The Conway-Kochen Argument and relativistic GRW Models Angelo Bassi ∗ Mathematisches Institut der Universita¨t Mu¨nchen, Theresienstr. 39, 80333 Mu¨nchen, Germany, The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, 34014 Trieste, Italy. GianCarlo Ghirardi † 7 0 Department of Theoretical Physics and I.N.F.N., Strada Costiera 11, 34014 Trieste, Italy, 0 The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, 34014 Trieste, Italy. 2 n a J 2 2 In a recent paper, Conway and Kochen proposed what is now known as the “Free 4 Will theorem” which, among other things, should prove the impossibility of combining v 9 GRW models with special relativity, i.e., of formulating relativistically invariant models of 0 2 spontaneouswavefunctioncollapse. Sincetheirargumentbasicallyamountstoanon-locality 0 proof for any theory aiming at reproducing quantum correlations, and since it was clear 1 6 sinceveryalongtimethatanyrelativisticcollapsemodelmustbenon-localinsomeway,we 0 h/ discuss why the theorem of Conway and Kochen does not affect the programof formulating p relativistic GRW models. - t n a u q : KEY WORDS: Free Will theorem, entangled states, non-locality, collapse models. v i X r a 1. THE CONWAY-KOCHEN ARGUMENT We briefly review the argument by Conway and Kochen [1]. Following their way of presenting it, we do not assume any particular theory underlying physical phenomena; we only assume that certainparticularphysicalsystemsexist—independentlyofhowtheyaredescribedbyanytheory— uponwhichmeasurementscanbemade;morespecifically,weassumethatthereexistsystemswhich we call “particles of spin 1”, upon which operations which we call “measurement of the square of the spin along the direction n” can beperformed (n denotes any direction in the three-dimensional ∗Electronic address: [email protected] †Electronic address: [email protected] 2 2 space); we call S the outcome of such a kind of measurement along the direction n. We assume n thatwhenx,y,z representanorthogonaltripleofdirections,thethreecorrespondingmeasurement- 1 operations can be simultaneously performed on the system ; we assume furthermore, in agreement with Quantum Mechanical rules, that: 2 S takes only the value 0 or 1, for any n; n (1) 2 2 2 For any orthogonal triple x,y,z, one has: S +S +S = 2, x y z whenever the three corresponding measurements are (simultaneously) performed. Trivially, prop- 2 2 2 erties (1) imply that one of S , S , S is 0, while two are 1. x y z We finally assume the standard formalism of special relativity, in particular concepts like the backward light-cone and space-like separated regions of spacetime. 1.1. The Postulates Conway and Kochen consider the following three axioms: TWIN: It is possible to produce two spin 1 particles which are in the state of “total spin 0”, meaningwiththisthatifameasurementofthesquareofthespinalongthedirectionnisperformed 2 on one particle, giving the outcome S , then a measurement of the square of the spin along the n 2 same direction n performed on the other particle gives the same outcome S . Moreover, such a n property does not depend on the relative position of the two particles and on the relative time on which the two experiments are performed; in particular, it holds true when the two measurements are space-like separated. FREE: Each experimenter can freely choose any direction n along which to perform the measure- ment. FIN: Information cannot travel at a speed greater than the speed of light. Inthelastaxiom, weusedtheterm“information” inanintuitive sense, withoutspecifyingwhat it means; though we do not like to resort to such a vague term in setting the axioms of any logical reasoning, we use it simply to adhere to the original formulation of [1]. 1 This property reflects the well known fact that, in the case of a spin 1 particle, the squares of the spin operators along threeorthogonal directions commuteamong themselves. 3 1.2. The Argument Let us consider two spin 1 particles which are in a state of total spin 0; let us label with “a” one of the two particles and with “b” the other one, and with “A” an experimenter who performs a measurement on a, and with “B” one who performs a measurement on b. Let n, m and ℓ be three orthogonal axis along which A decides to perform three practically simultaneous measurements of the square of the spin of particle a. We assume that the outcome of the measurement of the square of the spin of particle a along the direction n is a function of all the information α contained in the backward light-cone of the particle (with respect to the spacetime point where the measurement is performed), and of the 2 other two directions m and ℓ chosen by A: 2 2 S S (m,ℓ;α) (2) a:n a:n ≡ 2 2 2 (we have slightly changed the notation from S to S in order to distinguish when S refers to a n a:n n measurement performed on particle a and when it refers to a measurement performed on particle b, along the direction n). In a similar way, if B decides to simultaneously measure the square of the spin of particle b 2 along three orthogonal directions i,j,k, we assume that the outcome S may depend only on the i information β contained in the backward light-cone and on the other two directions j,k chosen by B: 2 2 S S (j,k;β) (3) b:i b:i ≡ Assume finally that the two experimenters perform their measurements at space-like separated regions, so that there cannot be any exchange of information among them before the two measure- ments are over. The argument now goes as follows. Because of TWIN, if A and B choose a common axis n, then the two outcomes must be perfectly correlated: 2 2 S (m,ℓ;α) = S (j,k;β), (4) a:n b:n for any direction n. But, because of FREE, B can choose to perform his measurement along any 2 orthogonal triple n,k,j, with n fixed. Because of FIN, such a choice cannot affect S (m,ℓ;α). a:n 2 The necessity of allowing, in principle, that the outcome along n depends on the other chosen directions follows from the so-called Kochen-Speckertheorem [2]. 4 This means that the outcome of the spin measurement along n cannot depend on which other two orthogonal directions j,k are chosen by the experimenter, i.e. : 2 2 S (j,k;β) = S (β). (5) b:n b:n 2 Obviously, a similar conclusion holds for S . a:n 2 2 The contradiction now arises because, as it is well known, a function like S (α) or S (β) a:n b:n 2 cannot exists [2]. As a consequence, the outcome S of the measurement of the square of the a:n spin of particle a (or b) cannot be uniquely determined by the past information contained in the backward light-cone from the region where A (or B) performs its measurement. 1.3. Comments The provocative conclusion of Conway and Kochen is that the particles’ response to the experi- ment is free, i.e. that the outcome of an experiment cannot be entirely determined by the previous information accessible to whom performs the measurement, but indeed the moral stemming out of the above argument is well known and less surprising: no local theory exists which fully agrees with quantum mechanical predictions. As amatter offact, whatConway andKochenhave shownis that the three postulates listed above, plus the existence of a functional relation between the outcomes of certain spin experiments and certain “information”, lead to a contradiction, from which they conclude that such a functional relation cannot exist. But indeed, after the work of Bell it is well known that the conclusion is a different one: Nature is non-local, i.e. FIN is wrong, if, as the authors seem to suggest, information—apart from its ambiguous meaning—includes everything which might possibly determine an event (in our case, the outcome of a certain experimental pro- cedure). And indeed, the above argument was first proposed by P. Heywood and M.L.G. Redhead [3] and by A. Stairs [4], further explored by H.R. Brown and G. Svetlichny [5] and subsequently generalized by A. Elby [6] (see also [13]); in these papers the above theorem is correctly presented asanon-locality proof,itsnovelty beingthatitdiffersfromtheoriginalproofbyBell, bycombining 3 ideas previously related only to contextuality . The reason why after Bell’s work one has to conclude that Nature is non-local is the following: Bell’s theorem does not require the existence of a functional relation between the outcomes of experiments and past information. It simply requires the FREE assumption, the analog of the TWIN axiom for 1/2-spin particles, and Bell’s definition of locality, which in some sense is the 3 Wethank S.L. Adlerand D. Du¨rr for having brought theabove papers to ourattention. 5 analog of the FIN axiom, even though it is expressed in clearer mathematical and physical terms. From these three axioms—we insist: without assuming any other functional dependence—one can derive an inequality which turns out to be violated by Nature. Accordingly, one of the three axioms must be wrong. Since TWIN has been experimentally verified and no-one is willing to deny FREE, then Bell’s locality must be violated: Nature, in a very precise sense, is non-local. Here we do not want to discuss the merits of the different proofs, whether Bell’s definition of locality is more or less general than those which have been subsequently proposed, included the above FIN axiom; the moral is basically the same as the one given by Bell’s theorem: any theory, in one way or another, must be non-local if it has to be empirically equivalent to Quantum Mechanics. We stress it once more: the series of experiments performed by A. Aspect [7, 8] proved that Nature, by violating Bell’s inequalities, violates Bell’s condition of locality: this means, as far as we understand physics now, that Nature is non-local, that something happening in some region of space can affect the state of far away systems. On the other hand, as proven in [9, 10] for a quantumtheory with thereduction postulate, such non-locality cannotbeusedto signalat aspeed greater than the speed of light; this is what has been called the “peaceful coexistence” between Quantum Mechanics and Special Relativity [11], which renders Quantum Mechanics compatible with Special Relativity in spite of its non-local character. Peopleworkingoncollapsemodelswereofcourseawareofthisnon-locality constraintwhichany collapse model, whether relativistic or not, has to obey in order to be compatible with Quantum Mechanics; thus, for them, the argument of Conway and Kochen does not come as a surprise. But there is something more. In applying their theorem to GRW models, Conway and Kochen mistakenly assumethattheresponseof aparticle (i.e. theoutcome of ameasurement) may depend only on the jumps which occurred in the backward light-cone of the spacetime point where the measurement occurs, since they regard the jumps as information which must fulfill FIN. But this cannot possibly be the right picture, even at a relativistic level, if the GRW model is to account for the nonlocal features of entangled quantum systems, which have been elucidated by Bell’s work; we now discuss this issue in more detail. 2. MODELS OF SPONTANEOUS WAVEFUNCTION COLLAPSE We briefly review some of the main features of the GRW model of spontaneous wavefunction collapse, which are relevant for the present discussion. We will first present the non relativistic 6 GRW model [14] and then comment on possible relativistic generalizations. 2.1. The GRW Model The starting point of the GRW model is that the wavefunction alone is the complete math- ematical description of all physical phenomena, it represents the maximum knowledge one can have, in principle, about the state of a physical system, both microscopic and macroscopic. Since macroscopic objects are always well localized in space, while the Schro¨dinger equation allows for superpositions of different macroscopic states, one has to modify the standard quantum evolution in order to provide a consistent and unified description of micro- and macro-phenomena. This is done in the following way. Let us consider a system of N particles; let be the Hilbert space associated to it and H the H standard quantum Hamiltonian of the system. The model is defined by the following postulates: 1. At random times, each particle experiences a sudden jump of the form: L (x)ψ n t ψ , (6) t −→ L (x)ψ n t k k where ψ is the statevector of the whole system at time t, immediately prior to the jump process. t L (x) is a linear operator which is conventionally chosen equal to: n α 3 α L (x) = 4 exp (q x)2 , (7) n r π −2 n− (cid:16) (cid:17) h i where α is a new parameter of the model which sets the the width of the localization process, and q is the position operator associated to the n-th particle. The random variable x corresponds to n the place where the jump occurs. 2. Between two consecutive jumps, the statevector evolves according to the standard Schro¨dinger equation. 3. The probability density for a jump taking place at the position x is given by: 2 p (x) L (x)ψ ; (8) n n t ≡ k k the probability density for the different particles are independent. 4. Finally, it is assumed that the jumps are distributed in time according to a Poissonian process with frequency λ, which is the second new parameter of the model. 4 The standard numerical values for α and λ are : 16 1 10 2 λ 10 sec α 10 cm , (9) − − − − ∼ ∼ 4 Recently S.L. Adlerproposed a radically different numerical valuefor thecollapse rate λ;see ref. [15]. 7 which have been chosen in such a way to guarantee a very good agreement of GRW with standard quantum mechanics and, at the same time, to ensure an almost instantaneous localization of the wavefunctionofclassicalmacro-objects,thussuppressingtheunwantedsuperpositionsofdifferently located macro-states. The evolution being stochastic, any initial state ψ0 evolves in time into an ensemble of states ψ (ω) , whereω labels thepossibledifferentways thejumpsmightoccur. Thestatistical operator t { } ρ associated to such an ensemble satisfy the following Lindblad-type equation: t N d i 3 ρ = [H,ρ ] λ ρ d x L (x)ρ L (x) . (10) dt t −~ t − (cid:18) t−Z n t n (cid:19) nX=1 TheGRWmodelandsimilarmodelswhichhaveappearedintheliteraturehavebeenextensively studied (see [16, 17] for a review of the subject); in particular, the following three important properties have been proved: At the microscopic level, quantum systems behave almost exactly as predicted by standard • Quantum Mechanics, the differences between the predictions of the GRW model and of QuantumMechanics beingsotiny thattheycannotbedetected withpresent-day technology. At the macroscopic level, wavefunctions of macro-objects are almost always very well local- • ized in space, so well localized that their centers of mass behave, for all practical purposes, like point-particles moving according to Newton’s laws. In a measurement-like situation, e.g. of the von Neumann type, GRW reproduces—as a • consequence of the modified dynamics—both the Born probability rule and the postulate of wave-packet reduction. Accordingly, models of spontaneous wavefunction collapse provide a unified description of all phys- ical phenomena, at least at the non-relativistic level, and a consistent solution to the measurement problem of Quantum Mechanics. 2.2. Features of the GRW Model There are some important properties of the GRW model, which all non-relativistic collapse models share, and which are relevant for the subsequent discussion. Non-linearity and stochasticity. The jump process is non-linear, since the probability of a jump taking place at x depends on the square norm of the statevector after the hitting. It is 8 also intrinsically stochastic; the model assumes that Nature is fundamentally random; needless to say, such a property is important in order to recover quantum probabilities when measurement situations are taken into account, but also for other reasons which will be clear soon. Non-locality. Themodelismanifestlynon-local; letustakeasanexamplethefollowingentangled state of two particles a and b: 1 ψ(xa,xb) = √2 [ψ∆1(xa)ψ∆2(xb) + ψ∆3(xa)ψ∆4(xb)], (11) where ψ∆ (x) is a normalized wavefunction well localized within the region ∆1 of space, and 1 similarly for the other three terms in (11); ∆1, ∆2, ∆3 and ∆4 label four regions which are arbitrarily far away from each other. Let us suppose that an experimenter A decides to measure the position of particle a, while an experimenter B decides to measure the position of particle b. The full initial state of the two particles plus the two apparata is: 1 ψbefore(xa,xb;yA,yB) = √2 [ψ∆1(xa)ψ∆2(xb) + ψ∆3(xa)ψ∆4(xb)]⊗φReady(yA)φReady(yB), (12) whereφ(y )andφ(y )denotethe(localized) statesof, letussay, thepointersofthetwoapparata, A A which initially are both in a state which is “ready” for the measurement. Now, let us suppose that the position of particle a is measured slightly before that of particle b. The dynamics of the GRW model tells that—because of the spontaneous jumps whose effect is amplified when a macro-object like a measuring apparatus enters into play—the final state of the pointer of the apparatus will be perfectly localized in space and will correspond either to the outcome : ∆1 or to the outcome : ∆3, each occurring with a probability almost identical to A A that given by the Born probability rule; let us suppose that the first possibility occurs. Then, the GRW dynamics implies that the initial state (12), after the first measurement, practically reduces to: ψafter(xa,xb;yA,yB) = ψ∆1(xa)ψ∆2(xb)⊗φA:∆1(yA)φReady(yB), (13) We see that, because of the (local) jumps which occurred on the pointer of the measuring device used by A to measure the position of particle a, also particle b has been almost instantaneously localized in space, in this case within ∆2, no matter how distant ∆2 is from ∆1; in fact, as we see from state (13), a subsequent measurement of the position of particle b will give (almost) certainly the outcome : ∆2. In a similar way, if the outcome of the first measurement had been : ∆3, B A then particle b would have been immediately localized around ∆4, and this would be confirmed by any subsequent measurement of its position. 9 Accordingly, there is a perfect and non-local correlation between the region where particle a is located after a measurement and the region where particle b is located by that same measurement doneonitsfaraway partner. Note, however, thatthejumpsactingonthepointer(anddetermining in this way the outcome of a measurement) are the consequence of a perfectly local interaction between the pointer and the stochastic background which enters in the dynamical evolution; it is only the entanglement between the two particles which renders the overall effect fundamentally non-local. The crucial point to understand is that this non-local feature of the collapse mechanism is 5 not a consequence of the fact that the GRW model is non-relativistic ; on the contrary, such a peculiar feature is necessary in order to reproduce the quantum correlations for EPR states like (11), which have been confirmed by all experiments. In other words, after the work of Bell and the experiments of Aspect which have shown that Nature is fundamentally non-local, any GRW model (whether relativistic or not) has to embody such a non-local behavior in order to reproduce quantum correlations. No faster-than-light. One might wonder whether the non-local character of the jump process might be used to send faster-than-light signals, but in [12] it has been proven that this is not possible, and the physical reason is quite simple to understand: since jumps are intrinsically random, they cannot be controlled to implement faster-than-light communication, and as soon as one averages over all possible jumps, their non-local character vanishes. Accordingly, like standard Quantum Mechanics, also the GRW model shares the “peaceful coexistence” between relativity and non-locality, which is one of the lessons we had lo learn from Bell’s inequalities. Indeed, Bell himself has stated [19]: “... I am particularly struck by the fact that the [GRW] model is as Lorentz invariant as it could be in the nonrelativistic version. It takes away the ground of my fear that any exact formulation of quantum mechanics must conflict with fundamental Lorentz invariance”. TheformalaspectsofthisnicefeatureofthetheoryconsistsinthefactthatGRWviolates Bell’s locality condition by violating outcome independence, just as standard nonrelativistic quantum mechanics does. Before proceeding it is useful to recall that Bell’s locality assumption has been proved [20, 21] to be equivalent to the conjunction of the two logically independent conditions, parameter independence and outcome independence. To clarify the matter let us fix our notation. 5 Indeed, it would be very easy to devise a local jump process, even for entangled states, which in any case would lead to a conflict with quantummechanical predictions. 10 We will denote by λ all parameters (which may include the quantum mechanical statevector or even to reduce to it alone) that specify completely the state of an individual physical system. For simplicity we will refer to a standard EPR-Bohm setup and we will denote by pAB(x,y;n,m) the λ joint probability of getting the outcome x in a measurement at A and y in a measurement at B. Obviously we assume that the experimenters at A and B can make a free-will choice of the directions n and m along which they will perform their measurements. They can also choose not to perform the measurement. Bell’s locality assumption can be expressed as pAB(x,y;n,m) = pA(x;n, )pB(y; ,m), (14) λ λ λ ∗ ∗ where the symbol appearing on the r.h.s. denotes that the corresponding measurement is not ∗ performed. As already anticipated, the above condition has been proved to be equivalent to the conjunction of the two following conditions: pA(x;n,m) = pA(x;n, ); pB(y;n,m) = pB(y; ,m) (15) λ λ λ λ ∗ ∗ and pAB(x,y;n,m) = pA(x;n,m)pB(y;n,m), (16) λ λ λ where we have denoted, e.g., by the symbol pA(x;n,m) the probability of getting, for the given λ settings n,m, the outcome x at A. The first conditions express Parameter independence, i.e., the requirement that the probability of getting an outcome at A(B) is independent of the setting chosen at B(A), while the last conditions (Outcome independence) expresses the requirement that the probability of an outcome at one wing does not depend on the outcome obtained at the other wing. We are now in the conditions of discussing briefly this point with reference to the twined state of total spin 0 of Conway and Kochen: 1 |φsingleti = √3 [|Sa:n = +1i|Sb:n = −1i+|Sa:n = −1i|Sb:n = +1i−|Sa:n = 0i|Sa:n = 0i], (17) the states Sa:n = +1 , Sa:n = 0 , Sa:n = 1 and the similar ones for particle b being the eigen- | i | i | − i states belonging to the indicated eigenvalues of the spin component along an arbitrary direction n. Let us now consider particle b. If particle a is not subjected to any measurement, the probabilities 2 6 of the two outcomes for S are : b:n 2 1 2 2 P(S = 1) = , P(S = 0) = . (18) b:n 3 b:n 3 6 Asalready remarked theGRW model gives practically thesame predictionsand hasthesame effectsas standard quantummechanics with thewave packet reduction postulate.