔ p ∈ κ(a). This proves (117). Suppose that a ≤ b and consider p ∈ κω (a). Applying (117) this gives p ∈ κ(a) and since κ(a) ⊆ κ(b) this gives p ∈ κ(b). Applying again (117) this gives p ∈ κω (b). Hence we have proven that κω (a) ⊆ κω (b). Suppose now that κω (a) ⊆ κω (b), and consider p ∈ κ(a). Applying (117) this gives p ∈ κω (a) and hence p ∈ κω (b). Applying again (117) this gives p ∈ κ(b). So we have proven that κ(a) ⊆ κ(b) and from this follows that a ≤ b. This proves (118). Suppose that κω (a) = κω (b) and consider p ∈ κ(a). Then we have p ∈ κω (a) and hence p ∈ κω (b). From this follows that p ∈ κ(b). This means that we have proven that κ(a) ⊆ κ(b). Since axiom 24 is satisﬁed for the state property space (Σ, L, ξ) we have a = b. This proves (119). We have p ∈ κω (∧i ai ) ⇔ p ∈ κ(∧i ai ) ⇔ p ∈ ∩i κ(ai ) ⇔ p ∈ κ(ai ) ∀ i ⇔ p ∈ κω (ai ) ∀ i ⇔ p ∈ ∩i κω (ai ). This proves (120). Suppose that p ⊥ω q. This means that there ω · αω is true exists a test α such that α · αω is true if the entity is in state p and α ω
· αω . This means that α is true if the entity is in state q. We have α · αω = α if the entity is in state p and α
is true if the entity is in state q. Hence p ⊥ q. Suppose now that p ⊥ q. This means that there exists a test α such that α is true if the entity is in state p and α
is true if the entity is in state q. Since p, q ∈ Σω we have that ω is actual and hence αω is true. Hence α · αω is true if the entity ω · αω is true if the entity is in state q. This means is in state p and α
· αω = α that p ⊥ω q. This proves (121). Suppose that a ∈ Lω . In this case a ∈ L and hence there exists an ortho test α testing a. Consider the test α · αω and a state q such that q ⊥ω a and q ∈ Σω . This means that q ⊥ω r ∀ r such that r ∈ κω (a). Hence q ⊥ r ∀ r such that r ∈ κ(a). Hence q ∈ κ(a)⊥ = κ(a⊥ ). This means that α
is true if the entity is in state q. Since q ∈ Σω we also have that αω is true ω · αω is true if the entity is in state if the entity is in state q. Hence α
· αω = α ω q. Note that property tested by α · αω is a⊥ ∧ ω. Hence we have proven that ⊥ ⊥ω ⊆ κω (a⊥ ∧ ω). Consider r ∈ Σω such that q ⊥ω a ⇒ q ∈ κω (a ∧ ω), or κ(a) ⊥ r ⊥ω a ∧ ω. This means that r ⊥ω s ∀ s such that a⊥ ∧ ω ∈ ξ(s). From this follows that r ⊥ s ∀ s such that a⊥ ∧ ω ∈ ξ(s). Hence r ∈ κ(a⊥ ∧ ω)⊥ . Since r ∈ Σω we have r ⊥ ω ⊥ , and hence r ⊥ a⊥ ∧ ω ⊥ , such that r ∈ κ(a⊥ ∧ ω ⊥ )⊥ . From (98) and (99) we know that κ(a⊥ ) = κ(a⊥ ∧ ω) ∪ κ(a⊥ ∧ ω ⊥ ), and hence κ(a) = κ(a⊥ )⊥ = κ(a⊥ ∧ ω)⊥ ∩ κ(a⊥ ∧ ω ⊥ )⊥ . Hence we have r ∈ κ(a). This proves that for the entity being in state r we have α is true, and hence α · αω is true. This proves that α · αω is an ortho test for the orthogonality relation ⊥ω . We can now denote a⊥ ∧ ω = a⊥ω . And it follows that we have proven (122). Remark that from (119) follows that axiom 24 is satisﬁed for (Σω , Lω , ξω ), and from (120) follows that axiom 27 is satisﬁed. From (122) follows that axiom 43 is satisﬁed for (Σω , Lω , ξω ).
Quantum Axiomatics
5.3
113
A decomposition theorem
To see in more detail in which way the classical and nonclassical parts are structured within the lattice L, we need to introduce some additional structures. DEFINITION 79 Direct union of state property spaces. Consider a set of state property spaces (Σω , Lω , ξω ) that all satisfy axioms 24, 27 and 43. The direct ∨ ω (Σω , Lω , ξω ) of these state property spaces is the state property space union ! ∨ ω Lω , ! ∨ ω ξω ), where (∪ω Σω , ! (i) ∪ω Σω is the disjoint union of the sets Σω ∨ ω Lω is the direct union of the lattices Lω , which means the set of sequences (ii) ! a = (aω )ω , such that (aω )ω ≤ (bω )ω
⇔
aω ≤ bω ∀ω ∈ Ω
(123)
(aω )ω ∧ (bω )ω (aω )ω ∨ (bω )ω
= =
(aω ∧ bω )ω (aω ∨ bω )ω
(124) (125)
(aω )⊥ ω
=
ω (a⊥ ω )ω
(126)
∨ ω ξω is deﬁned as follows: (iii) ! ∨ ω ξω : ∪ω Σω !
p
ω
→
∨ ω Lω ) P(!
→
{(aω )ω  a
(127)
ω
∈ ξ (p ), aω ∈ Lω ∀ ω = ω } ω
ω
(128)
and hence the corresponding Cartan map is the following ∨ ω κω : ! ∨ ω Lω !
→
∪ω Σω
(129)
(aω )ω
→
∪ω∈Ω κω (aω )
(130)
∨ ω∈Ω Lω We remark that if Lω are complete orthocomplemented lattices, then also ! is a complete orthocomplemented lattice. A fundamental decomposition theorem can now be proven.
THEOREM 80 Decomposition theorem. Consider the state property space (Σ, L, ξ), and suppose that axioms 24, 27 and 43 are satisﬁed. Then ∨ ω∈Ω (Σω , Lω , ξω ) (Σ, L, ξ) ∼ =!
(131)
where Ω is the set of classical states of (Σ, L, ξ), Σω is the set of states and Lω the lattice of properties of the nonclassical component state property space (Σω , Lω , ξω ). Proof. We use the notion of orthomorphism of state property systems, and need to prove that there exists an isomorphism of ortho state property systems between ∨ ω∈Ω (Σω , Lω , ξω ). From (107) it follows that m can be deﬁned in (Σ, L, ξ) and ! the following way: m:Σ → p
→
∪ω∈Ω Σω
(132)
p
(133)
114
Diederik Aerts
The function n is deﬁned in the following way: ∨ ω∈Ω Lω n:!
→
L
(134)
(aω )ω
→
∨ω∈Ω aω
(135)
∨ ω∈Ω Lω and The function m is a bijection by deﬁnition. Consider (aω )ω , (bω )ω ∈ ! suppose that n((aω )ω ) = n((bω )ω ), hence ∨ω∈Ω aω = ∨ω∈Ω bω . Then (∨ω∈Ω aω ) ∧ ω = (∨ω∈Ω bω ) ∧ ω ∀ ω ∈ Ω. From (101) it follows that (∨ω∈Ω aω ) ∧ ω = aω and (∨ω∈Ω bω ) ∧ ω = bω . Hence aω = bω ∀ ω ∈ Ω. As a consequence we have (aω )ω = (bω )ω . This proves that n is injective. Let us prove that n is surjective. Consider an arbitrary element a ∈ L. From (104) it follows that a = ∨ω∈Ω (a ∧ ω). ∨ ω∈Ω Lω . Then n((a ∧ ω)ω ) = a which proves Consider the element (a ∧ ω)ω ∈ ! that n is surjective. Hence we have proven that m as well as n are bijections. Let us show that we have an orthomorphism. We need to prove (67) and (72) hence: ∨ ω κω ((aω )ω ) ⇔ m(p) ∈ ! ⊥ ∨ ω κω ((aω )ω ) ⇔ m(p) ∈ !
p ∈ κ(∨ω∈Ω aω ) p ∈ κ((∨ω∈Ω aω )⊥ )
(136) (137)
∨ ω κω ((aω )ω ) = ∪ω∈Ω κω (aω ) = ∪ω∈Ω κ(aω ). On the other hand Let us calculate ! we have κ(n((aω )ω )) = κ(∨ω∈Ω aω ), and following (109), we have κ(∨ω∈Ω aω ) = ⊥ ∨ ω κω ((aω )ω ) = ∪ω∈Ω κ(aω ). This means that (136) is satisﬁed. We have ! ⊥ω ⊥ω ⊥ ⊥ ⊥ ∨ ω κω ((aω )ω ) = ∪ω κω (aω ) = ∪ω κω (a ∧ ω) = κ(∨ω∈Ω (a ∧ ω)) = κ(a ) = ! ⊥ κ((∨ω∈Ω aω ) ). This proves (137). Hence we have proven that (m, n) is an isomorphism of ortho state property spaces.
6 ADDITIONAL AXIOMS In the foregoing we have introduced three axioms. If these three axioms are satisﬁed we can decompose the state property space of an entity into its classical state property space such that for each classical state there is an underlying non classical property space describing the entity being in this classical state. We have proven that the classical state property space is isomorphic to the property space of classical physics. The underlying non classical state property spaces are however not necessarily isomorphic to the state property space of quantum mechanics. To make these underlying non classical state property spaces isomorphic to the state property space of quantum mechanics we need to introduce additional axioms.
6.1 The axiom of atomisticity The axiom of property determination makes the preorder relation on the set of properties L of a state property space into a partial order relation. The preorder relation existing on the set of states is not necessarily a partial order relation. This means that it is possible for two states p, q ∈ Σ to be diﬀerent states even if ξ(p) = ξ(q), which means that the properties which are actual if the entity is
Quantum Axiomatics
115
in state p are the same as the properties which are actual if the entity is in state q. The next axiom we introduce makes the preorder relation on Σ into a trivial order, i.e. p ≤ q iﬀ p = q. AXIOM 81 Atomisticity. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. We say that the axiom of atomisticity is satisﬁed if for p, q ∈ Σ we have ξ(q) ⊆ ξ(p) ⇒ p = q
(138)
DEFINITION 82. Consider a partially ordered set Z, ≤. We say that s ∈ Z is an atom, if whenever 0 ≤ a ≤ s we have a = 0 ot a = s. A lattice Z, ≤, is atomistic, if there exists a set of atoms A which is ordering. This means that for x, y ∈ Z we have x ≤ y ⇔ {s  s ∈ A, s ≤ x} ⊆ {s  s ∈ A, s ≤ y}. THEOREM 83. Consider a state property space (Σ, L, ξ) for which axioms 24, 27, 43 and 81 are satisﬁed. For p, q ∈ Σ and a, b ∈ L we have p≤q ⇒ p=q 0 ≤ a ≤ s(p) ⇒ a = 0 or a = s(p) a ≤ b ⇔ {s(p)  s(p) ≤ a} ⊆ {s(p)  s(p) ≤ b}
(139) (140) (141)
and L is an atomistic lattice with set of atoms the set of state properties A = {s(p)  p ∈ Σ}. Proof. Suppose p ≤ q, then we have ξ(q) ⊆ ξ(p), and hence p = q. This proves (139). Consider a such that 0 ≤ a ≤ s(p) = ∧b∈ξ(p) b. If a = 0 there exists q ∈ Σ such that a ∈ ξ(q). Hence we have s(p) ∈ ξ(q), and as a consequence we have b ∈ ξ(q) ∀ b ∈ ξ(p). Hence ξ(p) ⊆ ξ(q), and hence p = q. This implies that s(p) = s(q). Since s(q) ≤ a we have a = s(p). So we have proven (140), which means that s(p) is an atom of L. If a ≤ b we obviously have that {s(p)  s(p) ≤ a} ⊆ {s(p)  s(p) ≤ b}. Suppose that {s(p)  s(p) ≤ a} ⊆ {s(p)  s(p) ≤ b}. From (39) follows that a = ∨s(p)≤a s(p) ≤ ∨s(p)≤b s(p) = b, Hence we have proven (141), which means that A is an ordering set for L, and hence L is a complete orthocomplemented atomistic lattice. PROPOSITION 84. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. The classical state property space (Ω, C, ξc ) corresponding to (Σ, L, ξ) satisﬁes axiom 81. If (Σ, L, ξ) satisﬁes also axiom 81, then each non classical component (Σω , Lω , ξω ) of the decomposition of (Σ, L, ξ) satisﬁes axiom 81. Proof. Consider ω(p), ω(q) ∈ Ω such that ξc (ω(q)) ⊂ ξc (ω(p)). From (85) follows that then ω(p) ≤ ω(q), and hence ω(p) ∧ ω(q) = ω(p). Suppose now that ω(p) = ω(q), then from (83) follows that ω(p) ⊥ ω(q). But then ω(p) ∧ ω(q) = 0, which would lead to ω(p) = 0. This is not possible, and hence this proves that ω(p) =
116
Diederik Aerts
ω(q). Hence we have proven that (Ω, C, κc ) satisﬁes axiom 81. Suppose now that axiom 81 is satisﬁed for (Σ, L, ξ), and consider p, q ∈ Σω such that ξω (q) ⊆ ξω (p). From (116) follows then that p ≤ q and hence ξ(q) ⊆ ξ(p). Since axiom 81 is satisﬁed for (Σ, L, ξ) we have that p = q. Hence we have proven that axiom 81 is satisﬁed for (Σω , Lω , ξω ). PROPOSITION 85. Consider a state property space (Σ, L, ξ) for which axioms 24, 27, 43 and 81 are satisﬁed. A property a ∈ L is classical, hence a ∈ C, iﬀ a is a central element of the lattice L, i.e. x = (x ∧ a) ∨ (x ∧ a⊥ ) ∀ x ∈ L. The lattice of properties Lω of a non classical component property space (Σω , Lω , ξω ) is an irreducible lattice. Proof. From (98) follows that a classical property a is a central element of the lattice L. Consider a central element a of the lattice L, and an arbitrary state p ∈ Σ. Because axiom 81 is satisﬁed we have that s(p) is an atom of L. We have s(p) ∧ a ≤ s(p) and s(p) ∧ a⊥ ≤ s(p) and hence s(p) ∧ a = s(p) or s(p) ∧ a = 0, and s(p) ∧ a⊥ = s(p) or s(p) ∧ a⊥ = 0. Since a is a central element of L we have s(p) = (s(p) ∧ a) ∨ (s(p) ∧ a⊥ ), and hence we cannot have s(p) ∧ a = 0 and s(p) ∧ a⊥ = 0, which means that at least one of s(p) ∧ a = s(p) or s(p) ∧ a⊥ = s(p) is true. From this follows that s(p) ≤ a or s(p) ≤ a⊥ . Hence a ∈ ξ(p) or a⊥ ∈ ξ(p). Since axiom 43 is satisﬁed there exists an ortho test α testing a and hence α
testing a⊥ . From the foregoing follows that this ortho test is a classical test, and hence a is a classical property. Consider a central element a ∈ Lω of the lattice of properties of a non classical component (Σω , Lω , ξω ). Since a ∈ C, we have a = 0 or a = ω which proves that Lω is irreducible.
6.2 The axiom of weak modularity If we consider a closed subspace A ∈ L(H) of a Hilbert space H. The closed subspace A in itself is a Hilbert space. This means that we can consider the set L(A) of closed subspaces contained in A. One can prove that L(A) is a complete orthocomplemented lattice. The relative orthocomplementation B ⊥A for B ∈ L(A) is deﬁned as follows: B ⊥A = B ⊥ ∩ A. An important equality for ⊥A to be an orthocomplementaion is the following: (B ⊥A )⊥A = B. This gives B = (B ⊥ ∩ A)⊥ ∩ A or B = (B ∨ A⊥ ) ∧ A. This is the way the requirement of ‘weak modularity’ is usually introduced, hence more speciﬁcally: for B, A ∈ L and B ≤ A we have B = (B ∨ A⊥ ) ∧ A. We however want to introduce ‘weak modularity’ in an operational way. To formulate the following axiom we ﬁrst introduce the idea of relative state property space. PROPOSITION 86 Relative state property space. Suppose we have a state property space (Σ, L, ξ) and for a ∈ L we consider (Σ, L, ξ)a = (κ(a), La , ξa ) where La
=
{b  b ≤ a}
(142)
ξa : κ(a)
→
P(La )
(143)
p → ξ(p) ∩ La
Quantum Axiomatics
117
then (Σ, L, ξ)a is a state property space and for b ∈ La we have κa (b) = κ(b)
(144)
We call (Σ, L, ξ)a the state property space relative to property a ∈ L. Proof. Suppose the entity is in state p ∈ κ(a) and the property b ≤ a is actual. This means that b ∈ ξ(p). Since b ≤ a we have b ∈ ξ(p) ∩ La = ξa (p). On the other hand, suppose we have p ∈ κ(a) and b ≤ a, and b ∈ ξa (p). This means that b ∈ ξ(p), and hence, if the entity is in state p the property b is actual. This proves that (Σ, L, ξ)a is a state property space describing the same entity as the one described by the state property space (Σ, L, ξ). Suppose that p ∈ κa (b), and hence b ∈ ξa (p). This means that b ∈ ξ(p), and hence p ∈ κ(b). Hence we have κa (b) ⊆ κ(b). Suppose now that p ∈ κ(b) and hence b ∈ ξ(p). Since we have b ≤ a we also have b ∈ La , and hence b ∈ ξ(p) ∩ La = ξa (p). From this follows that p ∈ κa (b). Hence we have proven that κa (b) ⊆ κ(b). As a consequence we have κa (b) = κ(b). The operational meaning of the relative state property space is the following. We study the entity S in the special circumstance when we manage to keep the property a actual during the study. This means concretely that we can consider a test α testing a, and hence we consider only the states κ(a) of the entity which make this test true. PROPOSITION 87. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed, and a ∈ L. Then the relative state property space (Σ, L, ξ)a satisﬁes axioms 24 and 27. Proof. Consider b, c ∈ La such that κa (b) = κa (c). From (144) follows that κ(b) = κ(c), and since axiom 24 is satisﬁed for (Σ, L, ξ) we have b = c. This proves that axiom 24 is satisﬁed for (Σ, L, ξ)a . Consider b ∈ La . Since axiom 27 is satisﬁed for (Σ, L, ξ) there exists (bi )i ⊆ T such that κ(b) = ∩i κ(bi ). Consider Ta = {b ∧ a  b ∈ T }. Since b ≤ a we have b = b ∧ a. Hence κ(b) = κ(b) ∩ κ(a) and as a consequence we have κ(b) = ∩i κ(bi ) ∩ κ(a) = ∩i (κ(bi ) ∩ κ(a)) = ∩i κ(bi ∧ a). Consider now an arbitrary property b ∈ La . Since b ∈ L there exists (bi )i ⊆ T such that κ(b) = ∩i κ(bi ). This gives that κ(b) = ∩i κ(bi ∧ a) for (bi ∧ a)i ⊆ Ta . Hence we have proven that (Σ, L, ξ)a satisﬁes axiom 27. The axiom 43 of orthocomplementation is however not necessarily satisﬁed for a relative state property space. The next axiom, the axiom of weak modularity, is meant to make sure that also the axiom 43 of orthocomplementation is satisﬁed for an arbitrary relative state property space. Before we formulate the axiom of weak modularity, let us analyse why the axiom 43 of orthocomplementation is not necessarily satisﬁed for a relative state property space. We start by introducing the notion of relative inverse.
118
Diederik Aerts
DEFINITION 88 Relative inverse. Consider a state property space (Σ, L, ξ) and for a ∈ L the relative state property space (Σ, L, ξ)a . For a test β we consider the α test β ·α where α is the test testing the property a, and we introduce β · α = β ·α. α PROPOSITION 89. β · α is an inverse for tests testing properties of the relative state property space (Σ, L, ξ)a . Proof. Consider an arbitrary property b ∈ La and a test β testing b. Since b ≤ a α we have that also β · α tests property b. Obviously also β · α = β · α is a test α α α
testing a property of La , and we have β · α = β · α = β · α = β · α. This proves α that β · α deﬁnes an inverse on tests of properties of La . DEFINITION 90 Relative orthogonality. Consider a state property space (Σ, L, ξ), and for a ∈ L the relative state property space (Σ, L, ξ)a . Consider a test α testing a. For p, q ∈ κ(a) we say that p is relatively orthogonal to q with respect to α, and denote p ⊥α q, if there exists a test β · α such that β · α is true if the entity is α in state p and β · α is true if the entity is in state q. PROPOSITION 91. Consider a state property space (Σ, L, ξ), and for a ∈ L the relative state property space (Σ, L, ξ)a , and a test α testing a. For p, q ∈ κ(a) we have p ⊥α q ⇔ p ⊥ q
(145)
Proof. Suppose that p ⊥α q. This means that there exists β · α such that β · α is true if the entity is in state p and β · α is true if the entity is in state q. This means that β is true if the entity is in state p and β is true if the entity is in state q. Hence p ⊥ q. Suppose that p ⊥ q. This means that there exists a test γ such that γ is true if the entity is in state p and γ
is true if the entity is in state q. If we consider γ · α and γ
· α, then, since p, q ∈ κ(a) and α tests a, we have that γ · α is true if the entity is in state p and γ
· α is true if the entity is in state q. This proves that p ⊥α q.
PROPOSITION 92. Consider a state property space (Σ, L, ξ) for which 24, 27 and 43 are satisﬁed, and for a ∈ L the relative state property space (Σ, L, ξ)a . Axiom 43 is satisﬁed for the state property space (Σ, L, ξ)a if and only if for b ≤ a we have b = (b ∨ a⊥ ) ∧ a. Proof. Consider an ortho test α testing property a and an ortho test β testing property b and p ∈ κ(a) such that p ⊥α b. From proposition 91 follows that p ⊥ b and hence, since axiom 43 is satisﬁed for (Σ, L, ξ) we have β is true if the entity is in state p. Since p ∈ κ(a) we have that β · α is true if the entity is in state
Quantum Axiomatics
119
p. Hence p ∈ κ(b⊥ ∧ a). On the other hand, consider q ∈ κ(a) and q ⊥ b⊥ ∧ a.
is true if the entity Since b⊥ ∧ a ∈ L there exists an ortho test γ, such that γ
· α will test b, is in state q. This means that q ∈ κ((b⊥ ∧ a)⊥ ∧ a)). The test γ and hence be an ortho test for the relative inverse with respect to a, if and only if κ(b) = κ((b⊥ ∧ a)⊥ ∧ a)). This is equivalent to b = (b⊥ ∧ a)⊥ ∧ a) = (b ∨ a⊥ ) ∧ a. AXIOM 93 Weak modularity. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. We say that the axiom of weak modularity is satisﬁed if for a, b ∈ L we have b≤a
⇒
b = (b ∨ a⊥ ) ∧ a
(146)
Operationally this axiom means the following. Consider two properties b ≤ a, an ortho test β testing b, and a test α testing a. Then β · α tests b⊥ ∧ a. Consider
· α tests the property an ortho test γ ≈ β · α, then also γ tests b⊥ ∧ a. The test γ
· α also to (b⊥ ∧ a)⊥ ∧ a. The axiom of weak modularity means that we want γ test the property b. Hence we want γ
· α ≈ β. PROPOSITION 94. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. The classical state property space (Ω, C, ξc ) corresponding to (Σ, L, ξ) satisﬁes axiom 93. If (Σ, L, ξ) satisﬁes also axiom 93, then each non classical component (Σω , Lω , ξω ) of the decomposition of (Σ, L, ξ) satisﬁes axiom 93. Proof. Consider a, b ∈ C such that b ≤ a. From (102) follows that (b ∨ a⊥ ) ∧ a = (b∧a)∨(a⊥ ∧a) = b∨0 = b. This proves that (Ω, C, ξc ) satisﬁes axiom 93. Suppose now that axiom 93 is satisﬁed for (Σ, L, ξ) and consider a, b ∈ Lω such that b ≤ a. Using (122), and hence b⊥ω = b⊥ ∧ ω, we have (b ∨ a⊥ω ) ∧ a = (b⊥ω ∧ a)⊥ω ∧ a = (b⊥ ∧ω∧a)⊥ω ∧a = (b⊥ ∧a)⊥ω ∧a = (b⊥ ∧a)⊥ ∧ω∧a = (b⊥ ∧a)⊥ ∧a = (b∨a⊥ )∧a = b. This proves that axiom 93 is satisﬁed for (Σω , Lω , ξω ).
6.3
The axiom ‘the covering law’
The covering law is the root of the linear structure of quantum mechanics, which means that it is a very important axiom. In some sense it demands something similar to the axiom of atomisticity, but then for all parts of the lattice of properties. AXIOM 95 The covering law. Consider a state property space (Σ, L, ξ) for which axioms 24, 27, 43 and 81 are satisﬁed. For a, b ∈ L and p ∈ Σ we have s(p) ∧ a = 0 and a ≤ b ≤ a ∨ s(p) ⇒ b = a or b = a ∨ s(p)
(147)
PROPOSITION 96. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. The classical state property space (Ω, C, κc ) corresponding
120
Diederik Aerts
to (Σ, L, ξ) satisﬁes axiom 95. If (Σ, L, ξ) satisﬁes also axiom 95, then each non classical component (Σω , Lω , ξω ) of the decomposition of (Σ, L, ξ) satisﬁes axiom 95. Proof. Consider a, b ∈ C and ω(p) ∈ Ω, such that a∧ω(p) = 0 and a ≤ b ≤ a∨ω(p). Making use of theorem 66 we have κc (a) ⊆ κc (b) = κc (a ∨ ω(p)) = κc (a) ∪ {ω(p)}. Hence κc (b) = κc (a) or κc (b) = κc (a) ∪ {ω(p)}. From (87) follows then that b = a or b = a ∨ ω(p). This proves that (Ω, C, κc ) satisﬁes axiom 95. Suppose now that axiom 95 is satisﬁed for (Σ, L, ξ) and consider a, b ∈ Lω and p ∈ Σω such that a ∧ s(p) = 0 and a ≤ b ≤ a ∨ s(p). From axiom 95 being satisﬁed for (Σ, L, ξ) follows that b = a or b = a ∨ s(p), which proves that axiom 95 is satisﬁed for (Σω , Lω , ξω ).
6.4 The axiom of plane transitivity The seventh axiom that brings us directly to the structure of one of the three standard Hilbert spaces is much more recent [Aerts and van Steirteghem, 2000]. AXIOM 97 Plane transitivity. Consider a state property space (Σ, L, ξ) for which axioms 24, 27 and 43 are satisﬁed. The state property space is plane transitive if for an arbitrary classical state ω ∈ Ω and states p, q ∈ Σω there exist two distinct atoms s1 , s2 ∈ Lω and an automorphism (m, n) of (Σω , Lω , ξω ) such that n[0,s1 ∨s2 ] is the identity and m(p) = q. Both classical entities and quantum entities can be described by a state property space where the set of properties is a complete atomistic orthocomplemented lattice that satisﬁes the covering law, is weakly modular and plane transitive. In section 8 we consider the converse, namely how this structure leads us to classical physics and to quantum physics. But ﬁrst we want to look into one of the basic notions of quantum mechanics, namely the notion of superposition state. 7 SUPERPOSITION One of the aspects which is often put forward as the most characteristic feature of quantum mechanics is the existence of ‘superposition states’. In principle, the notion of superposition of states is intrinsically linked to the linearity of the Hilbert space. It is however possible to introduce it on a more fundamental level, which is what we will do in this section. DEFINITION 98 Superposition. Consider a state property space (Σ, L, ξ) and states p, q, r ∈ Σ. We say that r is a superposition of p and q if ξ(p) ∩ ξ(q) ⊆ ξ(r). More generally, for a set of states Γ ⊆ Σ we say that r is a superposition of Γ if ∩p∈Γ ξ(p) ⊆ ξ(r). We call ¯ = {r  r ∈ Σ, ∩p∈Γ ξ(p) ⊆ ξ(r)} Γ the superposition set corresponding to Γ ⊆ Σ.
(148)
Quantum Axiomatics
121
PROPOSITION 99. Consider a state property space (Σ, L, ξ) and Γ ⊆ Σ. For Γ, Δ ⊆ Σ we have ¯ Γ ¯⊆Δ ¯ Γ⊆Δ ⇒ Γ ¯ ¯ Γ = Γ Γ
⊆
(149) (150) (151)
which proves that ¯ is a closure operator. ¯ which proves Proof. We have ∩p∈Γ ξ(p) ⊆ ξ(r) ∀ r ∈ Γ, and hence that Γ ⊆ Γ, ¯ (149). Suppose we have Γ ⊆ Δ ⊆ Σ and consider r ∈ Γ. We have ∩p∈Δ ξ(p) ⊆ ¯ As a consequence we have Γ ¯ ⊆ Δ, ¯ which ∩p∈Γ ξ(p) ⊆ ξ(r) and hence r ∈ Δ. ¯ ¯ ¯ proves (150). From (149) follows that Γ ⊆ Γ. Consider r ∈ Γ, which means that ¯ Hence ∩p∈Γ¯ ξ(p) ⊆ ξ(r). We have ∩p∈Γ ξ(p) ⊆ ∩p∈Γ¯ ξ(p) ⊆ ξ(r), and hence r ∈ Γ. ¯ ⊆ Γ. ¯ This means that we have proven (151). we have that Γ PROPOSITION 100. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed, and Γ ⊆ Σ. We have ¯ ⇔ s(r) ≤ ∨p∈Γ s(p) r∈Γ
(152)
¯ which means that ∩p∈Γ ξ(p) ⊆ ξ(r). We have s(r) = Proof. Consider r ∈ Γ, ∧a∈ξ(r) a ≤ ∧a∈∩p∈Γ ξ(p) a = ∧a∈ξ(p)∀p∈Γ a = ∧s(p)≤a∀p∈Γ a = ∨p∈Γ s(p). Suppose now that s(r) ≤ ∨p∈Γ s(p). This means that ∧a∈∩p∈Γ ξ(p) a ≤ ∧a∈ξ(r) a and hence ¯ ∩p∈Γ ξ(p) ⊆ ξ(r), which proves that r ∈ Γ. DEFINITION 101 Superselection. Consider a state property space (Σ, L, ξ). We say that p, q ∈ Σ are separated by a superselection rule, and we denote p ssr q, if the only superpositions of p and q are contained in p or in q. Hence, if for r ∈ Σ such that ξ(p) ∩ ξ(q) ⊆ ξ(r) we have ξ(p) ⊆ ξ(r) or ξ(q) ⊆ ξ(r). PROPOSITION 102. Consider a state property space (Σ, L, ξ) for which axioms 24 and 27 are satisﬁed. For p, q ∈ Σ we have p ssr q ⇔ κ(s(p) ∨ s(q)) = {r  r ∈ Σ, r ≤ p or r ≤ q}
(153)
Proof. Suppose that p ssr q and consider r ∈ κ(s(p) ∨ s(q)) and hence s(r) ≤ s(p) ∨ s(q). From (152) follows that r ∈ {p, q}¯ = {r  r ∈ Σ, r ≤ p or r ≤ q}. Hence r ≤ p or r ≤ q. Suppose that κ(s(p) ∨ s(q)) = {r  r ∈ Σ, r ≤ p or r ≤ q} and consider r ∈ {p, q}¯. From (152) follows that s(r) ≤ s(p) ∨ s(q) and hence r ∈ κ(s(p) ∨ s(q)). As a consequence we have r ≤ p or r ≤ q. This proves that p ssr q.
122
Diederik Aerts
THEOREM 103. Consider a state property space (Σ, L, ξ) for which axioms 24, 27, 43, 81 and 93 are satisﬁed. For p, q ∈ Σ we have p ssr q ⇒ p = q or p ⊥ q
(154)
Proof. Suppose that p ssr q and p ⊥ q, and consider r such that r ∈ κ((s(p) ∨ s(q)) ∧ s(q)⊥ ). This means that r ∈ κ((s(p) ∨ s(q)) and r ∈ κ(s(q)⊥ ). From (153) we have that r ≤ p or r ≤ q, but since axiom 81 is satisﬁed, this gives r = p or r = q. However, since r ∈ κ(s(q)⊥ ) we cannot have r = q. Hence r = p. Hence we have proven that κ((s(p) ∨ s(q)) ∧ s(q)⊥ ) = κ(s(p) ∧ s(q)⊥ ). We have s(p) ∧ s(q)⊥ ≤ s(p) and since s(p) is an atom of L we have s(p) ∧ s(q)⊥ = s(p) or s(p) ∧ s(q)⊥ = 0. If s(p) ∧ s(q)⊥ = s(p) we have s(p) ≤ s(q)⊥ and hence p ⊥ q, which is not true. This means that s(p) ∧ s(q)⊥ = 0, and hence (s(p) ∨ s(q)) ∧ s(q)⊥ = 0. From this follows that (s(p)⊥ ∧ s(q)⊥ ) ∨ s(q) = I, and as a consequence we have ((s(p)⊥ ∧ s(q)⊥ ) ∨ s(q)) ∧ s(q)⊥ = s(q)⊥ . Since axiom 93 is satisﬁed we have ((s(p)⊥ ∧ s(q)⊥ ) ∨ s(q)) ∧ s(q)⊥ = s(p)⊥ ∧ s(q)⊥ . Hence we have s(q)⊥ = s(p)⊥ ∧ s(q)⊥ , and as a consequence s(q)⊥ ≤ s(p)⊥ . From this follows that s(p) ≤ s(q), and since s(q) is an atom of L we have s(p) = s(q), and hence p = q. THEOREM 104. Consider a state property space (Σ, L, ξ) for which axioms 24, 27, 43, 81 and 95 are satisﬁed. For p, q ∈ Σ we have p ssr q ⇒ p = q or p ⊥ q
(155)
Proof. In the same way as in the proof of theorem 103 we prove that (s(p)⊥ ∧ s(q)⊥ )∨s(q) = I if p ssr q and p ⊥ q. This means that I covers s(p)⊥ ∧s(q)⊥ . Since s(p)⊥ ∧s(q)⊥ ≤ s(q)⊥ ≤ I, and axiom 95 is satisﬁed, we have s(p)⊥ ∧s(q)⊥ = s(q)⊥ and hence s(q)⊥ ≤ s(p)⊥ . From this follows that s(p) ≤ s(q), and since s(q) is an atom, we have s(p) = s(q) and hence p = q. From these theorems follows that if axioms 24, 27, 43, 81 and 93 are satisﬁed or if axioms 24, 27, 43, 81 and 95 are satisﬁed, and two diﬀerent states p and q are separated by a superselection rule then they are orthogonal. It also means that for two diﬀerent states p and q that are not orthogonal there always exists a third state r which is a superposition of p and q. 8
HILBERT SPACE REPRESENTATIONS
In this section we make further steps to get closer to standard quantum mechanics in a complex Hilbert space. A ﬁrst step is based on Piron’s representation theorem for an irreducible complete orthocomplemented weakly modular lattice satisfying the covering law [Piron, 1964]. Piron proved that such a lattice can be represented as the set of closed subspaces of a generalized Hilbert space.
Quantum Axiomatics
8.1
123
Representation in Generalized Hilbert spaces
Starting from the general decomposition theorem 80 we have proven in section 5.3, and using the extra axioms 81, 93, 95 introduced in section 6, we can prove the following theorem for each one of the non classical components of the decomposition. THEOREM 105. Consider a state property space (Σ, L, ξ), and suppose that ax∨ ω∈Ω (Σω , Lω , ξω ) the ioms 24, 27, 43, 81, 93 and 95 are satisﬁed. Consider ! decomposition of (Σ, L, ξ) in its non classical components. For each nonclassical component (Σω , Lω , ξω ), which has at least four orthogonal states, there exists a vector space Vω , over a division ring Kω , with an involution of Kω , which means a function ∗
: Kω → Kω
(156)
such that for k, l ∈ Kω we have: (k ∗ )∗ (k · l)∗
= =
k l∗ · k ∗
(157) (158)
and an Hermitian product on Vω , which means a function , : V ω × V ω → Kω
(159)
such that for x, y, z ∈ Vω and k ∈ Kω we have: x + ky, z = x, z + kx, y x, y∗ = y, x x, x = 0 ⇔ x = 0
(160) (161) (162)
and such that for M ⊂ Vω we have: M ⊥ + (M ⊥ )⊥ = Vω
(163)
where M ⊥ = {y y ∈ Vω , y, x = 0, ∀x ∈ M }. Such a vector space is called a generalized Hilbert space or an orthomodular vector space. And we have that: (Σω , Lω , ξω ) ∼ = (R(Vω ), L(Vω ), νω )
(164)
where R(Vω ) is the set of rays of V , L(Vω ) is the set of biorthogonally closed subspaces (subspaces that are equal to their biorthogonal) of Vω , and νω makes correspond with each ray the set of biorthogonally closed subspaces that contain this ray. Proof. If axioms 24, 27, 43, 81, 93 and 95 are satisﬁed for (Σ, L, ξ), then from theorem 78 and propositions 84, 94 and 96 follows that (Σω , Lω , ξω ) satisﬁes axioms
124
Diederik Aerts
24, 27, 43, 81, 93 and 95. Hence the lattice Lω is a complete orthocomplemented atomistic weakly modular lattice satisfying the covering law. Furthermore from proposition 85 follows that Lω is irreducible, and since (Σω , Lω , ξω ) has at least four orthogonal states, it follows that Lω has at least four orthogonal atoms. This means that for Lω we can employ Piron’s representation theorem [Piron, 1964; Piron, 1976; Maeda and Maeda, 1970; Faure and Fr¨ olicher, 1995], and hence infer that there exists a vector space Vω , over a division ring Kω , with an involution ∗ of Kω and an Hermitian product , on Vω and such that for M ⊂ Vω we have: M ⊥ + (M ⊥ )⊥ = Vω
(165)
where M ⊥ = {y y ∈ Vω , y, x = 0, ∀x ∈ M }, and such that Lω ∼ = L(Vω )
(166)
where L(Vω ) is the set of all biorthogonal subspaces of Vω , i.e. L(Vω ) = {M  M ⊆ Vω , (M ⊥ )⊥ = M }
(167)
Each atom s(p) of the lattice Lω is represented by a ray, i.e. a one dimensional subspace of Vω . This means that Σω can be put equal to R(Vω ) the set of rays of the vector space Vω . If we deﬁne νω as the function from R(Vω ) to P(L(Vω )), that makes correspond with each ray the set of biorthogonally closed subspaces that contain this ray, we have proven that (R(Vω ), L(Vω ), νω ) is isomorphic to (Σω , Lω , ξω ).
8.2 Representation in classical Hilbert spaces Maria Pia Sol`er has proven that if Vω contains an inﬁnite orthonormal sequence, then K = R, C or H and Vω is the corresponding Hilbert space [Sol`er, 1995]. Holland has shown that it is enough to demand the existence of a nonzero λ ∈ K and an inﬁnite orthogonal sequence (en )n ∈ Vω such that < en , en >= λ for every n. To be precise, either (Vω , K, < ·, · >) or (Vω , K, − < ·, · >) is then a classical Hilbert space [Holland, 1995]. In [Aerts and van Steirteghem, 2000] we proved some alternatives to Sol`er’s result, by means of automorphisms of L(Vω ). PROPOSITION 106. Let (V, K, < ·, · >) be an orthomodular space and let L(V ) be the lattice of its closed subspaces. The following are equivalent: (1) (V, K, < ·, · >) is an inﬁnite dimensional Hilbert space over K = R, C or H. (2) V is inﬁnite–dimensional and given two orthogonal atoms x ¯, y¯ in L(V ), there is a unitary operator U such that U (¯ x) = y¯. (3) There exist a, b ∈ L(V ), where b is of dimension at least 2, and an ortholattice automorphism f of L(V ) such that f (a) a and f [0,b] is the identical map. (4) V is inﬁnite dimensional and given two orthogonal atoms x ¯, y¯ in L(V ) there exist distinct atoms x ¯1 , y¯2 and an ortholattice automorphism f of L(V ) such that x) = y¯. f [0,¯x1 ∨¯y2 is the identity and f (¯
Quantum Axiomatics
125
Condition (2) is Holland’s Ample Unitary Group axiom [Holland, 1995] and (3) is due to Mayet [Mayet, 1998]. Using the properties listed in section 2 of [Mayet, 1998], one can easily prove that (4) implies (2). THEOREM 107. Consider a state property space (Σ, L, ξ), and suppose that ax∨ ω∈Ω (Σω , Lω , ξω ) the ioms 24, 27, 43, 81, 93, 95 and 97 are satisﬁed. Consider ! decomposition of (Σ, L, ξ) in its non classical components. Each nonclassical component (Σω , Lω , ξω ), which has at least four orthogonal states, is isomorphic to the canonical state property space (Σ(Hω ), L(Hω ), ξHω ) where Hω are real, complex or quaternionic Hilbert spaces. Proof. An immediate consequence of proposition 106.
Theorem 107 proves that if axioms 24, 27, 43, 81, 93, 95 and 97 are satisﬁed, our theory reduces to standard quantum mechanics with super selection variables, and eventually on a quaternionic Hilbert space. BIBLIOGRAPHY [Aerts, 1981] D. Aerts. The One and the Many: Towards a Uniﬁcation of the Quantum and Classical Description of One and Many Physical Entities, Doctoral dissertation, Brussels Free University, 1981. [Aerts, 1982] D. Aerts. Description of many physical entities without the paradoxes encountered in quantum mechanics, Foundations of Physics, 12, 11311170, 1982. [Aerts, 1983] D. Aerts. Classical theories and non classical theories as a special case of a more general theory, Journal of Mathematical Physics, 24, 24412453, 1983. [Aerts, 1999a] D. Aerts. Foundations of quantum physics: a general realistic and operational approach, International Journal of Theoretical Physics, 38, 289358, 1999. Archive ref and link: quantph/0105109. [Aerts, 1999b] D. Aerts. Quantum mechanics: structures, axioms and paradoxes, in Quantum Mechanics and the Nature of Reality, eds. Aerts, D. and Pykacz, J., Kluwer Academic, Dordrecht, 1999. Archive ref and link: quantph/0106132. [Aerts, 2002] D. Aerts. Being and change: foundations of a realistic operational formalism, in Probing the Structure of Quantum Mechanics: Nonlinearity, Nonlocality, Probability and Axiomatics, eds. Aerts, D, Czachor, M. and Durt, T, World Scientiﬁc, Singapore, 2002. [Aerts et al., 1999] D. Aerts, E. Colebunders, A. Van der Voorde, and B. Van Steirteghem. State property systems and closure spaces: a study of categorical equivalence, International Journal of Theoretical Physics, 38, 359385, 1999. Archive ref and link: quantph/0105108. [Aerts et al., 2000] D. Aerts, E. Colebunders, A. Van der Voorde, and B. Van Steirteghem. On the amnestic modiﬁcation of the category of state property systems. Applied Categorical Structures, 10, 469480, 2000. [Aerts and van Steirteghem, 2000] D. Aerts and B. Van Steirteghem. Quantum axiomatics and a theorem of M. P. Sol`er, International Journal of Theoretical Physics, 39, 497502, 2000. Archive ref and link: quantph/0105107. [Aerts and Deses, 2002] D. Aerts and D. Deses. State property systems and closure spaces: extracting the classical and nonclassical parts. In D. Aerts, M. Czachor and T. Durt (Eds.), Probing the Structure of Quantum Mechanics: Nonlinearity, Nonlocality, Probability and Axiomatics (pp. 130148). Singapore: World Scientiﬁc, 2002. [Aerts and van Valckenborgh, 2002] D. Aerts and F. Valckenborgh. The linearity of quantum mechanics at stake: the description of separated quantum entities. In D. Aerts, M. Czachor and T. Durt (Eds.), Probing the Structure of Quantum Mechanics: Nonlinearity, Nonlocality, Probability and Axiomatics (pp. 2046). Singapore: World Scientiﬁc, 2002.
126
Diederik Aerts
[Aerts and Aerts, 2004] D. Aerts and S. Aerts. Towards a general operational and realistic framework for quantum mechanics and relativity theory. In A. C. Elitzur, S. Dolev and N. Kolenda (Eds.), Quo Vadis Quantum Mechanics? Possible Developments in Quantum Theory in the 21st Century (pp. 153208). New York: Springer, 2004. [Aerts and Deses, 2005] D. Aerts and D. Deses. State property systems and orthogonality. International Journal of Theoretical Physics, 44, 919929, 2005. [Aerts and Pulmannova, 2006] D. Aerts and S. Pulmannova. Representation of state property systems. Journal of Mathematical Physics, 47, 118, 2006. [Aerts et al., in press] D. Aerts, D. Deses, and A. Van der Voorde. Classicality and connectedness for state property systems and closure spaces. International Journal of Theoretical Physics, in press. [Birkhoﬀ and von Neumann, 1936] G. Birkhof and J. von Neumann. The logic of quantum mechanics, Annals of Mathematics, 37, 823843, 1936. [Faure and Fr¨ olicher, 1995] Cl.A. Faure and A. Fr¨ olicher. Dualities for inﬁnite–dimensional projective geometries, Geometriae Dedicata, 56, 225236, 1995. [Foulis et al., 1983] D. Foulis, C. Piron, and C. Randall. Realism, operationalism, and quantum mechanics, Foundations of Physics, 13, 813, 1983. [Holland, 1995] S. S. Holland, Jr. Orthomodularity in inﬁnite dimensions; a theorem of M. Sol`er, Bulletin of the American Mathematical Society, 32, 205234, 1995. [Keller, 1980] H. Keller. Ein nichtklassischer Hilbertscher Raum, Mathematische Zeitschrift, 172, 4149, 1980. [Mackey, 1963] G. Mackey. Mathematical Foundations of Quantum Mechanics, Benjamin, New York, 1963. [Maeda and Maeda, 1970] F. Maeda and S. Maeda. Theory of symmetric lattices, Springer–Verlag, Berlin, 1970. [Mayet, 1998] R. Mayet. Some characterizations of the underlying division ring of a Hilbert lattice by automorphisms, International Journal of Theoretical Physics, 37, 109114, 1998. [Mittelstaedt, 1963] P. Mittelstaedt. Philosophische Probleme der Modernen Physik, Bibliographisches Institut, Manheim, 1963. [Piron, 1964] C. Piron. Axiomatique quantique, Helvetica Physica Acta, 37, 439468, 1964. [Piron, 1976] C. Piron. Foundations of Quantum Physics, Benjamin, Massachusetts, 1976. [Randall and Foulis, 1983] C. Randall and D. Foulis. Properties and operational propositions in quantum mechanics, Foundations of Physics, 13, 835, 1983. [Sol`er, 1995] M. P. Sol`er. Characterization of Hilbert spaces by orthomodular spaces, Communications in Algebra, 23, 219243, 1995. [von Neumann, 1932] J. von Neumann. Mathematische Grundlagen der Quantenmechanik, Springer, Berlin, 1932. [Wigner, 1959] E. P. Wigner. Group Theory and its Applications to Quantum Mechanics of Atomic Spectra, Academic Press, New York, 1959.
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
127
QUANTUM LOGIC AND NONCLASSICAL LOGICS Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
1
INTRODUCTION
Classical logic is sometimes described as “the logic of an omniscient mind in a deterministic universe”. From an intuitive point of view the basic features of classical semantics can be summarized as follows: 1) any problem is semantically decided : for any sentence α, either α or its negation ¬α is true (excluded middle principle); at the same time, a sentence α and its negation ¬α cannot be both true (noncontradiction principle). 2) Meanings behave in a compositional way: the meaning of a compound expression is determined by the meanings of its parts. 3) Meanings are sharp and unambiguous. Some (possibly all) of these principles have been put in question by diﬀerent forms of nonclassical logic. In some signiﬁcant cases, the objective and descriptional notion of truth (which is characteristic of classical logic) has been replaced by an epistemic conception. Accordingly, truth has been identiﬁed with what is known by nonomniscient minds, acting in a universe that may be either deterministic or indeterministic. The ﬁrst choice is compatible with the intuitionistic approaches to logic and to mathematics, while the second choice represents the basic assumption of the quantum logical investigations. In both cases, the classical notion of truth has been replaced by the following relation: an information i forces us to assert the truth of a sentence α. One also brieﬂy says that the information i forces (or veriﬁes) the sentence α (and one usually writes: i = α). Should i represent a noncontradictory and complete informationsystem, our forcing relation would naturally collapse into the classical notion of truth. However, human information is generally incomplete and not necessarily consistent. As expected, in the case of physical theories, signiﬁcant pieces of information correspond to what is known by an observer about the physical systems under
128
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
investigation. In this connection, one usually speaks of physical states (brieﬂy, states). In the “happiest situations”, a state may represent a maximal knowledge of the observer: a piece of information that cannot be consistently extended to a richer information, in the framework of the theory. Even a hypothetical omniscient mind could not know more about the system in question (if the theory is correct). States of this kind are usually called pure states, both in classical and in quantum physics. Pieces of information that are not maximal are generally represented by mixtures of pure states (also called mixed states). There is an important diﬀerence that concerns the logical behavior of classical and of quantum pure states. In classical mechanics, maximality implies logical completeness: any pure state semantically decides any physical property (or event) that may hold for the system under investigation (in other words, the state attributes to the system either the property or its negation).This is in accordance not only with classical logic, but also with a number of important nonclassical logics (like intuitionistic logic), where any noncontradictory and maximal formal system is logically complete. Quantum pure states, instead, give rise to a somewhat “mysterious” divergence between maximality and logical completeness, which represents the origin of most logical anomalies of the quantum world. Although representing a maximal information, a quantum pure state is never logically complete. This is a consequence of Heisenberg’s uncertainty principle, according to which there are pairs of complementary events that cannot be simultaneously decided by any pure state. Both in classical and in quantum mechanics, physical states are represented by special kinds of mathematical objects. In classical mechanics (CM), a pure state of a single particle can be represented by a sequence of six real numbers (r1 , . . . , r6 ), where the ﬁrst three numbers correspond to the positioncoordinates, while the last ones are the momentumcomponents. The set IR6 of all sextuples of real numbers represents the phasespace for the particle in question. Similarly for the case of compound systems, consisting of a ﬁnite number n of particles. Hence, any pure state of a classical particlesystem is represented by a point of an appropriate phase space Σ. How to represent the physical events that may occur to a given particle? Following the standard ideas of classical (extensional) semantics, it is quite natural to assume that the such events are mathematically represented by suitable subsets of Σ. What about the structure of all events? As is well known, the power set of any set gives rise to a Boolean algebra. And also the set F(Σ) of all measurable subsets of Σ (which is more tractable than the full power set of Σ, from a measuretheoretic point of view) turns out to have a Boolean structure. Hence, we may refer to the following Boolean ﬁeld of sets: EV C = F(Σ), ∩, ∪, c , ∅, Σ , where the settheoretic operations ∩, ∪, c represent respectively the conjunction, the disjunction and the negation of classical events. As a consequence, the logic of CM turns out to be in perfect agreement with classical logic. Furthermore, pure states are logically complete: for any point p of the phasespace Σ and for any event E in F(Σ), either p ∈ E or p ∈ E c .
Quantum Logic and Nonclassical Logics
129
What happens in the case of quantum theory (QT)? As opposed to classical mechanics, QT is essentially probabilistic. A pure state generally assigns to a quantum event a probabilityvalue (a real number in the interval [0, 1]). As a consequence, a quantum event may be semantically indeterminate for a given pure state, and the excluded middle principle is violated. According to von Neumann’s axiomatization of QT, the mathematical interpretation of any quantum system is a complex separable Hilbert space H, which has, for QT, the role played by phasespaces in CM.1 Any pure state is mathematically represented by a unit vector ψ of H, while mixed states are represented by special operators called density operators. What about quantum events? After Birkhoﬀ and von Neumann’s celebrated article “The logic of quantum mechanics” (which is considered the birthdate of quantum logic) the mathematical representatives of quantum events are identiﬁed with the closed subspaces of the Hilbert space H associated to the quantum system S under investigation. Why are the mere subsets of H not adequate mathematical representatives for quantum events, as in the phasespace case? The reason depends on the superposition principle, which represents one of the basic dividing lines between the quantum and the classical case. As opposed to classical mechanics, in quantum mechanics, any unit vector, that is a linear combination of pure states, gives rise to a new pure state. Suppose two pure states ψ1 , ψ2 are orthogonal vectors and suppose that a pure state ψ is a linear combination (a superposition) of ψ1 , ψ2 . In other words: ψ = c1 ψ1 + c2 ψ2 (where c1 2 + c2 2 = 1). According to one of the basic axioms of QT (the so called Born rule), this means that a quantum system in state ψ might verify with probability c1 2 those events that are certain for state ψ1 (and are not certain for ψ) and might verify with are not certain probability c2 2 those events that are certain for state ψ2 (and for ψ). A similar relation also holds for inﬁnite superpositions i ci ψi . As a consequence, the mathematical representatives of events should be closed under ﬁnite and inﬁnite linear combinations. The closed subspaces of H are just the mathematical objects that can realize such a role. What will be the meaning of negation, conjunction and disjunction in the realm of quantum events? According to Birkhoﬀ and von Neumann: “The mathematical representative of the negative of any experimental proposition is the orthogonal complement of the mathematical representative of the proposition itself”. The orthogonal complement A of a closed subspace A is deﬁned as the set of all vectors that are orthogonal to all elements of A. In other words, ψ ∈ A iﬀ ψ ⊥ A iﬀ, for any ϕ ∈ A, the inner product ψϕ is 0.2 From the point of view of the physical interpretation, the orthogonal complement (called also orthocomplement) is particularly interesting, since it satisﬁes the following property: a pure state ψ assigns to an event A probability 1 (0, respectively) iﬀ ψ assigns to the orthocomplement of A probability 0 (1, respectively). As a consequence, one is 1 See 2 See
Deﬀ. 146, 148. Def.142.
130
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
dealing with an operation that inverts the two extreme probabilityvalues, which naturally correspond to the truthvalues Truth and Falsity (as in the classical truthtable of negation). As for conjunction, Birkhoﬀ and von Neumann point out that this can be still represented by the settheoretic intersection (as in the classical case). For, the intersection A ∩ B of two closed subspaces is again a closed subspace. Hence, we will obtain the usual truthtable for the connective and : ψ veriﬁes A ∩ B iﬀ ψ veriﬁes both members. Disjunction, however, cannot be represented here as a settheoretic union. For, generally, the union A ∪ B of two closed subspaces is not a closed subspace, except in special circumstances. In spite of this, we have at our disposal another good representative for the connective or : the supremum A∨B of two closed subspaces, that is the smallest closed subspace including both A and B. Of course, A ∨ B will include A ∪ B. As a consequence, we obtain the following structure: EV Q = C(H) , ∧ , ∨ , , 0 , 1 , where ∧ is the settheoretic intersection; ∨ , are deﬁned as above; while 0 and 1 represent, respectively, the null subspace (the singleton consisting of the null vector, which is the smallest possible subspace) and the total space H. The quantum event structure EV Q turns out to simulate a “quasiBoolean behavior”; however, it is not a Boolean algebra. Something very essential is missing. Conjunction and disjunction are not distributive. There are triplets of quantum events A, B, C such that: A ∧ (B ∨ C) = (A ∧ B) ∨ (A ∧ C). In fact, EV Q belongs to the variety of all orthocomplemented orthomodular lattices, that are not necessarily distributive.3 Structures of this kind are also called Hilbert lattices. By the onetoone correspondence between the set C(H) of all closed subspaces and the set Π(H) of all projectionoperators of H, the lattice based on Π(H) turns out to be isomorphic to the lattice based on C(H).4 Hence, quantum events can be equivalently represented either by closed subspaces or by projectionoperators of the Hilbert space associated to the quantum system under investigation.5 For any quantum system S, the pair (C(H), S) consisting of the set 3 See
Deﬀ.126 and 129. Deﬀ.157 and 149. 5 It is worthwhile recalling that von Neumann was not completely satisﬁed with the basic idea of the “quantum logical approach”, according to which the lattice of all closed subspaces of a Hilbert space should faithfully represent the structure of quantum events. The basic reason was the failure of modularity in the inﬁnite dimensional case (see Def.130). In fact, von Neumann considered this property as an essential condition for a frequencyinterpretation of quantum probabilities. An interesting analysis of von Neumann’s doubts and critiques can be found in [R´ edei, 1996]. 4 See
Quantum Logic and Nonclassical Logics
131
C(H) of all closed subspaces of H and of the set S of all possible (pure and mixed) states for S is usually called the eventstate system for S (which is equivalently represented also by the pair (Π(H), S)). Given a state, represented by a density operator ρ (∈ S) and an event, represented by a projectionoperator P (∈ Π(H)), the probability ρ(P ) that the physical system S in state ρ veriﬁes the event P is determined by the Bornrule, according to which: ρ(P ) = tr(ρP ), (where tr is the tracefunctional 6 ). A similarity space naturally arises in the framework of any eventstate system (C(H), S). DEFINITION 1. Two states ρ and σ are called similar (ρ ⊥ / σ) iﬀ there is no event A that is certain for ρ and impossible for σ (in other words, ρ(A) = 1 σ(A) = 0). One can easily show that ⊥ / is a a genuine similarity relation (reﬂexive and symmetric, but generally not transitive). Hence, any pair (S, ⊥ / ) represents an example of a similarity space. Any quantum similarity space (S, ⊥ / ) automatically determines a “twin space” (S, ⊥) (also called preclusivity space), where ⊥ is the negation of ⊥ / (in other words, ρ ⊥ σ iﬀ not ρ ⊥ / σ). Consider now the power set P(S) of the set of all states S. The preclusivity relation ⊥ permits one to deﬁne on P(S) a unary operation ⊥ (called the preclusive complement), which turns out to be a weak complement. For any set X of states: X ⊥ := {x ∈ S : ∀t ∈ X(s ⊥ t)} . The preclusive complement ⊥ satisﬁes the following properties for any sets X, Y of states: • X ⊆ X ⊥⊥ ; • X ⊆ Y implies Y ⊥ ⊆ X ⊥ ; • X ∩ X ⊥ = ∅. At the same time, the strong double negation principle (X ⊥⊥ ⊆ X) and the excluded middle principle ( X ∪ X ⊥ = S) generally fails. Consider now the map ⊥⊥ : P(S) → P(S) such that: X X ⊥⊥ , for any X ⊆ S. One can easily check that this map is a closure operator , satisfying the following conditions: ∅⊥⊥ = ∅; X ⊆ X ⊥⊥ ; X ⊥⊥ = X ⊥⊥⊥⊥ ; X ⊆ Y X ⊥⊥ ⊆ Y ⊥⊥ . 6 See
Def.158.
132
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
Consider then the set C(P(S)) of all closed elements of the power set of S. By deﬁnition, we have: X ∈ C(P(S)) iﬀ X = X ⊥⊥ . The elements of C(P(S)) are called closed sets of states. As we will see, such sets play a very signiﬁcant role for the semantics of quantum logic. Some important properties of the closed sets of states are described by the following lemmas, which refer to any eventstate system (C(H), S). THEOREM 2. The structure
C(P(S)) , ⊆ ,
⊥
, ∅, S
is a complete bounded ortholattice, where for any family {Xi }i∈I ⊆ C(P(S)): • the meet {Xi }i∈I exists and coincides with {Xi }i∈I ; • the join {Xi }i∈I exists and coincides with ( {Xi }i∈I )⊥⊥ ; • the preclusive complement
⊥
is an orthocomplement.7
The lattice C(P(S)) fails to be distributive. LEMMA 3. The lattice C(P(S)) is isomorphic to the lattice based on the set C(H) of all closed subspaces of H. LEMMA 4. Let X be any subset of S. Then, X is closed iﬀ X satisﬁes the following condition: ∀ρ[ρ ∈ X iﬀ ∀σ ⊥ / ρ∃τ ⊥ / σ(τ ∈ X)].
2 HOW QUANTUM LOGIC EMERGES FROM QUANTUM EVENTSTATE SYSTEMS Quantum similarity spaces have naturally suggested an abstract possible world semantics for quantum logic (ﬁrst proposed by Dishkant [Dishkant, 1972] and further developed by Goldblatt [Goldblatt, 1974]). Such semantics can be regarded as a “quantum variant” of the semantics proposed by Kripke for intuitionistic logic. Accordingly, one usually speaks of Kripkean semantics of quantum logic. As is well known, Kripkean models for intuitionistic logic are based on sets of possible worlds possibly correlated by an accessibility relation, which is reﬂexive and transitive. According to a canonical interpretation, the possible worlds of an intuitionistic Kripkean model, can be regarded as states of knowledge in progress. When a world j is accessible to another world i, the state of knowledge corresponding to j is more informative with respect to the state of knowledge represented by i. In this framework, knowledge is conservative: when a state of knowledge i 7 See
Def. 117126.
Quantum Logic and Nonclassical Logics
133
knows a given sentence, then all the states of knowledge that are accessible to i know the sentence in question. The Kripkean characterization of quantum logic is based on a diﬀerent intuitive idea. The possible worlds of a quantum logical model can be regarded as pieces of information that correspond to states of physical objects. What about the accessibility relation? This can be identiﬁed with the similarity relation that arises in a given eventstate system. From an intuitive point of view, one can easily understand the reason why semantic models based on similarity spaces are physically signiﬁcant. In fact, physical theories are not generally concerned with possible evolutions of states of knowledge with respect to a constant world (as happens in the case of intuitionistic logic), but rather with sets of physical situations that may be similar , where states of knowledge must single out some invariants. We will now brieﬂy sketch the basic concepts of an abstract possible world semantics for a weak form of quantum logic, that Dishkant had called minimal quantum logic, while Goldblatt’s preferred denomination is orthologic. This logic fails to satisfy an important property of quantum eventstructures: orthomodularity. Following Goldblatt’s terminology, we will distinguish orthologic (OL) from orthomodular quantum logic (OQL), which is often simply called quantum logic. The sentential language of both logics consists of sentential letters and of the following primitive connectives: ¬ (not), (and). The notion of sentence (or formula) is deﬁned in the expected way. We will use the following metavariables: p, q, ... for atomic sentences and α, β, γ, ... for sentences. The disjunction (or) is supposed to be deﬁned via de Morgan law (α β := ¬(¬α ¬β)). Consider an abstract similarity space (I, ⊥ / ), consisting of a nonempty set I of possible worlds and of a similarity relation ⊥ / (which is a reﬂexive and symmetric binary relation deﬁned on I). In semantic contexts, similarity relations are dealt with as special cases of accessibility relations, while similarity spaces are also called orthoframes. Given an orthoframe (I, ⊥ / ), we will use i, j, k, . . . as variables ranging over the set I of possible worlds. Like in the concrete Hilbertspace case, any abstract similarity space has a “twin space” that is an abstract preclusivity space. The preclusivity relation, that represents the negation of the accessibility relation ⊥, will be denoted by ⊥. In other words: i ⊥ j iﬀ not i ⊥ j. Whenever i ⊥ j we will say that j is inaccessible or orthogonal to i. On this basis, one can deﬁne a preclusive complement ⊥ on the power set P(I) of I: ∀X ⊆ I[X ⊥ := {i ∈ I : i ⊥ X}]. The following conditions hold: ⊥⊥
: P(I) → P(I) is a closure operator; • the structure C(P(I)) , ⊆ , ⊥ , ∅ , I based on the set of all closed subsets of • the map
134
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
I is an ortholattice.8 Hence, in particular,
⊥
is an orthocomplement;
• X is a closed subset of I iﬀ ∀i[i ∈ X iﬀ ∀j ⊥ i∃k ⊥ j(k ∈ X)]. In the framework of semantic applications, the closed subsets of I are usually called propositions of the orthoframe (I, ⊥). The following Lemma sums up some basic properties of propositions: LEMMA 5. Let (I, ⊥ / ) be an orthoframe. (i) I and ∅ are propositions; (ii) If X is any set of worlds, then X ⊥ is a proposition; (iii) If C is a family of propositions, then C is a proposition. The notion of Kripkean model for OL can be now deﬁned as follows: DEFINITION 6. Kripkean model for OL A Kripkean model for OL is a system K = I , ⊥ / , P r , V , where: (i) (I, ⊥ / ) is an orthoframe and P r is a set of propositions of the frame that contains ∅, I and is closed under the orthocomplement ⊥ and the settheoretic intersection ∩; (ii) V is a function that associates to any sentence α a proposition in P r, satisfying the following conditions: V (¬β) = V (β)⊥ ; V (β γ) = V (β) ∩ V (γ). Instead of i ∈ V (α), one usually writes i = α and reads: “the information i forces α to be true” (or brieﬂy, “α is true in the world i”). If T is a set of sentences, i = T will mean i = β for any β ∈ T . THEOREM 7. For any Kripkean model K and any sentence α: i = α iﬀ ∀j ⊥ i ∃k ⊥ j (k = α). LEMMA 8. In any Kripkean model K: (i) (ii)
i = ¬β iﬀ ∀j ⊥ i [j = / β]; i = β γ iﬀ i = β and i = γ.
On this basis, the notions of truth, logical truth, consequence, logical consequence are deﬁned in the expected way. DEFINITION 9. Truth and logical truth A sentence α is true in a model K = I , ⊥ / , P r , V (abbreviated =K α) iﬀ V (α) = I; 8 See
Def. 126.
Quantum Logic and Nonclassical Logics
135
α is a logical truth of OL (= α) iﬀ =K α for any model K. OL DEFINITION 10. Consequence in a model and logical consequence Let T be a set of sentences and let K be a model. A sentence α is a consequence in K of T (T =K α) iﬀ for any world i of K, i = T i = α. α) iﬀ for any model K, T =K α. A sentence α is a logical consequence of T (T = OL The set of propositions of any Kripkean model for OL gives rise to an ortholattice. As a consequence, Kripkean models for OL can be canonically transformed into algebraic models, where the meaning of any sentence is identiﬁed with an element of an ortholattice, while the connectives are interpreted as the corresponding latticeoperation. DEFINITION 11. Algebraic model for OL An algebraic model for OL is a pair A = B , v, consisting of an ortholattice B = B , ≤ , , 0 , 1 and a valuationfunction v that associates to any sentence α of the language an element in B, satisfying the following conditions: (i) v(¬β) = v(β) ; (ii) v(β γ) = v(β) ∧ v(γ). The notions of truth, logical truth, consequence and logical consequence are then deﬁned in the expected way. DEFINITION 12. Truth and logical truth A sentence α is true in a model A = B , v (abbreviated as =A α) iﬀ v(α) = 1; α) iﬀ for any algebraic model A = B , v, =A α. α is a logical truth of OL (= OL When for any β ∈ T, =A β, we will also write: =A T . DEFINITION 13. Consequence in a model and logical consequence Let T be a set of sentences and let A = B , v be a model. A sentence α is a consequence in A of T (T =A α) iﬀ for any element a of B: if for any β ∈ T , a ≤ v(β) then a ≤ v(α). α) iﬀ for any algebraic model A: A sentence α is a logical consequence of T (T = OL T =A α. One can prove that the Kripkean and the algebraic semantics characterize the same logic OL.9 In order to characterize orthomodular quantum logic (or quantum logic) one shall require a stronger condition both in the deﬁnition of Kripkean model and of algebraic model for OL: DEFINITION 14. Kripkean model for OQL A Kripkean model for OQL is a Kripkean model K = I , ⊥ / , P r , V for OL, where the set of propositions P r satisﬁes the orthomodular property: X ⊆ Y Y = X ∨ (X ⊥ ∧ Y ). 9 See
[Dalla Chiara and Giuntini, 2002].
136
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
DEFINITION 15. Algebraic model for OQL An algebraic model for OQL is an algebraic model A = B, v for OL, where B is an orthomodular lattice.10 We will indicate by QL either OL or OQL. Both logics are characterized by a deep asymmetry between conjunction and disjunction. By deﬁnition of Kripkean model, we have: • i = β γ iﬀ i = β and i = γ; • i = β γ iﬀ ∀j ⊥ i ∃k ⊥ j (k = β or k = γ). Hence, a disjunction may be true, even if both members are not true. A consequence of this asymmetry is the failure of the distributivity principle: α (β γ) = / QL (α β) (α γ). The semantic behavior of the quantum logical disjunction, which may appear prima facie somewhat strange, seems to reﬂect pretty well a number of concrete quantum situations. In quantum theory one is often dealing with alternatives that are semantically determined and true, while both members are, in principle, indeterminate. For instance, suppose we are referring to a spin onehalf particle (say an electron) whose spin in a certain direction may assume only two possible values: either up or down. Now, according to one of the uncertainty principles, the spin in the x direction (spinx ) and the spin in the y direction (spiny ) represent two incompatible quantities that cannot be simultaneously measured. Suppose an electron in state ψ veriﬁes the proposition “spinx is up”. As a consequence of the uncertainty principle both propositions “spiny is up” and “spiny is down” shall be indeterminate. However the disjunction “either spiny is up or spiny is down” must be true. As expected, the Kripkean models of OQL admit a quite natural realization in the framework of the Hilbert eventstate systems. Consider a quantum system S with associated Hilbert space H. Let (C(H), S(H)) be the eventstate system based on H. Consider now a sentential language LS for S, whose atomic sentences refer to quantum events in C(H). We can construct the following Kripkean model for S: / , P r , V , where: KS = I , ⊥ • I is the set S(H) of the states of S; • ⊥ / is the similarity relation that is deﬁned on S(H). In other words: i⊥ / j iﬀ not ∃E ∈ C(H)[i(E) = 1 and j(E) = 0]; • P r = C(P(S)) (= the set of all closed subsets of S(H)); 10 See
Def.129.
Quantum Logic and Nonclassical Logics
137
• for any atomic sentence p, V (p) is the closed subspace which p refers to. One immediately realizes that KS is a Kripkean model. For: • ⊥ / is a similarity relation (reﬂexive and symmetric); • P r is a set of propositions, because every element X of C(P(S)) is a closed set such that X = X ⊥⊥ . Furthermore, P r contains ∅ and I, and is closed under the operations ⊥ and ∩; • for any p, V (p) ∈ P r. Interestingly enough, the accessibility relation turns out to have the following physical meaning: i ⊥ / j iﬀ j is a state into which i can be transformed after the performance of a physical measurement that concerns an observable (i.e. a physical quantity) of the system (by application of von NeumannL¨ uders axiom, the so called “collapse of the wave function”). Let us now return to our general deﬁnition of Kripkean model for OQL. Apparently, orthomodularity has not been characterized in terms of properties of the accessibility relation. Hence, the following important question arises: is it possible to express the orthomodularity of the propositionstructure in an orthoframe (I, ⊥ / ) as an elementary (ﬁrstorder) property of the accessibility relation ⊥ /? In 1984, Goldblatt gave a negative answer to this question, proving that: orthomodularity is not elementary.11 Goldblatt’s theorem has revealed a kind of metalogical intractability of OQL. As a consequence of this negative result, properties like decidability and the ﬁnite model property (which are positively solved for OL) have stubbornly resisted to many attempts of solution in the case of OQL, and are still open problems. At the same time, OQL seems to have some logical advantages that are not shared by the weaker OL. For instance, interestingly enough, a conditional connective → turns out to be deﬁnable in terms of the primitive connectives of the quantum logical language. The most natural deﬁnition (originally proposed by Finch [1970] and Mittelstaedt [1972] and further investigated by Hardegree [1976] and other authors) is the following: α → β := ¬[α ¬(α β)] (which is equivalent to ¬α (α β)). 11 See
[Goldblatt, 1984].
138
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
In the quantum logical literature, such a conditional connective is often called Sasaki hook . Of course, in classical logic (by distributivity), the Sasaki hook is equivalent to the standard Philo’s conditional ¬α β. Notice that this classical conditional could not represent a “good” conditional for quantum logic, because it does not generally satisfy Modus Ponens. One can easily show that there are worlds i of a Kripkean model K such that: i = α; i = ¬α β; i = / β. The Sasaki hook, instead, turns out to be wellbehaved with respect to Modus Ponens, in the case of OQL (but not in the case of OL!). Both OL and OQL are axiomatizable logics. Many axiomatizations have been proposed: in the HilbertBernays style and in the Gentzenstyle (natural deduction and sequentcalculi).12 We present here a calculus (in the natural deduction style) which is a slight modiﬁcation of the version proposed by Goldblatt in 1974. This calculus (which has no axioms) is determined as a set of rules. Let T1 , . . . , Tn be ﬁnite or inﬁnite (possibly empty) sets of sentences. Any rule has the form T1 − α1 , . . . , Tn − αn T − α (if α1 has been inferred from T1 , . . . , αn has been inferred from Tn , then α can be inferred from T ). We call any expression of the form T − α a conﬁguration. The conﬁgurations T1 − α1 , . . . , Tn − αn represent the premisses of the rule, while T − α is the conclusion. As a limit case, we may have a rule in which the set of premisses is empty; in such a case we will speak of an improper rule. Instead of ∅ we will write T − α T − α; instead of ∅ − α, we will write − α. Rules of OL (OL1)
T ∪ {α} − α
(OL2)
T − α, T ∗ ∪ {α} − β T ∪ T ∗ − β
(OL3)
T ∪ {α β} − α
(elimination)
(OL4)
T ∪ {α β} − β
(elimination)
(identity) (transitivity)
12 An axiomatization of OQL in the HilbertBernays style see has been proposed by Hardegree in 1976 (see [Hardegree, 1976] ). Sequent calculi for diﬀerent forms of quantum logic have been investigated by Nishimura [1980] and by Battilotti and Sambin [1999]. See also [Battilotti and Faggian, 2002].
Quantum Logic and Nonclassical Logics
139
(OL5)
T − α, T − β T − α β
(introduction)
(OL6)
T ∪ {α, β} − γ T ∪ {α β} − γ
(introduction)
(OL7)
{α} − β, {α} − ¬β ¬α
(OL8)
T ∪ {α} − ¬¬α
(weak double negation)
(OL9)
T ∪ {¬¬α} − α
(strong double negation)
(OL10)
T ∪ {α ¬α} − β
(OL11)
{α} − β {¬β} − ¬α
(absurdity)
(Duns Scotus) (contraposition)
An axiomatization of OQL can be obtained by adding to the OLcalculus the following rule: (OQL)
α ¬(α ¬(α β)) − β.
(orthomodularity)
On this basis, all the standard syntactical notions (derivation, derivability, logical theorem) are deﬁned in the expected way. DEFINITION 16. Derivation A derivation of QL is a ﬁnite sequence of conﬁgurations T − α, where any element of the sequence is either the conclusion of an improper rule or the conclusion of a proper rule whose premisses are previous elements of the sequence. DEFINITION 17. Derivability A sentence α is derivable from T (T − QL α) iﬀ there is a derivation such that the conﬁguration T − α is the last element of the derivation. Instead of {α} − QL β we will write α − QL β. DEFINITION 18. Logical theorem A sentence α is a logical theorem of QL ( − QL α) iﬀ ∅ − QL α. A soundness and a completeness theorem have been proved for both logics with standard techniques (using the notion of canonical model )13 : THEOREM 19. Soundness theorem. T − QL α T =QL α. THEOREM 20. Completeness theorem T =QL α T − QL α. 13 See
[Dalla Chiara and Giuntini, 2002].
140
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
To what extent does orthomodular quantum logic represent a completely faithful abstraction from QT? As we have seen, the prototypical models of OQL that are interesting from the physical point of view are based on the class H of all Hilbert lattices. Let us call Hilbert quantum logic (HQL) the logic that is semantically characterized by H (both in the Kripkean and in the algebraic semantics). An important problem that has been discussed at length is the following: do OQL and HQL represent one and the same logic? In 1981 Greechie gave a negative answer to this question: there is an ortholatticetheoretical equation, the socalled orthoarguesian law 14 that holds in H, but fails in a particular orthomodular lattice. As a consequence, OQL does not represent a faithful logical abstraction from its quantum theoretical origin. The axiomatizability of HQL is still an open problem.
3 QUANTUM LOGIC VS POSITIVE, MINIMAL AND INTUITIONISTIC LOGIC Quantum logic represents a “singular point” in the variety of nonclassical logics, giving rise to a number of logical and metalogical anomalies. In order to understand some “strange” aspects of quantum logic a comparison with some logics that are at least as strong as positive logic (the positive fragment of intuitionistic logic) will be useful. Unlike QL, the language of positive logic (PL) contains as primitive connectives the conjunction , the disjunction and the conditional →. The notion of Kripkean model for PL is deﬁned as follows: DEFINITION 21. Kripkean model for PL A Kripkean model for PL is a system K = I , R , P r , V , where: (i) I is a nonempty set of possible worlds; (ii) R is a reﬂexive and transitive relation on I; (iii) P r is the set of all possible propositions; where a proposition is a set X of possible worlds that is Rclosed. In other words: X ∈ P r iﬀ X ⊆ I and ∀i, j ∈ I[i ∈ X and iRj j ∈ X]; (iv) V is a function that associates to any sentence α a proposition in P r, satisfying the following conditions: V (β γ) = V (β) ∩ V (γ); V (β γ) = V (β) ∪ V (γ); V (β → γ) = {i ∈ I : ∀j[iRj and j ∈ V (β) j ∈ V (γ)]}. We will write i = α instead of i ∈ V (α) (like in QL). The notions of truth, logical truth,consequence and logical consequence are then deﬁned as in the case of QL. 14 See
[Greechie, 1981]. See also [Kalmbach, 1983].
Quantum Logic and Nonclassical Logics
141
As is well known, two important nonclassical logics that are stronger than PL are minimal logic (ML) and intuitionistic logic (IL). Minimal logic can be characterized by adding to the language of LP a privileged atomic sentence f representing the Falsity. The negationconnective is then deﬁned as follows: ¬α := α → f . The Kripkean models of ML are deﬁned like in the case of PL. In any minimal model K, the proposition J = V (f ) (representing the meaning of the false sentence) is called the set of all absurd worlds. As a consequence, one immediately obtains: i = ¬α iﬀ ∀j[iRj and j = α j ∈ J]. Finally, intuitionistic logic is characterized by the class of all minimal models where the set J of all absurd worlds is empty. As a consequence, intuitionistic models turn out to satisfy the following condition: i = ¬α iﬀ ∀j[iRj j = / α]. We will now brieﬂy focus upon the basic diﬀerences between the quantum logical Kripkean models and the Kripkean models of PL, ML and IL (brieﬂy called knowledgein progress models). As we will see, these diﬀerences are responsible for the main logical and metalogical anomalies of QL. As we have seen, in any knowledgein progress model, the accessibility relation is reﬂexive and transitive, and propositions are closed with respect to the accessibility relation. As a consequence, truth turns out to be conservative: i = α and iRj j = α. Quantum logical models, instead, are based on similarity frames, where truth is not conservative and propositions X satisfy the following stability condition: i ∈ X iﬀ ∀j[i ⊥ / j ∃k(j ⊥ / k and k ∈ X)]. Interestingly enough, such condition turns out to characterize propositions also in the case of knowledgein progress models. In fact, one can prove that for any model K = I , R , P r , V and for any set of worlds X, the two following conditions are equivalent: I) X is Rclosed; II) i ∈ X iﬀ ∀j[iRj ∃k(jRk and k ∈ X)]. Of course, condition I) and II) are not equivalent in the case of QL. Let us now focus on the diﬀerent truthconditions for the logical connectives. Negation The truthcondition for the connective ¬ turns out to be the same for QL and for IL:
142
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
i = ¬α iﬀ for any world j accessible to i, j = / α. In both cases, one is dealing with a noncompositional situation: the truth of a negated sentence ¬α in a given world i essentially depends on the truthstatus of the positive sentence α in all the worlds that are accessible to i. Conjunction Conjunction is the only (primitive) connective that has a truthfunctional (compositional) behavior both in the quantum logical and in the knowledgein progress models: a conjunction is true in a given world i iﬀ both members are true in i (like in classical semantics). Disjunction The most important divergence between quantum logical and knowledgein progress models concerns the behavior of disjunction. The quantum logical disjunction is essentially noncompositional: a world i may verify a disjunction α β even if both members of the disjunction (α, β) are indeterminate for i. The truthstatus of α β in a given world depends on the truthstatus of α and β in other worlds, according to the following condition: i = α β iﬀ ∀j ⊥ / i∃k ⊥ / j[k = α or k = β]. The positive (minimal and intuitionistic) disjunction is, instead, truthfunctional (like in classical semantics): i = α β iﬀ i = α or i = β. It is worthwhile recalling that intuitionistic disjunction is strongly prime even from the prooftheoretic point of view. For, α β is an intuitionistic logical theorem iﬀ either α or β is an intuitionistic logical theorem. As is well known (unlike classical and quantum disjunction) the intuitionistic disjunction is not deﬁnable in terms of negation and of conjunction (via de Morgan law). Of course, a de Morgan disjunction dM can be trivially deﬁned also in the framework of IL: α dM β := ¬(¬α ¬β). We have: =IL α β → α dM β, but not the other way around. Apparently, the truth of the intuitionistic de Morgan disjunction, does not imply the truth of at least one member of the disjunction (as happens in the case of QL). This failure of truthfunctionality, however, is not suﬃcient to bring about a violation of distributivity. One can show that, in spite of their asymmetrical behavior, the intuitionistic connectives and dM do satisfy the distributivityrelations. In fact, quantum logical nondistributivity seems to be essentially connected with the nontransitive character of the accessibility relation.
Quantum Logic and Nonclassical Logics
143
Conditional The positive (minimal and intuitionistic) conditional is a primitive connective, whose behavior is governed by the following (nontruth functional) condition: i = α → β iﬀ ∀jRi[j = α j = β]. The quantum logical conditional is, instead, deﬁned in terms of ¬ and of : α → β := ¬[α ¬(α β)]. As a consequence one obtains that: i = α → β iﬀ ∀j ⊥ / i[j = α ∃k ⊥ / j(k = α and k = β)]. Such a condition (which is clearly weaker than the corresponding condition for the positive conditional) is responsible for most anomalies of the quantum logical conditional. For instance, the following laws that hold for positive conditionals are violated in OQL: α → (β → α); (α → (β → γ)) → ((α → β) → (α → γ)); (α → β) → ((β → γ) → (α → γ)); (α β → γ) → (α → (β → γ)); (α → (β → γ)) → (β → (α → γ)). Another interesting characteristic of the quantum logical conditional is a weak nonmonotonic behavior . For, we have: α → γ = / QL α β → γ. In 1975 Hardegree15 has suggested that such anomalous aspects might be explained by conjecturing that the quantum logical conditional represents a kind of counterfactual conditional. This hypothesis seems to be conﬁrmed by some signiﬁcant physical examples. Let us consider again the Kripkean models that are associated to a quantum system S. Following Hardegree, we restrict our attention to the case of pure states. As a consequence, we consider Kripkean models having the following form: / , P r , V , where : KS = I , ⊥ • I is the set of all pure states of S; • ⊥ / is the nonorthogonality relation deﬁned on I; 15 See
[Hardegree, 1975].
144
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
• P r is the set of all pure propositions of the eventstate system (C(H), S(H)). In other words: Z ∈ P r iﬀ Z is a closed set of pure states (i.e., such that Z = Z ⊥⊥ ); • V (p) is the pure proposition consisting of all pure states that assign probabilityvalue 1 to the quantum event which p refers to. Hardegree has shown that, in such a case, the conditional → turns out to receive a quite natural counterfactual interpretation (in the sense of Stalnaker 16 ). More precisely, one can deﬁne, for any sentence α of the language LS , a partial Stalnakerfunction fα in the following way: fα : Dom(fα ) → I, where: Dom(fα ) := {i ∈ I : i ⊥ V (α)} . In other words, fα is deﬁned exactly for all the pure states that are not orthogonal to the proposition of α. If i ∈ Dom(fα ), then: fα (i) := P V (α) i, where P V (α) is the projection determined by the closed subspace that is uniquely associated with the pure proposition V (α) . The following condition holds: i = α → β iﬀ either ∀j ⊥ i(j = / α) or fα (i) = β. From an intuitive point of view, one can say that fα (i) represents the “pure state nearest” to i, that veriﬁes α, where “nearest” is here deﬁned in terms of the metric of the Hilbert space H. By deﬁnition and in virtue of von Neumann L¨ uders axiom (the collapse of the wavefunction), fα (i) turns out to have the following physical meaning: it represents the transformation of state i after the performance of a measurement concerning the physical event expressed by α, provided the result was positive. As a consequence, one obtains: α → β is true in a state i iﬀ either α is impossible for i or the state into which i has been transformed after a positive αtest, veriﬁes β. As we have seen, the minimal and the intuitionistic negation can be deﬁned (in terms of → and of f ). Is such a deﬁnition possible also in the case of OQL? Suppose that the language of OQL contains an atomic sentence f such that for any orthomodular model K = I, ⊥ / , P r, V and any world i, i = / f . One can easily show that the following equivalence holds (for any K and for any i): i = ¬α iﬀ i = α → f . Hence, an alternative description of OQL might assume as primitive logical constants the false sentence f and the positive connectives and →. Then the 16 See
[Stalnaker, 1981].
Quantum Logic and Nonclassical Logics
145
following conditions should be required for any orthomodular model K and for any world i of K: i = / f; i = β γ iﬀ i = β and i = γ; i = β → γ iﬀ ∀j ⊥ / i[j = β ∃k ⊥ / j(k = β and k = γ)]. On this basis, both the quantum logical negation and the quantum logical disjunction can be dealt with as deﬁned logical constants. Needless to stress, such a construction would not be possible in the case of orthologic, that does not admit a wellbehaved conditional. 4
THE “LINDENBAUMANOMALY”
The consistencyproperty behaves diﬀerently in the framework of diﬀerent logics. Let L represent either QL or ML or IL. The following concepts characterize diﬀerent semantic aspects of the intuitive idea of consistency. Let T be a set of sentences of the logic L. DEFINITION 22. • T is veriﬁable in L iﬀ there is an Lmodel K such that =K T . • T is realizable in L iﬀ there is an Lmodel K and a world i of K such that i = T . • T is semantically noncontradictory in L iﬀ for any sentence α, T L α ¬α. • T is semantically αconsistent in L iﬀ T L α. • T is semantically consistent in L iﬀ there exists a sentence α such that T is semantically αconsistent. As is well known, IL gives rise to a strong relation between these diﬀerent aspects of the intuitive notion of consistency (relation that also holds in the case of CL). THEOREM 23. The following conditions are equivalent for IL: • T is veriﬁable. • T is realizable. • T is semantically noncontradictory. • T is semantically consistent. • T is semantically f consistent.
146
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
The equivalence between semantic noncontradictory and semantic consistent depends on the fact that intuitionistic logic is a scotian logic, where “ex absurdo sequitur quodlibet” (α ¬α =IL β and f =IL β). Minimal logic, instead, is an / ML β). important example of a nonscotian logic (for, α ¬α = / ML β and f = At the same time, the noncontradiction principle, which is an immediate consequence of Modus Ponens ( α (α → f ) → f ) holds for ML. Furthermore, the sentence f turns out to be equivalent to the contradictory sentence f ¬f (:= f (f → f )) and to any contradiction expressed in a negative form (¬α ¬¬α). As a consequence, the following weaker theorem holds for ML: THEOREM 24. • T is veriﬁable iﬀ T is realizable iﬀ T is semantically noncontradictory iﬀ T is f consistent. • If T is semantically noncontradictory, then T is semantically consistent, but generally not the other way around. On this basis, one can prove that both ML and IL satisfy a strong metalogical condition, that we will call preLindenbaum. Let L be either ML or IL. LEMMA 25. PreLindenbaum Lemma If T is semantically noncontradictory in L and T = / L ¬α, then T ∪ {α} is semantically noncontradictory in L. An important consequence of the preLindenbaum Lemma is a strong relation between two relevant metalogical properties: logical completeness and maximality. DEFINITION 26. • T is logically complete in the logic L iﬀ for any sentence α, either T =L α or T =L ¬α. • T is maximal in the logic L iﬀ for any sentence α, the set T ∪ {α} is semantically contradictory, if T = / L ¬α. Maximality and logical completeness turn out to be equivalent both in ML and in IL. THEOREM 27. A set T of sentences is maximal in L iﬀ T is logically complete in L. The proof of the lefttoright implication essentially uses the preLindenbaum Lemma. On this basis, both a weak and a strong version of the Lindenbaumtheorem can be proved for minimal and intuitionistic logic. Let L be either ML or IL. THEOREM 28. Weak Lindenbaum For any set of sentences T that is semantically noncontradictory in L, there exists a set of sentences T such that:
Quantum Logic and Nonclassical Logics
147
• T ⊆ T ; • T is semantically noncontradictory in L; • T is maximal in L. As is well known, the standard proof of the Lindenbaumtheorem refers to an enumeration of the sentences of the language (β1 , β2 , . . . ). On this basis, an inﬁnite sequence of set of sentences is deﬁned: T0 = T Tn ∪ {βn+1 } , if Tn ∪ {βn+1 } is semantically noncontradictory; Tn+1 = Tn , otherwise. By putting T = n Tn , one can prove that T is noncontradictory and maximal.17 By Lemma 25 one immediately obtains that also the strong version of the Lindenbaumtheorem holds for L. THEOREM 29. Strong Lindenbaum For any set of sentences T that is semantically noncontradictory in L, there exists a set of sentences T such that: • T ⊆ T ; • T is semantically noncontradictory in L; • T is logically complete in L. In other words, any noncontradictory set of sentences T can be extended to a set T , where any problem (expressed in the language) is semantically decided . What happens in the case of QL? As we already know, QL is a scotian logic (where α ¬α = β). Hence, semantically contradictory and semantically consistent collapse into one and the same concept (like in IL and in CL). At the same time (unlike ML and IL) veriﬁability and realizability split into two diﬀerent concepts. THEOREM 30. In QL: I) T is semantically noncontradictory iﬀ T is realizable. II) If T is veriﬁable, then T is realizable, but generally not the other way around. Proof. The proof of I) is trivial. As to II), veriﬁability trivially implies realizability. A counterexample to the opposite implication is given by the following sentence γ (which represents the negation of the afortiori principle): γ = ¬[α → (β → α)] = ¬¬[α ¬(α ¬(β ¬(β α)))]. 17 Of course, the proof makes essential use of the completeness theorem for the logic L (or, alternatively, of the semantic compactness theorem).
148
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
Consider a Kripkean model KS = I, ⊥ / , P r, V , associated to a quantum system S whose stateevent system is (S(H), C(H)), where H is the twodimensional Hilbert space C2 . Suppose that α and β are two atomic sentences such that V (α) and V (β) are propositions corresponding to two diﬀerent nonorthogonal onedimensional closed subspaces of H. One immediately obtains that: V (γ) = V (α) = ∅. Hence, there exists a pure state of S that veriﬁes γ, and consequently γ is realizable. At the same time, γ cannot be veriﬁable. For, one can easily show that the hypothesis that there exists a model K = I, ⊥ / , P r, V such that V (γ) = I, leads to a contradiction. As a consequence, one immediately obtains the failure of the strong Lindenbaumproperty for QL. Suppose, by contradiction, that any semantically noncontradictory T admits a semantically noncontradictory and logically complete extension T . One can easily show that any semantically noncontradictory and logically / , P r, V such complete T is veriﬁable by a classical Kripkean model K = I, ⊥ that: • I is a singleton set {i}, where i is any object; • ⊥ / is the identity relation on I; • P r is the powerset of I; {i} , if T =QL p; • V (p) = ∅, otherwise. Since T is semantically noncontradictory and logically complete, K is well deﬁned. Moreover we have: i = α iﬀ =K α iﬀ T =QL α. Consequently, T is veriﬁable. Although the strong version of the Lindenbaumtheorem is violated in QL, the weak version of the theorem is valid. In other words, any semantically noncontradictory set of sentences T can be extended to a set T that is semantically noncontradictory and maximal. The proof runs like in the case of ML and of IL. However (unlike ML, IL and CL) maximality does not imply logical completeness in QL. THEOREM 31. If T is logically complete (in QL), then T is maximal (in QL), but generally not the other way around. Proof. The proof that logical completeness implies maximality is trivial. Maximality, instead, cannot imply logical completeness. For, as we have seen, there are examples of semantically noncontradictory sentences γ that admit a maximal noncontradictory extension and do not admit any noncontradictory logically complete extension.
Quantum Logic and Nonclassical Logics
149
Also the preLindenbaum property is violated in QL. A counterexample can be obtained as follows. Put T = {β} and α = ¬(γ → β). We have β = / QL ¬α, because β = / QL γ → β, as shown by counterexamples in C2 . At the same time, T ∪ {α} is clearly semantically contradictory. Only the following weak preLindenbaum condition holds for QL. THEOREM 32. If T is semantically noncontradictory and T = / QL α, then there exists a set of sentences T such that: 1) T =QL ¬α; 2) T and T are logically compatible. In other words, for any sentence β, if / QL ¬β.18 T =QL β, then T =
5 QUANTUM LOGIC AND THE HIDDEN VARIABLE PROBLEM The failure of the Lindenbaum property in QL has represented a powerful metalogical tool that has been used to prove the impossibility of completing QT via some (noncontextual) hidden variable hypotheses.19 The debate concerning the question whether QT can be considered a physically complete account of microphenomena has a long and deep history. A turning point in this discussion has been the celebrated EinsteinBohr debate, with the ensuing charge of incompleteness raised by the EinsteinPodolskyRosen argument (EPR). As we already know, in the framework of orthodox QT, physical systems can be prepared in pure states that have, in general, positive dispersion for most physical quantities. In the EPR argument, the attention is focused on the question whether the account of the microphysical phenomena provided by QT is to be regarded as an exhaustive description of the physical reality to which those phenomena are supposed to refer, a question to which Einstein himself answered in the negative. There is a mathematical side of the completeness issue: the question becomes whether states with positive dispersion can be represented as a diﬀerent, dispersionfree, kind of states in a way that is consistent with the mathematical constraints of the quantum theoretical formalism. In his book on the mathematical foundations of quantum mechanics, von Neumann proved a celebrated “No go theorem” asserting the logical incompatibility between the quantum formalism and the existence of dispersionfree states (satisfying some general conditions). Already in the preface, von Neumann anticipates the program and the conclusion concerning the possibility of ‘neutralizing’ the statistical character of QT: There will be a detailed discussion of the problem as to whether it is possible to trace the statistical character of quantum mechanics to an ambiguity (i.e., incompleteness) in our description of nature. Indeed, 18 See 19 See,
[Dalla Chiara and Giuntini, 2002]. for instance, [Giuntini, 1991b].
150
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
such an interpretation would be a natural concomitant of the general principle that each probability statement arises from the incompleteness of our knowledge. This explanation “by hidden parameters” [...] has been proposed more than once. However, it will appear that this can scarcely succeed in a satisfactory way, or more precisely, such an explanation is incompatible with certain qualitative fundamental postulates of quantum mechanics. According to the advocates of hidden variables, QT is a physically incomplete theory. The intuitive idea that represents the common background to almost all hidden variable theories can be described in the following way: (I) the reason why a physical theory is statistical depends on the fact that the description provided by the states is incomplete. (II) It is possible to add a set Ξ of parameters (hidden variables) in such a way that • for every state s and for every ω ∈ Ξ, there exists a dispersionfree (dichotomous) state sω which semantically decides every property (event) of the physical system at issue; • the statistical predictions of the original theory should be recovered by averaging over these dichotomous states; • the algebraic structures determined by the properties (events) of the system should be preserved in the hidden variable extension. The hidden variable theories based on the assumptions (I) and (II) are usually called noncontextual , because they require the existence of a single space Ξ of hidden variables determining dispersionfree states. A weaker position is represented by the contextual hidden variable theories, according to which the choice of the hidden variable space depends on the physical quantity to be dealt with. As pointed out by Beltrametti and Cassinelli [1981]: Despite the absence of mathematical obstacles against contextual hidden variable theories, it must be stressed that their calling for completed states that are probability measures not on the whole proposition [event] lattice E but only on a subset of E is rather far from intuitive physical ideas of what a state of a physical system should be. Thus, contextual hidden variable theorists, in their search for the restoration of some classical deterministic aspects, have to pay, on other sides, in quite radical departures from properties of classical states. Von Neumann’s proof of his “No go theorem” was based on a general assumption that has been, later, considered too strong. The condition asserts the following: Let sω be a dispersionfree state and let A, B be two (possibly noncompatible) observables. Then, Exp(A + B, sω ) = Exp(A, sω ) + Exp(B, sω ).
Quantum Logic and Nonclassical Logics
151
In other words, the expectation functional Exp determined by the completed state sω is linear. In the late Sixties, Kochen and Specker published a series of articles, developing a purely logical argument for a “No go theorem,” such that von Neumann’s strong assumption can be relaxed.20 Kochen and Specker’s proof is based on a variant of quantum logic, that has been called partial classical logic (PCL). The basic semantic idea is the following: unlike orthologic and orthomodular quantum logic (which are total logics, because the meaning of any sentence is always deﬁned), molecular sentences of PCL can be semantically undeﬁned. From the semantic point of view, the crucial relation is represented by a compatibility relation, that may hold between the meanings of two sentences. As expected, the intended physical interpretation of the compatibility relation is the following: two sentences α and β have compatible meanings iﬀ α and β can be simultaneously tested. Models of PCL are special kinds of algebraic models based on partial Boolean algebras (weaker versions of Boolean algebras where the meet and the join are only deﬁned for pairs of compatible elements). All these investigations have revealed that there is a deep logical connection between the two following questions: • does a quantum system S admit a noncontextual hidden variable theory? • Does PCL satisfy a version of the Lindenbaum property with respect to the algebraic models concerning the events that may occur to the system S?
6 THE UNSHARP APPROACHES TO QT The essential indeterminism of QT gives rise to a kind of ambiguity of the quantum world. Such ambiguity can be investigated at diﬀerent levels. The ﬁrst level concerns the characteristic features of quantum pure states, which represent pieces of information that are at the same time maximal and logically incomplete. As we have seen, the divergence between maximality and logical completeness is the origin of most logical anomalies of the quantum phenomena. A second level of ambiguity is connected with a possibly fuzzy character of the physical events that are investigated. We can try and illustrate the diﬀerence between two “fuzzinesslevels” by referring to a nonscientiﬁc example. Let us consider the two following sentences, which apparently have no deﬁnite truthvalue: I) Hamlet is 1.70 meters tall; II) Brutus is an honourable man. The semantic uncertainty involved in the ﬁrst example seems to depend on the logical incompleteness of the individual concept associated to the name “Hamlet.” 20 See
Kochen and Specker [1965a; 1965b; 1967].
152
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
In other words, the property “being 1.70 meters tall” is a sharp property. However, our concept of Hamlet is not able to decide whether such a property is satisﬁed or not. Unlike real persons, literary characters have a number of indeterminate properties. On the contrary, the semantic uncertainty involved in the second example, is mainly caused by the ambiguity of the concept “honourable.” What does it mean “being honourable?” One need only recall how the ambiguity of the adjective “honourable” plays an important role in the famous Mark Anthony’s monologue in Shakespeare’s “Julius Caesar.” Now, orthodox QT generally takes into consideration examples of the ﬁrst kind (our ﬁrst level of fuzziness): events are sharp, while all semantic uncertainties are due to the logical incompleteness of the individual concepts, that correspond to pure states of quantum objects. This is the reason why orthodox QT is sometimes called sharp QT, in contrast with unsharp QT, which also investigates examples of the second kind (second level of fuzziness). Strangely enough, the abstract researches on fuzzy logics and on quantum structures have undergone quite independent developments for many decades during the 20th century. Only after the Eighties, there emerged an interesting convergence between the investigations about fuzzy and quantum structures, in the framework of the so called unsharp approach to quantum theory. In this connection a signiﬁcant conjecture has been proposed: perhaps some apparent mysteries of the quantum world should be described as special cases of some more general fuzzy phenomena, whose behavior has not yet been fully understood. In 1983 the German physicist G. Ludwig published the book Foundations of Quantum Mechanics, which has been later regarded as the birth of the unsharp approach to QT. Paradoxically enough, Ludwig has always been an “enemy” of quantum logic. In spite of this, his ideas have greatly contributed to the revival of the quantum logical investigations during the last two decades. Ludwig’s pioneering work has been further developed by many scholars (Kraus, Davies, Mittelstaedt, Busch, Lahti, Bugajski, Beltrametti, Nistic` o, Foulis, Bennett, Gudder, Greechie, Pulmannov´ a, Dvureˇcenskij, Rieˇcan, Rieˇcanova, Schroeck and many others including the authors of this chapter). The starting point of the unsharp approach is deeply connected with a general problem that naturally arises in the framework of Hilbert space QT. Let us consider the eventstate system (Π(H) , S(H)) of a quantum system S, where Π(H) is the set of projections, while S(H) is the set of density operators of the Hilbert space H (associated to S). One can ask the following question: do the sets Π(H) and S(H) correspond to an optimal possible choice of adequate mathematical representatives for the intuitive notions of event and of state, respectively? Consider ﬁrst the notion of state. Once Π(H) is ﬁxed, Gleason’s Theorem21 guarantees that any probability measure deﬁned on Π(H) is determined by a density operator of H (provided the dimension of H is greater than or equal to 3). Hence, S(H) corresponds to an optimal notion of state. Let us discuss then the notion of event and let us ask whether Π(H) represents 21 See
[Gleason, 1957] and [Dvureˇ censkij, 1993].
Quantum Logic and Nonclassical Logics
153
the largest set of operators assigned a probabilityvalue, according to the Born rule. The answer to this question is negative. One can easily recognize the existence of bounded linear operators E that are not projections and that satisfy the following condition: for any density operator ρ, tr(ρE) ∈ [0, 1].22 Recalling the Born rule, this means that such operators E “behave as possible events,” because any state assigns to them a probability value. An interesting example of this kind is represented by the operator 12 I (where I is the identity operator). One immediately realizes that 12 I is a linear bounded operator that is not a projection, because: 1 1 1 1 I I = I = I 2 2 4 2 (hence 12 I fails to be idempotent). At the same time, for any density operator ρ we have: 1 1 tr(ρ I) = . 2 2 Thus, 12 I seems to represent a totally indeterminate event, to which each state assigns probability 12 . Apparently, the event 12 I plays the role that, in fuzzy set theory, is played by the semitransparent fuzzy set 12 1 such that for any object x of the universe: 1 1 1(x) = . 2 2 This situation suggests that we liberalize the notion of quantum event and extend the set Π(H) to a new set of operators. Following Ludwig, the elements of this new set have been called eﬀects. The precise mathematical deﬁnition of eﬀect is the following: DEFINITION 33. Eﬀects An eﬀect of H is a bounded linear operator E that satisﬁes the following condition, for any density operator ρ: tr(ρE) ∈ [0, 1]. We denote by E(H) the set of all eﬀects of H. Clearly, E(H) properly includes Π(H). Because: • any projection satisﬁes the deﬁnition of eﬀect; • there are examples of eﬀects that are not projections (for instance the eﬀect 1 2 I, that is usually called the semitransparent eﬀect). 22 See
Deﬀ. 152, 153, 158.
154
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
By deﬁnition, eﬀects turn out to represent a kind of maximal mathematical representative for the notion of quantum event, in agreement with the basic statistical rule of QT (the Born rule). Unlike projections, eﬀects represent quite general mathematical objects that describe at the same time events and states. Let E be any eﬀect in E(H). The following conditions hold: • E represents a sharp event (∈ Π(H)) iﬀ E is idempotent (EE = E); • E is a density operator (representing a state) iﬀ tr(E) = 1; • E represents a pure state iﬀ E is at the same time a projection and a density operator.
6.1 Algebraic eﬀectstructures There are diﬀerent algebraic structures that can be induced on the set of all eﬀects in a Hilbert space. One immediately realizes that the set E(H) can be naturally structured as a regular involution bounded poset 23 : E(H) , ≤ , , 0 , 1 , where (i) ≤ is the natural order determined by the set of all density operators. In other words: E ≤ F iﬀ for any density operator ρ ∈ D(H), tr(ρE) ≤ tr(ρF ). (i.e., any state assigns to E a probabilityvalue that is less or equal than the probabilityvalue assigned to F ); (ii) E = 1 − E (where − is the standard operator diﬀerence); (iii) 0, 1 are the null projection (O) and the identity projection (I), respectively. One can easily check that: • ≤ is a partial order; •
is an involution;
• 0 and 1 are respectively the minimum and the maximum with respect to ≤; • the regularity condition holds. In other words: E ≤ E and F ≤ F implies E ≤ F . 23 See
Def. 117 124.
Quantum Logic and Nonclassical Logics
155
The eﬀect poset E(H) turns out to be properly fuzzy. The noncontradiction principle is violated: for instance the semitransparent eﬀect 12 I satisﬁes the following condition: 1 1 1 1 1 I ∧ ( I) = I ∧ I = I = 0. 2 2 2 2 2 This is one of the reasons why proper eﬀects (those that are not projections) may be regarded as representing unsharp physical events. Accordingly, we will also call the involution operation of an eﬀectstructure a fuzzy complement. At the same time, the eﬀectposet fails to be a lattice. As proved by Greechie and Gudder in 1996, some pairs of eﬀects have no meet.24 The eﬀect poset E(H) can be expanded to a richer structure, equipped with a new complement ∼ , that has an intuitionisticlike behavior. Such operation ∼ has been called the Brouwer complement.25 DEFINITION 34. The Brouwer complement ∀E ∈ E(H) : E ∼ = PKer(E) . In other words, the Brouwer complement of E is the projection operator PKer(E) whose range is Ker(E), the kernel of E.26 By deﬁnition, the Brouwer complement of an eﬀect is always a projection. In the particular case, when E is a projection, it turns out that E = E ∼ , in other words, the fuzzy and the intuitionistic complement collapse into one and the same operation. The structure E(H) , ≤ , , ∼ , 0 , 1 turns out to be a particular example of a kind of abstract structure that has been termed Brouwer Zadeh poset.27 The abstract deﬁnition of Brouwer Zadeh posets is the following: DEFINITION 35. Brouwer Zadeh poset A Brouwer Zadeh poset (or BZposet) is a structure B , ≤ , ,
∼
, 0 , 1 ,
where (i) B , ≤ , , 0 , 1 is a regular poset; (ii)
∼
is a unary operation that behaves like an intuitionistic complement:
(iia) a ∧ a∼ = 0; (iib) a ≤ a∼∼ ; (iic) a ≤ b implies b∼ ≤ a∼ . 24 See
[Gudder and Greechie, 1996]. [Cattaneo and Nistic´ o, 1989]. 26 The kernel of E is the set of all vectors of H that are transformed by E into the null vector. 27 See [Cattaneo and Nistic´ o, 1989]. 25 See
156
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
(iii) The following relation connects the fuzzy and the intuitionistic complement: a∼ = a∼∼ . Of course, any BZposet B , ≤ , , ∼ , 0 , 1 where the two complements and coincide, turns out to be an orthoposet (i.e. a bounded involution poset, where the involution satisﬁes the noncontradiction and the excluded middle principles). One can prove that the concrete eﬀectstructure
∼
E(H) , ≤ , ,
∼
, 0 , 1
is a Brouwer Zadeh poset, that is not an orthoposet. An interesting feature of the Brouwer Zadeh structures is the possibility to deﬁne two unary operations ν and μ, which turn out to behave as the modal operators necessarily and possibly, respectively. DEFINITION 36. The modal operators Let B , ≤ , , ∼ , 0 , 1 be a Brouwer Zadeh poset. ν(a) := a∼ ; μ(a) := a∼ . In other words, necessity is identiﬁed with the intuitionistic negation of the fuzzy negation, while possibility is identiﬁed with the fuzzy negation of the intuitionistic negation. The modal operators ν and μ turn out to have a typical S5 like behavior. For, the following conditions are satisﬁed: • ν(a) ≤ a Necessarily a implies a. • If a ≤ b, then ν(a) ≤ ν(b) If a implies b, then the necessity of a implies the necessity of b. • a ≤ ν(μ(a)) a implies the necessity of its possibility. • ν(ν(a)) = ν(a) Necessity is equivalent to the necessity of the necessity. • ν(μ(a)) = μ(a) The necessity of a possibility is equivalent to the possibility. Of course, in any BZposet B , ≤ , , ∼ , 0 , 1 where the two complements and ∼ coincide, we obtain a collapse of the modalities. In other terms, ν(a) = a = μ(a). Let us now return to concrete (Hilbertspace) Brouwer Zadeh posets E(H) , ≤ , ,
∼
, 0 , 1 ,
Quantum Logic and Nonclassical Logics
157
and consider the necessity ν(E) of a given eﬀect E (which may be either sharp or unsharp). One can easily prove the following lemma. LEMMA 37. (i) E is a projection iﬀ E = ν(E) = E ∼ = PKer(E ) . (ii) Let P be any projection. P ≤ E implies P ≤ ν(E). As a consequence, we can say that ν(E) represents a kind of “best sharp lower approximation of E.” Brouwer Zadeh posets do not represent the only interesting way of structuring the set of all concrete eﬀects. Other important structures that have naturally emerged from eﬀectsystems are eﬀect algebras and quantum MV algebras. Such structures (introduced in the late Eighties and in the Nineties) have represented a privileged object of research for the logicoalgebraic approach to QT at the turn of the century. We will ﬁrst sketch the deﬁnition of eﬀect algebra (also called unsharp orthoalgebras)28 . One is dealing with a particular kind of partial structure, equipped with a basic operation that is only deﬁned for special pairs of elements. From an intuitive point of view, such an operation can be regarded as an exclusive disjunction (aut), deﬁned for events that are logically incompatible. The abstract deﬁnition of eﬀect algebra is the following. DEFINITION 38. Eﬀect algebra An eﬀect algebra is a partial structure A = A , , 0 , 1 , where is a partial binary operation on A, and 0 and 1 are special distinct elements of A. When is deﬁned for a pair a , b ∈ A, we will write ∃(a b). The following conditions hold: (i) Weak commutativity ∃(a b) implies ∃(b a) and a b = b a; (ii) Weak associativity ∃(b c) and ∃(a (b c)) implies ∃(a b) and ∃((a b) c) and a (b c) = (a b) c; (iii) Strong excluded middle For any a, there exists a unique x such that a x = 1; (iv) Weak consistency ∃(a 1) implies a = 0. 28 See [Giuntini and Greuling, 1989; Foulis and Bennett, 1994; Dalla Chiara and Giuntini, 1994; Dvureˇ censkij, 2000; Dalla Chiara et al., 2004].
158
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
An orthogonality relation ⊥, a partial order relation ≤ and a generalized complement (which generally behaves as a fuzzy complement) can be deﬁned in any eﬀect algebra. DEFINITION 39. Let A , , 0 , 1 be an eﬀect algebra and let a, b ∈ A. (i) a ⊥ b iﬀ a b is deﬁned in A. (ii) a ≤ b iﬀ ∃c ∈ A such that a ⊥ c and b = a c. (iii) The generalized complement of a is the unique element a such that a a = 1. One can show that any eﬀect algebra A , , 0 , 1 gives rise to a bounded involution poset A , ≤ , , 0 , 1, where ≤ and are deﬁned according to Deﬁnition 39. (iii) The category of all eﬀect algebras turns out to be (categorically) equivalent to the category of all diﬀerence posets, which have been ﬁrst studied by Kˆ opka and Chovanec and further investigated by Pulmannov´ a and others.29 Eﬀect algebras represent weak examples of orthoalgebras, a category of partial structures that Foulis and Randall had introduced in 1981.30 Roughly, orthoalgebras are eﬀect algebras that satisfy the noncontradiction principle. In such algebras, the involution becomes an orthocomplementation. The precise mathematical deﬁnition is the following: DEFINITION 40. Orthoalgebras An orthoalgebra is an eﬀect algebra A , , 0 , 1 such that the following condition is satisﬁed: ∃(a a) implies a = 0
(Strong consistency).
In other words: 0 is the only element that is orthogonal to itself. One can easily realize that orthoalgebras always determine an orthoposet. Let A = A , , 0 , 1 be an orthoalgebra. The structure A , ≤ , , 0 , 1 (where ≤ and are the partial order and the generalized complement of A) is an orthoposet. For, given any a ∈ A, the inﬁmum a ∧ a exists and is equal to 0; equivalently, the supremum a ∨ a exists and is equal to 1. THEOREM 41. Any orthoalgebra A = A , , 0 , 1 satisﬁes the following condition: if a, b ∈ A and a ⊥ b, then a b is a minimal upper bound for a and b in A. 29 See 30 See
[Kˆ opka and Chovanec, 1994],[Pulmannov´ a, 1995]. [Foulis and Randall, 1981].
Quantum Logic and Nonclassical Logics
159
COROLLARY 42. Any orthoalgebra A = A , , 0 , 1 satisﬁes the following condition: for any a, b ∈ A such that a ⊥ b, if the supremum a ∨ b exists, then a ∨ b = a b. Orthoalgebras and orthomodular posets turn out to be deeply connected. Any orthomodular poset A , ≤ , , 0 , 1 determines an orthoalgebra A , , 0 , 1 , where: a b is deﬁned iﬀ a ≤ b . Furthermore, when deﬁned, a b = a ∨ b. At the same time, not every orthoalgebra is an orthomodular poset (as shown by Wright in 199031 ). Genuine examples of eﬀect algebras (which are not generally orthoalgebras) can be naturally obtained in the domain of fuzzy set systems. EXAMPLE 43. Eﬀect algebras of fuzzy sets Let B be the set of all fuzzy subsets of a universe U (in other words, B is the set of all functions assigning to any element of U a value in the real interval [0, 1] ). A partial operation can be deﬁned on B. For any f, g ∈ B: ∃(f g) iﬀ ∀x ∈ U : f (x) + g(x) ≤ 1, where + is the usual sum of real numbers. Furthermore: if ∃(f g), then f g := f + g, where: ∀x ∈ U {(f + g)(x) := f (x) + g(x)} . Let 1 be the classical characteristic function of the total set U , while 0 is the classical characteristic function of the empty set ∅. The structure B , , 0 , 1 is an eﬀect algebra. It turns out that the eﬀectalgebra generalized complement coincides with the fuzzy complement. In other words: ∀x ∈ U : f (x) = 1 − f (x). Furthermore, the eﬀectalgebra partial order relation coincides with the natural partial order of B. In other words: ∀x ∈ U [f (x) ≤ g(x)] iﬀ ∃h ∈ B[f ⊥ h and g = f h]. The eﬀect algebra B , , 0 , 1 is not an orthoalgebra, because the strong consistency condition is violated by some genuine fuzzy sets (such as the semitransparent fuzzy set 12 1 that assigns to any object x value 12 ). 31 See
[Wright, 1990].
160
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
How can we induce the structure of an eﬀect algebra on the set E(H) of all eﬀects of the Hilbert space H? As in the fuzzyset case, it is suﬃcient to deﬁne the partial sum as follows: ∃(E F ) iﬀ E + F ∈ E(H), where + is the usual sumoperator. Furthermore: E F := E + F, if ∃(E F ). It turns out that the structure E(H) , , O , I is an eﬀect algebra (called standard eﬀect algebra or Hilbert eﬀect algebra), where the generalized complement of any eﬀect E is just I − E. Furthermore, the eﬀectalgebra order relation coincides with the natural order deﬁned on E(H). In other words: ∀ρ ∈ D(H)[tr(ρE) ≤ tr(ρF )] iﬀ ∃G ∈ E(H)[E ⊥ G and F = E G]. At the same time, this structure fails to be an orthoalgebra. For instance, the semitransparent eﬀect 12 I gives rise to a counterexample to the strong consistency condition: 1 1 1 1 1 I = O and I I = I + ( I) = I. 2 2 2 2 2 Let us now turn to the other kind of structure that naturally emerges from concrete eﬀect systems. One is dealing with quantum MV algebras (QMV algebras): they are weak variants of MV algebras (which represent privileged abstractions from classical fuzzy set structures).32 Before introducing the notion of QMV algebra, it will be useful to sum up some basic properties of MV algebras. As is well known, the set of all fuzzy subsets of a given set X gives rise to a de Morgan lattice, where the noncontradiction and the excluded middle principles are possibly violated. In this framework, the lattice operations ( the meet ∧, the join ∨ and the fuzzy complement ) do not represent the only interesting fuzzy operations that can be deﬁned. An important role is played by a new kind of conjunction and disjunction, which have been ﬁrst investigated in the framework of L
ukasiewicz’ approach to many valued logics. These operations are usually called L ukasiewicz operations. The deﬁnition of L
ukasiewicz conjunction and disjunction in the framework of fuzzy set structures turns out to be quite natural. Fuzzy sets are nothing but generalized characteristic functions whose range is the real interval [0, 1]. Of course, [0, 1] is not closed under the ordinary real sum + (we may have x, y ∈ [0, 1] and x + y ∈ / [0, 1]). However, one can introduce a new operation ⊕, which is called truncated sum: ∀x, y ∈ [0, 1] {x ⊕ y := min(1, x + y)} . In other words, x ⊕ y is the ordinary sum x + y, whenever this sum belongs to the interval; otherwise x ⊕ y collapses into the maximum element 1. 32 See
[Giuntini, 1996].
Quantum Logic and Nonclassical Logics
161
One immediately realizes that [0, 1] is closed under the operation ⊕. Now, we can use the truncated sum in order to deﬁne the L
ukasiewicz disjunction between fuzzy sets (since no confusion is possible, it will be expedient to use the same symbol ⊕ both for the truncated sum and for the L
ukasiewicz disjunction). Let f, g be fuzzy subsets of a set X. The L
ukasiewicz disjunction ⊕ is deﬁned as follows: ∀x ∈ X {(f ⊕ g)(x) := f (x) ⊕ g(x) = min(1, f (x) + g(x))} . On this basis, the L
ukasiewicz conjunction # can be deﬁned, via de Morgan, in terms of ⊕ and : ∀x ∈ X {(f # g)(x) := (f ⊕ g ) (x)} . As a consequence, one obtains: (f # g)(x) = max(0, f (x) + g(x) − 1). From an intuitive point of view, the L
ukasiewicz operations and the lattice operations represent diﬀerent notions of conjunction and disjunction that can be used in a fuzzy situation. Consider two fuzzy sets f and g; they can be intuitively regarded as two ambiguous properties. The number f (x) represents the “degree of certainty” according to which the object x satisﬁes the property f . A similar comment holds for g and g(x). What does it mean that the object x satisﬁes the disjunctive property “f or g” with a given degree of certainty? If we interpret “or” as the lattice join, we assume the following choice: an object satisﬁes a disjunction to a degree that corresponds to the maximum between the degrees of the two members. If we, instead, interpret “or” as the L
ukasiewicz disjunction, we assume the following choice: the degrees of the members of the disjunction have to be summed in such a way that one never goes beyond the absolute certainty (the value 1). Of course, in the limitcase represented by crisp sets (i.e., classical characteristic functions) the L
ukasiewicz disjunction and the lattice join will coincide. Suppose x, y ∈ {0, 1}, then x ⊕ y = max(x, y). From the deﬁnitions, one immediately obtains that the L
ukasiewicz operations are not generally idempotent. It may happen: a ⊕ a = a and a # a = a. As noticed by Mundici33 , this is a typical semantic situation that seems to be governed by the principle “repetita iuvant!” (repetitions are useful!). Of course repetitions are really useful in all physical circumstances that are accompanied by a certain noise. As a consequence, ⊕ and # do not give rise to a lattice structure. At the same time, as with the lattice operations, they turn out to satisfy commutativity and associativity: f ⊕ g = g ⊕ f; f # g = g # f; 33 See
[Mundici, 1992].
162
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
f ⊕ (g ⊕ h) = (f ⊕ g) ⊕ h; f # (g # h) = (f # g) # h. Unlike the fuzzy lattice operations, the L
ukasiewicz conjunction and disjunction do satisfy both the excluded middle and the noncontradiction principle: f ⊕ f = 1; f # f = 0. Another important diﬀerence concerns the distributivity property. As opposed to the case of ∧ and ∨ (which satisfy distributivity in the fuzzy set environment), it may happen: f # (g ⊕ h) = (f # g) ⊕ (f # h); f ⊕ (g # h) = (f ⊕ g) # (f ⊕ h). What can be said about the relationships between the L
ukasiewicz operations and the lattice operations? Interestingly enough, the lattice operations turn out to be deﬁnable in terms of the fuzzy complement and of the L
ukasiewicz operations. For, we have: f ∧ g := (f ⊕ g ) # g; f ∨ g := (f # g ) ⊕ g. In this framework, the conjunction ∧ will be also called et, while the disjunction ∨ will be called vel . An interesting algebraic abstraction from fuzzy set structures can be obtained if we restrict our attention to the fuzzy complement, the lattice operations and the L
ukasiewicz operations. This gives rise to the abstract notion of an MV algebra (multivalued algebra), that Chang introduced in 1958 in order to provide an adequate semantic characterization for L
ukasiewicz’ manyvalued logics.34 MV algebras represent a weakening of Boolean algebras, where the notion of conjunction (disjunction) is split into two diﬀerent operations. The ﬁrst kind of operation behaves like a L
ukasiewicz conjunction (disjunction) and is generally nonidempotent; the second kind of operation is a latticemeet (join). These algebras are also equipped with a generalized complement. In this framework, the lattice operations turn out to be deﬁned in terms of the generalized complement and of the L
ukasiewicz operations. Whenever the two conjunctions (resp., disjunctions) collapse into one and the same operation, one obtains a Boolean algebra. Let us now recall the formal deﬁnition of MV algebra. DEFINITION 44. MV algebra 35 An MV algebra is a structure M = M , ⊕ , , 0 , 1 , where ⊕ is a binary operation, is a unary operation and 0, 1 are special distinct elements satisfying the following conditions: 34 See 35 See
[Chang, 1958; Chang, 1959]. [Mangani, 1973; Cignoli and D’Ottaviano, 2000].
Quantum Logic and Nonclassical Logics
163
(MV1) a ⊕ b = b ⊕ a; (MV2) a ⊕ (b ⊕ c) = (a ⊕ b) ⊕ c; (MV3) a ⊕ a = 1; (MV4) a ⊕ 0 = a; (MV5) a ⊕ 1 = 1; (MV6) a = a; (MV7) (a ⊕ b) ⊕ b = (b ⊕ a) ⊕ a.
ukasiewicz conjunction #, In any MV algebra M = M , ⊕ , , 0 , 1, the L the lattice operations ∧ and ∨, the L
ukasiewicz implication →L , the partial order relation ≤ can be deﬁned as follows: • a # b := (a ⊕ b ) ; • a ∧ b := (a ⊕ b ) # b; • a ∨ b := (a # b ) ⊕ b; • a →L b := a ⊕ b; • a ≤ b iﬀ a ∧ b = a. It is not diﬃcult to see that ∀a, b ∈ M : a ≤ b iﬀ a →L b = a ⊕ b = 1. Hence, the operation →L represents a well behaved conditional.36 LEMMA 45. Let M = M , ⊕ , , 0 , 1 be an MV algebra. Consider the structure M , ≤ , , 0 , 1 , where ≤ is the partial order deﬁned on M. Such structure is a distributive bounded involution lattice, where ∧ and ∨ represent the inﬁmum and the supremum, respectively. The noncontradiction principle (a ∧ a = 0) and the excluded middle (a ∨ a = 1) are possibly violated.37 A privileged example of MV algebra can be deﬁned by assuming as support the real interval [0, 1]. DEFINITION 46. The [0, 1]MV algebra The [0, 1]MV algebra is the structure M[0,1] = [0, 1] , ⊕ , , 0 , 1 , where 36 Generally, a binary operation → of a structure (which is at least a bounded poset) is considered a well behaved conditional, when: a ≤ b iﬀ a → b = 1, for any elements a and b. By assuming a natural logical interpretation, this means that the conditional a → b is “true” iﬀ the “implicationrelation” a ≤ b holds. 37 See, for instance, [Cignoli and D’Ottaviano, 2000].
164
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
• ⊕ is the truncated sum. In other words: ∀x, y ∈ [0, 1] {x ⊕ y = min(1, x + y)} ; • ∀x ∈ [0, 1] {x = 1 − x} ; • 0 = 0; • 1 = 1. One can easily realize that M[0,1] is a special example of MV algebra where: • the partial order ≤ is a total order (coinciding with the natural real order); • x ∧ y = min(x, y); • x ∨ y = max(x, y). Let us now return to the concrete eﬀectstructure E(H) , , 0 , 1. The partial operation can be naturally extended to a total operation ⊕ that behaves similarly to a truncated sum. For any E, F ∈ E(H), E + F if ∃(E F ), E ⊕ F := 1 otherwise. Furthermore, let us deﬁne: E := I − E. The structure E(H) , ⊕ , , 0 , 1 turns out to be “very close” to an MV algebra. However, something is missing: E(H) satisﬁes the ﬁrst six axioms of the deﬁnition of an MV algebra; at the same time one can easily check that the ﬁnal axiom (usually called “ Lukasiewicz axiom”) is violated. For instance, consider two nontrivial projections P, Q such that P is not orthogonal to Q and Q is not orthogonal to P . Then, by the deﬁnition of ⊕ given immediately above, we have that P ⊕ Q = I and Q ⊕ P = I. Hence, (P ⊕ Q) ⊕ Q = Q = P = (Q ⊕ P ) ⊕ P . As a consequence, the L
ukasiewicz axiom must be conveniently weakened to obtain an adequate description of concrete eﬀect structures. This can be done by means of the notion of quantum MV algebra (QMV algebra).38 As with MV algebras, QMV algebras are total structures having the following form: M = M , ⊕ , , 0 , 1 , where: (i) 0 , 1 represent the impossible and the certain object, respectively; (ii)
is the negationoperation;
(iii) ⊕ represents a disjunction (or ) which is generally nonidempotent (a⊕a = a). 38 See
[Giuntini, 1996].
Quantum Logic and Nonclassical Logics
165
A (generally nonidempotent) conjunction (and ) is then deﬁned via the de Morgan law: a # b := (a ⊕ b ) . On this basis, a pair consisting of an idempotent conjunction et and of an idempotent disjunction vel is then deﬁned. As we have already discussed, in any MV algebra such idempotent operations behave as a latticemeet and latticejoin, respectively. However, this is not the case for QMV algebras. As a consequence, in such a more general situation, we will denote the et operation by the symbol , while the vel will be indicated by . The deﬁnition of et and vel is as in the MVcase: a b := (a ⊕ b ) # b a b := (a # b ) ⊕ b. DEFINITION 47. QMV algebra A quantum MV algebra (QMV algebra) (QMV) is a structure M = M , ⊕ , , 0 , 1 , where ⊕ is a binary operation, is a unary operation, and 0, 1 are special distinct elements of M . For any a, b ∈ M : a # b := (a ⊕ b ) , a b := (a ⊕ b ) # b , a b := (a # b ) ⊕ b. Assume that the following conditions hold: (QMV1)
a ⊕ b = b ⊕ a;
(QMV2)
a ⊕ (b ⊕ c) = (a ⊕ b) ⊕ c;
(QMV3)
a ⊕ a = 1;
(QMV4)
a ⊕ 0 = a;
(QMV5)
a ⊕ 1 = 1;
(QMV6)
a = a;
(QMV7)
a ⊕ [(a b) (c a )] = (a ⊕ b) (a ⊕ c).
By Axioms (QMV3), (QMV1) and (QMV4), one immediately obtains that 0 = 1. The operations and of a QMV algebra M are generally noncommutative. As a consequence, they do not represent latticeoperations. It is not diﬃcult to prove that is commutative iﬀ is commutative iﬀ (MV7) of Deﬁnition 44 holds. From this it easily follows that a QMV algebra M is an MV algebra iﬀ or is commutative. At the same time (as in the MVcase), we can deﬁne in any QMV algebra M , ⊕ , , 0 , 1 the following relation: a ≤ b iﬀ a b = a.
166
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
The structure M , ≤ , , 0 , 1 turns out to be a bounded involution poset. One can prove that the concrete eﬀect structure E(H) , ⊕ , , 0 , 1 is a QMV algebra (which is not an MV algebra). 7 UNSHARP QUANTUM LOGICS Orthologic, orthomodular quantum logic and partial classical logic are all examples of sharp logics. Both the logical and the semantic version of the noncontradiction principle hold: • any contradiction α ¬α is always false;39 • a sentence α and its negation ¬α cannot both be true. Some unsharp forms of quantum logic have been proposed (in the late Eighties and in the Nineties) as natural logical abstractions from the eﬀectstate systems.40 The most obvious unsharp weakening of orthologic is represented by a logic that has been called paraconsistent quantum logic (brieﬂy, PQL).41 In the algebraic semantics, this logic is characterized by the class of all models based on a bounded involution lattice, where the noncontradiction principle (a ∧ a = 0) is possibly violated. In the Kripkean semantics, instead, PQL is characterized by the class of all models K = I , R , P r , V , where the accessibility relation R is symmetric (but not necessarily reﬂexive), while P r behaves as in the OL case (i.e., P r is a set of propositions that contains I, ∅ and is closed under the operations ∩ and ). Any pair I , R, where R is a symmetric relation on I, is called a symmetric frame. All the other semantic deﬁnitions are given as in the case of OL, mutatis mutandis. On this basis, one can show that our algebraic and Kripkean semantics characterize the same logic. Unlike OL and OQL, a world i of a PQLKripkean model may verify a contradiction. Since R is generally not reﬂexive, it may happen that i ∈ V (α) and i ⊥ V (α). Hence, i = α ¬α. In spite of this, a contradiction cannot be veriﬁed by all worlds of a model K. It is worthwhile noticing that, unlike OQL, the connective negation could not be deﬁned in terms of f and of → in the framework of PQL. In fact, should → satisfy Modus Ponens, we would immediately obtain for any i: i = ¬(α ¬α)(:= α (α → f ) → f ). Hilbertspace models for PQL can be constructed in a natural way. In the Kripkean semantics, consider the models based on the following frame E(H) − {0} , ⊥ , 39 Of
course, in the case of partial classical logic, contradictions are false only if deﬁned. [Dalla Chiara et al., 2004]. 41 See [Dalla Chiara and Giuntini, 1989]. 40 See
Quantum Logic and Nonclassical Logics
167
where ⊥ represents the nonorthogonality relation between eﬀects (E ⊥ F iﬀ E ≤ F ). Unlike the corresponding case involving projections, in this situation the accessibility relation is symmetric but generally nonreﬂexive. For instance, the semitransparent eﬀect 12 I (representing the prototypical ambiguous property) is a ﬁxed point of the generalized complement . Hence, 1 1 1 1 I ⊥ I and ( I) ⊥ ( I) . 2 2 2 2 From the physical point of view, possible worlds are here identiﬁed with possible pieces of information about the physical system under investigation. Any information may correspond to: • a pure state (a maximal information); • a proper mixture (a nonmaximal information); • a projection (a sharp property); • a proper eﬀect (an unsharp property). Thus, unlike the sharp models of orthomodular quantum logic, here possible worlds do not always correspond to states of the quantum system under investigation. As expected, violations of the noncontradiction principle will be determined by unsharp (ambiguous) pieces of knowledge. PQL represents a somewhat rough logical abstraction from the class of all eﬀectstructures. As we already know, a characteristic condition that holds in all eﬀect structures is the regularity property (which may fail in a generic PQLmodel). DEFINITION 48. An algebraic PQLmodel B , v is called regular iﬀ the bounded involution lattice B is regular.42 The regularity property can be naturally formulated also in the framework of the Kripkean semantics: DEFINITION 49. A PQLKripkean model I, R , P r , V is regular iﬀ its frame I , R is regular . In other words, ∀i, j ∈ I: i ⊥ i and j ⊥ j i ⊥ j. One can prove that a symmetric frame I, R is regular iﬀ the involutive bounded lattice of all propositions of I, R is regular. As a consequence, an algebraic model is regular iﬀ its Kripkean transformation is regular and viceversa (where the Kripkean [algebraic] transformation of an algebraic [Kripkean] model is deﬁned like in OL). On this basis one can introduce a proper extension of PQL: regular paraconsistent quantum logic (RPQL). Semantically RPQL is characterized by the class of all regular models (both in the algebraic and in the Kripkean semantics). An axiomatization of PQL can be obtained by dropping the absurdity rule and the Duns Scotus rule in the OL calculus. As with OL, the logic PQL satisﬁes the ﬁnite model property and is consequently decidable. 42 See
Deﬀ. 124, 125.
168
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
The calculus for RPQL is obtained by adding to the PQLcalculus the following rule: α ¬α − β ¬β Kleene rule A completeness theorem for both PQL and RPQL can be proved, similarly to the case of OL. From the logical point of view, an interesting feature of PQL is represented by the fact that this logic is a common sublogic in a wide class of important logics. In particular, PQL is a sublogic of Girard’s linear logic, of L
ukasiewicz’ inﬁnitely manyvalued logic and of some relevant logics.43 An interesting question concerns the relation between PQL and the orthomodular property. Let B = B, ≤ , , 0 , 1 be an ortholattice. It is well known that the following three conditions (expressing possible deﬁnitions of the orthomodular property) turn out to be equivalent: (i) ∀a, b ∈ B: a ≤ b b = a ∨ (a ∧ b); (ii) ∀a, b ∈ B: a ≤ b and a ∧ b = 0 a = b; (iii) ∀a, b ∈ B: a ∧ (a ∨ (a ∧ b)) ≤ b. However, this equivalence breaks down in the case of bounded involution lattices. One can only prove: LEMMA 50. Let B be a bounded involution lattice. If B satisﬁes condition (i), then B satisﬁes conditions (ii) and (iii). LEMMA 51. Any bounded involution lattice B that satisﬁes condition (iii) is an ortholattice. As a consequence, we can conclude that there exists no proper orthomodular paraconsistent quantum logic when orthomodularity is understood in the sense (i) or (iii). A residual possibility for a proper paraconsistent quantum logic to be orthomodular is orthomodularity in the sense (ii). In fact, there are examples of lattices that are orthomodular (ii) but not orthomodular (i).44 Hilbert space models for orthomodular paraconsistent quantum logic can be constructed in the algebraic semantics by taking as support the following proper subset of the set of all eﬀects: Ec (H) := {aI : a ∈ [0, 1]} ∪ Π(H). In other words, a possible meaning of a sentence is either a sharp event (projection) or an unsharp eﬀect that can be represented as a multiple of the universal event (I). Hence all proper unsharp eﬀects are supposed to have a very special form. We will call the elements of Ec (H) eccentric eﬀects. 43 See 44 See
also Section 9. [Giuntini, 1990].
Quantum Logic and Nonclassical Logics
169
Unlike the case of E(H) (which is not a lattice), the set Ec (H) of all eccentric eﬀects turns out to be closed under ∧ and ∨. As a consequence, Ec (H) determines an orthomodular regular bounded involution lattice, where the partial order is the partial order of E(H) restricted to Ec (H), while the fuzzy complement is deﬁned like in the class of all eﬀects (E := I − E). As we have seen, PQL is expressed in the same language of orthologic and of orthomodular quantum logic, representing a weakening thereof. The Brouwer Zadeh structures (emerging from the concrete eﬀectstate systems) have suggested a stronger example of unsharp quantum logic, called Brouwer Zadeh logic (also fuzzy intuitionistic logic). As expected, a characteristic property of Brouwer Zadeh logic is a splitting of the connective “not” into two forms of negation: a fuzzylike negation, that gives rise to a paraconsistent behaviour, and an intuitionisticlike negation. The fuzzy “not” represents a weak negation, that inverts the two extreme truthvalues (truth and falsity), satisﬁes the double negation principle but generally violates the noncontradiction principle. The second “not” is a stronger negation, a kind of necessitation of the fuzzy “not”. The language of Brouwer Zadeh logic (BZL) is an extension of the language of QL. The primitive connectives are: the conjunction (), the fuzzy negation (¬), the intuitionistic negation (∼). Disjunction is metatheoretically deﬁned in terms of conjunction and of the fuzzy negation: α β := ¬(¬α ¬β) . A necessity operator is deﬁned in terms of the intuitionistic and of the fuzzy negation: Lα :=∼ ¬α . A possibility operator is deﬁned in terms of the necessity operator and of the fuzzy negation: M α := ¬L¬α . As happens to OL, OQL and PQL, also BZL can be characterized by an algebraic and by a Kripkean semantics. DEFINITION 52. Algebraic model for BZL An algebraic model of BZL is a pair B , v, consisting of a BZlattice B , ≤ , ,∼ , 0 , 1 and a valuationfunction v that associates to any sentence α an element in B, satisfying the following conditions: (i) v(¬β) = v(β) (ii) v(∼ β) = v(β)∼ (iii) v(β γ) = v(β) ∧ v(γ). The deﬁnitions of truth, consequence in an algebraic realization for BZL, logical truth and logical consequence are given similarly to the case of OL.
170
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
A characteristic feature of the Kripkean semantics for BZL is the use of frames with two accessibility relations. DEFINITION 53. Kripkean model for BZL A Kripkean model for BZL is a system K = I , ⊥ , ∼ ⊥ , P r , V where: (i) I , ⊥ , ∼ ⊥ is a frame with a non empty set I of possible worlds and two accessibility relations: ⊥ (the fuzzy accessibility relation) and ∼ ⊥ (the intuitionistic accessibility relation). Two worlds i , j are called fuzzyaccessible iﬀ i ⊥ j. They are called intuitionisticallyaccessible iﬀ i ∼ ⊥ j. Instead of not (i ⊥ j) and not (i ∼ ⊥ j), we will write i ⊥ j and i ∼ ⊥ j, respectively. The following conditions are required for the two accessibility relations: (ia) I, ⊥ is a regular symmetric frame; (ib) any world is fuzzyaccessible to at least one world: ∀i ∃j(i ⊥ j) . (ic) I, ∼ ⊥ is an orthoframe; (id) Fuzzy accessibility implies intuitionistic accessibility: i ⊥ j
i∼ ⊥ j.
(ie) Any world i has a kind of “twinworld” j such that for any world k: (a) i ∼ ⊥ k iﬀ j ∼ ⊥ k (b) i ∼ ⊥ k j ⊥ k. For any set X of worlds, the fuzzy orthogonal set X ⊥ is deﬁned like in OL: X ⊥ = {i ∈ I : ∀j ∈ X ( i ⊥ j)} . Similarly, the intuitionistic orthogonal set X ∼ is deﬁned as follows: X ∼ = {i ∈ I  ∀j ∈ X (i ∼ ⊥ j)} . The notion of proposition is deﬁned like in OL: a set of worlds X is a proposition iﬀ X = X ⊥⊥ . One can prove that for any set of worlds X, both X ⊥ and X ∼ are propositions. Further, like in OL, X ∧ Y (the greatest proposition included in the propositions X and Y ) is X ∩ Y , while X ∨ Y (the smallest proposition including X and Y ) is (X ∪ Y )⊥⊥ . (ii) P r is a set of propositions that contains I, and is closed under
⊥ ∼
,
, ∧.
(iii) V associates to any sentence a proposition in P r according to the following conditions: V (¬β) = V (β)⊥ ;
Quantum Logic and Nonclassical Logics
171
V (∼ β) = V (β)∼ ; V (β γ) = V (β) ∧ V (γ). THEOREM 54. Let I , ⊥ , ∼ ⊥ be a BZframe (i.e. a frame satisfying the conditions of Deﬁnition 53) and let P r0 be the set of all propositions of the frame. Then, the structure P r0 , ⊆ , ⊥ , ∼ , ∅ , I is a complete BZlattice such that for any set Γ ⊆ P r0 : ⊥⊥
Γ= Γ and Γ= Γ . As a consequence, the propositionstructure P r , ⊆ , ⊥ , ∼ , ∅ , I of a BZLmodel, turns out to be a BZlattice. The deﬁnitions of truth, consequence in a Kripkean model, logical truth and logical consequence, are given similarly to the case of OL. One can prove, with standard techniques, that the algebraic and the Kripkean semantics for BZL characterize the same logic. We will now introduce a calculus that represents an adequate axiomatization for the logic BZL. The most intuitive way to formulate our calculus is to present it as a modal extension of the axiomatic version of regular paraconsistent quantum logic RPQL. (Recall that the modal operators of BZL are deﬁned as follows: Lα :=∼ ¬α; M α := ¬L¬α). Rules of BZL. The BZLcalculus includes, besides the rules of RPQL the following modal rules: (BZ1)
Lα − α
(BZ2)
Lα − LLα
(BZ3)
M Lα − Lα
(BZ4)
α − β Lα − Lβ
(BZ5)
Lα Lβ − L(α β)
(BZ6)
∅ − ¬(Lα ¬Lα)
The rules (BZ1)–(BZ5) give rise to a S5 –like modal behaviour. The rule (BZ6) (the noncontradiction principle for necessitated sentences) is, of course, trivial in any classical modal system. One can [Dalla Chiara and Giuntini, 2002] prove a soundness and completeness Theorem with respect to the Kripkean semantics (by an appropriate modiﬁcation of the corresponding proofs for OL).
172
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
THEOREM 55. Soundness theorem T − α. BZL α T = BZL THEOREM 56. Completeness theorem T =BZLα T − BZL α. Characteristic logical properties of BZL are the following: (a) like in PQL, the distributive principles, Duns Scotus, the noncontradiction and the excluded middle principles break down for the fuzzy negation; (b) like in intuitionistic logic: =BZL∼ (α ∼ α);
/=BZLα ∼ α ;
α =BZL∼∼ α ;
∼∼ α /=BZLα ;
∼∼∼ α =BZL∼ α ; α =BZLβ ∼ β =BZL∼ α ; (c) furthermore: ∼ α =BZL¬α ;
¬α /=BZL∼ α ;
¬ ∼ α =BZL∼∼ α ;
We have seen that concrete eﬀectsystems give also rise to examples of partial algebraic structures, where the basic operations are not always deﬁned. How to give a semantic characterization for a logic that corresponds to the class of all eﬀect algebras? Such a logic has been called unsharp partial quantum logic (UPaQL). The language of UPaQL consists of a set of atomic sentences and of two primi+ (aut). The set of tive connectives: the negation ¬ and the exclusive disjunction ∨ sentences is deﬁned in the usual way. A conjunction is metalinguistically deﬁned, via de Morgan law: + ¬β). α∧. β := ¬(¬α ∨ The intuitive idea underlying the semantics for UPaQL is the following: disjunctions and conjunctions are always considered “legitimate” from a mere lin+ guistic point of view. However, semantically, a disjunction α ∨ β will have the intended meaning only in the “appropriate cases:” where the values of α and β + β will have any are orthogonal in the corresponding eﬀect algebra. Otherwise, α ∨ meaning whatsoever (generally not connected with the meanings of α and β). As is well known, a similar semantic “trick” is used in some classical treatments of the description operator ι (“the unique individual satisfying a given property”; for instance, “the present king of Italy”). Apparently, one is dealing with a diﬀerent idea with respect to the semantics of partial classical logic (PaCL), where the meaning of a sentence is not necessarily deﬁned. One has proved that UPaQL is an axiomatizable logic.45 45 See
[Dalla Chiara and Giuntini, 2002].
Quantum Logic and Nonclassical Logics
173
Also the theory of QMV algebras has naturally suggested the semantic characterization of another form of quantum logic (called L ukasiewicz quantum logic (LQL)), which generalizes both OQL and Lℵ ( Lukasiewicz’ inﬁnite many valued logic). The language of L QL contains the same primitive connectives as UPaQL + , ¬). The conjunction (∧. ) is deﬁned via the de Morgan law (as with UPaQL). (∨ ∨ Furthermore, a new pair of conjunction ( ∧ ∧ ) and disjunction ( ∨ ) connectives are deﬁned as follows: + . β α∧ ∧ β := (α ∨ ¬β)∧ ∨ β := ¬(¬α ∧ α∨ ∧ ¬β)
L QL can be easily axiomatized by means of a calculus that simply mimics the axioms of QMV algebras.46 8
QUANTUM LOGIC AND PARACONSISTENT LOGIC
As we have seen, the orthodox quantum logics whose investigation can be traced back to the work by Birkhoﬀ and von Neumann, such as orthologic or orthomodular quantum logic, are not paraconsistent: both logics validate the principle of ex absurdo quodlibet (any sentence β follows from a contradiction α ¬α) and the law of noncontradiction ¬ (α ¬α), although the acceptance of the latter is clearly compatible with paraconsistency47 . Thus, any inconsistent theory in either OL or OQL is bound to be trivial. We have learnt, however, that other quantum logics — in particular, unsharp logics like PQL — have a marked paraconsistent character. As a consequence, a question naturally arises: should a “good” quantum logic be paraconsistent? At least two diﬀerent stances can be adopted in this connection: • On the one hand, we might believe that orthodox quantum logics are inadequate to account for some aspects of quantum theory which are better captured through recourse to a paraconsistent logic.48 For example, if we liberalize the notion of quantum event in such a way as to allow for proper eﬀects as mathematical counterparts of unsharp properties, we end up adopting one of the paraconsistent quantum logics (either RPQL or BZL or L QL or UPaQL). • On the other hand, we might claim that orthodox quantum logics — whether or not they adequately capture every single aspect of quantum theory — are, in a way, already paraconsistent in themselves. For example, it is remarked in [Restall, 2002] that both OL and OQL (unlike, for example, 46 See
[Dalla Chiara et al., 2004]. the famous anecdote of the judge who had been given the task of settling a controversy between two litigants asserting opposite claims. After hearing the former’s pleading, he decreed he was in the right; yet, when the latter set out his reasons, he conceded that he also was in the right. As his assistant protested that no two parties asserting mutually contradictory statements can both be in the right, he exclaimed: “You’re in the right, too!”. Similarly, it is possible for a logic to uphold the noncontradiction principle while allowing for nontrivial inconsistent theories. 48 See [Dalla Chiara et al., 2004], [da Costa et al., 2006]. 47 Recall
174
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
propositional intuitionistic logic) tolerate classical inconsistencies, because the falsumsentence f does not follow therein from the classically inconsistent formula α (β γ) ¬((α β) (α γ)). Having discussed at some length in the preceding sections the interaction between the unsharp approach to quantum logic and paraconsistency, we will focus here on another application of paraconsistent logic to quantum theory, viz. the logical treatment of the complementarity problem. According to Bohr’s interpretation, quantum theory contains pairs of complementary sentences which, although not mutually contradictory as such, jointly entail a contradiction. Examples could be “p is a corpuscle” and “p is a wave”, where p refers to a designated photon — the former sentence is not the negation of the latter, but it entails the negation of the latter. If the logic underlying our physical theory is classical logic, which does not support nontrivial inconsistent theories, the triviality of quantum theory readily follows. A ﬁrst attempt to cope with this problem was done as early as in 1937 by P. DestouchesF´evrier, who (informally) introduced a threevalued logic by adjoining to the classical values True and False a third one (Absolutely False), meant to act as the truth value of the conjunctions of complementary sentences [DestouchesF´evrier, 1937]. Here, “absolutely false” should not be understood as “necessarily false” or “deﬁnitely false”, but rather as “meaningless” or “ungrammatical”, i.e. as expressing the fact that strictly speaking complementary sentences could not be conjoined. A paraconsistent version of this logic — where not only the value True, but also the value Absolutely False is designated — can be found in [da Costa and Krause, 2005]; the authors claim this modiﬁcation of F´evrier’s logic can meet some of the objections levelled against it in the literature. The same authors also devised another possible way out of the logical problem aroused by the complementarity phenomenon [da Costa et al., 2006]: they introduced a paraconsistent logical consequence relation P , termed paraclassical and speciﬁed in the following guise: DEFINITION 57. Paraclassical consequence Let L be the language of CL (classical sentential logic), containing the connectives ¬, , , → and let T be a set of sentences of L. A sentence α (of L) is called a Pconsequence of T (in symbols, T P α) iﬀ: P1 α ∈ T , or P2 α is a classical tautology, or P3 there exists a classically consistent subset T ⊆ T such that T CL α (where CL is the classical consequence relation). The notions of Pinconsistency and Ptriviality are introduced in the expected way.
Quantum Logic and Nonclassical Logics
175
DEFINITION 58. A set T of sentences of L is Pinconsistent iﬀ there exists a sentence α s.t. T P α and T P ¬α; it is Ptrivial iﬀ T P α for every formula α.49 Pinconsistency and Ptriviality are distinct concepts: if p is an atomic sentence, the set {p, ¬p} is Pinconsistent, but not Ptrivial. Also, Ptriviality and Pinconsistency imply, respectively, classical triviality and classical inconsistency, but the converse relations do not generally hold. Complementary sentences easily ﬁnd a home within this framework. First of all, one deﬁnes a Ctheory in the language L as a set of Lsentences closed under the relation P , and checks that there exist classically inconsistent Ctheories which are not Ptrivial. With these ingredients at our disposal, we can introduce a proper deﬁnition of complementary sentences: DEFINITION 59. Let T be a Ctheory in the language L. Two sentences α and β of L are said to be T complementary iﬀ: 1) T P α and T P β; 2) there is a sentence γ of L such that: T, α P γ and T, β P ¬γ. It follows from the deﬁnitions that Ctheories containing complementary theorems are classically inconsistent, but not Ptrivial: if α and β are complementary theorems and γ is such that T, α P γ and T, β P ¬γ, then in general γ ¬γ does not follow from T . Da Costa et al. conclude by remarking: (Complementary) theories are closer to those theories scientists actually use in their everyday activity than those theories with the classical concept of deduction. In other words, paraclassical logic (and paraconsistent logics in general) seems to ﬁt more accurately the way scientists reason when stating their theories [da Costa et al., 2006].
9
9.1
QUANTUM LOGIC AND LINEAR LOGIC
A brief survey of linear logic
Since Heyting, followers of the constructive approach to logic have singled out the notion of proof as the fundamental concept of their discipline, stressing at the same time that what really matters is not whether a given formula is provable in a certain logical calculus, but how it can be proved therein. In other words, the emphasis is not on the outcome, but on the process, the idea being that we must sharply distinguish between diﬀerent proofs of one and the same formula. Given these basic tenets, it does no harm to identify a formula with the set of its proofs, so that a proof of α from the assumptions α1 , ..., αn — seen as a method for converting any given proofs of α1 , ..., αn into a proof of α — boils down to a function f (x1 , ..., xn ) which associates to elements ai ∈ αi a result f (a1 , ..., an ) ∈ α. 49 A terminological remark: here “trivial” corresponds to “semantically inconsistent” (in the sense of Def. 22), while “inconsistent” corresponds to “semantically contradictory”.
176
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
This much is already implicitly contained in the general lines of the socalled BHK (BrouwerHeytingKolmogorov) interpretation of intuitionistic logic. In the 1960’s, W.A. Howard [Howard, 1980] added to it the identiﬁcation of intuitionistic natural deduction proofs with terms of typed lambda calculus. In a nutshell, a proof of the formula α is associated with a term of type α, whence it becomes possible to spell out the computational content of the inference rules in the {, →}fragment of the intuitionistic natural deduction calculus: • if t and s are terms having respective types α and β, then t, s (the pairing of t and s) is a term of type α β; • if t is a term of type α β, then π1 (t) and π2 (t) (the ﬁrst and second projections of t) are terms having respective types α and β; • if x is a variable of type α and t is a term of type β, then λx.t (the abstraction of t w.r.t. x) is a term of type α → β; • if t and s are terms having respective types α → β and α, then ts (the application of t to s) is a term of type β. The ensuing correspondence between intuitionistic natural deduction proofs and terms in the lambda calculus with projection and pairing functors can be seen as a fullﬂedged isomorphism (and it is indeed referred to as the CurryHoward isomorphism) in that there is a perfect match between the notions of conversion, normality and reduction introduced in the two frameworks. In the light of the CurryHoward isomorphism, it was readily acknowledged that the problem of ﬁnding a “semantics of proofs” for a given constructive logic and the problem of providing lambda calculus (or, for that matter, functional programming) with a semantical interpretation were two sides of the same coin. In Dana Scott’s domain theory, a ﬁrst attempt to accomplish this task, a type α was interpreted by means of a particular topological space. In the mid 1980’s, on the other hand, JeanYves Girard [Girard, 1987] introduced for this purpose the notion of coherent space over a set X. In the spirit of the BHK tradition, Girard identiﬁes each formula with the set of its proofs, which are in turn represented as “coherent” sets of information bits. More precisely: • the points in X are supposed to represent atomic information tokens; • a coherence relation between information tokens is deﬁned; • sets of pairwise coherent tokens correspond to proofs; • a coherent space represents a formula, viewed as the set of its proofs. DEFINITION 60. Polar Let X be a set and let α ⊆ P(X). The polar of α (indicated by ¬α) is deﬁned as follows: ¬α := {Y ⊆ X : ∀Z ∈ α(card(Y ∩ Z) ≤ 1)} .
Quantum Logic and Nonclassical Logics
177
In other words, given a family α of subsets of X, ¬α is the family of all subsets of X whose intersection with all elements of α is either the empty set or a singleton. DEFINITION 61. Coherent space A coherent space over the set X is a family α of subsets of X s.t. α = ¬¬α. EXAMPLE 62. Let X = {a, b, c}. • α = {∅ , {a} , {b} , {c} , {a, b}} is a coherent space over X; • β = {∅ , {b, c} , {a, b, c}} is not a coherent space over X. A coherence relation ⊥ /α can be now deﬁned in any coherent space α. DEFINITION 63. If α is a coherent space over X and a, b ∈ X, we set: a⊥ /α b iﬀ {a, b} ∈ α. Interestingly enough, coherent spaces turn out to be deeply connected with orthoframes. LEMMA 64. Let X be a set. There is a 11 correspondence between coherent spaces over X and orthoframes with universe X. Proof. If α is a coherent space over X, we set, in accordance with Deﬁnition 63: /α ) . F α = (X, ⊥ Conversely, let F = (X, ⊥ /) be an orthoframe with universe X; we deﬁne: αF = {Y ⊆ X : a ⊥ b for any a, b ∈ Y } . /α is clearly symThe system F α is an orthoframe. For, the induced relation ⊥ metric and is likewise reﬂexive in that α = ¬¬α contains the singletons of all tokens. A moment’s reﬂection also shows that αF is a coherent space over X: it is a family of subsets of X, and it equals its own double polar. It therefore suﬃces F α to show that: F = F α and α = αF . As to the former claim, just remark that F /b. As regards the latter, a⊥ /αF b iﬀ {a, b} ∈ α iﬀ a ⊥ αF
α
= {Y ⊆ X : a ⊥ /α b for any a, b ∈ Y } = {Y ⊆ X : {a, b} ∈ α for any a, b ∈ Y } = ¬¬α = α.
Deﬁnition 60 already provides us with a semantics of proofs for the negation connective. What about the other logical connectives? Let us consider conjunction ﬁrst. Here, Girard departs from the intuitionistic tradition inasmuch as he distinguishes two nonequivalent proof conditions for a conjunctive sentence:
178
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
• in a sense, I have a proof of a conjunctive sentence when I can assemble together two diﬀerent proofs, one for each conjunct; • yet, in a diﬀerent sense, I can prove a conjunctive sentence when I have just one proof, which however suﬃces to yield any one of the two conjuncts. From a constructive viewpoint, diﬀerent proof conditions correspond to diﬀerent logical constants; thus, classical (or, for that matter, intuitionistic) conjunction “splits up” into a multiplicative conjunction ⊗ and an additive conjunction $. Translating the above into the formal jargon of coherent spaces, we are led to the following DEFINITION 65. Let α and β be two coherent spaces, respectively over X and over Y (which we assume without loss of generality to be disjoint sets50 ). We deﬁne: α ⊗ β := {Z ⊆ X × Y : there exist W ∈ α, V ∈ β s.t. Z ⊆ V × W } ; α $ β := {Z ∪ W : Z ∈ α, W ∈ β} . One can prove that α ⊗ β is a coherent space over X × Y , while α $ β is a coherent space over X ∪ Y . Let us now come to implication, which can be deﬁned out of negation and multiplicative conjunction as ¬ (α ⊗ ¬β). This does not quite seem to tally with the usual picture of a proof for an implicative sentence, i.e. of a method (function) for extracting a proof for the consequent from a proof for the antecedent. The following deﬁnition and fact, however, provide the required bridge: DEFINITION 66. Let α, β be coherent spaces. A function f from α to β is called linear whenever it preserves disjoint unions. LEMMA 67. If α, β are coherent spaces, respectively over X and Y , Z ∈ α and W ∈ α → β, then (Z) = {b ∈ Y : a, b ∈ W for some a ∈ Z} W is a linear function from α to β, and is a member of β, W (Z) ∈ β for every Z ∈ α . α→β= W :W The resulting implication is accordingly termed linear implication. In addition, Girard considers a coherent space α β corresponding to intuitionistic implication, which is decomposed in terms of linear implication and a necessitylike unary operator, !. More precisely, the orthoframe associated to the coherent space α β is isomorphic (as a ﬁrst order structure) to the orthoframe associated to !α → β. Having seen how implication can be dealt with, we observe that disjunction, like conjunction, is aﬀected by ambiguity. In fact: 50 If they are not, it suﬃces to consider disjoint bijective copies of X and Y , built in the standard way.
Quantum Logic and Nonclassical Logics
179
• I have a proof of a disjunctive sentence when I have a method for converting a proof for the negation of any disjunct into a proof for the other; • yet, in a diﬀerent sense, I can prove a disjunctive sentence when I have a proof for one of the two disjuncts. Thus, once again, we have to distinguish between a multiplicative disjunction ⊕ and an additive disjunction : DEFINITION 68. Let α and β be two coherent spaces, respectively over X and over Y (which we assume to be disjoint sets). We deﬁne: α ⊕ β := ¬α → β; α β := α ∪ β. One can prove that α ⊕ β is a coherent space over X × Y , while α β is a coherent space over X ∪ Y . How can we make intuitive sense of this logic? A possible option is viewing formulas as concrete resources that, once they are consumed in a deduction to get some conclusion, cannot be recycled or reused. Formulas of the form !α, on the other hand, represent “ideal” resources that can be reused at will. Thus, while the availability of an intuitionistic implication α β means that using as many α’s as I might need I can get one β, the availability of a linear implication α → β expresses the fact that, using just one α, I can get one β — something that perfectly squares with the coherent space isomorphism pointed out above. We can also view the other compound formulas of linear logic as concrete resources: for example, α ⊗ β expresses the availability of both resource α and resource β, while α $ β expresses the availability of any one between such resources — i.e. we can have just one between α and β and not both, but we are in a position to choose which51 . In his 1987 seminal paper, Girard sets up a sequent calculus for his newly discovered logic, henceforth called LL.52 Unlike the calculi we have considered in the previous sections, a sequent calculus for a logic L is based on axioms and rules that govern the behavior of sequents. Any sequent has the form Γ ⇒ Δ, where Γ and Δ are ﬁnite (possibly empty) multisets of formulas.53 Axioms are 51 A standard example goes like this: suppose that a cup of coﬀee and a newspaper cost 1 Euro each. Thus, for 1 Euro I can get a cup of coﬀee and () a newspaper, but for the same amount I cannot get a cup of coﬀee and (⊗) a newspaper. 52 One caveat, though: our notation for logical constants does not exactly match Girard’s original one. For example, he resorts to the symbol ⊕ to denote additive, not multiplicative, disjunction. 53 A multiset is a set of pairs such that the ﬁrst element of every pair denotes an object, while the second element denotes the multiplicity of the occurrences of our object. Two multisets are equal iﬀ all their pairs are equal, that is all their objects together with their multiplicities are equal.
180
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
particular sequents. Any rule has the form: Γ1 ⇒ Δ1 , . . . , Γn ⇒ Δn Γ⇒Δ where Γ1 ⇒ Δ1 , . . . Γn ⇒ Δn are the premisses of the rule while Γ ⇒ Δ is the conclusion. Rules can be either structural or operational . Operational rules introduce a new connective either on the left or on the right side of a sequent. Accordingly, one usually speaks of left and of right introduction rule. For example, consider Gentzen’s calculus LK for classical logic. The left introduction rule for the connective can be written as follows: α, Γ ⇒ Δ α β, Γ ⇒ Δ
β, Γ ⇒ Δ ( L) α β, Γ ⇒ Δ
Structural rules, instead, only deal with the structure of sequents (orders, repetitions, etc.). Two important examples of structural rules are weakening and contraction: Γ⇒Δ Γ, Π, Π ⇒ Δ, Δ, Σ (weakening); (contraction) Γ, Π ⇒ Σ, Δ Γ, Π ⇒ Δ, Σ, A derivation (in the logic L) is a sequence of sequents where any element is either an axiom or the conclusion of a rule whose premisses are previous elements of the sequence. A sequent Γ ⇒ Δ is said to be derivable in the logic L (abbreviated L Γ ⇒ Δ ) iﬀ Γ ⇒ Δ is the last element of a derivation of L.54 The completeness (and soundness) theorem for Gentzen’s LK can be formulated as follows: LK {α1 , . . . , αn } ⇒ {β1 , . . . , βm } iﬀ α1 . . . αn → β1 . . . βm is a logical truth of classical logic. Girard’s LL diﬀers from LK in that: • LL has no weakening or contraction rules — so that its language contains additive and multiplicative versions of both conjunction and disjunction, as well as of the verum and falsum constants. The absence of weakening and contraction is readily explained in terms of the resource interpretation: for example, contraction would say that if you can get a resource β by using n copies of the resource α, you can still get β by using just one copy of α, which is just wishful thinking. • The language of LL includes the modalities of course! (!) and why not? (?). The addition of modalities reintroduces weakening and contraction for special classes of formulas; more precisely, a formula preceded by an exclamation mark can be introduced through unrestricted left weakening and 54 Since a rule has generally many premisses, derivations can be also conveniently represented as special treeconﬁgurations.
Quantum Logic and Nonclassical Logics
181
contraction inferences, while a formula preceded by a question mark can be introduced through unrestricted right weakening and contraction inferences. The modalities abide at the same time by structural rules and modal rules: if we have a look at the sequent calculi for modal logics, in fact, it is easy to see that ! behaves as an S4 necessity operator, and ? as an S4 possibility operator. The importance of the modalities is fully realized if we take into account the fact that Girard is admittedly not interested in setting up a logic which is weaker than classical or intuitionistic logic: he rather aims for a logic which permits a better analysis of proofs through a stricter control of structural rules. Modalities are there precisely to recapture the deductive power of weakening and contraction, an aim which is attained — in a sense — by showing that both classical logic and intuitionistic logic can be embedded into linear logic. Conﬁning ourselves to classical logic, Girard provides in fact translations of multisets of classical formulas into multisets of linear formulas in such a way that Γ ⇒ Δ is provable in LK iﬀ a sequent compounded out of an appropriate translation of its formulas is provable in LL. Beside this desirable aspect, modalities have their down side too: they are to blame for the undecidability of propositional linear logic [Lincoln et al., 1992]. The language L1 of LL contains the connectives ⊗, ⊕, →, $, (binary), ¬, !, ? (unary), %, ⊥, f , t (nullary). The notations !Γ and ?Γ mean that the modality at issue must be preﬁxed to each element of the given sequence; L0 will denote the language obtained by barring from L1 the modalities ! and ?.
9.2
Quantum logic vs linear logic: proof theory
A prominent feature shared by quantum logics and linear logic alike is the failure of lattice distributivity. This means that any adequate proof system for these logics — for the sake of deﬁniteness, say any sequent calculus — should clamp in some way or another the mechanism which allows the proof of α (β γ) ⇒ (α β) (α γ) in LK:55 β⇒β γ⇒γ α⇒α α⇒α α, β ⇒ α, α γ α, β ⇒ β, α γ α, γ ⇒ α β, α α, γ ⇒ α β, γ α, β ⇒ α β, α γ α, γ ⇒ α β, α γ α, β γ ⇒ α β, α γ α (β γ), α (β γ) ⇒ (α β) (α γ), (α β) (α γ) α (β γ) ⇒ (α β) (α γ) As we have seen, in LL — whose operational rules for the additives are the same as we used in the above LK proof — we attain such an aim by doing away with the structural rules of weakening and contraction outright. In the various sequent systems which have been suggested for quantum logics, on the other hand, one 55 The converse half corresponds to an inequality which is valid in the class of general lattices and, therefore, holds both in linear logic and in all mainstream quantum logics.
182
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
Axioms and rules of LL α ⇒ α (ID) ⊥, Γ ⇒ Δ (⊥L)
Γ ⇒ Δ, % (%R)
f ⇒ (f L)
Γ⇒Δ (f R) Γ ⇒ Δ, f
Γ⇒Δ (tL) t, Γ ⇒ Δ
⇒ t (tR)
α, β, Γ ⇒ Δ ( ⊗ L) α ⊗ β, Γ ⇒ Δ
Γ ⇒ Δ, α Π ⇒ Σ, β ( ⊗ R) Γ, Π ⇒ Δ, Σ, α ⊗ β
α, Γ ⇒ Δ β, Π ⇒ Σ ( ⊕ L) α ⊕ β, Γ, Π ⇒ Δ, Σ
Γ ⇒ Δ, α, β ( ⊕ R) Γ ⇒ Δ, α ⊕ β
α, Γ ⇒ Δ α $ β, Γ ⇒ Δ
Γ ⇒ Δ, α Γ ⇒ Δ, β ( $ R) Γ ⇒ Δ, α $ β
β, Γ ⇒ Δ ( $ L) α $ β, Γ ⇒ Δ
α, Γ ⇒ Δ β, Γ ⇒ Δ ( L) α β, Γ ⇒ Δ
Γ ⇒ Δ, α Γ ⇒ Δ, α β
Γ ⇒ Δ, α β, Π ⇒ Σ ( → L) α → β, Γ, Π ⇒ Δ, Σ
α, Γ ⇒ Δ, β ( → R) Γ ⇒ Δ, α → β
Γ ⇒ Δ, α (¬L) ¬α, Γ ⇒ Δ
α, Γ ⇒ Δ (¬R) Γ ⇒ Δ, ¬α
α, Γ ⇒ Δ (!L) !α, Γ ⇒ Δ
!Γ ⇒?Δ, α (!R) !Γ ⇒?Δ, !α
Γ⇒Δ (!W) !α, Γ ⇒ Δ
!α, !α, Γ ⇒ Δ (!C) !α, Γ ⇒ Δ
α, !Γ ⇒?Δ (?L) ?α, !Γ ⇒?Δ
Γ ⇒ Δ, α (?R) Γ ⇒ Δ, ?α
Γ⇒Δ (?W) Γ ⇒ Δ, ?α
Γ ⇒ Δ, ?α, ?α (?C) Γ ⇒ Δ, ?α
Γ ⇒ Δ, β ( R) Γ ⇒ Δ, α β
Quantum Logic and Nonclassical Logics
183
usually manages to avoid distribution by means of appropriate restrictions on side utte Monting, formulas.56 For example, the calculus for orthologic implicit in [Sch¨ 1981]57 is exactly like LK formulated in the {¬, , }language, except for the fact that sequents can contain at most two formulas. Weakening and contraction rules are there indeed, but many applications thereof are blocked because they “take up too much room”. Several steps in the displayed proof of distribution, for example, turn out to be incorrect by such a standard. If we tighten even further the mentioned constraint, by allowing only sequents which contain exactly two formulas, we obtain a calculus for an unbounded version of paraconsistent quantum logic (where all weakening and contraction inferences clearly become unfeasible, leading to the unprovability of sequents like α ¬α ⇒ β or α ⇒ β ¬β).58 There is, however, a diﬀerent standpoint we could take while comparing the proof theories of linear logic and quantum logics. One could, indeed, search for a common abstraction of such logics, i.e. try to formulate a more general logic from which they can be obtained through the addition of special postulates. This perspective has been suggested in [Sambin et al., 2000] and [Battilotti and Faggian, 2002], where the resulting logic is called basic logic59 . Three general assumptions lay the groundwork for the development of basic logic: the principles of reﬂection, symmetry, and visibility. The principle of reﬂection amounts, roughly speaking, to the assumption that in the framework of sequent calculi each propositional connective reﬂects at the level of the object language a link between assertions formulated in an appropriate metalanguage. Such assertions are compounded out of basic assertions of the form α is by means of the links and and yields, which are suﬃcient to express all the metalinguistic assertions occurring in a sequent calculus:
56 In any operational rule, the formula in the conclusion that contains the connective introduced by the rule in question is called the principal formula; the formulas in the premisses that are the components of the formula introduced by the rule are called the auxiliary formulas. All other formulas occurring in the rule are called side formulas (or also contextformulas). 57 More precisely, it is a calculus which generates the involution lattice inequalities valid in all ortholattices. 58 In the algebraic semantics, this logic is characterized by the class of all models based on involution lattices that are not necessarily bounded. Hence, truth and logical truth cannot be deﬁned in this framework; at the same time, consequence and logical consequence are deﬁned in the usual way. 59 This label, as a matter of fact, is all too overworked in the ﬁeld of nonclassical logics. To the best of our knowledge, there are three more established logics with this same name (a fuzzy logic, introduced by Hajek; a relevant logic, introduced by Routley; and a constructive logic, introduced by Visser).
184
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
the sequent calculus conﬁguration
abbreviates
the complex assertion
α1 , ..., αn
→
(α1 is) and...and (αn is)
Γ⇒Δ
→
Γ yields Δ
Γ⇒Δ Π⇒Σ Λ⇒Θ
→
(Γ ⇒ Δ) and (Π ⇒ Σ) yields (Λ ⇒ Θ)
The truth conditions for a given connective are given by its deﬁnitional equation, a metalinguistic biconditional which relates a deﬁniendum, containing the connective, and a deﬁniens, not containing it. Deﬁnitional equations provide the justiﬁcation for the inference rules of the calculus, obtained by ”solving” such equations according to a general method60 . More precisely, every connective has a formation rule, derived from the direction of the deﬁnitional equation which gives suﬃcient conditions for asserting a formula containing the connective at issue, and a reﬂection rule, which licenses deductions from an already available formula containing the connective. For example, here are the deﬁnitional equations for the multiplicative (⊗) and the additive ($) conjunction of basic logic: (DE⊗) (α ⊗ β is) yields Γ iﬀ (α is and β is) yields Γ; (DE$) Γ yields (α $ β is) iﬀ (Γ yields (α is)) and (Γ yields (β is)). Remark that multiplicative conjunction reﬂects an and link within the scope of a yields link, while additive conjunction reﬂects a principal and link. Solving such deﬁnitional equations one gets the rules Γ⇒α Δ⇒β ( ⊗ R) Γ, Δ ⇒ α ⊗ β
α, β ⇒ Γ ( ⊗ F) α⊗β ⇒Γ Γ⇒α Γ⇒β ( $ F) Γ⇒α$β
α⇒Γ α$β ⇒Γ
β⇒Γ ( $ R) α$β ⇒Γ
One can observe how the formation rule for multiplicative conjunction operates on the antecedent of the sequent, whereas the reﬂection rule operates on the succedent; the rules for additive conjunction, instead, follow the opposite pattern. All the other connectives of basic logic can be likewise divided into left connectives and right connectives, according as they behave like ⊗ or like $ under the mentioned respect. Every left connective is matched by a symmetric right connective: for example, ⊗ is matched by a right multiplicative disjunction ⊕, while $ is matched by a left additive disjunction . Furthermore, one can notice that all the rules of basic logic satisfy, like in the example cited above, a visibility requirement: there are no side formulas on the same side of either the principal, or the auxiliary formulas. Said otherwise, basic logic adds to the control of structural rules typical 60 for
the details see e.g. [Sambin et al., 2000].
Quantum Logic and Nonclassical Logics
185
of linear logic a control of side formulas, both on the left and on the right of the arrow. How can we get back from here to where we started, namely to linear and quantum logics? Well, we can extend basic logic in at least three ways: relaxing the visibility constraints, either on the left (L) or on the right (R) of the arrow, or restoring the structural rules of weakening and contraction (S). According to the chosen path, we get: • BL, intuitionistic linear logic without exponentials61 ; • BR, “dual intuitionistic” linear logic without exponentials; • BS, basic orthologic, an expansion of paraconsistent quantum logic; • BLR, linear logic without exponentials; • BLS, intuitionistic logic; • BRS, “dual intuitionistic” logic; • BLRS, classical logic. In particular, a calculus for PQL can be obtained by: (i) conﬁning ourselves to the {$, } fragment of BS; (ii) introducing negation deﬁnitionally: in other words, we start from positive literals (atomic sentences) and negative literals (the negations of such) and we inductively stipulate that: ¬¬p := p; ¬(α $ β) := ¬α ¬β; ¬(α β) := ¬α $ ¬β. This calculus is open to further simpliﬁcation once we consider that weakening and contraction rules are redundant in it: the calculus admits elimination of contraction, while left and right weakening are respectively simulated by ($R) and (R). In addition, we can turn it into a sequent system for orthologic if we restore the weakening and contraction rules and we adjoin two rules of transfer — a means to displace whole multisets of formulae across the arrow: Γ⇒Δ (tr1) Γ, ¬Δ ⇒
Γ⇒Δ (tr2) ⇒ ¬Γ, Δ
The visibility and symmetry properties enable to overcome in a simple way two problems that usually make cut elimination for orthologic so complicated: respectively, the constraints on contexts and negation. For a presentation of the cut elimination proof, see e.g. [Battilotti and Faggian, 2002]. Another signiﬁcant connection between the proof theories of linear logic and quantum logics can be found in the formalism of proofnets. Girard [Girard, 1987] introduced proofnets in order to amend three ﬂaws of sequent calculi: 61 To be precise, what we obtain is an expansion of this logic by a nonassociative multiplicative disjunction.
186
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
• Some “informal” proofs of linear logical sequents are represented in LL by more than one proof; formally speaking, they count as distinct proofs only because the order of application of the rules is diﬀerent. For example: α⇒α β⇒β α, β ⇒ α ⊗ β D: β ⇒ ¬α, α ⊗ β
α⇒α ⇒ α, ¬α β⇒β D : β ⇒ ¬α, α ⊗ β
We might therefore want to ﬁnd a calculus where these “bureaucratical variants” correspond to one and the same proof, as it would seem appropriate. • Sequent proofs are extremely redundant, because side formulas keep being copied again and again with each inference of the proof itself. • The cut elimination procedure is highly nondeterministic. For example, consider the proof: .. .. . . Γ ⇒ α, β, σ σ, γ, δ ⇒ Δ Γ ⇒ α ⊕ β, σ σ, γ ⊗ δ ⇒ Δ γ ⊗ δ, Γ ⇒ Δ, α ⊕ β The indicated cut can be pushed upwards and replaced by a cut whose premisses are Γ ⇒ α, β, σ and σ, γ, δ ⇒ Δ. But what comes next? We can either go on with a (⊗L) inference and then apply the (⊕R) rule, or proceed the other way around. Both choices are equally legitimate. The calculus of proofnets is free of these shortcomings and works perfectly well at least for the constantfree multiplicative fragment of linear logic. Put brieﬂy, one associates sequent proofs to special undirected graphs, in such a way that a graph actually represents a sequent proof if and only if it satisﬁes a simple, purely geometrical, criterion (namely, if and only if all the graphs one obtains from it by omitting edges according to a speciﬁed procedure are acyclic and connected: [Danos and Reignier, 1989]). This is all well and good so far as we remain conﬁned within the multiplicative fragment; however, a completely satisfactory way to ﬁt additives into the picture has not yet been found, despite several attempts by [Girard, 1996], [Tortora De Falco, 2003], [Hughes and van Glabbeek, 2003]. Restall and Paoli [Restall and Paoli, 2005] reversed, so to speak, the perspective and provided a simple calculus of proofnets for the linear logical additives alone, i.e. for the logic of general (possibly nondistributive) lattices, also giving a purely geometrical correctness criterion. Even though the possibility to extend such a calculus to the multiplicative fragment remains to be explored, this approach yielded a bonus — viz., a proofnet formulation for a number of quantum logics expressed in the language of involution lattices, such as an unbounded version of paraconsistent quantum logic and orthologic.
Quantum Logic and Nonclassical Logics
9.3
187
Quantum logic vs linear logic: semantics of provability
Although Girard repeatedly pointed out that, in his opinion, the only meaningful semantics for a logical system is its semantics of proofs, in his 1987 paper he does provide linear logic with a more traditional, Kripkestyle semantics of provability, which he dubs phase semantics. This circumstance suggests to highlight another noteworthy similarity between linear and quantum logics. As we already know, both orthologic (or weakenings thereof) and linear logic share the failure of lattice distributivity. In particular, the fragment of linear logic that includes just negation and the additive connectives is nothing but a version of the paraconsistent quantum logic PQL, whose algebraic counterpart, as we have seen, is the class of bounded involution lattices, and whose proper Kripkean counterpart is the class of symmetric frames. In linear logic, however, we have to model a wider range of connectives, including the multiplicatives and the modalities. True to form, adequate Kripkean models for full linear logic can be obtained by tinkering with symmetric frames in an appropriate way. Let us now ﬁll the details, which we partly draw from [Gallier, 1991]. For the sake of simplicity, we will leave the modalities aside; throughout this section, therefore, LL will not denote the whole calculus, but only its fragment corresponding to the language L0 . DEFINITION 69. Monoidal symmetric frame A monoidal symmetric frame is a ﬁrst order structure I = (I, •, 1, ⊥ /) such that (I, ⊥ /) is a symmetric frame, (I, •, 1) is an Abelian monoid62 and, for every a, b, c ∈ I, a ⊥ b • c iﬀ a • b ⊥ c Intuitively, the elements of I are information tokens which may or may not verify a given formula; the monoidal operation can be understood as composition of information bits; 1 represents a “true” piece of information; ﬁnally, that a is in the relation ⊥ / with b should be taken to mean that the information in a does not conﬂict with the information in b. The additional monoidal operation turns out to be the right key to access linear logical multiplicative connectives. Recalling the notion of closure of a subset in a symmetric frame63 , and recalling the notion of generalized subset product in a monoid: XY = {a • b : a ∈ X, b ∈ Y } we can deﬁne the following operations between subsets of a monoidal symmetric frame I: X Y = (XY )⊥⊥ ; X Y = (X ⊥ Y ⊥ )⊥ ; X Y = (X Y ⊥ )⊥ . 62 See
Def.137. a symmetric frame, deﬁne X ⊥ := {a ∈ X : a ⊥ X}. The operation be a closure operator. See Section 1. 63 Given
⊥⊥
turns out to
188
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
It is immediate to verify that X Y and X Y are closed subsets of I, that X Y = (X ⊥ Y ⊥ )⊥ , and that X Y = (XY ⊥ )⊥ . Furthermore, the deﬁnition of implication makes sense in the light of the following LEMMA 70. Let X, Y be closed subsets of I and a ∈ I. We have: a ∈ X Y iﬀ for every b ∈ I, if b ∈ X then a • b ∈ Y . Monoidal symmetric frames can be used to interpret the formulas and sequents of linear logic. Basically, we associate to each formula the set of information bits which verify it — formally speaking, a closed subset in a monoidal symmetric frame. DEFINITION 71. Monoidal symmetric model A monoidal symmetric model is a system M = I, •, 1, ⊥, v, where (i) I, •, 1, ⊥ is a monoidal symmetric frame; (ii) v is a function that associates to any sentence α a closed subset of I, satisfying the following conditions: v(¬α) = v(α)⊥ ; v(α ⊕ β) = v(α) v(β); v(α $ β) = v(α) ∩ v(β); ⊥⊥ v(t) = {1} ; v(%) = I;
v(α ⊗ β) = v(α) v(β); v(α → β) = v(α) v(β); ⊥⊥ v(α β) = (v(α) ∪ v(β)) ; v(f ) = {a ∈ I : a ⊥ 1} ; v(⊥) = I ⊥ .
DEFINITION 72. Truth and logical truth A formula α is true in a model M = I, •, 1, ⊥, v (abbreviated =M α) iﬀ 1 ∈ v(α); α is a logical truth (or a valid formula) of LL (=LL α) iﬀ for any model M, =M α. DEFINITION 73. Let Γ be we deﬁne: − t (Γ) = + t (Γ) =
a ﬁnite, possibly empty multiset of formulas in L0 ; α1 ⊗ ... ⊗ αn , if Γ = {α1 , ..., αn } ; t, otherwise. α1 ⊕ ... ⊕ αn , if Γ = {α1 , ..., αn } ; f , otherwise.
LEMMA 74. LL Γ ⇒ Δ iﬀ LL ⇒ t− (Γ) → t+ (Δ). We have the following completeness theorem: THEOREM 75. LL Γ ⇒ Δ iﬀ t− (Γ) → t+ (Δ) is valid.64 In Girard’s paper, phase semantics is not introduced in terms of monoidal symmetric frames, but rather in terms of a slightly diﬀerent class of structures, called phase structures. Although we preferred the alternative presentation in that it simpliﬁes comparison with the relational semantics for quantum logics, both perspectives are readily seen to be absolutely equivalent. 64 See
[Girard, 1987].
Quantum Logic and Nonclassical Logics
9.4
189
Quantum logic vs linear logic: semantics of proofs
Girard’s attitude towards quantum logic is utterly disparaging. Rather than abstracting away from the structures of quantum mechanics in order to get a logic — a process which is bound to yield not a “theory of reasoning”, but only a game of algebra written in a diﬀerent guise — he claims that one should go the other way around and use quantum structures to interpret logical proofs, in particular linear logical proofs [Girard, 1999], [Girard, 2004a], [Girard, 2004b]. Once we strip away from it similar anathemas — which are invariably, we think, the most infelicitous component of Girard’s contributions — this particular variant of denotational semantics may be worth a mention. Thus, we now turn to expound its main features. Some linear logical proofs have a probabilistic ﬂavour. Consider e.g. a possible proof for the sequent ⇒ α β. How could we have obtained it? Well, for sure we applied the (R) rule to get to it, but we can only guess as to the form of its premiss: the best we can say is that there is a 50% chance for ⇒ α and a 50% chance for ⇒ β (the 5050 assignment being a default one, of course). Thus, we cannot maintain with certainty that the information tokens constituting a proof of ⇒ α belong to the envisaged proof of ⇒ α β; we can only assign a probability to their so doing. This leads us to revise our previous concept of a proof as a “crisp” set of information tokens belonging to a set X — or, in other words, as a (characteristic) function from X to {0, 1}. A more realistic proof model will rather be a function from X to the whole real interval [0, 1], intuitively representing a probability assignment which outputs, for each information token, a “likelihood measure” of its membership in the given proof. Formally: DEFINITION 76. Let X be a ﬁnite non empty set and let α ⊆ IR+X (where IR+X is the set of all functions from X to the positive reals). The polar of α (denoted by ¬α), is deﬁned as follows: ¬α = {f ∈ IR+X : for every g ∈ α,
f (x) · g(x) ≤ 1}.
x∈X
The rationale for this deﬁnition is clear enough: if f, g are the characteristic functions, respectively, of the sets Y, Z, then f (x) · g(x) is nothing but the x∈X
cardinality of Y ∩ Z. Deﬁnition 76 is therefore a natural generalization of Deﬁnition 60, meaning that two proofs are “orthogonal” whenever, allowing for mutual compensation eﬀects, there is a maximal probability that they share at most one piece of information. DEFINITION 77. A probabilistic coherent space over the set X is a set α ⊆ IR+X s.t. α = ¬¬α. Let us now examine the treatment of additive connectives within this framework. Since there is nothing probabilistic about additive conjunction, we may simply generalize the classical deﬁnition to our realvalued function setting; to
190
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
cope with additive disjunction, on the other hand, we need to identify each one of its proofs with a probability assignment that weighs up the respective chances of its stemming from a proof of the former or from a proof of the latter disjunct. We are thus led to the following deﬁnitions: DEFINITION 78. If f ∈ IR+X and g ∈ IR+Y , where X, Y are disjoint sets, then f ∪ g ∈ IR+X∪Y is so deﬁned: f (a), if a ∈ X; (f ∪ g) (a) = g(a), if a ∈ Y . DEFINITION 79. Let α and β be two probabilistic coherent spaces, respectively over X and over Y (which we assume to be disjoint sets). We deﬁne: α $ β = {f ∪ g : f ∈ α, g ∈ β} ; α β = {λf ∪ (1 − λ)g : f ∈ α , g ∈ β , 0 ≤ λ ≤ 1} . This deﬁnition of additive conjunction squares with the standard coherent space framework in that one can prove that α β = ¬¬(α ∪ β), as expected. The standard coherent space perspective dictates that a proof for an implicative sentence be a function which extracts a proof for the consequent from a proof for the antecedent; here, however, proofs are themselves functions — it follows that the objects which correspond to proofs of implications will be functionals. Yet, this is not exactly what we want: to parallel our development of coherent spaces in the previous section, we need to simulate each linear functional from IR+X to IR+Y by means of a function in IR+X×Y . In other words, we want an implication whose antecedent is a probabilistic space α over X and whose consequent is a probabilistic space β over Y to be a space α → β over X × Y whose members are “linear traces” of functionals from α to β. h : IR+X → IR+Y is deﬁned in such DEFINITION 80. If h : X × Y → IR+ , then
+X and for every a ∈ X, a way that, for every f ∈ IR
h (f ) (a) = h((x, a)) · f (x) x∈X
(Remenber that X is a ﬁnite set!) The next lemma guarantees the adequacy of the previous deﬁnition: LEMMA 81. The map sending h to
h is a bijection from IR+X×Y onto the set of +X +Y to IR . linear functions from IR A fairly straightforward adaptation of the classical case thus gives: DEFINITION 82. Let α, β be probabilistic coherent spaces, respectively over X and Y . We deﬁne α → β over X × Y as follows: α→β= h:
h (f ) ∈ β for every f ∈ α .
Quantum Logic and Nonclassical Logics
191
One then proceeds to show that α → β is indeed a probabilistic coherent space, a fact that carries along the obvious deﬁnitions for the multiplicatives: DEFINITION 83. Let α and β be two probabilistic coherent spaces, respectively over X and over Y (which we assume to be disjoint sets). We deﬁne: α ⊗ β = ¬(α → ¬β); α ⊕ β = ¬α → β. It is not infrequent to ﬁnd in the linear logical literature the following analogy: formulas are to proofs as states (of a system) are to transitions between states. This perspective suggests a rather natural generalization of the above, obtained by formalizing linear logical proofs not through realvalued functions, but by means of selfadjoint operators on a complex ﬁnitedimensional Hilbert space — the idea being that each state is identiﬁed with the set of all transitions leading to it. DEFINITION 84. Let HX , HY be complex ﬁnitedimensional Hilbert spaces. We denote by L (HX , HY ) the set of all linear maps from HX to HY , abbreviating L (HX , HX ) by L (HX ). By HL(HX ) we mean the set of all selfadjoint operators on HX . The set HL(HX ) can be turned in the usual way into a vector space (over IR), endowed with an inner product ` a la HilbertSchmidt: U  V = tr(U V ). DEFINITION 85. Let α ⊆ HL(HX ) . The polar of α (denoted by ¬α), is deﬁned as follows: ¬α = {U : for every V ∈ α , 0 ≤ U  V ≤ 1} As far as we stay within the domain of positive reals, the inner product of two operators is a measure of their orthogonality: the lower the product, “the more orthogonal” they are. DEFINITION 86. A quantum coherent space over the space HX is a set α ⊆ HL(HX ) s.t. α = ¬¬α. Recall that, in standard coherent space semantics, additive (respectively, multiplicative) coherent spaces were deﬁned over the disjoint union (respectively, the cartesian product) of the sets underlying the constituent spaces. Here, disjoint union is replaced by direct sum of vector spaces, while tensor product plays the role of cartesian product 65 . We begin with the additives: DEFINITION 87. Let α and β be two quantum coherent spaces over HX and HY , respectively. α $ β and α β are spaces over the direct sum HX ⊕ HY , deﬁned as 65 See
Deﬀ. 161 and 162
192
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
follows
66
:
α $ β = {U : PX U PX = V ⊕ O and PY U PY = O ⊕ W, for V ∈ α , W ∈ β} ; α β = {λ(V ⊕ O) + (1 − λ)(O ⊕ W ) : V ∈ α, W ∈ β, 0 ≤ λ ≤ 1} . While it is fairly clear why disjunction is treated like this (cp. Deﬁnition 79), the deﬁnition of conjunction deserves a bit of elucidation. Intuitively, PX U PX and PY U PY can be thought of as the “restrictions” of the operator U to the subspaces HX and HY , respectively. Indeed, if U is a selfadjoint operator on HX ⊕ HY of the form U1 ⊕ U2 , then, for any ψ ∈ HX and for any ϕ ∈ HY : PX U PX (ψ ⊕ ϕ) = U1 (ψ) ⊕ O and PY U PY (ψ ⊕ ϕ) = O ⊕ U2 (ϕ). Accordingly, if we deﬁne PX U PX ∪ PY U PY = PX U PX + PY U PY , we obtain in this case: (PX U PX ∪ PY U PY )(ψ ⊕ ϕ) = U1 (ψ) ⊕ U2 (ϕ). Notice, however, that, in general, not every selfadjoint operator on HX ⊕ HY has the form U1 ⊕ U2 . As regards the multiplicatives, our starting point is the observation that, if HX and HY are ﬁnitedimensional Hilbert spaces, the space of the linear maps from the space of the operators over HX to the space of the operators over HY is isomorphic to the space of the operators over the tensor product HX ⊗ HY .
is a linear map belonging to DEFINITION 88. If U ∈ L (HX ⊗ HY ), then U L (L (HX ) , L (HY )), deﬁned in such a way that, ∀ V ∈ L (HX ), ∀ ψ, ϕ ∈ HY :
(V ) (ψ)ϕ = tr(U · (V ⊗ Wψ,ϕ )), U where, ∀ χ ∈ HY : Wψ,ϕ (χ) = χψϕ. Restricting ourselves to selfadjoint operators we get, in full analogy with the probabilistic case:
is a bijection from HL(HX ⊗HY ) onto LEMMA The map sending U to U L(H )89. L(H X Y) . ,H L H 66 P X and PY denote the projection operators onto the spaces HX and HY , respectively. In other terms, for any ψ ∈ HX and for any ϕ ∈ HY : PX (ψ ⊕ ϕ) = ψ ⊕ 0 and PY (ψ ⊕ ϕ) = 0 ⊕ ϕ. If U and V are selfadjoint operators on HX and HY , respectively, and λ ∈ IR, then U ⊕ V and λ(U ⊕ V ) are (selfadjoint) operators on HX ⊕ HY such that for any ψ ∈ HX and for any ϕ ∈ HY : (U ⊕ V )(ψ ⊕ ϕ) = U (ψ) ⊕ V (ϕ) and λ(U ⊕ V )(ψ ⊕ ϕ) = λU (ψ) ⊕ λV (ϕ).
Quantum Logic and Nonclassical Logics
193
Thus, as expected, DEFINITION 90. Let α, β be quantum coherent spaces, respectively over HX and HY . We deﬁne α → β over HX ⊗ HY as follows:
(V ) ∈ β for every V ∈ α . α→β= U :U One then goes on to show that α → β is indeed a quantum coherent space, a fact that leads once again to the appropriate deﬁnitions for the multiplicatives: DEFINITION 91. Let α and β be two quantum coherent spaces, respectively over HX and over HY . We deﬁne: α ⊗ β = ¬(α → ¬β); α ⊕ β = ¬α → β.
10
QUANTUM LOGIC AND QUANTUM COMPUTATION
Quantum computation has suggested new forms of quantum logic that have been called quantum computational logics.67 The main diﬀerence between these new logics and traditional (sharp and unsharp) quantum logics concerns a basic semantic question: how to represent the meanings of the sentences of a given language? As we have learnt, the answer given by “orthodox” quantum logic was the following: the meanings of the elementary experimental sentences of quantum theory have to be regarded as determined by appropriate sets of states of quantum objects (mathematically represented by closed subspaces of a Hilbert space). The answer given in the framework of quantum computational logics is quite diﬀerent. The meaning of a sentence is identiﬁed with a quantum information quantity: a system of qubits or, more generally, a mixture of systems of qubits.68 Two kinds of quantum computational semantics have been investigated: • A compositional semantics, where (like in classical logic) the meaning of a compound sentence is determined by the meanings of its parts. • A holistic semantics, which makes essential use of the characteristic “holistic” features of the quantumtheoretic formalism. Hence, in this framework, the meaning of a compound sentence generally determines the meanings of its parts, but not the other way around. Let us ﬁrst recall some basic deﬁnitions of quantum computation. Consider the twodimensional Hilbert space C2 (where any vector ψ is represented by a pair of complex numbers). Let B (1) = {0, 1} be the canonical orthonormal basis for C2 , where 0 = (1, 0) and 1 = (0, 1). 67 See
[Dalla Chiara et al., 2003]. logical approaches have investigated some interesting relations between quantum computation and linear logic. See, for instance, [Pratt, 1993],[Selinger, 2004],[van Tonder, 2004]. 68 Other
194
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
DEFINITION 92. Qubit A qubit is a unit vector ψ of the Hilbert space C2 . From an intuitive point of view, a qubit can be regarded as a quantum variant of the classical notion of bit: a kind of “quantum perhaps”. In this framework, the two basiselements 0 and 1 represent the two classical bits 0 and 1, respectively. From a physical point of view, a qubit represents a state of a single particle, carrying an atomic piece of quantum information. In order to carry the information stocked by n qubits, we need of course a compound system, consisting of n particles. DEFINITION 93. Quregister An nqubit system (also called nquregister ) is a unit vector in the nfold tensor product Hilbert space ⊗n C2 := C2 ⊗ . . . ⊗ C2 (where ⊗1 C2 := C2 ).69 !" # n−times
We will use x, y, . . . as variables ranging over the set {0, 1}. At the same time, x, y, . . . will range over the basis B(1) . Any factorized unit vector x1 ⊗. . .⊗xn of the space ⊗n C2 will be called a classical register. Instead of x1 ⊗ . . . ⊗ xn we will simply write x1 , . . . , xn . The set B (n) of all classical registers is an orthonormal basis for the space ⊗n C2 . Quregisters are pure states: maximal pieces of information about the particles under consideration. In quantum computation (as well as in quantum theory), one cannot help referring also to mixed states (pieces of information that are not maximal and might be enriched). In the framework of quantum computation, mixed states (mathematically represented by density operators of an appropriate Hilbert space) are also called qumixes. DEFINITION 94. Qumix A qumix is a density operator of ⊗n C2 (where n ≥ 1).70 Needless to say, quregisters correspond to particular qumixes that are pure states (i.e. projections onto onedimensional closed subspaces of a given ⊗n Cn ). We will indicate by D(⊗n C2 ) the set of all density operators of ⊗n C2 . Hence the ∞ set D = n=1 D(⊗n C2 ) will represent the set of all possible qumixes. A classical register x1 , . . . , xn is called true, when xn = 1; false, otherwise. The idea is that any classical register corresponds to a classical truthvalue that is determined by its last element. Hence, in particular, the bit 1 corresponds to the truthvalue Truth, while the bit 0 corresponds to the truthvalue Falsity. On this basis, we can identify, in any space ⊗n C2 , two special projection(n) (n) operators (P1 and P0 ) that represent, in this framework, the Truthproperty (n) and the Falsityproperty, respectively. The projection P1 is determined by the (n) closed subspace spanned by the set of all true registers, while P0 is determined by the closed subspace spanned by the set of all false registers. As we already know, in quantum theory, projections have the role of mathematical representatives of 69 See 70 See
Def. 161. Def. 159.
Quantum Logic and Nonclassical Logics
195
possible events (or physical properties) of the quantum objects under investigation. Hence, it turns out that Truth and Falsity behave here as special cases of quantum events. As a consequence, one can naturally apply the Born rule that determines the probabilityvalue that a quantum system in a given state satisﬁes a physical property. Consider any qumix ρ, which represents a possible state of a quantum system in the space ⊗n C2 . By applying the Born rule, we obtain that the probabilityvalue (n) that a physical system in state ρ satisﬁes the Truthproperty P1 is the number (n) 71 tr(ρP1 ) (where tr is the trace functional ). This suggests the following natural deﬁnition of the notion of probability of a given qumix. DEFINITION 95. Probability of a qumix For any qumix ρ ∈ D(⊗n C2 ): (n)
p(ρ) = tr(ρP1 ). From an intuitive point of view, p(ρ) represents the probability that the information stocked by the qumix ρ is true. In the particular case where ρ corresponds to the qubit ψ = c0 0 + c1 1, we obtain that p(ρ) = c1 2 . Given a quregister ψ, we will also write p(ψ) instead of p(Pψ ), where Pψ is the density operator represented by the projection onto the onedimensional subspace spanned by the vector ψ.
10.1
Quantum gates
In quantum computation, quantum logical gates (brieﬂy, gates) are unitary operators that transform quregisters into quregisters.72 Being unitary, gates represent characteristic reversible transformations. The canonical gates (which are studied in the literature) can be naturally generalized to qumixes. Generally, gates correspond to some basic logical operations that admit a reversible behavior. We will consider here the following gates: the negation, the PetriToﬀoli gate and the square root of the negation. Let us ﬁrst describe these gates in the framework of quregisters. DEFINITION 96. The negation For any n ≥ 1, the negation on ⊗n C2 is the linear operator Not(n) such that for every element x1 , . . . , xn of the basis B (n) : Not(n) (x1 , . . . , xn ) := x1 , . . . , xn−1 ⊗ 1 − xn . In other words, Not(n) inverts the value of the last element of any basisvector of ⊗n C2 . 71 See 72 See
Def. 158. Def. 160.
196
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
DEFINITION 97. The PetriToﬀoli gate For any n ≥ 1 and any m ≥ 1 the PetriToﬀoli gate is the linear operator T(n,m,1) deﬁned on ⊗n+m+1 C2 such that for every element x1 , . . . , xn ⊗ y1 , . . . , ym ⊗ z of the basis B (n+m+1) : T(n,m,1) (x1 , . . . , xn ⊗y1 , . . . , ym ⊗z) := x1 , . . . , xn ⊗y1 , . . . , ym ⊗xn ym ⊕ z, where ⊕ represents the sum modulo 2. One can easily show that both Not(n) and T(n,m,1) are unitary operators. Consider now the set R of all quregisters ψ. The gates Not and T can be uniformly deﬁned on this set in the expected way: Not(ψ) := Not(n) (ψ),
if ψ ∈ ⊗n C2
T(ψ ⊗ ϕ ⊗ χ) := T(n,m,1) (ψ ⊗ ϕ ⊗ χ), if ψ ∈ ⊗n C2 , ϕ ∈ ⊗m C2 and χ ∈ C2 . On this basis, a conjunction And, a disjunction Or, can be deﬁned for any pair of quregisters ψ and ϕ: And(ψ, ϕ) := T(ψ ⊗ ϕ ⊗ 0); Or(ψ, ϕ) := Not(And(Not(ψ), Not(ϕ))). Clearly, 0 represents an “ancilla” in the deﬁnition of And. The quantum logical gates we have considered so far are, in a sense, “semiclassical”. A quantum logical behavior only emerges in the case where our gates are applied to superpositions. When restricted to classical registers, such operators turn out to behave as classical (reversible) truthfunctions. We will now consider an important example of a genuine quantum gate that transforms classical registers (elements of B(n) ) into quregisters that are superpositions. This gate is the square root of the negation. DEFINITION 98. The square root of the negation For any n ≥ 1, the square root of the negation on ⊗n C2 is the linear operator √ (n) Not such that for every element x1 , . . . , xn of the basis B (n) : √ (n) 1 Not (x1 , . . . , xn ) := x1 , . . . , xn−1 ⊗ ((1 + i)xn + (1 − i)1 − xn ), 2 √ where i := −1. √ (n) One can easily show that Not is a unitary operator. The basic property of √ (n) Not is the following: for any ψ ∈ ⊗n C2 ,
√
(n)
Not
√ (n) ( Not (ψ)) = Not(n) (ψ).
In other words, applying twice the square root of the negation means negating.
Quantum Logic and Nonclassical Logics
197
√ (n) From a logical point of view, Not can be regarded as a “tentative partial negation” (a kind of “half negation”) that transforms precise pieces of information into maximally uncertain ones. For, we have: √ √ (1) (1) 1 p( Not (1)) = = p( Not (0)). 2 √ As expected, also Not can be uniformly deﬁned on the set R of all quregisters. √ Interestingly enough, the gate Not seems to represent a typically quantum logical operation that does not admit any counterpart either in classical logic or in standard fuzzy logics. THEOREM 99. [Dalla Chiara and Leporini, 2005] 1. There is no function f : {0, 1} → {0, 1} such that for any x ∈ {0, 1} : f (f (x)) = 1 − x. 2. There is no continuous function f : [0, 1] → [0, 1] such that for any x ∈ [0, 1] : f (f (x)) = 1 − x. The gates considered so far can be naturally generalized to qumixes [Gudder, 2003 to density operators, we will write: NOT, √ √ ]. When our gates will be applied NOT, T, AND, OR (instead of Not, Not, T, And, Or). DEFINITION 100. The negation For any qumix ρ ∈ D(⊗n C2 ), NOT(n) (ρ) := Not(n) ρ Not(n) . DEFINITION 101. The square root of the negation For any qumix ρ ∈ D(⊗n C2 ), √ √ (n) (n) √ (n)∗ NOT (ρ) := Not ρ Not , √ √ (n)∗ (n) is the adjoint of Not .73 where Not √ (n) It is easy to see that for any n ∈ N+ , both NOT(n) (ρ) and NOT (ρ) are qumixes of D(⊗n C2 ). DEFINITION 102. The conjunction Let ρ ∈ D(⊗n C2 ) and σ ∈ D(⊗m C2 ). (1)
(1)
AND(n,m,1) (ρ, σ) = T(n,m,1) (ρ, σ, P0 ) := T(n,m,1) (ρ ⊗ σ ⊗ P0 )T(n,m,1) . √ Like in the quregistercase, the gates NOT, NOT, T, AND, OR can be uniformly deﬁned on the set D of all qumixes. An interesting preorder relation can be deﬁned on the set of all qumixes. DEFINITION 103. Preorder ρ & σ iﬀ the following conditions hold: 73 See
Def. 155.
198
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
(i) p(ρ) ≤ p(σ); √ √ (ii) p( NOT(σ)) ≤ p( NOT(ρ)). One immediately shows that & is reﬂexive and transitive, but not antisymmetric. Counterexamples can be easily found in D(C2 ).
10.2 The compositional quantum computational semantics Both the compositional and the holistic semantics are based on the following intuitive idea: any sentence α of the language is interpreted as an appropriate qumix, that generally depends on the logical form of α; at the same time, the logical connectives are interpreted as special operations deﬁned in terms of gates. We will consider a minimal (sentential) quantum computational language L that contains a privileged atomic sentence f (whose intended interpretation is the truthvalue Falsity) and the following primitive connectives: the negation (¬), the square root √ (which corresponds to the Petriof the negation ( ¬), a ternary conjunction Toﬀoli gate). For any sentences α and β, the expression (α, β, f ) is a sentence of L. In this framework, the usual conjunction α β is dealt with as metalinguistic abbreviation for the ternary conjunction (α, β, f ). The disjunction connective () is supposed to be deﬁned via de Morgan (α β := ¬(¬α ¬β)). This minimal quantum computational language can be extended to richer languages containing other primitive connectives. We will ﬁrst introduce the notion of compositional quantum computational model (brieﬂy, compositional QCmodel or qumixmodel ). DEFINITION 104. Compositional QCmodel A compositional QCmodel of L is a map Qum that associates a qumix to any sentence α⎧ of L, satisfying the following conditions: ⎪ a density operator of D(C2 ) if α is an atomic sentence; ⎪ ⎪ ⎪ ⎪ ⎪ if α = f ; ⎨P0 Qum(α) := NOT(Qum(β)) if α = ¬β; ⎪ √ √ ⎪ ⎪ NOT(Qum(β)) if α = ¬β; ⎪ ⎪ ⎪ ⎩T(Qum(β), Qum(γ), Qum(f )) if α = (β, γ, f ). The concept of compositional QCmodel seems to have a “quasi intensional” feature: the meaning Qum(α) of the sentence α partially reﬂects the logical form of α. In fact, the dimension of the Hilbert space where Qum(α) “lives” depends on the number of occurrences of atomic sentences in α. DEFINITION 105. The atomic complexity of α ⎧ ⎪ 1 if α is an atomic sentence; ⎨ √ At(α) = At(β) if α = ¬β or α = ¬β; ⎪ ⎩ At(β) + At(γ) + 1 if α = (β, γ, f ). LEMMA 106. If At(α) = n, then Qum(α) ∈ D(⊗n C2 ).
Quantum Logic and Nonclassical Logics
199
We can say that the space ⊗At(α) C2 represents the semantic space where all possible meanings of α should “live”. Accordingly we will also write Hα instead of ⊗At(α) C2 . Given a model Qum, any sentence α has a natural probabilityvalue, which can be also regarded as its extensional meaning with respect to Qum. DEFINITION 107. The probabilityvalue of α in a model Qum pQum (α) := p(Qum(α)). As we have learnt, qumixes are preordered by the relation &. This suggests a natural deﬁnition of a logical consequence relation. DEFINITION 108. Consequence in a model Qum A sentence β is a consequence in a model Qum of a sentence α (α =Qum β) iﬀ Qum(α) & Qum(β). DEFINITION 109. Logical consequence A sentence β is a logical consequence of a sentence α (α = β) iﬀ for any model Qum, α =Qum β. We call quantum computational logic (QCL) the logic that is semantically characterized by the logical consequence relation we have just deﬁned. Hence, β is a logical consequence of α in the logic QCL (α =QCL β) iﬀ β is a consequence of α in any model Qum. So far we have considered models, where the meaning of any sentence is represented by a qumix. A natural question arises: do density operators have an essential role in characterizing the logic QCL? This question has a negative answer. In fact, one can prove that quregisters are suﬃcient for our logical aims in the case of the minimal quantum computational language L. Let us ﬁrst introduce the notion of qubitmodel. DEFINITION 110. Qubitmodel A qubitmodel of L is a function Qub that associates a quregister to any sentence α of L, satisfying the following conditions: ⎧ ⎪ a qubit in C2 if α is an atomic sentence; ⎪ ⎪ ⎪ ⎪ ⎪ 0 if α = f ; ⎨ Qub(α) := Not(Qub(β)) if α = ¬β; ⎪ √ √ ⎪ ⎪ Not(Qub(β)) if α = ¬β; ⎪ ⎪ ⎪ ⎩T(Qub(β), Qub(γ), Qub(f )) if α = (β, γ, f ). The notions of consequence and logical consequence are deﬁned like in the case of qumixmodels, mutatis mutandis. One can prove that the qubitsemantics and the qumixsemantics characterize the same logic.74 Quantum computational logic turns out to be a nonstandard form of quantum logic. Conjunction and disjunction do not correspond to lattice operations, because 74 See
[Dalla Chiara et al., 2003].
200
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
they are not generally idempotent (α QCL αα, αα QCL α). Unlike the usual (sharp and unsharp) versions of quantum logic, the weak distributivity principle breaks down ((α β) (α γ) QCL α (β γ)). At the same time, the strong distributivity, that is violated in orthodox quantum logic, is here valid (α (β γ) =QCL (α β) (α γ)). Both the excluded middle and the noncontradiction principles are violated. As a consequence, one can say that the logic arising from quantum computation represents, in a sense, a new example of fuzzy logic. The axiomatizability of QCL is an open problem.
10.3 Quantum trees The meaning and the probabilityvalue of any sentence α can be naturally described (and calculated) by means of a special conﬁguration called quantum tree, that illustrates a kind of reversible transformation of the atomic subformulas of α. The notion of quantum tree can be dealt with either in the framework of the qubitsemantics or in the framework of the qumixsemantics. In the ﬁrst case quantum trees will be called qubit trees, while in the second case we will speak of qumix trees. Any sentence α can be naturally decomposed into its parts, giving rise to a special conﬁguration called the syntactical tree of α ( indicated by ST reeα ). Roughly,75 ST reeα can be represented as a sequence of levels: Levelk (α) ... Level1 (α), where: • each Leveli (α) (with 1 ≤ i ≤ k) is a sequence of subformulas of α; • the bottom level (Level1 (α)) consists of α; • the top level (Levelk (α)) is the sequence of all atomic occurrences of α; • for any i (with 1 ≤ i < k), Leveli+1 (α) is the sequence obtained by dropping the principal connective 76 in all molecular formulas occurring at Leveli (α), and by repeating all the atomic sentences that possibly occur at Leveli (α). As an example, consider the following sentence: α = q ¬q = syntactical tree of α is the following conﬁguration: 75 A
(q, ¬q, f ). The
formal deﬁnition of syntactical8tree can be found in [Dalla Chiara and Leporini, 2005]. > ¬, if α = ¬β; V :V , if α = (β, γ, f ).
Quantum Logic and Nonclassical Logics
201
Level3 (α) = (q, q, f ); Level2 (α) = (q, ¬q, f ); Level1 (α) = ( (q, ¬q, f )).
By Height of α (indicated by Height(α)we mean the number of levels of the syntactical tree of α. For instance, Height( (q, ¬q, f )) = 3. The syntactical tree of α (which represents a purely syntactical object) uniquely determines a sequence of gates that are all deﬁned on the semantic space of α. We will call this gatesequence the qubit tree of α. Consider a sentence α such that At(α) = t and Height(α) = k. Let Levelij (α) represent the jth node of Leveli (α). Each Levelij (α) (where 1 ≤ i < Height(α)) can be naturally associated to a unitary operator Opji , according to the following operatorrule: ⎧ (1) ⎪ if Levelij (α) is an atomic sentence; ⎪ ⎪I ⎪ ⎨Not(r) if Levelij (α) = ¬β and At(β) = r; Opji := √ √ ⎪ Not(r) if Levelij (α) = ¬β and At(β) = r; ⎪ ⎪ ⎪ ⎩T(r,s,1) if Levelj (α) = (β, γ, f ), At(β) = r and At(γ) = s, i
where I(1) is the identity operator of C2 . On this basis, one can associate a gate Gα i to each Leveli (α) (such that 1 ≤ i < Height(α)): Leveli (α) % Gα := Opji , i j=1
where Leveli (α) is the length of the sequence Leveli (α). Being the tensor product of unitary operators, every Gα i turns out to be a are deﬁned on the same space, unitary operator. One can easily show that all Gα i Hα . DEFINITION 111. The qubit tree of α The qubit tree of α (denoted by QT reeα ) is the gatesequence α (Gα 1 , . . . , GHeight(α)−1 )
that is uniquely determined by the syntactical tree of α. As an example, consider again the sentence: α = (q, ¬q, f ). In order to construct the qubit tree of α, let us ﬁrst determine the operators Opji corresponding to each node of Streeα . We will obtain: • Op11 = T(1,1,1) , because (q, ¬q, f ) is connected with (q, ¬q, f ) (at Level2 (α));
202
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
• Op12 = I(1) , because q is connected with q (at Level3 (α)); • Op22 = Not(1) , because ¬q is connected with q (at Level3 (α)); • Op32 = I(1) , because f is connected with f (at Level3 (α)). α The qubit tree of α is represented by the gatesequence (Gα 1 , G2 ), where: 1 (1,1,1) Gα ; 1 = Op1 = T 1 2 3 (1) Gα ⊗ Not(1) ⊗ I(1) . 2 = Op2 ⊗ Op2 ⊗ Op2 = I
As we have seen, qubit trees consist of unitary operators (which can be applied to quregisters). The notion of qubit tree can be naturally generalized to qumixes. In such a case we will speak of qumix trees, and we will call quantum tree either α a qubit tree or a qumix tree. Let (Gα 1 , . . . , Gk−1 ) be the qubit tree of α. We can deﬁne the following sequence of functions on the set D(Hα ): D
α α∗ Gα 1 (ρ) = G1 ρ G1
... D
D
α α∗ Gα k−1 (ρ) = Gk−1 ρ Gk−1 .
One can easily prove that, for any ρ ∈ D(Hα ) and for any i (1 ≤ i ≤ k − 1), α Gα i (ρ) is a density operator of D(H ). The sequence QumT reeα = (D Gα 1,...,
D
Gα k−1 )
is called the qumix tree of α. D α Consider now a sentence α and let (D Gα 1 , . . . , Gk−1 ) be the qumix tree of α. Any choice of a qumix ρ in Hα determines a sequence (ρk , . . . , ρ1 ) of qumixes of Hα , where: ρk = ρ ρk−1 =
D
Gα k−1 (ρk )
... ρ1 =
D
Gα 1 (ρ2 )
The qumix ρk can be regarded as a possible inputinformation concerning the atomic parts of α, while ρ1 represents the outputinformation about α, given the inputinformation ρk . Each ρi corresponds to the information about Leveli (α), given the inputinformation ρk . How to determine an information about the parts of α under a given input? It is natural to apply the standard quantumtheoretic rule that determines the states of the parts of a compound system. Suppose that: Leveli (α) = (βi1 , . . . βir )
Quantum Logic and Nonclassical Logics
203
We have: Hα = Hβi1 ⊗ . . . ⊗ Hβir We know that QumT reeα and the choice of an input ρk (in Hα ) determine a sequence of qumixes: ρk Levelk (α) = (q1 , . . . qt ) ... ρi Leveli (α) = (βi1 , . . . , βir ) ... ρ1 Level1 (α) = (α) Consider redj (ρi ), the reduced state of ρi with respect to the jth subsystem (where 1 ≤ j ≤ r).77 From a semantic point of view, this state can be regarded as a contextual information about βij (the subformula of α occurring at the jth position at Leveli (α)) under the input ψk . Apparently, a contextual information about a subformula is generally a mixture. An interesting situation arises when the qumix ρk , representing a global information about the atomic parts of α, is an entangled pure state.78 As an example, consider the sentence α = ¬ (q, ¬q, f ) (which represents an example of the noncontradiction principle formalized in the quantum computational language). The inputinformation might be the following entangled state: 1 1 ψ4 = √ 110 + √ 000 Level4 (α) = (q, q, f ) 2 2 The reduced states of ψ4 turn out to be the following: (1) (1) red1 ( √12 110 + √12 000) = 12 P0 + 12 P1 red2 ( √12 110 +
√1 000) 2 √1 000) 2
(1)
= 12 P0
(1)
+ 12 P1
(1)
red3 ( √12 110 + = P0 Hence, the contextual information about both occurrences of q is the (proper) mixture 1 (1) 1 (1) P + P1 . 2 0 2 recall that redj (ρi ) is the unique density operator that satisﬁes the following condition: β for any selfadjoint operator Aj of H ij , 77 We
tr(redj (ρi )Aj ) = tr(ρi (I1 ⊗ . . . ⊗ Ij−1 ⊗ Aj ⊗ Ij+1 ⊗ . . . ⊗ Ir )), (where Ih is the identity operator of Hβih ). As a consequence, ρi and redj (ρi ) are statistically equivalent with respect to the jth subsystem of the compound system described by ρi . 78 As is well known, the basic features of an entangled state ψ are the following: 1) ψ is a maximal information (a pure state) that describes a compound physical system S; 2) the pieces of information determined by ψ about the parts of S are, generally, nonmaximal (proper mixtures). Hence, the information about the whole is more precise than the information about the parts.
204
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli (1)
At the same time, the contextual information about f is the false projection P0 . Quantum trees can be naturally regarded as examples of quantum circuits that compute outputs under given inputs. Since both qubit trees and qumix trees are determined by the syntactical tree of a given sentence, one can also say that any sentence of the quantum computational language plays the role of an intuitive and “economical” description of a quantum circuit.
10.4 Holistic semantics As we have seen, in the compositional quantum computational semantics, the meaning of a molecular sentence is determined by the meanings of its parts (like in classical logic). In this framework, the inputinformation about the top level of the syntactical tree of a sentence α is always associated to a factorized state ρ1 ⊗ . . . ⊗ ρt , where t is the atomic complexity of α and ρ1 , . . . , ρt are qumixes of C2 . As a consequence, the meaning of a molecular α cannot be a pure state, if the meanings of some atomic parts of α are proper mixtures. The holistic quantum compositional semantics 79 is based on a more “liberal” assumption: the input information about the toplevel of the syntactical tree of α can be represented by any qumix “living” in the semantic space of α. As a consequence, the meanings of all levels of ST reeα are not, generally, factorized states. Suppose that: Leveli (α) = (β1 , . . . , βr ). As we already know, the space Hα can be naturally regarded as the Hilbert space of a compound physical system consisting of r parts (mathematically represented by the spaces Hβ1 , . . . , Hβr ), where each part may be compound. On this basis, for any qumix ρi (associated to Leveli (α)) and for any node Levelij (α), we can consider the reduced state redj (ρi ) with respect to the jth subsystem of the system described by ρi . From an intuitive point of view, redj (ρi ) describes the jth subsystem on the basis of the global information ρi . Since Leveli (α) = (β1 , . . . , βr ), the qumix redj (ρi ) (which is a density operator of the space Hβj ) represents a possible meaning of the sentence βj . We can now introduce the basic deﬁnitions of the holistic semantics. Unlike compositional models, a holistic quantum computational model is a function Hol that assigns to any sentence α of the quantum computational language a global meaning that cannot be generally inferred from the meanings of the parts of α. Of course, the function Hol shall respect the logical form of α. In order to deﬁne the concept of holistic quantum computational model , we will ﬁrst introduce the notions of atomic holistic model and of tree holistic model . DEFINITION 112. Atomic holistic model An atomic holistic model is a map HolAt that associates a qumix to any sentence α of L , satisfying the following conditions: 79 In
[Dalla Chiara and Leporini, 2005] we have presented a weaker version of holistic semantics.
Quantum Logic and Nonclassical Logics
205
(1) HolAt (α) ∈ D(Hα ); (2) Let At(α) = n and LevelHeigth(α) = q1 , . . . , qn . Then, (2.1) if qj = f , then redj (HolAt (α)) = P0 ; (2.2) if qj and qh are two occurrences in α of the same atomic sentence, then redj (HolAt (α)) = redh (HolAt (α)). Apparently, HolAt (α) represents a global interpretation of the atomic sentences occurring in α. At the same time, redj (HolAt (α)), the reduced state of the compound system (described by HolAt (α)) with respect to the jth subsystem, represents a contextual meaning of qj with respect to the global meaning HolAt (α). Conditions (2.1) and (2.2) guarantee that HolAt (α) is well behaved. For, the contextual meaning of f is always the Falsity, while two diﬀerent occurrences (in α) of the same atomic sentence have the same contextual meaning. The map HolAt (which assigns a meaning to the toplevel of the syntactical tree of any sentence α) can be naturally extended to a map HolT ree that assigns a meaning to each level of the syntactical tree of any α, following the prescriptions of the qumix tree of α. Consider a sentence α such that: D α QumT reeα = (D Gα 1 , . . . , GHeigth(α)−1 ).
The map HolT ree is deﬁned as follows: HolT ree (LevelHeigth(α) ) = HolAt (α) HolT ree (Leveli (α)) =
D
T ree Gα (Leveli+1 (α)) i (Hol
(where Heigth(α) > i ≥ 1). On this basis, one can naturally deﬁne the notion of holistic (quantum computational ) model of L. DEFINITION 113. Holistic model A map Hol that assigns to any sentence α a qumix of the space Hα is called a holistic (quantum computational) model of L iﬀ there exists an atomic holistic model HolAt s.t.: Hol(α) = HolT ree (Level1 (α)), where HolT ree is the extension of HolAt . Given a sentence γ, Hol determines the contextual meaning, with respect to the context Hol(γ), of any occurrence of a subformula β in γ. DEFINITION 114. Contextual meaning of a node Let β be a subformula of γ occurring at the j − th position of the i − th level of the syntactical tree of γ. We indicate by β[ij ] the node of ST reeγ corresponding
206
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
to such occurrence. The contextual meaning of β[ij ] with respect to the context Hol(γ) is deﬁned as follows: Holγ (β[ij ]) := redj (HolT ree (Leveli (γ))). Hence, we have: Holγ (γ) = HolT ree (Level1 (γ)) = Hol(γ). Suppose that β[ij ] and β[hk ] are two nodes of the syntactical tree of γ, representing two occurrences of the same subformula β. One can show that: Holγ (β[ij ]) = Holγ (β[hk ]). In other words, two diﬀerent occurrences of one and the same subformula in a sentence γ receive the same contextual meaning with respect to the context Hol(γ). On this basis, one can deﬁne the contextual meaning of a subformula β of γ, with respect to the context Hol(γ): Holγ (β) := Holγ (β[ij ]), where β[ij ] is any occurrence of β at a node of ST reeγ . Suppose now that β is a subformula of two diﬀerent formulas γ and δ. Generally, we have: Holγ (β) = Holδ (β). In other words, sentences may receive diﬀerent contextual meanings in diﬀerent contexts! Apparently, Holγ is a partial function that only assigns meanings to the subformulas of γ. Given a sentence γ, we will call the partial function Holγ a contextual holistic model of the language. As expected, compositional models turn out to be limitcases of holistic models. One can easily prove that Hol represents a compositional model iﬀ the following condition is satisﬁed for any sentence α: HolAt (α) = Hol(q1 ) ⊗ . . . ⊗ Hol(qt ), where q1 , . . . , qt are the atomic sentences occurring in α. Unlike holistic models, compositional models are, of course, contextindependent. Suppose that β is a subformula of two diﬀerent formulas γ and δ, and let Hol represent a compositional model. We have: Holγ (β) = Holδ (β) = Hol(β). The notion of logical consequence in the framework of the holistic quantum computational semantics can be now deﬁned in a natural way.
Quantum Logic and Nonclassical Logics
207
Let us ﬁrst deﬁne the notion of consequence in a given contextual model . DEFINITION 115. Consequence in a given contextual model Holγ A sentence β is a consequence of a sentence α in a given contextual model Holγ (α =Holγ β) iﬀ 1. α and β are subformulas of γ; 2. Holγ (α) & Holγ (β) (where & is the preorder relation deﬁned in Def. 103). DEFINITION 116. Logical consequence (in the holistic semantics) A sentence β is a consequence of a sentence α (in the holistic semantics) iﬀ for any sentence γ such that α and β are subformulas of γ and for any Hol, α =Holγ β. We call HQCL the logic that is semantically characterized by the logical consequence relation we have just deﬁned. Hence, α =HQCL β iﬀ for any sentence γ such that α and β are subformulas of γ and for any Hol, α =Holγ β. Although the basic ideas of the holistic and of the compositional quantum computational semantics are quite diﬀerent, one can prove that HQCL and QCL are the same logic.80 In other words, for any sentences α and β, α =HQCL β iﬀ α =QCL β. This means that the logics (formalized in our “poor” sentential languages) are not able to capture the diﬀerence between an analytical and a holistic semantic procedure. The holistic quantum computational semantics provides a formalism that might represent a useful abstract tool for describing gestaltic semantic patterns, which arise in a number of diﬀerent rational and perceptual activities. As we have seen, an important role in this game is played by the notion of tensor product, which is mainly responsible for most holistic quantum phenomena. The compositional and analytical features of classical semantics (and of many other nonclassical approaches) are, instead, generally based on cartesian products. In this connection, an interesting question arises: to what extent is it possible (and reasonable) to try and generalize the tensorproduct formalism to some abstract semantic situations, that might be quite independent of the notion of Hilbert space? 11
MATHEMATICAL APPENDIX
We give here a survey of the deﬁnitions of some basic mathematical concepts that play a fundamental role in quantum logic. 80 See
[Dalla Chiara et al., 2006].
208
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
11.1 Algebraic structures DEFINITION 117. Poset A partially ordered set (called also poset) is a structure B = B , ≤ , where: B (the support of the structure) is a nonempty set and ≤ is a partial order relation on B. In other words, ≤ satisﬁes the following conditions for all a, b, c ∈ B: (i) a ≤ a (reﬂexivity); (ii) a ≤ b and b ≤ a implies a = b (antisymmetry); (iii) a ≤ b and b ≤ c implies a ≤ c (transitivity). DEFINITION 118. Chain Let B = B , ≤ be a poset. A chain in B is a subset C ⊆ B such that ∀a, b ∈ C: a ≤ b or b ≤ a. DEFINITION 119. Bounded poset A bounded poset is a structure B = B , ≤ , 0 , 1 , where: (i) B , ≤ is a poset; (ii) 0 and 1 are special elements of B: the minimum and the maximum with respect to ≤. In other words, for all b ∈ B: 0 ≤ b and b ≤ 1. DEFINITION 120. Lattice A lattice is a poset B = B , ≤ in which any pair of elements a, b has a meet a ∧ b (also called inﬁmum) and a join a ∨ b (also called supremum) such that: (i) a ∧ b ≤ a, b, and ∀c ∈ B: c ≤ a, b implies c ≤ a ∧ b; (ii) a, b ≤ a∨ b , and ∀c ∈ B: a, b ≤ c implies a ∨ b ≤ c. In any lattice the following condition holds: a ≤ b iﬀ a ∧ b = a iﬀ a ∨ b = b. DEFINITION 121. Complemented lattice A complemented lattice is a bounded lattice B where: ∀a ∈ B ∃b ∈ B such that a ∧ b = 0 and a ∨ b = 1. Let X be any set of elements of a lattice B. If existing, the inﬁmum X and the supremum X are the elements of B that satisfy the following conditions:
Quantum Logic and Nonclassical Logics
(ia) ∀a ∈ X :
209
X ≤ a; (ib) ∀c ∈ B : ∀a ∈ X[c ≤ a] implies c ≤ X; (iia) ∀a ∈ X : a ≤ X; (iib) ∀c ∈ B : ∀a ∈ X[a ≤ c] implies X ≤ c. On can show that, when they exist the inﬁmum and the supremum are unique. A lattice iscomplete iﬀ for any set of elements X the inﬁmum X and the supremum X exist. A lattice is σcomplete iﬀ for any countable set of elements X the inﬁmum X and the supremum X exist. In many situations, a poset (or a lattice) is closed under a unary operation that represents a weak form of logical negation. Such a ﬁner structure is represented by a bounded involution poset. DEFINITION 122. Bounded involution poset A bounded involution poset is a structure B = B , ≤ , , 0 , 1 where: (i) B , ≤ , 0 , 1 is a bounded poset; (ii)
is a unary operation (called involution or generalized complement) that satisﬁes the following conditions: (a) a = a
(double negation);
(b) a ≤ b implies b ≤ a
(contraposition).
The presence of a negationoperation permits us to deﬁne an orthogonality relation ⊥, that may hold between two elements of a bounded involution poset. DEFINITION 123. Orthogonality Let a and b belong to a bounded involution poset. The object a is orthogonal to the object b (indicated by a ⊥ b) iﬀ a ≤ b . A set of elements S is called a pairwise orthogonal set iﬀ ∀a, b ∈ S such that a = b, a ⊥ b. A maximal set of pairwise orthogonal elements is a set of pairwise orthogonal elements that is not a proper subset of any set of pairwise orthogonal elements. When a is not orthogonal to b we write: a ⊥ b. The orthogonality relation ⊥ is sometimes also called preclusivity; while its negation ⊥ is also called accessibility. Since, by deﬁnition of bounded involution poset, a ≤ b implies b ≤ a (contraposition) and a = a (double negation), one immediately obtains that ⊥ is a symmetric relation. Notice that 0 ⊥ 0 and that ⊥ is not necessarily irreﬂexive. It may happen that an object a (diﬀerent from the null object 0) is orthogonal to itself: a ⊥ a (because a ≤ a ). Objects of this kind are called selfinconsistent. Suppose we have two selfinconsistent objects a and b, and let us ask whether in such a case a is necessarily orthogonal to b. Generally, the answer to this question is negative. There are examples of bounded involution posets such that for some objects a and b:
210
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
a ⊥ a and b ⊥ b and a ⊥ b. DEFINITION 124. Kleene poset A bounded involution poset is a Kleene poset (or also a regular poset) iﬀ it satisﬁes the Kleene condition for any pair of elements a and b: a ⊥ a and b ⊥ b implies a ⊥ b. DEFINITION 125. Bounded involution lattice A bounded involution lattice is a bounded involution poset that is also a lattice. A Kleene lattice (or regular lattice) is a Kleene poset that is also a lattice. One can prove that a bounded involution lattice is regular iﬀ a ∧ a ≤ b ∧ b (for any pair of elements a and b). Generally, bounded involution lattices and Kleene lattices may violate both the noncontradiction principle and the excluded middle. In other words, it may happen that: a ∧ a = 0 and a ∨ a = 1. DEFINITION 126. Orthoposet and ortholattice An orthoposet is a bounded involution poset B = B , ≤ , , 0 , 1 that satisﬁes the conditions: (i) a ∧ a = 0 (noncontradiction principle); (ii) a ∨ a = 1 (excluded middle principle). An ortholattice is an orthoposet that is also a lattice. The involution operation of an orthoposet (ortholattice) is also called orthocomplementation (or shortly orthocomplement). A σorthocomplete orthoposet (σorthocomplete ortholattice) is an orthoposet (ortholattice) B such that for any countable set {ai }i∈I of pairwise orthogonal elements the supremum {ai }i∈I exists in B. DEFINITION 127. Distributive lattice A lattice B = B , ∧ , ∨ is distributive iﬀ the meet ∧ is distributed over the join ∨ and vice versa. In other words: (i) a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c); (ii) a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c). Distributive involution lattices are also called de Morgan lattices. In this framework, Boolean algebras can be then deﬁned as particular examples of de Morgan lattices. DEFINITION 128. Boolean algebra A Boolean algebra is a structure B = B , ∧ , ∨ , , 0 , 1
Quantum Logic and Nonclassical Logics
211
that is at the same time an ortholattice and a de Morgan lattice. In other words, Boolean algebras are distributive ortholattices. DEFINITION 129. Orthomodular poset and orthomodular lattice An orthomodular poset is an orthoposet B = B , ≤ , , 0 , 1 that satisﬁes the following conditions: (i) ∀a, b ∈ B, a ⊥ b implies a ∨ b ∈ B; (ii) ∀a, b ∈ B, a ≤ b implies b = a ∨ (a ∨ b ) . An orthomodular lattice is an orthomodular poset that is also a lattice. Clearly, any distributive ortholattice (i.e., any Boolean algebra), is orthomodular. DEFINITION 130. Modularity A lattice B is called modular iﬀ ∀a, b ∈ B, a ≤ b implies ∀c ∈ B[a ∨ (b ∧ c) = (a ∨ b) ∧ (a ∨ c)]. Every modular ortholattice is orthomodular, but not the other way around. Furthermore, any distributive lattice is modular. A bounded poset (lattice) B may contain some special elements, called atoms. DEFINITION 131. Atom An element b of B is called an atom of B iﬀ b covers 0. In other words, b = 0 and ∀c ∈ B: c ≤ b implies c = 0 or c = b. Apparently, atoms are nonzero elements such that no other element lies between them and the latticeminimum. DEFINITION 132. Atomicity A bounded poset B is atomic iﬀ ∀a ∈ B − {0} there exists an atom b such that b ≤ a. Of course, any ﬁnite bounded poset is atomic. At the same time, there are examples of inﬁnite bounded posets that are atomless (and hence nonatomic), the real interval [0, 1] being the most familiar example. It turns out that any atomic orthomodular lattice B is atomistic in the sense that any element can be represented as the supremum of a set of atoms, i.e., for any element a there exists a set {bi }i∈I of atoms such that a = {bi }i∈I . DEFINITION 133. Covering property A lattice B satisﬁes the covering property iﬀ ∀a, b ∈ B: if a covers a ∧ b, then a ∨ b covers b.
212
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
It turns out that an atomic lattice B has the covering property iﬀ for every atom a of B and for every element b ∈ B such that a ∧ b = 0, the element a ∨ b covers b. One of the most signiﬁcant quantum relations, compatibility, admits a purely algebraic deﬁnition. DEFINITION 134. Compatibility Let B be an orthomodular lattice and let a and b be elements of B. The element a is called compatible with the element b iﬀ a = (a ∧ b ) ∨ (a ∧ b). One can show that the compatibility relation is symmetric. The proof uses the orthomodular property in an essential way. Clearly, if B is a Boolean algebra, then any element is compatible with any other element by distributivity. One can prove that a, b are compatible in the orthomodular lattice B iﬀ the subalgebra of B generated by {a, b} is Boolean. DEFINITION 135. Irreducibility Let B be an orthomodular lattice. B is said to be irreducible iﬀ {a ∈ B : ∀b ∈ B (a is compatible with b)} = {0, 1} . If B is not irreducible, it is called reducible. DEFINITION 136. Separability An orthomodular lattice B is called separable iﬀ every set of pairwise orthogonal elements of B is countable. DEFINITION 137. Group A group is a structure G = G , + , − , 0, where + is a binary operation, − is a unary operation, 0 is a special element. The following conditions hold: (i) G , + , 0 is a monoid . In other words, (a) the operation + is associative: a + (b + c) = (a + b) + c; (b) 0 is the neutral element: a + 0 = a; (ii) ∀a ∈ G, −a is the inverse of a: a + (−a) = 0. An Abelian monoid (group) is a monoid (group) in which the operation + is commutative: a + b = b + a. DEFINITION 138. Ring A ring is a structure D = D , + , · , − , 0 that satisﬁes the following conditions:
Quantum Logic and Nonclassical Logics
213
(i) D , + , 0 is an Abelian group; (ii) the operation · is associative: a · (b · c) = (a · b) · c; (iii) the operation · distributes over + on both sides, i.e., ∀a, b, c ∈ D: (a) a · (b + c) = (a · b) + (a · c); (b) (a + b) · c = (a · c) + (b · c). If there is an element 1 in D that is neutral for · (i.e., if D , · , 1 is a monoid), then the ring is called a ring with unity. A ring is trivial in case it has only one element, otherwise it is nontrivial . It is easy to see that a ring with unity is nontrivial iﬀ 0 = 1. A commutative ring is a ring in which the operation · is commutative. DEFINITION 139. Division ring A division ring is a nontrivial ring D with unity such that any nonzero element is invertible; in other words, for any a ∈ D (a = 0), there is an element b ∈ D such that a · b = b · a = 1. DEFINITION 140. Field A ﬁeld is a commutative division ring. Both the real numbers (IR) and the complex numbers (C) give rise to a ﬁeld. An example of a genuine division ring (where · is not commutative) is given by the quaternions (Q).
11.2
Hilbert spaces
DEFINITION 141. Vector space A Vector space over a division ring D is a structure V = V , + , − , · , 0 that satisﬁes the following conditions: (i) V , + , − , 0 (the vector structure) is an Abelian group, where 0 (the null vector ) is the neutral element; (ii) for any element a of the division ring D and any vector ϕ of V , aϕ (the scalar product of a and ϕ) is a vector in V . The following conditions hold for any a, b ∈ D and for any ϕ, ψ ∈ V : (a) a(ϕ + ψ) = (aϕ) + (aψ); (b) (a + b)ϕ = (aϕ) + (bϕ); (c) a(bϕ) = (a · b)ϕ; (d) 1ϕ = ϕ.
214
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
The elements (vectors) of a vector space V are indicated by ϕ, ψ, χ, . . ., while a, b, c, . . . represent elements (scalars) of the division ring D. Any ﬁnite sum of vectors ψ1 , . . . , ψn is indicated by ψ1 + . . . + ψn (or i∈K ψi , when K = {1, . . . , n}.) On this basis, one can introduce the notion of preHilbert space. Hilbert spaces are then deﬁned as special cases of preHilbert spaces. We will only consider preHilbert spaces (and Hilbert spaces) whose division ring is either IR or C. DEFINITION 142. PreHilbert space Let D be the ﬁeld of the real or the complex numbers. A preHilbert space over D is a vector space V over D, equipped with an inner product .. that associates to any pair of vectors ϕ, ψ ∈ V an element ϕψ ∈ D. The following conditions are satisﬁed for any ϕ, ψ, χ ∈ V and any a ∈ D: (i) ϕϕ ≥ 0; (ii) ϕϕ = 0 iﬀ ϕ = 0; (iii) ψaϕ = aψϕ; (iv) ϕψ + χ = ϕψ + ϕχ; (v) ϕψ = ψϕ∗ , where tion if D = C.
∗
is the identity if D = IR, and the complex conjuga
The inner product .. permits one to generalize some geometrical notions of ordinary 3dimensional spaces. DEFINITION 143. Norm of a vector The norm 'ϕ' of a vector ϕ is the number ϕϕ1/2 . A unit (or normalized) vector is a vector ψ such that 'ψ' = 1. Two vectors ϕ, ψ are called orthogonal iﬀ ϕψ = 0. DEFINITION 144. Orthonormal set of vectors A set {ψi }i∈I of vectors is called orthonormal iﬀ its elements are pairwise orthogonal unit vectors. In other words: (i) ∀i, j ∈ I(i = j) : ψi ψj = 0; (ii) ∀i ∈ I : 'ψi ' = 1. The norm '.' induces a metric d on the preHilbert space V: d(ψ, ϕ) := 'ψ − ϕ'. We say that a sequence {ψi }i∈N of vectors in V converges in norm (or simply converges) to a vector ϕ of V iﬀ limi→∞ d(ψi , ϕ) = 0. In other words, ∀ ε > 0 ∃n ∈ N ∀k > n : d(ψk , ϕ) < ε. A Cauchy sequence is a sequence {ψi }i∈N of vectors in V such that ∀ε > 0 ∃n ∈ N ∀h > n ∀k > n : d(ψh , ψk ) < ε.
Quantum Logic and Nonclassical Logics
215
It is easy to see that whenever a sequence {ψi }i∈N of vectors in V converges to a vector ϕ of V , then {ψi }i∈N is a Cauchy sequence. The crucial question is the converse one: which are the preHilbert spaces in which every Cauchy sequence converges to an element in the space? DEFINITION 145. Metrically complete preHilbert space A preHilbert space V with inner product .. is metrically complete with respect to the metric d induced by .. iﬀ every Cauchy sequence of vectors in V converges to a vector of V . DEFINITION 146. Hilbert space A Hilbert space is a metrically complete preHilbert space. A real (complex ) Hilbert space is a Hilbert space whose division ring is IR (C). The notion of preHilbert space (Hilbert space) can be generalized to the case where the division ring is represented by Q (the division ring of all quaternions). Consider a Hilbert space H over a division ring D. DEFINITION 147. (Hilbert) linear combination Let {ψi }i∈I be a set of vectors of H and let {ai }i∈I ⊆ D. A vector ψ is called a (Hilbert) linear combination (or superposition) of {ψi }i∈I (with scalars {ai }i∈I ) iﬀ ∀ε ∈ IR+ there is a ﬁnite set J ⊆ I such that for any ﬁnite subset K of I including J: 'ψ − ai ψi ' ≤ ε. i∈K
Apparently, when existing, the linear combination of {ϕi }i∈I (with scalars {ai }i∈I ) is unique. We denote it by i∈I ai ψi . When no confusion is possible, the index set I will be omitted. DEFINITION 148. Orthonormal basis An orthonormal basis of H is a maximal orthonormal set {ψi }i∈I of H. In other words, {ψi }i∈I is an orthonormal set such that no orthonormal set includes {ψi }i∈I as a proper subset. One can prove that every Hilbert space H has an orthonormal basis and that all orthonormal bases of H have the same cardinality. The dimension of H is then deﬁned as the cardinal number of any basis of H. Let {ψi }i∈I be any orthonormal basis of H. One can prove that every vector ϕ of H can be expressed in the following form: ψi ϕψi . ϕ = i∈I
Hence, ϕ is a linear combination of {ψi }i∈I with scalars ψi ϕ (the scalars ψi ϕ are also called Fourier coeﬃcients.) A Hilbert space H is called separable iﬀ H has a countable orthonormal basis. In the following, we will always refer to separable Hilbert spaces. DEFINITION 149. Closed subspace
216
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
A closed subspace of H is a set X of vectors that satisﬁes the following conditions: (i) X is a subspace of H. In other words, X is closed under ﬁnite linear combinations. Hence, ψ, ϕ ∈ X implies aψ + bϕ ∈ X; (ii) X is closed under limits of Cauchy sequences. In other words: if each element of a Cauchy sequence of vectors belongs to X, then also the limit of the sequence belongs to X. The set of all closed subspaces of H is indicated by C(H). For any vector ψ, we indicate by [ψ] the unique 1dimensional closed subspace that contains ψ. DEFINITION 150. Operator An operator of H is a map A : Dom(H) → H, where Dom(A) (the domain of A) is a subset of H. DEFINITION 151. Densely deﬁned operator A densely deﬁned operator of H is an operator A that satisﬁes the following condition: ∀ε ∈ IR+ ∀ψ ∈ H ∃ϕ ∈ Dom(A) [d(ψ, ϕ) < ε], where d represents the metric induced by ... DEFINITION 152. Linear operator A linear operator on H is an operator A that satisﬁes the following conditions: (i) Dom(A) is a closed subspace of H; (ii) ∀ψ, ϕ ∈ Dom(A) ∀a, b ∈ D : A(aψ + bϕ) = aAψ + bAϕ. In other words, a characteristic of linear operators is preserving the linear combinations. DEFINITION 153. Bounded operator A linear operator A is called bounded iﬀ there exists a positive real number a such that ∀ψ ∈ H : 'Aψ' ≤ a'ψ'. The set B(H) of all bounded operators of H turns out to be closed under the operator sum, the operator product and the scalar product. In other words, if A ∈ B(H) and B ∈ B(H), then A + B ∈ B(H) and A.B ∈ B(H) ; for any scalar a, if B ∈ B(H), then aB ∈ B(H). DEFINITION 154. Positive operator A bounded operator A is called positive iﬀ ∀ψ ∈ H : ψAψ ≥ 0. DEFINITION 155. The adjoint operator
Quantum Logic and Nonclassical Logics
217
Let A be a densely deﬁned linear operator of H. The adjoint of A is the unique operator A∗ such that ∀ψ ∈ Dom(A) ∀ϕ ∈ Dom(A∗ ) : Aψϕ = ψA∗ ϕ . DEFINITION 156. Selfadjoint operator A selfadjoint operator is a densely deﬁned linear operator A such that A = A∗ . If A is selfadjoint, then ∀ψ, ϕ ∈ Dom(A) : Aψϕ = ψAϕ. If A is selfadjoint and everywhere deﬁned (i.e., Dom(A) = H), then A is bounded. DEFINITION 157. Projection operator A projection operator is an everywhere deﬁned selfadjoint operator P that satisﬁes the idempotence property: ∀ψ ∈ H : P ψ = P P ψ. There are two special projections O and I called the zero (or null projection) and the identity projection which are deﬁned as follows: ∀ψ ∈ H, Oψ = 0 and Iψ = ψ. Any projection other than O and I is called a nontrivial projection. Thus, P is a projection operator if Dom(P ) = H and P = P 2 = P ∗ . The set of all projection operators will be indicated by Π(H). One can prove that the set C(H) of all closed subspaces and the set Π(H) of all projections of H are in onetoone correspondence. Let X be a closed subspace of H. By the projection theorem every vector ψ ∈ H can be uniquely expressed as a linear combination ψ1 + ψ2 , where ψ1 ∈ X and ψ2 is orthogonal to any vector of X. Accordingly, we can deﬁne an operator PX on H such that ∀ψ ∈ H : PX ψ = ψ1 (in other words, PX transforms any vector ψ into the “Xcomponent” of ψ) It turns out that PX is a projection operator of H. Conversely, we can associate to any projection P its range, XP = {ψ : ∃ϕ(P ϕ = ψ)} , which turns out to be a closed subspace of H. For any closed subspace X and for any projection P , the following conditions hold: X(PX ) = X; P(XP ) = P. DEFINITION 158. The trace functional Let {ψi }i∈I be any orthonormal basis for H and let A be a positive operator. The trace of A (indicated by tr(A)) is deﬁned as follows: ψi Aψi . tr(A) := i
218
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
One can prove that the deﬁnition of tr is independent of the choice of the basis. For any positive operator A, there exists a unique positive operator B such that: B 2 = A. If A is a (not necessarily positive) bounded operator, then A∗ A is positive. Let A be the unique positive operator such that A2 = A∗ A. A bounded operator A is called a traceclass operator iﬀ tr(A) < ∞. DEFINITION 159. Density operator A density operator is a positive, selfadjoint, traceclass operator ρ such that tr(ρ) = 1. It is easy to see that, for any vector ψ, the projection P[ψ] onto the 1dimensional closed subspace [ψ] is a density operator. DEFINITION 160. Unitary operator A unitary operator is a linear operator U such that: • Dom(U ) = H; • U U ∗ = U ∗ U = I. One can show that the unitary operators U are precisely the operators that preserve the inner product. In other words, for any ψ, ϕ ∈ H : ψϕ = U ψU ϕ . Any pair of Hilbert spaces H1 , H2 gives rise to two new Hilbert spaces H1 ⊗ H2 and H1 ⊕ H2 that represent the tensor product of H1 and H2 and the direct sum of H1 and H2 , respectively. Tensor products play an important role for the mathematical representation of compound quantum systems. They are also systematically used in the mathematical formalism of quantum computation. DEFINITION 161. Tensor product Hilbert space Let H1 and H2 be two Hilbert spaces over the same ﬁeld D (the real or the complex numbers). A Hilbert space H is the tensor product of H1 and H2 iﬀ the following conditions are satisﬁed: (i) there exists a map (called tensor product) from the cartesian product H1 ×H2 into H that satisﬁes the following conditions: (a) the tensor product ⊗ is linear in each “slot”; in other words, ∀ψ , ϕ ∈ H1 ∀χ, δ ∈ H2 ∀a, b ∈ D: (a1) (aψ + bϕ) ⊗ χ = (aψ) ⊗ χ + (bϕ) ⊗ χ; (a2) ψ ⊗ (aχ + bδ) = ψ ⊗ (aχ) + ψ ⊗ (bδ); (b) the external product with a scalar carries across the tensor product; in other words, ∀ψ ∈ H1 ∀ϕ ∈ H2 ∀a ∈ D: a(ψ ⊗ ϕ) = (aψ) ⊗ ϕ = ψ ⊗ (aϕ).
Quantum Logic and Nonclassical Logics
219
(ii) every vector of H can be expressed as a linear combination of vectors of the set {ϕ ⊗ ψ : ϕ ∈ H1 , ψ ∈ H2 }. One can show that the tensor product is unique up to isomorphism. As required by condition (ii), every vector of H1 ⊗ H2 can be expressed as a linear combination of vectors of the form ψ ⊗ ϕ (where ψ ∈ H1 , ϕ ∈ H2 ). At the same time, there are vectors of H1 ⊗ H2 that cannot be written as a single product ψ ⊗ ϕ for any ψ ∈ H1 , ϕ ∈ H2 . These vectors are called non factorized. If {ψi }i∈I and {ϕj }j∈J are orthonormal bases for H1 and H2 , respectively, then the set {ϕi ⊗ ψj : i ∈ I , j ∈ J} is an orthonormal basis of the tensor product Hilbert space. In particular, if {ψi , . . . , ψn } and {ϕ1 , . . . ϕm } are orthonormal bases of the ﬁnite dimensional Hilbert spaces H1 , H2 , then every vector ψ ∈ H1 ⊗ H2 can be written as ψ =
m n
aij ψi ⊗ ϕj .
i=1 j=1
This shows that the dimension of the tensor product Hilbert space H1 ⊗ H2 is the product of the dimensions of H1 and H2 . DEFINITION 162. Direct sum Hilbert space Let H1 and H2 be two Hilbert spaces over the same ﬁeld D (the real or the complex numbers), with inner products · ·1 and · ·2 , respectively. The direct sum of H1 and H2 (denoted by H1 ⊕ H2 ) is the vector space based on the cartesian product H1 × H2 , where the vector operations are deﬁned as follows ∀ ψ1 , ϕ1 ∈ H1 , ∀ψ2 , ϕ2 ∈ H2 , ∀a ∈ D: (i) (ψ1 , ψ2 ) + (ϕ1 , ϕ2 ) = (ψ1 + ϕ1 , ψ2 + ϕ2 ); (ii) a(ψ1 , ψ2 ) = (aψ1 , aψ2 ). The inner product · · of H1 ⊕ H2 is deﬁned in the following way: (iii) (ψ1 , ψ2 )(ϕ1 , ϕ2 ) = ψ1 ϕ1 1 + ψ2 ϕ2 2 . One can easily show that H1 ⊕ H2 is a Hilbert space. Instead of (ψ1 , ψ2 ) we will write ψ1 ⊕ ψ2 . Clearly, the subspaces H1 × {0} and {0} × H2 of H1 ⊕ H2 are isomorphic to H1 and H2 , respectively. ACKNOWLEDGEMENTS We warmly thank Giuseppe Sergioli for his precious suggestions concerning Section 9.
220
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
BIBLIOGRAPHY [Aerts, 2000] D. Aerts and B. Van Steirteghem, Quantum axiomatics and a theorem of M.P. Sol´ er, International Journal of Theoretical Physics 39 (2000), 497–502. [Altenkirch and Grattage, 2005] T. Altenkirch and J. Grattage, A functional quantum programming language, Proceedings of the XX Annual Symposium on Logic and Computer Science, IEEE, 2005. [Arrighi and Dowek, 2004] P. Arrighi and G. Dowek, Operational semantics for formal tensorial calculus, Languages (P. Selinger, ed.), Turku Centre for Computer Science, 2004. [Battilotti, 1998] G. Battilotti, Embedding classical logic into basic orthologic with a primitive modality, Logic Journal of the IGPL, 6 (1998), 383–402. [Battilotti and Faggian, 2002] G. Battilotti and C. Faggian, Quantum logic and the cube of logics, Handbook of Philosophical Logic (D. M. Gabbay and F. Guenthner, eds.), vol. 6, Kluwer Academic Publishers, Dordrecht, 2002, pp. 213–226, section 17 in [Dalla Chiara and Giuntini, 2002]. [Battilotti and Sambin, 1999] G. Battilotti and G. Sambin, Basic logic and the cube of its extensions, Logic and Foundations of Mathematics (A. Cantini, E. Casari, and P. Minari, eds.), Kluwer Academic Publishers, Dordrecht, 1999, pp. 165–186. [Beltrametti and Bugajski, 1995] E. Beltrametti and S. Bugajski, A classical extension of quantum mechanics, Journal of Physics A: Mathematical and General 28 (1995), 247–261. [Beltrametti and Bugajski, 1997] E. Beltrametti and S. Bugajski. Eﬀect algebras and statistical physical theories, Journal of Mathematical Physics 38 (1997), 3020–3030. [Beltrametti and Cassinelli, 1981] E. Beltrametti and G. Cassinelli, The logic of quantum mechanics, Encyclopedia of Mathematics and its Applications, vol. 15, AddisonWesley, Reading, 1981. [Bennett and Foulis, 1997] M. K. Bennett and D. J. Foulis, Interval and scale eﬀect algebras, Advances in Mathematics 19 (1997), 200–215. [Birkhoﬀ, 1940] G. Birkhoﬀ, Lattice Theory, 3rd (new) ed., Colloquium Publications, vol. 25, American Mathematical Society, Providence, 1967. [Birkhoﬀ and von Neumann, 1936] G. Birkhoﬀ and J. von Neumann, The logic of quantum mechanics, Annals of Mathematics 37 (1936), 823843, in [von Neumann, 1961b]. [Bruns et al., 1990] G. Bruns, R. J. Greechie, J. Harding, and M. Roddy, Completions of orthomodular lattices, Order 7 (1990), 789–807. [Bub, 1999] J. Bub, Interpreting the quantum world, Cambridge University Press, Cambridge, 1999. [Bugajski, 1993] S. Bugajski, Delinearization of quantum logic, International Journal of Theoretical Physics 32 (1993), 389–398. [Busch, 1985] P. Busch, Elements of unsharp reality in the EPR experiment, Symposium on the Foundations of Modern Physics (P. Lahti and P. Mittelstaedt, eds.), World Scientiﬁc, Singapore, 1985, pp. 343–357. [Busch et al., 1995] P. Busch, M. Grabowski, and P. Lahti, Operational quantum mechanics, Lectures Notes in Physics, no. m31, Springer, Berlin, 1995. [Busch et al., 1991] P. Busch, P. Lahti, and P. Mittelstaedt, The quantum theory of measurement, Lectures Notes in Physics, no. m2, Springer, Berlin, 1991. [Cattaneo, 1993] G. Cattaneo, Fuzzy quantum logic II: the logics of unsharp quantum mechanics, International Journal of Theoretical Physics 32 (1993), 1709–1734. [Cattaneo, 1997] G. Cattaneo. A uniﬁed framework for the algebra of unsharp quantum mechanics, International Journal of Theoretical Physics 36 (1997), 3085–3117. [Cattaneo et al., 1999] G. Cattaneo, M. L. Dalla Chiara, and R. Giuntini, How many notions of ’sharp’ ?, International Journal of Theoretical Physics 38 (1999), 3153–3161. [Cattaneo et al., 1989] G. Cattaneo, C. Garola, and G. Nistic´ o, Preparationeﬀect versus questionproposition structures, Physics Essays 2 (1989), 197–216. [Cattaneo and Giuntini, 1995] G. Cattaneo and R. Giuntini, Some results on BZ structures from hilbertian unsharp quantum physics, Foundations of Physics 25 (1995), 1147–1182. [Cattaneo and Gudder, 1999] G. Cattaneo and S. P. Gudder, Algebraic structures arising in axiomatic unsharp quantum physics, Foundations of Physics 29 (1999), 1607–1637. [Cattaneo and Laudisa, 1994] G. Cattaneo and F. Laudisa, Axiomatic unsharp quantum theory (from Mackey to Ludwig), Foundations of Physics 24 (1994), 631–683.
Quantum Logic and Nonclassical Logics
221
[Cattaneo and Nistic´ o, 1989] G. Cattaneo and G. Nistic´ o, BrouwerZadeh posets and threevalued L ukasiewicz posets, Fuzzy Sets and Systems 33 (1986), 165–190. [Chang, 1958] C. C. Chang, Algebraic analysis of many valued logics, Transactions of the American Mathematical Society 88 (1958), 74–80. [Chang, 1959] C. C. Chang. A new proof of the completeness of L ukasiewicz axioms, Transactions of the American Mathematical Society 93 (1959), 467–490. [Cignoli and D’Ottaviano, 2000] R. Cignoli, I. M. L. D’Ottaviano, and D. Mundici, Algebraic foundations of manyvalued reasoning, Trends in Logic, vol. 7, Kluwer Academic Publishers, Dordrecht, 2000. [Coecke et al., 2000] B. Coecke, D. Moore and A. Wilce, Operational quantum logic: An overview, Current Research in Operational Quantum Logic (B. Coecke, D. Moore and A. Wilce, eds.), Kluwer, Dordrecht, 2000, pp. 136. [Cutland and Gibbins, 1982] N. J. Cutland and P .F. Gibbins, A regular sequent calculus for quantum logic in which ∧ and ∨ are dual, Logique et Analyse — Nouvelle Serie  25 (1982), no. 45, 221–248. [da Costa and Krause, 2005] N.C.A. da Costa, D. Krause, Remarks on the applications of paraconsistent logic to physics, Filosoﬁa, Ciencia e Historia (M. Pietrocola and O. Freire, eds.), Discurso Editorial, S. Paulo, 2005, pp. 337359. [da Costa et al., 2006] N.C.A. da Costa, D. Krause and O. Bueno, Paraconsistent logic and paraconsistency: Technical and philosophical developments, Handbook of Philosophy of Science (D. Jacquette, ed.), Elsevier, Amsterdam, 2006. [Dalla Chiara, 1981] M. L. Dalla Chiara, Some metalogical pathologies of quantum logic, Current Issues in Quantum Logic (E. Beltrametti and B. van Fraassen, eds.), Ettore Majorana International Science Series, vol. 8, Plenum, New York, 1981, pp. 147–159. [Dalla Chiara and Giuntini, 1989] M. L. Dalla Chiara and R. Giuntini, Paraconsistent quantum logics, Foundations of Physics 19 (1989), 891–904. [Dalla Chiara and Giuntini, 1994] M. L. Dalla Chiara and R. Giuntini. Unsharp quantum logics, Foundations of Physics 24 (1994), 1161–1177. [Dalla Chiara and Giuntini, 1995] M. L. Dalla Chiara and R. Giuntini. The logics of orthoalgebras, Studia Logica 55 (1995), 3–22. [Dalla Chiara and Giuntini, 1999] M. L. Dalla Chiara and R. Giuntini, L ukasiewicz theory of truth, from the quantum logical point of view, Alfred Tarski and the Vienna Circle (J. Wol´enski and E. K¨ ohler, eds.), Kluwer, Dordrecht, 1999, pp. 127–134. [Dalla Chiara and Giuntini, 2002] M. L. Dalla Chiara and R. Giuntini. Quantum logics, Handbook of Philosophical Logic (D.M. Gabbay and F. Guenthner, eds.), vol. 6, Kluwer Academic Publishers, Dordrecht, 2002, pp. 129–228. [Dalla Chiara et al., 2004] M. L. Dalla Chiara, R. Giuntini, R. Greechie, Reasoning in quantum theory. Sharp and unsharp quantum logics, Kluwer Academic Publishers, Dordrecht, 2004. [Dalla Chiara et al., 2003] M. L. Dalla Chiara, R. Giuntini, and R. Leporini, Quantum Computational Logics. A Survey, Studia Logica, 21 (2003), 213–255. [Dalla Chiara et al., 2004a] M. L. Dalla Chiara, R. Giuntini and R. Leporini, Quantum computational logics and Fock space semantics, International Journal of Quantum Information, 2 (2004), 1–8. [Dalla Chiara and Leporini, 2005] M. L. Dalla Chiara, R. Giuntini and R. Leporini, Logics from quantum computation, International Journal of Quantum Information, 3 (2005), 293–337. [Dalla Chiara et al., 2006] M. L. Dalla Chiara, R. Giuntini and R. Leporini, Compositional and Holistic Quantum Computational Semantics, to appear in Natural Computing, 2006. [Danos and Reignier, 1989] V. Danos and L. Reignier, The structure of multiplicatives, Archive for Mathematical Logic 28 (1989), 181203. [Davies, 1976] E. B. Davies, Quantum theory of open systems, Academic, New York, 1976. [DestouchesF´evrier, 1937] P. DestouchesF´evrier, Les relations d’incertitude de Heisenberg et la logique, Comptes Rendus de l’Acad´emie de Sciences 204 (1937), 481483. [Dishkant, 1972] H. Dishkant, Semantics of the minimal logic of quantum mechanics, Studia Logica 30 (1972), 17–29. [DunnHardegree, 2001] J. M. Dunn and G. M. Hardegree, Algebraic methods in philosophical logic, Clarendon Press, Oxford, 2001. [Dvureˇ censkij, 1993] A. Dvureˇ censkij, Gleason’s theorem and its applications, Mathematics and its Applications, no. 60, Kluwer, Dordrecht, 1993.
222
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
[Dvureˇ censkij, 1997] A. Dvureˇ censkij, Measures and ⊥decomposable measures of eﬀects of a Hilbert space, Atti del Seminario Matematico e Fisico dell’Universit` a di Modena 45 (1997), 259–288. [Dvureˇ censkij, 2000] A. Dvureˇ censkij, New trends in quantum structures, Mathematics and Its Applications, vol. 516, Kluwer Academic Publishers, Dordrecht, 2000. [Einstein et al., 1035] A. Einstein, B. Podolsky, and N. Rosen, Can quantummechanical description of reality be considered complete?, Physical Review 47 (1935), 777–780. [Engesser and Gabbay, 2002] K. Engesser and D. Gabbay, Quantum logic, Hilbert space, revision theory, Artiﬁcial Intelligence 136 (2002), 61–100. [Faggian, 1998] C. Faggian, Classical proofs via basic logic, Computer Science Logic 11th International Workshop, CSL’97 (M. Nielson and W. Thomas, eds.), Lecture Notes in Computer Science, vol. 1414, Springer Verlag, 1998, pp. 203–219. [Faggian and Sambin, 1997] C. Faggian and G. Sambin, From basic logic to quantum logics with cutelimination, International Journal of Theoretical Physics 12 (1997), 31–37. [Finch, 1970] P. D. Finch, Quantum logic as an implication algebra, Bulletin of the Australian Mathematical Society 2 (1970), 101–106. [Fitting, 1969] M. Fitting, Intuitionistic logic, model theory and forcing, NorthHolland, Amsterdam, 1969. [Foulis, 1999] D. J. Foulis, A halfcentury of quantum logic, what have we learned?, Quantum Structures and the Nature of Reality (D. Aerts and J. Pykacz, eds.), vol. 7, Kluwer Academic Publishers, Dordrecht, 1999, pp. 1–36. [Foulis, 2000] D. J. Foulis, MV and Heyting eﬀect algebras, Foundations of Physics 30 (2000), 1687–1706. [Foulis and Bennett, 1994] D. J. Foulis and M. K. Bennett, Eﬀect algebras and unsharp quantum logics, Foundations of Physics 24 (1994), 1325–1346. [Foulis and Greechie, 2000] D. J. Foulis and R. J. Greechie, Speciﬁcation of ﬁnite eﬀect algebras, International Journal of Theoretical Physics 39 (2000), 665–676. [Foulis and Randall, 1981] D. J. Foulis and C. H. Randall, Empirical logic and tensor product, Interpretation and Foundations of Quantum Mechanics, Grundlagen der exakten Naturwissenschaften, vol. 5, Bibliographisches Institut, Mannheim, 1981, pp. 9–20. [Foulis and Randall, 1983] D. J. Foulis and C. H. Randall, Properties and operational propositions in quantum mechanics, Foundations of Physics 13 (1983), 843–857. [Foulis et al., 1996] D.J. Foulis, R.J. Greechie, M. L. Dalla Chiara, and R. Giuntini, Quantum Logic, Encyclopedia of Applied Physics (G. Trigg, ed.), vol. 15, VCH Publishers, 1996, pp. 229–255. [Gallier, 1991] J. Gallier, Constructive Logics. II: Linear Logic and Proof Nets, Research Report no. 9, Digital PRL, 1991. [Garola, 1980] C. Garola, Propositions and orthocomplementation in quantum logic, International Journal of Theoretical Physics 19 (1980), 369–378. [Garola, 1985] C. Garola, Embedding of posets into lattices in quantum logic, International Journal of Theoretical Physics 24 (1985), 423–433. [Gibbins, 1985] P. F. Gibbins, A userfriendly quantum logic, LogiqueetAnalyse.NouvelleSerie 28 (1985), 353–362. [Gibbins, 1987] P. F. Gibbins, Particles and paradoxes — the limits of quantum logic, Cambridge University Press, Cambridge, 1987. [Girard, 1987] J. Y. Girard, Linear logic, Theoretical Computer Science 50 (1987), 1–102. [Girard, 1996] J.Y. Girard, Proofnets: The parallel syntax for proof theory, Logic and Algebra (A. Ursini and P. Aglian´ o, eds.), Dekker, New York, 1996, pp. 97124. [Girard, 1999] J.Y. Girard, Coherent Banach spaces: A continuous denotational semantics, Theoretical Computer Science 227 (1999), 275297. [Girard, 2004a] J.Y. Girard, Between logic and quantic: A tract, Linear Logic in Computer Science (P. Ruet, J. Ehrhard, J.Y. Girard and P. Scott, eds.), Cambridge University Press, Cambridge, 2004, pp. 346390. [Girard, 2004b] J.Y. Girard, Logique a ` Rome, electronic notes, Ch. 17, 2004. [Giuntini, 1990] R. Giuntini, BrouwerZadeh logic and the operational approach to quantum mechanics, Foundations of Physics 20 (1990), 701–714. [Giuntini, 1991a] R. Giuntini, Quantum logic and hidden variables, Grundlagen der exakten Naturwissenschaften, no. 8, Bibliographisches Institut, Mannheim, 1991.
Quantum Logic and Nonclassical Logics
223
[Giuntini, 1991b] R. Giuntini, A semantical investigation on BrouwerZadeh logic, Journal of Philosophical Logic 20 (1991), 411–433. [Giuntini, 1992] R. Giuntini, BrouwerZadeh logic, decidability and bimodal systems, Studia Logica 51 (1992), 97–112. [Giuntini, 1993] R. Giuntini, Threevalued BrouwerZadeh logic, International Journal of Theoretical Physics 32 (1993), 1875–1887. [Giuntini, 1995a] R. Giuntini, Quasilinear QMV algebras, International Journal of Theoretical Physics 34 (1995), 1397–1407. [Giuntini, 1995b] R. Giuntini, Unsharp orthoalgebras and quantum MV algebras, The Foundations of Quantum Mechanics — Historical Analysis and Open Questions (C. Garola and A. Rossi, eds.), Kluwer, Dordrecht, 1995, pp. 325–337. [Giuntini, 1996] R. Giuntini, Quantum MV algebras, Studia Logica 56 (1996), 393–417. [Giuntini, 2000] R. Giuntini, An independent axiomatization of QMV algebras, The Foundations of Quantum Mechanics (C. Garola and A. Rossi, eds.), World Scientiﬁc, Singapore, 2000. [Giuntini, 2002] R. Giuntini, Weakly linear QMV algebras, Algebra Universalis, 53, 1, 2005, 4572. [Giuntini and Greuling, 1989] R. Giuntini and H. Greuling, Toward an unsharp language for unsharp properties, Foundations of Physics 19 (1989), 931–945. [Giuntini and Pulmannov´ a, 2000] R. Giuntini and S. Pulmannov´ a, Ideals and congruences in QMV algebras, Communications in Algebra 28 (2000), 1567–1592. [Gleason, 1957] A. M. Gleason, Measures on the closed subspaces of a Hilbert space, Journal of Mathematics and Mechanics 6 (1957), 885–893. [Goldblatt, 1974] R. Goldblatt, Semantical analysis of orthologic, Journal of Philosophical Logic 3 (1974), 19–35. [Goldblatt, 1984] R. Goldblatt, Orthomodularity is not elementary, The Journal of Symbolic Logic 49 (1984), 401–404. [Greechie, 1968] R. J. Greechie, On the structure of orthomodular lattices satisfying the chain condition, Journal of Combinatorial Theory 4 (1968), 210–218. [Greechie, 1974] R. J. Greechie, Some results from the combinatorial approach to quantum logic, Synthese 29 (1974), 113–127. [Greechie, 1981] R. J. Greechie, A nonstandard quantum logic with a strong set of states, Current Issues in Quantum Logic (E. Beltrametti and B. van Fraassen, eds.), Ettore Majorana International Science Series, vol. 8, Plenum, New York, 1981, pp. 375–380. [Greechie and Foulis, 1995] R. J. Greechie and D. J. Foulis, The transition to eﬀect algebras, International Journal of Theoretical Physics 34 (1995), 1369–1382. [Gudder, 1979] S. P. Gudder, A survey of axiomatic quantum mechanics, The LogicoAlgebraic Approach to Quantum Mechanics (C. A. Hooker, ed.), vol. II, Reidel, Dordrecht, 1979, pp. 323–363. [Gudder, 1995] S. P. Gudder, Total extensions of eﬀect algebras, Foundations of Physics Letters 8 (1995), 243–252. [Gudder, 1998] S. P. Gudder, Sharply dominating eﬀect algebras, Tatra Mountains Mathematical Publications 15 (1998), 23–30. [Gudder, 2003] S. Gudder, Quantum computational logic, International Journal of Theoretical Physics 42 (2003), 39–47. [Gudder and Greechie, 1996] S. P. Gudder and R. J. Greechie, Eﬀect algebra counterexamples, Mathematica Slovaca 46 (1996), 317–325. [H´ ajek, 1998] P. H´ ajek, Metamathematics of fuzzy logic, Trends in Logic, vol. 4, Kluwer Academic Publishers, Dordrecht, 1998. [Halmos, 1951] P. R. Halmos, Introduction to Hilbert space and the theory of spectral multiplicity, Chelsea, New York, 1951. [Hardegree, 1975] G. M. Hardegree, Stalnaker conditionals and quantum logic, Journal of Philsophical Logic 4 (1975), 399–421. [Hardegree, 1976] G. M. Hardegree, The conditional in quantum logic, Logic and Probability in Quantum Mechanics (P. Suppes, ed.), Reidel, Dordrecht, 1976, pp. 55–72. [Hardegree, 1981] G. M. Hardegree, An axiom system for orthomodular quantum logic, Studia Logica 40 (1981), 1–12. [Holland, 1995] S. S. Holland, Orthomodularity in inﬁnite dimensions: a theorem of M. Sol` er, Bulletin of the American Mathematical Society 32 (1995), 205–232.
224
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
[Howard, 1980] W.A. Howard, The formulaeastypes notion of construction, To H.B. Curry: Essays on Combinatory Logic, Lambda Calculus, and Formalism (J.R. Hindley and J. Seldin, eds.), Academic Press, New York, 1980, pp. 479490. [Hughes, 1985] R. I. G. Hughes, Semantic alternatives in partial Boolean quantum logic, Journal of Philosophical Logic 14 (1985), 411–446. [Hughes, 1987] R. I. G. Hughes, The structure and interpretation of quantum mechanics, Cambridge University Press, Cambridge, 1987. [Hughes and van Glabbeek, 2003] D.J.D. Hughes, R.J. van Glabbeek, Proof nets for unit free multiplicative additive linear logic (extended abstract), Proceedings of the XVIII Annual Symposium on Logic and Computer Science, IEEE, 2003, pp. 110. [Jammer, 1974] M. Jammer, The philosophy of quantum mechanics, WileyInterscience, New York, 1974. [Jauch, 1968] J. M. Jauch, Foundations of quantum mechanics, AddisonWesley, London, 1968. [Kalmbach, 1983] G. Kalmbach, Orthomodular lattices, Academic Press, New York, 1983. [Keller, 1980] H. A. Keller, Ein nichtklassischer hilbertscher Raum, Mathematische Zeitschrift 172 (1980), 41–49. [Kochen and Specker, 1965a] S. Kochen and E. P. Specker, The calculus of partial propositional functions, Proceedings of the 1964 International Congress for Logic, Methodology and Philosophy of Science (Y. BarHillel, ed.), NorthHolland, Amsterdam, 1965, pp. 45–57. [Kochen and Specker, 1965b] S. Kochen and E. P. Specker, Logical structures arising in quantum theory, The Theory of Models (J. Addison, L. Henkin, and A. Tarski, eds.), NorthHolland, Amsterdam, 1965, pp. 177–189. [Kochen and Specker, 1967] S. Kochen and E. P. Specker, The problem of hidden variables in quantum mechanics, Journal of Mathematics and Mechanics 17 (1967), 59–87. [Kˆ opka and Chovanec, 1994] F. Kˆ opka and F. Chovanec, Dposets, Mathematica Slovaca 44 (1994), 21–34. [Kraus, 1983] K. Kraus, States, eﬀects and operations, Lecture Notes in Physics, vol. 190, Springer, Berlin, 1983. [Lincoln et al., 1992] P.D. Lincoln, J.C. Mitchell, A. Scedrov, N. Shankar, Decision problems for propositional linear logic, Annals of Pure and Applied Logic 56 (1992), 239311. [Ludwig, 1983] G. Ludwig, Foundations of quantum mechanics, vol. 1, Springer, Berlin, 1983. [L ukasiewicz, 1936] J. L ukasiewicz, Logistic and philosophy, Selected Work (L. Borkowski, ed.), NorthHolland, Amsterdam, 1970, pp. 218–235. [L ukasiewicz, 1946] J. L ukasiewicz, On determinism, Selected Work (L. Borkowski, ed.), NorthHolland, Amsterdam, 1970, pp. 110–128. [L ukasiewicz, 1970] J. L ukasiewicz, On threevalued logic, Selected Work (L. Borkowski, ed.), NorthHolland, Amsterdam, 1970. [Mackey, 1957] G. Mackey, The Mathematical Foundations of Quantum Mechanics, Benjamin, New York, 1957. [Mangani, 1973] P. Mangani, Su certe algebre connesse con logiche a pi` u valori, Bollettino Unione Matematica Italiana 8 (1973), 68–78. [Minari, 1987] P. Minari, On the algebraic and kripkean logical consequence relation for orthomodular quantum logic, Reports on Mathematical Logic 21 (1987), 47–54. [Mittelstaedt, 1972] P. Mittelstaedt, On the interpretation of the lattice of subspaces of Hilbert space as a propositional calculus, Zeitschrift f¨ ur Naturforschung, Dordrecht, textbf27a (1972), 1358–1362. [Mittelstaedt, 1978] P. Mittelstaedt, Quantum logic, Reidel, Dordrecht, 1978. [Mittelstaedt, 1985] P. Mittelstaedt (ed.), Recent developments in quantum logic, Grundlagen der exakten Naturwissenschaften, no. 6, Bibliographisches Institut, Mannheim, 1985. [Mundici, 1992] D. Mundici, The logic of Ulam’s game with lies, Knowledge, Belief and Strategic Interaction (C. Bicchieri and M. L. Dalla Chiara, eds.), Cambridge University Press, Cambridge, 1992. [Navara, 1999] M. Navara, Two descriptions of state spaces of orthomodular structures, International Journal of Theoretical Physics 38 (1999), 3163–3178. [Neubrunn and Rieˇcan, 1997] T. Neubrunn and B. Rieˇcan, Integral, measure and ordering, Kluwer Academic Publishers, Dordrecht, 1997. [Nielsen and Chuang, 2000] M.A. Nielsen and I.L. Chuang, Quantum computation and quantum information, Cambridge University Press, Cambridge, 2000.
Quantum Logic and Nonclassical Logics
225
[Nishimura, 1980] H. Nishimura, Sequential method in quantum logic, Journal of Symbolic Logic 45 (1980), 339–352. [Nishimura, 1994] H. Nishimura, Proof theory for minimal quantum logic I and II, International Journal of Theoretical Physics 33 (1994), 102–113, 1427–1443. [Paoli, 2002] F. Paoli, Substructural logics: A primer, Trends in Logic, vol. 13, Kluwer Academic Publishers, 2002. [Peres, 1995] A. Peres, Quantum theory: Concepts and methods, Kluwer Academic Publishers, Dordrecht, 1995. [Piron, 1976] C. Piron, Foundations of quantum physics, W. A. Benjamin, Reading, 1976. [Pitowsky, 1989] I. Pitowsky, Quantum probability — quantum logic, Lectures Notes in Physics, no. 321, Springer, Berlin, 1989. [Pratt, 1993] V. Pratt, Linear logic for generalized quantum mechanics, Workshop on Physics and Computation (PhysComp’92) (Dallas), IEEE, 1993, pp. 166–180. [Pt´ ak and Pulmannov´ a, 1991] P. Pt´ ak and S. Pulmannov´ a, Orthomodular structures as quantum logics, Fundamental Theories of Physics, no. 44, Kluwer, Dordrecht, 1991. [Pulmannov´ a, 1995] S. Pulmannov´ a, Representation of Dposets, International Journal of Theoretical Physics 34 (1995), 1689–1696. [Putnam, 1969] H. Putnam, Is logic empirical?, Boston Studies in the Philosophy of Science (R. S. Cohen and M. W. Wartofsky, eds.), vol. 5, Reidel, Dordrecht, 1969, pp. 216–241. [Pykacz, 2000] J. Pykacz, L ukasiewicz operations in fuzzy set theories and manyvalued representations of quantum logics, Foundations of Physics, 30 (2000), 1503–1524. [R´ edei, 2001] M. R´ edei, Von Neumann’s concept of quantum logic and quantum probability, in [R´ edei and Stoeltzner, 2001]. [R´ edei, 1999] M. R´ edei, “Unsolved problems in mathematics” J. von Neumann’s address to the International Congress of Mathematicians Amsterdam, September 29, 1954, The Mathematical Intelligencer 21 (1999), 712. [R´ edei, 1998] M. R´ edei, Quantum logic in algebraic approach, Kluwer Academic Publishers, Dordrecht, Holland, 1998. [R´ edei, 1996] M. R´ edei, Why John von Neumann did not like the Hilbert space formalism of quantum mechanics (and what he liked instead), Studies in the History and Philosophy of Modern Physics 27 (1996), 493510. [R´ edei, forthcoming] M. R´ edei: The birth of quantum logic, History and Philosophy of Logic, forthcoming. [R´ edei and Stoeltzner, 2001] M. R´ edei and M. St¨ oltzner, John von Neumann and the Foundations of Quantum Physics, M. St¨ oltzner, M. R´ edei (eds.), Kluwer Academic Publishers, Dordrecht, Boston, London 2001. [Redhead, 1987] M. Redhead, Incompleteness, nonlocality and realism — a prolegomenon to the philosophy of quantum mechanics, Clarendon Press, Oxford, 1987. [Reed and Simon, 1972] M. Reed and B. Simon, Methods of modern mathematical physics, vol. I, Academic Press, New York, 1972. [Restall, 2002] G. Restall, Paraconsistency everywhere, Notre Dame Journal of Formal Logic 43 (2002), 147156. [Restall and Paoli, 2005] G. Restall, F. Paoli, The geometry of nondistributive logics, Journal of Symbolic Logic 70 (2005), 11081126. [Rosenthal, 1990] K. I. Rosenthal, Quantales and their applications, Longman, New York, 1990. [Sambin et al., 2000] G. Sambin, G. Battilotti, and C. Faggian, Basic logic: reﬂection, symmetry, visibility, The Journal of Symbolic Logic 65 (2000), 979–1013. [Schroeck, 1996] F. E. Schroeck, Quantum mechanics on phase space, Fundamental Theories of Physics, vol. 74, Kluwer Academic Publishers, Dordrecht, 1996. [Sch¨ utte Monting, 1981] J. Sch¨ ulte Monting, Cut elimination and word problem for varieties of lattices, Algebra Universalis 12 (1981), 290321. [Selinger, 2004] P. Selinger, Towards a quantum programming language, Mathematical Structures in Computer Science 14 (2004), 527586. [Smets, 2001] S. Smets, The logic of physical properties in static and dynamic perspective, PhD Thesis, Vrije Universiteit Brussels, 2001. [Sol` er, 1995] M. P. Sol` er, Characterization of Hilbert spaces by orthomodular spaces, Communications in Algebra 23 (1995), 219–243.
226
Gianpiero Cattaneo, Maria Luisa Dalla Chiara, Roberto Giuntini and Francesco Paoli
[Stalnaker, 1981] R. Stalnaker, A theory of conditionals, Ifs. Conditionals, Belief, Decision, Chance, and Time (W. Harper, G. Pearce, and R. Stalnaker, eds.), Reidel, Dordrecht, 1981, pp. 41–55. [Svozil, 1998] K. Svozil, Quantum logic, Springer, Singapore, 1998. [Takeuti, 1981] G. Takeuti, Quantum set theory, Current Issues in Quantum Logic (E. G. Beltrametti and B. C. van Fraassen, eds.), Ettore Majorana International Science Series, vol. 8, Plenum, New York, 1981, pp. 303–322. [Tamura, 1988] S. Tamura, A Gentzen formulation without the cut rule for ortholattices, Kobe Journal of Mathematics 5 (1988), 133–150. [Tortora De Falco, 2003] L. Tortora De Falco, The additive multiboxes, Annals of Pure and Applied Logic 120 (2003), 489524. [van Fraassen, 1991] B. van Fraassen, Quantum mechanics. An empiricist view, Clarendon Press, Oxford, 1991. [Varadarajan, 1985] V. S. Varadarajan, Geometry of quantum theory, 2 ed., Springer, Berlin, 1985. [van Tonder, 2004] A. van Tonder, A lambda calculus for quantum computation, SIAM Journal of Computing 33 (2004), 11091135. [von Neumann, 1927a] J. von Neumann, Mathematische Begr¨ undung der Quantenmechanik, G¨ ottinger Nachrichten (1927), 157, in [von Neumann, 1962], pp. 151207. [von Neumann, 1927b] J. von Neumann, Wahrscheinlichkeitstheoretischer Aufbau der Quantenmechanik, G¨ ottinger Nachrichten (1927), 245272, in [von Neumann, 1962], pp. 208235. [von Neumann, 1927c] J. von Neumann, Thermodynamik quantenmechanischer Gesamtheiten, G¨ ottinger Nachrichten (1927), 245272, in [von Neumann, 1962],pp. 236254. [von Neumann, 1932] J. von Neumann, Mathematische Grundlagen der Quantenmechanik, Dover Publications, New York, 1943 (ﬁrst American Edition; ﬁrst edition: Springer Verlag, Heidelberg, 1932). [von Neumann, 1962] J. von Neumann, Collected Works Vol. I. Logic, Theory of Sets and Quantum Mechanics , A.H. Taub (ed.), Pergamon Press, 1962. [von Neumann, 1961] J. von Neumann, Collected Works Vol. III. Rings of Operators, A.H. Taub (ed.), Pergamon Press, 1961. [von Neumann, 1961b] J. von Neumann, Collected Works Vol. IV. Continuous Geometry and Other Topics, A.H. Taub (ed.), Pergamon Press, 1961. [von Neumann, 1981] J. von Neumann,Continuous Geometries with Transition Probability, Memoirs of the American Mathematical Society 34 No. 252 (1981) 1210. [Wright, 1990] R. Wright, Generalized urn models, Foundations of Physics 20 (1990), 881–903. [Zadeh, 1965] L. Zadeh, Fuzzy sets, Information and Control 8 (1965), 338–353. [Zierler, 1961] N. Zierler, Axioms for nonrelativistic quantum mechanics, Paciﬁc Journal of Mathematics 11 (1961), 1151–1169.
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
227
GENTZEN METHODS IN QUANTUM LOGIC Hirokazu Nishimura
1
INTRODUCTION
Since Birkhoﬀ and von Neumann [Birkhoﬀ and von Neumann, 1936] a new area of logical investigation has grown up under the name of quantum logic. During its early days emphasis was put exclusively on its algebraic aspects. A new impetus came from Dishkant and Goldblatt’s ([Dishkant, 1972], [Dishkant, 1977] and [Goldblatt, 1974]) remarkable discovery on the relationship between ortholattices and the Brouwerian modal logic B in the 1970’s, which is comparable to Mckinsey and Tarski’s [McKinsey and Tarski, 1948] translation of intuitionistic logic into the modal logic S4. As the semantics of possible worlds has been one of the main tools in modal logic since Kripke [Kripke, 1963], the discovery naturally admitted to a Kripkian relational semantics of minimal quantum logic. Since it was then well known that there is a close relationship between Gentzenstyle formulations of modal logics and their Kripkian relational semantics (cf. [Nishimura, 1983] and [Sato, 1977]), Nishimura [Nishimura, 1980] was driven on closing days of the 1970’s to a Gentzenstyle formulation of minimal quantum logic, which regrettably failed to enjoy the cutelimination theorem. A more natural Gentzenstyle formulation of minimal quantum logic with closer inspection on its relationship to the relational semantics was given by Cutland and Gibbins [Cutland and Gibbins, 1982], but it still failed to acquiesce in the cutelimination property. The ﬁrst cutfree Gentzenstyle formulation of minimal quantum logic was presented by Tamura [Tamura, 1988], though it suﬀered from unnecessary clumsiness, which made his system appear more esoteric than it really was. A ﬁnal step was taken again by Nishimura ([Nishimura, 1994a] and [Nishimura, 1994b]), which was followed by Takano’s [Takano, 1995] signiﬁcant remark that the inference rule from a sequent to its contraposition is redundant. The ﬁrst stage of the story has thus ended, and the principal objective in this paper is to present its fruits to a novice thoroughly. In Section 2 we will present our cutfree Gentzenstyle sequential system GMQL. We will remark, following [Cutland and Gibbins, 1982], that admitting unrestricted (cut) as an inference rule would force our system GMQL to degenerated into classical logic. In Section 3 we will show, following [Cutland and Gibbins, 1982], that the inference rule from a sequent to its contraposition is admissible in GMQL. In Section 4 we will establish the fundamental fact that the negation is involutive with respect to its prooftheoretical behaviors. In Section 5 the desired cutelimination theorem is to be demonstrated. The ﬁnal section is devoted to the
228
Hirokazu Nishimura
completeness theorem with respect to the relational semantics of Dishkant and Goldblatt. The reader may wonder what is to be the second stage of the story. We will give two suggestions. The modal logics S4 and B stand to the modal logic S5 in opposite directions, but they are complementary against S5, as may be illustrated in the following ﬁgure: S5 ( ) S4 B The complementarity of the modal logics S4 and B corresponds to the following complementarity of intuitionistic logic and minimal quantum logic against classical logic, as may be illustrated in the following ﬁgure: classical logic (
)
intuitionistic logic
minimal quantum logic
Logics between classical logic and intuitionistic logic have been studied vigorously under the name of intermediate logics. It would be interesting to investigate logics between classical logic and minimal quantum logic, among which you can ﬁnd quantum logic. The other intriguing topic for future study is a semantical proof of the cutelimination theorem of GMQL. In other words, it would be interesting to give a proof of the completeness theorem with respect to the Kripkian relational semantics without any recourse to the cutelimination theorem, which would surely open a new area of research. 2 MINIMAL QUANTUM LOGIC IN GENTZEN STYLE The sequential system GMQL that we have enunciated for minimal quantum logic in our [Nishimura, 1994a] and that has then been elaborated by Takano in [Takano, 1995] consists of the following inference rules: Γ→Δ π, Γ → Δ, Σ
(extension)
α, Γ → Δ α ∧ β, Γ → Δ
β, Γ → Δ α ∧ β, Γ → Δ
(∧ →)
Γ → Δ, α Γ → Δ, α ∨ β
Γ → Δ, β Γ → Δ, α ∨ β
(→ ∨)
Gentzen Methods in Quantum Logic
α→Δ β→Δ α∨β →Δ
(∨ →)
Γ→α Γ→β Γ→α∧β
(→ ∧)
Γ→Δ Δ , Γ →
( →)
Γ→Δ → Δ, Γ
(→ )
α, Γ → Δ α , Γ → Δ
( →)
Γ → Δ, α Γ → Δ, α
(→ )
229
α , Γ → Δ (α ∨ β) , Γ → Δ
β, Γ → Δ (α ∨ β) , Γ → Δ
(∨ →)
Γ → Δ, α Γ → Δ, (α ∧ β)
Γ → Δ, β Γ → Δ, (α ∧ β)
(→ ∧ )
α → Δ β → Δ (α ∧ β) → Δ,
(∧ →)
Γ → α Γ → β Γ → (α ∨ β)
(→ ∨ )
Γ → α Γ → β α ∨ β, Γ →
(∨ → )
α → Δ β → Δ → Δ, α ∧ β
( → ∧)
Now some notational and terminological comments are in order. In this paper we adopt (negation), ∧ (conjunction), and ∨ (disjunction) as primitive logical symbols. Propositional variables are denoted by p, q, ... , while wﬀs (wellformed formulas), also called formulas, are denoted by α, β, .... The grade of a wﬀ α, denoted by G(α), is deﬁned inductively as follows: 1. G(p) = 0 for any propositional variable p. 2. G(α ) = G(α) + 1. 3. G(α ∧ β) = G(α ∨ β) = G(α) + G(β) + 2.
230
Hirokazu Nishimura
Finite (possibly empty) sets of wﬀs are denoted by Γ, Δ, Π, .... Given a ﬁnite set Γ of wﬀs, Γ denotes the set {α α ∈ Γ}. A sequent Γ → Δ means the ordered pair (Γ, Δ) of ﬁnite sets Γ and Δ of wﬀs, while the sets Γ and Δ are called the antecedent and the succedent of the sequent Γ → Δ, respectively. Such selfexplanatory notations as Π, Γ → Δ, Σ for Π ∪ Γ → Δ ∪ Σ are used freely. A sequent of the form α → α is called an axiom sequent. Given a sequent Γ → Δ, the sequent Δ → Γ is called the contraposition of Γ → Δ. The notion of a proof P of a sequent Γ → Δ with length n is deﬁned inductively as follows: 1. Any axiom sequent α → α is a proof of itself with length 0. 2. If P is a proof of a sequent Γ → Δ with length n and Γ→Δ Π→Σ is an instance of an inference rule of GMQL, then P Π→Σ is a proof of the sequent Π → Σ with length n + 1. 3. If Pi is a proof of a sequent Γi → Δi with length ni (i = 1, 2) and Γ1 → Δ1 Γ2 → Δ2 Π→Σ is an instance of an inference rule of GMQL, then P1 P 2 Π→Σ is a proof of the sequent Π → Σ with length max{n1 , n2 } + 1. The length of a proof P is denoted by l(P ). A sequent Γ → Δ is said to be provable if it has a proof. Otherwise it is called consistent. Although our cutfree sequential system GMQL does not satisfy the socalled subformula property in its strict sense, it gives a decision procedure for the word problem of free ortholattices once the completeness theorem is established, for which it suﬃces to note that G(α ) < G((α ∧ β) ) and G(β ) < G((α ∧ β) ) for the rule (∧ →) by way of example. For algebraic and semantical decision procedures, the reader is referred to [Bruns, 1976], [Goldblatt, 1974] and [Goldblatt, 1975]. Fortunately, minimal quantum logic enjoys these three kinds of decision procedures. However, algebraic and semantical approaches to the decision problem of quantum logic have not succeeded so far. This is why we should try the third one.
Gentzen Methods in Quantum Logic
231
Generally speaking, (cut) is the inference rule of the following form: Γ1 → Δ1 , α α, Γ2 → Δ2 Γ1 , Γ2 → Δ1 , Δ2
(cut)
However, following [Cutland and Gibbins, 1982], we should remark that the inference rule (cut) in such an unrestricted form forces our system GMQL to degenerate into classical logic. In other words, we have PROPOSITION 1. If we add the inference rule (cut) to the system GMQL, then we obtain classical logic. Schematically, we have GMQL+(cut) = classical logic Proof. It suﬃces to show that the following three rules are admissible in GMQL+(cut): Γ → Δ, α α , Γ → Δ
(* → )c
α, Γ → Δ Γ → Δ, α
( → )c
Γ → Δ, α Γ → Δ, β Γ → Δ, α ∧ β Since we have
α→α Γ → Δ, α α, α → α , Γ → Δ
( → ∧)c
( →) (cut)
the inference rule (* →)c is admissible in GMQL+(cut). Similarly, since we have α→α → α , α
(→ )
α, Γ → Δ Γ → Δ, α
(cut)
the inference rule (→ )c is admissible in GMQL+(cut). Now we deal with the last inference rule (→ ∧)c . The sequents Γ, Δ → α and Γ, Δ → β follow from the sequents Γ → Δ, α and Γ → Δ, β respectively by a ﬁnite number of applications of the inference rule (* →)c . Now we have Γ, Δ → α Γ, Δ → β Γ, Δ → α ∧ β
(→ ∧)
The sequent Γ → Δ , α ∧ β follows from the sequent Γ, Δ → α ∧ β by a ﬁnite number of applications of the inference rule (→ )c . Since we have γ→γ (→ ) → γ, γ γ → γ
( →)c
232
Hirokazu Nishimura
we are sure that the sequent γ → γ is provable in GMQL+(cut) for any γ ∈ Δ. Therefore the desired sequent Γ → Δ, α∧β follows from the sequent Γ → Δ , α∧β by a ﬁnite number of applications of the inference rule (cut). This is the reason why Cutland and Gibbins [Cutland and Gibbins, 1982] should have proposed (cut) in the following restricted form: Γ → Δ1 , α α → Δ2 Γ → Δ1 , Δ2
(cut1)
Γ1 → α α, Γ2 → Δ Γ 1 , Γ2 → Δ
(cut2)
The wﬀ α in (cut1) and (cut2) is called the cut formula. Both (cut1) and (cut2) are called (cut)q as a whole. Roughly speaking, if we deprive our system GMQL of the inference rules (∨ →), (→ ∧ ), (∧ →), (→ ∨ ), (∨ → ), and ( → ∧) and we agree to admit the inference rules (cut1) and (cut2), then we obtain the system of Cutland and Gibbins [Cutland and Gibbins, 1982]. We will prove in Section 5 that the inference rules (cut1) and (cut2) are admissible in GMQL. Tamura [Tamura, 1988] gave a cutfree system by exploiting the legacy of Cutland and Gibbins [Cutland and Gibbins, 1982] but incorporating their inference rules surely except (cut1) and (cut2) into his system in an unnecessarily restricted manner. This unreasonable restriction forced him in the proof of the cutelimination theorem to combine wﬀs in the antecedent of a sequent by conjunction and wﬀs in its succedent by disjunction, and then to dissolve such unnatural combinations. Such a proof is not compatible with Gentzen’s [Gentzen, 1935] original philosophy and aesthetics, and is to be avoided if possible. Furthermore, the conceptual signiﬁcance of Lemma 4 in Tamura’s [Tamura, 1988] paper remained vague at best there. This is distilled into the duality theorem in Section 4, which is followed by the socalled cutelimination theorem in Section 5. Our original GMQL, proposed in [Nishimura, 1994a], contains the following inference rule besides the above ones: Γ→Δ Δ → Γ
( → )
It was pointed out by Takano [Takano, 1995] that the rule is redundant, which is the topic of the succeeding section. 3
THE CONTRAPOSITION THEOREM
The principal objective in this section is to show the following theorem on the lines of Takano [Takano, 1995]. THEOREM 2. The following inference rule is admissible in GMQL. Γ→Δ Δ → Γ
( → )
Gentzen Methods in Quantum Logic
233
To establish the above theorem, we introduce an auxiliary formal system to be denoted by GMQL# and to be obtained from GMQL by admitting not only sequents α → α but also sequents α, α → and → α, α as axiom sequents while deleting the inference rules ( →) and (→ ) and adding the following two inference rules: Γ→α Γ→β (∧ →)# (α ∧ β) , Γ → α→Δ β→Δ (→ ∨ )# → Δ, (α ∨ β) We need three lemmas so as to establish the equivalence of GMQL and GMQL# . LEMMA 3. 1. If a sequent α , Γ → Δ is provable in GMQL# , then so is α, Γ → Δ. 2. If a sequent Γ → Δ, α is provable in GMQL# , then so is Γ → Δ, α. Proof. We prove only the ﬁrst statement by induction on the length l(P ) of a proof P of the sequent α , Γ → Δ, while leaving a similar treatment of the second statement to the reader. Our treatment is divided into several cases, some of which are again divided into several subcases. 1. The case that the sequent α , Γ → Δ is an axiom sequent: We divide this case into three subcases. (a) The subcase that the sequent α , Γ → Δ is α → α : Since we have α→α (→ ) α → α the sequent α → α is also provable. (b) The subcase that the sequent α , Γ → Δ is α , α →: The sequent α, α → is an axiom, and so is provable. (c) The subcase that the sequent α , Γ → Δ is α , α →: Since we have α, α → ( →) α, α → the sequent α, α → is also provable. 2. The case that the last inference is (extension): The last inference has one of the following two forms: Γ1 → Δ1 α , Γ2 , Γ1 → Δ1 , Δ2
(extension)
234
Hirokazu Nishimura
α , Γ1 → Δ1 α , Γ2 , Γ1 → Δ1 , Δ2
(extension)
In the former case, since we have Γ1 → Δ1 α, Γ2 , Γ1 → Δ1 , Δ2
(extension)
the sequent α, Γ2 , Γ1 → Δ1 , Δ2 is also provable. In the latter case, since the sequent α, Γ1 → Δ1 is provable by induction hypothesis and we have α, Γ1 → Δ1 α, Γ2 , Γ1 → Δ1 , Δ2
(extension)
the sequent α, Γ2 , Γ1 → Δ1 , Δ2 is also provable. 3. The case that the last inference is ( →): We divide this case into two subcases according as the principal formula of the last inference is α or not. (a) The subcase that the principal formula of the last inference is α : The last inference has one of the following two forms: α, Γ → Δ α , Γ → Δ α , α, Γ → Δ α , Γ → Δ
( →) ( →)
In the former case the sequent α, Γ → Δ is palpably provable, while in the latter case it should be provable by induction hypothesis. (b) The subcase that the principal formula of the last inference is not α : The last inference has the form α , β, Γ1 → Δ α , β , Γ1 → Δ
( →)
Since the sequent α, β, Γ1 → Δ is provable by induction hypothesis and we have α, β, Γ1 → Δ ( →) α, β , Γ1 → Δ the sequent α, β , Γ1 → Δ is also provable. 4. The case that the last inference is neither (extension) nor ( →):Similar to the subcase (3b). LEMMA 4. The inference rule ( → ) is admissible in GMQL# .
Gentzen Methods in Quantum Logic
235
Proof. We will prove that if a sequent Γ → Δ is provable in GMQL# , then its contraposition Δ → Γ is also provable in GMQL# . The proof is carried out by induction on the length of a proof P of the given sequent Γ → Δ. Our treatment is divided into several cases. 1. The case that the given sequent Γ → Δ is an axiom sequent: The sequent Γ → Δ has one of the following three forms α → α, α , α → and → α, α , whose contrapositions are also axioms α → α , → α , α and α , α →. 2. The case that the last inference in P is (extension), (∧ →), (→ ∧), (∨ →), (→ ∨), ( →) or (→ ): All these cases can be dealt with similarly, so we deal only with the case that the last inference is (→ ∧) as follows: Γ→α Γ→β Γ→α∧β
(→ ∧)
Since the sequents α → Γ and β → Γ are provable by induction hypothesis and we have β → Γ α → Γ (∧ →) (α ∧ β) → Γ the sequent (α ∧ β) → Γ is also provable. 3. The case that the last inference in P is either (∧ →)# or (→ ∨ )# : Here we deal only with the former case, leaving a similar treatment of the latter case to the reader. So we suppose that the last inference in P is Γ1 → α Γ1 → β (α ∧ β) , Γ1 →
(∧ →)#
Since the sequents α → Γ1 and β → Γ1 are provable by induction hypothesis and we have β → Γ1 α → Γ1 ( → ∧) → Γ1 , α ∧ β ( → ) → Γ1 , (α ∧ β) we are sure that the sequent → Γ1 , (α ∧ β) is also provable. 4. The case that the last inference in P is either ( → ∧) or (∨ → ): Here we deal only with the former case, leaving a similar treatment of the latter case to the reader. So we suppose that the last inference in P is β → Δ1 α → Δ1 → Δ1 , α ∧ β
( → ∧)
The sequents Δ1 → α and Δ1 → β are provable by induction hypothesis, which imply by Lemma 3 that the sequents Δ1 → α and Δ1 → β are also provable. Since we have Δ1 → α Δ1 → β (α ∧ β) , Δ1 →
(∧ →)#
236
Hirokazu Nishimura
we are sure that the sequent (α ∧ β) , Δ1 → is also provable. 5. The case that the last inference in P is (∧ →), (→ ∧ ), (∨ →) or (→ ∨ ): Here we deal only with the ﬁrst case, leaving similar treatments of the remaining three cases to the reader. So we suppose that the last inference in P is α → Δ1 β → Δ1 (∧ →) (α ∧ β) → Δ1 The sequents Δ1 → α and Δ1 → β are provable by induction hypothesis, which imply by Lemma 3 that the sequents Δ1 → α and Δ1 → β are also provable. Since we have Δ1 → α Δ1 → β ( → ∧) Δ1 → α ∧ β Δ1 → (α ∧ β)
( → )
we are sure that the sequent Δ1 → (α ∧ β) is also provable. LEMMA 5. 1. If a sequent Γ → Δ is provable in GMQL# , then so is Δ , Γ →. 2. If a sequent Γ → Δ is provable in GMQL# , then so is → Δ, Γ . Proof. The proof is by induction on the length of a proof P of the given sequent Γ → Δ. We deal only with the ﬁrst statement, leaving a similar treatment of the second treatment to the reader. Our treatment is divided into several cases. 1. The case that the given sequent Γ → Δ is an axiom sequent: The sequent Γ → Δ is one of the three forms α → α, α , α → and → α, α . Then the sequent Δ , Γ → is one of the two forms α , α → and α , α →, both of which are axioms. 2. The case that the last inference in P is (extension), (∧ →), (→ ∨), ( →), (→ ) or (∨ →): Here we deal only with the third case, leaving similar treatments of the remaining ﬁve cases to the reader. Thus the last inference of P is of the following form: Γ → Δ1 , α Γ → Δ1 , α ∨ β
(→ ∨)
Since the sequent α , Δ1 , Γ → is provable by induction hypothesis and we have α , Δ1 , Γ → (∨ →) (α ∨ β) , Δ1 , Γ → we are sure that the sequent (α ∨ β) , Δ1 , Γ → is also provable.
Gentzen Methods in Quantum Logic
237
3. The case that the last inference in P is (→ ∧ ): The last inference in P is in the following form: Γ → Δ1 , α (→ ∧ ) Γ → Δ1 , (α ∧ β) The sequent α , Δ1 , Γ → is provable by induction hypothesis, which implies by Lemma 3 that the sequent α, Δ1 , Γ → is also provable. Since we have α, Δ1 , Γ → (∧ →) α ∧ β, Δ1 , Γ → (α ∧ β) , Δ1 , Γ →
( →)
we are sure that the sequent (α ∧ β) , Δ1 , Γ → is also provable. 4. The case that the last inference in P is (→ ∧): The last inference in P is of the following form: Γ→α Γ→β (→ ∧) Γ→α∧β Since we have Γ→α Γ→β (∧ →)# (α ∧ β) , Γ → we are sure that the sequent (α ∧ β) , Γ → is also provable. 5. The case that the last inference in P is (→ ∨ ): The last inference in P is of the following form: Γ → α Γ → β (→ ∨ ) Γ → (α ∨ β) Since we have Γ → α Γ → β (∨ → ) α ∨ β, Γ → ( →) (α ∨ β) , Γ → we are sure that the sequent (α ∨ β) , Γ → is also provable. 6. The case that the last inference in P is either ( → ∧) or (∧ →): The last inference in P is one of the following two forms: α → Δ1 β → Δ1 → Δ1 , α ∧ β
( → ∧)
α → Δ1 β → Δ1 (∧ →) (α ∧ β) → Δ1 In both cases, the sequents Δ1 → α and Δ1 → β are provable by Lemma 4, which implies by dint of Lemma 3 that the sequents Δ1 → α and Δ1 → β are also provable. Since we have Δ1 → α Δ1 → β (α ∧ β) , Δ1 →
(∧ →)#
we are sure that the sequent (α ∨ β) , Γ → is also provable.
238
Hirokazu Nishimura
7. The case that the last inference in P is (∨ →): The last inference in P is of the following form: α→Δ β→Δ (∨ →) α∨β →Δ The sequents Δ → α and Δ → β are provable by Lemma 4. Since we have Δ → α Δ → β (∨ → ) Δ , α ∨ β → we are sure that the sequent Δ , α ∨ β → is also provable. 8. The case that the last inference in P is (→ ∨ )# : The last inference in P is of the following form: α→Δ β→Δ (→ ∨ )# → Δ, (α ∨ β) The sequents Δ → α and Δ → β are provable by Lemma 4. Since we have Δ → α Δ → β (∨ → ) Δ , α ∨ β → ( →) Δ , (α ∨ β) → 9. The case that the last inference in P is either (∨ → ) or (∧ →)# : There is nothing to prove, for the succedent Δ of the given sequent Γ → Δ is empty. Now we are ready to present a proof of the main theorem. THEOREM 6. A sequent Γ → Δ is provable in GMQL# iﬀ it is provable in GMQL. Proof. 1. First we deal with the onlyif part. Since
and
α→α α , α →
( →)
α→α → α, α
(→ )
sequents α, α → and → α, α are provable in GMQL. Since Γ→α Γ→β (→ ∧) Γ→α∧β ( →) (α ∧ β) , Γ →
Gentzen Methods in Quantum Logic
and
239
α→Δ β→Δ (∨ →) α∧β →Δ (→ ) → Δ, (α ∨ β)
the inferene rules (∧ →)# and (→ ∨ )# are admissible in GMQL. Thus the onlyif part has been established 2. The if part follows directly from Lemmas 4 and 5 Our desired theorem at the beginning of this section follows at once from the above theorem. Since the inference rule ( → ) is admissible in GMQL, we will often take it as a basic inference rule of the system GMQL. 4 THE DUALITY THEOREM Two wﬀs α and β are said to be provably equivalent, in notation α + β, if for any ﬁnite sets Γ and Δ of wﬀs we have that 1. the sequent α, Γ → Δ is provable iﬀ the sequent β, Γ → Δ is provable; and 2. the sequent Γ → Δ, α is provable iﬀ the sequent Γ → Δ, β is provable. It is easy to see that this is indeed an equivalence relation among wﬀs. We will show that it is even a congruence relation. THEOREM 7. (The fundamental theorem of provability equivalence). If α1 + β1 and α2 + β2 , then α1 ∼ β1 , α1 ∧ α2 + β1 ∧ β2 , and α1 ∨ α2 + β1 ∨ β2 . Proof. If γ, δ1 , ..., δn are wﬀs and p1 , ..., pn are distinct propositional variables, we write γ[δ1 /p1 , ..., δn /pn ] for the wﬀ obtained from γ by replacing every occurrence of pi by δi (1 ≤ i ≤ n). Whenever we use this notation, it will always be assumed that the propositional variables at issue are distinct. The theorem follows readily from the following two statements: 1. If δ1 + σ1 , ..., δn + σn and a sequent γ[δ1 /p1 , ..., δn /pn ], Γ → Δ has a proof P with l(P ) ≤ m, then the sequent γ[σ1 /p1 , ..., σn /pn ], Γ → Δ is also provable. 2. If δ1 + σ1 , ..., δn + σn and a sequent Γ → Δ, γ[δ1 /p1 , ..., δn /pn ] has a proof P with l(P ) ≤ m, then the sequent Γ → Δ, γ[σ1 /p1 , ..., σn /pn ] is also provable. These two statements are proved simultaneously by double induction principally on G(γ) and secondly on m. The proof is divided into cases according to which inference rule is used as the last inference in P . The details are safely left to the reader.
240
Hirokazu Nishimura
THEOREM 8. (The ﬁrst duality theorem). If α + β, then α + β . Proof. It suﬃces to show the following claim: CLAIM 9. 1. If a sequent α, Γ → Δ is provable, then the sequent β , Γ → Δ is also provable. 2. If a sequent Γ → Δ, α is provable, then the sequent Γ → Δ, β is also provable. 3. If a sequent α , Γ → Δ is provable, then the sequent β, Γ → Δ is also provable. 4. If a sequent Γ → Δ, α is provable, then the sequent Γ → Δ, β is also provable. It is easy to see that the ﬁrst and second statements of the above claim follow at once from a simple application of the inference rules ( →) and (→ ), respectively, while 3 and 4 of the above claim follow at once from the following, ostensibly more general statement. CLAIM 10. If α1 + β1 , ..., αn + βn , αn+1 + βn+1 , ..., αn+m + βn+m and a sequent , ..., αn+m has a proof P with l(P ) ≤ k then the sequent α1 , ..., αn , Γ → Δ, αn+1 β1 , ..., βn , Γ → Δ, βn+1 , ..., βn+m is also provable. We will prove Claim 10 by induction on k. The proof is divided into cases according to which inference rule is used in the last step of P . To make the notation simpler, we proceed as if n = 1 and m = 0, leaving safely easy but due modiﬁcations to the reader. In dealing with the rules (∧ →), (→ ∨), (∨ →) and (→ ∧ ), each of which consists of two forms, we treat only one of them. 1. The case that the sequent α1 , Γ → Δ is an axiom sequent: It must be that α1 → α1 . Since β1 → β1 is an axiom sequent and α1 + β1 by assumption, the sequent β1 → α1 is provable, which implies that the sequent β1 → α1 is also provable as follows: β1 → α1 (→ ) β1 → α1 2. The case that the last inference of the proof of the sequent α1 , Γ → Δ is (extension), (∧ →), (→ ∨), (∧ →), (→ ), (∨ →), (→ ∧ ), (→ ∨ ), or (∨ → ): All the cases can be dealt with similarly, so here we deal only with the case in which the last inference of the proof is (→ ∧) as follows: α1 , Γ → γ α1 , Γ → β α1 , Γ → β ∧ γ
(→ ∧)
Gentzen Methods in Quantum Logic
241
By the induction hypothesis the sequents β1 , Γ → β and β1 , Γ → γ are provable, which gives the desired result as follows: β1 , Γ → γ β1 , Γ → β β1 , Γ → β ∧ γ
(→ ∧)
3. The case that the last inference of the proof of α1 , Γ → Δ is ( →): Then the last inference is one of the following two forms. α1 , Γ → Δ α1 , Γ → Δ
( → ∧)
α1 , β , Γ1 → Δ α1 , β , Γ1 → Δ
( →)
In the former case the sequent β1 , Γ → Δ is provable for α1 + β1 and the sequent α1 , Γ → Δ is provable by assumption. In the latter case the sequent β1 , β, Γ1 → Δ is provable by the induction hypothesis, which implies that the sequent β1 , β , Γ1 → Δ is provable as follows: β1 , β, Γ1 → Δ β1 , β , Γ1 → Δ
( →)
4. The case that the last inference of the proof of the sequent α1 , Γ → Δ is ( →): This case is divided into several subcases according to how the upper sequent of ( →) is obtained. (a) The case that the upper sequent of ( →) is an axiom sequent: In this case the axiom sequent must be α1 → α1 , so the proof that we must consider is as follows: α1 → α1 α1 , α1 →
( →)
Since the sequent α1 → α1 is an axiom sequent and α1 + β1 by assumption, the sequent β1 → α1 is provable, which implies that the desired sequent β1 , α1 → is also provable as follows: β1 , α1 → α1 , β1 →
( →)
(b) The case that the upper sequent of ( →) is obtained as the lower sequent of (extension), (∧ →), ( →), or (∨ →): All these cases can be dealt with similarly, so here we consider only the case of ( →), in which the last two steps of the proof go as follows: β, Γ2 → α1 , Γ1 ( →) β , Γ2 → α1 , Γ1 α1 , Γ1 , β , Γ2 →
( →)
242
Hirokazu Nishimura
The sequent α1 , Γ1 , β, Γ2 → has a shorter proof than the sequent α1 , Γ1 , β , Γ2 →, as follows: β, Γ2 , → α1 , Γ1 ( →) α1 , Γ1 , β, Γ2 → Therefore the sequent β1 , Γ1 , β, Γ2 → is provable by the induction hypothesis, which implies that the desired sequent β1 , Γ1 , β , Γ2 → is also provable as follows: β1 , Γ1 , β, Γ2 → β1 , Γ1 , β , Γ2 →
( →)
(c) The case that the upper sequent of ( →) is obtained as the lower sequent of (→ ): The last two steps of the proof that we must consider can be supposed to be one of the following two forms: Γ2 → β, Γ1 Γ2 → β , Γ1 β , Γ1 , Γ2 →
(→ ) ( →)
Γ2 → α1 , β, Γ1 (→ ) Γ2 → α1 , β , Γ1 α1 , β , Γ1 , Γ2 →
( →)
In the former case α1 is supposed to be β , Since the latter case can be dealt with in a similar manner to the case (2), here we deal with the former case, in which the sequent α1 , Γ1 , Γ2 → is provable with a shorter proof than that of the sequent α1 , Γ1 , Γ2 → as follows: Γ2 , → β, Γ1 β , Γ1 , Γ2 →
( →)
Thus the desired sequent β1 , Γ2 , Γ2 → is also provable by hypothesis, (d) The case that the upper sequent of ( →) is obtained as the lower sequent of (→ ∨): The last two steps of the proof go as follows: Γ2 → α1 , β, Γ1 (→ ∨) Γ2 → α1 , β ∨ γ, Γ1 α1 , (β ∨ γ) , Γ1 , Γ2 →
( →)
The sequent α1 , β , Γ1 , Γ2 → has a shorter proof than the sequent α1 , (β ∨ γ) , Γ1 , Γ2 → as follows: Γ2 , → α1 , β, Γ1 →
α1 , β , Γ1 , Γ2
( →)
Gentzen Methods in Quantum Logic
243
Therefore the sequent β1 , β , Γ1 , Γ2 → is provable by the induction hypothesis, which implies that the desired sequent β1 , (β ∨ γ) , Γ1 , Γ2 → is provable as follows: β1 , β , Γ1 , Γ2 → β1 , (β ∨ γ) , Γ1 , Γ2 →
(∨ →)
(e) The case that the upper sequent of ( →) is obtained as the lower sequent of (→ ∧ ); The last two steps of the proof are of one of the following two forms: Γ2 → α1 , β , Γ1 (→ ∧ ) Γ2 → α1 , (β ∧ γ) , Γ1 α1 , (β ∧ γ) , Γ1 , Γ2 → Γ 2 → β , Γ1 (→ ∧ ) Γ2 → (β ∧ γ) , Γ1 (β ∧ γ) , Γ1 , Γ2 →
( →)
( →)
In the latter case α1 is assumed to be β ∧ γ. Here we deal only with the former case, leaving a similar treatment of the latter case to the reader. The sequent α1 , β , Γ1 , Γ2 → has a shorter proof than the sequent α1 , (β ∧ γ) , Γ1 , Γ2 → as follows: Γ2 , → α1 , β , Γ1 α1 , β , Γ1 , Γ2 →
( →)
This implies by the induction hypothesis that the sequent β1 , β, Γ1 , Γ2 → is also provable. Thus the desired sequent β1 , (β ∧ γ) , Γ1 , Γ2 → is also provable, as follows: β1 , β, Γ1 , Γ2 → (∧ →) β1 , β ∧ γ, Γ1 , Γ2 → β1 , (β ∧ γ) , Γ1 , Γ2 →
( →)
(f) The case that the upper sequent of ( →) is obtained as the lower sequent of (∨ →): The last two steps of the proof that we must consider go as follows: γ → α1 , Γ1 β → α1 , Γ1 (∨ →) β ∨ γ → α1 , Γ1 ( →) α1 , Γ1 , β ∨ γ → The sequents α1 , Γ1 → β and α1 , Γ1 → γ are provable with shorter proofs than that of α1 , Γ1 , β ∨ γ → as follows β → α1 , Γ1 → β
α1 , Γ1
( → )
244
Hirokazu Nishimura
γ → α1 , Γ1 → γ
α1 , Γ1
( → )
Therefore the sequents β1 , Γ1 → β and β1 , Γ1 → γ are provable by the induction hypothesis, which implies that the desired sequent β ∨ γ, β1 , Γ1 → is also provable as follows: β1 , Γ1 → β β1 , Γ1 → γ β ∨ γ, β1 , Γ1 →
(∨ → )
(g) The case that the upper sequent of ( →) is obtained as the lower sequent of (∧ →): The last two steps of the proof that we must consider go as follows: γ → α1 , Γ1 β → α1 , Γ1 (∧ →) (β ∧ γ) → α1 , Γ1 ( →) α1 , Γ1 , (β ∨ γ) → The sequents α1 , Γ1 → β and α1 , Γ1 → γ are provable with shorter proofs than that of the sequent α1 , Γ1 , (β ∨ γ) → as follows: β → α1 , Γ1 → β
( → )
γ → α1 , Γ1 → γ
( → )
α1 , Γ1
α1 , Γ1
Thus the sequents β1 , Γ1 → β and β1 , Γ1 → γ are provable by the induction hypothesis, which implies that the desired sequent (β∧γ) , β1 , Γ1 → is also provable as follows: β1 , Γ1 → β β1 , Γ1 → γ β1 , Γ1 → β ∧ γ, (β ∧ γ) , β1 , Γ1 →
(→ ∧)
( →)
(h) The case that the upper sequent of ( →) is obtained as the lower sequent of (→ ∨ ): The last two steps of the proof that we must consider go as follows: Γ1 → γ Γ1 → β (→ ∨ ) Γ1 → (β ∨ γ) ( →) (β ∨ γ) , Γ1 → Here α1 is supposed to be β ∨ γ. The sequent β ∨ γ, Γ1 → is provable as follows: Γ1 → γ Γ1 → β (∨ → ) β ∨ γ, Γ1 → Since β1 + α1 by assumption, the desired sequent β1 , Γ1 → is also provable.
Gentzen Methods in Quantum Logic
245
(i) The case that the upper sequent of ( →) is obtained as the lower sequent of ( → ∧): The last two steps of the proof that we must consider go as follows: β → α1 , Γ1 γ → α1 , Γ1 ( → ∧) → β ∧ γ, α1 , Γ1 ( →) (β ∧ γ) , α1 , Γ1 → The sequents α1 , Γ1 → β and α1 , Γ1 → γ are provable with shorter proofs than that of the sequent (β ∧ γ) , α1 , Γ1 → as follows: β → α1 , Γ1 → β
( → )
γ → α1 , Γ1 → γ
( → )
α1 , Γ1
α1 , Γ1
Thus the sequents β1 , Γ1 → β and β1 , Γ1 → γ are provable by the induction hypothesis, which implies that the desired sequent (β∧γ) , β1 , Γ1 → is also provable as follows: β1 , Γ1 → β
β1 , Γ1 → γ β1 , Γ1 → β ∧ γ (β ∧ γ) , β1 , Γ1 →
(→ ∧)
( →)
(j) The case that the upper sequent of ( →) is obtained as the lower sequent of ( → ): The last two steps of the proof that we must consider go as follows: α1 , Γ1 → Γ2 ( → ) Γ2 → α1 , Γ1 ( →) α1 , Γ1 , Γ2 → Since the sequent α1 , Γ1 → Γ2 is provable and α1 + β1 by assumption, β1 , Γ1 → Γ2 is also provable, which implies that the desired sequent β1 , Γ1 , Γ2 → is provable, as follows: β1 , Γ1 → Γ2 β1 , Γ1 , Γ2 → β1 , Γ1 , Γ2 →
( →)
( →)
5. The case that the upper sequent of ( →) is obtained as the lower sequent of (→ ): We can proceed similarly to (4j). The case that the last inference of the proof of the sequent α1 , Γ → Δ is ( → ): This case is divided into several subcases according to how the upper sequent of ( → ) is obtained. (a) The case that the upper sequent of ( → ) is an axiom sequent: The treatment of this case is similar to (4a) and is safely left to the reader.
246
Hirokazu Nishimura
(b) The case that the upper sequent of ( → ) is obtained as the lower sequent of (extension): This case can safely be left to the reader. (c) The case that the upper sequent of ( → ) is obtained as the lower sequent of (∧ →): The last two steps of the proof that we must consider go as follows: α, Δ1 → α1 , Γ1 (∧ →) α ∧ β, Δ1 → α1 , Γ1 α1 , Γ1 → (α ∧ β) , Δ1
( → )
The sequent α1 , Γ1 → α , Δ1 is provable with a shorter proof than that of α1 , Γ1 → (α ∧ β) , Δ1 as follows: α, Δ1 → α1 , Δ1 α1 , Γ1 → α , Δ1
( → )
Thus the sequent β1 , Γ1 → α , Δ1 is provable by the induction hypothesis, which implies that the desired sequent β1 , Γ1 → (α ∧ β) , Δ1 is also provable as follows: β1 , Γ1 → α , Δ1 β1 , Γ1 → (α ∧ β) , Δ1
(→ ∧ )
(d) The case that the upper sequent of ( → ) is obtained as the lower sequent of (→ ∨): The treatment is similar to (5c) and is safely left to the reader. (e) The case that the upper sequent of ( → ) is obtained as the lower sequent of (∨ →): The last two steps of the proof that we have to consider go as follows: β → α1 , Γ1 α → α1 , Γ1 α ∨ β → α1 , Γ1 α1 , Γ1 → (α ∨ β)
(∨ →)
( → )
The sequents α1 , Γ1 → α and α1 , Γ1 → β are provable with shorter proofs than that of α1 , Γ1 → (α ∨ β) as follows: α → α1 , Γ1 α1 , Γ1 → α
( → )
β → α1 , Γ1 α1 , Γ1 → β
( → )
Thus the sequents β1 , Γ1 → α and β1 , Γ1 → β are provable by the induction hypothesis, which implies that the desired sequent α1 , Γ1 → (α ∨ β) is provable, as follows: β1 , Γ1 → β β1 , Γ1 → α β1 , Γ1 → (α ∨ β)
(→ ∨ )
Gentzen Methods in Quantum Logic
247
(f) The case that the upper sequent of ( → ) is the lower sequent of (→ ): The last two steps of the proof that we should consider can be supposed to be one of the following two forms: α1 , Γ1 → Γ2 → α1 , Γ1 , Γ2 α1 , Γ1 , Γ2 →
(→ ) ( → )
Γ1 → α1 , Γ2 (→ ) → α1 , Γ1 , Γ2 α1 , Γ1 , Γ2 →
( → )
In the former case the sequent α1 , Γ1 , Γ2 → is provable as follows: α1 , Γ1 → Γ2 ( →) α1 , Γ1 , Γ2 → α1 , Γ1 , Γ2 →
( →)
Since α1 + β1 by assumption, the desired sequent β1 , Γ1 , Γ2 → is also provable. As for the latter case, the sequent α1 , Γ2 → Γ1 is provable with a shorter proof than that of the sequent α1 , Γ1 , Γ2 →, as follows: Γ1 → α1 , Γ2 α1 , Γ2 → Γ1
( → )
By the induction hypothesis the sequent β1 , Γ2 → Γ1 is also provable, which implies that the desired sequent β1 , Γ1 , Γ2 → is provable, as follows: β1 , Γ2 → Γ1 ( →) β1 , Γ1 , Γ2 → (g) The case that the upper sequent of ( → ) is obtained as the lower sequent of ( →) or (→ ): The treatment is similar to (4c) and is safely left to the reader. (h) The case that the upper sequent of ( → ) is the lower sequent of another ( → ): The last two steps of the proof that we have to consider go as follows: α1 , Γ1 → Δ1 ( → ) Δ1 → α1 , Γ1 ( → ) α1 , Γ1 → Δ1 Since the sequent α1 , Γ1 → Δ1 has a shorter proof than the sequent α1 , Γ1 → Δ1 , the sequent β1 , Γ1 → Δ1 is also provable by the induction hypothesis, which implies that the desired sequent β1 , Γ1 → Δ1 is also provable, as follows β1 , Γ1 → Δ1 ( →) β1 , Γ1 → Δ1 β1 , Γ1 → Δ1
(→ )
248
Hirokazu Nishimura
(i) The case that the upper sequent of ( → ) is obtained as the lower sequent of (∨ →), (→ ∧ ), (∧ →), or (→ ∨ ): These four cases can be . dealt with similarly, so here we deal only with the case of (→ ∨ ), in which the last two steps of the proof that we must consider go as follows: Δ1 → β Δ1 → α (→ ∨ ) Δ1 → (α ∨ β) ( → ) (α ∨ β) → Δ1 Here α1 is supposed to be α ∨ β. The sequents α → Δ1 and β → Δ1 are provable with shorter proofs than that of (α ∨ β) → Δ1 as follows: Δ1 → α α → Δ1
( → )
Δ1 → β β → Δ1
( → )
Therefore the sequents α → Δ1 and β → Δ1 are provable by the induction hypothesis, which implies that the sequent α ∨ β → Δ1 is also provable, as follows: α → Δ1 β → Δ1 α ∨ β → Δ1
(∨ →)
Since β1 + α1 = α ∨ β by assumption, the desired sequent β1 → Δ1 is provable. (j) The case that the upper sequent of ( → ) is obtained as the lower sequent of ( → ∧): The last two steps of the proof that we must consider go as follows: β → α1 , Γ1 α → α1 , Γ1 → α1 , Γ1 , α ∧ β α1 , Γ1 , (α ∧ β) →
( → ∧)
( → )
The sequents Γ1 , α1 → α and Γ1 , α1 → β are provable with shorter proofs than that of α1 , Γ1 , (α ∧ β) → as follows: α → α1 , Γ1 α1 , Γ1 → α
( → )
β → α1 , Γ1 α1 , Γ1 → β
( → )
By the induction hypothesis the sequents β1 , Γ1 → α and β1 , Γ1 → β are provable, which implies that the desired sequent β1 , Γ1 , (α ∧ β) → is also provable, as follows: β1 , Γ1 → α β1 , Γ1 → β (→ ∧) β1 , Γ1 → α ∧ β β1 , Γ1 , (α ∧ β) →
( →)
Gentzen Methods in Quantum Logic
249
COROLLARY 11. 1. If a sequent Γ, Π → Δ, Σ is provable, then the sequent Δ , Σ → Γ , Π is also provable. 2. If a sequent Γ → Δ is provable, then the sequent Γ, Δ → is provable. 3. If a sequent Γ → Δ is provable, then the sequent → Γ, Δ is also provable. Proof. By Theorem 8 it suﬃces only to take into account the rules ( → ), ( →), and (→ ). THEOREM 12. (The second duality theorem). If α1 + β1 and α2 + β2 , then α1 ∧ α2 + (β1 ∨ β2 ) and α1 ∨ α2 + (β1 ∧ β2 ) . Proof. First we show the following claim: CLAIM 13. 1. If a sequent α1 ∧ α2 , Γ → Δ is provable, then the sequent (β1 ∨ β2 ) , Γ → Δ is also provable. 2. If a sequent Γ → Δ, α1 ∧ α2 is provable, then the sequent Γ → Δ, (β1 ∨ β2 ) is also provable. 3. If a sequent α1 ∨ α2 , Γ → Δ is provable, then the sequent (β1 ∧ β2 ), Γ → Δ is also provable. 4. If a sequent Γ → Δ, α1 ∨ α2 is provable, then the sequent Γ → Δ, (β1 ∧ β2 ) is also provable. Here we deal only with the second statement in the above claim, leaving the remaining three statements to the reader. The proof is carried out by induction on the construction of a proof P of the sequent Γ → Δ, α1 ∧ α2 . Here we deal only with the critical case in which the last inference is (→ ∧) as follows: Γ → α1 Γ → α2 Γ → α1 ∧ α2
(→ ∧)
Since α1 + β1 and α2 + β2 by assumption, the sequents Γ → β1 and Γ → β2 are provable, which implies that the sequent Γ → (β1 ∨ β2 ) is is provable, as follows: Γ → β1 Γ → β2 (→) ( → ) β1 → Γ β2 → Γ β1 ∨ β2 → Γ Γ → (β1 ∨ β2 )
(∨ →)
( → )
Therefore the sequent Γ → (β1 ∨ β2 ) s provable by Theorem 8. To establish the remaining half of the theorem smoothly, we introduce a useful notion weaker than provability equivalence. A wﬀ β is said to be provably dominated by a wﬀ α, in notation α β, if we have that for any ﬁnite sets Γ rand Δ of wﬀs:
250
Hirokazu Nishimura
1. Whenever the sequent α, Γ → δ is provable, the sequent β, Γ → Δ is also provable. 2. Whenever the sequent Γ → Δ, α is provable, the sequent Γ → Δ, β is also provable. →δ1 and We notice that what we have really proved in Claim 13 is that if γ1 →δ2 , then γ1 ∧γ2 →(δ1 ∨δ2 ) and γ1 ∨γ2 →(δ1 ∧δ2 ) . Similarly, what we have really γ2 →δ1 and γ2 →δ2 , then γ1 →δ1 , γ1 ∧ proved in the proof of Theorem 7 is that if γ1 →δ1 ∧ δ2 nd γ1 ∨ γ2 →δ1 ∨ δ2 , while what we have really proved in the proof of γ2 →β. It is easy to see that two wﬀs α and β Theorem 8 is that if α →β, , then α are provably equivalent iﬀ each of them is provably dominated by the other. Thus, to conclude the proof of the theorem, it suﬃces to notice that →(β1 ∨ β2 ) →(α1 ∧ α2 ) →α1 ∧ α2 α1 ∧ α2 COROLLARY 14. 1. If α1 + β1 and α2 + β2 , then α1 ∧ α2 + (β1 ∨ β2 ) and α1 ∨ α2 + (β1 ∧ β2 ) . Proof. By Theorems 7, 8, and 12, we have that α1 ∧ α2 + (α1 ∨ α2 ) + (β1 ∨ β2 ) and α1 ∨ α2 + (α1 ∧ α2 ) + (β1 ∧ β2 ) . 5 THE CUTELIMINATION THEOREM THEOREM 15. A sequent α, β, Γ → Δ is provable iﬀ the sequent α ∧ β, Γ → Δ is provable. Similarly, a sequent Π → Σ, γ, δ is provable iﬀ the sequent Π → Σ, γ ∨ δ is provable. Proof. For both statements, the onlyif part follows readily from (∧ →) or (→ ∨). The if part can be established by induction on the construction of a proof of α ∧ β, Γ → Δ or Γ → Δ, α ∨ β. COROLLARY 16. A sequent α , β , Γ → Δ is provable iﬀ the sequent (α∨β) , Γ → Δ is provable. Similarly, a sequent Π → Σ, γ , δ is provable iﬀ the sequent Π → Σ, (γ ∧ δ) is provable. Proof. Follows from Corollary 1 and Theorem 15.
THEOREM 17. If a sequent α ∨ β, Γ → Δ is provable, then the sequents α, Γ → Δ and β, Γ → Δ are provable. Similarly, if a sequent Π → Σ, γ ∧ δ is provable, then the sequents Π → Σ, γ and Π → Σ, δ are provable.
Gentzen Methods in Quantum Logic
251
Proof. By induction on the construction of a proof of α ∨ β, Γ → Δ or Π → Σ, γ ∧ δ. Here we deal only with the case that the last step of a proof of a sequent α ∨ β, Γ → Δ is (∨ → ). So it must be one of the following two forms. Γ → α Γ → β (∨ → ) α ∨ β, Γ → α ∨ β, Γ1 → σ α ∨ β, Γ1 → ρ (∨ → ) α ∨ β, σ ∨ ρ, Γ1 → In the former case the sequents α , Γ → and β , Γ → are provable by ( →). So the desired sequents α, Γ → and β, Γ → are provable by Theorem 8. In the latter case the sequents α, Γ1 → σ , β, Γ1 → σ , and β, Γ1 → ρ , are provable by the induction hypothesis. So the desired sequents α, σ ∨ ρ, Γ1 → and β, σ ∧ ρ, Γ1 → are provable as follows: α, Γ1 → σ α, Γ1 → ρ (∨ → ) α, σ ∨ ρ, Γ1 → β, Γ1 → σ β, Γ1 → ρ (∨ → ) β, σ ∨ ρ, Γ1 → COROLLARY 18. If a sequent (α ∧ β) , Γ → Δ is provable, then the sequents α , Γ → Δ and β , Γ → Δ are provable. Similarly, if a sequent Π → Σ, (γ ∨ δ) is provable, then the sequents Π → Σ, γ and Π → Σ, δ are provable. Proof. This follows from Theorem 17 and Corollary 1.
THEOREM 19. (The cutelimination theorem). If sequents Γ1 → Δ1 , α and α, Γ2 → Δ2 are provable with either Δ1 = ∅ or Γ2 = ∅, then the sequent Γ1 , Γ2 → Σ1 , Δ2 is also provable. In other words, (cut)q is permissible in GMQL. Proof. Suppose that the sequents Γ1 → Δ1 , α and α, Γ2 → Δ2 have proofs P1 and P2 , respectively. We prove the theorem by double induction principally on G(α) and secondarily on l(P1 ) + l(P2 ). By Theorem 12 we can assume that there is no occurrence of the disjunction symbol ∨ in P1 or P2 . As in the proof of Theorem 8, whenever we are forced to deal with the rules (∧ →) or (→ ∧ ), each of which consists of two forms, only one of them is treated. Our proof is divided into several cases according to which inference rule is used in the last step of P1 or P2 as follows: 1. The case that one of the sequents Γ1 → Δ1 , α and α, Γ2 → Δ2 is an axiom sequent: There is nothing to prove.
252
Hirokazu Nishimura
2. The case that one of the sequents Γ1 → Δ1 , α and α, Γ2 → Δ2 is obtained as the lower sequent of (extension): Here we deal only with the case that the former sequent Γ1 → Δ1 , α is obtained as the lower sequent of (extension), leaving the dual case to the reader. Then the last step of the proof P1 is in one of the following two forms: Γ11 → Δ11 , α (extension) Γ11 , Γ12 → Δ11 , Δ12 , α Γ11 → Δ11 (extension) Γ11 , Γ12 → Δ11 , Δ12 , α In the former case the desired sequent Γ11 , Γ12 , Γ2 → Δ11 , Δ12 , Δ2 is provable by induction hypothesis as follows. α, Γ2 → Δ2 Γ11 → Δ11 , α (cut)q Γ11 , Γ2 → Δ11 , Δ2 Γ11 , Γ12 , Γ2 → Δ11 , Δ12 , Δ2
(extension)
In the latter case the desired sequent Γ11 , Γ12 , Γ2 → Δ11 , Δ12 , Δ2 is obtained as follows. Γ11 → Δ11 (extension) Γ11 , Γ12 , Γ2 → Δ11 , Δ12 , Δ2 3. The case that either the sequent Γ1 → Δ1 , α is obtained as the lower sequent of one of the inference rules ( →) and (∧ →) or the sequent α, Γ2 → Δ2 is obtained as the lower sequent of one of the inference rules (→ ) and (→ ∧ ): Here we deal only with the case that the sequent α, Γ2 → Δ2 is obtained as the lower sequent of (→ ∧ ), leaving the remaining three cases to the reader. So the last step of P2 is of the following form: α, Γ2 → Σ2 , β α, Γ2 → Σ2 , (β ∨ γ)
(→ ∧ )
The desired sequent Γ1 , Γ2 → Δ1 , Σ2 , (β ∧ γ) is provable by induction hypothesis as follows: α, Γ2 → Σ2 , β Γ1 → Δ1 , α (cut)q Γ1 , Γ2 → Δ1 , Σ2 , β Γ1 , Γ2 → Δ1 , Σ2 , (β ∧ γ)
(→ ∧ )
4. The case that either the sequent Γ1 → Δ1 , α is obtained as the lower sequent of (∧ →) or the sequent α, Γ2 → Δ2 is obtained as the lower sequent of (→ ∧): Here we deal only with the former case, leaving a similar treatment of the latter case to the reader. So the last step of P1 goes as follows:
Gentzen Methods in Quantum Logic
β → Δ1 , α γ → Δ1 , α (β ∧ γ) → Δ1 , α
253
(∧ →)
If γ2 = ∅, then the desired sequent (β ∧ γ) → Δ1 , Δ2 is provable by the induction hypothesis as follows: β → Δ1 , α α → Δ2 β1 → Δ1 , Δ2
γ → Δ1 , α α → Δ2 γ2 → Δ1 , Δ2 (β ∧ γ) → Δ1 , Δ2 (cut)q
(cut)q
(∧ →)
Unless Γ2 = ∅, the situation can be classiﬁed into cases according to which inference rule is used in the last step of P2 . If Γ2 = ∅ and it is not the case that the last inference of P2 is (→ ∧), the situation is subsumed under the cases that have been or will be dealt with. If Γ2 = ∅ and the last inference of P2 is (→ ∧), then surely Γ1 = ∅, so that the situation can be handled dually to the case that Γ2 = ∅. 5. The case that either the sequent Γ1 → Δ1 , α is obtained as the lower sequent of one of the inference rules (→ ) and (→ ∧ ) or the sequent α, Γ2 → Δ2 is obtained as the lower sequent of one of the inference rules ( →) and (∧ →): Here we deal only with the case that the sequent Γ1 → Δ1 , α is obtained as the lower sequent of (→ ∧ ), leaving the remaining three cases to the reader. So the last step of P1 is in one of the following two forms: Γ1 → Σ, β , α Γ1 → Σ, (β ∧ γ) , α
(→ ∧ )
Γ1 → Δ1 , β Γ1 → Δ1 , (β ∧ γ)
(→ ∧ )
In the latter case α is supposed to be (β ∧ γ) . In the former case the (cut)q at issue is an instance of (cut1), so that Γ2 = ∅, and the desired sequent Γ1 → Σ, (β ∧ γ) , Δ2 is provable by induction hypothesis, as follows: α → Δ2 Γ1 → Σ, β , α Γ1 → Σ, β Γ1 → Σ, (β ∧ γ)
(cut)q
(→ ∧ )
As for the latter case, the cut formula is (β∧γ) , and the sequent, β , Γ2 → Δ2 is provable by Corollary 18. Thus the desired sequent Γ1 , Γ2 → Δ1 , Δ2 is provable by the induction hypothesis, as follows: β , Γ2 → Δ2 Γ1 → Δ1 , β Γ1 , Γ2 → Δ1 , Δ2
(cut)q
254
Hirokazu Nishimura
6. The case that either the sequent Γ1 → Δ1 , α is obtained as the lower sequent of (→ ∧) or the sequent α, Γ2 → Δ2 is obtained as the lower sequent . of (∧ →): Here we deal only with the latter case, leaving a similar treatment of the latter to the reader. So the last step of P2 goes as follows: γ → Δ2 β → Δ2 (β ∧ γ) → Δ2
(∧ →)
Here α is supposed to be (β∧γ) , and the (cut)q at issue is an instance of (cut1) with the cut formula (β ∧ γ) . By Corollary 8 the sequent Γ1 → Δ1 , β , γ is provable, so that the desired sequent Γ1 → Δ1 , Δ2 is also provable, as follows: β → Δ2 Γ1 → Δ1 , β , γ Γ1 → Δ1 , Δ2 , γ γ → Δ2 Γ1 → Δ1 , Δ2
(cut)q (cut)q
7. The case that the sequent Γ1 → Δ1 , α is obtained as the lower sequent of ( → ∧): The last step of P1 is in one of the following two forms: β → Σ, α γ → Σ, α → Σ, β ∧ γ, α γ → Δ1 β → Δ1 → Δ1 , β ∧ γ
( → ∧)
( → ∧)
In the latter case α is assumed to be β ∧ γ. First we deal with the former case, in which the (cut)q at issue is (cut1) so that Γ2 = ∅. Then the desired sequent → Σ, β ∧ γ, Δ2 is provable by the induction hypothesis as follows: α → Δ2 β → Σ, α β1 → Σ, Δ2
γ → Σ, α α → Δ2 γ → Σ, Δ2 → Σ, β ∧ γ, Δ2
(cut)q
(cut)q
( → ∧)
As for the latter case, suppose ﬁrst that Δ1 = ∅, so that Γ2 = ∅. Then the sequents Δ1 → β and Δ1 → γ are provable by Corollary 11, while the sequent β, γ → Δ1 , Δ2 is provable by Theorem 15. Thus the sequent → Δ1 , Δ2 is provable by the induction hypothesis, as follows: Δ1
Δ1 → β β, γ → Δ2 → γ γ, Δ1 → Δ2 Δ1 → Δ2 → Δ1 , Δ2
(cut)q (cut)q
(→ )
Thus the desired sequent → Δ1 , Δ2 is provable by Theorem 8. If Δ1 = ∅, then the sequents → β and → γ are provable by Corollary 11, while the
Gentzen Methods in Quantum Logic
255
sequent β, γ, Γ2 → Δ2 is provable by Theorem 15. Thus the desired sequent Γ2 → Δ2 is provable by the induction hypothesis as follows: →β →γ
β, γ, Γ2 → Δ2 γ, Γ2 → Δ2 Γ2 → Δ2
(cut)q (cut)q
8. The case that one of the sequents Γ1 → Δ1 , α and α, Γ2 → Δ2 is obtained as the lower sequent of (→ ): Here we deal only with the case that the sequent Γ1 → Δ1 , α is obtained as the lower sequent of (→ ), leaving the dual case to the reader. So the last step of the proof P1 is in one of the following two forms: Δ12 → Δ11 , α (→ ) → Δ11 , Δ12 , α Δ12 , β → Δ11 → Δ11 , Δ12 , β
(→ )
In the latter case α is supposed to be β . First we deal with the former case. If Γ2 = ∅, then the desired sequent → Δ11 , Δ12 , Δ2 is provable by the induction hypothesis, as follows: α → Δ2 Δ12 → Δ11 , α Δ12 → Δ11 , Δ2 → Δ11 , Δ12 , Δ2
(cut)q
(→ )
If Γ2 = ∅, then α is of the form γ and the sequent Δ12 → Δ11 , α is γ → γ . The sequents γ → and Δ2 → γ are provable by Corollary 11, which implies that the sequent Δ2 → is also provable by the induction hypothesis, as follows: γ→ Δ2 → γ (cut)q Δ2 → By Corollary 11 the sequent → Δ2 is provable, which implies that the desired sequent → Δ1 , Δ2 is provable as follows: → Δ2 → Δ1 , Δ2
(extension)
Now we deal with the latter case. If Γ2 = ∅, then the sequent Δ2 → β is provable by Corollary 11, and the sequent → Δ11 , Δ12 , Δ2 is also provable by the induction hypothesis as follows: . Δ2 → β Δ12 , β → Δ11 (cut)q Δ12 , Δ2 → Δ11 → Δ11 , Δ12 , Δ2
(→ )
256
Hirokazu Nishimura
Thus the desired sequent → Δ11 , Δ12 , Δ2 is provable by Theorem 8. If Γ2 = ∅, then the sequent Δ12 , β → Δ11 must be β → or β → β , the latter of which implies by Corollary 11 that the sequent β → is provable. Thus in any case the sequent β → is provable. Since the sequent Δ2 → Γ2 , β is provable by Corollary 11, the sequent Δ2 → Γ2 is provable by induction hypothesis as follows: β→ Δ2 → Γ2 , β Δ2 → Γ2
(cut)q
Therefore the sequent Γ2 → Δ2 is provable by Corollary 11, which implies that the desired sequent Γ2 → Δ1 , Δ2 is provable as follows: Γ2 → Δ2 Γ2 → Δ1 , Δ2
(extension)
9. The case that both the sequent Γ1 → Δ1 , α and the sequent α, Γ2 → Δ2 are obtained as the lower sequent of ( → ): The last steps of the proofs P1 and P2 go as follows: Σ1 , β → Π1 ( → ) Π1 → Σ1 , β Σ2 , → β, Π2 β , Π2 → Σ2
( → )
In the above α is supposed to be β . The desired sequent Π1 , Π2 → Σ1 , Σ2 is provable by the induction hypothesis as follows: Σ2 → Π2 , β β, Σ1 → Π1 (cut) Σ1 , Σ2 → Π1 , Π2 ( → ) Π1 , Π2 → Σ1 , Σ2 6 THE COMPLETENESS THEOREM An Oframe is a pair (X, ⊥) of a nonempty set X and an orthogonality relation (i.e., an irreﬂexive and symmetric binary relation) on X. Given Y ⊆ X, we write Y ⊥ for the set {x ∈ Xx ⊥ y for any y ∈ Y }. A subset Y of X is said to be ⊥closed if Y = Y ⊥⊥ . An Omodel is a triple (X, ⊥, D), where (X, ⊥) is an Oframe and D assigns to each propositional variable p a ⊥closed subset D(p) of X. The notation 'α' for a wﬀ α is deﬁned inductively as follows: 1. 'p' = D(p) for any propositional variable p.
Gentzen Methods in Quantum Logic
257
2. 'α ∧ β' = 'α' ∩ 'β. 3. 'α ' = α⊥ . 4. 'α ∨ β' = 'α ∧ β ' = ('α'⊥ ∩ 'β⊥ )⊥ . Given x ∈ X and a wﬀ α, we write V (α; x) = 1 if x ∈ 'α' and V (α; x) = 0 if x ∈ / 'α'. Given x ∈ X and a sequent Γ → Δ, we write V (Γ → Δ; x) = 1 if x ∈ {'α' α ∈ Γ} and x ∈ / (∪{'β'⊥ β ∈ Δ})⊥ , and V (Γ → Δ; x) = 0 otherwise. A sequent Γ → Δ is said to be realizable if there exists an Omodel (X, ⊥, D) and some x ∈ X such that V (Γ → Δ; x) = 1. The sequent Γ → Δ is called valid otherwise. THEOREM 20. (The soundness theorem). If a sequent Γ → Δ is provable, then it is valid. Proof. By induction on the construction of a proof of the sequent Γ → Δ.
A set Ω of wﬀs is said to be admissible if it satisﬁes the following conditions: 1. If p is a propositional variable and p ∈ Ω, then p ∈ Ω. 2. If α ∈ Ω and β is a subformula of α, then β ∈ Ω. 3. If (α ∨ β) ∈ Ω, then (α ∧ β ) ∈ Ω. A ﬁnite set Γ of wﬀs is said to be inconsistent if for some wﬀ α, both of the sequents Γ → α and Γ → α are provable. Otherwise the set Γ is said to be consistent. LEMMA 21. A ﬁnite set Γ of wﬀs is inconsistent iﬀ the sequent Γ → is provable. Proof. The if part is obvious. The onlyif part can be shown easily as follows:
Γ → α Γ → α Γ → α ∧ α
(→ ∧) Γ→
α→α ( →) α, α → α ∧ α →
(∧ →) (cut)q
Given an admissible set Ω of wﬀs, the Ωcanonical Omodel M(Ω) = (XΩ , ⊥Ω , DΩ ) is deﬁned as follows: 1. XΩ is the set of all the consistent subsets of Ω. 2. For any Γ1 , Γ2 ∈ XΩ , Γ1 ⊥Ω Γ2 iﬀ for some α ∈ Ω, either: (a) both of the sequents Γ1 → α and Γ1 → α are provable, or (b) both of the sequents Γ1 → α and Γ2 → α are provable.
258
Hirokazu Nishimura
3. If p ∈ / Ω, then DΩ (p) = ∅, while if p ∈ Ω, then DΩ (p) consists of all the consistent subsets Γ of Ω such that the sequent Γ → p is provable. THEOREM 22. M(Ω) is an Omodel. Proof. Obviously the relation ⊥Ω is symmetric. That the relation ⊥Ω is irreﬂexive follows from our assumption that every element of XΩ is a consistent set of wﬀs. Now it remains to show that DΩ (p) is ⊥Ω closed for any propositional variable p. Unless p ∈ Ω, there is nothing to prove. So let p ∈ Ω. Let Γ be an element of XΩ such that the sequent Γ → p is not provable. Suppose for the sake of contradiction that the set {p } is inconsistent, which implies by Lemma 21 that the sequent p → is provable. By Corollary 11 the sequent → p is provable, which implies by (extension) that the sequent Γ → p is provable. This is a contradiction. So {p } ∈ XΩ . Suppose, for the sake of contradiction, that for some α ∈ Ω, either both of the sequents Γ → α and p → α are provable or both of the sequents Γ → α and p → α are provable. Here we deal only with the former case, leaving a similar treatment of the latter to the reader. By Corollary 11 the sequent α → p is provable, which implies by (cut) that the sequent Γ → p is provable. This is a contradiction. Thus it cannot be the case that Γ ⊥Ω {p }, while for any Δ ∈ XΩ such that the sequent Δ → p is provable, Δ ⊥Ω {p}. This implies that the set of all Δ ∈ XΩ such that the sequent Δ → p is provable is ⊥Ω closed. The disjunction grade of a wﬀ α, denoted by G∨ (α), is deﬁned inductively as follows: 1. G∨ (p) = o for any propositional variable p. 2. G∨ (α ) = G∨ (α). 3. G∨ (α ∧ β) = G∨ (α) + G∨ (β). 4. G∨ (α ∨ β) = G∨ (α) + G∨ (β) + 1. THEOREM 23. (The fundamental theorem for M(Ω)). For any α ∈ Ω and any Γ ∈ XΩ , the sequent Γ → α is provable iﬀ Γ ∈ 'α in M(Ω). Proof. The proof is carried out by double induction principally on G∨ (α) and secondarily on G(α). The proof is divided into several cases. 1. In the case that α is a propositional variable: It follows from the deﬁnition of DΩ . 2. In the case that α = β for some wﬀ β: If Γ → β is provable, then Γ ⊥Ω 'β' by induction hypothesis, which implies that Γ ∈ 'β '. Suppose, for the sake of contradiction, that the set {β} is inconsistent, which implies by Lemma 21 that the sequent β → is provable. Thus the sequent Γ → β is provable as follows:
Gentzen Methods in Quantum Logic
β→ (→ ) → β Γ → β
259
(extension)
This is a contradiction. So it must be the case that {β} ∈ XΩ . Suppose, for the sake of contradiction, that for some γ ∈ Ω, either both of the sequents Γ → γ and β → γ are provable or both of the sequents Γ → γ and β → γ are provable. Here we deal only with the former case, leaving safely a similar treatment of the latter to the reader. The desired contradiction is obtained as follows: Γ → γ
β→γ γ → β Γ→β
( → ) (cut)q
Thus it cannot be the case that Γ ⊥ Ω {β}. Since {β} ∈ 'β' by induction hypothesis, this means that Γ ∈ / 'β' ⊥ Ω = 'β '. 3. In the case that α is of the form β ∧ γ for some wﬀs β and γ: If the sequent Γ → α is provable, then both of the sequents Γ → β and Γ → γ are provable by Theorem 17, which implies by induction hypothesis that Γ ∈ 'β' and Γ ∈ 'γ'. So Γ ∈ 'β' ∩ 'γ' = 'β ∧ γ'. Unless the sequent Γ → α is provable, suppose, for the sake of contradiction, that both of the sequents Γ → β and Γ → γ are provable. The desired conclusion is obtained as follows: Γ→β Γ→γ Γ→β∧γ
(→ ∧)
Thus one of the sequents Γ → β and Γ → γ is consistent, which implies by induction hypothesis that Γ ∈ / 'β' or Γ ∈ / 'γ'. So Γ ∈ / 'β ∧ γ' = 'β' ∩ 'γ'. 4. In the case that α is of the form β ∨ γ for some wﬀs β and γ: Use Theorem 12. THEOREM 24. (The completeness theorem). A sequent Γ → Δ is realizable iﬀ it is consistent. Proof. The onlyif part is the soundness theorem already established. To see the if part, take an admissible set Ω such that Γ ∪ {β1 ∨ ... ∨ βn } ⊆ Ω, where Δ = {β1 , ..., βn }. By Theorem 15 the sequent Γ → Δ is consistent iﬀ the sequent Γ → β1 ∨...∨βn is consistent. The desired conclusion follows readily from Theorem 23. We remark in passing that in the proof of Theorem 24 it does not matter how to insert parentheses in β1 ∨ ... ∨ βn .
260
Hirokazu Nishimura
BIBLIOGRAPHY [Birkhoﬀ and von Neumann, 1936] G. Birkhoﬀ and J. von Neumann. The logic of quantum mechanics, Annals of Mathematics, 37, 823843, 1936. [Bruns, 1976] G. Bruns. Free ortholattices, Canadian Journal of Mathematics, 28, 977985, 1976. [Cutland and Gibbins, 1982] N. J. Cutland and P. F. Gibbins. A regular sequent calculus for quantum logic in which ∧ and ∨ are dual, Logique et Analyse, 99, 221248, 1982. [Dalla Chiara, 1986] M. L. Dalla Chiara. Quantum logic, in Handbook of Philosophical Logic, D. Gabbay and F. Guenthner, eds., Reidel, Dordrecht, Volume III, pp. 427469, 1986. [Dishkant, 1972] H. Dishkant. Semantics for the minimal logic of quantum mechanics, Studia Logica, 30, 2336, 1972. [Dishkant, 1977] H. Dishkant. Imbedding of the quantum logic in the modal system of Brower, Journal of Symbolic Logic, 42, 321328, 1977. [Foulis and Randall, 1971] D. J. Foulis and C. H. Randall. Lexicographic orthogonality, Journal of Combinatorial Theory, 11, 157162, 1971. [Gentzen, 1935] G. Gentzen. Untersuchungen ﬁber das logische Schliessen, I and II, Mathematische Zeitschrift, 39, 176210, 405431, 1935. [Goldblatt, 1974] R. I. Goldblatt. Semantical analysis of orthologic, Journal of Philosophical Logic, 3, 1936, 1974. [Goldblatt, 1975] R. I. Goldblatt. The Stone space of an ortholattice, Bulletin of the London Mathematical Society, 7, 4548. 1975. [Goldblatt, 1984] R. I. Goldblatt. Orthomodularity is not elementary. Journal of Symbolic Logic, 49, 401404, 1984. [Jammer, 1974] M. Jammer. The Philosophy of Quantum Mechanics, Wiley, New York, 1974. [Kripke, 1963] S. Kripke. Semantical analysis of modal logic, I, normal modal propositional calculi, Z. Math. Logik Grundlagen Math., 9, 6796, 1963. [Maeda, 1980] S. Maeda. Lattice Theory and Quantum Logic, Maki, Tokyo [in Japanese], 1980. [McKinsey and Tarski, 1948] J. C. McKinsey and A. Tarski. Some theorems about the sentential calculi of Lewis and Heyting, Journal of Symbolic Logic, 13, 115, 1948. [Nishimura, 1980] H. Nishimura. Sequential method in quantum logic, Journal of Symbolic Logic, 45, 339352, 1980. [Nishimura, 1983] H. Nishimura. A cutfree sequential system for the propositional modal logic of ﬁnite chains, Publications of RIMS, Kyoto University, 19, 305316, 1983. [Nishimura, 1994a] H. Nishimura. Proof theory for minimal quantum logic I, International Journal of Theoretical Physics, 33, 103113, 1994. [Nishimura, 1994b] H. Nishimura. Proof theory for minimal quantum logic I, International Journal of Theoretical Physics, 33, 14271443, 1994. [Randall and Foulis, 1970] C. H. Randall and D. J. Foulis. An approach to empirical logic, American Mathematical Monthly, 77, 363374, 1970. [Sato, 1977] M. Sato. A study of Kripketype models for some modal logics by Gentzen’s sequential methods, Publications of RIMS, Kyoto University, 13, 381468, 1977. [Takano, 1995] M. Takano. Proof theory for minimal quantum logic:a remark, International Journal of Theoretical Physics, 34, 649654, 1995. [Takeuti, 1975] G. Takeuti. Proof Theory, NorthHolland, Amsterdam, 1975. [Tamura, 1988] S. Tamura. A Gentzen formulation without the cut rule for ortholattices, Kobe Journal of Mathematics, 5, 133150, 1988.
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
261
CATEGORICAL QUANTUM MECHANICS Samson Abramsky and Bob Coecke
1
INTRODUCTION
Our aim is to revisit the mathematical foundations of quantum mechanics from a novel point of view. The standard axiomatic presentation of quantum mechanics in terms of Hilbert spaces, essentially due to von Neumann [1932], has provided the mathematical bedrock of the subject for over 70 years. Why, then, might it be worthwhile to revisit it now? First and foremost, the advent of quantum information and computation (QIC) as a major ﬁeld of study has breathed new life into basic quantum mechanics, asking new kinds of questions and making new demands on the theory, and at the same time reawakening interest in the foundations of quantum mechanics. As one key example, consider the changing perceptions of quantum entanglement and its consequences. The initial realization that this phenomenon, so disturbing from the perspective of classical physics, was implicit in the quantummechanical formalism came with the EPR Gedankenexperiment of the 1930’s [Einstein et al., 1935], in the guise of a “paradox”. By the 1960’s, the paradox had become a theorem — Bell’s theorem [Bell, 1964], demonstrating that nonlocality was an essential feature of quantum mechanics, and opening entanglement to experimental conﬁrmation. By the 1990’s, entanglement had become a feature, used in quantum teleportation [Bennett et al., 1993], in protocols for quantum key distribution [Ekert, 1991], and, more generally, understood as a computational and informatic resource [Bouwmeester et al., 2001].
1.1
The Need for HighLevel Methods
The current tools available for developing quantum algorithms and protocols, and more broadly the whole ﬁeld of quantum information and computation, are deﬁcient in two main respects. Firstly, they are too lowlevel. One ﬁnds a plethora of ad hoc calculations with ‘bras’ and ‘kets’, normalizing constants, matrices etc. The arguments for the beneﬁts of a highlevel, conceptual approach to designing and reasoning about quantum computational systems are just as compelling as for classical computation. In particular, we have in mind the hardlearned lessons from Computer Science of the importance of compositionality, types, abstraction, and the use of tools from algebra and logic in the design and analysis of complex informatic processes.
262
Samson Abramsky and Bob Coecke
At a more fundamental level, the standard mathematical framework for quantum mechanics is actually insuﬃciently comprehensive for informatic purposes. In describing a protocol such as teleportation, or any quantum process in which the outcome of a measurement is used to determine subsequent actions, the von Neumann formalism leaves feedback of information from the classical or macroscopic level back to the quantum implicit and informal, and hence not subject to rigorous analysis and proof. As quantum protocols and computations grow more elaborate and complex, this point is likely to prove of increasing importance. Furthermore, there are many fundamental issues in QIC which remain very much open. The current lowlevel methods seem unlikely to provide an adequate basis for addressing them. For example: • What are the precise structural relationships between superposition, entanglement and mixedness as quantum informatic resources? Or, more generally, • Which features of quantum mechanics account for diﬀerences in computational and informatic power as compared to classical computation? • How do quantum and classical information interact with each other, and with a spatiotemporal causal structure? • Which quantum control features (e.g. iteration) are possible and what additional computational power can they provide? • What is the precise logical status and axiomatics of NoCloning and NoDeleting, and more generally, of the quantum mechanical formalism as a whole? These questions gain additional force from the fact that a variety of diﬀerent quantum computational architectures and informationprocessing scenarios are beginning to emerge. While at ﬁrst it seemed that the notions of Quantum Turing Machine [Deutsch, 1985] and the quantum circuit model [Deutsch, 1989] could supply canonical analogues of the classical computational models, recently some very diﬀerent models for quantum computation have emerged, e.g. Raussendorf and Briegel’s oneway quantum computing model [Raussendorf and Briegel, 2001; Raussendorf et al., 2003] and measurement based quantum computing in general [Jozsa, 2005], adiabatic quantum computing [Farhi et al., 2000], topological quantum computing [Freedman et al., 2004], etc. These new models have features which are both theoretically and experimentally of great interest, and the methods developed to date for the circuit model of quantum computation do not carry over straightforwardly to them. In this situation, we can have no conﬁdence that a comprehensive paradigm has yet been found. It is more than likely that we have overlooked many new ways of letting a quantum system compute. Thus there is a need to design structures and develop methods and tools which apply to these nonstandard quantum computational models . We must also address
Categorical Quantum Mechanics
263
the question of how the various models compare — can they be interpreted in each other, and which computational and physical properties are preserved by such interpretations?
1.2
HighLevel Methods for Quantum Foundations
Although our initial motivation came from quantum information and computation, in our view the development of highlevel methods is potentially of great signiﬁcance for the development of the foundations of quantum mechanics, and of fundamental physical theories in general. We shall not enter into an extended discussion of this here, but simply mention some of the main points: • By identifying the fundamental mathematical structures at work, at a more general and abstract level than that aﬀorded by Hilbert spaces, we can hope to gain new structural insights, and new ideas for how various physical features can be related and combined. • We get a new perspective on the logical structure of quantum mechanics, radically diﬀerent to the traditional approaches to quantum logic. • We get a new perspective on “NoGo” theorems, and new tools for formulating general results applying to whole classes of physical theories. • Our structural tools yield an eﬀective calculational formalism based on a diagrammatic calculus, for which automated software toolsupport is currently being developed. This is not only useful for quantum information and computation, it may also yield new ways of probing key foundational issues. Again, this mirrors what has become the common experience in Computer Science. In the age of QIC, Gedankenexperiments turn into programs! We shall take up some of these issues again in the concluding sections.
1.3
Outline of the Approach
We shall use category theory as the mathematical setting for our approach. This should be no surprise. Category theory is the language of modern structural mathematics, and the fact that it is not more widely used in current foundational studies is a regrettable consequence of the sociology of knowledge and the encumbrances of tradition. Computer Science, once again, leads the way in the applications of category theory; abstract ideas can be very practical! We shall assume a modest familiarity with basic notions of category theory, including symmetric monoidal categories. Apart from standard references such as [MacLane, 1998], a number of introductions and tutorials speciﬁcally on the use of monoidal categories in physics are now available [Abramsky and Tzevelekos, 2008; Baez and Stay, 2008; Coecke and Paquette, 2008]. More advanced textbooks in the area are [Kock, 2003; Street, 2007].
264
Samson Abramsky and Bob Coecke
We shall give an axiomatic presentation of quantum mechanics at the abstract level of strongly compact closed categories with biproducts — of which the standard von Neumann presentation in terms of Hilbert spaces is but one example. Remarkably enough, all the essential features of modern quantum protocols such as quantum teleportation [Bennett et al., 1993], logicgate teleportation [Gottesman ˙ and Chuang, 1999], and entanglement swapping [Zukowski et al., 1993] —which exploit quantum mechanical eﬀects in an essential way— ﬁnd natural counterparts at this abstract level. More speciﬁcally: • The basic structure of a symmetric monoidal category allows compound systems to be described in a resourcesensitive fashion (cf. the ‘no cloning’ [Dieks, 1982; Wootters and Zurek, 1982] and ‘no deleting’ [Pati and Braunstein, 2000] theorems of quantum mechanics). • The compact closed structure allows preparations and measurements of entangled states to be described, and their key properties to be proved. • The strong compact closed structure brings in the central notions of adjoint, unitarity and sesquilinear inner product —allowing an involution such as complex conjugation to play a role— and it gives rise to a twodimensional generalization of Dirac’s braket calculus [Dirac, 1947], in which the structure of compound systems is fully articulated, rather than merely implicitly encoded by labelling of basis states. • Biproducts allow probabilistic branching due to measurements, classical communication and superpositions to be captured. Moreover, from the combination of the—apparently purely qualitative—structures of strong compact closure and biproducts there emerge scalars and a Born rule. We are then able to use this abstract setting to give precise formulations of quantum teleportation, logic gate teleportation, and entanglement swapping, and to prove correctness of these protocols — for example, proving correctness of teleportation means showing that the ﬁnal state of Bob’s qubit equals the initial state of Alice’s qubit.
1.4 Development of the Ideas A ﬁrst step in the development of these ideas was taken in [Abramsky and Coecke, 2003], where it was recognized that compactclosed structure could be expressed in terms of bipartite projectors in Hilbert space, thus in principle enabling the structural description of information ﬂows in entangled quantum systems. In [Coecke, 2003] an extensive analysis of a range of quantum protocols was carried out concretely, in terms of Hilbert spaces, with a highly suggestive but informal graphical notation of informationﬂow paths through networks of projectors. The decisive step in the development of the categorical approach was taken in [Abramsky and Coecke, 2004], with [Abramsky and Coecke, 2005] as a supplement improving the
Categorical Quantum Mechanics
265
deﬁnition of strongly compact closed category. The present article is essentially an extended and revised version of [Abramsky and Coecke, 2004]. There have been numerous subsequent developments in the programme of categorical quantum mechanics since [Abramsky and Coecke, 2004]. We shall provide an overview of the main developments in Section 7, but the underlying programme as set out in [Abramsky and Coecke, 2004] still stands, and we hope that the present article will serve as a useful record of this approach in its original conception.
1.5
Related Work
To set our approach in context, we compare and contrast it with some related approaches.
Quantum Logic Firstly, we discuss the relationship with quantum logic as traditionally conceived, i.e. the study of lattices abstracted from the lattice of closed linear subspaces of Hilbert space [Birkhoﬀ and von Neumann, 1936]. We shall not emphasize the connections to logic in the present article, but in fact our categorical axiomatics can be seen as the algebraic or semantic counterpart to a logical type theory for quantum processes. This type theory has a resourcesensitive character, in the same sense as Linear logic [Girard, 1987] — and this is directly motivated by the nocloning and nodeleting principles of quantum information. The correspondence of our formalism to a logical system, in which a notion of proofnet (a graphical representation of multipleconclusion proofs) gives a diagrammatics for morphisms in the free strongly compact closed category with biproducts, and simpliﬁcation of diagrams corresponds to cutelimination, is developed in detail in [Abramsky and Duncan, 2006]. This kind of connection with logic belongs to the prooftheory side of logic, and more speciﬁcally to the CurryHoward correspondence, and the threeway connection between logic, computation and categories which has been a staple of categorical logic, and of logical methods in computer science, for the past three decades [Lambek and Scott, 1986; Abramsky and Tzevelekos, 2008]. The key point is that we are concerned with the direct mathematical representation of quantum processes. By contrast, traditional quantum logic is concerned with quantum propositions, which express properties of quantum systems. There are many other diﬀerences. For example, compound systems and the tensor product are central to our approach, while quantum logic has struggled to accommodate these key features of quantum mechanics in a mathematically satisfactory fashion. However, connections between our approach and the traditional setting of orthomodular posets and lattices have been made by John Harding [2007; 2008].
266
Samson Abramsky and Bob Coecke
Categories in Physics There are by now several approaches to using category theory in physics. For comparison, we mention the following: • [Baez and Dolan, 1995; Crane, 2006]. Higherdimensional categories, TQFT’s, categoriﬁcation, etc. • [Isham and Butterﬁeld, 1998; Do¨ering and Isham, 2007]. The topostheoretic approach. Comparison with the topos approach The topos approach aims ambitiously at providing a general framework for the formulation of physical theories. It is still in an early stage of development. Nevertheless, we can make some clear comparisons. Our approach monoidal linear processes geometry of proofs
vs. vs. vs. vs.
Topos approach cartesian intuitionistic propositions geometric logic
Rather as in our comparison with quantum logic, the topos approach is primarily concerned with quantum propositions, whereas we are concerned directly with the representation of quantum processes. Our underlying logical setting is linear, theirs is cartesian, supporting the intuitionistic logic of toposes. It is an interesting topic for future work to relate, and perhaps even usefully combine, these approaches. Comparison with the ncategories approach The ncategories approach is mainly motivated by the quest for quantum gravity. In our approach, we emphasize the following key features which are essentially absent from the ncategories work: • operational aspects • the interplay of quantum and classical • compositionality • open vs. closed systems. These are important for applications to quantum informatics, but also of foundational signiﬁcance. There are nevertheless some intriguing similarities and possible connections, notably in the rˆ ole played by Frobenius algebras, which we will mention brieﬂy in the context of our approach in Section 7.
Categorical Quantum Mechanics
1.6
267
Outline of the Article
In Section 2, we shall give a rapid review of quantum mechanics and some quantum protocols such as teleportation. In Sections 3, 4 and 5, we shall present the main ingredients of the formalism: compact and strongly compact categories, and biproducts. In Section 6, we shall show how quantum mechanics can be axiomatized in this setting, and how the formalism can be applied to the complete speciﬁcation and veriﬁcation of a number of important quantum protocols. In Section 7 we shall review some of the main developments and advances made within the categorical quantum mechanics programme since [Abramsky and Coecke, 2004], thus giving a picture of the current state of the art. 2 REVIEW OF QUANTUM MECHANICS AND TELEPORTATION In this paper, we shall only consider ﬁnitary quantum mechanics, in which all Hilbert spaces are ﬁnitedimensional. This is standard in most current discussions of quantum computation and information [Nielsen and Chuang, 2000], and corresponds physically to considering only observables with ﬁnite spectra, such as spin. (We refer brieﬂy to the extension of our approach to the inﬁnitedimensional case in the Conclusion.) Finitary quantum theory has the following basic ingredients (for more details, consult standard texts such as [Isham, 1995]). 1. The state space of the system is represented as a ﬁnitedimensional Hilbert space H, i.e. a ﬁnitedimensional complex vector space with a ‘sesquilinear’ innerproduct written φ  ψ, which is conjugatelinear in the ﬁrst argument and linear in the second. A state of a quantum system corresponds to a onedimensional subspace A of H, and is standardly represented by a vector ψ ∈ A of unit norm. 2. For informatic purposes, the basic type is that of qubits, namely 2dimensional Hilbert space, equipped with a computational basis {0, 1}. 3. Compound systems are described by tensor products of the component systems. It is here that the key phenomenon of entanglement arises, since the general form of a vector in H1 ⊗ H2 is n
αi · φi ⊗ ψi
i=1
Such a vector may encode correlations between the ﬁrst and second components of the system, and cannot simply be resolved into a pair of vectors in the component spaces. The adjoint to a linear map f : H1 → H2 is the linear map f † : H2 → H1 such that, for all φ ∈ H2 and ψ ∈ H1 , φ  f (ψ)H2 = f † (φ)  ψH1 .
268
Samson Abramsky and Bob Coecke
Unitary transformations are linear isomorphisms U : H1 → H2 such that U −1= U † : H2 → H1 . Note that all such transformations preserve the inner product since, for all φ, ψ ∈ H1 , U (φ)  U (ψ)H2 = (U † U )(φ)  ψH1 = φ  ψH1 . Selfadjoint operators are linear transformations M : H → H such that M = M † . 4. The basic data transformations are represented by unitary transformations. Note that all such data transformations are necessarily reversible. 5. The measurements which can be performed on the system are represented by selfadjoint operators. The act of measurement itself consists of two parts: 5a. The observer is informed about the measurement outcome, which is a value xi in the spectrum σ(M ) of the corresponding selfadjoint operator M . For convenience we assume σ(M ) to be nondegenerate (linearly independent eigenvectors have distinct eigenvalues). 5b. The state of the system undergoes a change, represented by the action of the projector Pi arising from the spectral decomposition M = x1 · P1 + . . . + xn · Pn In this spectral decomposition the projectors Pi : H → H are idempotent, selfadjoint, and mutually orthogonal Pi ◦ Pi = Pi
Pi = P†i
Pi ◦ Pj = 0, i = j.
This spectral decomposition always exists and is unique by the spectral theorem for selfadjoint operators. By our assumption that σ(M ) was nondegenerate each projector Pi has a onedimensional subspace of H as its ﬁxpoint set (which equals its image). The probability of xi ∈ σ(M ) being the actual outcome is given by the Born rule which does not depend on the value of xi but on Pi and the system state ψ, explicitly Prob(Pi , ψ) = ψ  Pi (ψ) . The status of the Born rule within our abstract setting will emerge in Section 8. The derivable notions of mixed states and nonprojective measurements will not play a signiﬁcant rˆ ole in this paper. The values x1 , . . . , xn are in eﬀect merely labels distinguishing the projectors P1 , . . . , Pn in the above sum. Hence we can abstract over them and think of a
Categorical Quantum Mechanics
269
measurement as a list of n mutually orthogonal projectors (P1 , . . . , Pn ) where n is the dimension of the Hilbert space. Although reallife experiments in many cases destroy the system (e.g. any measurement of a photon’s location destroys it) measurements always have the same shape in the quantum formalism. When distinguishing between ‘measurements which preserve the system’ and ‘measurements which destroy the system’ it would make sense to decompose a measurement explicitly in two components: • Observation consists of receiving the information on the outcome of the measurement, to be thought of as speciﬁcation of the index i of the outcomeprojector Pi in the above list. Measurements which destroy the system can be seen as ‘observation only’. • Preparation consists of producing the state Pi (ψ). In our abstract setting these arise naturally as the two ‘building blocks’ which are used to construct projectors and measurements. We now discuss some important quantum protocols which we chose because of the key rˆ ole entanglement plays in them — they involve both initially entangled states, and measurements against a basis of entangled states.
2.1
Quantum teleportation
The quantum teleportation protocol [Bennett et al., 1993] (see also [Coecke, 2003, §2.3&§3.3]) involves three qubits a, b and c and two spatial regions A (for “Alice”) and B (for “Bob”).
A
B
φ
x ∈ B2
Ux
MBell
time
φ
a
6
00+11
b
c
270
Samson Abramsky and Bob Coecke
Qubit a is in a state φ and located in A. Qubits b and c form an ‘EPRpair’, that is, their joint state is 00 + 11. We assume that these qubits are initially in B e.g. Bob created them. After spatial relocation so that a and b are located in A, while c is positioned in B, or in other words, “Bob sends qubit b to Alice”, we can start the actual teleportation of qubit a. Alice performs a Bellbase measurement on a and b at A, that is, a measurement such that each Pi projects on one of the onedimensional subspaces spanned by a vector in the Bell basis: b1 :=
00+11 √ 2
b2 :=
01+10 √ 2
b3 :=
00−11 √ 2
b4 :=
01−10 √ . 2
This measurement can be of the type ‘observation only’. Alice observes the outcome of the measurement and “sends these two classical bits (x ∈ B2 ) to Bob”. Depending on which classical bits he receives Bob then performs one of the unitary transformations & ' & ' & ' & ' 1 0 0 1 1 0 0 −1 β1 := β2 := β3 := β4 := 0 1 1 0 0 −1 1 0 on c — β1 , β2 , β3 are all selfinverse while β4−1 = −β4 . The ﬁnal state of c proves to be φ as well. (Because of the measurement, a no longer has this state — the information in the source has been ‘destroyed’ in transferring it to the target). Note that the state of a constitutes continuous data —an arbitrary pair of complex numbers (α, β) satisfying α2 + β2 = 1— while the actual physical data transmission only involved two classical bits. We will be able to derive this fact in our abstract setting. Teleportation is simply the most basic of a family of quantum protocols, and already illustrates the basic ideas, in particular the use of preparations of entangled states as carriers for information ﬂow, performing measurements to propagate information, using classical information to control branching behaviour to ensure the required behaviour despite quantum indeterminacy, and performing local data transformations using unitary operations. (Local here means that we apply these operations only at A or at B, which are assumed to be spatially separated, and not simultaneously at both). Since in quantum teleportation a continuous variable has been transmitted while the actual classical communication involved only two bits, besides this classical information ﬂow there has to exist some kind of quantum information ﬂow. The nature of this quantum ﬂow has been analyzed by one of the authors in [Coecke, 2003; Coecke, 2004], building on the joint work in [Abramsky and Coecke, 2003]. We recover those results in our abstract setting (see Subsection 3.5), which also reveals additional ‘ﬁne structure’. To identify it we have to separate it from the classical information ﬂow. Therefore we decompose the protocol into: 1. a tree with the operations as nodes, and with branching caused by the indeterminism of measurements; 2. a network of the operations in terms of the order they are applied and the subsystem to which they apply.
Categorical Quantum Mechanics
U00
U01
U10
00 01 10
271
U11
...
11
...
MBell 00+11
...
a
b
c
The nodes in the tree are connected to the boxes in the network by their temporal coincidence. Classical communication is encoded in the tree as the dependency of operations on the branch they are in. For each path from the root of the tree to a leaf, by ‘ﬁlling in the operations on the included nodes in the corresponding boxes of the network’, we obtain an entanglement network, that is, a network Ux Px
6 time
00+11
a
b
c
for each of the four values x takes. A component Px of an observation will be referred to as an observational branch. It will be these networks, from which we have removed the classical information ﬂow, that we will study in Subsection 3.5. (There is a clear analogy with the idea of unfolding a Petri net into its set of ‘processes’ [Petri, 1977]). The classical information ﬂow will be reintroduced in Section 5.
2.2
Logic gate teleportation
Logic gate teleportation [Gottesman and Chuang, 1999] (see also [Coecke, 2003]§3.3) generalizes the above protocol in that b and c are initially not necessarily an EPRpair but may be in some other (not arbitrary) entangled state Ψ. Due to this modiﬁcation the ﬁnal state of c is not φ but fΨ (φ) where fΨ is a linear map which depends on Ψ. As shown in [Gottesman and Chuang, 1999], when this construction is applied to the situation where a, b and c are each a pair of qubits rather than a single qubit, it provides a universal quantum computational primitive which is moreover faulttolerant [Shor, 1996] and enables the construction of a quantum computer based on single qubit unitary operations, Bellbase measurements and only one kind of prepared state (socalled GHZ states). The connection between Ψ, fΨ and the unitary corrections UΨ,x will emerge straightforwardly in our abstract setting.
272
Samson Abramsky and Bob Coecke
2.3 Entanglement swapping ˙ Entanglement swapping [Zukowski et al., 1993] (see also [Coecke, 2003]§6.2) is another modiﬁcation of the teleportation protocol where a is not in a state φ but is a qubit in an EPRpair together with an ancillary qubit d. The result is that after the protocol c forms an EPRpair with d. If the measurement on a and b is nondestructive, we can also perform a unitary operation on a, resulting in a and b also constituting an EPRpair. Hence we have ‘swapped’ entanglement: a
00+11
d
a
;
00+11
b
d 00+11
c
b
00+11
c
In this case the entanglement networks have the shape: Ux’
Ux Px
00+11
d
a
6 time
00+11
b
c
Why this protocol works will again emerge straightforwardly from our abstract setting, as will generalizations of this protocol which have a much more sophisticated compositional content (see Subsection 3.5).
3
COMPACT CLOSED CATEGORIES AND THE LOGIC OF ENTANGLEMENT
3.1 Monoidal Categories Recall that a symmetric monoidal category consists of a category C, a bifunctorial tensor − ⊗ − : C × C → C, a unit object I, and natural isomorphisms λA : A + I ⊗ A
ρA : A + A ⊗ I
αA,B,C : A ⊗ (B ⊗ C) + (A ⊗ B) ⊗ C σA,B : A ⊗ B + B ⊗ A which satisfy certain coherence conditions [MacLane, 1998].
Categorical Quantum Mechanics
273
Examples The following two examples are of particular importance and will recur through this section. 1. The category FdVecK , of ﬁnitedimensional vector spaces over a ﬁeld K and linear maps. The tensor product is the usual construction on vector spaces. The unit of the tensor is K, considered as a onedimensional vector space over itself. 2. The category Rel of sets and relations, with cartesian product as the ‘tensor’, and a oneelement set as the unit. Note that cartesian product is not the categorical product in Rel. The Logic of Tensor Product Tensor can express independent or concurrent actions (mathematically: bifunctoriality): A1 ⊗ A2 1 ⊗ f2 ? A1 ⊗ B2
f1 ⊗1
B1 ⊗ A2 1 ⊗ f2
?  B1 ⊗ B2 f1 ⊗ 1
But tensor is not a categorical product, in the sense that we cannot reconstruct an ‘element’ of the tensor from its components. This turns out to comprise the absence of diagonals and projections: Δ
π
i Ai A −→ A ⊗ A A1 ⊗ A2 −→ Cf. AA∧A A1 ∧ A2 Ai
Hence monoidal categories provide a setting for resourcesensitive logics such as Linear Logic [Girard, 1987]. NoCloning and NoDeleting are built in! Any symmetric monoidal category can be viewed as a setting for describing processes in a resource sensitive way, closed under sequential and parallel composition
3.2
The ‘miracle’ of scalars
A key step in the development of the categorical axiomatics for Quantum Mechanics was the recognition that the notion of scalar is meaningful in great generality — in fact, in any monoidal (not necessarily symmetric) category. Let (C, ⊗, I, λ, α, σ) be a monoidal category . We deﬁne a scalar in C to be a morphism s : I → I, i.e. an endomorphism of the tensor unit. EXAMPLE 1. In FdVecK , linear maps K → K are uniquely determined by the image of 1, and hence correspond biuniquely to elements of K ; composition corresponds to multiplication of scalars. In Rel, there are just two scalars, corresponding to the Boolean values 0, 1.
274
Samson Abramsky and Bob Coecke
The (multiplicative) monoid of scalars is then just the endomorphism monoid C(I, I). The ﬁrst key point is the elementary but beautiful observation by Kelly and Laplaza [Kelly and Laplaza, 1980] that this monoid is always commutative. LEMMA 2. C(I, I) is a commutative monoid Proof: ρI 
I 6
I ⊗ I ====== I ⊗ I 6
ρI 
I
s ⊗t
I⊗I
? I⊗I 6
1⊗t
t ? I
λI
 I
1⊗t
s⊗1
s
λ−1 I
λ−1 I
t ?  I 6
s⊗1
?  I ⊗ I ====== I ⊗ I
ρ−1 I
s  I
using the coherence equation λI = ρI .
The second point is that a good notion of scalar multiplication exists at this level of generality. That is, each scalar s : I → I induces a natural transformation sA : A
+
I ⊗A
s ⊗ 1A
I ⊗A
+
A.
with the naturality square A
sA A
f
f ? B
sB
?  B
We write s • f for f ◦ sA = sB ◦ f . Note that 1•f = f s • (t • f ) = (s ◦ t) • f (s • g) ◦ (t • f ) = (s ◦ t) • (g ◦ f ) (s • f ) ⊗ (t • g) = (s ◦ t) • (f ⊗ g) which exactly generalizes the multiplicative part of the usual properties of scalar multiplication. Thus scalars act globally on the whole category.
Categorical Quantum Mechanics
3.3
275
Compact Closure
A category C is ∗autonomous [Barr, 1979] if it is symmetric monoidal, and comes equipped with a full and faithful functor ( )∗ : Cop → C such that a bijection C(A ⊗ B, C ∗ ) + C(A, (B ⊗ C)∗ ) exists which is natural in all variables. Hence a ∗autonomous category is closed, with A B := (A ⊗ B ∗ )∗ . These ∗autonomous categories provide a categorical semantics for the multiplicative fragment of linear logic [Seely, 1989]. A compact closed category [Kelly, 1972] is a ∗autonomous category with a selfdual tensor, i.e. with natural isomorphisms uA,B : (A ⊗ B)∗ + A∗ ⊗ B ∗ It follows that
uI : I∗ + I .
A B + A∗ ⊗ B .
A very diﬀerent deﬁnition arises when one considers a symmetric monoidal category as a oneobject bicategory. In this context, compact closure simply means that every object A, qua 1cell of the bicategory, has a speciﬁed adjoint [Kelly and Laplaza, 1980]. DEFINITION 3 KellyLaplaza. A compact closed category is a symmetric monoidal category in which to each object A a dual object A∗ , a unit ηA : I → A∗ ⊗ A and a counit
A : A ⊗ A∗ → I
are assigned, in such a way that the diagram A
ρA
A⊗I
1A ? A
1A ⊗ ηA
A ⊗ (A∗ ⊗ A)
αA,A∗,A
λ−1 A
? I⊗A (A ⊗ A∗ ) ⊗ A A ⊗ 1A
and the dual one for A∗ both commute.
276
Samson Abramsky and Bob Coecke
Examples The symmetric monoidal categories (Rel, ×) of sets, relations and cartesian product and (FdVecK , ⊗) of ﬁnitedimensional vector spaces over a ﬁeld K, linear maps and tensor product are both compact closed. In (Rel, ×), we simply set X ∗ = X. Taking a onepoint set {∗} as the unit for ×, and writing R∪ for the converse of a relation R: ηX = ∪ X = {(∗, (x, x))  x ∈ X} . For (FdVecK , ⊗), we take V ∗ to be the dual space of linear functionals on V . The unit and counit in (FdVecK , ⊗) are ηV : K → V ∗ ⊗ V :: 1 →
i=n
e¯i ⊗ ei
V : V ⊗ V ∗ → K :: ei ⊗ e¯j → e¯j (ei )
and
i=1
¯i is the linear functional where n is the dimension of V , {ei }i=n i=1 is a basis of V and e in V ∗ determined by e¯j (ei ) = δij . DEFINITION 4. The name f and the coname f of a morphism f : A → B in a compact closed category are A∗⊗A 6 ηA
1A∗⊗f 
A∗ ⊗B 
f
f A⊗B ∗
I
I  6 B
 B ⊗B ∗ f ⊗1B ∗
For R ∈ Rel(X, Y ) we have R = {(∗, (x, y))  xRy, x ∈ X, y ∈ Y } and R = {((x, y), ∗)  xRy, x ∈ X, y ∈ Y } and for f ∈ FdVecK (V, W ) with (mij ) the matrix of f in bases {eVi }i=n i=1 and j=m } of V and W respectively {eW j j=1 i,j=n,m
f : K → V ∗ ⊗ W :: 1 →
mij · e¯Vi ⊗ eW j
i,j=1
and f : V ⊗ W ∗ → K :: eVi ⊗ e¯W j → mij . Given f : A → B in any compact closed category C we can deﬁne f ∗ : B ∗ → A∗
Categorical Quantum Mechanics
277
as ∗ λB
B∗
I ⊗ B∗
∗ ηA ⊗ 1B
A∗ ⊗ A ⊗ B ∗
f∗
1A∗ ⊗ f ⊗ 1B ∗
? A∗
ρ−1 A∗
A∗ ⊗ I 1A∗ ⊗ B
? A∗ ⊗ B ⊗ B ∗
This operation ( )∗ is functorial and makes Deﬁnition 3 coincide with the one given at the beginning of this section. It then follows by C(A ⊗ B ∗ , I) + C(A, B) + C(I, A∗ ⊗ B) that every morphism of type I → A∗ ⊗ B is the name of some morphism of type A → B and every morphism of type A ⊗ B ∗ → I is the coname of some morphism of type A → B. In the case of the unit and the counit we have ηA = 1A
and
A = 1A .
For R ∈ Rel(X, Y ) the dual is the converse, R∗ = R∪ ∈ Rel(Y, X), and for f ∈ FdVecK (V, W ), the dual is f ∗ : W ∗ → V ∗ :: φ → φ ◦ f . The following holds by general properties of adjoints and symmetry of the tensor [Kelly and Laplaza, 1980]§6. PROPOSITION 5. In a compact closed category C there is a natural isomorphism dA : A∗∗ + A and the diagrams A∗ ⊗ A
σA∗,A
1A∗ ⊗ d−1 A ? A∗ ⊗ A∗∗
A ⊗ A∗ A
A∗
?  I
I
ηA∗
dA ⊗ 1A∗
ηA ? A∗ ⊗ A
A∗∗ ⊗ A∗
?  A ⊗ A∗
σA∗,A
commute for all objects A of C. Graphical representation. Complex algebraic expressions for morphisms in symmetric monoidal categories can rapidly become hard to read. Graphical representations exploit twodimensionality, with the vertical dimension corresponding to composition and the horizontal to the monoidal tensor, and provide more intuitive presentations of morphisms. We depict objects by wires, morphisms by boxes
278
Samson Abramsky and Bob Coecke
with input and output wires, composition by connecting outputs to inputs, and the monoidal tensor by locating boxes sidebyside. We distinguish between an object and its dual in terms of directions of the wires. In particular, g ◦ f , f ⊗ g, f and f will respectively be depicted by
g f
f
f
g
f
Implicit in the use of this graphical notation is that we assume we are working in a strict monoidal category, in which the unit and associativity isomorphisms are identities. We can always do this because of the coherence theorem for monoidal categories [MacLane, 1998]. Similarly, strictness is assumed for the duality in compact closed categories: A∗∗ = A,
(A ⊗ B)∗ = A∗ ⊗ B ∗ ,
I∗ = I .
Pointers to references on diagrammatic representations and corresponding calculi are in Section 7.8.
3.4 Key lemmas The following Lemmas constitute the core of our interpretation of entanglement in compact closed categories. It was however observed by Radha Jagadeesan [2004] that they can be shown in arbitrary ∗autonomous categories using some of the results in [Cockett and Seely, 1997]. LEMMA 6 absorption. For A
 B
 C we have that
f
g
(1A∗⊗ g) ◦ f = g ◦ f .
Proof: Straightforward by Deﬁnition 4. In a picture,
g
= go f
f LEMMA 7 Compositionality. For A
 B
f
 C we have that
g
λ−1 C ◦ (f ⊗ 1C ) ◦ (1A ⊗ g) ◦ ρA = g ◦ f .
Categorical Quantum Mechanics
279
Proof: g◦f
 C 6
ρA
f ⊗ 1C
g
⊗

C

1A
∗⊗
ηB
⊗
1B
⊗
B∗
⊗
⊗
1C
f
1A
B ⊗ B∗ ⊗ C
⊗
⊗
f ⊗

B
1B ⊗ ηB
1B
∗
∗⊗
 B ⊗ B∗ ⊗ B
B ⊗ 1B
 I⊗B −
λB
Compact closedness

?
1
ρB

B
1B
?
B⊗I
g
1I ⊗ g
g
A ⊗ B∗ ⊗ B
f ⊗ 1I
 I⊗C 6

A ⊗ B∗ ⊗ C
B

1A ⊗ g
λ
Lemma 7 A⊗I
f
− C 1

A
B
 B
1B
The top trapezoid is the statement of the Lemma. The diagram uses bifunctoriality and naturality of ρ and λ. In a picture,
f
=
g
LEMMA 8 Compositional CUT. For A
g
 B
f
f
 C
g
 D we have that
h
(ρ−1 A ⊗ 1D ∗) ◦ (1A∗ ⊗ g ⊗1D ) ◦ (f ⊗ h) ◦ ρI = h ◦ g ◦ f .
Proof:
280
Samson Abramsky and Bob Coecke
h ◦ g ◦ f
 A∗⊗D 6
f
ηA ⊗1I
?
A∗⊗A ⊗I
6 1A∗⊗h
1I
1A∗⊗f ⊗1I
Lemma 7

A∗⊗B ⊗I
ρA∗ ⊗B
∗
D 1
λ
A∗⊗B 1A∗⊗(h◦g◦f )
f
ρA
⊗
∗
∗
⊗
A
1A
?
−
⊗
A∗⊗I⊗D

ηA
⊗
1A∗ ⊗g⊗1D
A∗⊗B ⊗C ∗⊗D
g)
f ⊗h 
1A ∗ ⊗( h◦
I⊗I
1A
ρ
− A 1
Lemma 8
⊗
ρI
1D
∗

I
A∗⊗A
 A∗⊗A
1A∗⊗A
The top trapezoid is the statement of the Lemma. The diagram uses Lemma 7 and naturality of ρ and λ. In a picture,
g
= f
h o go f
h
Discussion. On the right hand side of Lemma 7 we have g ◦ f , that is, we ﬁrst apply f and then g, while on the left hand side we ﬁrst apply the coname of g, and then the coname of f . In Lemma 8 there is a similar, seemingly ‘acausal’ inversion of the order of application, as g gets inserted between h and f . For completeness we add the following ‘backward’ absorption lemma, which again involves a reversal of the composition order. LEMMA 9 backward absorption. For C
 A
g
 B we have that
f
(g ∗ ⊗ 1A∗) ◦ f = f ◦ g. Proof: This follows by unfolding the deﬁnition of g ∗ , then using naturality of λA∗ , λI = ρI , and ﬁnally Lemma 8. In a picture,
Categorical Quantum Mechanics
g*
281
= f og
f
The obvious analogues of Lemma 6 and 9 for conames also hold.
3.5
Quantum information ﬂow in entanglement networks
We claim that Lemmas 6, 7 and 8 capture the quantum information ﬂow in the (logicgate) teleportation and entanglement swapping protocols. We shall provide a full interpretation of ﬁnitary quantum mechanics in Section 6 but for now the following rule suﬃces: • We interpret preparation of an entangled state as a name and an observational branch as a coname. For an entanglement network of teleportationtype shape, applying Lemma 7 yields U ◦ λ−1 C ◦ (f ⊗ 1) ◦ ((1 ⊗ g) ◦ ρA ) = U ◦ g ◦ f . In a picture,
U f
U
=
g
g f
We make the information ﬂow more explicit in the following version of the same picture:
U f
U
= g
g f
282
Samson Abramsky and Bob Coecke
Note that the quantum information seems to ﬂow ‘following the line’ while being acted on by the functions whose name or coname labels the boxes (and this fact remains valid for much more complex networks [Coecke, 2003]). Teleporting the input requires U ◦ g ◦ f = 1A — we assume all functions have type A → A. Logicgate teleportation of h : A → B requires U ◦ g ◦ f = h. We calculate this explicitly in Rel. For initial state x ∈ X after preparing S ⊆ {∗} × (Y × Z) we obtain {x} × {(y, z)  ∗ S(y, z)} as the state of the system. For observational branch R ⊆ (X × Y ) × {∗} we have that z ∈ Z is the output iﬀ R × 1Z receives an input (x, y, z) ∈ X × Y × Z such that (x, y)R ∗ . Since ∗ S(y, z) ⇔ ySz and (x, y)R ∗ ⇔ xRy we indeed obtain x(R; S)z. This illustrates that the compositionality is due to a mechanism of imposing constraints between the components of the tuples. In FdVecC the vector space of all linear maps of type V → W is V W and hence by V ∗ ⊗ W + V W we have a bijective correspondence between linear maps f : V → W and vectors Ψ ∈ V ∗ ⊗ W (see also [Coecke, 2003; Coecke, 2004]): 1 Ψf = √ · f (1) 2
√ f = 2 · Ψf − .
and
In particular we have for the Bell base: 1 bi = √ · βi (1) 2
√ βi = 2 · bi − .
and
Setting g := β1 = 1V , f := βi and U := βi−1 indeed yields βi−1 ◦ 1A ◦ βi = 1A , which expresses the correctness of the teleportation protocol along each branch. Setting g := h and f := βi for logicgate teleportation requires Ui to satisfy Ui ◦ h ◦ βi = h that is h ◦ βi = U † ◦ h (since U has to be unitary). Hence we have derived the laws of logicgate teleportation — one should compare this calculation to the size of the calculation in Hilbert space. Deriving the swapping protocol using Lemma 6 and Lemma 8 proceeds analogously to the derivation of the teleportation protocol.
γi1
β1i 1
γi βi 1
= 1
1 1
– the two triangles within the dashed line stand for γi ◦ βi . We obtain two distinct ﬂows due to the fact that a nondestructive measurement is involved.
Categorical Quantum Mechanics
γi1
283
β1i 1
γi
=
βi
How γi has to relate to βi such that they make up a true projector will be discussed in Section 6. For a general entanglement network of the swappingtype (without unitary correction and observational branching) by Lemma 8 we obtain the following ‘reduction’:
g rewrite
f
h
h o go f
This picture, and the underlying algebraic property expressed by Lemma 3.5, is in fact directly related to CutElimination in the logic corresponding to compactclosed categories. If one turns the above picture upsidedown, and interprets names as Axiomlinks and conames as Cutlinks, then one has a normalization rule for proofnets. This perspective is developed in [Abramsky and Duncan, 2006].
4 STRONGLY COMPACT CLOSED CATEGORIES AND 2DIMENSIONAL DIRAC NOTATION The key example In Section 3 we analysed the compact closed structure of FdVecK , where we took the dual of a vector space V to be the vector space of its linear functionals V ∗ . In the case that V is equipped with an inner product we can reﬁne this analysis. We discuss this for the key example of ﬁnitedimensional Hilbert spaces, i.e. ﬁnitedimensional complex vector spaces with a sesquilinear inner product: the inner product is linear in the second argument, and φ  ψ = ψ  φ which implies that it is conjugatelinear rather than linear in its ﬁrst argument. We organize these spaces into a category FdHilb, where the morphisms are linear maps. Note that we do not require morphisms to preserve the inner product.
284
Samson Abramsky and Bob Coecke
This category provides the basic setting for ﬁnitedimensional quantum mechanics and for quantum information and computation.1 In the setting of Hilbert spaces, we can replace the dual space by a more elementary construction. In a Hilbert space, each linear functional ψ¯ : H → C is witnessed by some ψ ∈ H such that ψ¯ = ψ  · . This however does not induce an isomorphism between H and H∗ , due to the conjugatelinearity of the inner ¯ product in its ﬁrst argument. This leads us to introduce the conjugate space H of a Hilbert space H: this has the same additive group of vectors as H, while the scalar multiplication and inner product are “twisted” by complex conjugation: ¯ •H φ α •H¯ φ := α
φ  ψH¯ := ψ  φH
¯ since H and H ¯ have the same orthornormal bases, and we We can deﬁne H∗ = H, can deﬁne the counit by ¯ → C :: φ ⊗ ψ → ψ  φH H : H ⊗ H which is indeed (bi)linear rather than sesquilinear! Note that ¯¯ = H, H
¯. A ⊗ B = A¯ ⊗ B
4.1 Why compact closure does not suﬃce Note that the categories FdHilb and FdVecC are equivalent! This immediately suggests that some additional categorical structure must be identiﬁed to reﬂect the rˆ ole of the inner product. A further reason for seeking additional categorical structure is to reﬂect the centrally important notion of adjoint in Hilbert spaces: f A  B f† A B
f φ  ψB = φ  f † ψA
This is not the same as the dual — the types are diﬀerent! In “degenerate” CCC’s in which A∗ = A, e.g. Rel or real innerproduct spaces, we have f ∗ = f † . In Hilbert spaces, the isomorphism A + A∗ is not linear, but conjugate linear : ¯ • φ  − λ • φ  − = λ and hence does not live in the category Hilb at all! 1 Much of quantum information is concerned with completely positive maps acting on density matrices. An account of this extended setting in terms of a general categorical construction within our framework is discussed in Section 7.
Categorical Quantum Mechanics
4.2
285
Solution: Strong Compact Closure
A key observation is this: the assignment H → H∗ on objects has a covariant functorial extension f → f∗ , which is essentially identity on morphisms; and then we can deﬁne f † = (f ∗ )∗ = (f∗ )∗ . Concretely, in terms of matrices ()∗ is transpose, ()∗ is complex conjugation, and the adjoint is the conjugate transpose. Each of these three operations can be expressed in terms of the other two. For example, f ∗ = (f † )∗ . All three of these operations are important in articulating the foundational structure of quantum mechanics. All three can be presented at the abstract level as functors, as we shall now show.
4.3
Axiomatization of Strong Compact Closure
We shall adopt the most concise and elegant axiomatization of strongly compact closed categories, which takes the adjoint as primitive, following [Abramsky and Coecke, 2005]. It is convenient to build the deﬁnition up in several stages, as in [Selinger, 2007]. DEFINITION 10. A dagger category is a category C equipped with an identityonobjects, contravariant, strictly involutive functor f → f † : 1† = 1,
(g ◦ f )† = f † ◦ g † ,
f †† = f .
We deﬁne an arrow f : A → B in a dagger category to be unitary if it is an isomorphism such that f −1 = f † . An endomorphism f : A → A is selfadjoint if f = f †. DEFINITION 11. A dagger symmetric monoidal category (C, ⊗, I, λ, ρ, α, σ, †) combines dagger and symmetric monoidal structure, with the requirement that the natural isomorphisms λ, ρ, α, σ are componentwise unitary, and moreover that † is a strong monoidal functor: (f ⊗ g)† = f † ⊗ g † . Finally we come to the main deﬁnition. DEFINITION 12. A strongly compact closed category is a dagger symmetric monoidal category which is compact closed, and such that the following diagram commutes: ηA 
I
A∗ ⊗ A
A
†
σA∗ ,A 
? A ⊗ A∗
286
Samson Abramsky and Bob Coecke
This implies that the counit is deﬁnable from the unit and the adjoint: † ◦ σA,A∗ A = ηA
and similarly the unit can be deﬁned from the counit and the adjoint. Furthermore, it is in fact possible to replace the two commuting diagrams required in the deﬁnition of compact closure by one. We refer to [Abramsky and Coecke, 2005] for the details. DEFINITION 13. In any strongly compact closed category C, we can deﬁne a covariant monoidal functor A → A∗ ,
f : A → B → f∗ = (f † )∗ : A∗ → B ∗ .
Examples Our central example is of course FdHilb. Any compact closed category such as Rel, in which ( )∗ is the identity on objects, is trivially strongly compact closed (we just take f † = f ∗ ). Note that in this case f∗ = f ∗∗ = f . Thus in Rel the adjoint is relational converse. The category of ﬁnitedimensional real inner product spaces and linear maps, with A = A∗ , oﬀers another example of this situation. A construction of free strongly compact closed categories over dagger categories is given in [Abramsky, 2005]. Scalars Selfadjoint scalars s = s† in strongly compact closed categories are of special interest. In the case of FdHilb, these are the positive reals R+ . The passage from s to ss† , which is selfadjoint, will track the passage in quantum mechanics from amplitudes to probabilities.
4.4 Inner Products and Dirac Notation With the adjoint available, it is straightforward to interpret Dirac notation — the indispensable everyday notation of quantum mechanics and quantum information. A ket is simply an arrow ψ : I → A, which we can write as ψ for emphasis. We think of kets as states, of a given type of system A. The corresponding bra will then be ψ † : A → I, which we can think of as a costate. Example In FdHilb, a linear map f : C → H can be identiﬁed with the vector f (1) = ψ ∈ H: by linearity, all other values of f are determined by ψ. Even better, we can identify f with its image, which is the ray or onedimensional subspace of H spanned by ψ — the proper notion of (pure) state of a quantum system. DEFINITION 14. Given ψ, φ : I → A we deﬁne their abstract inner product ψ  φ as ψ † ◦ φ : I −→ I .
Categorical Quantum Mechanics
287
Note that this is a scalar, as it should be. In FdHilb, this deﬁnition coincides with the usual inner product. In Rel, for x, y ⊆ {∗} × X: x  y = 1I , x ∩ y = ∅
x  y = 0I , x ∩ y = ∅.
and
We now show that two of the basic properties of adjoints in FdHilb hold in generality in the abstract setting. PROPOSITION 15. For ψ : I → A, φ : I → B and f : B → A we have f † ◦ ψ  φB = ψ  f ◦ φA . Proof: f † ◦ ψ  φ = (f † ◦ ψ)† ◦ φ = ψ † ◦ f ◦ φ = ψ  f ◦ φ.
PROPOSITION 16. Unitary morphisms U : A → B preserve the inner product, that is for all ψ, φ : I → A we have U ◦ ψ  U ◦ φB = ψ  φA . Proof: By Proposition 15, U ◦ ψ  U ◦ φB = U † ◦ U ◦ ψ  φA = ψ  φA . Finally, we show how the inner product can be deﬁned in terms of the ‘complex conjugate’ functor ()∗ . PROPOSITION 17. For ψ, φ : I → A we have: ψ  φA = I
ρI 
I⊗I
1I ⊗uI
I ⊗ I∗
φ⊗ψA ∗ A ⊗ A∗ I.
Proof: Since uI = ρ−1 I∗ ◦ ηI by naturality of ρ we have ηI = ρI∗ ◦ ρ−1 I∗ ◦ ηI = ρI∗ ◦ uI = (uI ⊗ 1I ) ◦ ρI where we use ρ−1 = ρ† and similarly we obtain I = ρ†I ◦ (1I ⊗ u†I ). Hence by 1I = u†I ◦ uI and the analogues to Lemmas 6 and 9 for the counit we obtain ψ † ◦ φ = ρ†I ◦ ((ψ † ◦ φ) ⊗ 1I ) ◦ ρI = I ◦ (ψ † ⊗ 1I∗ ) ◦ (φ ⊗ 1I∗ ) ◦ †I = ψ † ◦ (φ ⊗ 1I∗ ) ◦ †I = I ◦ (1I ⊗ ψ∗ ) ◦ (φ ⊗ 1I∗ ) ◦ †I which is equal to I ◦ (φ ⊗ ψ∗ ) ◦ (1I ⊗ uI ) ◦ ρI .
4.5
Dissection of the bipartite projector
Projectors are a basic building block in the von Neumannstyle foundations of quantum mechanics, and in standard approaches to quantum logic. It is a notable feature of our approach that we are able, at the abstract level of strongly compact
288
Samson Abramsky and Bob Coecke
closed categories, to delineate a ﬁnestructure of bipartite projectors, which can be applied directly to the analysis of information ﬂow in quantum protocols. We deﬁne a projector on an object A in a strongly compact closed category to be an arrow P : A → A which is idempotent and selfadjoint: P2 = P,
P = P† .
PROPOSITION 18. Suppose we have a state ψ : I → A which is normalized, meaning ψ  ψ = 1I . Then the ‘ketbra’ ψψ = ψ ◦ ψ † : A → A is a projector. Proof: Selfadjointness is clear. For idempotence: ψ ◦ ψ † ◦ ψ ◦ ψ † = ψ  ψ • ψ ◦ ψ † = ψ ◦ ψ † . We now want to apply this idea in a more reﬁned form to a state ψ : I → A∗ ⊗ B of a compound system. Note that, by MapState duality: C(I, A∗ ⊗ B) ≡ C(A, B) any such state ψ corresponds biuniquely to the name of a map f : A → B, i.e. ψ = f . This arrow witnesses an information ﬂow from A to B, and we will use this to expose the information ﬂow inherent in the corresponding projector. Explicitly, we deﬁne Pf := f ◦ (f )† = f ◦ f∗ : A∗ ⊗ B → A∗ ⊗ B , that is, we have an assignment P : C(I, A∗ ⊗ B) −→ C(A∗ ⊗ B, A∗ ⊗ B) :: Ψ → Ψ ◦ Ψ† from bipartite elements to bipartite projectors. Note that the strong compact closed structure is essential in order to deﬁne Pf as an endomorphism. We can normalize these projectors Pf by considering sf • Pf for sf := (f∗ ◦ f )−1 (provided this inverse exists in C(I, I)), yielding (sf • Pf ) ◦ (sf • Pf ) = sf • (f ◦ (sf • (f∗ ◦ f )) ◦ f∗ ) = sf • Pf , and also (sf • Pf ) ◦ f = f
and
f∗ ◦ (sf • Pf ) = f∗ .
A picture corresponding to this decomposed bipartite projector is:
Categorical Quantum Mechanics
289
f
f†
4.6
Trace
Another essential mathematical instrument in quantum mechanics is the trace of a linear map. In quantum information, extensive use is made of the more general notion of partial trace, which is used to trace out a subsystem of a compound system. A general categorical axiomatization of the notion of partial trace has been given by Joyal, Street and Verity [Joyal, Street and Verity, 1996]. A trace in a symmetric monoidal category C is a family of functions TrU A,B : C(A ⊗ U, B ⊗ U ) −→ C(A, B) for objects A, B, U of C, satisfying a number of axioms, for which we refer to [Joyal, Street and Verity, 1996]. This specializes to yield the total trace for endomorphisms by taking A = B = I. In this case, Tr(f ) = TrU I,I (f ) : I → I is a scalar. Expected properties such as the invariance of the trace under cyclic permutations Tr(g ◦ f ) = Tr(f ◦ g) follow from the general axioms. Any compact closed category carries a canonical (in fact, a unique) trace. The deﬁnition can be given slightly more elegantly in the strongly compact closed case. For an endomorphism f : A → A, the total trace is deﬁned by Tr(f ) = A ◦ (f ⊗ 1A∗ ) ◦ †A . More generally, if f : A ⊗ C → B ⊗ C, TrC A,B (f ) : A → B is deﬁned to be: A
ρA 
A⊗I
∗ 1⊗†C C  A⊗C ⊗C ∗ f ⊗1C B ⊗C ⊗C ∗ 1⊗B ⊗I
ρ−1 B
 B.
These deﬁnitions give rise to the standard notions of trace and partial trace in FdHilb.
290
Samson Abramsky and Bob Coecke
5
BIPRODUCTS, BRANCHING AND MEASUREMENTS
As we have seen, many of the basic ingredients of quantum mechanics are present in strongly compact closed categories. What is lacking is the ability to express the probabilistic branching arising from measurements, and the information ﬂows from quantum to classical and back. We shall ﬁnd this ﬁnal piece of expressive power in a rather standard piece of additional categorical structure, namely biproducts.
5.1 Biproducts Biproducts have been studied as part of the structure of Abelian categories. For further details, and proofs of the general results we shall cite in this subsection, see e.g. [Mitchell, 1965; MacLane, 1998]. Recall that a zero object in a category is one which is both initial and terminal. If 0 is a zero object, there is an arrow  0
0A,B : A
 B
between any pair of objects A and B. Let C be a category with a zero object and binary products and coproducts. Any arrow A1 with injections qi : Ai → A1 written uniquely as a matrix
( *
f
A2
A1
)
A2
A2 and projections pj : A1 &
f11 f12
f21 f22
+
A2 → Aj can be
'
where fij := pj ◦ f ◦ qi : Ai → Aj . If the arrow & ' 1 0 0 1 is an isomorphism for all A1 , A2 , then we say that C has biproducts, and write A ⊕ B for the biproduct of A and B. PROPOSITION 19 Semiadditivity. If C has biproducts, then we can deﬁne an operation of addition on each homset C(A, B) by A Δ ? A⊕A
f + gB 6 ∇  B⊕B f ⊕g
Categorical Quantum Mechanics
291
for f, g : A → B, where Δ = 1A , 1A and ∇ = [1B , 1B ] are respectively the diagonal and codiagonal. This operation is associative and commutative, with 0AB as an identity. Moreover, composition is bilinear with respect to this additive structure. Thus C is enriched over abelian monoids. Because of this automatic enrichment of categories with biproducts over abelian monoids, we say that such a category is semiadditive. PROPOSITION 20. If C has biproducts, ,k=n we can choose projections p1 , . . ., pn and injections q1 , . . ., qn for each k=1 Ak satisfying pj ◦ qi = δij
and
k=n
qk ◦ pk = 1Lk Ak
k=1
where δii = 1Ai , and δij = 0Ai ,Aj , i = j.
5.2
Strongly compact closed categories with biproducts
We now come to the full mathematical structure we shall use as a setting for ﬁnitary quantum mechanics: namely strongly compact closed categories with biproducts. A ﬁrst point is that, because of the strongly selfdual nature of compact closed categories, weaker assumptions suﬃce in order to guarantee the presence of biproducts. The following elegant result is due to Robin Houston [2006], and was in fact directly motivated by [Abramsky and Coecke, 2004], the precursor to the present article. THEOREM 21. Let C be a monoidal category with ﬁnite products and coproducts, and suppose that for every object A of C, the functor A ⊗ − preserves products and the functor − ⊗ A preserves coproducts. Then C has ﬁnite biproducts. Because a compact closed category is closed and selfdual, the existence of products implies that of coproducts, and vice versa, and the functor − ⊗ A is a left adjoint and hence preserves coproducts. Moreover, since A∗ B + A∗∗ ⊗ B + A ⊗ B, the functor A ⊗ − is a right adjoint and preserves products. Hence this result specializes to the following: PROPOSITION 22. If C is a compact closed category with either products or coproducts, then it has biproducts, and hence is semiadditive. Examples There are many examples of compact closed categories with biproducts: the category of relations for a regular category with stable disjoint coproducts; the category of ﬁnitely generated projective modules over a commutative ring; the category of ﬁnitely generated free semimodules over a commutative semiring; and the category of free semimodules over a complete commutative semiring are all semiadditive compact closed categories. Examples have also arisen in a Computer Science context in the ﬁrst author’s work on Interaction Categories [Abramsky,
292
Samson Abramsky and Bob Coecke
Gay and Nagarajan, 1995]. Compact closed categories with biproducts with additional assumptions, in particular that the category is abelian, have been studied in the mathematical literature on Tannakian categories [Deligne, 1990]. In the case of strongly compact closed categories, we need a coherence condition between the dagger and the biproduct structure. We say that a category is strongly compact closed with biproducts if we can choose biproduct structures pi , qi as in Proposition 20 such that p†i = qi for i = 1, . . . , n. PROPOSITION 23. If C is strongly compact closed with biproducts, then k=n k=1
p†k ◦ pk =
k=n
qk ◦ qk† = 1Lk Ak .
k=1
Moreover, there are natural isomorphisms νA,B : (A ⊕ B)∗ + A∗ ⊕ B ∗
and
νI : 0∗ + 0 ,
and ( )† preserves biproducts and hence is additive: (f ⊕ g)† = f † ⊕ g † ,
(f + g)† = f † + g †
and
0†A,B = 0B,A .
Examples Examples of semiadditive strongly compact closed categories are the category (Rel, ×, +), where the biproduct is the disjoint union, and the category (FdHilb, ⊗, ⊕), where the biproduct is the direct sum. Distributivity and classical information ﬂow As we have already seen, in a compact closed category with biproducts, tensor distributes over the biproduct. This abstractseeming observation in fact plays a crucial rˆ ole in the representation of classical information ﬂow. To understand this, consider a quantum system A ⊗ B, composed from subsystems A(lice) and B(ob). Now suppose that Alice performs a local measurement, which we will represent as resolving her part of the system into say A1 ⊕ A2 . Here the biproduct is used to represent the diﬀerent branches of the measurement. At this point, by the functorial properties of ⊕, Alice can perform actions f1 ⊕f2 , which depend on which branch of the measurement has been taken. The global state of the system is (A1 ⊕ A2 ) ⊗ B, and as things stand Bob has no access to this measurement outcome. However, under distributivity we have (A1 ⊕ A2 ) ⊗ B + (A1 ⊗ B) ⊕ (A2 ⊗ B) which corresponds to propagating the classical information as to the measurement outcome ‘outwards’, so that it is now accessible to Bob, who can perform an action depending on this outcome, of the form 1A ⊗ (g1 ⊕ g2 ). We shall record distributivity in an explicit form for future use.
Categorical Quantum Mechanics
293
PROPOSITION 24 Distributivity of ⊗ over ⊕. In any monoidal closed category there is a right distributivity natural isomorphism τA,B,C : A ⊗ (B ⊕ C) + (A ⊗ B) ⊕ (A ⊗ C), which is explicitly deﬁned as τA,·,· := 1A ⊗ p1 , 1A ⊗ p2
−1 τA,·,· := [1A ⊗ q1 , 1A ⊗ q2 ] .
and
A left distributivity isomorphism υA,B,C : (A ⊕ B) ⊗ C + (A ⊗ C) ⊕ (A ⊗ C) can be deﬁned similarly. Semiring of scalars. In a strongly compact closed category with biproducts, the scalars form a commutative semiring. Moreover, scalar multiplication satisﬁes the usual additive properties (s1 + s2 ) • f = s1 • f + s2 • f,
0•f =0
as well as the multiplicative ones. For Hilbert spaces, this commutative semiring is the ﬁeld of complex numbers. In Rel the commutative semiring of scalars is the Boolean semiring {0, 1}, with disjunction as sum. Matrix calculus. We can write any arrow of the form f : A ⊕ B → C ⊕ D as a matrix & C,D ' p1 ◦ f ◦ q1A,B pC,D ◦ f ◦ q2A,B 1 . Mf := pC,D ◦ f ◦ q1A,B pC,D ◦ f ◦ q2A,B 2 2 The sum f + g of such morphisms corresponds to the matrix sum Mf + Mg and composition g ◦ f corresponds to matrix multiplication Mg · Mf . Hence categories with biproducts admit a matrix calculus.
5.3
Spectral Decompositions
We deﬁne a spectral decomposition of an object A to be a unitary isomorphism U :A→
i=n 
Ai .
i=1
(Here the ‘spectrum’ is just the set of indices 1, . . . , n). Given a spectral decomposition U , we deﬁne morphisms ψj := U † ◦ qj : Aj → A
πj := ψj† = pj ◦ U : A → Aj ,
and
diagramatically
qj
i=1
? i=n Ai
U
ψj A
Aj
pj
πj
?  Aj
294
Samson Abramsky and Bob Coecke
and ﬁnally projectors Pj := ψj ◦ πj : A → A . These projectors are selfadjoint P†j = (ψj ◦ πj )† = πj† ◦ ψj† = ψj ◦ πj = Pj idempotent and orthogonal A ◦ Pi . Pi ◦ Pj = ψi ◦ πi ◦ ψj ◦ πj = ψi ◦ δij ◦ πj = δij
Moreover, they yield a resolution of the identity: i=n
Pi
=
i=1
i=n
ψ i ◦ πi =
i=1
=
i=n
U † ◦ q i ◦ pi ◦ U
i=1
i=n U† ◦ ( qi ◦ pi ) ◦ U = U −1 ◦ 1Li Ai ◦ U = 1A . i=1
5.4 Bases and dimension
,i=n Writing n · X for types of the shape i=1 X it follows by selfduality of the tensor unit I that ∗ −1 ◦ (n · uI ) : n · I + (n · I) . νI,...,I A basis for an object A is a unitary isomorphism base : n · I → A . Given bases baseA and baseB for objects A and B respectively we can deﬁne the matrix (mij ) of any morphism f : A → B in those two bases as the matrix of base†B ◦ f ◦ baseA : nA · I → nB · I . PROPOSITION 25. Given f : A → B, baseA : nA · I → A and baseB : nB · I → A ) of f † in these bases is the conjugate transpose of the matrix (mij ) the matrix (mij of f . Proof: mij = pi ◦ base†A ◦ f † ◦ baseB ◦ qj = (pj ◦ base†B ◦ f ◦ baseA ◦ qi )† = m†ji .
If in addition to the assumptions of Proposition 15 and Proposition 16 there exist bases for A and B, we can prove converses to both of them. PROPOSITION 26. If there exist bases for A and B then f : A → B is the adjoint to g : B → A if and only if f ◦ ψ  φB = ψ  g ◦ φA
Categorical Quantum Mechanics
295
for all ψ : I → A and φ : I → B. ) the matrix of g in the given bases. Proof: Let (mij ) be the matrix of f † and (mij
mij
= = = = = = =
pi ◦ base†A ◦f † ◦ baseB ◦ qj f ◦ baseA ◦ qi  baseB ◦ qj B f ◦ ψ  φB ψ  g ◦ φA baseA ◦ qi  g◦ baseB ◦ qj A pi ◦ base†A ◦g ◦ baseB ◦ qj mij .
Hence the matrix elements of g and f † coincide so g and f † are equal. The converse is Proposition 15. PROPOSITION 27. If there exist bases for A and B then a morphism U : A → B is unitary if and only if it preserves the inner product, that is for all ψ, φ : I → A we have U ◦ ψ  U ◦ φB = ψ  φA . Proof: We have U −1◦ ψ  φA = U ◦ U −1◦ ψ  U ◦ φB = ψ  U ◦ φB and hence by Proposition 26, U † = U −1 . The converse is given by Proposition 16. Note also that when a basis is available we can assign to ψ † : A → I and φ : I → A matrices ⎞ ⎛ φ1 † ⎜ .. ⎟ and ψ1 · · · ψn† ⎝ . ⎠ φn respectively, and the inner product becomes ⎛ ψ  φ =
ψ1†
···
ψn†
⎞ φ1 i=n ⎜ . ⎟ ψi† ◦ φi . ⎝ .. ⎠ = i=1 φn
Dimension Interestingly, two diﬀerent notions of dimension arise in our setting. We assign an integer dimension dim(A) ∈ N to an object A provided there exists a base base : dim(A) · I → A . Alternatively, we introduce the scalar dimension as dims (A) := Tr(1A ) = A ◦ †A ∈ C(I, I). We also have: dims (I) = 1I
dims (A∗ ) = dims (A)
dims (A ⊗ B) = dims (A)dims (B)
296
Samson Abramsky and Bob Coecke
In FdVecK these notions of dimension coincide, in the sense that dims (V ) is multiplication with the scalar dim(V ). In Rel the integer dimension corresponds to the cardinality of the set, and is only welldeﬁned for ﬁnite sets, while dims (X) always exists; however, dims (X) can only take two values, 0I and 1I , and the two notions of dimension diverge for sets of cardinality greater than 1.
5.5 Towards a representation theorem As the results in this section have shown, under the assumption of biproducts we can replicate many of the familiar linearalgebraic calculations in Hilbert spaces. One may wonder how far we really are from Hilbert spaces. The deep results by Deligne [1990] and DoplicherRoberts [1989] on Tannakian categories, the latter directly motivated by algebraic quantum ﬁeld theory, show that under additional assumptions, in particular that the category is abelian as well as compact closed, we obtain a representation into ﬁnitedimensional modules over the ring of scalars. One would like to see a similar result in the case of strongly compact closed categories with biproducts, with the conclusion being a representation into innerproduct spaces. 6 ABSTRACT QUANTUM MECHANICS: AXIOMATICS AND QUANTUM PROTOCOLS We can identify the basic ingredients of ﬁnitary quantum mechanics in any semiadditive strongly compact closed category. 1. A state space is represented by an object A. 2. A basic variable (‘type of qubits’) is a state space Q with a given unitary isomorphism baseQ : I ⊕ I → Q which we call the computational basis of Q. By using the isomorphism n · I + (n · I)∗ described in Section 5, we also obtain a computational basis for Q∗ . 3. A compound system for which the subsystems are described by A and B respectively is described by A ⊗ B. If we have computational bases baseA and baseB , then we deﬁne baseA⊗B := (baseA ⊗ baseB ) ◦ d−1 nm where dnm : n · I ⊗ m · I + (nm) · I is the canonical isomorphism constructed using ﬁrst the left distributivity isomorphism υ, and then the right distributivity isomorphism τ , to give the usual lexicographicallyordered computational basis for the tensor product.
Categorical Quantum Mechanics
297
4. Basic data transformations are unitary isomorphisms. 5a. A preparation in a state space A is a morphism ψ : I → A for which there exists a unitary U : I ⊕ B → A such that ψ A

I
U
q1 ? I⊕B commutes. 5b. Consider a spectral decomposition
U :A→
i=n 
Ai
i=1
with associated projectors Pj . This gives rise to the nondestructive measurement Pi i=n i=1 : A → n · A. The projectors Pi : A → A for i = 1, . . . , n are called the measurement branches. This measurement is nondegenerate if Ai = I for all i = 1, . . . , n. In this case we refer to U itself as a destructive measurement or observation. The morphisms πi = pi ◦ U : A → I for i = 1, . . . , n are called observation branches. Note that the type of a nondestructive measurement makes it explicit that it is an operation which involves a nondeterministic transition (by contrast with the standard Hilbert space quantum mechanical formalism). 6a. Explicit biproducts represent the branching arising from the indeterminacy of measurement outcomes. Hence an operation f acting on an explicit biproduct A ⊕ B should itself be an explicit biproduct, i.e. we want f = f1 ⊕ f2 : A ⊕ B → C ⊕ D , for f1 : A → C and f2 : B → D. The dependency of fi on the branch it is in captures local classical communication. The full force of nonlocal classical communication is enabled by Proposition 24. 6b. Distributivity isomorphisms represent nonlocal classical communication.
298
Samson Abramsky and Bob Coecke
To see this, suppose e.g. that we have a compound system Q ⊗ A, and we (nondestructively) measure the qubit in the ﬁrst component, obtaining a new system state described by (Q ⊕ Q) ⊗ A. At this point, we know ‘locally’, i.e. at the site of the ﬁrst component, what the measurement outcome is, but we have not propagated this information to the rest of the system A. However, after applying the distributivity isomorphism (Q ⊕ Q) ⊗ A + (Q ⊗ A) ⊕ (Q ⊗ A) the information about the outcome of the measurement on the ﬁrst qubit has been propagated globally throughout the system, and we can perform operations on A depending on the measurement outcome, e.g. (1Q ⊗ U0 ) ⊕ (1Q ⊗ U1 ) where U0 , U1 are the operations we wish to perform on A in the event that the outcome of the measurement we performed on Q was 0 or 1 respectively.
6.1 The Born rule We now show how the Born rule, which is the key quantitative feature of quantum mechanics, emerges automatically from our abstract setting. ,i=n For a preparation ψ : I → A and spectral decomposition U : A → i=1 Ai , with corresponding nondestructive measurement Pi i=n i=1 : A → n · A , we can consider the protocol I
ψ A
Pi i=n i=1 
n · A.
We deﬁne scalars Prob(Pi , ψ) := ψ  Pi  ψ = ψ † ◦ Pi ◦ ψ . PROPOSITION 28. With notation as above, Prob(Pi , ψ) = (Prob(Pi , ψ))† and
i=n
Prob(Pi , ψ) = 1 .
i=1
Hence we think of the scalar Prob(Pj , ψ) as ‘the probability of obtaining the j’th outcome of the measurement Pi i=n i=1 on the state ψ’. Proof: From the deﬁnitions of preparation and the projectors, there are unitaries U , V such that Prob(Pi , ψ) = (V ◦ q1 )† ◦ U † ◦ qi ◦ pi ◦ U ◦ V ◦ q1
Categorical Quantum Mechanics
299
for each i. Hence i=n
Prob(Pi , ψ)
i=n
=
i=1
p1 ◦ V † ◦ U † ◦ q i ◦ pi ◦ U ◦ V ◦ q 1
i=1 †
†
p1 ◦ V ◦ U ◦
=
n
q i ◦ pi ◦ U ◦ V ◦ q 1
i=1
p1 ◦ V
=
−1
◦U
−1
◦ 1n·I ◦ U ◦ V ◦ q1 = p1 ◦ q1 = 1I .
Moreover, since by deﬁnition Pj = πj† ◦ πj , we can rewrite the Born rule expression as Prob(Pj , ψ) = ψ † ◦ Pj ◦ ψ = ψ † ◦ πj† ◦ πj ◦ ψ = (πj ◦ ψ)† ◦ πj ◦ ψ = s†j ◦ sj for some scalar sj ∈ C(I, I). Thus sj can be thought of as the ‘probability amplitude’ giving rise to the probability s†j ◦ sj , which is of course selfadjoint. If we consider the protocol ψ A
I
πi i=n i=1 
n · I.
which involves an observation πi i=n i=1 , then these scalars sj correspond to the branches I
ψ A
πj I.
We now turn to the description of the quantum protocols previously discussed in Section 2 within our framework. In each case, we shall give a complete description of the protocol, including the quantumtoclassical information ﬂows arising from measurements, and the subsequent classicaltoquantum ﬂows corresponding to the classical communications and the actions depending on these performed as steps in the protocols. We shall in each case verify the correctness of the protocol, by proving that a certain diagram commutes. Thus these case studies provide evidence for the expressiveness and eﬀectiveness of the framework. Our general axiomatic development allows for considerable generality. The standard von Neumann axiomatization ﬁts Quantum Mechanics perfectly, with no room to spare. Our basic setting of strongly compact closed categories with biproducts is general enough to allow very diﬀerent models such as Rel, the category of sets and relations. When we consider speciﬁc protocols such as teleportation, a kind of ‘Reverse Arithmetic’ (by analogy with Reverse Mathematics [Simpson, 1999]) arises. That is, we can characterize what requirements are placed on the semiring of scalars C(I, I) (where I is the tensor unit) in order for the protocol to be realized. This is often much less than requiring that this be the ﬁeld of complex numbers, but in the speciﬁc cases which we shall consider, the requirements are suﬃcient to exclude Rel.
300
Samson Abramsky and Bob Coecke
6.2 Quantum teleportation DEFINITION 29. A teleportation base is a scalar s together with a morphism prebaseT : 4 · I → Q∗ ⊗ Q such that: • baseT := s • prebaseT is unitary. • the four maps βj : Q → Q, where βj is deﬁned by βj := prebaseT ◦ qj , are unitary. • 2s† s = 1. The morphisms s • βj are the base vectors of the teleportation base. A teleportation base is a Bell base when the Bell base maps β1 , β2 , β3 , β4 : Q → Q satisfy2 β1 = 1Q
⊕ β2 = σQ
β3 = β3†
⊕ β4 = σQ ◦ β3
where ⊕ ⊕ := baseQ ◦ σI,I ◦ base−1 σQ Q .
A teleportation base deﬁnes a teleportation observation ∗ s† • βi i=4 i=1 : Q ⊗ Q → 4 · I .
To emphasize the identity of the individual qubits we label the three copies of Q we shall consider as Qa , Qb , Qc . We also use labelled identities, e.g. 1bc : Qb → Qc , and labelled Bell bases. Finally, we introduce Δ4ac := s† s • 1ac i=4 i=1 : Qa → 4 · Qc as the labelled, weighted diagonal. This expresses the intended behaviour of teleportation, namely that the input qubit is propagated to the output along each branch of the protocol, with ‘weight’ s† s, corresponding to the probability amplitude for that branch. Note that the sum of the corresponding probabilities is 4(s† s)† s† s = (2s† s)(2s† s) = 1 . 2 This choice of axioms is suﬃcient for our purposes. One might prefer to axiomatize a notion of Bell base such that the corresponding Bell base maps are exactly the Pauli matrices — note that this would introduce a coeﬃcient i in β4 .
Categorical Quantum Mechanics
301
THEOREM 30. The following diagram commutes. Qa =============== Qa ρa
import unknown state
? Qa ⊗ I 1a ⊗ (s • 1bc)
produce EPRpair
? Qa ⊗ (Q∗b ⊗ Qc ) αa,b,c
spatial delocation
? (Qa ⊗ Q∗b ) ⊗ Qc
Δ4ac
s† • βiabi=4 i=1 ⊗1c
teleportation observation
? (4 · I) ⊗ Qc
4 · λ−1 ◦ υc c
classical communication
? 4 · Qc ,i=4
c −1 i=1 (βi )
unitary correction
? ? 4 · Qc ============ 4 · Qc The righthandside of the above diagram is our formal description of the teleportation protocol; the commutativity of the diagram expresses the correctness of the protocol. Hence any strongly compact closed category with biproducts admits quantum teleportation provided it contains a teleportation base. If we do a Bellbase observation then the corresponding unitary corrections are βi−1= βi for i ∈ {1, 2, 3}
and
⊕ β4−1= β3 ◦ σQ .
Proof: For each j ∈ {1, 2, 3, 4} we have a commutative diagram of the form below. The top trapezoid is the statement of the Theorem. We ignore the scalars – which
302
Samson Abramsky and Bob Coecke
cancel out against each other – in this proof. 1ac i

4 · Qc
Qa ⊗I
1a ⊗1bc
β a
λc
⊗1
c
)
c
i
c pI⊗Q j
pIj ⊗1c
 ?  I ⊗ Qc

1I⊗Qc
?
I ⊗ Qc
c pQ j

λ−1 c
c pQ j
? Qc
c
a j c

λ−1 c
j b
(β
β
−
)
j

1
? Qa
(β
I β abi ⊗1c 4 · λ−1 c i ∗  (4 · I) ⊗ Qcpi ⊗1c 4 · Qc Qa ⊗Qb⊗Qc i 4 · (I⊗Qc )
1Qa Qc
i
L
ρa
Quantum teleportation
i
−
1

Qa
1ac

?
Qc
We use the universal property of the product, naturality of λ and the explicit form of the natural isomorphism υc := pIi ⊗1c i=4 i=1 . In the speciﬁc case of a Bellbase † ⊕ † ⊕ ⊕ ⊕ † ⊕ ◦ β3 )† = β3† ◦ (σQ ) = β3 ◦ σQ . observation we use 1Q = 1Q , (σQ ) = σQ and (σQ Although in Rel teleportation works for ‘individual observational branches’ it fails to admit the full teleportation protocol since there are only two automorphisms of Q (which is just a twoelement set, i.e. the type of ‘classical bits’), and hence there is no teleportation base. We now consider suﬃcient conditions on the ambient category C for a teleportation base to exist. We remark ﬁrstly that if C(I, I) contains an additive inverse for 1, then it is a ring, and moreover all additive inverses exist in each homset C(A, B), so C is enriched over Abelian groups. Suppose then that C(I, I) is a ring with 1 = −1. We can deﬁne a morphism prebaseT = baseQ∗ ⊗Q ◦ M : 4 · I → Q∗ ⊗ Q where M is the endomorphism of 4 · I determined by the matrix ⎛ ⎞ 1 0 1 0 ⎜ 0 1 0 1 ⎟ ⎜ ⎟ ⎝ 0 1 0 −1 ⎠ 1 0 −1 0 The corresponding morphisms βj will have 2 × 2 matrices determined by the columns of this 4 × 4 matrix, and will be unitary. If C(I, I) furthermore contains a scalar s satisfying 2s† s = 1, then s • prebaseT is unitary, and the conditions for a teleportation base are fulﬁlled. Suppose we start with a ring R containing an element s satisfying 2s2 = 1. (Examples are plentiful, e.g. any subring of C,
Categorical Quantum Mechanics
303
√ or of Q( 2), containing √12 ). The category of ﬁnitely generated free Rmodules and Rlinear maps is strongly compact closed with biproducts, and admits a teleportation base (in which s will appear as a scalar with s = s† ), hence realizes teleportation.
6.3
Logicgate teleportation
Logic gate teleportation of qubits requires only a minor modiﬁcation as compared to the teleportation protocol. THEOREM 31. Let unitary morphism f : Q → Q be such that for each i ∈ {1, 2, 3, 4} a morphism ϕi (f ) : Q → Q satisfying f ◦ βi = ϕi (f ) ◦ f exists. The diagram of Theorem 30 with the modiﬁcations made below commutes. .. .. .. .. .. .. .. .. .. ..
.. .. ..? Qa ⊗ I 1a ⊗ (s • f )
produce f state
? Qa ⊗ (Q∗b ⊗ Qc ) .. .. .. .. .. .. .. .. .? 4 · Qc
Δ4ac ◦f
,i=4
−1 i=1 (ϕi (f ))
unitary correction
? ? 4 · Qc ========= 4 · Qc The righthandside of the diagram is our formal description of logicgate teleportation of f : Q → Q; the commutativity of the diagram under the stated conditions expresses the correctness of logicgate teleportation for qubits. Proof: The top trapezoid is the statement of the Theorem. The a, b and clabels are the same as in the proof of teleportation. For each j ∈ {1, 2, 3, 4} we have a
304
Samson Abramsky and Bob Coecke
diagram of the form below. Again we ignore the scalars in this proof. f i

4·Q

?

β
j
λ−1 Q
⊗1
λQ
Q
pIj ⊗1Q
 ?  I⊗Q
1I⊗Q

pQ j
?
I⊗Q

λ−1 Q
pQ j
? Q
ϕ
1

−
)
j(
f)
(f ϕj
Q
?
ϕ
i(
pI⊗Q j
◦f

f ◦ βj
Q
L
I 4 · λ−1 1Q ⊗f Q i Q  Q⊗Q∗⊗Q βi i ⊗1 Q (4 · I)⊗Q pi ⊗14 · (I⊗Q) 4·Q
Lemma 7
1Q
i
ρQ
Logicgate teleportation Q⊗I
Q
f)
−
1
Q
Q

f
? Q
This twodimensional case does not yet provide a universal computational primitive, which requires teleportation of Q ⊗ Qgates [Gottesman and Chuang, 1999]. We present the example of teleportation of a CNOT gate [Gottesman and Chuang, 1999] (see also [Coecke, 2003] Section 3.3). Given a Bell base we deﬁne a CNOT gate as one which acts as follows on tensors of the Bell base maps: ⊕
⊕ ⊕ = (σQ ⊗ σQ ) ◦ CNOT
CNOT ◦ (β3
⊗ 1Q ) = (β3 ⊗ 1Q ) ◦ CNOT
CNOT ◦ (σQ ⊗ 1Q )
⊕
CNOT ◦ (1Q ⊗ σQ )
⊕ = (1Q ⊗ σQ ) ◦ CNOT
CNOT ◦ (1Q
⊗ β3 ) = (β3 ⊗ β3 ) ◦ CNOT
CNOT ◦ (1Q
⊗ β4 ) = (β3 ⊗ β4 ) ◦ CNOT
It follows from this that CNOT ◦ (β4
⊕ ⊗ 1Q ) = (β4 ⊗ σQ ) ◦ CNOT
from which in turn it follows by bifunctoriality of the tensor that the required unitary corrections factor into single qubit actions, for which we introduce a notation by setting CNOT ◦ (βi
⊗ 1Q ) = ϕ1 (βi ) ◦ CNOT
CNOT ◦ (1Q
⊗ βi ) = ϕ2 (βi ) ◦ CNOT
The reader can verify that for 42 · (Qc1⊗Qc2 ) := 4 · (4 · (Qc1⊗Qc2 )) and
2
i=4 2 Δ4ac := s†s • s†s •1ac i=4 i=1 i=1 : Qa1⊗Qa2 → 4 · (Qc1⊗Qc2 )
the following diagram commutes.
Categorical Quantum Mechanics
305
Qa1⊗Qa2 ================= Qa1⊗Qa2
import unknown state
ρa
?
(Qa1⊗Qa2 ) ⊗ I 1a ⊗ (s2 • CNOT)
produce CNOTstate ?
∗
(Qa1⊗Qa2 ) ⊗ ((Qb1⊗Qb2 ) ⊗ (Qc1⊗Qc2 ))
spatial delocation
(α, σ)◦ (1a ⊗ (ub ⊗1c ))
?
∗ 1
∗ 2
((Qa1⊗Qb ) ⊗ (Qc1⊗Qc2 )) ⊗ (Qa2⊗Qb ) (s†•βi 1 1i=4 i=1 ⊗1c )⊗12 a b
1st observation ?
∗ 2
((4 · I) ⊗ (Qc1⊗Qc2 )) ⊗ (Qa2⊗Qb ) 2
Δ4ac ◦ CNOT
((4 · λ−1 c )◦υc ) ⊗ 12
1st communication ?
∗ 2
(4 · (Qc1⊗Qc2 )) ⊗ (Qa2⊗Qb ) `Li=4
c −1 i=1 (ϕ1 (βi ))
´
1st correction
⊗12
?
∗ 2
(4 · (Qc1⊗Qc2 )) ⊗ (Qa2⊗Qb ) (4 · 1c )⊗s† •βi 2 2i=4 i=1 a b
2nd observation ?
(4 · (Qc1⊗Qc2 )) ⊗ (4 · I) (4 · ρ−1 4c )◦τ4c
2nd communication ?
(4 · (4 · (Qc1⊗Qc2 ))) Li=4 2
?
i=1 (4
· ϕc2 (βi ))−1 2
2nd correction ?
4 · (Qc1⊗Qc2 ) ============ 4 · (Qc1⊗Qc2 )
306
Samson Abramsky and Bob Coecke
6.4 Entanglement swapping THEOREM 32. Setting γi
:= (βi )∗
Pi
:= s† s • (γi ◦ βi ) ,i=4 ∗ −1 −1 ∗ := i=1 (1b ⊗ γi ) ⊗ (1d ⊗ βi )
ζiac
Θab := 1∗d ⊗ Pi i=4 i=1 ⊗1c Ωab := s† s3 • (1ba⊗1dc)i=4 i=1 the following diagram commutes. I ⊗ I ============== I ⊗ I s2 • (1da⊗1bc)
produce EPRpairs
? (Q∗d ⊗ Qa ) ⊗ (Q∗b ⊗ Qc ) α
spatial delocation
? Q∗d ⊗ (Qa ⊗ Q∗b ) ⊗ Qc Ωab
Θab Bellbase measurement ? Q∗d ⊗ (4 · (Qa ⊗ Q∗b )) ⊗ Qc (4 · (α, σ)) ◦ (τ, υ) classical communication ? 4 · ((Q∗b ⊗ Qa )⊗(Q∗d ⊗ Qc )) ζiac
unitary correction
? ? 4 · ((Q∗b ⊗ Qa ) ⊗(Q∗d ⊗ Qc )) The righthandside of the above diagram is our formal description of the entanglement swapping protocol. Proof: The top trapezoid is the statement of the Theorem. We have a diagram of the form below for each j ∈ {1, 2, 3, 4}. To simplify the notation of the types
Categorical Quantum Mechanics
307
we set (a∗, b, c∗, d) for Q∗a ⊗ Qb ⊗ Q∗c ⊗ Qd etc. Again we ignore the scalars in this proof. a
⊗
1 b
Entanglement swapping
c 
Θab
(d∗, a, b∗, c) 1∗
Lemma 8
ρI
(d∗, c)
d
⊗
β j
λd∗ ⊗ 1c
−1
1c
(d∗, I , c)
λI
Lemma 6
p (a,b ∗
)
?
a⊗
1∗ d⊗ − β
pj
?  (b∗, a, d∗, c)
σ
?
(d∗, a, b∗, c) ======== (d∗, a, b∗, c) 1 b
4·(b∗, a, d∗, c) pj
c?
 (d∗, a, b∗, c)
L. 6
4·σ 
pj
⊗1
1∗d ⊗γj⊗1c
1∗d ⊗1ab⊗βj−1
(I , d∗, c)
I

4·(d∗, a, b∗, c)
?  (b∗, a, d∗, c) 
σ
(τ,υ) j
⊗
ρ−1 d∗ ⊗1c

?
(d∗, 4·(a, b∗ ), c) 1∗ d⊗
‡
βj
 4·(b∗, a, d∗, c) 
6
1ba ⊗ 1dci
1 d
I⊗I
σ
σ −1
 ?  (b∗, a, d∗, c)
j 1
? I⊗I
1ba ⊗ 1dc
where −1 ∗ (1∗b ⊗γ−1 := i ⊗1d ⊗βi )
,
−1 ∗ ‡ := 1∗b ⊗γ−1 j ⊗1d ⊗βj
−1 ∗ and . := 1∗d⊗γ−1 j ⊗1b ⊗βj
i
We use γi = (βi )∗ rather than βi to make Pi an endomorphism and hence a projector. The general deﬁnition of a ‘bipartite entanglement projector’ is Pf := f ◦ f∗ = f ◦ f † ◦ σA∗,B : A∗ ⊗ B → A∗ ⊗ B for f : A → B, so in fact Pi = P(βi )∗ . 7 EXTENSIONS AND FURTHER DEVELOPMENTS Since its ﬁrst publication in 2004, a number of elaborations on the categorical quantum axiomatics described above have been proposed, by ourselves in collaboration with members of our group, Ross Duncan, Dusko Pavlovic, Eric Oliver Paquette, Simon Perdrix and Bill Edwards, and also by others elsewhere, most notably Peter Selinger and Jamie Vicary. We shall present some of the main developments.
308
Samson Abramsky and Bob Coecke
7.1 Projective structure We shall discuss our ﬁrst topic at considerably greater length than the others we shall cover in this survey. The main reason for this is that it concerns the passage to a projective point of view, which makes for an evident comparison with the standard approaches to quantum logic going back to [Birkhoﬀ and von Neumann, 1936]. Thus it seems appropriate to go into some detail in our coverage of this topic, in the context of the Handbook in which this article will appear. The axiomatics we have given corresponds to the pure state picture of quantum mechanics. The very fact that we can faithfully carry out linearalgebraic calculations using the semiadditive structure provided by biproducts means that states will typically carry redundant global phases, as is the case for vectors in Hilbert spaces. Eliminating these means ‘going projective’. The quantum logic tradition provides one way of doing so [Birkhoﬀ and von Neumann, 1936]. Given a Hilbert space one eliminates global scalars by passing to the projection lattice. The nonBoolean nature of the resulting lattice is then taken to be characteristic for quantum behaviour. This leads one then to consider certain classes of nondistributive lattices as ‘quantum structures’. Bir kho ﬀ v
Hilbert space
on
kill redundant global scalars
Ne
um
? lattice of subspaces
go abstract
ann
 nondistributive lattices
It is wellknown that there is no obvious counterpart for the Hilbert space tensor product when passing to these more general classes of lattices. This is one reason why Birkhoﬀvon Neumann style quantum logic never penetrated the mainstream physics community, and is particularly unfortunate in the light of the important role that the tensor product plays in quantum information and computation. But one can also start from the whole category of ﬁnite dimensional Hilbert spaces and linear maps FdHilb. Then we can consider ‘strongly compact closed categories + some additive structure’ as its appropriate abstraction, and hope to ﬁnd some abstractly valid counterpart to ‘elimination of redundant global scalars’.
FdHilb
go abstract our
app roa
 ‘vectorial’ strong compact closure kill redundant global scalars
ch
 ? ‘projective’ strong compact closure
Categorical Quantum Mechanics
309
The major advantage which such a construction has is that the tensor product is now part of the mathematical object under consideration, and hence will not be lost in the passage from vectorial spaces to projective ones. This passage was realised in [Coecke, 2007a] as follows. For morphisms in FdHilb, i.e. linear maps, if we have that f = eiθ · g with θ ∈ [0, 2π[ for f, g : H1 → H2 , then f ⊗ f † = eiθ· g ⊗ (eiθ· g)† = eiθ· g ⊗ e−iθ· g † = g ⊗ g † . Now in abstract generality, given a strongly compact closed category C, we can deﬁne a new category WProj(C) with the same objects as those of C, but with 5 6 4 WProj(C)(A, B) := f ⊗ f † 5 f ∈ C(A, B) as homsets and in which composition is given by (f ⊗ f † ) ¯ ◦ (g ⊗ g † ) := (f ◦ g) ⊗ (f ◦ g)†. One easily shows that WProj(C) is again a strongly compact closed category. The abstract counterpart to elimination of global phases is expressed by the following propositions. PROPOSITION 33. [Coecke, 2007a] For morphisms f , g and scalars s, t in a strongly compact closed category, we have s • f = t • g ∧ s ◦ s† = t ◦ t† = 1I
=⇒ f ⊗ f † = g ⊗ g † .
PROPOSITION 34. [Coecke, 2007a] For morphisms f and g in a strongly compact closed category with scalars S we have f ⊗ f † = g ⊗ g†
=⇒
∃s, t ∈ S. s • f = t • g ∧ s ◦ s† = t ◦ t† .
In particular we can set s := (f )† ◦ f
t := (g)† ◦ f .
and
While the ﬁrst proposition is straightforward, the second one is somewhat more surprising. It admits a simple graphical proof. We represent units by dark triangles and their adjoints by the same triangle but depicted upside down. Other morphisms are depicted by square boxes as before, with the exception of scalars which are depicted by ‘diamonds’. The scalar s := (f )† ◦ f is depicted as
f† f
310
Samson Abramsky and Bob Coecke
Bifunctoriality means that we can move these boxes upward and downward, and naturality provides additional modes of movement, e.g. scalars admit arbitrary movements. Now, given that f ⊗ f † = g ⊗ g † , that is, in a picture,
=
f†
f
g†
g
we need to show that s • f = t • g and s ◦ s† = t ◦ t† for some choice of scalars s and t, that is, in a picture,
=
f
t
s
g
s†
=
s
t
t†
The choice that we will make for s and t is
f†
t
=:
f
g†
=:
s
f
Then we indeed have s • f = t • g since in
f
f
=
f†
g†
g
f
the areas within the dotted line are equal by assumption. We also have that s ◦ s† = t ◦ t† since
Categorical Quantum Mechanics
f†
f
f
=
f†
311
g†
f†
f
g
which completes the proof. As expected, biproducts do not survive the passage from C to WProj(C) but the weaker structure which results still suﬃces for a comprehensive description of the protocols we have discussed in this article. In particular, the distributivity natural isomorphisms dist0,l : A ⊗ 0 + 0
distl : A ⊗ (B ⊕ C) + (A ⊗ B) ⊕ (A ⊗ C)
dist0,r : 0 ⊗ A + 0
distr : (B ⊕ C) ⊗ A + (B ⊗ A) ⊕ (C ⊗ A) .
carry over to WProj(C). Details can found in [Coecke, 2007a]. Our framework also allows a precise general statement of the incompatibility of biproducts with projective structure. Call a strongly compact closed category projective iﬀ equality of projections implies equality of the corresponding states, that is, (1) ∀ ψ, φ : I → A .
ψ ◦ ψ † = φ ◦ φ† =⇒ ψ = φ .
PROPOSITION 35. [Coecke, 2007a] If a strongly compact closed category with biproducts is projective and the semiring of scalars admits negatives, i.e. is a ring, then we have 1 = −1, that is, there are no nontrivial negatives. Having no nontrivial negatives of course obstructs the description of interference.
7.2
Mixed states and Completely Positive Maps
The categorical axiomatics set out in this article primarily refers to the purestate picture of quantum mechanics. However, for many purposes, in particular those of quantum information, it is mixed states, acted on by completely positive maps, which provide the most appropriate setting. Peter Selinger [2007] proposed a general categorical construction, directly in the framework of the categorical axiomatics of [Abramsky and Coecke, 2004] which has been described in this article, to capture the passage from the pure states to the mixed states picture. The construction proceeds as follows. Given any strongly compact closed category C we deﬁne a new category CPM(C) with the same objects as C but with morphisms given by CPM(C)(A, B) :=
312
Samson Abramsky and Bob Coecke
4
5 6 (1B ⊗ C ⊗ 1B ∗ ) ◦ (1B⊗C ⊗ σB ∗ ,C ∗ ) ◦ (f ⊗ f∗ ) 5 f ∈ C(A, B ⊗ C)
where for simplicity we assume that the monoidal structure is strict. Composition in CPM(C) is inherited pointwise from C. The morphisms of the category CPM(FdHilb) are exactly the completely positive maps, and the morphisms in the homset CPM(FdHilb)(C, H) are exactly the selfadjoint operators with positive trace on H. The category WProj(C) faithfully embeds in CPM(C) by setting f ⊗ f † → f ⊗ f∗ . Metaphorically, we have CPM(C) density operators = . WProj(C) projectors For more details on the CPMconstruction we refer the reader to [Selinger, 2007]. Recently it was shown that the CPMconstruction does not require strong compact closure, but only dagger symmetric monoidal structure. Details are in [Coecke, 2007]. An axiomatic presentation of categories of completely positive maps is given in [Coecke, 2008].
7.3 Generalised NoCloning and NoDeleting theorems The NoCloning theorem [Dieks, 1982; Wootters and Zurek, 1982] is a basic limitative result for quantum mechanics, with particular signiﬁcance for quantum information. It says that there is no unitary operation which makes perfect copies of an unknown (pure) quantum state. A stronger form of this result is the NoBroadcasting theorem [Barnum et al., 1996], which applies to mixed states. There is also a NoDeleting theorem [Pati and Braunstein, 2000]. The categorical and logical framework which we have described provides new possibilities for exploring the structure, scope and limits of of quantum information processing, and the features which distinguish it from its classical counterpart. One area where some striking progress has already been made is the axiomatics of NoCloning and NoDeleting. It is possible to delimit the classicalquantum boundary here in quite a subtle way. On the one hand, we have the strongly compact closed structure which is present in the usual Hilbert space setting for QIC, and which we have shown accounts in generality for the phenomena of entanglement. Suppose we were to assume that either copying or deleting were available in a strongly compact closed category as uniform operations. Mathematically, a uniform copying operation means a natural diagonal ΔA : A → A ⊗ A i.e. a monoidal natural transformation, which moreover is coassociative and co
Categorical Quantum Mechanics
313
commutative:
αA,A,A ?  A⊗A  (A ⊗ A) ⊗ A Δ Δ⊗1
A
ΔA⊗A σA,A ? A⊗A

A
Δ1 ⊗Δ A⊗A A ⊗ (A ⊗ A)
Δ
A 7 7 7 7 7 7 7 7 7 7
Thinking of the diagonal associated with the usual cartesian product, one sees immediately that cocommutativity and coassociativity are basic requirements for a copying operation: if I have two copies of the same thing, it does not matter which order they come in, and if I produce three copies by iterating the copying operation, which copy I choose to perform the second copying operation on is immaterial. Naturality, on the other hand, corresponds essentially to basisindependence in the Hilbert space setting; it says that the operation exists ‘for logical reasons’, in a representationindependent form. We have shown recently that under these assumptions the category trivializes; in other words, that this combination of quantum and classical features is inconsistent, leading to a collapse of the structure. The precise form of the result is that under these hypotheses every endomorphism in the category is a scalar multiple of the identity. Similar generalizations of the NoDeleting theorem [Pati and Braunstein, 2000] and the NoBroadcasting theorem [Barnum et al., 1996] also hold. Papers on these results are in preparation. One striking feature of these results is that they are visibly in the same genre as a wellknown result by Joyal in categorical logic [Lambek and Scott, 1986] showing that a ‘Boolean cartesian closed category’ trivializes, which provides a major roadblock to the computational interpretation of classical logic. In fact, they strengthen Joyal’s result, insofar as the assumption of a full categorical product (diagonals and projections) in the presence of a classical duality is weakened. This shows a heretofore unsuspected connection between limitative results in proof theory and NoGo theorems in quantum mechanics. Another interesting point is the way that this result is delicately poised. The basis structures to be discussed in the next subsection do assume commutative comonoid structures existing in strongly compact closed categories—indeed with considerable additional properties, such as the Frobenius identity. Not only is this consistent, such structures correspond to a major feature of Hilbert spaces, namely orthonormal bases. The point is that there are many such bases for a given Hilbert space, and none are canonical. Indeed, the choice of basis corresponds to the choice of measurement setup, to be made by a ‘classical observer’. The key ingredient which leads to inconsistency, and which basis structures lack, is naturality, which, as we have already suggested, stands as an abstract proxy for basisindependence.
314
Samson Abramsky and Bob Coecke
7.4 Basis Structures and Classical Information In this article, an approach to measurements and classical information has been developed based on biproducts. This emphasizes the branching structure of measurements due to their probabilistic outcomes. One may distinguish the ‘multiplicative’ from the ‘additive’ levels of our axiomatization (using the terminology of Linear logic [Girard, 1987]). The multiplicative, purely tensorial level of strongly compact closed categories shows, among other things, how a remarkable amount of multilinear algebra, encompassing much of the structure needed for quantum mechanics and quantum information, can be done without any substrate of linear algebra. Moreover, this level of the axiomatization carries a very nice diagrammatic calculus, which we have sampled informally. In general, the return on structural insights gained from the axiomatization seems very good. The additive level of biproducts reinstates a linear (or ‘semilinear’) level of structure, albeit with fairly weak assumptions, and there is more of a sense of recapitulating familiar deﬁnitions and calculations. While a diagrammatic calculus is still available here (see [Abramsky and Duncan, 2006]), it is subject to a combinatorial unwieldiness familiar from process algebra in Computer Science [Milner, 1989] (cf. the ‘Expansion Theorem’). An alternative approach to measurements and classical information has been developed in a series of papers [Coecke and Pavlovic, 2007; Coecke and Paquette, 2006; Coecke, Pavlovic and Vicary, 2008a; Coecke, Paquette and Perdrix, 2008] under various names, the best of which is probably ‘basis structure’. Starting from the standard idea that a measurement setup corresponds to a choice of orthonormal basis, the aim is to achieve an axiomatization of the notion of basis as an additional structure. Of course, the notion of basis developed in Section 5 has all the right properties, but it is deﬁned in terms of biproducts, while the aim here is to achieve an axiomatization purely at the multiplicative level. This is done in an interesting way, bringing the informatic perspective to the fore. One can see the choice of a basis as determining a notion of ‘classical data’, namely the basis vectors. These vectors are subject to the classical operations of copying and deleting, so in a sense classical data, deﬁned with respect to a particular choice of basis, stands as a contrapositive to the NoCloning and NoDeleting theorems. Concretely, having chosen a basis {i} on a Hilbert space H, we can deﬁne linear maps H −→ H ⊗ H :: i → ii,
H −→ C :: i → 1
which do correctly copy and delete the basis vectors (the ‘classical data’), although not of course the other vectors. These considerations lead to the following deﬁnition. A basis structure on an object A in a strongly compact closed category is a commutative comonoid structure on A Copy : A → A ⊗ A,
Delete : A → I
Categorical Quantum Mechanics
315
subject to a number of additional axioms, the most notable of which is the Frobenius identity [Carboni and Walters, 1987]. In FdHilb these structures exactly correspond to orthonormal bases [Coecke, Pavlovic and Vicary, 2008a], which justiﬁes their name and interpretation. Quantum measurements can be deﬁned relative to these structures, as selfadjoint EilenbergMoore coalgebras for the comonads induced by the above comonoids [Coecke and Pavlovic, 2007]. In FdHilb these indeed correspond exactly to projective spectra. The EilenbergMoore coalgebra square A
M easure
1X ⊗ M easure
M easure ? X ⊗A
 X ⊗A
Copy ⊗ 1A
?  X ⊗X ⊗A
can be seen as an operational expression of von Neumann’s projection postulate in a resourcesensitive setting: measuring twice is the same as measuring once and then copying the measurement outcome. This abstract notion of measurement admits generalisation to POVMs and PMVMs, for which a generalised Naimark dilatation theorem can be proved at the abstract level [Coecke and Paquette, 2006]. Within CPM(C) the decoherence aspect of quantum measurement, which, concretely in FdHilb, is the completely positive map which eliminates nondiagonal elements relative to the measurement basis, arises as Copy ◦ Copy † : X ⊗ X → X ⊗ X where X is now taken to be selfdual, that is, X = X ∗ . These basis structures not only allow for classical data, measurement and control operations to be described but also provide useful expressiveness when discussing multipartite states and unitaries. For example, they capture GHZstates in a canonical fashion [Coecke and Pavlovic, 2007], and enable an elegant description of the statetransfer protocol [Coecke, Paquette and Perdrix, 2008]. Vicary showed that if one drops the cocommutativity requirement of basis structures in FdHilb, then, rather than all orthonormal bases, one ﬁnds exactly all ﬁnite dimensional C*algebras [Vicary, 2008]. The fact that these multiplicative basis structures allow measurements to be expressed without any explicit account of branching may be compared to the way that the pure λcalculus can be used to encode booleans and conditionals [Barendregt, 1984]. From this perspective, explicit branching can be seen to have its merits, while modelchecking [Clarke et al., 1999] has done much to ameliorate the combinatorial unwieldiness mentioned above. It is likely that further insights will be gained by a deeper understanding of the relationships between the additive and multiplicative levels.
316
Samson Abramsky and Bob Coecke
7.5 Complementary observables and phases A further step is taken in [Coecke and Duncan, 2008], where abstract counterparts to (relative) phases are deﬁned. Given a basis structure on X and a point ψ : I → X its action on X is deﬁned to be the morphism Δ(ψ) := Copy † ◦ (ψ ⊗ 1X ) : X → X. From the axioms of basis structures it follows that the set of all these actions on the homset C(I, X) is a commutative monoid. Deﬁning unbiassed points as those ψ ∈ C(I, X) for which Δ(ψ) is unitary, the corresponding set of unbiassed actions is always an abelian group, which we call the phase group. In the case of the qubit in FdHilb the phase group corresponds to the equator of the Bloch sphere, that is, indeed, to relative phase data. Also in [Coecke and Duncan, 2008] an axiomatics is proposed for complementary observables. It is shown that for all known constructions of complementary quantum observables, the corresponding basis structures obey a ‘scaled’ variant of the bialgebra laws. This scaled bialgebra structure together with the phase group is suﬃciently expressive to describe all linear maps, hence all mutipartite states and unitary operators, in FdHilb. It provides an abstract means to reason about quantum circuits and to translate between quantum computational models, such as the circuit model and the measurementbased model. As an application, a description is given of the quantum Fourier transform, the key ingredient of Shor’s factoring algorithm [Shor, 1994], the bestknown example of a quantum algorithm.
7.6 The quantum harmonic oscillator Jamie Vicary [2007] gave a purely categorical treatment of the quantum harmonic oscillator, directly in the setting described in this article, of strongly compact closed categories with biproducts. In Linear logic terminology, he introduced an ‘exponential level’ of structure, corresponding to Fock space. This provides a monoidal adjunction that encodes the raising and lowering operators into a cocommutative comonoid. Generalised coherent states arise through the homset isomorphisms deﬁning the adjunction, and it is shown that they are eigenstates of the lowering operators. Similar results were independently obtained in [Fiore, 2007] in an abstract ‘formal power series’ context, with a motivation stemming from Joyal’s theory of species.
7.7 Automated quantum reasoning The structures uncovered by the research programme we have described provide a basis for the design of software tools for automated reasoning about quantum phenomena, protocols and algorithms. Several MSc students at Oxford University
Categorical Quantum Mechanics
317
Computing Laboratory have designed and implemented such tools for their Masters Thesis projects. An ongoing highlevel comprehensive approach has recently be initiated by Lucas Dixon and Ross Duncan [2008].
7.8
Diagrammatic reasoning
We have used a diagrammatic notation for tensor categories in an informal fashion. In fact, this diagrammatic notation, which can be traced back at least to Penrose [1971], was made fully formal by Joyal and Street [1991]; topological applications can be found in [Turaev, 1994]. The various structures which have arisen in the above discussion, such as strong compact closure, biproducts, dagger Frobenius comonoids, phase groups, scaled bialgebras, and the exponential structures used in the description of the quantum harmonic oscilator, all admit intuitive diagrammatic presentations in this tradition. References on these include [Abramsky and Duncan, 2006; Coecke and Paquette, 2006; Coecke and Duncan, 2008; Vicary, 2007]. Tutorial introductions to these diagrammatic calculi are given in [Coecke and Paquette, 2008; Selinger, 2008a] These diagrammatic calculi provide very eﬀective tools for the communication of the structural ideas. The software tools mentioned in the previous subsection all support the presentation and manipulation of such diagrams as their interface to the user.
7.9
Free constructions
In [Abramsky, 2005] a number of free constructions are described in a simple, synthetic and conceptual manner, including the free strongly compact closed category over a dagger category, and the free traced monoidal category. The KellyLapalaza [1980] construction of the free compact closed category is recovered in a structured and conceptual fashion. These descriptions of free categories in simple combinatorial terms provide a basis for the use of diagrammatic calculi as discussed in the previous subsection.
7.10
TemperleyLieb algebra and connections to knot theory and topological quantum ﬁeld theory
Our basic categorical setting has been that of symmetric monoidal categories. If we weaken the assumption of symmetry, to braided or pivotal categories, we come into immediate contact with a wide swathe of developments relating to knot theory, topology, topological quantum ﬁeld theories, quantum groups, etc. We refer to [Freyd and Yetter, 1989; Kaufmann, 1991; Turaev, 1994; Yetter, 2001; Kock, 2003; Street, 2007] for a panorama of some of the related literature. In [Abramsky, 2007], connections are made between the categorical axiomatics for quantum mechanics developed in this article, and the TemperleyLieb algebra,
318
Samson Abramsky and Bob Coecke
which plays a central rˆ ole in the Jones polynomial and ensuing developments. For illustration, we show the deﬁning relations of the TemperleyLieb algebra, in the diagrammatic form introduced by Kauﬀman:
=
U1 U2 U1 = U1
=
U12 = δU1
=
U1 U3 = U3 U1
The relationship with the diagrammatic notation we have been using should be reasonably clear, The ‘cups’ and ‘caps’ in the above diagrams correspond to the triangles we have used to depict units and counits. An important mediating rˆ ole is played by the geometry of interaction [Girard, 1989; Abramsky, 1996], which provides a mathematical model of information ﬂow in logic (cutelimination of proofs) and computation (normalization of λterms). The TemperleyLieb algebra is essentially the (free) planar version of our quantum setting; and new connections are made between logic and geometry in [Abramsky, 2007]. For example, a simple, direct description of the TemperleyLieb algebra, with no use of quotients, is given in [Abramsky, 2007]. This leads in turn to full completeness results for various noncommutative logics. Moreover, planarity is shown to be an invariant of the information ﬂow analysis of cut elimination. This leads to a number of interesting new kinds of questions: • It seems in practice that few naturally occurring quantum protocols require the use of the symmetry maps. (For example, none of those described in this paper do). How much of Quantum Informatics can be done ‘in the plane’ ? What is the signiﬁcance of this constraint? • Beyond the planar world we have braiding, which carries 3dimensional geometric information. Does this information have some computational significance? Some ideas in this direction have been explored by Kauﬀman and Lomonaco [2002], but no clear understanding has yet been achieved. • Beyond this, we have the general setting of Topological Quantum Field Theories [Witten, 1988; Atiyah, 1998] and related notions. This may be relevant to Quantum Informatic concerns in (at least) two ways: 1. A novel and promising paradigm of Topological Quantum Computing has recently been proposed [Freedman et al., 2004]. 2. The issues arising from distributed quantum computing, quantum security protocols etc. mean that the interactions between quantum informatics and spatiotemporal structure will need to be considered.
Categorical Quantum Mechanics
7.11
319
Logical syntax
In [Abramsky and Duncan, 2006] a strongly normalising proofnet calculus corresponding to the logic of strongly compact closed categories with biproducts is presented. The calculus is a full and faithful representation of the free strongly compact closed category with biproducts on a given category with an involution. This syntax can be used to represent and reason about quantum processes. In [Duncan, 2006] this is extended to a description of the free strongly compact category generated by a monoidal category. This is applied to the description of the measurement calculus of [Danos et al., 2007].
7.12
Completeness
In [Selinger, 2008] Selinger showed that ﬁnitedimensional Hilbert spaces are equationally complete for strongly compact closed categories. This result shows that if we want to verify an equation expressed purely in the language of strongly compact closed categories, then it suﬃces to verify that it holds for Hilbert spaces.
7.13
Toy quantum categories
In [Coecke and Edwards, 2008] it is shown that Spekkens’ wellknown ‘toy model’ of quantum mechanics described in [Spekkens, 2007] can be regarded as an instance of the categorical quantum axiomatics. The category Spek is deﬁned to be the dagger symmetric monoidal subcategory of Rel generated by those objects whose cardinality is a power of 4, the symmetry group on 4 elements, and a wellchosen copyingdeleting pair for the 4 element set. BIBLIOGRAPHY [Abramsky, 1996] S. Abramsky. Retracing some paths in process algebra. In: Proceedings of CONCUR 96, Lectures Notes in Computer Science Volume 1119, pp. 1–17. SpringerVerlag, 1996. [Abramsky, 2004] S. Abramsky. Highlevel methods for quantum computation and information. In: Proceedings of the 19th Annual IEEE Symposium on Logic in Computer Science, IEEE Computer Science Press, 2004. [Abramsky, 2005] S. Abramsky. Abstract scalars, loops, and free traced and strongly compact closed categories. In: Proceedings of CALCO 2005, pages 1–31. Springer Lecture Notes in Computer Science 3629, 2005. [Abramsky, 2007] S. Abramsky. TemperleyLieb algebra: From knot theory to logic and computation via quantum mechanics. In: Mathematics of Quantum Computing and Technology, G. Chen, L. Kauﬀman and S. Lamonaco (eds), pages 415–458. Taylor and Francis, 2007. [Abramsky, Blute and Panangaden, 1999] S. Abramsky, R. Blute and P. Panangaden. Nuclear and trace ideals in tensored ∗categories. Journal of Pure and Applied Algebra 143, 3–47, 1999. [Abramsky and Coecke, 2003] S. Abramsky and B. Coecke. Physical traces: Quantum vs. classical information processing. Electronic Notes in Theoretical Computer Science 69. arXiv:cs/0207057, 2003. [Abramsky and Coecke, 2004] S. Abramsky and B. Coecke. A categorical semantics of quantum protocols. In: Proceedings of 19th IEEE conference on Logic in Computer Science, pages 415– 425. IEEE Press. arXiv:quantph/0402130, 2004.
320
Samson Abramsky and Bob Coecke
[Abramsky and Coecke, 2005] S. Abramsky and B. Coecke. Abstract physical traces. Theory and Applications of Categories 14, 111–124. http://www.tac.mta.ca/tac/volumes/14/6/1406abs.html, 2005. [Abramsky and Duncan, 2006] S. Abramsky and R. W. Duncan. A categorical quantum logic. Mathematical Structures in Computer Science 16, 469–489. arXiv:quantph/0512114, 2006 [Abramsky, Gay and Nagarajan, 1995] S. Abramsky, S. J. Gay and R. Nagarajan. Interaction categories and the foundations of typed concurrent programming. In: Deductive Program Design: Proceedings of the 1994 Marktoberdorf International Summer School, NATO Advanced Science Institutes Series F, pages 35–113. SpringerVerlag, 1995. [Abramsky and Tzevelekos, 2008] S. Abramsky and N. Tzevelekos. Introduction to categories and categorical logic. In: New Structures for Physics, B. Coecke, Ed, Springer lecture Notes in Physics, 2008. [Atiyah, 1998] M. F. Atiyah. Topological quantum ﬁeld theory. Publications Math´ematiques de l’IHES 68, 175–186, 1998. [Baez, 2006] J. C. Baez. Quantum quandaries: a categorytheoretic perspective. In: The Structural Foundations of Quantum Gravity, D. Rickles, S. French and J. T. Saatsi (Eds), pages 240–266. Oxford University Press. arXiv:quantph/0404040, 2006 [Baez and Dolan, 1995] J. C. Baez and J. Dolan. Higherdimensional algebra and topological quantum ﬁeld theory. Journal of Mathematical Physics 36, 60736105. arXiv:qalg/9503002, 1995 [Baez and Stay, 2008] J. C. Baez and M. Stay (2008) Physics, topology, logic and computation: A Rosetta Stone. In: New Structures for Physics, B. Coecke, Ed, Springer lecture Notes in Physics, 2008. [Barendregt, 1984] H. Barendregt. The Lambda Calculus: Revised Edition. Elsevier, 1984. [Barnum et al., 1996] H. Barnum, C. M. Caves, C. A. Fuchs, R. Jozsa, and B. Schumacher. Noncommuting mixed states cannot be broadcast. Physical Review Letters, 76, 2818–2821, 1996. [Barr, 1979] M. Barr. ∗Autonomous Categories. Lecture Notes in Mathematics 752, Springer, 1979. [Bell, 1964] J. S. Bell. On the Einstein Podolsky Rosen paradox. Physics 1, 195, 1964. [Bennett et al., 1993] C. H. Bennett, G. Brassard, C. Cr´epeau, R. Jozsa, A. Peres and W. K. Wooters. Teleporting an unknown quantum state via dual classical and EinsteinPodolskyRosen channels. Physical Review Letters 70, 1895–1899, 1993. [Birkhoﬀ and von Neumann, 1936] G. Birkhoﬀ and J. von Neumann. The logic of quantum mechanics. Annals of Mathematics 37, 823–843, 1936. [Bouwmeester et al., 2001] D. Bouwmeester, A. Ekert and A. Zeilinger, eds. The Physics of Quantum Information. SpringerVerlag, 2001. [Carboni and Walters, 1987] A. Carboni and R. F. C. Walters. Cartesian bicategories I. Journal of Pure and Applied Algebra 49, 11–32, 1987. [Clarke et al., 1999] E. M. Clarke, O. Grumberg and D. Peled. Model Checking. Springer, 1999. [Cockett and Seely, 1997] R. Cockett and R. A. G. Seely. Weakly distributive categories. Journal of Pure and Applied Algebra 114, 85–131, 1997. [Coecke, 2003] B. Coecke. The logic of entanglement. An invitation. Technical Report PRGRR0312. web.comlab.ox.ac.uk/oucl/publications/tr/rr0312.html, 2003. [Coecke, 2004] B. Coecke. The logic of entanglement. arXiv:quantph/0402014, 2004. [Coecke, 2007] B. Coecke. Complete positivity without positivity and without compactness. Oxford University Computing Laboratory Research Report PRGRR0705, 2007. web.comlab.ox.ac.uk/ oucl/publications/tr/rr0705.html [Coecke, 2007a] B. Coecke. Delinearizing linearity: projective quantum axiomatics from strong compact closure. Electronic Notes in Theoretical Computer Science 170, 47–72, 2007. arXiv:quantph/0506134 [Coecke, 2008] B. Coecke. Axiomatic description of mixed states from Selinger’s CPMconstruction. Proceedings of QPL’06. Electronic Notes in Theoretical Computer Science (to appear, 2008). [Coecke and Duncan, 2008] B. Coecke and R. W. Duncan. Interacting quantum observables. In: Proceedings of the 35th International Colloquium on Automata, Languages and Programming (ICALP), pages 298–310, Lecture Notes in Computer Science 5126, SpringerVerlag, 2008.
Categorical Quantum Mechanics
321
[Coecke and Edwards, 2008] B. Coecke and W. A. Edwards. Toy quantum categories. In: Proc. Quantum Physics and Logic/Development of Computational Models (QPLDCM). Electronic Notes in Theoretical Computer Science, to appear 2008. [Coecke and Paquette, 2006] B. Coecke and E. O. Paquette. POVMs and Naimark’s theorem without sums. Electronic Notes in Theoretical Computer Science (To appear). arXiv:quantph/0608072, 2006. [Coecke and Paquette, 2008] B. Coecke and E. O. Paquette. Categories for the practising physicist. In: New Structures for Physics, B. Coecke, Ed, Springer lecture Notes in Physics, 2008. [Coecke, Paquette and Perdrix, 2008] B. Coecke, E. O. Paquette and S. Perdrix. Bases in diagramatic quantum protocols. In: Proc. Mathematical Foundations for Programming Semantics XXIV, to appear, 2008. [Coecke and Pavlovic, 2007] B. Coecke and D. Pavlovic. Quantum measurements without sums. In: Mathematics of Quantum Computing and Technology, G. Chen, L. Kauﬀman and S. Lamonaco (eds), pages 567–604. Taylor and Francis, 2007. arXiv:quantph/0608035. [Coecke, Pavlovic and Vicary, 2008a] B. Coecke, D. Pavlovic and J. Vicary. Commutative †Frobenius algebras in FdHilb are bases. In: New Structures for Physics, B. Coecke, ed, Springer Lecture Notes in Physics, 2008. [Crane, 2006] L Crane. Categorical geometry and the mathematical foundations of quantum general relativity, 2006. arXiv:grqc/0602120v [Danos et al., 2007] V. Danos, E. Kasheﬁ and P. Panangaden. The Measurement Calculus. Journal of the ACM 54, 2007. arXiv:quantph/0412135 [Deligne, 1990] P. Deligne. Cat´ egories tannakiennes. In: The Grothendieck Festschrift Volume II, Progress in Mathematics 87, pages 111–196. Birkh¨ auser, 1990. [Deutsch, 1985] D. Deutsch. Quantum theory, the ChurchTuring principle and the universal quantum computer. Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences 400, 97–117, 1985. [Deutsch, 1989] D. Deutsch. Quantum computational networks. Proceedings of the Royal Society of London, Series A, Mathematical and Physical Sciences 425, 73–90, 1989. [Dieks, 1982] D. G. B. J. Dieks. Communication by EPR devices. Physics Letters A 92, 271–272, 1982. [Dirac, 1947] P. A. M. Dirac, (1947) The Principles of Quantum Mechanics (third edition). Oxford University Press, 1947. [Dixon and Duncan, 2008] L. Dixon and R. W. Duncan. Extending graphical representations for compact closed categories with applications to symbolic quantum computation. Proceedings of AISC’08, Lecture Notes in Artiﬁcial Intelligence, Springer, 2008. [Do¨ ering and Isham, 2007] A. Do¨ ering and C. J. Isham. A topos foundation for theories of physics: I. Formal languages for physics, 2007. arXiv:quantph/0703060 [Doplicher and Roberts, 1989] S. Doplicher and J. E. Roberts. A new duality theory for compact groups. Inventiones Mathematicae 98, 157, 1989. [Duncan, 2003] R. W. Duncan. Quantum entanglement and multiplicative linear logic. Transfer Report Oxford University, November 2003, 2003. [Duncan, 2006] R. W. Duncan. Types for Quantum Computing. D.Phil. thesis. University of Oxford, 2006. [Einstein et al., 1935] A. Einstein, B. Podolsky and N. Rosen. Can quantummechanical description of physical reality be considered complete? Physical Review 47, 777, 1935. [Ekert, 1991] A. K. Ekert. Quantum cryptography based on Bell’s theorem. Physical review letters 67, 661–663, 1991. [Farhi et al., 2000] E. Farhi, J. Goldstone, S. Gutmann and M. Sipser (2000) Quantum computation by adiabatic evolution, 2000. arXiv:quantph/0001106 [Fiore, 2007] M. Fiore. An axiomatics and a combinatorial model for creation/annihilation operators and diﬀerential structure. Talk given at the workshop on Categorical Quantum Logic, Oxford University, Jul. 2007. Slides available at http://se10.comlab.ox.ac.uk:8080/ FOCS/CQL program en.html [Freedman et al., 2004] M. H. Freedman, A. Kitaev, M. J. Larsen and Z. Wang. Topological quantum computation, 2004. arXiv:quantph/0101025 [Freyd and Yetter, 1989] P. Freyd and D. Yetter. Braided compact closed categories with applications to lowdimensional topology. Advances in Mathematics 77, 156–182, 1989. [Girard, 1987] J.Y. Girard. Linear logic. Theoretical Computer Science 50, 1–102, 1987.
322
Samson Abramsky and Bob Coecke
[Girard, 1989] J.Y. Girard. Geometry of Interaction I: Interpretation of System F. In: Logic Colloquium ’88, R. Ferro et al., pp. 221–260. NorthHolland, 1989. [Gottesman and Chuang, 1999] D. Gottesman and I. L. Chuang. Quantum teleportation is a universal computational primitive. Nature 402, 390–393, 1999. arXiv:quantph/9908010 [Harding, 2007] J. Harding, Some quantum logic and a few categories. Talk given at the workshop on Categorical Quantum Logic, Oxford University, Jul. 2007. Slides available at http://se10.comlab.ox.ac.uk:8080/FOCS/CQL program en.html [Harding, 2008] J. Harding. Orthomodularity in dagger biproduct categories. Preprint, 2008. [Houston, 2006] R. Houston. Finite products are biproducts in a compact closed category. Journal of Pure and Applied Algebra 212, 2006. arXiv:math/0604542 [Isham, 1995] C. J. Isham/ Lectures on Quantum Theory. Imperial College Press, 1995. [Isham and Butterﬁeld, 1998] C. J. Isham and J. Butterﬁeld. Topos perspective on the KochenSpecker theorem: I. Quantum states as generalized valuations. International Journal of Theoretical Physics 37, 2669–2733, 1998. arXiv:quantph/9803055 [Jagadeesan, 2004] R. Jagadeesan. Private communication, 2004. [Joyal and Street, 1991] A. Joyal and R. Street. The geometry of tensor calculus I. Advances in Mathematics 88, 55–112, 1991. [Joyal, Street and Verity, 1996] A. Joyal, R. Street and D. Verity. Traced monoidal categories. Proceedings of the Cambridge Philosophical Society 119, 447–468, 1996. [Jozsa, 2005] R. Jozsa. An introduction to measurement based quantum computation, 2005. arXiv:quantph/0508124. [Kaufmann, 1991] L. H. Kauﬀman. Knots and Physics. World Scientiﬁc, 1991. [Kauﬀman and Lomonaco, 2002] L. H. Kauﬀman and S. J. Lomonaco Jr. Quantum entanglement and topological entanglement. New Journal of Physics 4, 1–18, 2002. [Kauﬀman, 2005] L. H. Kauﬀman. Teleportation topology. Optics and Spectroscopy 99, 227– 232, 2005. arXiv:quantph/0407224 [Kelly, 1972] G. M. Kelly. Manyvariable functorial calculus. In: Coherence in Categories, G. M. Kelly, M. L. Laplaza, G. Lewis and S. Mac Lane (eds), pages 66–105. Lecture notes in Mathematics 281, SpringerVerlag, 1972. [Kelly and Laplaza, 1980] G. M. Kelly and M. L. Laplaza. Coherence for compact closed categories. Journal of Pure and Applied Algebra 19, 193–213, 1980. [Kock, 2003] J. Kock. Frobenius Algebras and 2D Topological Quantum Field Theories. Cambridge University Press, 2003. [Lambek and Scott, 1986] J. Lambek and P. J. Scott. Introduction to HigherOrder Categorical Logic. Cambridge University Press, 1986. [MacLane, 1998] S. MacLane. Categories for the Working Mathematician. 2nd edition. Springer, 1998. [Mackey, 1963] G. M. Mackey. The mathematical foundations of quantum mechanics. W. A. Benjamin, 1963. [Milner, 1989] R. Milner. Communication and Concurrency. Prentice Hall, 1989. [Mitchell, 1965] B. Mitchell. Theory of Categories. Academic Press, 1965. [Nielsen and Chuang, 2000] M. A. Nielsen and L. Chuang. Quantum computation and quantum information. Cambridge University Press, 2000. [Pati and Braunstein, 2000] A. K. Pati and S. L. Braunstein. Impossibility of deleting an unknown quantum state. Nature 404, 164–165, 2000. arXiv:quantph/9911090 [Penrose, 1971] R. Penrose. Applications of negative dimensional tensors. In: Combinatorial Mathematics and its Applications, pages 221–244. Academic Press, 1971. [Petri, 1977] C. A. Petri. Nonsequential processes. Technical Report ISF775, GMD, StAugustin, 1977. [Piron, 1976] C. Piron. Foundations of Quantum Physics. W. A. Benjamin, 1976. [Raussendorf and Briegel, 2001] R. Raussendorf and H.J. Briegel. A oneway quantum computer. Physical Review Letters 86, 5188, 2001. [Raussendorf et al., 2003] R. Raussendorf, D. E. Browne and H.J. Briegel. Measurementbased quantum computation on cluster states. Physical Review A 68, 022312, 2003. arXiv:quantph/ 0301052. [Seely, 1989] R. A. G. Seely. Linear logic, ∗autonomous categories and cofree algebras. In: Categories in Computer Science and Logic, Contemporary Mathematics 92, 1989. [Selinger, 2004] P. Selinger. Towards a quantum programming language. Mathematical Structures in Computer Science 14, 527–586, 2004.
Categorical Quantum Mechanics
323
[Selinger, 2007] P. Selinger. Dagger compact categories and completely positive maps. Electronic Notes in Theoretical Computer Science 170, 139–163. Extended version to appear in New Structures for Physics, B. Coecke (ed). Springer Lecture Notes in Physics, 2007. [Selinger, 2008] P. Selinger. Finite dimensional Hilbert spaces are complete for dagger compact closed categories. In: Proc. Quantum Physics and Logic/Development of Computational Models (QPLDCM). Electronic Notes in Theoretical Computer Science, to appear, 2008. [Selinger, 2008a] P. Selinger. TBA. In: New Structures for Physics, B. Coecke, Ed, Springer lecture Notes in Physics, 2008. [Shor, 1994] P. W. Shor. Algorithms for quantum computation: discrete logarithms and factoring. Proceedings of the 35th Annual IEEE Symposium on Foundations of Computer Science, IEEE Computer Science Press, 1994. [Shor, 1996] P. W. Shor. Faulttolerant quantum computation. In: Proceedings of the 37nd Annual Symposium on Foundations of Computer Science, pages 56–65. IEEE Computer Society Press, 1996. [Simpson, 1999] S. G. Simpson. Subsystems of SecondOrder Arithmetic. SpringerVerlag, 1999. [Spekkens, 2007] R. Spekkens. Evidence for the epistemic view of quantum states: A toy theory. Physical Review A 75, 032110, 2007. arXiv:quantph/0401052 [Street, 2007] R. Street. Quantum Groups: A Path to Current Algebra. Cambridge UP, 207. [Turaev, 1994] V. Turaev. Quantum Invariants of Knots and 3Manifolds, de Gruyter, 1994. [van Tonder, 2003] A. van Tonder. Quantum computation, categorical semantics, and linear logic, 2003. arXiv:quantph/0312174 [Vicary, 2007] J. Vicary. A categorical framework for the quantum harmonic oscillator. International Journal of Theoretical Physics, to appear, 2007. arXiv:0706.0711 [Vicary, 2008] J. Vicary (2008) Categorical formulation of C*algebras. In: Proc. Quantum Physics and Logic/Development of Computational Models (QPLDCM). Electronic Notes in Theoretical Computer Science, to appear, 2008. [von Neumann, 1932] J. von Neumann. Mathematische grundlagen der quantenmechanik. SpringerVerlag, 1932. [Witten, 1988] E. Witten. Topological quantum ﬁeld theory. Communications in Mathematical Physics 117, 353–386, 1988. [Wootters and Zurek, 1982] W. K. Wootters and W. Zurek. A single quantum cannot be cloned. Nature 299, 802–803, 1982. [Yetter, 2001] D. N. Yetter. Functorial Knot Theory. Categories of Tangles, Coherence, Categorical Deformations, and Topological Invariants. World Scientiﬁc, 2001. ˙ ˙ [Zukowski et al., 1993] M. Zukowski, A. Zeilinger, M. A. Horne and A. K. Ekert. Eventreadydetectors’ Bell experiment via entanglement swapping. Physical Review Letters 71, 4287– 4290, 1993.
This page intentionally left blank
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
325
EXTENDING CLASSICAL LOGIC FOR REASONING ABOUT QUANTUM SYSTEMS Rohit Chadha, Paulo Mateus, Am´ılcar Sernadas, and Cristina Sernadas
1
INTRODUCTION
A new logic EQPL (exogenous quantum propositional logic) was proposed in [Mateus and Sernadas, 2004a; Mateus and Sernadas, 2004b; Mateus and Sernadas, 2006] for modeling and reasoning about quantum systems, embodying all that is stated in the relevant Postulates of quantum physics (as presented, for instance, in [CohenTannoudji et al., 1977; Nielsen and Chuang, 2000]). The logic was designed from the semantics upwards, starting with the key idea of adopting superpositions of classical models as the models of the proposed quantum logic. This novel approach to quantum reasoning is diﬀerent from the mainstream approach [Foulis, 1999; Chiara et al., 2004]. The latter, as initially proposed by Birkhoﬀ and von Neumann [Birkhoﬀ and von Neumann, 1936], focuses on the lattice of closed subspaces of a Hilbert space and replaces the classical connectives by new connectives representing the latticetheoretic operations. The former adopts superpositions of classical models as the models of the quantum logic, leading to a natural extension of the classical language containing the classical connectives (just as modal languages are extensions of the classical language). Furthermore, EQPL allows quantitative reasoning about amplitudes and probabilities, being in this respect much closer to the possible worlds logics for probability reasoning than to the mainstream quantum logics. Finally, EQPL is designed to reason about ﬁnite collections of qubits and, therefore, it is suitable for applications in quantum computation and information. The models of EQPL are superpositions of classical valuations that correspond to unit vectors expressed in the computational basis of the Hilbert space resulting from the tensor product of the independent qubit systems. Therefore, in EQPL we can express a wide range of properties of states of such a ﬁnite collection of qubits. For example, we can impose that some qubits are independent of (that is, not entangled with) other qubits; we can prescribe the amplitudes of a speciﬁc quantum state; we can assert the probability of a classical outcome after a projective measurement over the computational basis; and, we can also impose classical constraints on the admissible quantum states.
326
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
Herein, we concentrate on presenting a decidable fragment of EQPL by suitably relaxing the semantic structures of EQPL. Instead of considering Hilbert spaces we work with inner product spaces over an arbitrary real closed ﬁeld and its algebraic closure. The decidability results from the fact that the ﬁrst order theory of such ﬁelds is decidable [Tarski, 1948; Hodges, 1993; Basu et al., 2003]. This technique was inspired by related work on probabilistic logic [Abadi and Halpern, 1994]. Furthermore, the decidable fragment of EQPL so established turns out to be strongly complete although we concentrate on weak completeness. The price we have to pay for decidability is a weak arithmetic language — we loose the analytic aspects of complex numbers. The exogenous approach to extending a given logic is discussed and illustrated in Section 2. Section 3 presents dEQPL step by step: design options, models, language and its interpretation, sound axiomatization, and some useful metatheorems. In Section 4 we show that dEQPL is weakly complete and decidable. The proof of weak completeness can easily be adapted to a proof of strong completeness but we refrained to do so since our primary interest is in applications involving ﬁnitely presented theories. We illustrate the use of dEQPL with two worked examples in Section 5. First we reason about a Bell state. Afterwards, we reason about the quantum teleportation protocol proposed in [Bennett et al., 1993]. Finally, in Section 6 we assess what was achieved and provide an outlook of further developments of the proposed approach to quantum reasoning. 2 EXOGENOUS APPROACH The exogenous semantics approach to enriching a given logic roughly consists of taking as models of the new logic sets of models of the original logic, possibly together with some additional structure. This general mechanism for building new logics is described in detail in [Mateus et al., 2005; Caleiro et al., 2006]. The ﬁrst example of the approach appeared in the context of probabilistic logics [Nilsson, 1986; Nilsson, 1993], although by then not yet recognized as a general construction. The adjective “exogenous” is used as a counterpoint to “endogenous”. For instance, in order to enrich some given logic with probabilistic reasoning it may be convenient to tinker with the models of the original logic. This endogenous approach has been used extensively. For example, the domains of ﬁrstorder structures are endowed with probability measures in [Halpern, 1990]. Other examples include labeling the accessibility pairs with probabilities in the case of Kripke structures [Hansson and Jonsson, 1995] for reasoning about probabilistic transition systems. By not tinkering with the original models and only adding some additional structure on collections of those models as they are, the exogenous approach has the potential for providing general mechanisms for enriching a given logic with some additional reasoning dimension. As we shall see, in our case the exogenous approach has the advantage of closely guiding the design of the language around the underlying concepts of quantum physics while keeping the classical connectives.
Extending Classical Logic for Reasoning about Quantum Systems
327
The exogenous approach of collecting the original models as proposed in [Mateus and Sernadas, 2004a; Mateus and Sernadas, 2004b] is inspired by the possible worlds semantics of modal logic [Kripke, 1963]. It is also akin to the society semantics for manyvalued logic [Carnielli and LimaMarques, 1999] and to the possible translations semantics for paraconsistent logic [Carnielli, 2000]. The possible worlds approach also plays a role in probabilistic logic [Nilsson, 1986; Nilsson, 1993; Bacchus, 1990b; Bacchus, 1990a; Fagin et al., 1990; Abadi and Halpern, 1994; Chadha et al., 2007]. As an introductory example of the exogenous approach, we brieﬂy explain how a probabilistic logic can be obtained from classical propositional logic, following closely [Mateus et al., 2005]. Since quantum reasoning subsumes probabilistic reasoning, this example will also be useful for our purposes. However, before we proceed to explain the probabilistic logic, we ﬁrst concentrate on a fragment of the probabilistic logic called global propositional logic. Global logic is also a fragment of the quantum logic proposed in this paper. We start by taking a set Π of propositional symbols. From a semantic point of view, the models of global logic are sets of valuations over Π. The language of global logic consists of: • Classical propositional formulas constructed from Π using the classical connectives ⊥ and ⇒. • Global formulas constructed from the classical propositional formulas by the global connectives ⊥ ⊥ and . The global connectives mimic the classical connectives in a sense which we will make precise shortly. The satisfaction relation between the semantic models and the formulas is as follows. A model V (V is some set of “classical” valuations) of the global logic satisﬁes a classical propositional formula α if every classical valuation v ∈ V satisﬁes α. Therefore, any classical tautology is a global tautology. Analogous to the case of classical logic, a global valuation V satisﬁes the global formula γ1 γ2 if either V satisﬁes γ2 or V does not satisfy γ1 . The global connective ⊥ ⊥ is never satisﬁed. Clearly this is a copy of the classical propositional logic and indeed, if we replace the classical connectives in a classical tautology by their global counterparts we will get a global tautology. As we just saw, there are two copies of the classical propositional logic in the global logic. A natural question to ask is whether the two copies are necessarily distinct. The answer is yes and while the connectives ⊥ and ⊥ ⊥ collapse, it is not the case with the two implications. However, there is a relation between those two and if V satisﬁes α1 ⇒ α2 then V also satisﬁes α1 α2 . The reverse does not hold in general. There is a sound and strongly complete axiomatization for global logic which contains ﬁve axiom schemas and an inference rule. One axiom schema says that every classical tautology is a global tautology while the other says that replacing classical connectives by their global counterparts results in a global tautology. One
328
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
axiom identiﬁes ⊥ and ⊥ ⊥, a second one axiomatizes the relation between the two implications that we mentioned above, and the last one says that the classical and global conjunctions (global conjunction is introduced as usual) collapse. The inference rule is the global counterpart of modus ponens. Global logic is the ﬁrst step towards creating the exogenous probabilistic logic. The probabilistic logic is obtained “exogenously” by assigning probabilities to each of the classical valuations in a global valuation V . This allows us to reason about the probability that a classical propositional formula is true in V : the probability of ϕ is the sum of the probabilities of the valuations that satisfy ϕ. Given a set Π of propositional symbols, the language of the probabilistic logic consists of: • Classical propositional formulas constructed from Π using the classical connectives ⊥ and ⇒. • A set of terms that include:  realvalued variables and real computable numbers;  probability terms denoting probabilities of classical formulas; and  sum and product of terms. • Comparison formulas of the form t1 ≤ t2 where t1 and t2 are terms. • Formulas constructed from classical propositional formulas and comparison formulas using the global connectives ⊥ ⊥ and . A model for the probabilistic logic, that is a probabilistic valuation, contains a global valuation along with a probability measure which assigns to each classical valuation a real value between 0 and 1. As explained, this gives us an interpretation of the probability terms in the language. The satisfaction of classical formulas is the same as in the global logic. Observe that if V satisﬁes a classical formula α then the probability of α being true is 1 regardless of the probability measure on V . Hence, the probability of a classical tautology in any model is always 1. In order to interpret the variables, the model also contains an assignment of variables to real numbers. This helps to interpret the terms and the comparison formulas in the natural way. The interpretation of the global connectives is the same as before. An axiomatization of probabilistic logic is obtained by extending the axiomatization for global logic as follows. The connection between the classical connectives and probability terms is obtained by three axioms: 1. The probability of any classical tautology is 1. 2. If the probability of the classical formula α1 ∧ α2 is 0 then the probability of α1 ∨ α2 is the sum of the probabilities of α1 and α2 . This is the ﬁnite additivity of probability measures.
Extending Classical Logic for Reasoning about Quantum Systems
329
3. If the probability of the classical formula α1 ⇒ α2 is 1 then the probability of α1 is less than the probability of α2 . This is the monotonicity property of probability measures. For the comparison formulas, an oracle is used which gives the valid comparison formulas. The axiomatization is sound and weakly complete modulo the oracle. However, even with the oracle strong completeness fails as the logic is not compact. The development of the exogenous quantum logic herein follows the same lines as the development of the probabilistic one. Instead of assigning probabilities, we assign amplitudes to the classical valuations in a global valuation. The classical valuations themselves represent the computational basis of the qubits in a quantum system. In fact, we are only interested in quantum systems composed of a ﬁnite number of qubits since applications in quantum computation and information only deal with such systems. A superposition of these classical valuations will then give the state of the quantum system. We will explicitly have terms in the language to interpret these amplitudes and they will be at the core of the design of our language. We postpone the detailed discussion of the language and the logic to Section 3. The resulting quantum logic is a decidable fragment of the logic in [Mateus and Sernadas, 2006]. These quantum logics obtained using the exogenous approach are philosophically closer to some probabilistic logics (like [Fagin et al., 1990; Abadi and Halpern, 1994]) than to the mainstream quantum logics in the tradition of Birkhoﬀ and von Neumann [Birkhoﬀ and von Neumann, 1936; Foulis, 1999; Chiara et al., 2004]. Both types of quantum logic are motivated by semantic considerations, albeit very diﬀerent ones. The mainstream quantum logics are based on the idea of replacing the Boolean algebras of truth values by the more relaxed notion of orthomodular lattices. Thus, they end up with non classical connectives reﬂecting the properties of meets and joins of those lattices. The exogenous quantum logics are based on the idea of replacing classical valuations by superpositions of classical valuations while preserving the classical connectives. On the other hand, in both types of quantum logic a formula and a propositional symbol in particular denotes a subspace of the Hilbert space at hand. However, in the exogenous quantum logics a quantum system is assumed to be composed of n qubits and, hence, the underlying Hilbert space has dimension 2n . Our semantics of quantum logic, although inspired by modal logic, is also completely diﬀerent from the alternative Kripke semantics given to mainstream quantum logics (as ﬁrst proposed in [Dishkant, 1972]). That Kripke semantics is based on orthomodular lattices. The quantum logic proposed in [van der Meyden and Patra, 2003b; van der Meyden and Patra, 2003a; Patra, 2008] is also inspired by probabilistic logics [Fagin et al., 1990] and capitalizes on some techniques ﬁrst proposed for those logics, but it has aspects of both mainstream quantum logics and exogenous quantum logics. In short, it is a classical logic of probabilistic measurements over a quantum system where quantum formulas denote projectors, quantum negation stands for orthogonal complement and quantum conjunction stands for composition.
330
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
Note also that no amplitude terms appear in [van der Meyden and Patra, 2003b; van der Meyden and Patra, 2003a] contrarily to exogenous quantum logics where amplitudes replace probabilities as the central concept. The tensor product plays a key role in the exogenous quantum logics as it does in the categorical semantics proposed in [Abramsky and Coecke, 2004; Abramsky and Duncan, 2006]. However, in our logics we still use the concrete characterization of tensor product of qubits (represented in our language by the propositional symbols). 3 DECIDABLE FRAGMENT OF EQPL We start by discussing design issues, and then proceed to introduce the logic.
3.1 Design issues In this section, we shall discuss how the Postulates of quantum mechanics [CohenTannoudji et al., 1977] guided the design of the proposed logic, and give a brief introduction to the relevant concepts and results. The ﬁrst Postulate of quantum mechanics states: Postulate 1: Every isolated quantum system is described by a Hilbert space. The states of the quantum system are the unit vectors of the corresponding Hilbert space. Please recall that a Hilbert space is a complete inner product space over C (the ﬁeld of complex numbers). In quantum computation and information the quantum systems are composed of qubits. For example, the states of an isolated qubit are vectors of the form z0 0 + z1 1 where z0 , z1 ∈ C and z0 2 + z1 2 = 1. In other words, they are unit vectors in the (unique up to isomorphism) Hilbert space of dimension two. As pointed out in the introduction, instead of working with a Hilbert space we shall consider a “generalized” inner product space over the algebraic closure of an arbitrary real closed ﬁeld. This design decision has the advantage that the resulting logic is decidable. It is possible to work with Hilbert spaces and still get a weaklycomplete calculus as was the case in EQPL [Mateus and Sernadas, 2006], a previous version of the logic developed herein. Indeed, the logic deﬁned here identiﬁes a decidable fragment of EQPL, and hence we shall call it dEQPL. In addition to being decidable, dEQPL turns out to be strongly complete and, therefore, compact. In fact, the source of the non compactness of EQPL mentioned in [Mateus and Sernadas, 2006] was in its arithmetic component. We shall now brieﬂy review some deﬁnitions and results concerning real closed ﬁelds and their algebraic closures. DEFINITION 1 Real closed ﬁelds. An ordered ﬁeld K = (K, +, ., 1, 0, ≤) is said to be a real closed ﬁeld if the following hold:
Extending Classical Logic for Reasoning about Quantum Systems
331
• Every nonnegative element of the K has a square root in K. • Any polynomial of odd degree with coeﬃcients in K has at least one solution in K. We shall use K1 , K2 , . . . to range over real closed ﬁelds and k1 , k2 , . . . to range over the elements of a real closed ﬁeld. The set of real numbers with the usual multiplication, addition and order constitute a real closed ﬁeld. The set of computable real numbers with the same operations is another example of a real closed ﬁeld. The algebraic closure of a real closed ﬁeld K = (K, +, ×, 1, 0, ≤) is obtained by adjoining an element δ to K such δ 2 + 1 = 0. The algebraic closure, denoted by K(δ), is a twodimensional vector space over K. Each element in K(δ) is of the form k1 + k2 δ where k1 , k2 ∈ K. The addition and multiplication are deﬁned as: (k1 + k2 δ) + (k1 + k2 δ) = (k1 + k1 δ) + (k2 + k2 δ) = (k1 .k1 − k2 .k2 ) + (k1 .k2 + k1 .k2 δ) (k1 + k2 δ).(k1 + k2 δ) where −k2 .k2 is the additive inverse of k2 .k2 We shall use c1 , c2 , . . . to range over the elements of K(δ). For example, the ﬁeld of complex numbers is the algebraic closure of the set of real numbers with δ = i. The standard notion of conjugation, absolute value and real and imaginary parts from complex numbers can be generalized to K(δ) as follows: Re(k1 + k2 δ) Im(k1 + k2 δ) k1 + k2 δ k1 + k2 δ
= = = =
k1 k2 k12 + k22 k1 + (−k2 )δ where −k2 is the additive inverse of k2
The conjugation allows us to generalize the notion of inner product and normed vector space over C to an arbitrary K(δ) as follows: DEFINITION 2 K(δ)inner product space. A K(δ)inner product space is a vector space W over the ﬁeld K(δ) together with a map ·, · : W × W → K(δ) such that for all w, w1 , w2 ∈ V and k ∈ K(δ), the following hold: 1. w, w1 + w2 = w, w1 + w, w2 . 2. w, w ∈ K and w, w ≥ 0. 3. w, w = 0 if and only if w = 0. 4. w1 , w2 = w2 , w1 .
332
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
5. w1 , cw2 = cw1 , w2 . DEFINITION 3 K(δ)normed vector space. A K(δ)normed space is a vector space W over the ﬁeld K(δ) together with a map . : W × W → K such that for all w, w1 , w2 ∈ V and k ∈ K, the following hold 1. w ≥ 0. 2. w = 0 if and only if w = 0. 3. cw = cw where c is the absolute value of c. 4. w1 + w2  ≤ w1  + w2 . We shall say that a vector w is a unit vector if w = 1. As in the case of inner product spaces over complex numbers, a K(δ)−inner product space (W, ·, ·) gives rise to a norm by letting: 8 w = w, w. For example, the ﬁeld K(δ) together with the map: c1 , c2 = c1 .c√2 is itself a K(δ)inner product space. In this case, the resulting norm ( c = c.c ) is the absolute value function. Any Hilbert space is a Cinner product space. However, we shall model quantum systems as K(δ)−inner product spaces instead of Hilbert spaces, and the ﬁeld K(δ) will be a part of our semantic structure. Therefore, any theorem we prove in the logic would remain valid if we had just used Hilbert spaces. It is also worthwhile to point out that, unlike Hilbert spaces, K(δ)−inner product spaces in general may not have an analytical structure. So, we will not be able to express properties that necessarily depend upon the analytical structure1 . Moreover, as the logic is intended to be applied for quantum computation and information, we shall work only with a special kind of K(δ)−inner product spaces that are deﬁned by free construction from ﬁnite sets: DEFINITION 4 Free K(δ)inner product space. Given an arbitrary ﬁnite set B, we can construct the free K(δ)inner product space HK(δ) (B) as: • Each element of HK(δ) (B) is a map ψ : B → K(δ). • ψ1 + ψ2 is pointwise addition, i.e., (ψ1 + ψ2 )(b) = ψ1 (b) + ψ2 (b). 1 For
example, we cannot deﬁne the exponential function on an arbitrary K(δ).
Extending Classical Logic for Reasoning about Quantum Systems
333
• cψ is pointwise scalar multiplication, i.e., (cψ)(b) = c (ψ(b)). • The inner product is given by2 ψ1 ψ2 =
ψ1 (b) ψ2 (b).
b∈B
The dimension of the vector space HK(δ) (B) is the cardinality of the set B. Given b ∈ B, let b ∈ HK(δ) (B) be the vector deﬁned as b(b) = 1 and b(b1 ) = 0 for every b1 = b. It can be easily checked that the set {b : b ∈ B} forms a basis of the vector space HK(δ) (B). Furthermore, it is the case that bb = 1 and bb1 = 0 for every b = b1 . For obvious reasons, we say that {b : b ∈ B} is an orthonormal basis of HK(δ) (B). This basis plays an important role in the semantics of dEQPL and for this reason we will henceforth refer to it as being the canonical basis of HK(δ) . A natural question that arises in this context is how do we choose B. The answer lies in our interest in quantum systems composed of qubits. As mentioned before, the states of an isolated qubit are vectors of the form z0 0 + z1 1 where z0 , z1 ∈ C and z0 2 + z1 2 = 1. The set of states can be identiﬁed with (upto isomorphism) the unit vectors in the free Cinner product HC (B) where B is an set of 2 elements. Keeping this is mind, it is natural to represent a qubit by a propositional symbol (henceforth called a qubit symbol) and take B in this case to be the set of two possible classical valuations of the qubit symbol: 0 that assigns false to the qubit symbol and 1 that assigns true to it. Similarly, the states of a isolated pair of qubits are of the form z00 00+z01 01+ z10 10 + z11 11, where z00 , z10 , z01 , z11 ∈ C and z00 2 + z01 2 + z10 2 + z11 2 = 1. The set of states in this case can be identiﬁed with the unit vectors in the free Cinner product HC (B) where B is the set of the four classical valuations over the pair of qubit symbols representing the two qubits at hand. The pattern becomes clear, and in general, we will ﬁx a ﬁnite set of qubit symbols 3 : qB = {qbk : 0 < k ≤ n}. These will represent the n qubits in our system. As we need to work with the algebraic closure of arbitrary real closed ﬁelds, the states in our systems will be unit vectors in the free K(δ)inner product space HK(δ) (2qB ), where 2qB is the set 2 We adopt here the Dirac notation, given its widespread use by the community of quantum physics and computation. 3 In [Mateus and Sernadas, 2006], the set of qubits was inﬁnite. However, the set was restricted when judgments were considered.
334
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
of 2n possible classical valuations of the n qubit symbols. We shall call these unit vectors K(δ)quantum valuations over the set qB. Another characteristic of quantum systems that we are likely to encounter in applications in computation and information is that they will be built from independent subsystems. We shall model the subsystems by partitioning the set qB, and a semantic structure will contain this partition. Each member of the partition, henceforth called a component, will then model the qubits of an independent subsystem. If A ⊆ qB is a component, then the states of the A subsystem will be quantum valuations over A, i.e., unit vectors in HK(δ) (2A ). If S is the partition, then the semantic structure also includes a collection {ϕA : A ∈ S}, where ϕA is a quantum valuation over A. These represent the states of the subsystems. In addition to reasoning about component subsystems, we also need to reason about bigger subsystems. The sets of qubits of bigger subsystems are given by union of qubits of the component subsystems. Therefore, given a partition S of qB, we deﬁne Alg(S) = {∪i Ai : Ai ∈ S}. A member F ∈ Alg(S) models the qubits of the component systems. It is easy to see that Alg(S) satisﬁes the following properties4 : 1. ∅, qB ∈ S. 2. G ∈ S implies that qB \ G ∈ S. 3. G1 , G2 ∈ S implies that G1 ∪ G2 ∈ S We also need a way to construct the states of subsystems from smaller ones. For this, we take recourse to the second Postulate of quantum mechanics: Postulate 2: The Hilbert space of a quantum system composed of a ﬁnite number of independent components is the tensor product of the component Hilbert spaces. Therefore, for instance, the state of a subsystem composed of two independent subsystems is the “tensor product” of the states of the subsystem. Of course, we remember that we are not working with Hilbert spaces. Therefore, we need a deﬁnition of a K(δ)tensor product. For this, we will assume that the reader is familiar with tensor products of vector spaces. Given two vector (K)(δ) vector spaces W1 and W2 , we shall denote the tensor product by W1 ⊗ W2 . Please recall that the vector space W1 ⊗ W2 is generated by vectors of form w1 ⊗ w2 where w1 ∈ W1 and w2 ∈ W2 . We are ready to deﬁne K(δ)tensor products: DEFINITION 5 K(δ)tensor product. The tensor product of two K(δ)inner product spaces (W1 , ·, ·1 ) and (W2 , ·, ·2 ), is the pair (W1 ⊗ W2 , ·, ·), where ·, · is deﬁned as: ai vi ⊗ wi , bj vj ⊗ wj = ai bj vi , vj wi , wj i 4 These
j
i,j
properties deﬁne a structure often called an algebra in probability theory.
Extending Classical Logic for Reasoning about Quantum Systems
335
Observe also that given w ∈ W1 ⊗ W2 it is not always possible to ﬁnd w1 ∈ W1 and w2 ∈ W2 such that w = w1 ⊗ w2 . Furthermore, when that factorization is possible it is not necessarily unique. Please also observe that in our case, the Kvector spaces over the set of qubits A are generated by vectors v where v is a classical valuation over A. Therefore, if S is the partition of qB in our model and A1 , A2 ∈ S then the subsystem composed of A1 and A2 will be generated by vectors of the form v1 ⊗v2 where vi ∈ H(2Ai ). We will identify v1 ⊗ v2 with the vector v1 v2 ∈ H(2A1 ∪A2 ) where v1 v2 is the unique valuation that extends v1 and v2 . Furthermore, the state of subsystem composed of A1 and A2 is the tensor product ψA1 ⊗ ψA2 . (Please note that the tensor product of two unit vectors is again a unit vector.) When given a quantum state ψ ∈ HK(δ) (2qB ) and non empty G qB, we say that the qubits in G are not entangled with the other qubits if there are ψ1 ∈ HK(δ) (2G ) and ψ2 ∈ HK(δ) (2qB\G ) such that ψ = ψ1 ⊗ ψ2 . Therefore, given any G ∈ Alg(S), the qubits in G are not entangled with the other qubits, thanks to the way we build the whole state of the system from the states of the components. Hence, qubits taken from any two independent components of the system are not entangled in every possible quantum state. Please note also that (contrarily to what was adopted in [Mateus and Sernadas, 2006]) we do not require that each component state be non factorisable. This relaxation of the notion of quantum structure had no impact on the entailment relation. Another key concept in the design of our logic is the concept of logical amplitudes. Given a K(δ)quantum valuation ψ and a classical valuation v, the inner product vψ is said to be the logic amplitude of ψ for v. As we shall see, these logical amplitudes are at the core of dEQPL. These amplitudes appear in two ways in the structure which we discuss below. It is also sometimes convenient to work with V 2qb , as we may want to impose classical constraints on the quantum valuations. For example, we may want to impose (qb1 ∨ qb2 ) requiring states to have (logical) amplitude zero for every classical valuation not satisfying this classical formula. In our semantics structures, we shall therefore explicitly have a set V ⊆ 2qB and we shall call V the set of admissible classical valuations. Furthermore, for any v ∈ V , we will require that the amplitude vψ = 0 where ψ is the quantum state of the full system. Note also that every subset A of qB can be identiﬁed with a classical valuation v over qB: v assigns true to qb if and only if qb ∈ A. This, of course, can be G generalized. Any set A ⊂ G ⊂ qB can be identiﬁed with a classical valuation vA G over G: vA assigns true to all elements of A and false to all elements of G \ A. Finally, we also have a collection of K(δ) values ν = {νGA }G⊆qB, A⊆G in the G ψG where semantic structure. We impose that if G ⊂ Alg(S) then νGA = vA ψG is the state of the subsystem composed of qubits modeled by G. In other words, they are logic amplitudes when the qubits in G constitute an independent subsystem. It should be stressed that these values are not always physically meaningful. A
336
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
term νGA is meaningful only if G ∈ Alg(S). The others are nevertheless useful for our purposes and help to avoid partial denotation maps. We are now ready to assemble the diﬀerent pieces of our semantic structure: DEFINITION 6 Quantum structure. A quantum structure over qB is a tuple w = (K, δ, V, S, ψ, ν) where: • K is a real closed ﬁeld and K(δ) is its algebraic closure; • V is a nonempty subset of 2qB ; • S is a partition of qB; • ψ = {ψS }S∈S where each ψS is a unit vector of HS . We extend ψ to Alg(S) as follows: 1. ψ∅ = 1; 2. ψS1 ∪···∪Sn = ψS1 ⊗ · · · ⊗ ψSn ; • vψqB = 0 if v ∈ V ; G ψG if G ∈ Alg(S). In particular, • ν : {νGA }G⊆qB,A⊆G where νGA = vA ν∅∅ = 1.
The proposed quantum logic will be interpreted over these quantum structures. Obviously, we have some redundancy in the notion of quantum structure, namely, ψ can be reconstructed from ν. However, this redundancy pays oﬀ in ease of use and in clarifying the connection to quantum physics. The ﬁrst two Postulates were suﬃcient to guide us in the task of setting up the notion of quantum structures over which we shall be able to deﬁne the semantics of dEQPL. Now, we turn our attention to the Postulates concerning measurements of physical quantities. Postulate 3: Every measurable physical quantity of an isolated quantum system is described by an observable acting on its Hilbert space.
Please recall that an observable is a Hermitian operator such that the direct sum of its eigensubspaces coincides with the underlying Hilbert space. Also recall that the spectrum Ω of a Hermitian operator (set of its eigenvalues) is a subset of the set of real numbers, R. For each e ∈ Ω, we denote the corresponding eigensubspace by He , and the projector onto the subspace Ee by Pe . It might seem at ﬁrst that we need to extend the deﬁnition of Hermitian operators to an arbitrary K(δ) as Hermitian operators are usually deﬁned over Hilbert
Extending Classical Logic for Reasoning about Quantum Systems
337
spaces. However, as we shall see shortly, fortunately that is not required. This is because we do not have constructs in the language for denoting such measurement operators. In order to use Postulate 3, we need to consider Postulate 4. Postulate 4: The possible outcomes of the measurement of a physical quantity are the eigenvalues of the corresponding observable. When the physical quantity is measured using observable A on a system in a state ψ, the resulting outcomes A = (Ω, EΩ , μA are ruled by the probability space Pψ ψ ) where (in the case A has a countable spectrum)  Ω is the spectrum of the observable A,  EΩ is ℘Ω the powerset of Ω , and  μA ψ : EΩ → R is the probability measure deﬁned as μA ψ (E) =
2
Pe ψ .
e∈E
For the applications in quantum computation and information that we have in mind, only logic projective measurements are relevant. Given a quantum system with the set of qubits qB and a set of classical valuations V , these are measurements A such that:  The spectrum of A is equipotent 5 to V , i.e., there is a bijection between the spectrum of A and V .  If we identify V with the spectrum of A then for each v ∈ V , the corresponding eigenspace Hv is generated by the vector v. The projector Pv is the operator vv, i.e., Pv ψ = v, ψ v for each vector ψ ∈ HC (2qB ). Postulate 4 then tells us that the stochastic result of making a logic projective measurement A given a quantum structure w = (K, δ, V, S, ψ, ν) is described by the ﬁnite probability space Pw = (V, ℘V, μw ) where for each U ⊆ V : 2 (1) μw (U ) = vψ . v∈U
For example, if the quantum system is in the particular state α00ω1 00ω1 + α01ω2 01ω2 + α01ω3 01ω3 + α10ω4 10ω4 5 The chosen bijection depends on how the qubits are physically implemented. For example, when implementing a qubit using the spin of an electron, we may impose that spin + 12 corresponds to true and spin − 12 corresponds to false. But, as we shall see, the semantics of EQPL does not depend on the choice of the bijection, as long as one exists. The same happens in the case of classical logic — its semantics does not depend on how bits are implemented. The details of which voltages correspond to which truth values are irrelevant.
338
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
then the probability of observing the ﬁrst two qubits qb0 , qb1 in the classical valuation 01 (here we take V as {00ω1 , 00ω2 , 00ω3 , 00ω4 }) is given by α01ω2 2 + α01ω3 2 . We have probability terms in the language of the proposed logic and Equation 1 is all that we need from Postulates 3 and 4 for interpreting them as we shall see in Section 3.2. Once again, we recall that we are working with an arbitrary real closed ﬁeld. Given a quantum structure w = (K, δ, V, S, ψ, ν), we deﬁne the probability map μw : ℘(V ) → K as: 2 (2) μw (U ) = vψ . v∈U
The essential diﬀerence between Equation 1 and 2 is that summands in the former are real numbers while the summands of the latter one are elements of a real closed ﬁeld given by the quantum structure. It is easy to check that μV deﬁned in Equation 2 satisﬁes the “usual” ﬁnite probability axioms: 1. μV (∅) = 0 and μV (V ) = 1, and 2. μV (U1 ∪ U2 ) = μV (U1 ) + μV (U2 ) if U1 and U2 are disjoint sets. Therefore, given a quantum 9 structure w, we have the means for interpreting dEQPL terms of the form ( α) that denote probabilities. Finally, although irrelevant to the design of dEQPL, we mention en passant Postulate 5 that rules how quantum systems evolve beyond measurements: Postulate 5: Excluding measurements, the evolution of a quantum system is described by unitary transformations. This last Postulate becomes relevant only when designing a dynamical extension of the logic (see for instance [Mateus and Sernadas, 2004b]).
3.2 Language and semantics There are two kinds of terms in dEQPL, one denoting elements of real closed ﬁeld in the quantum structure and the other denoting elements in its algebraic closure. The formulas of dEQPL, henceforth called quantum formulas, are constructed from classical propositional formulas, formulas denoting subsystem and comparison formulas (comparing terms denoting elements of real closed ﬁelds) using global connectives introduced in Section 2. We present language of dEQPL in Table 1 using an abstract version of BNF notation [Naur, 1963] for a compact presentation of inductive deﬁnitions. We discuss the language in detail below. The ﬁrst syntactic category is classical formulas. Please recall that we ﬁxed a ﬁnite set of qubit symbols qB. Classical formulas are built from qubit symbols in qB using the classical disjunctive connectives, falsum ⊥ and implication ⇒. As usual,
Extending Classical Logic for Reasoning about Quantum Systems
339
Classical formulas α := ⊥ qb (α ⇒ α) Term language (with the proviso A ⊆ G ⊆ qB) 9 t := x 0 1 (t + t) (t t) Re(u) Im(u) u ( α) u := z %GA t + it u (u + u) (u u) (α u; u) Quantum formulas (with the proviso F ⊆ qB): γ := α (t ≤ t) [F ] ⊥ ⊥ (γ γ)
Table 1. Language of dEQPL other classical connectives like ¬, ∧, ∨, ⇔ and % are introduced as abbreviations. We denote the set of qubit symbols occurring in α by qB(α), and say that a classical formula α is over a set S of qubit symbols if qB(α) ⊆ S. For the term language, we pick two disjoint denumerable sets of variables. The ﬁrst set of variables X = {xk : k ∈ N} is interpreted in the real closed ﬁeld of the quantum structure, and the second set Z = {zk : k ∈ N} is interpreted in the closure of the real closed ﬁelds. As we shall see in Section 5, variables are often useful for applications that we have in mind. There are two syntactic categories t and u for terms, which are mutually deﬁned. The syntactic category t denotes the elements of a real closed ﬁeld and u denotes the elements of its closure respectively. We will often abuse the notation by saying that t is a real term and u is a complex term. Most of the term constructs are selfexplanatory and already motivated in the previous section. The term %GA denotes the logical amplitude νGA in the quan9 tum structure, and henceforth will be called an amplitude term. The term ( α) denotes the probability that classical formula α holds for an outcome of a logical projective measurement, and will be called a probability term. The denotation of the alternative term (α u1 ; u2 ) will be the value denoted by u1 if α is true, and the value denoted by u2 otherwise. As usual, we may deﬁne the notion of occurrence of a term t1 in a term t, and the notion of replacing zero or more occurrences of terms t1 in t by t2 . If x, t, z and u are sequences of real variables, real terms, complex variables and complex terms respectively, we will write t{x/t, z/u} to mean the real term obtained by substituting all occurrences of xi by ti and all occurrences of zj by uj . The complex term u{x/t, z/u} is similarly deﬁned. The quantum formulas are built from classical formulas α, subsystem formulas [F ] and comparison formulas (t ≤ t) using the connectives ⊥ ⊥ and . The formulas consisting of just the classical formulas, subsystem and comparison formulas are
340
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
called quantum atoms, and the set of quantum atoms shall henceforth be called qAtom. We shall use δ, δ to range over elements of qAtom. Please note that quantum bottom ⊥ ⊥ and quantum implication are global connectives and should not be confused with their classical (local) counterparts. The notion of occurrence of a term t in a quantum formula γ can be easily deﬁned. However, we have to be careful while deﬁning the notion of occurrence of a quantum formula γ in the quantum formula γ1 . This is because we want γ to occur as a quantum subformula of γ1 and rule out situations where γ occurs as classical subformula. More precisely, we deﬁne γ1 qoccurs in γ inductively as: • if γ is a classical formula, a comparison formula, a subsystem formula, or ⊥ ⊥, then γ1 qoccurs in γ if and only if γ1 is γ and; • if γ is γ γ then γ1 qoccurs in γ if and only if one of the following holds: – γ1 is γ, or – γ1 qoccurs in γ , or – γ1 qoccurs in γ . The notion of replacing zero or more qoccurrences of a quantum formula γ1 in γ by γ can now be suitably deﬁned. For example, the classical formula qb qoccurs in (qb qb1 ) and replacing one qoccurrence of qb by qb2 will yield the quantum formula (qb2 qb1 ). On the other hand qb does not qoccur in (qb ⇒ qb1 ) (qb is a classical subformula and not quantum subformula). The replacement qb by qb2 in (qb ⇒ qb1 ) has no eﬀect. Similarly, qb does not qoccur in [{qb}]. For clarity sake, we shall often drop parenthesis in formulas and terms if it does not lead to ambiguity. As expected, other quantum connectives will be introduced as abbreviations. However, before introducing a whole set of useful abbreviations, we present the semantics of the language. The language is interpreted in a quantum structure as deﬁned in Section 3.1. Given a quantum structure w = (K, δ, V, S, ψ, ν), recall that K is a real closed ﬁeld with K(δ) as its algebraic closure, V is a set of valuations over qB, S is a partition of qB, ψ is a collection of K(δ)quantum states, and ν is a collection of amplitude terms. We shall assume the semantics of classical propositional logic, and say that v c α if the classical valuation v satisﬁes the classical formula α. For interpreting the probability terms, we shall use the probability map μw : ℘(V ) → K deﬁned in Section 3.1 as: 2 vψ . μw (U ) = v∈U
For the probability terms, we shall also need the extent at a set V of classical formulas over S deﬁned as: αV = {v ∈ V : v c α}.
Extending Classical Logic for Reasoning about Quantum Systems
341
For interpreting the variables, we need the concept of an assignment. Given a real closed ﬁeld K, a Kassignment ρ is a map such that ρ(x) ∈ K for each x ∈ X and ρ(z) ∈ K(δ) for each z ∈ Z. Please note that when K is clear from the context, we shall drop K. Given a quantum structure w = (K, δ, V, S, ψ, ν) and a Kassignment ρ. The denotation of terms and satisfaction of quantum formulas at w and ρ and is inductively deﬁned in Table 2 (omitting the obvious ones). Denotation of terms [[x]]wρ [[t19 + it2 ]]wρ [[( α)]]wρ [[z]]wρ [[%GA ]]wρ [[(α u1 ; u2 )]]wρ
= = = = = =
ρ(x) [[t1 ]]wρ + δ[[t2 ]]wρ μw (αV ) ρ(z) νGA [[u1 ]]wρ if αV = V [[u2 ]]wρ otherwise
Satisfaction of quantum formulas wρ α iﬀ αV = V w wρ (t1 ≤ t2 ) iﬀ [[t1 ]]w ρ ≤ [[t2 ]]ρ wρ [A] iﬀ A ∈ Alg(S) wρ ⊥ ⊥ wρ (γ1 γ2 ) iﬀ wρ γ1 or wρ γ2
Table 2. Semantics of dEQPL Please observe that the set V is suﬃcient to interpret the classical formulas, and the partition S is suﬃcient to interpret the subsystem formulas. The Kassignment ρ is suﬃcient to interpret a useful sublanguage of the formulas deﬁned as: κ := (a ≤ a) ⊥ ⊥ (κ κ) a := x 0 1 (a + a) (a a) Re(b) Im(b) b b := z a + ia b (b + b) (b b) Henceforth, the terms of this sublanguage will be called arithmetical terms and the formulas will be called arithmetical formulas. We may use the satisfaction relation to deﬁne entailment as expected: we say that a set of quantum formulas Γ entails a quantum formula η, written Γ η, if wρ η for every w and ρ satisfying every element of Γ. We say a quantum formula η is valid when it is entailed by the empty set of quantum formulas. Please note also that the metatheorem of entailment holds: Γ, η1 η2 iﬀ Γ (η1 η2 ). That is, quantum implication internalizes the notion of quantum entailment. The following are some examples of entailment:
342
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
[G1 ], [G2 ] α [G]
(¬ α) ( α) (α1 ∧ α2 ) ≡ (α1 $ α2 ) [G91 ∩ G2 ] ((α) = 1) (( A⊆G %GA 2 ) = 1)
We shall now present some useful abbreviations, and give some small examples.
3.3 Abbreviations and examples As anticipated, the proposed quantum language with the semantics above is rich enough to express interesting properties of quantum systems. To this end, it is quite useful to introduce other operations, connectives and modalities through abbreviations. We start with some additional quantum connectives: • quantum negation: ( γ) for (γ ⊥ ⊥); • quantum disjunction: (γ1 γ2 ) for (( γ1 ) γ2 ); • quantum conjunction: (γ1 $ γ2 ) for ((( γ1 ) ( γ2 ))); • quantum equivalence: (γ1 ≡ γ2 ) for ((γ1 γ2 ) $ (γ2 γ1 )). It is also useful to introduce some additional comparison formulas: • (t1 < t2 ) for ((t1 ≤ t2 ) $ ((t2 ≤ t1 ))); • (t1 = t2 ) for ((t1 ≤ t2 ) $ (t2 ≤ t1 )); • (u1 = u2 ) for ((Re(u1 ) = Re(u2 )) $ (Im(u1 ) = Im(u2 ))) Please note that the only constants in our term language are 0 and 1. As every real closed ﬁeld K has characteristic 0, we can embed a copy of rationals in K. It is also possible to take square roots of positive numbers. Hence, it will be useful to use the following abbreviations (with the proviso n > 0): • (t = n) for t = • (t =
m n)
• (t1 =
√
((1 + (1 + . . . ...))) !" # ; n times
for ((m.t) = n);
t2 ) for ((t2 ≥ 0) $ (t1 2 = t2 )).
Given A ⊆ G ⊆ qB, the following classical formula will also be useful: • (∧G A) is ((∧qbk ∈ A qbk ) ∧ (∧qbk ∈ G\A (¬ qbk )).
Extending Classical Logic for Reasoning about Quantum Systems
343
The classical formula (∧G A) speciﬁes the unique classical valuation that satisﬁes all the qubit symbols in A and does not satisfy the qubit symbols in G \ A. We will often need this classical formula in the case the set G is the full set of qubit symbols qB. Therefore, we will often use the following abbreviation • (∧A) for (∧qB A). The logical amplitude terms, %GA , are easily extendible to any classical formula as (with the provisos qB(α) ⊆ G and A ⊆ G ⊆ qB): • αGA for (((∧G A) ⇒ α) %GA ; 0). Intuitively, the amplitude term αGA coincides with %GA when the valuation ∧G A satisﬁes with α and is 0 otherwise. We will often use this term in the case G is the full set of qubit symbols qB. Therefore, the following abbreviation will also be useful: • αA for αqBA . We introduce a couple of probability modalities as abbreviations: 9 • (♦α) for (0 < ( α)); 9 • (α) for (1 = ( α)). Finally, we can also deﬁne a quantum modality as an abbreviation: • ([G]♦ α : u) for ([G] $ (u > 0) $ (A⊆G (αGA = u))). Intuitively ([G]♦ α : u) is true iﬀ G is a subsystem, there is a subset A of G such that the classical valuation ∧G A satisﬁes α and the logical amplitude %GA takes the nonzero value u. We discuss a small example where we demonstrate the usefulness of dEQPL to specify properties of a quantum system. We postpone the discussion of more involved examples to Section 5. Consider the following variant of Schr¨ odinger’s cat. The attributes of the cat that we consider are: being inside or outside the box, alive or dead, and moving or still. We choose three qubit symbols qb0 , qb1 , qb2 to represent these attributes. For the sake of readability, we use catinbox, catalive and catmoving instead of the symbols qb0 , qb1 and qb2 respectively. The following dEQPL formulas constrain the state of the cat at diﬀerent levels of detail: 1. [catinbox, catalive, catmoving]; 2. (catmoving ⇒ catalive); 3. ((♦ catalive) $ (♦ (¬ catalive))); 4. ([catalive]); 9 5. (( catalive) = 13 );
344
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
9 6. ([catalive, catmoving] $ (( catalive ∧ catmoving) = 16 ) 9 $ (( catalive ∧ (¬ catmoving)) = 16 ) 9 $ (( (¬ catalive) ∧ (¬ catmoving)) = 23 )). Please observe that all the above assertions are consistent with each other. Intuitively, the ﬁrst assertion states that the qubits catinbox, catalive and catmoving form a subsystem and therefore, are not entangled with the other qubits of the cat system. The second is a classical constraint on the set of admissible valuations: if the cat is moving then it is alive. The third assertion is a consequence of the famous paradox: the cat can be in a state where it is possible that the cat is alive and it is possible that the cat is dead. The fourth assertion states that the qubit catalive is necessarily entangled with other qubits. The ﬁfth assertion states that the cat is in a state where the probability of observing it alive (after collapsing the wave function) is 13 . Finally, the sixth assertion states that the qubits catalive, catmoving are not entangled with other qubits, and that the cat is in quantum state where: the probability of observing it alive and moving is 16 , the probability of observing it alive and not moving is 16 , and the probability of observing it dead (and, thus also not moving by second assertion) is 23 .
3.4 The axiomatization We shall present a Hilbertstyle axiomatization of the dEQPL. We need two new concepts for the axiomatization, one of quantum tautology and the second of a valid arithmetical formula. Let P be a countable set of propositional symbols disjoint from qB. Given a classical formula β over P, let βq be the syntactic entity obtained by replacing all occurrences of ⊥ by ⊥ ⊥ and ⇒ by . A quantum formula σ is said to be a quantum tautology if there is a classical tautology β over P and a map σ : P → qAtom such that σ coincides with βq σ where βq σ is the quantum formula obtained from βq by replacing each p ∈ P by σ(p). For instance, the quantum formula ((x1 ≤ x2 ) (x1 ≤ x2 )) is tautological (obtained, for example, from the classical tautology p ⇒ p). Please recall that an arithmetical formula in the dEQPL is any formula that does not have probability terms, amplitude terms, alternative terms, classical formulas and subsystem formulas. As noted in Section 3.2, given an quantum structure with K0 as the underlying real closed ﬁeld, a K0 assignment is enough to interpret all arithmetical formulas. We say that an arithmetical formula κ is a valid arithmetical formula if it holds for any assignment that maps variables into an arbitrary real closed ﬁeld K. Clearly, a valid arithmetical formula holds for all semantic structures of dEQPL. It is a wellknown fact from the theory of quantiﬁer elimination [Hodges, 1993; Basu et al., 2003] that the set of valid arithmetical formulas so deﬁned is decidable6 . However, we shall not go into details of this 6 For the arithmetical sublanguage, we may treat the global connectives as classical connectives.
Extending Classical Logic for Reasoning about Quantum Systems
345
result as we want to focus our attention on reasoning about quantum aspects only. The axioms and inference rules of dEQPL are listed in Table 3. In total, we have two inference rules and sixteen axioms. The two inference rules are modus ponens for classical implication CMP and modus ponens for quantum implication QMP7 . The axioms are better understood in the following groups.
Axioms [CTaut] [QTaut]
α for each classical tautology α γ for each quantum tautology γ
[Lift⇒] [Eqv⊥] [Ref$]
((α1 ⇒ α2 ) (α1 α2 )) (⊥≡ ⊥ ⊥) ((α1 $ α2 ) (α1 ∧ α2 ))
[Sub∅] [Sub∪] [Sub\]
[∅] ([G1 ] ([G2 ] [G1 ∪ G2 ])) ([G] ≡ [qB \ G])
[RCF]
κ{x/t , z/u} where κ is a valid arithmetical formula, x, z, t and u are sequences of real variables, complex variables, real terms and complex terms respectively
[If%] [If⊥]
(α ((α u1 ; u2 ) = u1 )) (( α) ((α u1 ; u2 ) = u2 ))
[Empty] [NAdm] [Unit] [Mul]
(%∅∅ = 1) ((¬(∧A)) (%qBA = 0)) ([G] (( A⊆G %GA 2 ) = 1)) (([G1 ] $ [G2 ]) (%G1 ∪G2 A1 ∪A2 = %G1 A1 %G2 A2 )) where G1 ∩ G2 = ∅, A1 ⊆ G1 and A2 ⊆ G2
[Prob]
9 (( α) = ( A αA 2 ))
Inference rules [CMP] α1 , (α1 ⇒ α2 ) α2 [QMP] γ1 , (γ1 γ2 ) γ2 Table 3. Axioms for dEQPL 7 Actually,
CMP can be derived from QMP and Lift⇒.
346
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
We have as axioms the classical tautologies and the quantum tautologies (CTaut and QTaut, respectively). Since the set of classical tautologies and the set of quantum tautologies are both recursive, there is no need to spell out the details of tautological reasoning. The axioms Lift⇒, Eqv⊥ and Ref$ are suﬃcient to relate (local) classical reasoning and (global) quantum tautological reasoning. These are exactly the axioms that relate classical connectives and global connectives in global logic (see Section 2). We refer to [Mateus et al., 2005] for more details. The axioms Sub, Sub∪, and Sub\ are enough to reason about subsystems. Together, they impose that subsystems are closed under settheoretic operations (closure under intersection and set diﬀerence appear as theorems). The axiom RCF says that if κ is a valid arithmetical formula, then any formula obtained by replacing variables with the terms of dEQPL is a tautology. Since the set of valid arithmetical formulas is recursive, we refrain from spelling out the details. The axioms If% and If⊥ are selfexplanatory, and will be used in the completeness proof to remove alternative terms. The axioms Empty, NAdm, Unit and Mul rule logical amplitudes. Each of them closely reﬂects a property of our semantic structures. The axiom empty says that the logical amplitude %∅∅ is always 1. The axiom Unit says that the state of each subsystem is a unit vector. The axiom NAdm says that the amplitude of a nonadmissible classical valuation is 0. The axiom Mul says that the state of a system composed of two subssystems is a tensor product of the two subsystems. Finally, the axiom Prob relates probabilities and amplitudes, closely following Postulate 4 of quantum mechanics. As expected, we say that a formula γ is a theorem, written γ, if we can build a derivation of γ from the axioms using the inference rules. We say that a (possibly inﬁnite) set of formulas Γ derives γ, written Γ γ, if we can build a derivation of γ from axioms and the inference rules using formulas in Γ as hypothesis. As an illustration of the axiomatization, we establish the following theorems: PROPOSITION 7. For any classical formulas α1 , α2 , we have
[Lift∧] [PUnit]
Proof. Derivation of [Lift∧]:
(α91 ∧ α2 ) (α1 $ α2 ). (( %) = 1).
Extending Classical Logic for Reasoning about Quantum Systems
347
1
(α1 ∧ α2 ) ⇒ α1
CTaut
2
((α1 ∧ α2 ) ⇒ α1 ) ((α1 ∧ α2 ) α1 )
Lift⇒
3
(α1 ∧ α2 ) α1
QMP:1,2
4
(α1 ∧ α2 ) ⇒ α2
CTaut
5
((α1 ∧ α2 ) ⇒ α2 ) ((α1 ∧ α2 ) α2 )
Lift⇒
6
(α1 ∧ α2 ) α2
QMP:4,5
7
((α1 ∧ α2 ) α1 ) (((α1 ∧ α2 ) α2 ) ((α1 ∧ α2 ) (α1 α2 )))
Qtaut
8
((α1 ∧ α2 ) α2 ) ((α1 ∧ α2 ) (α1 α2 ))
QMP:3,8
9
(α1 ∧ α2 ) (α1 α2 )
QMP:6,8
Derivation of [PUnit] 1
[∅]
Sub∅
2
[∅] [qB]
Sub\
3
[qB]
QMP:1,2
4
([qB] ((
5
((
6 7 8 9
P A⊆qB
R
P A⊆qB
Unit
qBA 2 ) = 1)
(( ) = (
P
R
A⊆qB
((( ) = ( (((
qBA 2 ) = 1))
P A⊆qB
QMP:3,4
qBA 2 ))
P A⊆qB
Prob
qBA 2 )) (((
P A⊆qB
R
qBA 2 ) = 1) (( ) = 1)))
R
qBA 2 ) = 1) (( ) = 1))
R
(( ) = 1)
RCF QMP:6,7 QMP:5,8
We ﬁnish this section with a list of interesting theorems. The ﬁrst three shall be proved in Section 3.6 using the metatheorems of the logic. The ﬁrst two relate local equivalence and negation with their global counterparts, while the third one says subsystems are closed under set intersection and the fourth one says that subsystems are closed under set diﬀerence. [Lift⇔] [Lift¬] [Sub∩] [SubDiﬀ]
(α1 ⇔ α2 ) (α1 ≡ α2 ). ¬ α α. F ([G1 ] ([G2 ] [G1 ∩ G2 ])). F ([G1 ] ([G2 ] [G1 \ G2 ])).
The following theorems give some insight on the major properties of logical amplitudes.
348
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
[AAdd] [AMon] [ASoE] [ANec] [AMExc]
(((α1 ∨ α2 )G + (α1 ∧ α2 )G ) = (α1 G + α2 G )). ((α1 ⇒ α2 ) (α1 G  ≤ α2 G )). ((α1 ⇔ α2 ) (α1 G = α2 G )). (α (αG = %G )). ((αG + (¬ α)G ) = %G ) .
The ﬁrst of the following theorems about probability after measurements just states ﬁnite additivity. The second relates logical reasoning with probability reasoning (monotonicity). These two theorems and the theorem PUnit illustrated in Proposition 7 are axioms in the exogenous probabilistic logic discussed in Section 2. 9 9 9 9 [PAdd] ((( (α1 ∨ α2 )) +9 ( (α1 ∧ 9α2 ))) = (( α1 ) + ( α2 ))). [PMon] ((α1 ⇒ α2 ) (( α1 ) ≤ ( α2 ))). The following theorems show that probability modalities behave as normal modalities. [PNec] [PNorm]
(α (α)). (((α ⇒ α )) ((α) (α ))).
The quantum modalities also behave as normal modalities. [QNorm] [QMon] [QCong]
(([G]♦ (α ∨ α ) : u) ≡ (([G]♦ α : u) ([G]♦ α : u))) . ((α ⇒ α ) (([G]♦ α : u) ([G]♦ α : u))). ((u = u ) (([G]♦ α : u) ([G]♦ α : u ))) .
3.5 Soundness We now show that the calculus is strongly sound, i.e., if Γ γ then Γ γ. It suﬃces to show that each of the axioms is valid, i.e., if γ1 is an axiom, then every semantic structure satisﬁes γ1 . LEMMA 8. The axiom QTaut is valid. Proof. Assume that β is a classical tautology over the set of propositional symbols P and let σ : P → qAtom be a map from P into quantum atoms. We show that βq σ is valid in all models of dEQPL. Take an arbitrary quantum structure w = (K, δ, V, S, ψ, ν), and consider the classical valuation v over P such that 1 if wρ σ(p) . v (p) = 0 otherwise We show that for any classical formula β over P v β iﬀ wρ βq σ
Extending Classical Logic for Reasoning about Quantum Systems
349
by induction on the structure of β as follows. If β is a propositional symbol then it follows from the deﬁnition of v . The case where β is the connective ⊥ is immediate. If β is (β1 ⇒ β2 ), then v satisﬁes β2 or v does not satisfy β1 . If v satisﬁes β2 then by induction hypothesis wρ (β2 )q . If v does not satisfy β1 , then by induction hypothesis once again, v (β1 )q . Therefore, in either case, wρ (β1 )q (β2 )q . Now, note that βq is (β1 )q (β2 )q . The lemma now follows by observing that v β. LEMMA 9. The axioms are valid. Proof. The axioms CTaut, Eqv⊥, RCF, If%, If⊥, Empty, Sub∅, Sub∪ and Sub\ are easy to show. For the rest, let w = (K, δ, V, S, ψ, ν) be a quantum structure and ρ a Kassignment. We consider the other axioms one by one: • Lift⇒. Assume that wρ (α1 ⇒ α2 ). Then, by deﬁnition, all classical valuations in V must satisfy (α1 ⇒ α2 ). Therefore, if all classical valuations in V satisfy α1 they must satisfy α2 also. Hence, either α1 V = V or α2 V = V . We conclude, by deﬁnition, w α1 α2 . • Ref$. Similar to the axiom Lift⇒. • NAdm. Assume that wρ (¬(∧A)). This means that the classical valqB that assigns 1 to the qubit symbols in A and 0 to all other uation vA qB ψqB = 0 and hence qubits is not an element of V . Therefore, νqBA = vA wρ %qBA = 0. 9 • Prob. Using the deﬁnition [[( α)]]wρ = μw (αV ) = v∈α vψqB 2 , it V suﬃces to show that
vψqB 2 = [[ αA 2 ]]wρ . A
v∈αV
Please note that [[αA ]]wρ =
νqBA 0
qB if vA ∈ αV . otherwise
qB ψqB . Therefore, Also, by deﬁnition, νqBA = vA
[[ αA 2 ]]wρ = A
qB vA ψqB 2
qB vA ∈αV
qB We conclude by observing that every v is a vA for some unique A ⊆ qB.
350
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas G • Unit. Assume that wρ [G]. Then G ∈ Alg(S). Please note that {vA : G A ⊆ G} forms an orthonormal basis of H(2 ). Hence, G G vA ψG vA . ψG = A⊆G G Again, by deﬁnition, vA ψG = νGA and so G ψG = νGA vA . A⊆G
Since ψG is a unit vector, we get
νGA 2 = 1.
A⊆G
We conclude by noting that [[%GA ]]wρ = νGA by deﬁnition. • Mul. Assume that wρ [G1 ] $ [G2 ] where G1 ∩ G2 = ∅. Then G1 , G2 ∈ Alg(S). The deﬁnition of quantum structure says that ψG1 ∪G2 = ψG1 ⊗ ψG2 . G1 ∪G2 G1 G2 = vA ⊗ vA . The deﬁnition of tensor product says that vA 1 ∪A2 1 2
The deﬁnition of quantum structure gives G1 ∪G2 νG1 ∪G2 A1 ∪A2 = vA ψG 1 ∪A2
1 ∪G2
.
The deﬁnition of tensor product then gives, G1 G2 ⊗ vA  ψG1 ⊗ ψG2 G νG1 ∪G2 A1 ∪A2 = vA 1 2
1 ∪G2
G1 G2 = vA ψG vA ψG . 1 2 1
2
G1 G2 ψG and νG2 A2 is vA ψG . We conclude by observing that νG1 A1 is vA 1 2 1
2
THEOREM 10 Soundness. The proof system of dEQPL is sound. Proof. The proof now follows by induction on the number of steps in the derivation.
3.6 Metatheorems We now prove some useful metatheorems for dEQPL. We start by showing that the inference rule Hypothetical Syllogism holds for dEQPL. LEMMA 11 Hypothetical Syllogism. Let γ1 , γ2 , γ3 be quantum formulas. Then, [HypSyl] Γ γ1 γ2 and Γ γ2 γ3 imply Γ γ1 γ3 .
Extending Classical Logic for Reasoning about Quantum Systems
351
Proof. Observe that by QTaut, (γ1 γ2 ) ((γ2 γ3 ) (γ1 γ3 )). The proposition follows by using two instances of QMP.
The inference rule HypSyl is a useful rule as illustrated in the derivation of the theorem Lift¬ below: PROPOSITION 12. For any classical formula α, we have [Lift¬]
¬ α α.
Proof. 1
((⊥ ⊥ ⊥) (⊥ ⊥ ⊥))
Eqv⊥
2
((⊥ ⊥ ⊥) (⊥ ⊥ ⊥)) (⊥ ⊥ ⊥)
QTaut
3
(⊥ ⊥ ⊥)
QMP: 1,2
4
(⊥ ⊥ ⊥) ((α ⊥) (α ⊥ ⊥))
QTaut
5
(α ⊥) (α ⊥ ⊥)
QMP: 3,4
6
(α ⇒ ⊥) (α ⊥)
Lift⇒
7
(α ⇒ ⊥) (α ⊥ ⊥)
HypSyl: 5,6
The axiomatization also enjoys the metatheorem of deduction: THEOREM 13 Metatheorem of deduction. Let Γ be a set of quantum formulas and γ1 , γ2 be quantum formulas. Then, Γ ∪ {γ1 } γ2 iﬀ Γ γ1 γ2 . Proof. (←) Assume that Γ γ1 γ2 . Let Π be a proof of the derivation Γ γ1 γ2 and assume that the length of Π is n. We can extend Π to obtain Γ ∪ {γ1 } γ2 as follows: n n+1 n+2
γ1 γ2 γ1 γ2
Π Hyp QMP: n,n+1
(→) Assume that Γ ∪ {γ1 } γ2 . We will prove Γ γ1 γ2 by induction on n, the length of proof of Γ ∪ {γ1 } γ2 . The base step n = 1 will be subsumed by the inductive step. In the inductive step, we consider the last rule applied. There are three cases:
352
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
• γ2 is either an hypothesis or an axiom. In this case: 1 2 3
γ2 γ2 (γ1 γ2 ) γ1 γ2
axiom or hypothesis QTaut QMP: 1,2
• γ2 is obtained from γ and γ γ2 by QMP where γ and γ γ2 are also derived from Γ ∪ {γ1 }. Then, by the induction hypothesis,  Γ γ1 γ;  Γ γ1 (γ γ2 ) Let Π1 and Π2 be the proofs of Γ γ1 γ and Γ γ1 (γ γ2 ) of lengths m1 and m2 , respectively. Let m3 be m1 + m2 . The proof of Γ γ1 γ2 is as follows: m1 . m3 . m3 + 1. m3 + 2. m3 + 3.
γ1 γ γ1 (γ γ2 ) (γ1 (γ γ2 )) ((γ1 γ) (γ1 γ2 )) (γ1 γ) (γ1 γ2 ) γ1 γ2
Π1 Π2 QTaut QMP: m3 ,m3 + 1 QMP: m1 ,m3 + 2
• γ2 is obtained from γ and γ ⇒ γ2 by CMP where γ and γ ⇒ γ2 are also derived from Γ ∪ {γ1 }. Then, by the induction hypothesis,  Γ γ1 γ;  Γ γ1 (γ ⇒ γ2 ). By the axiom Lift⇒ we also have Γ (γ ⇒ γ2 ) (γ γ2 ). By hypothetical syllogism (Lemma 11) we also have Γ γ1 (γ γ2 ). The proof now proceeds as in the previous case. We get as a corollary: COROLLARY 14 Metatheorem of reductio ad absurdum. Let Γ be a set of quantum formulas and γ be a quantum formula. Then, If Γ ∪ {γ} ⊥ ⊥ then Γ γ. We use the metatheorem of equivalence to derive the following theorems: PROPOSITION 15. For every classical formulas α1 and α2 and subsets G1 , G2 ∈ qB, we have [Lift≡] [Sub∩]
(α1 ⇔ α2 ) (α1 ≡ α2 ). F ([G1 ] ([G2 ] [G1 ∩ G2 ])).
Extending Classical Logic for Reasoning about Quantum Systems
353
Proof. We shall use metatheorem of deduction to show each of the theorems: Lift≡. It suﬃces to show that (α1 ⇔ α2 ) (α1 ≡ α2 ) 1
(α1 ⇒ α2 ) ∧ (α2 ⇒ α1 )
Hyp
2
((α1 ⇒ α2 ) ∧ (α2 ⇒ α1 )) ⇒ (α1 ⇒ α2 )
CTaut
3
(α1 ⇒ α2 )
CMP: 1,2
4
(α1 ⇒ α2 ) (α1 α2 )
Lift⇒
5
(α1 α2 )
QMP: 3, 4
6
((α1 ⇒ α2 ) ∧ (α2 ⇒ α1 )) ⇒ (α2 ⇒ α1 )
CTaut
7
(α2 ⇒ α1 )
CMP: 1,2
8
(α2 ⇒ α1 ) (α2 α1 )
Lift⇒
9
(α2 α1 )
QMP: 7, 8
10
(α1 α2 ) ((α2 α1 ) (α1 ≡ α2 ))
QTaut
11
(α2 α1 ) (α1 ≡ α2 )
QMP: 5,10
12
(α1 ≡ α2 )
QMP: 9,11
Sub∩. It suﬃces to show that [G1 ], [G2 ] [G1 ∩ G2 ] 1
[G1 ]
Hyp
2
[G1 ] [qB \ G1 ]
Sub\
3
[qB \ G1 ]
QMP: 1,2
4
[G2 ]
Hyp
5
[G2 ] [qB \ G2 ]
Sub\
6
[qB \ G2 ]
QMP: 4,5
7
[qB \ G1 ] ([qB \ G2 ] [qb \ (G1 ∩ G2 )])
Sub∪
8
[qB \ G2 ] [qb \ (G1 ∩ G2 )]
QMP: 3,7
9
[qb \ (G1 ∩ G2 )]
QMP: 6,8
10
[qb \ (G1 ∩ G2 )] [G1 ∩ G2 ])
Sub\
11
[G1 ∩ G2 ]
QMP: 9,10
We also have the principles of substitution of equal terms and equivalent formulas.
354
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
THEOREM 16 Principle of substitution of equal terms. Given a quantum formula γ, two real terms t1 and t2 , let γ be a quantum formula obtained from γ by replacing zero or more occurrences of t1 in γ1 by t2 . Then, t1 = t2 (γ ≡ γ ). Proof. The proof is by a straightforward induction on the structure of γ. We note that in the case where γ is t ≤ t , we use the axiom RCF. The other cases are immediate. Substitution of equivalent terms preserves quantum equivalence: THEOREM 17 Principle of substitution of equivalent formulas. Given three quantum formulas γ, γ1 and γ2 , let γ be obtained from γ by replacing zero or more qoccurrences of γ1 in γ by γ2 . Then, (γ1 ≡ γ2 ) (γ ≡ γ ). Proof. The case γ1 does not qoccur in γ is trivial. We just consider the case in which γ1 has at least one qoccurrence in γ and γ is obtained by replacement of at least one such qoccurrence. The proof is carried out by induction on the structure of γ. There are two cases: 1. γ is a quantum atom or ⊥ ⊥. Then γ1 is γ, γ1 qoccurs in γ exactly once, and replacement of qoccurrence of γ1 in γ by γ2 yields γ2 . Hence, in that case γ is γ2 . So the theorem holds trivially by the following assertion (justiﬁed by the axiom QTaut): (γ1 ≡ γ2 ) (γ1 ≡ γ2 ). 2. γ is γa γb . Then there are two cases. • γ1 is γ. Then the theorem follows as in the previous case. • γ1 qoccurs in γa or γb (it may occur in both). Let γ be γa γb where γa and γb are obtained by replacing zero or more occurrences of γa and γb respectively. Then, by the induction hypothesis we have (γ1 ≡ γ2 ) (γa ≡ γa ) and We show that as follows.
(γ1 ≡ γ2 ) (γb ≡ γb ). (γ1 ≡ γ2 ), γ γ
Extending Classical Logic for Reasoning about Quantum Systems
1 2 3 4 5 6
γ1 ≡ γ2 γa γb γb γb γa γb γa γa γa γb
355
Hyp Hyp 1, Induction Hypothesis HypSyl: 2, 3 1, Induction Hypothesis HypSyl: 4, 5
We can show similarly that (γ1 ≡ γ2 ), γ γ. The theorem now follows from metatheorem of deduction. We get as a corollary that substitution of classically equivalent formulas preserves quantum equivalence: COROLLARY 18. Given a quantum formula γ, two classical formulas α1 , α2 , let γ be obtained from γ by replacing zero or more qoccurrences of α1 in γ by α2 . Then (α1 ⇔ α2 ) (γ ≡ γ ). Proof. We observe that by Lift⇔, we have (α1 ⇔ α2 ) (α1 ≡ α2 ). The result then follows from principle of substitution of equivalent formulas and hypothetical syllogism. Please note that we are only concerned with occurrence of classical formulas only as quantum subformulas and not as classical formulas. Indeed, replacement of a classical formula by a quantum formula may not always yield valid a quantum formula. Even in the case it yields a valid quantum formula, the principle of substitution does not hold. For example, let α1 be qb1 , α2 be qb2 and γ be qb3 . Now, consider the quantum formula: (qb1 ≡ qb2 ) ((qb1 ⇒ qb3 ) ≡ (qb2 ⇒ qb3 )). Let V be the set of two valuations v1 , v2 such that: • v1 (qb1 ) = v1 (qb3 ) = 0, v1 (qb2 ) = 1; • v2 (qb1 ) = v2 (qb3 ) = 1, v2 (qb2 ) = 0. Any quantum structure with V as the set of valuations would then invalidate the above quantum formula.
356
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
4 COMPLETENESS AND DECIDABILITY We shall prove weak completeness of dEQPL — if Γ is a ﬁnite set of quantum formulas, then Γ γ implies that Γ γ. As our proof system enjoys principle of deduction, it suﬃces to demonstrate weak completeness when the set Γ is empty. The proof of weak completeness will go handinhand with the proof of decidability, and can be adapted to a proof of strong completeness as we will sketch later. The proof of weak completeness essentially follows the proof in [Mateus and Sernadas, 2006], which in turn was inspired by the FaginMegiddoHalpern technique for probabilistic logic [Fagin et al., 1990]. The main diﬀerence is in the way the subsystem formulas are treated here. The other diﬀerence is that the proof is carried out in a manner so as to facilitate the proof of decidability. The central result in the proof is the Model Existence Lemma, namely, if γ is consistent then there is a quantum structure w and an assignment ρ such that wρ γ. A quantum formula γ is said to be consistent if ( γ). It will suﬃce to show that the model existence lemma holds for specials kinds of quantum formula, namely quantum molecular formulas. A quantum molecular formula is a quantum disjunction of quantum literals (a quantum literal is either a quantum atom or the quantum negation of a quantum atom). Please recall that quantum atoms are classical formulas, comparison terms and subsystem assertions. The ﬁrst steps in the proof of the Model Existence Lemma are to remove the probability and alternative terms using the axioms Prob, If% and If⊥. Next, we use the weak completeness of classical propositional logic to construct the set of valuations V in the envisaged quantum structure. The partition S is constructed by considering the subsystem literals in the quantum molecule, and the construction is guided by the fact that subsystems are closed under set operations (axioms Sub∅, Sub∪ and Sub\). The logical amplitudes νGA are constructed by ﬁrst adding all consistent equations using the axioms NAdm, Unit, Empty and Mul, and then “solving” for the (in)equations in the quantum molecule using RCF. Before proceeding with carrying out the above outline, we start with a few abbreviations and notations. We introduce the following abbreviation where Q ⊂ qAtom and D ⊆ Q: • ( Q D) for (( μ∈D μ) $ ( μ∈(Q\D) ( μ))). We shall say that D is the positive part of the quantum molecule ( Q D) and that Q \ D is its negative part. Given a molecule η, we denote by η + and η − the positive and negative parts respectively. We denote by ηc the conjunction of the classical literals in η. In a similar way we deﬁne η≤ and ηs . As is the case with classical propositional logic, every dEQPL formula has a quantum disjunctive normal form. A quantum formula is said to be in quantum disjunctive normal form if it is a disjunction of quantum molecules. PROPOSITION 19. Every quantum formula is equivalent to a quantum disjunctive normal form. Furthermore, there is an algorithm that computes the quantum
Extending Classical Logic for Reasoning about Quantum Systems
357
disjunctive normal form. Proof. It is easier to prove a stronger result. That is, we show that any quantum formula η has both a quantum disjunctive normal form and a quantum conjunctive normal form. We say that η is in quantum conjunctive normal form if it is a quantum conjunction of quantum disjunctions of literals. The proof is constructive and follows by induction on the structure of the quantum formula as in the case of classical logic. The construction also gives the algorithm for computing the normal forms. From now on we will assume that every quantum formula is in quantum disjunctive normal form. The following proposition will ensure that to decide consistency of a quantum formula we only need to check if one of its molecules is consistent. PROPOSITION 20. A quantum formula is consistent iﬀ one of its molecules is consistent. Proof. (⇒) It suﬃces to show that the quantum disjunction of two inconsistent quantum formulas γ1 and γ2 is inconsistent. If γ1 and γ2 are inconsistent then ( γ1 ) and ( γ2 ). We can easily show that in this case (γ1 γ2 ) as follows: 1
( γ1 ) (( γ2 ) (γ1 γ2 ))
QTaut
2
( γ1 )
Hyp
3
( γ2 )
Hyp
4
( γ2 ) (γ1 γ2 )
QMP: 1,2
5
(γ1 γ2 )
QMP: 3,4
Hence the formula, (γ1 γ2 ) is inconsistent. (⇐) Assume that η is inconsistent. Then ( η). Let η be η1 . . . ηn . By QTaut, ( η) ≡ ( η1 $ . . . $ ηn ). Using QMP and QTaut we can easily show that ηi is inconsistent for i = 1, . . . , n. The ﬁrst step in the proof is to remove the probability terms. PROPOSITION 21. Given a quantum molecule η, there is a η such that η has no probability terms and η ≡ η . Furthermore, there is an algorithm that computes η . 9 Proof.Let η be a molecule. For every probability term of the form ( α) replace it by ( A αA 2 )). Then by axiom Prob and the principle of substitution of equal terms, the resulting formula is equivalent to η. The following proposition allows us to remove alternative terms in quantum molecules.
358
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
PROPOSITION 22. A quantum molecule η is consistent iﬀ there is a consistent quantum molecule η such that η has no alternative terms and (η η). Moreover, if there is an algorithm for deciding the consistency of quantum molecules without alternative terms then there is an algorithm for deciding the consistency of quantum molecules. Proof. The existence of a consistent η such that (η η) clearly implies the consistency of η. For the other direction, consider an ordering α0 , . . . , αm of the guards of alternative terms occurring in η. Let αi0 be αi and αi1 be αi for i = 0, . . . , m. Given b0 . . . bm ∈ {0, 1}m , let bm . ηb0 ...bm := η $ α0b0 $ . . . $ αm
Using QTaut we get, η≡
: b0 ...bm
ηb0 ...bm .
∈{0,1}m
Observe that, using the axioms If% and If⊥ and the principle of substitution of equal terms, each ηb0 ...bm is equivalent to a formula in which the alternative (αi u0i ; u1i ) is replaced by ubi i . Let η b0 ...bm be the resulting formula. Therefore, η≡
:
η b0 ...bm .
b0 ...bm ∈{0,1}m
With a reasoning similar to the one in Proposition 20, we conclude that η is consistent iﬀ η b0 ...bm is consistent for some b0 . . . bm ∈ {0, 1}m . Please note that η b0 ...bm η for each b0 . . . bm ∈ {0, 1}m . Finally, as the construction of each η b0 ...bm can be deﬁned by an algorithm, we get the proposition. We shall now build the set of classical valuations V . Given a classical formula α and a nonempty set of valuations V , we write V c α if every element of V classically satisﬁes α. We say that V c η if V c α for every α ∈ ηc+ and V c β for every β ∈ ηc− . We will consider only a special kind of molecular formulas which will allow us to deal with the restrictions imposed by the axiom NAdm. Please recall that given A ⊆ qB, vA is the valuation that assigns true to qubit symbols in A and false to qubit symbols in qb \ A. A molecular formula η is said to be maximal with respect to admissible classical valuations if for every subset A of qB and set of valuations V such that V c η, we have: / V iﬀ (¬(∧A)) ∈ ηc+ . vA ∈ The following proposition ensures that it suﬃces to consider molecular formulas maximally consistent with classical valuations.
Extending Classical Logic for Reasoning about Quantum Systems
359
PROPOSITION 23. A molecule η is consistent iﬀ there is a consistent molecule η such that η is maximal with respect to admissible classical valuations and η η. Moreover, if there is an algorithm for deciding the consistency of quantum molecules maximal with respect to admissible valuations then there is an algorithm for deciding consistency of quantum molecules. Proof. Let A1 , . . . , Am be an ordering of the subsets of qB. Let A0i be (¬(∧Ai )) and A1i be (¬(∧Ai )) for i = 0, . . . , m. Given b0 . . . bm ∈ {0, 1}m , let ηb0 ...bm := η $ Ab00 $ . . . $ Abmm . Using QTaut, η≡
:
ηb0 ...bm .
b0 ...bm ∈{0,1}m
With a reasoning similar to the one in Proposition 20, we can conclude that η is consistent iﬀ ηb0 ...bm is consistent for some b0 . . . bm ∈ {0, 1}m . Please note that ηb0 ...bm η for each b0 . . . bm ∈ {0, 1}m . We claim that each ηb0 ...bm is maximal with respect to admissible valuations. Fix one ηb0 ...bm . Let V be a set of valuations such that V c ηb0 ...bm . We will show that vAi ∈ V iﬀ (¬(∧Ai )) ∈ (ηb0 ...bm )+ c . Clearly if (¬(∧Ai )) ∈ (ηb0 ...bm )+ c then vAi ∈ V . For the other part, if vAi ∈ V it suﬃces to show that bi = 0. Suppose that bi = 1. Then V c (¬(∧Ai )). That means there is v ∈ V such that v c (¬(∧Ai )). This means that v c ∧Ai which in turn implies that v is equal to vAi . Therefore vAi ∈ V contradicting the assumption bi = 1. As the construction of ηb0 ...bm can be deﬁned by an algorithm, we get the proposition. We will say that η is gsatisﬁable if there is a set of valuations V such that V c η. Given a consistent molecule η, we now construct V such that V c η as follows. LEMMA 24 gsatisﬁability. If η is consistent then η is gsatisﬁable. Furthermore, there is an algorithm to decide if η is gsatisﬁable. Proof. Let V be the set of valuations v such that v c α for every α ∈ ηc+ . This set can be computed since the set of qubit symbols is ﬁnite. If V is empty then η is not gsatisﬁable. If V is not empty, then η is gsatisﬁable iﬀ V c β for every β ∈ ηc− . As V and ηc− are ﬁnite sets, this gives us an algorithm to check if η is gsatisﬁable. Assume that η is a consistent formula. Please note that using the theorem Lift∧ and the principle of substitution, it is easy to show that if η is consistent then (∧α∈ηc+ α) is consistent as a classical propositional formula.
360
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
We show that η is gsatisﬁable. As (∧α∈ηc+ α) is consistent (in propositional logic), there is a classical valuation v that satisﬁes every α. As above, let V be the set of valuations v such that v c α for every α ∈ ηc+ . It suﬃces to show that V c β for every β ∈ ηc− . We proceed by contradiction. Assume that there is β ∈ ηc− such that V c β. Fix one such β say β0 . Therefore, by construction of V , we get: ⎛⎛ ⎞ ⎞ c ⎝⎝ α⎠ ⇒ β0 ⎠ . α∈ηc+
So, by CTaut we get:
⎛⎛
⎝⎝
⎞
⎞
α⎠ ⇒ β0 ⎠ .
α∈ηc+
Thus, by Lift⇒, we obtain ⎛⎛
⎝⎝
⎞
⎞
α⎠ β0 ⎠ .
α∈ηc+
Thus, by Ref$ and QTaut (transitivity of ) we get ⎛⎛ ⎞ ⎞ ⎝⎝ α⎠ β0 ⎠ . α∈ηc+
Therefore, by QTaut (right weakening of ) ⎛⎛ ⎞ ⎛ ⎞⎞ : ⎝⎝ α⎠ ⎝ β ⎠⎠ α∈ηc+
leading to
⎛ ⎛⎛ ⎝ ⎝⎝
α∈ηc+
β∈ηc−
⎞
⎛
α⎠ $ ⎝
⎞⎞⎞ ( β)⎠⎠⎠
β∈ηc−
by several obvious tautological steps. That is, we have ( η), contradicting the consistency of η. Please observe that if η has neither probability nor alternative terms then η as constructed in the above proof also does not have probability and alternative terms. Given a subsystem formula [G] and a partition S of the set of qubits, we write S s [G] if G ∈ Alg(S). We say that S s η if S s [G] for every [G] ∈ ηs+
Extending Classical Logic for Reasoning about Quantum Systems
361
and S s [G] for every [G] ∈ ηs− . We will say that η is ssatisﬁable if there is a partition S such that S s η. We construct the partition S in the proof of Model Existence Lemma as follows. LEMMA 25 ssatisﬁability. If η is consistent then η is ssatisﬁable. There is an algorithm to decide if η is ssatisﬁable. Proof. Assume that η is consistent. We will show that η is ssatisﬁable. Please recall that an algebra of sets on a domain X is a nonempty collection of subsets of X closed under complements and unions. Let Alg(ηs+ ) be the smallest algebra on qB containing ηs+ . Find the minimal elements for Alg(ηs+ ): a set G ∈ Alg(ηs+ ) is minimal if G ⊆ G and G ∈ Alg(ηs+ ) implies that G is either the empty set or G itself. Take S to be the set of minimal elements of Alg(ηs+ ) (it can be easily shown that they form a partition). Therefore, by construction, S s [G] for every [G] ∈ ηs+ . / Alg(ηs+ ). We proceed by contradiction If [H] ∈ ηs− then we need to show [H] ∈ + and assume H ∈ Alg(ηs ). Then H = H1 ∪ . . . ∪ Hm , where either Hi ∈ ηs+ or qB \ Hi ∈ ηs+ for each 1 ≤ i ≤ m. Using the axioms QTaut, Sub\ and Sub∪, we can show that η ≡ η $ [H]. Now as [H] ∈ ηs− , we get η ≡ η $ [H] ≡ η $ [H] $ [H]. Now, by QTaut, (η $ [H] $ [H]) ⊥ ⊥. Then, by principle of substitution of equivalent formulas, we get ( η). This contradicts the consistency of η. The algorithm for checking the sconsistency is as follows. Take ηs+ and generate the algebra Alg(ηs+ ) with them. This algebra can be computed since the set of qubit symbols is ﬁnite. The formula η is ssatisﬁable iﬀ G ∈ / Alg(ηs+ ) for every − [G] ∈ ηs . This can be checked by an algorithm again as the set of qubits is ﬁnite. We are now ready to construct the model (the amplitudes νGA will be constructed in the proof). We need some auxiliary deﬁnitions. Recall that assignments are enough to interpret the arithmetical formulas. Let κ be a quantum conjunction of comparison literals. Let K be a real closed ﬁeld with algebraic closure K(δ) and ρ be a Kassignment. We say that K(δ), ρ i κ if • [[s]]ρ ≤ [[t]]ρ if s ≤ t ∈ κ+ ;
362
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
• [[s]]ρ ≤ [[t]]ρ if s ≤ t ∈ κ− . We say that ρ is a solution of κ in K(δ). We say that κ is ≤consistent if there is a real closed ﬁeld K with algebraic closure K(δ), and a Kassignment ρ such that K(δ), ρ i κ. Please note that the theory of elimination of quantiﬁers ensures that there is an algorithm to decide the ≤consistency [Hodges, 1993; Basu et al., 2003]. THEOREM 26 Model Existence Theorem. If the molecule η is consistent then there is a quantum structure w = (K, δ, V, S, ψ, ν) and a Kassignment ρ such that wρ η. Proof. As a result of Propositions 21 and 22, we can assume that η does not have any probability and alternative terms and is maximally consistent with respect to admissible valuations. Using Lemma 24 and Lemma 25, we ﬁnd V and S such that V c η and S s η. We can show that η ≡ (η $ [G]∈Alg(S) [G]) using axioms Sub∅, Sub∪ and Sub\. Please observe that the axiom Unit allows us to establish for every [G] ∈ Alg(S): η( %GA 2 = 1). A⊆G
Let η1 be the formula
η$
(
%GA 2 = 1).
G∈Alg(S) A⊆G
As a result we get that (η1 ≡ η). We also get as a result of the axiom NAdm, for every (¬(∧A)) occurring in η: η1 (%qBA = 0). Let η2 be the formula
η1 $
(%qBA = 0).
(¬(∧A)) in ηc+
As a result of axiom Mul, for every G1 , G2 , A1 , A2 such that G1 , G2 ∈ Alg(S), A1 ⊆ G1 and A2 ⊆ G2 , we get η2 (%G1 ∪G2 A1 ∪A2 = %G1 A1 %G2 A2 ). Let η3 be the formula η2 $
G1 ,G2 ∈Alg(S)
A1 ⊆G1 ,A2 ⊆G2
(%G1 ∪G2 A1 ∪A2 = %G1 A1 %G2 A2 ).
Extending Classical Logic for Reasoning about Quantum Systems
363
The axiom Empty gives us η3 (%∅∅ = 1). Let η • be the formula η3 $ (%∅∅ = 1). Observe that we can show: (η ≡ η • ). Please recall that η • ≤ is the conjunction of the (in)equations in η • . Let ηR be the formula obtained from η • by replacing each term of the form %GA by a fresh variable zGA . Please observe that η • is ηR {zGA / %GA }. Now, either there is a real closed ﬁeld K with K(δ) as its algebraic closure, and a Kassignment ρ such that K(δ), ρ i (ηR )≤ or not. If there is no such K and ρ then it must be the case that (ηR )≤ is a valid arithmetic formula. So, by axiom RCF, ((ηR )≤ ){zGA / %GA }. However, the formula ((ηR )≤ ){zGA / %GA } is ( η≤ ) and this will imply that η is inconsistent. Therefore there are K(δ) and ρ such that K(δ), ρ i (ηR )≤ . We ﬁx such a K, K(δ) and ρ. We now construct ψ = {ψS }S∈S as follows: • ψ[∅] = 1; • Let νSA = ρ(zSA ) for every S ∈ S and A ⊆ S. Then, ψ[S] =
S νSA vA .
A⊆S
We construct ν = {νGA }G⊂qB,A⊆G as follows: νGA =
ρ(zGA ) if zGA is a variable in ηR . 0 otherwise
G Please note that, by construction νGA = vA ψ[G] if G ∈ Alg(S). Let w be (K, δ, V, S, ψ, ν). We can easily show that w is a quantum structure and wρ η.
The decidability of consistency of molecular formulas follows as a corollary to the proof of the Model Existence Lemma. COROLLARY 27. There is an algorithm to decide if a quantum molecule η is consistent.
364
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
Proof. As a result of Propositions 21 and 22, we can assume that η does not have any probability and alternative terms and is maximally consistent with respect to admissible valuations. Now as a result of the model existence lemma, all we need to do is to check if there is a quantum structure w such that w η. We refer to the proof of model existence lemma. We ﬁrst check if η is gsatisﬁable and ssatisﬁable which is algorithmic by Lemmas 24 and 25. If not then η is not consistent. Otherwise, let V and S be as in the proof of the model existence lemma. Now, we construct ηR as in the same proof. Note that the construction is algorithmic. We check if (ηR )≤ is ≤consistent or not. If it is not the case then η is not consistent. If (ηR )≤ is ≤consistent then we can construct w as in that proof such that w η. Therefore η will be consistent if (ηR )≤ is ≤consistent. Please note any formula γ is equivalent to a disjunction of quantum molecular formulas. Furthermore, if γ is consistent, so is one of its molecules, say η. Theorem 26 gives a quantum structure w and an assignment ρ such that wρ = η. As η is a quantum molecule of Γ we get easily wρ = γ. Hence, if any quantum formula γ is consistent then γ has a model. We can now deduce the weak completeness of dEQPL in the standard way. THEOREM 28 Completeness. The proof system of dEQPL is weakly complete, i.e., γ implies γ. Proof. We prove completeness by contradiction. Assume that γ. So by Qtaut and QMP, we have (( γ)). Therefore, γ is consistent, and hence there is a quantum structure w and an assignment ρ such that wρ = γ. Therefore, wρ = γ. Finally, we get the decidability of dEQPL. THEOREM 29 Decidability. The set of theorems is decidable. Proof. As a result of soundness and completeness we have, η iﬀ η is inconsistent. We can decide consistency of a formula by Corollary 27, Proposition 19 and Proposition 20. We ﬁnish this section by observing two things. The ﬁrst observation is that the proof of weak completeness can be adapted to a proof of strong completeness as follows. The key in the proof is again the Model Existence Lemma. Given a possibly inﬁnite consistent set of quantum formulas Γ, we construct a maximally consistent set (the usual HenkinLindenbaum construction). Next, by looking at the classical formulas in Γ, we construct V using the strong completeness of propositional logic. The construction of the partition S is by considering the subsystem literals in Γ and is similar to the one in the above proof. Finally, just as in the proof above, we replace the amplitude terms in comparisonliterals by fresh
Extending Classical Logic for Reasoning about Quantum Systems
365
variables and “solve” the resulting equations using the strong completeness of ﬁrstorder logic (note that as Γ is maximal all the maximally consistent information about logical amplitudes is already in Γ). The second observation is that in our semantic structures, if G is the set of qubits of a subsystem then the qubits in G are necessarily not entangled with the rest. That is, the following is a theorem in dEQPL:
[G]
(%qB(A1 ∪A2 ) = %GA1 %(qB\G)A2 ).
A1 ⊆G , A2 ⊆qB\G
In EQPL [Mateus and Sernadas, 2006], the reverse implication was also true. That is in [Mateus and Sernadas, 2006], it was the case that G is a subsystem if and only if the qubits in G are not entangled with the rest. We can extend our results to such semantic structures by considering the (ﬁnite) set of formulas Γ = {γG  G ⊆ qB} where
γG := ([G] ≡ (
(%qB(A1 ∪A2 ) = %GA1 %(qB\G)A2 ))).
A1 ⊆G , A2 ⊆qB\G
Clearly Γ γ if and only γ holds in all the semantic structures where every set of qubits not entangled with the rest forms a subsystem. If were to augment our axiom system with elements of Γ, then γ is a theorem in the augmented axiomatization if and only if Γ γ. The weak completeness and decidability in the augmented system then follow from the results of this section.
5
APPLICATION EXAMPLES
As it is, dEQPL is appropriate for reasoning about quantum states only. For reasoning about the evolution of quantum systems through the application of measurements and unitary transformations we will need to extend it towards a dynamic logic, as already sketched in [Mateus and Sernadas, 2004a; Mateus and Sernadas, 2004b]. Herein, we ﬁrst illustrate how dEQPL can be used to reason about a Bell state. Afterwards, we turn our attention to quantum teleportation and outline there some of the relevant constructs of the envisaged dynamic logic. In the following examples, we write F as an abbreviation for the vector (%F A )A⊆F assuming the lexicographic ordering of the subsets of F . We may also abbreviate {qbk1 , . . . , qbkm } by qbk1 ,...,km in amplitude terms.
5.1
Reasoning about Bell states
Bell states were ﬁrst discussed by Einstein, Podolsky and Rosen [Einstein et al., 1935] and have been very useful in designing quantum protocols. An independent
366
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
subsystem composed of a pair of qubits is said to be in a Bell state if they are maximally entangled. For instance, 1 ψ = √ (10 − 01) 2 is a Bell state. In order to represent this pair in our logic, we choose two qubit symbols, say qb0 and qb1 . The fact that these qubits are independent from other qubits can be written as γind := [qb0 , qb1 ]. We can express the state as the following formula 1 γEPR := (qb01 = √ (0, −1, 1, 0)). 2 We can use our logic to derive that these qubits are necessarily entangled, that is, neither qb0 nor qb1 form an independent subsystem. In other words we will show that γind , γEPR [qb0 ] $ [qb1 ]. The proof will follow by applying the metatheorem theorem of deduction. In particular, we show ⊥, γind , γEPR , [qb0 ] ⊥ as follows: 1 [qb0 , qb1 ]
Hyp
2 [qb0 ]
Hyp
3 ([qb0 , qb1 ] ([qb0 ] [qb1 ]))
SubDiﬀ
4 ([qb0 ] [qb1 ]))
QMP: 1,3
5 [qb1 ]
QMP :2,4
6 (qb
01 ∅
= − √1 ) (qb
= 0) ((qb
01 qb0
01 qb1
2
=
1 √ ) 2
(qb
7 (γ1 γ2 ) γ2 8 (qb
01 qb0
= − √1 ) (qb
10 qb
= qb
11 qb
= qb
01 qb1
01 qb1
2
= qb
01 qb1
= 0)) Hyp QTaut
9 qb
01 qb0
01 qb01
0 qb0 0∅
qb
1∅
qb
0 qb0
1 qb1
qb
1 qb1
=
1 √ ) 2
(qb
01 qb01
= 0)
QMP: 6,7 Mul: 2,5 Mul: 2,5 Mul: 2,5
12 ⊥
RCF: 8–11
13 ⊥ ⊥
Eqv⊥: 12
Extending Classical Logic for Reasoning about Quantum Systems
367
Therefore, by metatheorem of deduction, we get γind , γEPR [qb0 ]. In a similar way, we can derive γind , γEPR [qb1 ], and consequently, we get γind , γEPR [qb0 ] $ [qb1 ]. In the next section, we consider a protocol which uses this Bell state to achieve teleportation.
5.2
Reasoning about quantum teleportation
A protocol for quantum teleportation was ﬁrst proposed in [Bennett et al., 1993]. The idea is to move a qubit from one agent to another who share an entangled pair of qubits while exchanging only classical information. Before describing and verifying the protocol we need to extend dEQPL with some features from dynamic logic. Namely, we shall use formulas, called Hoare triples for historical reasons [Hoare, 1969], of the form {γ1 } P {γ2 } where γ1 and γ2 are dEQPL formulas and P is a quantum program denoting some composition of unitary transformations and measurements. It is often useful to reserve some qubits that are always in a classical state. Let us call them classical bits and use the symbols cb1 , . . . , cbm to range over them. We shall avoid going into the details of the quantum program language and semantics, better left to a speciﬁc paper on a dynamic extension of dEQPL. However, we shall provide the needed intuitions. Namely, the Hoare triple above means that if the system is in a quantum state satisfying γ1 then after running P it reaches a state satisfying γ2 . The protocol in [Bennett et al., 1993] uses three qubits, say qb0 , qb1 and qb2 plus two classical bits cb0 and cb2 . The purpose is to transfer the quantum state of qb0 to qb1 , using qb2 and the classical bits as auxiliary variables. Initially, qb1 and qb2 will be prepared in a Bell state not entangled with qb0 . Afterwards, a measurement of qb0 and qb2 is made (by Alice). Note that this measurement will also aﬀect qubit 1 because it is entangled with qubit 2. The classical bits are used to store the result of measuring the corresponding qubits. The classical information to be exchanged is precisely the contents of the classical bits after the measurement. Finally, this information is used (by Bob) to decide which unitary transformation to apply on qb1 in order to achieve the required state. In short,
368
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
the protocol QTP is as follows: Mqb02 ; IF cb02 = (1, 0, 0, 0) cb02 = (0, 1, 0, 0) cb02 = (0, 0, 1, 0) cb02 = (0, 0, 0, 1)
→ → → →
−Iqb1 −Zqb1 Xqb1 −Xqb1 Zqb1
FI where I is the identity operator and X and Z are the standard Pauli operators (not and phase ﬂip, respectively). The initial state of the system (after preparing the qubits 1 and 2) is assumed to comply with: 1 γinit := [qb0 ] $ (qb12 = √ (0, 1, −1, 0)) $ (qb0 = (z0 , z1 )) . 2 Observe that we are not constraining the state of qubit 0. We just need to refer to it which we achieve by using the (rigid) variables z0 and z1 . Note also that in such a state the qubits 1 and 2 are entangled. Actually, they are in a Bell state as discussed in the previous example. We want the ﬁnal state of the system (after running the protocol) to comply with: γﬁn := [qb1 ] $ (qb1 = (z0 , z1 )) . In other words, we want to establish: Spec := {γinit } QTP {γﬁn } . To this end, it is enough to assume that the measurement operator Mqb02 complies with the following non probabilistic speciﬁcation: {γinit } Mqb02 {4k=1 γk } where γ1 γ2 γ3 γ4
:= (cb02 = (1, 0, 0, 0)) := (cb02 = (0, 1, 0, 0)) := (cb02 = (0, 0, 1, 0)) := (cb02 = (0, 0, 0, 1))
$ $ $ $
(qb02 = (qb02 = (qb02 = (qb02 =
√1 (1, 0, 0, 1)) $ (qb = −(z0 , z1 )); 1 2 √1 (−1, 0, 0, 1)) $ (qb = (−z0 , z1 )); 1 2 √1 (0, 1, 1, 0)) $ (qb = (z1 , z0 )); 1 2 √1 (0, −1, 1, 0)) $ (qb = (z1 , −z0 )). 1 2
Observe that the IF part of the protocol QTP complies with: {γk } IF {γﬁn } for k = 1, . . . , 4. Therefore, we can derive Spec using the traditional composition rules of dynamic logic.
Extending Classical Logic for Reasoning about Quantum Systems
369
6 CONCLUDING REMARKS A decidable quantum logic allowing us to reason about amplitudes of quantum states and probabilities of classical outcomes was obtained as a fragment of EQPL. Decidability was achieved by relaxing the semantics, replacing Hilbert spaces by inner product spaces over arbitrary real closed ﬁelds and their algebraic closures. The proof of decidability was carried out hand in hand with the proof of weak completeness and follows the FaginHalpernMegiddo technique (originally proposed for probabilistic logics [Fagin et al., 1990; Abadi and Halpern, 1994]). We envision to use this decidable quantum logic in the speciﬁcation and veriﬁcation of quantum procedures and protocols, either via model checking or theorem proving. To this end, the hardness of the proposed decision algorithm needs to be analyzed. We also intend to enrich this decidable quantum logic with Hoare triples as outlined in Section 5 and in [Mateus and Sernadas, 2004a; Mateus and Sernadas, 2004b]. Temporal extensions of dEQPL should also be explored to reason about liveness and progress properties of quantum computations. Another interesting line of research would be to develop a ﬁrstorder quantum logic based on the exogenous semantics approach. Both EQPL and dEQPL allow us to express amplitudes of pure quantum states of collections of qubits, so these logics are not insensitive to the global phase of the quantum state. One may argue that it should be insensitive since no physical measurement will ever be able to distinguish two quantum states that are equivalent up to global phase. We decided to leave dEQPL as it is (that is, sensitive to global phase) for two reasons. In practice, physicists and quantum computer scientists need to work with both levels of abstraction. Sometimes they want to work with states as unit vectors and other times they want to abstract away the global phase. So, a calculus supporting the former level of abstraction is also useful. The second reason is a consequence of the fact that forgetting global phase requires a major semantic shift. Indeed, it is better solved by identifying a quantum state with a density operator working on the underlying inner product space, that is, working with probabilistic ensembles or mixed quantum states in general. Such a shift toward a semantics based on density operators will lead to a quite diﬀerent quantum logic (but still extending classical logic by applying the exogenous approach) that will also be useful for reasoning about quantum systems evolving under partial tracing, besides unitary transformations and measurements. Clearly, this is yet another line of research that will deserve attention. The relationship between the exogenous quantum logics and the more traditional quantum logics (based on the original Birkhoﬀ and von Neumann proposal) should be further explored. At the preliminary stage of work in this direction, it seems that most of the qualitative assertions possible in the latter can be made in the former and that the latter can be easily extended with quantitative aspects of the former. In other words, it seems feasible to combine the two quantum logics into a single logic by using ﬁbring techniques [Gabbay, 1996; Caleiro et al., 2005].
370
R. Chadha, P. Mateus, A. Sernadas and C. Sernadas
ACKNOWLEDGMENTS The authors wish to express their gratitude to the regular participants in the QCI Seminar at SQIGIT (formerly at CLC), and also to Dave Marker and Anand Pillay for their help on real closed ﬁelds and their algebraic completion. This work was partially supported by FCT and FEDER through POCTI, namely via the QuantLog POCTI/MAT/55796/2004 (Quantum Logic), KLog PTDC/MAT/68723/2006 (Kleistic Logic) and QSec PTDC/EIA/67661/2006 (Quantum Security) projects. BIBLIOGRAPHY [Abadi and Halpern, 1994] M. Abadi and J. Y. Halpern. Decidability and expressiveness for ﬁrstorder logics of probability. Information and Computation, 112(1):1–36, 1994. [Abramsky and Coecke, 2004] S. Abramsky and B. Coecke. A categorical semantics of quantum protocols. In Proceedings of the 19th Annual IEEE Symposium on Logic in Computer Science (LICS 2004), pages 415–425. IEEE Computer Science Press, 2004. Extended version at arXiv:quantph/0402130). [Abramsky and Duncan, 2006] S. Abramsky and R. Duncan. A categorical quantum logic. Mathematical Structures in Computer Science, 16(3):469–489, 2006. [Bacchus, 1990a] F. Bacchus. On probability distributions over possible worlds. In Uncertainty in Artiﬁcial Intelligence, 4, volume 9 of Machine Intelligence and Pattern Recognition, pages 217–226. NorthHolland, 1990. [Bacchus, 1990b] F. Bacchus. Representing and Reasoning with Probabilistic Knowledge. MIT Press Series in Artiﬁcial Intelligence. MIT Press, 1990. [Basu et al., 2003] S. Basu, R. Pollack, and R. MarieFran¸coise. Algorithms in Real Algebraic Geometry. Springer, 2003. [Bennett et al., 1993] C. H. Bennett, G. Brassard, C. Cr´epeau, R. Jozsa, A. Peres, and W. K. Wootters. Teleporting an unkown quantum state via dual classical and einsteinpodolskyrosen channels. Physical Review Letters, 70(13):1895–1899, 1993. [Birkhoﬀ and von Neumann, 1936] G. Birkhoﬀ and J. von Neumann. The logic of quantum mechanics. Annals of Mathematics, 37(4):823–843, 1936. [Caleiro et al., 2005] C. Caleiro, A. Sernadas, and C. Sernadas. Fibring logics: Past, present and future. In S. Artemov, H. Barringer, A. S. d’Avila Garcez, L. C. Lamb, and J. Woods, editors, We Will Show Them: Essays in Honour of Dov Gabbay, Volume One, pages 363–388. College Publications, 2005. [Caleiro et al., 2006] C. Caleiro, P. Mateus, A. Sernadas, and C. Sernadas. Quantum institutions. In K. Futatsugi, J.P. Jouannaud, and J. Meseguer, editors, Algebra, Meaning, and Computation — Essays Dedicated to Joseph A. Goguen on the Occasion of His 65th Birthday, volume 4060 of Lecture Notes in Computer Science, pages 50–64. SpringerVerlag, 2006. [Carnielli and LimaMarques, 1999] W. A. Carnielli and M. LimaMarques. Society semantics and multiplevalued logics. In Advances in Contemporary Logic and Computer Science (Salvador, 1996), volume 235 of Contemporary Mathematics, pages 33–52. AMS, 1999. [Carnielli, 2000] W. A. Carnielli. Possibletranslations semantics for paraconsistent logics. In Frontiers of Paraconsistent Logic (Ghent, 1997), volume 8 of Studies in Logic and Computation, pages 149–163. Research Studies Press, 2000. [Chadha et al., 2007] R. Chadha, L. CruzFilipe, P. Mateus, and A. Sernadas. Reasoning about probabilistic sequential programs. Theoretical Computer Science, 379(12):142–165, 2007. [Chiara et al., 2004] M. L. D. Chiara, R. Giuntini, and R. Greechie. Reasoning in Quantum Theory. Kluwer Academic Publishers, 2004. [CohenTannoudji et al., 1977] C. CohenTannoudji, B. Diu, and F. Lalo¨e. Quantum Mechanics. John Wiley, 1977. [Dishkant, 1972] H. Dishkant. Semantics of the minimal logic of quantum mechanics. Studia Logica, 30:23–32, 1972. [Einstein et al., 1935] A. Einstein, B. Podolsky, and N. Rosen. Can quantummechanical description of physical reality be considered complete? Physical Review, 47:777–780, 1935.
Extending Classical Logic for Reasoning about Quantum Systems
371
[Fagin et al., 1990] R. Fagin, J. Y. Halpern, and N. Megiddo. A logic for reasoning about probabilities. Information and Computation, 87(12):78–128, 1990. [Foulis, 1999] D. J. Foulis. A halfcentury of quantum logic. What have we learned? In Quantum Structures and the Nature of Reality, volume 7 of Einstein Meets Magritte, pages 1–36. Kluwer Acad. Publ., 1999. [Gabbay, 1996] D. M. Gabbay. Fibred semantics and the weaving of logics: Part 1. Journal of Symbolic Logic, 61(4):1057–1120, 1996. [Halpern, 1990] J. Y. Halpern. An analysis of ﬁrstorder logics of probability. Artiﬁcial Intelligence, 46:311–350, 1990. [Hansson and Jonsson, 1995] H. Hansson and B. Jonsson. A logic for reasoning about time and reliability. Formal Aspects of Computing, 6:512–535, 1995. [Hoare, 1969] C. Hoare. An axiomatic basis for computer programming. Communications of the ACM, 12:576–583, 1969. [Hodges, 1993] W. Hodges. Model Theory. Cambridge University Press, 1993. [Kripke, 1963] S. A. Kripke. Semantical analysis of modal logic. I. Normal modal propositional calculi. Zeitschrift f¨ ur Mathematische Logik und Grundlagen der Mathematik, 9:67–96, 1963. [Mateus and Sernadas, 2004a] P. Mateus and A. Sernadas. Exogenous quantum logic. In W. A. Carnielli, F. M. Dion´ısio, and P. Mateus, editors, Proceedings of CombLog’04, Workshop on Combination of Logics: Theory and Applications, pages 141–149, 1049001 Lisboa, Portugal, 2004. Departamento de Matem´ atica, Instituto Superior T´ecnico. Extended abstract. [Mateus and Sernadas, 2004b] P. Mateus and A. Sernadas. Reasoning about quantum systems. In J. Alferes and J. Leite, editors, Logics in Artiﬁcial Intelligence, Ninth European Conference, JELIA’04, volume 3229 of Lecture Notes in Artiﬁcial Intelligence, pages 239–251. SpringerVerlag, 2004. [Mateus and Sernadas, 2006] P. Mateus and A. Sernadas. Weakly complete axiomatization of exogenous quantum propositional logic. Information and Computation, 204(5):771–794, 2006. [Mateus et al., 2005] P. Mateus, A. Sernadas, and C. Sernadas. Exogenous semantics approach to enriching logics. In G. Sica, editor, Essays on the Foundations of Mathematics and Logic, volume 1 of Advanced Studies in Mathematics and Logic, pages 165–194. Polimetrica, 2005. [Naur, 1963] P. Naur. Revised report on the algorithmic language Algol 60. The Computer Journal, 5:349–367, 1963. [Nielsen and Chuang, 2000] M. A. Nielsen and I. L. Chuang. Quantum Computation and Quantum Information. Cambridge University Press, 2000. [Nilsson, 1986] N. J. Nilsson. Probabilistic logic. Artiﬁcial Intelligence, 28(1):71–87, 1986. [Nilsson, 1993] N. J. Nilsson. Probabilistic logic revisited. Artiﬁcial Intelligence, 59(12):39–42, 1993. [Patra, 2008] M. Patra. A logic for quantum computation and classical simulation of quantum algorithms. International Journal of Quantum Information, 6(2), 2008. In print. [Tarski, 1948] A. Tarski. A decision method for elementary algebra and geometry. Manuscript. Santa Monica, CA: RAND Corp., 1948. Republished as A Decision Method for Elementary Algebra and Geometry, 2nd ed. Berkeley, CA: University of California Press, 1951. [van der Meyden and Patra, 2003a] R. van der Meyden and M. Patra. Knowledge in quantum systems. In M. Tennenholtz, editor, Theoretical Aspects of Rationality and Knowledge, pages 104–117. ACM, 2003. [van der Meyden and Patra, 2003b] R. van der Meyden and M. Patra. A logic for probability in quantum systems. In M. Baaz and J. A. Makowsky, editors, Computer Science Logic, volume 2803 of Lecture Notes in Computer Science, pages 427–440. SpringerVerlag, 2003.
This page intentionally left blank
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
373
` SOLER’S THEOREM Alexander Prestel Sol`er’s Theorem gives an axiomatic characterization (in algebraic terms) of inﬁnite dimensional Hilbert spaces over the reals, the complex, and the quaternions. At the same time it gives a characterization of the lattices that are isomorphic to the lattice of closed subspaces of the just mentioned Hilbert spaces. These lattices play an important role in quantum logic (see [Holland, 1995]) and, more generally, in Hilbert space logic (see [Engesser and Gabbay, 2002]). More about the history of Sol`er’s Theorem and its consequences in areas like Baer *rings, inﬁnite dimensional projective geometry, orthomodular lattices, and the logic of quantum mechanics can be found in S. Holland’s exposition [Holland, 1995]. The aim of this paper is to provide a complete proof of Sol`er’s Theorem for the reader of the ‘Handbook of Quantum Logic and Quantum Structures’.1 Sol`er’s Theorem deals with inﬁnite dimensional hermitian spaces (E, ) over an arbitrary skew ﬁeld K which are orthomodular, i.e., every subspace X of E satisﬁes (1)
X = (X ⊥ )⊥ ⇒ E = X ⊕ X ⊥ .
The theorem can be stated as follows (for deﬁnitions see below): Sol` er’s Theorem Let (K,∗ ) be a skew ﬁeld together with an (anti)involution , and let (E, ) be an inﬁnite dimensional hermitian vector space over (K,∗ ). If (E, ) is orthomodular and contains an inﬁnite orthonormal sequence (en )n∈N , then
∗
(i) K = R, C, or H where in the ﬁrst case ∗ is the identity and in the case of C and the quaternions H,∗ is the canonical conjugation, (ii) (E, ) is a Hilbert space over R, C or H, resp. The main part of this theorem actually is (i). Once we know that K is the ﬁeld of real or complex numbers, or the skew ﬁeld H of the quaternions, it is not diﬃcult to prove (ii) (cf. the end of Section 2). While the (skew) ﬁelds R, C and H carry a canonical metric that makes them complete, there is no mention of any metric on 1 The
proof presented here is identical with that of [Prestel, 1995].
374
Alexander Prestel
K in the assumption of the theorem. Not even any topology is mentioned. Nevertheless, the seemingly ‘algebraic’ conditions of orthomodularity and the existence of an inﬁnite orthonormal sequence will lead to the surprising fact that K is R, C or H. It has been a long standing open problem whether orthomodularity of (E, ) would already force K to be one of R, C or H. In 1980, ﬁnally, H. Keller (see [Keller, 1980]) constructed “nonclassical Hilbert spaces”, i.e., orthomodular hermitian spaces, not isomorphic to one of the classical Hilbert spaces. Only in 1995, M.P. Sol`er showed in her Ph.D. thesis (cf [Sol`er, 1995]) that adding the existence of an inﬁnite orthonormal sequences expelles all the nonclassical Hilbert spaces. In sections 2 to 4 below we shall ﬁrst treat the commutative case, i.e., we let K be a commutative ﬁeld. Under this additional assumption we give a complete proof of Sol`er’s Theorem. In Section 4 we deal with the general case, i.e., we let K be an arbitrary skew ﬁeld. The proof in this case needs a few reﬁnements of the earlier ones which we shall explain. The structure of the proof in the noncommutative case, however, is essentially the same as that in the commutative case.
1
PRELIMINARIES AND STRUCTURE OF THE PROOF
Let K be a (commutative) ﬁeld and
∗
(α + β)∗ = α∗ + β ∗ ,
: K → K an involution on K, i.e. (αβ)∗ = α∗ β ∗ ,
∗
satisﬁes
α∗∗ = α
for all α, β ∈ K. The identity on R and the complex conjugation − on C are our standard examples for such an involution. Furthermore, let E be an inﬁnite dimensional Kvector space and : E ×E → K a hermitian form on E, i.e. satisﬁes < αx + βy, z > = α < x, z > + β < y, z > < z, αx + βy > = α∗ < z, x > + β ∗ < z, y > < x, z > = < z, x >∗ for all α, β ∈ K and x, y, z ∈ E, The pair (E, ) is called a hermitian space if is a hermitian form on E. The hermitian form is called anisotropic if for x ∈ E, < x, x > = 0 implies x = 0. As usual, we deﬁne orthogonality x ⊥ y by < x, y >= 0 for vectors x, y ∈ E. The orthogonal space U ⊥ to a subset U of E is deﬁned by U ⊥ = {x ∈ E x ⊥ u for all u ∈ U }. We simply write U ⊥⊥ for (U ⊥ )⊥ . The space U ⊥⊥ is called the closure of U , and U is called closed if U ⊥⊥ = U . In case (E, ) is anisotropic, every ﬁnite dimensional
Sol` er’s Theorem
375
subspace of E is closed. Since U ⊥⊥⊥ = U ⊥ , U ⊥ is also closed. Thus, in particular, orthomodularity of (E, ) may be equivalently expressed by (1)
E = U ⊥ ⊕ U ⊥⊥ for all subspaces U of E.
Two important consequences of orthomodularity of a hermitian space are (2)
if U and V are orthogonal subspaces of E then (U + V )⊥⊥ = U ⊥⊥ + V ⊥⊥ and U ⊥⊥ ⊥ V ⊥⊥ ,
(3)
if U is a closed subspace of E, then U together with the restriction of to U × U is an orthomodular space, too.
Applying (1) to U = {x} we see that orthomodular spaces are anisotropic. A ﬁnite dimensional hermitian space is orthomodular if and only if it is anisotropic. It is also wellknown that any RHilbert space and any CHilbert space is orthomodular (w.r.t. its deﬁning inner product). The content of the above theorem is that there are no other examples of hermitian spaces (E, ) which are orthomodular and contain an inﬁnite sequence (en )n∈N which is orthonormal, i.e. for all n, m ∈ N we have: (4)
< en , en >= 1 and en ⊥ em for n = m.
As H. Keller has shown in [Keller, 1980], inﬁnite dimensional orthomodular spaces exist which do not contain any orthonormal sequence, hence cannot be real or complex Hilbert spaces. Now let us explain the three steps into which we will divide the proof of the theorem. In Step 1 we will show that is ‘positive deﬁnite’, i.e. we will show that the ﬁxed ﬁeld F = {α ∈ K α∗ = α} of ‘symmetric’ elements admits an ordering ≤ such that < x, x >≥ 0 for all x ∈ E. (Note that < x, x > is symmetric.) In fact, the ordering on F will be given by (5)
α ≤ β iﬀ β − α ∈ P,
where P = {< x, x >  x ∈ E} is the set of ‘lengths’ of vectors x ∈ E. In Step 2 we will show ﬁrst that ≤ is archimedean on F , i.e. (6)
to every α ∈ P there exists n ∈ N such that α ≤ n.
At this point we make use of the fact that, in order to prove that (5) deﬁnes an ordering, it suﬃces to know here that ≤ is a semiordering (as introduced in
376
Alexander Prestel
[Prestel, 1984] by the author), i.e. ≤ linearly orders F such that in addition we have for α, β ∈ F : (7)
0 ≤ α, 0 ≤ β =⇒ 0 ≤ α + β 0 ≤ α =⇒ 0 ≤ αβ 2 0≤1
In fact, an archimedean semiordering is already an ordering, i.e., it satisﬁes in addition 0 ≤ α, 0 ≤ β =⇒ 0 ≤ α · β for all α, β ∈ F ([Prestel, 1984], Theorem 1.20). Thus, in Step 1 it therefore suﬃces to show that (5) deﬁnes a semiordering on F . The linearity of ≤ is obtained from Sol`er’s main Lemma 5, which also gives the archimedeanity (Lemma 6). Actually, the linearity of ≤, i.e. P ∪ −P = F , is not important in the commutative case, since it may be simply obtained by maximalizing a subset P of F satisfying P + P ⊂ P, P F 2 ⊂ P, P ∩ −P = {0} and 1 ∈ P by Zorn’s Lemma to some P0 , w.r.t. to these properties. The maximal object P0 then satisﬁes P0 ∪ −P0 = F (see [Prestel, 1984], Lemma 1.13). In the noncommutative case, however, this method may not work (see the remarks at the end of Section 6). As it is wellknown, an archimedean ordered ﬁeld (F, ≤) contains (an isomorphic) copy of the rational number ﬁeld Q as a dense subﬁeld. Thus, it suﬃces to prove that every Dedekind cut is realized in (F, ≤) in order to ﬁnd that F is isomorphic to R. This is done in Lemma 7. Since F is the ﬁxed ﬁeld of the involution ∗ , we have [K : F ] ≤ 2. Hence K = R and ∗ =id or K = C and ∗ is the complex conjugation. Now we know that K is R or C and is positive deﬁnite, i.e., (E, ) is a preHilbert space. Thus, in Step 3 it remains to prove that every orthomodular preHilbert space is complete. This, however, can already be found in the literature (see [Maeda and Maeda, 1970], Theorem 34.9). The argument runs as follows. ˆ be the completion of E w.r.t. the metric induced by . Given a ∈ E ˆ Let E we have to show that a already belongs to E. Choosing c ∈ E suitably such that ˆ such that a ⊥ b and c = a + b. By standard arguments < a, c >= 0, we ﬁnd b ∈ E we then ﬁnd sequences (an )n∈N and (bn )n∈N in E converging to a and b resp. such that an ⊥ bm for all n, m ∈ N. If we then take A = {bn  n ∈ N}⊥ we see that A⊥⊥ = A in E and thus the modularity (1) of E gives us a decomposition c = c1 + c2 with c1 ∈ A and c2 ∈ A⊥ . ˆ we trivially have c1 ∈ A and Denoting by A the topological closure of A in E by continuity of also a ∈ A. Again by continuity arguments we see that b and ˆ Since E ˆ is a Hilbert space, the closed subspace A yields c2 belong to (A)⊥ in E. a decomposition ˆ = A ⊕ (A)⊥ . E
Sol` er’s Theorem
377
Thus, it follows from a + b = c = c1 + c2 that a = c1 ∈ E.
2 CONSTRUCTION OF THE SEMIORDERING ON F Let (E, ) be an orthomodular hermitian space and (fi )i∈I be some orthogonal sequence in E, i.e., fi ⊥ fj for i = j. Then the map x −→ (< x, fi >)i∈I ⊥ ⊥
is Klinear and injective on the subspace U = (fi )i∈I of E. In fact, if ⊥ < x, fi >= 0 for all i ∈ I, then x ∈ (fi )i∈I . Thus, for x ∈ U we get x ⊥ x which implies x = 0, since is anisotropic. Hence, every vector x from U is uniquely determined by the sequence (< x, fi >)i∈I . We therefore write (1)
x=
i∈I
αi fi with αi :=
< x, fi > < fi , fi >
and call αi the ‘Fouriercoeﬃcient’ of x w.r.t. (fi )i∈I . It should be clear, however, that in (1) is not an inﬁnite sum converging w.r.t. some metric. It is only a formal notation, expressing nothing else than: x is the unique vector in U satisfying < x, fi >= αi < fi , fi > for all i ∈ I. In case I is a ﬁnite set, in (1) is actually a ﬁnite sum and the identity in (1) is familiar. As an ordered ﬁeld, K clearly will have characteristic zero. This, however, is already clear from the existence of an inﬁnite orthonormal sequence (en )n∈N and the anisotropy of . In fact, for any prime p 0 =< e1 + ... + ep , e1 + ... + ep >= p. The next proposition contains one main tool for constructing the desired ordering on F . PROPOSITION 1. Let (fn )n∈N and (gn )n∈N be two orthonormal sequences such ⊥ ⊥ that fn ⊥ gm for all n, m ∈ N. Then to every vector x = n αn fn ∈ (fn )n∈N a ⊥ ⊥ vector y = n αn gn ∈ (gn )n∈N exists such that < x, x >=< y, y >. Proof. Since char K = 2, the systems (fn )n∈N ∪ (gn )n∈N and (fn + gn )n∈N ∪ (fn − ⊥ ⊥ gn )n∈N generate the same linear subspace H of E. From (fn )n∈N ⊂ H ⊥⊥ and (2) we get x = x1 + x2 with ⊥ ⊥ αn (fn + gn ) ∈ (fn + gn )n∈N x1 = n
x2
=
n
αn (fn − gn ) ∈ (fn − gn )n∈N
⊥ ⊥
378
Alexander Prestel
(Note that (fn + gn ) ⊥ (fm − gm ) for all n, m ∈ N.) For αn we ﬁnd αn =
< x, fn + gn > < x2 , fn + gn > 1 < x1 , fn + gn > = − = αn < fn + gn , fn + gn > 2 2 2
Similarly, we get αn = 12 αn . ⊥ ⊥ From (fn + gn )n∈N ⊂ H ⊥⊥ and (2) we next get 2x1 = n αn (fn +gn ) = y1 +y2 with ⊥ ⊥ y1 = βn fn ∈ (fn )n∈N n
y2
=
⊥ ⊥
βn gn ∈ (gn )n∈N .
n
For βn we ﬁnd βn =< y1 , fn >=< 2x1 , fn >=< x1 , (fn + gn ) + (fn − gn ) >=< x1 , fn + gn >= αn and similarly βn =< y2 , gn >=< 2x1 , gn >=< x1 , (fn + gn ) − (fn − gn ) >= αn . Thus, we ﬁrst see that y1 = x , and, taking ⊥ ⊥ αn gn ∈ (gn )n∈N , y := y2 = n
we then ﬁnd < y, y >=< x, x > from y = 2x1 − x and < x, x1 >=< x1 , x1 >= < x1 , x >. In order to apply this proposition we will make the ASSUMPTION 2. E = (en )n∈N
⊥ ⊥
is orthomodular.
By (3) this is without restriction since by Step 3 of Section 2 it suﬃces to prove that F ∼ = R. LEMMA 3. Let (E, ) satisfy Assumption 2 and let (fn )n∈N , (gn ) n∈N be two othonormal sequences in E. Then the map assigning to every x = n αn fn ∈ ⊥ ⊥ ⊥ ⊥ (fn )n∈N the vector y = n αn gn ∈ (gn )n∈N deﬁnes an isometry of the sub⊥ ⊥ ⊥ ⊥ spaces (fn )n∈N and (gn )n∈N of E. Proof. The only diﬀerence of this lemma to Proposition 1 is that the sequences (fn )n∈N and (gn )n∈N may not be orthogonal to each other. ⊥ ⊥ Let Ei = (e4n+i )n∈N with i ∈ {0, 1, 2, 3}. Then
Sol` er’s Theorem
(2)
379
E = E ⊕ E with E = E1 ⊕ E2 and E = E3 ⊕ E4 .
From Proposition 1 we immediately get the isometries E ∼ = E , E ∼ = E1 , E ∼ = E2 . Together we therefore also obtain E = E ⊕ E ∼ = E . = E1 ⊕ E2 = E and E ∼ Now we are able to deduce the lemma from Proposition 1. In fact, we ﬁrst move the sequence (fn )n∈N from E to E and the sequence (gn )n∈N from E to E . Then we apply Proposition 1. From this lemma we get two easy consequences which will be used in the next sections. COROLLARY 4. Let (E, ) satisfy Assumption 2. Then: (a) E contains a subspace U isomorphic to the inﬁnite orthogonal sum E ⊕ E ⊕ E ⊕ ... (b) There is no vector x ∈ E having Fouriercoeﬃcient < x, en >= 1 for all n ∈ N. Proof. (a) Let N = m∈N Im be a disjoint partition of N into inﬁnite sets In . Then by Lemma 3, every closure (Im )⊥⊥ is isometric to E. Now let U = I1⊥⊥ + I2⊥⊥ + I3⊥⊥ + ... (b) Assume x ∈ E has < x, en >= 1 for all n ∈ N. Observing that E = ⊥ ⊥ ⊥ ⊥ ⊥ ⊥ (e2n )n∈N + (e2n+1 )n∈N we get x = x1 + x2 with x1 ∈ (e2n )n∈N and x2 ∈ ⊥ ⊥ (e2n+1 )n∈N . By Lemma 3 we get < x, x >=< x1 , x1 >=< x2 , x2 >. Now x1 ⊥ x2 yields the contradiction < x, x >=< x1 , x1 > + < x2 , x2 >= 2 < x, x > . We are now able to prove that the set of lengths P = {< x, x >  x ∈ E} satisﬁes the following properties: (3)
(i)
P +P ⊂P
(ii) (iii)
PF2 ⊂ P P ∩ −P = {0}
(iv)
1∈P
380
Alexander Prestel
Obviously, (ii) and (iv) are trivial. To see (i), let α =< x, x > and β = < y, y >. By (2) choose x ∈ E and y ∈ E such that < x , x >=< x, x > and < y , y >=< y, y >. Since E and E are orthogonal, we get < x + y , x + y >= α + β. For proving (iii), let α =< x, x > and −α =< y, y >. Now we ﬁnd < x + y , x + y >= α − α = 0. Since is anisotropic, we have x + y = 0. Thus, x = y = 0 and α = 0. From (3) we see that α ≤ β ⇐⇒ β − α ∈ P deﬁnes a partial order on F which satisﬁes (7). Thus, ≤ is be a semiordering on F , if we can prove its linearity, i.e. α ≤ β or β ≤ α for all α, β ∈ F . This clearly is equivalent to P ∪ −P = F . The next lemma will easily imply linearity of ≤. This lemma actually is the heart of Sol`er’s Thesis. LEMMA 5. Let (E, ) satisfy Assumption 2 and let α ∈ F \{0, ±1}. Then either n 2 −1 or the the vector a = n α en exists in E with length < a, a >= (1 − α ) −n −2 −1 a = n α en exists in E with length < a , a >= (1 − α ) . Proof. We deﬁne for all n ∈ N: xn =
n
αi ei
yn = xn −
,
i=0
an = x2n+1 − α2 x2n ,
1 e0 1 − α2
bn = y2n+2 − α2 y2n+1
From these deﬁnitions we ﬁnd < xn , yn > < xn−1 , yn−1 >
=
1 + α2n < xn−1 , yn−1 >−1
=
1 + α2n (1 −
1 + α2 + α4 + ... + α2(n−1) )−1 = α2 1 − α2
An easy computation shows an ⊥ bm for all n, m ∈ N. The subspace A = (bn )n∈N
⊥
is closed. Hence orthomodularity of E yields E = A ⊕ A⊥ .
Thus, we get a ∈ A and b ∈ A⊥ such that 1 e0 = a − b. 1 − α2 Clearly, a ⊥ an and b ⊥ bn for all n ∈ N. We now set (αn + εn )en a= n
Sol` er’s Theorem
381
where εn is suitably chosen from K. The following computations then yield recursive conditions on the εn : 0
= < b, an >=< b, x2n+1 − α2 x2n > = < a, x2n+1 − α2 x2n > −(1 − α2 )−1 < e0 , x2n+1 − α2 x2n > = < a, y2n+2 − α2 y2n+1 − α2n+2 e2n+2 + α2 α2n+1 e2n+1 + e0 ) − (1 − α2 )−1 (1 − α) = −(α2n+2 + ε2n+2 )α2n+2 + (α2n+1 + ε2n+1 )α2n+3 + (1 + ε0 ) − 1 = −ε2n+2 α2n+2 + ε2n+1 α2n+3 + ε0
Hence ε2n+2 = ε2n+1 α + ε0 α−(2n+2)
(5) Similarly we get 0
= < a, y2n+2 − α2 y2n+1 > = < b, x2n+3 − α2 x2n+2 − α2n+3 e2n+3 + α2n+4 e2n+2 − e0 > +(1 − (1 − α2 )−1 ) = −ε2n+3 α2n+3 + ε2n+2 α2n+4 − ε0
and thus ε2n+3 = ε2n+2 α − ε0 α−(2n+3)
(6) Finally
0 =< b, a0 >=< a −
1 e0 , (1 − α2 )e0 + αe1 >= ε0 (1 − α2 ) + ε1 α 1 − α2
gives ε1 = −ε0 α−1 (1 − α2 )
(7)
If now ε0 = 0, the recursive conditions (5), (6) and (7) imply εn = 0 for all n ∈ N. Hence, in this case we get αn en . a= n
In case ε0 = 0, by Lemma 3 and the recursion formulas (5) and (6) the following vector exists in E: a
=
(αn + εn )en − α
n≥2
=
=
n≥2
(αn−1 + εn−1 )en
n≥2
(α + αεn−1 + (−1)n α−n ε0 )en − α n
n≥2
n≥2
n −n
((−1) α
ε0 )en
(αn−1 + εn−1 )en
382
Alexander Prestel
Thus, again by Lemma 3 the vector a =
α−n en
n
exists in E. This proves the diﬃcult part of the Lemma. The computation of < a, a > or < a , a > resp. is easy (note that we shall use Lemma 3 in the last step):
< a, a > =
1+
αn en ,
n≥1
=
αn en = 1 + α2
n≥1
αn−1 en ,
n≥1
αn−1 en
n≥1
1+α αn en+1 , αn en+1 = 1 + α2 αn en , αn en 2
n
n
n
n
This clearly gives < a, a > =
1 1 − α2
and similarly < a , a > =
1 1 − α−2
From this lemma we get the linearity of ≤ as follows. For every α ∈ F \{0, ±1} we have 1 ∈P 1 − α2
1 ∈ P. 1 − α−2
or
Using (3)(ii) twice we ﬁnd for every α ∈ F : 1 − α2 ∈ P
α2 − 1 ∈ P.
or
Now let β ∈ F \{−1} be given and take α = 4β ∈P (β + 1)2
2 β+1
− 1. This clearly yields
4β ∈P (β + 1)2
or
−
or
− β ∈ P.
Again by (3)(ii) we ﬁnally ﬁnd β∈P Thus linearity of ≤ is proved.
Sol` er’s Theorem
383
3 ARCHIMEDEANITY AND COMPLETENESS OF (F, ≤) The next lemma shows that ≤ is archimedean. Thus by [Prestel, 1984], Theorem 1.20, ≤ is even an ordering and therefore F may be considered as a subﬁeld of R with ≤ induced by the ordinary ordering of R. LEMMA 6. Let (E, ) satisfy Assumption 2 and let ≤ be a semiordering on the ﬁxed ﬁeld F such that < x, x >≥ 0 for all x ∈ E. Then ≤ is archimedean. Proof. Assume that ≤ is not archimedean. Then γ ∈ F exists such that n < γ for all n ∈ N. Then by the laws of semiorderings (cf.[Prestel, 1984], Lemma 1.18), we have 0 < 2n δ < 1 for all n ∈ N in case we take δ = γ −1 . Thus by Lemma 5 a vector hn exists such that < hn , hn >= (1− (2n δ)2 )−1 . Scaling hn by 1− (2n δ)2 we obtain some vector hn with < hn , hn >= 1 − (2n δ)2 for each n ∈ N. By Corollary 4(a) we may also assume that hn ⊥ hm for all n = m. Moreover we may assume that there exists an orthonormal sequence (en )n∈N orthogonal to all vectors hn . Then the vectors fn = hn + 2n δen obviously have length < fn , fn >= 1 and are pairwise orthogonal. Thus (fn )n∈N is an orthonormal sequence in E. Now by Lemma 5 together with Lemma 3 the vector 2−n en x = n ⊥ ⊥
exists in (en )n∈N . From the orthomodularity of E we ﬁnd by (2.1) E = (fn )n∈N ⊥ ⊥
⊥ ⊥
+ (fn )n∈N
⊥ ⊥
and thus x = f + g with f ∈ (fn )n∈N and g ∈ (fn )n∈N . Computing the Fourier coeﬃcients of f w.r.t. the orthonormal system (fn )n∈N , we ﬁnd < f, fn > = < x , fn >=< x , hn + (2n δ)en > = < x , 2n δen >= 2n δ < x , en >= δ . Thus δ −1 f would have Fourier coeﬃcients 1 w.r.t. (fn )n∈N which contradicts Corollary 4 (b) (together with Lemma 3). LEMMA 7. Let (E, ) satisfy Assumption 2. Then every Dedekind cut is realized in the archimedean ordered ﬁeld (F, ≤). Proof. It clearly suﬃces that every Dedekind cut in the interval from 0 to 16 5 is realized. Taking α = 34 in Lemma 5 we see that the vector x = n αn en exists in E and has length 1 16 = < x, x >= = α2n . 5 1 − α2 n
384
Alexander Prestel
Now let β be any real number from the interval [0, 16 5 ]. It is not diﬃcult to see that a subset M of N exists such that α2n (1) β= n∈M
From (2) we get E = (en )n∈M
⊥ ⊥
+ (en )n∈M /
⊥ ⊥
. ⊥ ⊥
⊥ ⊥
Thus x admits a decomposition x = y + z with y ∈ (en )n∈M and z ∈ (en )n∈M . / We claim that < y, y > realizes the Dedekind cut determined by the real number β of (1) in the interval [0, 16 5 ]. In fact, if I is any ﬁnite subset of M we ﬁnd again from (2) that (en )n∈M
⊥ ⊥
= (en )n∈I
⊥ ⊥
+ (en )n∈M \I
⊥ ⊥
.
Thus y admits a representation y = y + y with y ∈ (en )n∈I ⊥ ⊥ (en )n∈M \I . Since (en )n∈I is a ﬁnite orthonormal system, we get < y, y >=
⊥ ⊥
and y ∈
α2n + < y , y >
n∈I
and, hence, in particular
(2)
α2n ≤ < y, y > .
n∈I
Arguing similarly for the vector z which has length < z, z >= 16 5 − β, we ﬁnd for every ﬁnite subset J of N\M
(3)
n∈M /
α2n =
α2n ≤ < z, z > .
n∈J
From (2) and (3) we ﬁnally obtain
(4)
n∈I
α2n ≤ < y, y >≤
16 2n − α . 5 n∈J
Since in (4) the left hand and the right hand side both are rational numbers converging to β in R for increasing sets I and J, and since Q is dense in F (by Lemma 6), the length < y, y > realizes in F the Dedekind cut determined by β.
Sol` er’s Theorem
4
385
THE NONCOMMUTATIVE CASE
As alraedy mentioned at the beginning of this paper, M.P. Sol`er’s Thesis also covers Hilbert spaces over the quaterions H. We will now explain this and also show how some slight changes in the above proof of the commutative case give the theorem in full generality. Let now K be a skew ﬁeld (commutative or not) and ∗ : K → K an involution2 , i.e. for all α, β ∈ K we have (α + β)∗ = α∗ + β ∗ ,
(αβ)∗ = β ∗ α∗ ,
α∗∗ = α
The main noncommutative example is the skew ﬁeld K = H = R + Ri + RjRk of quaterions with i2 = j 2 = −1, ij = k = −ji, and (α + βi + γj + δk)∗ = α − βi − γj − δk for all α, β, γ, δ ∈ R. Now the set F = {α ∈ Kα∗ = α} of symmetric elements of K need no longer be a subskewﬁeld of K, in fact, F need not be closed under multiplication. Next let E be an inﬁnite dimensional Kvector space and : E × E → K a hermitian form on E which now has the properties
< αx + βy, z > = α < x, z > +β < y, z > < z, αx + βy > = < z, x > α∗ + < z, y > β ∗ < x, z > = < z, x >∗ for all α, β ∈ K and x, y, z ∈ E. All the other notions like orthogonality, anisotropy, orthomodularity and orthonormal sequences are deﬁned as in the commutative case. The Theorem of Sol`er then says that the only inﬁnite dimensional hermitian Kvector spaces (E, ) which are orthomodular and admit an orthonormal sequence (en )n∈N are the usual Hilbert spaces over R, C, and H. The proof of this theorem can be obtained from the one above by observing the following modiﬁcations. We ﬁrst observe that Lemma 3 and Corollary 4 also hold in the generalized situation.3 The proofs are literally the same. Also Sol`er’s main lemma (Lemma 5) now holds. The proof, however, is more involved (see [Sol`er, 1995], Lemma 4, 2 Sometimes
also called an antiinvolution an ‘isometry’ of subspaces U and V we here mean a Klinear isomorphism f satisfying < f (x), f (x) >=< x, x > for all x ∈ U . In the commutative case this clearly implies < f (x), f (y) >=< x, y > for all x, y ∈ U . 3 By
386
Alexander Prestel
and its proof). Nevertheless, it only uses Lemma 3 and Corollary 4, with a slight generalization in (b): (b’) Let α = 0 and (fn )n∈N be a sequence of pairwise orthogonal vectors such that ⊥ ⊥ < fn , fn >= α for all n ∈ N. Then there is no vector f ∈ (fn )n∈N having all Fourier coeﬃcients equal to 1 w.r.t. (fn )n∈N . The proof of (b’) is just the same as that of (b) using (3.2). From Lemma 3 and Corollary 4 we obtain that the set P = {< x, x > x ∈ E} of lengths deﬁnes a Baerordering on K, i.e. α ≤ β iﬀ β − α ∈ P, linearly orders F and satisﬁes 0 ≤ α, 0 ≤ β 0≤α
(1)
⇒ 0≤α+β ⇒ 0 ≤ γαγ ∗ 0≤1
for all α, β ∈ F and γ ∈ K. All these properties are proved as in Section 3. Concerning linearity, i.e. 0 ≤ α or α ≤ 0 for all α ∈ F , one should be aware of the fact that L = Q(α) is a commutative subskewﬁeld of K, and that ≤ restricted to L therefore yields a semiordering. This very fact also allows us to use the proof of Lemma 6 in order to see that ≤ is an archimedean Baerordering of K, i.e. 0 ≤ λ < n1 for all n ∈ N\{0} can only hold for λ = 0. At this point we make use of S. Holland’s result (see [Holland, 1977], Theorem 2 and Corollary 3) that a skew ﬁeld K with involution ∗ which admits an archimedean Baerordering is isomorphic to a subskewﬁeld of R, C, or H and is actually equal to R, C, or H in case every Dedekind cut is realized in F . This last fact, however, follows literally as in Lemma 7. In [Holland, 1995] Holland actually shows that a noncommutative skew ﬁeld K with involution ∗ can not be generated as a ring by the set F of symmetric elements of K. But then a theorem from Dieudonn´e (cf. [Holland, 1995], Lemma 1) states that K is a generalized quaterion algebra K = Z(αβ) over its center Z, that is K = Z + Zi + Zj + Zk with i2 = α,
j 2 = β,
ij = −ji = k, and
(α0 + α1 i + α2 j + α3 k)∗ = α0 − α1 i − α2 j − α3 k for all α0 , . . . , α3 ∈ Z. It now follows that F = Z is a commutative subﬁeld of Kand thus by the above arguments has to be isomorphic to R. This then clearly implies that K = H.
Sol` er’s Theorem
387
Now that we know that the ﬁeld K of scalars is R, C, or H, and that is positive deﬁnite, the completeness of (E, ) follows as in Step 3 of Section 2. Remark. It should be pointed out that the linearity of the Baerordering ≤ deﬁned by P was essential in the above arguments. If we did not have P ∪ −P = F , i.e. ≤ is only a partial ordering of F satisfying (1), it is not known whether P could be extended to some Baerordering on K or not. BIBLIOGRAPHY [Engesser and Gabbay, 2002] K. Engesser and D. M. Gabbay. Quantum logic, Hilbert space, revision theory. Artiﬁcial Intelligence, 136 (2002), 61100. [Holland, 1977] S. S. Holland, Jr. Orderings and square roots in ∗ ﬁelds. J. of Algebra, 46(1977), 207219. [Holland, 1995] S. S. Holland, Jr. Orthomodularity in inﬁnite dimensions; a theorem of M. Sol`er. Bull. AMS, 32 (1995), 205234. [Keller, 1980] H. A. Keller. Ein nicht klassischer Hilbertscher Raum. Math.Z., 172(1980), 4149. [Keller et al., 1998] H. A. Keller, U. M. K¨ unzi, and M. P. Sol`er. Orthomodular spaces. In: Orthogonal geometry in inﬁnite dimensional vector spaces. Bayreuther Mathematische Schriften, 53 (1998), 171250. [Maeda and Maeda, 1970] F. Maeda and S. Maeda. Theory of symmetric lattices. SpringerVerlag, Heidelberg, 1970. [Prestel, 1984] A. Prestel. Lectures on formally real ﬁelds. Lecture Notes in Math. 1093 (Springer), 1984. [Prestel, 1995] A. Prestel. On Sol`er’s characterization of Hilbert spaces. Manuscripta Math., 86 (1995), 225238. [Sol` er, 1995] M. P. Sol` er. Characterization of Hilbert spaces by orthomodular spaces. Communications in Algebra, 23 (1995), 219243.
This page intentionally left blank
HANDBOOK OF QUANTUM LOGIC AND QUANTUM STRUCTURES: QUANTUM LOGIC Edited by K. Engesser, D. M. Gabbay and D. Lehmann © 2009 Elsevier B.V. All rights reserved
389
OPERATIONAL QUANTUM LOGIC: A SURVEY AND ANALYSIS David J. Moore and Frank Valckenborgh
1
INTRODUCTION
One of the basic assumptions of standard quantum physics is that the collection of experimentally veriﬁable propositions that are associated with a quantum system has a logicoalgebraic structure isomorphic with the lattice of closed linear subspaces of some complex Hilbert space or, dually, with the lattice of orthogonal projections on these subspaces. The quantum logic approach to the mathematical and conceptual foundations of quantum mechanics attempts to provide an advanced perspective on the rather highlevel languages of standard textbook classical and quantum theory, in the hope that the resulting theory leads to additional simpliﬁcations in the conceptual framework of conventional quantum theory. For example, for a slightly idiosyncratic basic textbook on elementary quantum mechanics with an operational twist, the intrepid reader can consult the blue book by Constantin Piron 1998). The goal of the particular branch of operational quantum logic — the subject of this work — consists ﬁrst and foremost in developing a suitable global framework in which the general discussion of physical processes can take place. The driving principle here is the conviction that the primitive terms of a candidate mathematical framework in which at a second stage more concrete models for physical phenomena can be developed, should be based, as much as possible, on concrete physical notions and operations. In other words, the abstract language should be developed in terms of operationally welldeﬁned entities, closely related to the actual practice of the experimental physicist, rather than on mathematically possibly more convenient or elegant but experimentally less motivated terms. That is, in the development of a suitable framework we have to adopt the more concrete line of thinking of the experimental physicist, where the characterisation of a system depends on how we can act upon that system. One then exploits explicitly the empirical fact that there exists a certain class of systems, socalled measuring instruments, that are capable of undergoing macroscopically observational changes triggered by their interaction with single microsystems (Kraus, 1983). By way of contrast, in the construction of a mathematical theory the choice of primitive terms is usually motivated internally (how else?) by the elegance and simplicity of the resulting theory; in physics this choice then should ideally be justiﬁed externally.
390
David J. Moore and Frank Valckenborgh
Instead of deﬁning properties of a physical system by reference to its putative internal structure, one thus proceeds by analyzing its external relationships with other systems; this information is then used to attribute properties to the system under investigation. We will argue that an operational approach — at least from a methodological point of view — on fundamental physical concepts such as system and particle, in particular composite particle, evolution and dynamics, space and spacetime, and so on, possibly also leads to additional clariﬁcations and maybe even a reformulation of the foundations of physical theory. More concretely, we will indicate how the resulting language encompasses many of the essentials of both classical physics and quantum physics, and so is obviously of considerable independent interest. From a philosophical perspective, it is important to remark that the word operational in this context has to be considered as a pragmatic attitude — in which we privilege mathematical terms possessing a deﬁnite physical heuristic in the development of a framework theory overarching the plethora of speciﬁc models — rather than a doctrine. Speciﬁcally, we do not commit ourselves to a deﬁnite metaphysical position when we insist on developing a general mathematical language by abstracting from concrete physical operations. On the contrary, one of the important underlying motivations to do so is based upon the conviction that many of the interpretational issues that haunt standard quantum mechanics, in particular those associated with the wavemechanical formalism, will present themselves in a diﬀerent light when the primitive terms of the formalism have a clear physical interpretation. For that matter, the Q in OQL seems to have more historical roots, when the necessity for a deeper analysis of the descriptional framework of physical systems became more acute with the investigation of microsystems. Finally, trying to analyze the meaning of the L would almost certainly stir up an academic hornet’s nest, so we happily refrain from doing so, positions of members of the quantum logic research community covering the whole available spectrum from a purely algebraic perspective on the usage of the word ‘logic’ to much stronger philosophical commitments. We have to emphasise, however, that OQL has an explicit physical origin, contrary to the metaphysical basis of empirical quantum logic sensu Finkelstein (1968) and Putnam (1968). It is also from this perspective that at least some part of the general operational methodology and philosophy of the researchers active in the thriving domains of quantum information science and quantum computation is actually quite close in spirit to the conceptual methodology used by OQL, the behaviour of individual physical systems rather than statistical ensembles of such objects moving more into the focus of attention. On the other hand, it is only fair to mention that there are also important diﬀerences. Speciﬁcally, we will argue that at least the original incarnations of OQL have some diﬃculty when confronted with the notions of composite systems on the one hand, and with the explicit sequential composition of quantum processes on the other. Indeed, the explicit incorporation of multistage quantum processes in more general terms seems to require some form of typing, which reﬂects a conditional aspect in performing consecutive operations, due to
Operational Quantum Logic: A Survey and Analysis
391
the probabilistic character of measurements on quantum systems (at least from the perspective of the system itself). In other words, both the serial composition and parallel composition of physical operations are usually not explicitly accounted for. In this context, we will also have another look at the important work of Pool (1968a, b). Distinct approaches to formulating an adequate framework for a better understanding of physical phenomena attribute basic physical signiﬁcance to diﬀerent concepts, reﬂecting the fact that various notions play an important conceptual role in physical thinking. Consequently, it is clear that a priori various choices for the primitive terms can be made, and this handbook reﬂects some parts of the corresponding spectrum of possibilities. Most authors ascribe a fundamental importance to the two concepts of state and observable; the physical interpretation of these notions, however, varies considerably from one author to another. Let us brieﬂy discuss these notions from an operationally oriented perspective, more details about the particular incarnations in our perception of OQL being referred to a later section. The abstract notion of a state that can be assigned to a physical system is a very natural one from the more static perspective that focuses on suﬃciently stable physical objects in the here and now, but it is not necessarily restricted to this setting. States come in two conceptually distinct ﬂavours. In principle, knowledge of the state allows one to predict the experimental behaviour to the best of one’s knowledge, but often only in probabilistic terms. Indeed, it is one of the essential features of quantum physics that for each state — even states that apparently are not reducible to more fundamental states — there exist many observables with highly nontrivial probability distributions. Contrary to the orthodox ensemblebased interpretation of quantum mechanics, we claim that pure states can be construed — and that it is a conceptual advantage to do so — as abstract names that encode the possible singular realisations of a given particular physical system. In other words, pure states are admittedly somewhat idealised objects that adopt the somewhat privileged role of being able to refer to individual physical systems in an ensemble, at least under mildly favourable conditions. As a consequence, there will always exist some kind of duality between the set of pure states and the collection of properties attributed to a physical system, reﬂecting the wellestablished practice among physicists to describe a general system either by its properties or by its states. In fact, on the one hand knowledge of the pure state of the individual system allows one to predict its behaviour, or its properties, in the best possible way, and on the other hand we expect that the knowledge of all its actual properties determines the pure state completely. Often, it is suﬃcient to know a particular subset of its actual properties to identify the state; such a collection of properties is then statedetermining. Alternatively, mixed states are often and meaningfully regarded as a convenient heuristic that summarises the probabilistic behaviour of an ensemble of systems that can be subjected to various experimental protocols. We can always interpret these, with some care, as a probabilistic mixture of pure states; the mixed state
392
David J. Moore and Frank Valckenborgh
then also incorporates the maximal information we can assign a priori to the individual objects in the ensemble. Mathematically speaking, for a given physical system we then have to assume the existence of an abstract collection of states, say M, in this generalised sense, that forms a convex set, the pure states corresponding with the extreme points, say ∂e M. In practice, pure or mixed states can often be assigned to physical systems by a physically welldeﬁned preparation procedure. Sometimes however, this is not the case, and we are forced to attribute a state to an object that is beyond our immediate control, based on our experience with objects that can be prepared in a certain state. Observables on the other hand can be construed as encoding the abstract relation between some physically relevant collection of properties of a measurement device and corresponding properties assigned to our physical system. For example, after performing an experiment — an interaction between a physical system and the appropriate device — the position of a pointer of the measurement device should better reﬂect a possible property of the system under investigation. Operationally speaking, properties assigned to a system will then be construed as candidate elements of reality corresponding to deﬁnite experimental projects that are deﬁned for that particular physical system. From a mathematical perspective, the collection of physically useful properties associated with the measurement device can be represented by a boolean algebra of some sort, and this structure should then be faithfully represented also at the level of the description of the physical system. Often, such a correspondence will be subject to additional constraints at the level of the measurement devices, due to symmetry considerations, and such constraints should then be appropriately incorporated, also with regard to the system. Standard examples are given by the position and momentum observables with respect to a given reference frame, coordinates and intervals of space and time being deﬁned in a physically standardised way (Audoin & Guinot, 2001). In many approaches, a distinguished role is played by the subset of twovalued observables. In conventional quantum mechanics, such observables will correspond with pairs of orthogonal projections in the complex Hilbert space. In our view, a general operational attitude — methodologically speaking — towards the formulation of physical theories has other advantages. In fact, one is inclined to critically examine the various concepts that otherwise would be taken for granted. For example, the worldview of many working physicists is still (dangerously?) close to the classical Descartian one that pictures physical particles as pointlike particles that move — possibly under the inﬂuence of ﬁelds of some sort — in an otherwise independent spatiotemporal containing arena (Piron, 2002). We think that one of the lessons that we have to learn from the progress of technology is to take nonlocality seriously, although it is admittedly very diﬃcult for us macroscopic beings to think in terms of individual microscopic objects that have an extension but no parts, acting as a single whole. Speciﬁcally, neutron interferometric experiments performed during the last decennia of the previous century indicate that our current notion of spacetime is more subtle than anticipated. Indeed, individual neutrons can be manipulated in spatially
Operational Quantum Logic: A Survey and Analysis
393
separated regions, hence at least for a microsystem the property of being localised in a certain region of space cannot be decided with certainty from the knowledge of its state. As some sort of an afterthought, we can also add that it may be useful to regard the notion of a physical system as being ‘composed’ of more fundamental entities in an operational light. For the sake of completeness, for the convenience of the reader, and to make the paper as selfcontained as possible, we have included many of the proofs of at least the simpler statements, so that the reader is not only spared the interruption of the train of thought by feeling the need to show a particular result before proceeding with the text, but also the task of having to look up the widely dispersed and — with the progress of time — not necessarily easily available original sources, both papers and monographs, if she or he wishes to do so. 2
SOME HISTORICAL REMARKS
The subject of quantum logic started essentially with the monumental treatise of John von Neumann (1932, 1955), in which the complex Hilbert space model of quantum physics was developed and its structural, logical and physical implications explored. One can argue that the intellectual heritage of this work was the essential dominance of the complex Hilbert space model for the mathematical description and the physical and philosophical analysis of all physical systems, a priori without any additional restrictions. It is, however, rarely appreciated that von Neumann himself started expressing his dissatisfaction with the resulting Hilbert space framework not long after (R´edei, 1996). One manifestation of this eventually led to the pioneering paper — coauthored by Garrett Birkhoﬀ — that was appropriately titled “The Logic of Quantum Mechanics”, published in 1936 in the Annals of Mathematics (Birkhoﬀ & von Neumann, 1936). In this paper, the authors take a more structural view, based upon a deeper analysis of the “logical” structure one may expect to ﬁnd in general physical theories, and in particular quantum mechanics, for the description of physical systems for which the collection of experimentally veriﬁable propositions does not seem to conform to the laws of classical logical discourse. In this paper, the idea was presented that the primitive mathematical building blocks (and the relations connecting them) that are used in the abstract mathematical description of physical systems, should take into account the basic structure of some sort of more concrete and admittedly slightly idealised operational calculus about physical systems, in which one tries to isolate key aspects of the general behaviour of such systems in an experimental context. In particular, notions of “experimental proposition”, “observation space”, “experimental implication” and the like gained some ground in developing the structural foundations of a physical theory. This perspective led almost automatically to a shift of the burden of the conceptual weight to the projective structure associated with the traditional complex Hilbert space model, the linear structure becoming a secondary — though technically extremely useful — construction, in view of the ﬁrst fundamental theorem
394
David J. Moore and Frank Valckenborgh
of projective geometry. A more detailed analysis of this part of the story is given in this handbook by the paper by Isar Stubbe & Bart Van Steirteghem (2007). In 1957, George W. Mackey, in an inﬂuential article in the American Mathematical Monthly that was later expanded in an important monograph (Mackey, 1957, 1963), sketched a probabilistic framework for the mathematical foundations of quantum (and classical) mechanics. In his setup, he starts with an abstract set of observables O that can be eﬀectuated on a physical system; this set is closed under the action of real Borel functions, which represent measurements of the same observable followed by a simple computation of f . States attributed to this system / M, where the latter set denotes the collection are conceived as functions μ : O of probability measures on the Borel or Lebesgue sets in R. An important role is played by Mackey’s questions, which correspond 4 6 with the twovalued observables that take their values in the outcome set 0, 1 . Each observable A generates a family of questions of the form χE (A), where χE is the characteristic function of some Borel set E ⊆ R. The whole mathematical structure is then determined by the restriction of the states to the set of questions. Mackey then observes that one can partially order the set of questions by the prescription 4 6 4 6 (1) Q1 ≤ Q2 iﬀ ∀ μ : μ(Q1 )( 1 ) ≤ μ(Q2 )( 1 ) and deﬁne a relation of orthogonality on this set by setting Q1 ⊥ Q2 iﬀ Q1 ≤ 1−Q2 . Notice the reliance on probabilistic arguments to deﬁne this order relation. In his 1963 monograph, this framework was reﬁned and extended into a system of nine axioms for the mathematical foundations of quantum mechanics. His Axiom VII is somewhat remarkable and ad hoc in the sense that it explicitly demands that the partially ordered set of all questions associated with a quantum system should be isomorphic with the lattice of all closed subspaces of a separable, inﬁnitedimensional complex Hilbert space. Parts of this work were generalised in an important paper by Varadarajan (1962) into a more general probabilistic formalism, that considerable extended the axiomatic model that was proposed by Kolmogorov (1956) to take into account the phenomenology associated with quantum systems. More explicitly, the author proceeds in developing an experimentally motivated probability calculus on orthomodular σorthoposets. In particular, he developed and thoroughly analysed the notion of observables — generalised random variables with respect to his framework — and their mathematical characterisation in terms of Boolean subalgebras. In addition, he generalised the functional calculus of the more conventional quantum mechanical observables in this setting, and the related problem of simultaneous measurability for such observables. Some of the central tenets of the axiomatic structure of quantum theory as it was developed by the socalled Geneva school from the early sixties or so consist in the aﬃrmation that it is possible to erect the framework of quantum theory on the basis of the notion of certainty instead of probability, and in the assumption that and the physical motivation why the collection of experimentally veriﬁable propositions has the structure of a lattice and not merely a partially ordered set of some sort (Piron, 1964; Jauch, 1968; Jauch & Piron, 1969; Piron, 1976). Remarkably, the
Operational Quantum Logic: A Survey and Analysis
395
axiomatic framework that was developed by this group of researchers is explicitly nonprobabilistic; in the words of Jauch & Piron (1969): “If one introduces probability at this stage of the axiomatics one has diﬃculties of avoiding the criticism of Einstein that a state is not an attribute of an individual system but merely the statistical property of a homogeneous ensemble of similarly prepared identical systems.” This desideratum requires that any probabilistic characteristics associated with the system itself have to be eliminated right from the start, and so the restriction to pure states only as part of the primitive objects of the framework is essential in this program. Speciﬁcally, the aims of Piron were at least twofold: (1) Develop a general theory valid for both classical and quantal systems; (2) Justify the usual formalisms for these two extremal cases by an intrinsic characterisation of the structure of their sets of observables. This program culminates in Piron’s celebrated representation theorem, about which more will be said in the next section (Piron, 1964, 1976). At the end of the day, one then likes to reconstruct explicitly the usual concrete particle models of classical Hamiltonian mechanics and Hilbert space quantum physics in this light. Hereto, one has to appeal to more advanced techniques of group representation theory. The by now classical work of Mackey on the induced representations of locally compact groups has proven instrumental in this respect; for a rather leisurely overview, see Mackey (1968). The operational statistics approach that was developed in an extensive series of papers by Charles Randall and Dave Foulis in Amherst is in many respects a parallel branch in the evolutionary tree of foundational approaches (see Foulis & Randall, 1972; Randall & Foulis, 1973). For a more elaborate analysis of this work and its more recent development, see the contribution on test spaces by Alex Wilce (2008) in this handbook. For the relation between the Geneva school approach and the Amherst approach, see Foulis, Piron & Randall (1983) and Coecke, Moore & Wilce (2000) for a more general discussion. For the sake of completeness, let us also mention that a third highly operationally motivated sibling was developed by the school of Ludwig, starting from the early 1960’s (see, for example Ludwig, 1983). For another account of parts of this approach, the reader can consult the monograph by Kraus (1983). In the light of the recent explosion of the domains of quantum computation and quantum information science, new axiom systems have been proposed that emphasise various concepts that stem from the more specialised perspectives of this type of research. Here is a quick anthology of some more recent attempts: Fuchs (2001); Hardy (2001); D’Ariano (2007a, b), and there are undoubtedly many others. Without going into details, we only remark that — to avoid some of the conceptual and methodological pitfalls of the past on the one hand, and to highlight the successes achieved and the weaknesses exposed in the operational quantum logic approach to the foundations of the physical sciences on the other — we consider it useful and appropriate also from this more specialised perspective
396
David J. Moore and Frank Valckenborgh
to give a reasonably detailed annotated exposition of what has been achieved so far by this part of the quantum logic community. 3 QUANTUM OBJECTS The empirical basis for the statistical laws of quantum physics consists in the reproducibility of the relative frequencies of the results one obtains when subjecting microsystems, prepared according to a given experimental protocol, to various measurement procedures. It is indeed an empirically wellestablished fact that there exist macroscopic devices capable of undergoing macroscopically observable changes when interacting with such microsystems (Kraus, 1983). Experience also tells us that the outcome of a single experiment is not determined completely by the speciﬁcations of the preparation procedure and the macroscopic change that occurred in the measuring instrument after the interaction between the system and the device. As we have already indicated, there exist various established schools in the quantum logic enterprise and they attribute basic physical signiﬁcance to diﬀerent concepts as the primitive building blocks. A priori, various choices can be made: in general, a physical system can be described by a class of events, propositions, properties, operations, etc., and a class of states. Sometimes the events alone are seen as fundamental and states are considered as derived entities, sometimes both events and states are considered as primitive, and sometimes the collection of states plays a central role. In operational quantum logic, the Geneva school approach occupies a central position. The central observation of this approach is that the abstract sets ∂e M of pure states and L of properties each admit an abstract mathematical structure, induced by concrete physical considerations on the concrete primitive notions of particular physical system and deﬁnite experimental project. It is important to remark that these properties can be introduced in a nonprobabilistic way. In the following development, we will ﬁrst work in the apparently slightly more general probabilistically oriented framework, and subsequently add the reﬁnements that are particular to the Geneva approach. The highly idealised concept of a physical system is central in physics, and usually taken for granted, without much further comments. As such, the potential diﬃculties it embodies tend to escape attention and scrutiny. From the more objectoriented viewpoint that we adopt in this section, however, it may be wise to add a few comments on this potentially problematic notion. Indeed, it is important to clearly express the ambiguities inherent in any precisiﬁcation of a physical concept rather than trying to ignore or minimise their cognitive importance and consequences. A physical system is usually conceived as a suﬃciently wellcircumscribed part of reality external to the physicist, in the sense that its interaction with its surroundings can either be ignored or modeled in an eﬀective way. An extreme case corresponds with the notion of a closed system, where the interaction with the external world can be totally neglected. In practice however, it is hard, if not impossible, to isolate physical systems completely from their en
Operational Quantum Logic: A Survey and Analysis
397
vironment, and the classical thermodynamic concept of an open system is perhaps a more appropriate term in the other extreme case where the very identity of the system becomes somewhat blurred. For example, we can choose to investigate the properties of a certain part of spacetime, identiﬁed with respect to some reference frame that attributes coordinates of space and time to potential events, and then electromagnetic radiation generated by the environment is often an issue. Another yet not unrelated reason why we have to be careful in this case, is the experimental reality of nonlocality, that manifests itself in some wellknown cases (see for example, Rauch et al., 1975). The typical adventures of the paradigmatic couple Alice and Bob dramatically illustrates some of these aspects. Of course, it is part of conventional scientiﬁc methodology that some idealisation is inescapable in the elaboration of our physical models, considered as abstract and hence partial reﬂections of concrete world situations. In this text, we will take the point of view that the state attributed to an individual member of a closed or almost closed physical system is always a pure state. This then can also be taken as a characterization of the type of physical systems that we favour in this development. On the other hand, if we happen to encounter a physical system where this would not be the case, we may be inclined to embark on a more comprehensive investigation of that particular phenomenon. Physical systems can be investigated by their potential interaction with a privileged class of macroscopic probing devices. Notice that the inherently active nature of experimentation, which implies that probing a system will in general perturb and sometimes even destroy it entirely, is an essential part of modern scientiﬁc methodology. A deﬁnite experimental project — also called question or sometimes test — then consists of a complete experimental protocol: a measurement device and instructions on how to properly use it, together with a rule for interpreting its possible results in terms of two alternatives only: either the positive result, identiﬁed with a particular predeﬁned set of conﬁgurations of the measurement device before the actual measurement takes place, would be obtained (yes), or not (no). We emphasise that this notion explicitly refers to the empirical existence of independent macroscopic arrangements that can interact with single individual samples of the physical system under investigation, and that can leave an objective macroscopic eﬀect due to such a singular interaction; this eﬀect is then interpreted as the occurrence of the positive result for the measurement. In other words, this approach is explicitly relational in that systems are characterised by their interaction — hypothetical or de facto — with other systems. The class of deﬁnite experimental projects relative to the system will be denoted by Q. To avoid any confusion and possible conﬂation with Mackey’s notion of a question, we will avoid the use of the term “question” in the context of the Geneva approach. Observe that the notion of a deﬁnite experimental project is suﬃciently general, since each multioutcome experiment can be chopped up into a collection of twovalued tests. We say that a deﬁnite experimental project α ∈ Q is certain for a given singular realisation of a particular physical system (with state μ ∈ M) when the positive result would be obtained with certainty, should the experiment be (properly) exe
398
David J. Moore and Frank Valckenborgh
cuted. For notational brevity, we shall encode this relationship between the set of all possible states M and the class of deﬁnite experimental projects Q succinctly as μ α, or sometimes also T (μ, α); alternatively, we use the notation F (μ, α) whenever the negative result would obtain with certainty. Note the counterfactual locution in this statement! It implies that the notion of certainty has a sense before, after, and even in the absence of an experiment. The crucial point is that deﬁnite experimental projects are hypothetical, so that their certainty or otherwise can be regarded as an objective feature of the particular physical system. In practice, we often convince ourselves that a deﬁnite experimental project is certain for some preparation of a physical system by running the experiment on a number of samples of the physical system; if the positive result is always obtained, we have the right to claim that similarly prepared new samples will also yield a positive result, if we would actually run the experiment (Aerts, 1983). The emphasis of the founders of the Geneva approach on the notion of certainty, in contrast to probability, at this early level of the formalisation circumvents Einstein’s criticism on conventional quantum mechanics that a state is not an attribute of an individual system but rather of a homogeneous ensemble of similarly prepared systems (Jauch & Piron, 1969). The class Q can then be endowed with an elementary mathematical structure by exploiting these concrete epistemological considerations. Speciﬁcally, there exists a physically natural preorder relation — encoding phenomenological implication — where we say that α & β when α certain implies the certainty of β, for any preparation or realisation of the particular physical system under investigation. Symbolically, (2) α & β
iﬀ
∀ μ ∈ M : μ α implies μ β
The advantage of this counterfactual notion of certainty is that it allows one to give a sensible justiﬁcation for claiming that more than one deﬁnite experimental project is certain for a given single sample of a particular physical system, even when the various experimental conditions are incompatible. From an operational perspective then, the problem is the following: Given a family of not necessarily compatible experimental protocols, is it possible to construct a deﬁnite experimental pr