aa r X i v : . [ m a t h - ph ] N ov Indeterminism and Undecidability
Klaas Landsman
Department of Mathematics, Institute for Mathematics, Astrophysics, andParticle Physics (IMAPP), Radboud University, Nijmegen, The NetherlandsEmail: [email protected]
Dedicated to the memory of Michael Redhead (1929–2020)
Abstract
The aim of this paper is to argue that the (alleged) indeterminism of quantum me-chanics, claimed by adherents of the Copenhagen interpretation since Born (1926),can be proved from Chaitin’s follow-up to G¨odel’s (first) incompleteness theorem. Incomparison, Bell’s (1964) theorem as well as the so-called free will theorem–originallydue to Heywood and Redhead (1983)–left two loopholes for deterministic hidden vari-able theories, namely giving up either locality (more precisely: local contextuality, asin Bohmian mechanics) or free choice (i.e. uncorrelated measurement settings, as in ’tHooft’s cellular automaton interpretation of quantum mechanics). The main point isthat Bell and others did not exploit the full empirical content of quantum mechanics,which consists of long series of outcomes of repeated measurements (idealized as in-finite binary sequences): their arguments only used the long-run relative frequenciesderived from such series, and hence merely asked hidden variable theories to repro-duce single-case Born probabilities defined by certain entangled bipartite states. Ifwe idealize binary outcome strings of a fair quantum coin flip as infinite sequences,quantum mechanics predicts that these typically (i.e. almost surely) have a propertycalled in logic, which is much stronger than uncomputability. This isthe key to my claim, which is admittedly based on a stronger (yet compelling) notionof determinism than what is common in the literature on hidden variable theories.
Contents
Introduction: G¨odel and Bell
While prima facie totally unrelated, G¨odel’s theorem (1931) in mathematical logic andBell’s theorem (1964) in physics share a number of fairly unusual features (for theorems ): • Despite their very considerable technical and conceptual difficulty, both results areextremely famous and have caught the popular imagination like few others in science. • Though welcome in principle–in their teens, many people including the author wereintrigued by books with titles like
G¨odel, Escher Bach: An Eternal Golden Braid and
The Dancing Wu-Li Masters: An Overview of the New Physics , both of whichappeared in 1979–this imagination has fostered wild claims to the effect that G¨odelproved that the mind cannot be a computer or even that God exists, whilst Bellallegedly showed that reality does not exist. Both theorems (apparently throughrather different means) supposedly also supported the validity of Zen Buddhism. • However, even among professional mathematicians (logicians excepted) few wouldbe able to correctly state the content of G¨odel’s theorem when asked on the spot,let alone provide a correct proof, and similarly for Bell’s theorem among physicists. • Nonetheless, many professionals will be aware of the general feeling that G¨odel insome sense shattered the great mathematician Hilbert’s dream of what the foun-dations of mathematics should look like, whilst there is similar consensus that Belldealt a lethal blow to Einstein’s physical world view–though ironically, G¨odel workedin the spirit and formalism of Hilbert’s proof theory, much as Bell largely agreed withEinstein’s views about quantum mechanics and about physics in general. • Both experts and amateurs seem to agree that G¨odel’s theorem and Bell’s theorempenetrate the very core of the respective disciplines of mathematics and physics.In this light, anyone interested in both of these disciplines will want to know what theseresults have to do with each other, especially since mathematics underwrites physics (orat least is its language). At first sight this connection looks remote. Roughly speaking:
1. G¨odel proved that any consistent mathematical theory (formalized as an axiomatic-deductive system in which proofs could in principle be carried out mechanicallyby a computer) that contains enough arithmetic is incomplete (in that arithmeticsentences ϕ exist for which neither ϕ nor its negation can be proved).2. Bell showed that if a deterministic “hidden variable” theory underneath (and com-patible with) quantum mechanics exists, then this theory cannot be local (in thesense that the hidden state, if known, could be used for superluminal signaling). In fact there are two incompleteness theorems in logic due to G¨odel (see footnotes 2 and 5) and two theorems on quantum mechanics due to Bell (Brown & Timpson, 2014; Wiseman, 2014), but for reasons tofollow in this essay I am mainly interested in the first ones, of both authors, except for a few side remarks. See Franz´en (2005) for an excellent first introduction to G¨odel’s theorems, combined with a fair anddetailed critique of its abuses, including overstatements by both amateurs and experts (a similar guide tothe use and abuse of Bell’s theorems remains to be written), and Smith (2013) for a possible second go. Yanofsky (2013) nicely discusses both theorems in the context of the limits of science and reason. Both reformulations are a bit anachronistic and purpose-made. See G¨odel (1931) and Bell (1964)!
PrincipiaMathematica of Russell and Whitehead and the axioms for set theory proposed earlierby Zermelo, Fraenkel, and von Neumann. Though relegated to a footnote, the shadowof
Hilbert’s program , aimed to prove the consistency of mathematics (ultimately based onCantor’s set theory) using absolutely reliable, “finitist” means, clearly loomed large, too. Bell, on the other hand, tried to understand if the de Broglie–Bohm pilot wave theory ,which was meant to be a deterministic theory of particle motion reproducing all predictionsof quantum mechanics, necessarily had to be non-local: Bell’s answer, then, was “yes.” In turn, the circumstances in which G¨odel and Bell operated had a long pedigree inthe quest for certainty in mathematics and for determinism in physics , respectively. Theformer had even been challenged at least three times: first, by the transition from Euclid’smathematics to Newton’s; second, by the set-theoretic paradoxes discovered around 1900by Russell and others (which ultimately resulted from attempts to make Newton’s calculusrigorous by grounding it in analysis, and in turn founding analysis in the real numbers andhence in set theory), and third, by Brouwer’s challenge to “classical” mathematics, whichhe tried to replace by “intuitionistic” mathematics (both Hilbert and G¨odel were influencedby Brouwer, though contrecoeur : neither shared his overall philosophy of mathematics).In physics (and more generally), what Hacking (1990, Chapter 2) calls the doctrine ofnecessity , which thus far–barring a few exceptions–had pervaded European thought, beganto erode in the 19th century, culminating in the invention of quantum mechanics between1900–1930 and notably in its probability interpretation as expressed by Born (1926): Thus Schr¨odinger’s quantum mechanics gives a very definite answer to the question ofthe outcome of a collision; however, this does not involve any causal relationship. Oneobtains no answer to the question “what is the state after the collision,” but only tothe question “how probable is a specific outcome of the collision”. (. . . ) This raises theentire problem of determinism. From the standpoint of our quantum mechanics, thereis no quantity that could causally establish the outcome of a collision in each individualcase; however, so far we are not aware of any experimental clue to the effect that thereare internal properties of atoms that enforce some particular outcome. Should wehope to discover such properties that determine individual outcomes later (perhapsphases of the internal atomic motions)? (. . . ) I myself tend to relinquish determinismin the atomic world. (Born, 1926, p. 866, translation by the present author) In a letter to Born dated December 4, 1926, Einstein’s famously replied that ‘God doesnot play dice’ (‘Jedenfalls bin ich ¨uberzeugt, daß der nicht w¨urfelt’). Within ten yearsEinstein saw a link with locality, and Bell (1964) and later papers followed up on this. G¨odel’s second incompleteness theorem shows that one example of ϕ is the (coded) statement that theconsistency of the theory can be proved within the theory. This is often taken to refute Hilbert’s program,but even among experts it seems controversial if it really does so. For Hilbert’s program and its role inG¨odel’s theorems see e.g. Zach (2001), Tait (2005), Sieg (2013), and Tapp (2013). Greenstein (2019) is a popular book on the history and interpretation of Bell’s work. Scholarly analysesinclude Redhead (1989), Butterfield (1992), Werner & Wolf (2001), and the papers cited in footnote 1. Some vocal researchers calim that Bell and Einstein were primarily interested in locality and realism,determinism being a secondary (or no) issue, but the historical record is ambiguous; more generally, over10,000 papers about Bell’s theorems show that Bell can be interpreted in almost equally many ways. Butthis controversy is a moot point: whatever his own (or Einstein’s) intentions, Bell’s (1964) theorem putsconstraints on possible deterministic underpinnings of quantum mechanics, and that is how I take it. For an overall survey of this theme see Kline (1980). This phase in the history of quantum mechanics is described by Mehra & Rechenberg (2000). Randomness and its unprovability
This precise history has a major impact on my argument, since it shows that right from thebeginning the kind of randomness that Born (probably preceded by Pauli and followedby Bohr, Heisenberg, Jordan, Dirac, von Neumann, and most of the other pioneers ofquantum mechanics except Einstein, de Broglie, and Schr¨odinger) argued for as beingproduced by quantum mechanics, was antipodal to determinism . Thus randomness inquantum mechanics was identified with indeterminism , and hence attempts (like the deBroglie–Bohm pilot wave theory) to undermine the “Copenhagen” claim of randomnesslooked for deterministic (and arguably realistic ) theories underneath quantum mechanics.Although “undecidability” may sound a bit like “indeterminism”, the analogy betweenthe quests for certainty in mathematics and for determinism in physics (and their allegedundermining by G¨odel’s and Bell’s theorems, respectively) may sound rather superficial.To find common ground more effort is needed to bringing these theorems together. First, some of its “romantic” aspects have to be removed from G¨odel’s theorem, no-tably its reliance on self-reference, although admittedly this was the key to both G¨odel’soriginal example of an undecidable sentence ϕ (which in a cryptic way expresses its ownunprovability) and his proof, in which an axiomatic theory that includes arithmetic isarithmetized through a numerical encoding scheme so as to be able to “talk about itself”.Though later proofs of G¨odel’s theorem also use numerical encodings of mathematicalexpressions (such as symbols, sentences, proofs, and computer programs), this is donein order to make recursion theory (initially a theory of functions f : N → N ) availableto a wider context, rather than to exploit self-reference. Each computably enumerablebut uncomputable subset E ⊂ N leads to undecidable statements (very rarely in main-stream mathematics), namely those for which the sentence n / ∈ E is true but unprovable. Chaitin’s (first) incompleteness theorem (Theorem B.1 in Appendix B), which will playan important role in my reasoning, is an example of this. To understand this theorem andits background we return to the history of 20th century mathematics and physics.Hilbert influenced this history in many ways, of which the sixth problem on hisfamous list of 23 mathematical problems from 1900 is particularly relevant here: thisproblem concerns the ‘ Mathematical Treatment of the Axioms of Physics, especially thetheory of probabilities and mechanics ’ (Hilbert, 1902). This problem influenced our topicin two initially independent ways, which now come together. First, the problem inspiredvon Neumann (1932) to develop his mathematical axiomatization of quantum mechanics,which still forms the basis of all mathematically rigorous work on this theory. In particular,he initiated the literature on hidden variable theories (see § See Landsman (2020) for the view that randomness is a family resemblance (in that it lacks a meaningcommon to all its applications) with the special feature that its various uses are always defined antipodally. Also cf. Breuer (2001), Calude (2004), Svozil, Calude & Stay (2004), , Szangolies (2018). A subset E ⊂ N is computably enumerable (c.e.) if it is the image of a computable function f : N → N ,and computable if its characteristic function 1 E is computable, which is true iff both E and N \ E are c.e. This is true for physics almost as much as it is (more famously) for mathematics, since Hilbert playeda major role in the mathematization of the two great theories of twentieth century physics, i.e. generalrelativity (Corry, 2004; Renn, 2007) and quantum mechanics (R´edei & St¨oltzner, 2001; Landsman, 2021). σ = 001101010111010010100011010111 (2.1)is as probable as a “deterministic” string like σ = 111111111111111111111111111111 . (2.2)In other words, their probabilities say little or nothing about the “randomness” of indi-vidual outcomes. Imposing statistical properties helps but is not enough to guaranteerandomness. It is slightly easier to explain this in base 10, to which I therefore switch fora moment. If we call a sequence x Borel normal if each possible string σ in x has relativefrequency 10 −| σ | , where | σ | is the length of σ (so that each digit 0 , . . . , Champernowne’s number . . . can be shown to be Borel normal. The decimal expansion of π is conjectured to be Borelnormal, too (and has been empirically verified to be so in billions of decimals), but thesenumbers are hardly random: they are computable, which is one version of “deterministic”.Any sound definition of randomness (for binary strings or sequences) has to navigatebetween Scylla and Charybdis: if the definition is too weak (such as Borel normality),counterexamples will undermine it (such as Champernowne’s number), but if it is toostrong (such as being lawless, like Brouwer’s choice sequences), it will not hold almostsurely in a 50-50 Bernoulli process (Moschovakis, 2016). As an example of such soundnavigation, Solomonoff, Kolmogorov, Martin-L¨of, Chaitin, Levin, Solovay, Schnorr, andothers developed the algorithmic theory of randomness (Li & Vit´anyi, 2008). The basicidea is that a string or sequence is random iff its shortest description is the sequence itself,but the notion of a description has to made precise to avoid Berry’s paradox : The Berry number is the smallest positive integer that cannot be described in lessthan eighteen words.
The paradox, then, is that on the one hand this number must exist, since only finitely manyintegers can be described in less than eighteen words and hence the set of such numbersmust have a lower bound, while on the other hand Berry’s number cannot exists by itsown definition. In the case at hand, the notion of a description is sharpened by askingit to be computable , so that, roughly speaking (see appendix B for technical details),we call a (finite) binary string σ (Kolomogorov) random if the length of the shortestcomputer program generating σ is at least as long as σ itself, and call an (infinite) binarysequence x (Levin–Chaitin) random or if its (sufficiently long) finite truncationsare Kolomogorov random. At last, for finite strings σ Chaitin’s (first) incompletenesstheorem states that although countably many strings σ are random, this can be proved only for finitely many of these, whereas for infinite sequences x his (second) incompletenesstheorem says that if such a sequence is random, only finitely many of its digits can becomputed (see Theorems B.1 and B.4 for precise statements). Thus randomness is elusive . This is one of innumerable paradoxes of natural language, which leads to an incompleteness theoremonce the notion of a description has been appropriately formalized in mathematics, much as G¨odel’s firstincompleteness theorem turns the the liar’s paradox into a theorem. Rethinking Bell’s theorem
In order to locate Bell’s (1964) theorem in the literature on quantum mechanics and(in)determinism, I recall that Hilbert’s sixth problem inspired both the work of von Misesand Kolmogorov that eventually gave rise to the algorithmic theory of randomness, and (Hilbert’s postdoc) von Neumann’s work on the mathematical foundations of quantummechanics, culminating in his book (von Neumann, 1932). One of his results was thatthere can be no nonzero function λ : H n ( C ) → R (where H n ( C ) is the space of hermitian n × n matrices, seen as the observables of a quantum-mechanical n -level system) that is:1. dispersion-free (i.e. λ ( a ) = λ ( a ) for each a ∈ H n ( C ));2. linear (i.e. λ ( sa + tb ) = sλ ( a ) + tλ ( b ) for all s, t ∈ R and a, b ∈ H n ( C )).Unfortunately, von Neumann interpreted this correct, non-circular, and interesting resultas a proof that quantum mechanics is complete in the sense that there can be no hiddenvariables in the sense of Born (1926), i.e. ‘properties that determine individual outcomes’.The reason this does not follow is twofold. First, the proof relies on a tacit assump-tion that later came to be called non-contextuality , namely that the value λ ( a ) of someobservable a only depends on a , whereas measurement ideology `a la Bohr (1935) suggeststhat it may depend on a measurement context , formalized as a further set of observablescommuting with a (unless a is maximal such a set is far from unique). Second, thoughnatural, the linearity assumption is very strong and excludes even eigenvalues of a .This second point was remedied by the Kochen–Specker theorem , who weakened vonNeumann’s linearity assumption to linearity on commuting observables , which at least in-corporates eigenvalues and is even found so appealing that the Kochen–Specker is generallytaken to exclude non-contextual hidden variable theories. See also Appendix C.The final step in the series of attempts, initiated by von Neumann, to exclude hiddenvariables by showing that subject to reasonable assumptions the corresponding value attri-butions cannot exist even independently of any statistical considerations, is the so-called free will theorem . In the wake of the renowned “ epr ” paper (Einstein, Podolsky andRosen, 1935) the setting has now become bipartite (i.e. Alice and Bob who are spacelikeseparated each perform experiments on a correlated state) and the non-contextuality as-sumption is weakened to local contextuality : the outcomes of Alice’s measurements areindependent of any choice of measurements Bob might perform, and vice versa . Thusher value attributions λ ( a | context) may well be contextual, as long as the observablescommuting with the one she measures (i.e. a ), which form a context to a , are local to her. A second line of research, which goes back at least to de Broglie (1928), was influ-entially taken up by Bohm (1952), and most recently includes ’t Hooft (2016), assumesthe possibility of non-contextual value attributions and tries to make these compatiblewith the Born rule of quantum mechanics. Bell (1964) was primarily concerned with suchtheories, asking himself if a deterministic theory like Bohm’s was necessarily non-local. See also Bub (2011), Dieks (2016), and forthcoming work by Chris Mitsch for balanced accounts. The idea of contextuality was first formulated by Grete Hermann (Crull & Bacciagaluppi, 2016). See Kochen & Specker (1967). Ironically, his followers attribute this theorem to Bell (1966), althoughthe result is just a technical sharpening of von Neumann’s result they so vehemently ridicule. For a deepphilosophical analysis of the Kochen–Specker theorem, as well as of Bell’s theorems, see Redhead (1989). See appendix C. This theorem is originally due to Heywood & Redhead (1983), with follow-ups by Stairs(1983), Brown & Svetlichny (1990), and Clifton (1993), but it was named and made famous by Conway &Kochen (2009), whose main contribution was an emphasis on free will (Landsman, 2017, Chapter 6). Since Alice and Bob are spacelike separated their observables commute (Einstein locality).
6n Bell’s analysis, which takes place in the bipartite ( epr ) setting, the quantum-mechanicalprobabilities are obtained by formally averaging over the set of hidden variables, i.e., P ψ ( F = x, G = y | A = a, B = b ) = Z Λ dµ ψ ( λ ) P λ ( F = x, G = y | A = a, B = b ) . (3.1)Here ψ is some (explicitly identified) quantum state of a correlated pair of (typically)2-level quantum systems (which may be either optical, where the degree of freedom ishelicity, or massive, where the degree of freedom is spin), F is an observable measured byAlice defined by her choice of setting a , likewise G for Bob defined by his setting b , withpossible outcomes x ∈ { , } , likewise y ∈ x, y ) if the correlated system has been prepared in the state ψ ; the expression P λ ( · · · ) on the right-hand side is the probability of the outcome ( x, y ) ifthe unknown hidden variable or state equals λ , and finally, µ ψ is some probability measureon the space Λ of hidden states supposedly provided by the theory for each state ψ .We now say that the hidden variable theory supplying the above quantities is: • deterministic if the probabilities P λ ( F = x, G = y | A = a, B = b ) equal 0 or 1; • locally contextual if the expression P λ ( F = x | A = a, B = b ) = X y =0 , P λ ( F = x, G = y | A = a, B = b ); (3.2)is independent of b , whilst the corresponding expression P λ ( G = y | A = a, B = b ) = X x =0 , P λ ( F = x, G = y | A = a, B = b ) , (3.3)is independent of a . That is, the probabilities of Alice’s outcomes are independentof Bob’s settings, and vice versa . This locality property seems very reasonable andin fact it follows from special relativity, for if Bob chooses his settings just before hismeasurement, there is a frame of reference in which Alice measures before Bob haschosen his settings, and vice versa . In turn, this is equivalent to the property thateven if she knew the value of λ , Alice could not signal to Bob, and vice versa . Bell proved that a hidden variable theory cannot satisfy (3.1) and be both deterministicand locally contextual (which explained why Bohm’s theory had to be non-local). Makinghis tacit assumption that experimental settings can be “freely” chosen explicit , we obtain: Theorem 3.1
The conjunction of the following properties is inconsistent:1. determinism ;2. quantum mechanics , i.e. the Born rule for P ψ ( F = x, G = y | A = a, B = b ) ;3. local contextuality ;4. free choice , i.e. (statistical) independence of the measurement settings a and b fromeach other and from the hidden variable λ (given the probability measure µ ψ ). In quantum mechanics the left-hand side of (3.1) satisfies this locality condition for any state ψ . See Landsman (2017), § Are deterministic hidden variable theories deterministic?
Although the assumptions have a slightly different meaning, the free will theorem leadsto the same result as Bell’s theorem (see Appendix C), so that the (no) hidden variabletradition initiated by von Neumann, which culminates in the former, coalesces with the(positive) hidden variable tradition going back to de Broglie, shown its place by the latter.Thusly there are the obvious four (minimal) ways out of the contradiction in Theorem 3.1: • Copenhagen (i.e. mainstream) quantum mechanics rejects determinism; • Valentini (2019) rejects the Born rule and hence qm (see the end of § • Bohmians reject local contextuality; • ’t Hooft (2016) rejects free choice.We focus on the last two options, so that determinism and quantum mechanics (i.e. theBorn rule) are kept. In both cases the Born rule is recovered by averaging the hiddenvariable with respect to a probability measure µ ψ on the space of hidden variables, givensome (pure) quantum state ψ . The difference is that in Bohmian mechanics the totalstate (which consists of the hidden configuration plus the “pilot wave” ψ ) determines themeasurement outcomes given the settings , whereas in ’t Hooft’s theory the hidden variabledetermines the outcomes as well as the settings. More specifically: • In Bohmian mechanics the hidden variable is position q , and dµ ψ ( q ) = | ψ ( q ) | dq isthe Born probability for outcome q with respect to the expansion | ψ i = R dq ψ ( q ) | q i . • In ’t Hooft’s theory the hidden variable is a basis vector | m i in some separableHilbert space H ( m ∈ N ), and once again the measure µ ψ ( m ) = | c m | is given bythe Born probability for outcome m with respect to the expansion | ψ i = P m c m | m i .Thus the hidden variables (i.e. q ∈ Q and m ∈ N , respectively) have familiar quantum-mechanical interpretations and also their compatibility measures are precisely the Bornmeasures for the quantum state ψ . In this light, we may ask to what extent these hiddenvariable theories are truly deterministic, as their adherents claim them to be. Since theargument does not rely on entanglement and hence on a bipartite experiment, we may aswell work with a quantum coin toss. The settings of the experiments are then fixed, so thatwe may treat Bohmian mechanics and ’t Hooft’s theory on the same footing. Idealizingto an infinite run, one has an outcome sequence x : N →
2. Standard (Copenhagen)quantum mechanics refuses to say anything about its origin, but nonetheless it does makevery specific predictions about x . The basis of these predictions is the following theorem,whose notation and proof are explained in Appendix A. One may think of a fair quantumcoin, in which σ ( a ) = 2 = { , } and µ a (0) = µ a (1) = 1 /
2, and which probabilistically isindistinguishable from a fair classical coin (which in my view cannot exist, cf. § There is a subtle difference between Bohmian mechanics as reviewed by e.g. Goldstein (2017), and deBroglie’s original pilot wave theory (Valentini, 2019). This difference is immaterial for my discussion. In Bohmian mechanics, the hidden state q ∈ Q just pertains to the particles undergoing measurement,whilst the settings a are supposed to be “freely chosen” for each measurement (and in particular areindependent of q ). The outcome is then fixed by a and q . In ’t Hooft’s theory, the hidden state x ∈ X of “the world” determines the settings as well as the outcomes. Beyond the issue raised in the main text,Bohmians (but not ’t Hooft!) therefore have an additional problem, namely the origin of the settings(which are simply left out of the theory). This weakens their case for determinism even further. heorem 4.1 The following procedures for repeated identical independent measurementsare equivalent (in giving the same possible outcome sequences with the same probabilities):1. Quantum mechanics is applied to the whole run, described as a single quantum-mechanical experiment with a single classically recorded outcome sequence;2. Quantum mechanics is applied to single experiments (with classically recorded out-comes), upon which classical probability theory takes over to combine these.Either way, the (purely theoretical) Born probability µ a for single outcomes induces theinfinite Bernoulli process probability µ ∞ a on the space σ ( a ) N of infinite outcome sequences. Theorem B.3 in Appendix B then implies:
Corollary 4.2
With respect to the “fair” probability measure P ∞ on N almost everyoutcome sequence x of an infinitely often repeated fair quantum coin flip is 1-random. In hidden variable theories, on the other hand, x factors through Λ, that is, there arefunctions h : N → Λ and g : Λ → x = g ◦ h . Hidden variable theoriesdo provide g , i.e. describe the outcome of any experiment given the value of the hiddenvariable λ ∈ Λ. However, what about h , that is, the specification of the value of the hiddenvariable λ in each run of the experiment? There are just the following two scenarios:1. The function h is provided by the hidden variable theory. In that case, since thetheory is supposed to be deterministic, h explicitly gives the values λ n = h ( n ) foreach n ∈ N (i.e. experiment no. n in the run). Since g is also given, this meansthat x is given by the theory. By Theorem B.4 (i.e. Chaitin’s second incompletenesstheorem), the outcome sequence cannot be 1-random, against Corollary 4.2.2. The function h is not provided by the hidden variable theory. In that case, the theoryfails to determine the outcome of any specific experiment and just provides averagesof outcomes. My conclusion would be that, except for some kind of a “story”, nothinghas been gained over quantum mechanics, but hidden variable theorists argue thattheir theories cannot be expected to provide initial conditions (for experiments), andclaim that the randomness in measurement outcomes originates in the randomnessof the initial conditions of the experiment. But then the question arises what elseprovides these conditions, and hence our function h . The point here is that in order torecover the predictions of quantum mechanics as meant in Corollary 4.2, the function h must sample the Born measure (in its guise of the compatibility measure µ ψ on Λ),in the sense of “randomly” picking elements from Λ, distributed according to µ ψ , cf.(A.2). This, in turn, should guarantee that the sequences x = g ◦ h mimic fair coinflips. Since g is supposed to be given, this implies that the randomness properties of x must entirely originate in h . This origin cannot be deterministic, since in that casewe are back to the contradictory scenario 1 above. Hence h must come from someunknown external random process in nature that our hidden variable theories invokeas a kind of an oracle. In my view the need for such a random oracle underminestheir purpose and makes them self-defeating. Every way you look at this you lose! The Bohmians are divided on the origin of their compatibility measure, referred to in this context asthe quantum equilibrium distribution , cf. D¨urr, Goldstein, & Zanghi (1992) against Valentini (2019). Theorigin of µ ψ is not my concern, which is the need to randomly sample it and the justification for doing so. Conclusion and discussion
We may summarize the discussion in the previous section as follows: Theorem 5.1
For any hidden variable theory T the following properties are incompatible:1. Determinism: T states the outcome of the measurement of any observable a giventhe value λ ∈ Λ of the hidden variable via a function g : Λ → σ ( a ) and providesthese values for each experiment; for an infinite run this is done via some function h : N → Λ , so that T provides the outcome sequence x : N → σ ( a ) through x = g ◦ h .2. Born rule:
Outcome sequences are almost surely 1-random. (cf. Corollary 4.2).
The proof is short. According to the first clause T states the entire outcome sequence x .By Chaitin’s incompleteness theorem B.4 this is incompatible with the second clause. (cid:3) In order to understand Theorem 5.1 and its proof it may be helpful to note that in classicalcoin tossing the role of the hidden state is also played by the initial conditions (cf. Diaconis& Skyrms, 2018 Chapter 1, Appendix 2). The 50-50 chances (allegedly) making the coinfair are obtained by averaging over the initial conditions, i.e., by sampling. By the samearguments, this sampling cannot be deterministic, i.e. given by a function like h , forotherwise the outcome sequences appropriate to a fair coin would not obtain: it mustbe done in a genuinely random way and hence by appeal to an external random process.This is impossible classically, so that–unless they have a quantum-mechanical seed– fairclassical coins do not exist , as confirmed by Diaconis & Skyrms (2018, Chapter 1).I conclude that deterministic hidden variable theories compatible with quantum me-chanics do not exist. The reason that Bell’s (1964) theorem and the free will theoremleave two loopholes for determinism (i.e. local contextuality and no free choice) is thattheir compatibility condition with quantum mechanics is stated too weakly: the theoryis only required to reproduce certain single-case (Born) probabilities, as opposed to theproperties of typical outcome sequences (from which the said probabilities are extractedas long-run frequencies). This reason this approach is still partly successful lies in theclever use of entangled states. If one rejects the second requirement on determinism inTheorem 5.1, Bell’s theorem and the free will theorem still provide useful constraints ondeterministic hidden variable theories, but as shown in the previous section such a rejectionnecessitates an appeal to an unknown random process and hence seems self-defeating.Let us now consider the role of the idealization to infinite outcome sequences and seewhat happens if the experimental runs are finite. Once again, via Theorem 4.1 the Bornrule predicts that outcome strings will be Kolmogorov random with high probability. Anydeterministic theory (in the sense of Theorem 5.1) provides an explicit description (say inZFC) of the outcomes, whose randomness would be provable from this description. Butthis is precluded by Chaitin’s first incompleteness theorem (i.e. Theorem B.4), now in therole played by his second incompleteness theorem in the infinite case. (cid:3) In stating the second condition I have taken σ ( a ) = { , } with 50-50 Born probabilities, but this canbe generalized to other spectra and probability measures. See Downey & Hirschfeldt (2010), § In other words, we examine whether Earman’s principle is satisfied, cf. footnote 37. To make this argument completely rigorous one would need to define what a “description” providedby a deterministic theory means logically. There is a logical characterization of deterministic theories(Montague, 1974), and there are some arguments to the effect that the evolution laws in deterministictheories should be computable, cf. Earman (1986), Chapter 11, and Pour-El & Richards (2016), passim ,but this literature makes no direct reference to output strings or sequences of the kind we analyze and in epr –Bohm experiment local deterministichidden variable theories predict correlations that satisfy the Bell inequalities, whereason suitable settings quantum mechanics predicts (and experiment shows) that typical out-come sequences violate these inequalities. Now a disproof of some deterministic hiddenvariable theory T cannot perhaps be expected to show that all quantum-mechanical out-come sequences violate the predictions of the hidden variable theory (indeed they do not,albeit with low probability), but it should identify at least a sufficiently large numberof typical (i.e. random) sequences. However, even in the finite case this identification isimpossible by Theorem B.1, so that the false predictions of T cannot really be confrontedwith the correct predictions of quantum mechanics. Thus the unprovability of their false-hood condemns deterministic hidden variable theories, and perhaps even determinism asa whole, to a zombie-like existence in a twilight zone comparable with the Dutch situationaround selling soft drugs: although this is forbidden by law, it is (officially) not prosecuted.The situation would change drastically if deterministic hidden variable theories gaveup their compatibility with the Born rule (on which my entire reasoning is based), as forexample Valentini (2019) has argued in case of the de Broglie–Bohm pilot wave theory. Forit is this compatibility requirement that kills such theories, which could leave zombie-domif only they were brave enough to challenge the Born rule. This might open the door tosuperluminal signaling and worse, but on the other hand the possibility of violating theBorn rule would also provide a new context for deriving it, e.g. as a dynamical equilibriumcondition (as may be the case for the Broglie–Bohm theory, if Valentini is right).I would personally expect that the Born rule is emergent from some lower-level theory,which equally well suggests that it is valid in some limit only, rather than absolutely. The author is grateful to Jacob Barandes, Jeremy Butterfield, Cristian Calude, Erik Curiel, John Earman,Bas Terwijn, and Noson Yanofsky, as well as to members of seminar audiences and especially readers ofthe first version of this essay on the FQXi website for very helpful comments and corrections. He is evenmore grateful to the late Michael Redhead, for his exemplary approach to the foundations of physics.any case the identification of “deterministic” with “computable” is obscure even in situations where thelatter concept is well defined. For example, if we stipulate that h : N → Λ is computable (and likewise g : Λ →
2) then the above appeal to Chaitin’s first incompleteness theorem is not even necessary, butthis seems too easy. A somewhat circular solution, proposed by Scriven (1957), is to simply say that T is deterministic iff the output strings or sequences it describes are not random, but this begs for a moreexplicit characterization. One might naively expect such a characterization to come from the arithmeticalhierarchy (found in any book on computability): if, as before, we identify 2 N with the power set P ( N ) of N ,then S ⊂ N is called arithmetical if there is a formula ψ ( x ) in PA (Peano Arithmetic) such that n ∈ S iff N (cid:15) ψ ( n ), that is, ψ ( n ) is true in the usual sense. We may then classify the arithmetical subsets throughthe logical form of ψ , assumed in prenex normal form (i.e., all quantifiers have been moved to the left): S isin Σ = Π iff ψ has no quantifiers or only bounded quantifiers (in which case S is computable), and thenrecursively S ∈ Σ n +1 iff ψ ( x ) = ∃ y ϕ ( x, y ) with ϕ ∈ Π n , and ϕ ∈ Π n +1 iff ψ ( x ) = ∀ y ϕ ( x, y ) with ϕ ∈ Σ n .Here any singly quantified expression ∃ y ϕ ( x, y ) may be replaced by ∃ y · · · ∃ y k ϕ ( x, y , . . . , y k ) and likewisefor ∀ y . By convention Σ n ⊂ Σ n +1 and Π n ⊂ Π n +1 , and ∆ n := Σ n ∩ Π n . Since in classical logic ∀ y ϕ ( x, y )is equivalent to ¬∃ y ¬ ϕ ( x, y ), it follows that Π n sets are the complements of Σ n sets. One would then liketo locate deterministic theories somewhere in this hierarchy, preferably above the computable ∆ . Theidea of a hidden variable (namely y ) suggests Σ and closure under complementation (it would be crazy ifsome deterministic theory prefers ones over zeros) then leads to ∆ , but this equals ∆ . The next level ∆ is impossible since this already contains 1-random sets like Chaitin’s Ω. Hence more research is needed. For Bell’s proof it is irrelevant whether or not some hidden variable is able to sample the compatibilitymeasure, since the Bell inequalities follow from pointwise bounds, cf. Landsman (2017), eq. (6.119). The Born rule
The Born measure is a probability measure µ a on the spectrum σ ( a ) of a (bounded) self-adjoint operator a on some Hilbert space H , defined as follows by any state ω on B ( H ): Theorem A.1
Let H be a Hilbert space, let a ∗ = a ∈ B ( H ) , and let ω be a state on B ( H ) . There exists a unique probability measure µ a on the spectrum σ ( a ) of a such that ω ( f ( a )) = Z σ ( a ) dµ a ( λ ) f ( λ ) , for all f ∈ C ( σ ( a )) . (A.1)The Born measure is a mathematical construction; what is its relationship to experiment?This relationship must be the source of the (alleged) randomness of quantum mechanics,for the Schr¨odinger equation is deterministic. We start by postulating, as usual, that µ a (∆) is the (single case) “probability” that measurement of the observable a in the state ω (which jointly give rise to the pertinent Born measure µ a ) gives a result λ ∈ ∆ ⊂ σ ( a ).Here I identify single-case “probabilities” with numbers (consistent with the probabilitycalculus) provided by theory , upon which long-run frequencies provide empirical evidence for the theory in question, but do not define probabilities. The Born measure is a casein point: these probabilities are theoretically given , but have to be empirically verified bylong runs of independent experiments. In other words, by the results reviewed below suchexperiments provide numbers whose role it is to test the Born rule as a hypothesis. Thisis justified by the following sampling theorem (strong law of large numbers): for any(measurable) subset ∆ ⊂ σ ( a ) and any sequence ( x n ) ∈ σ ( a ) N we have µ ∞ a -almost surely:lim N →∞ N (1 ∆ ( x ) + · · · + 1 ∆ ( x N )) = µ a (∆) . (A.2) Proof of Theorem 4.1 . Let a = a ∗ ∈ B ( H ), where H is a Hilbert space and B ( H ) is thealgebra of all bounded operators on H , and let σ ( a ) be the spectrum of a . For simplicity(and since this is enough for our applications, where H = C ) I assume dim( H ) < ∞ , sothat σ ( a ) simply consists of the eigenvalues λ i of a (which may be degenerate). Let usfirst consider a finite number N of identical measurements of a (a “run”). The first optionin the theorem corresponds to a simultaneous measurement of the commuting operators a = a ⊗ H ⊗ · · · ⊗ H ; (A.3) · · · a N = 1 H ⊗ · · · ⊗ H ⊗ a, (A.4)all defined on the N -fold tensor product H N ≡ H ⊗ N of H with itself. To put thisin a broader perspective, consider any set ( a , . . . , a N ) ≡ a of commuting operators on any Hilbert space K (of which (A.3) - (A.4) is obviously a special case with K = H N ).These operators have a joint spectrum σ ( a ), whose elements are the joint eigenvalues λ = ( λ , . . . , λ N ), defined by the property that there exists a nonzero joint eigenvector ψ ∈ K such that a i ψ = λ i ψ for all i = 1 , . . . , N ; clearly, σ ( a ) = { λ ∈ σ ( a ) × · · · × σ ( a n ) | e λ ≡ e (1) λ · · · e ( n ) λ n = 0 } ⊆ σ ( a ) × · · · × σ ( a N ) , (A.5) Here a state ω is a positive normalized linear functional on B ( H ), as in the C*-algebraic approach toquantum mechanics (Haag, 1992; Landsman, 2017). One may think of expectation values ω ( a ) = Tr ( ρa ),where ρ is a density operator on H , with the special case ω ( a ) = h ψ, aψ i , where ψ ∈ H is a unit vector. This can even be replaced by a single measurement, see Landsman (2017), Corollary A.20. e ( i ) λ i is the spectral projection of a i on the eigenspace for the eigenvalue λ i ∈ σ ( a i ).Von Neumann’s Born rule for the probability of finding λ ∈ σ ( a ) then simply reads p a ( λ ) = ω ( e λ ) , (A.6)where ω is the state on B ( K ) with respect to which the Born probability is defined. Ifdim( K ) < ∞ , as I assume, we always have ω ( a ) = Tr ( ρa ) for some density operator ρ ,and for a general Hilbert space K this is the case iff the state ω is normal on B ( K ). For(normal) pure states we have ρ = | ψ ih ψ | for some unit vector ψ ∈ K , in which case p a ( λ ) = h ψ, e λ ψ i . (A.7)The Born rule (A.6) is similar to the single-operator case (Landsman, 2017, § thecontinuous functional calculus gives a Gelfand isomorphism of commutative C*-algebras C ∗ ( a, K ) ∼ = C ( σ ( a )) , (A.8)under which the restriction of the state ω , originally defined on B ( K ), to its commutativeC*-subalgebra C ∗ ( a ) defines a probability measure µ a on the joint spectrum σ ( a ) via theRiesz isomorphism. This is the Born measure, whose probabilities are given by (A.6). Forthe case (A.3) - (A.4) we have equality in (A.5); since in that case σ ( a i ) = σ ( a ), we obtain σ ( a ) = σ ( a ) N , (A.9)and therefore, for all λ i ∈ σ ( a ) and states ω on B ( H N ), the Born rule (A.6) becomes p a ( λ , . . . , λ N ) = ω ( e λ ⊗ · · · ⊗ e λ N ) . (A.10)Now take a state ω on B ( H ). Reflecting the idea that ω is the state on B ( H N ) in which N independent measurements of a ∈ B ( H ) in the state ω are carried out, choose ω = ω N , (A.11)the state on B ( H N ) defined by linear extension of its action on elementary tensors: ω N ( b ⊗ · · · ⊗ b n ) = ω ( b ) · · · ω N ( b N ) . (A.12)It follows that ω N ( e λ ⊗ · · · ⊗ e λ N ) = ω ( e λ ) · · · ω ( e λ N ) = p a ( λ ) · · · p a ( λ N ) , (A.13)so that the joint probability of the outcome ( λ , . . . , λ N ) ∈ σ ( a ) is simply p ~a ( λ , . . . , λ N ) = p a ( λ ) · · · p a ( λ N ) . (A.14)Since these are precisely the probabilities for option 2 (i.e. the Bernoulli process), i.e., µ a = µ Na , (A.15)this proves the claim for N < ∞ . To describe the limit N → ∞ , let B be any C*-algebrawith unit 1 B ; below I take B = B ( H ), B = C ∗ ( a, H ), or B = C ( σ ( a )). We now take A N = B ⊗ N , (A.16) The uses of states themselves may be justified by Gleason’s theorem (Landsman, 2017, §§ The Born rule for commuting operators follows from the single operator case (Landsman, 2017, § N -fold tensor product of B with itself. The special cases above may be rewritten as B ( H ) ⊗ N ∼ = B ( H N ); (A.17) C ∗ ( a, H ) ⊗ N ∼ = C ∗ ( a , . . . , a N , H N ); (A.18) C ( σ ( a )) ⊗ N ∼ = C ( σ ( a ) × · · · × σ ( a )) , (A.19)with N copies of H and σ ( a ), respectively, and in (A.18) the a i are given by (A.3) -(A.4). We may then wonder if these algebras have a limit as N → ∞ . They do, but itis not unique and depends on the choice of observables, that is, of the infinite sequences a = ( a , a , . . . ), with a N ∈ A N , that are supposed to have a limit. One possibility is totake sequences a for which there exists M ∈ N and a M ∈ A M such that for each N ≥ M , a N = a M ⊗ B · · · ⊗ B , (A.20)with N − M copies of 1 B . On that choice, one obtains the infinite tensor product B ⊗∞ , seeLandsman (2017), § C.14. The limit of (A.17) in this sense is B ( H ⊗∞ ), where H ⊗∞ is vonNeumann’s ‘complete’ infinite tensor product of Hilbert spaces, in which C ∗ ( a, H ) ⊗∞ is the C*-algebra generated by ( a , a , . . . ) and the unit on H ⊗∞ . The limit of (A.19) is C ( σ ( a )) ⊗∞ ∼ = C ( σ ( a ) N ) , (A.21)where σ ( a ) N , which we previously saw as a measure space (as a special case of X N forgeneral compact Hausdorff spaces X ), is now seen as a topological space with the producttopology, in which it is compact. As in the finite case, we have an isomorphism C ∗ ( a, H ) ⊗∞ ∼ = C ( σ ( a )) ⊗∞ , (A.22)and hence, on the given identifications, we obtain an isomorphism of C*-algebras C ∗ ( a , a , . . . , H ⊗∞ ) ∼ = C ( σ ( a ) N ) . (A.23)It follows from the definition of the infinite tensor products used here that each state ω on B defines a state ω ∞ on B ⊗∞ . Take B = B ( H ) and restrict ω ∞ , which a priori is a stateon B ( H ⊗∞ ), to its commutative C*-subalgebra C ∗ ( a , a , . . . , H ⊗∞ ). The isomorphism(A.23) then gives a probability measure µ a on the compact space σ ( a ) N , where the label a now refers to the infinite set of commuting operators ( a , a , . . . ) on H ⊗∞ . To computethis measure, I use (A.1) and the fact that by construction functions of the type f ( λ , λ , . . . ) = f ( N ) ( λ , . . . , λ N ) , (A.24)where N < ∞ and f ( N ) ∈ C ( σ ( a ) N ), are dense in C ( σ ( a ) N ) (with respect to the appropri-ate supremum-norm), and that in turn finite linear combinations of factorized functions f ( N ) ( λ , . . . , λ N ) = f ( λ ) · · · f N ( λ N ) are dense in C ( σ ( a ) N ). It follows from this that µ a = µ ∞ a . (A.25)Since this generalizes (A.15) to N = ∞ , the proof of Theorem 4.1 is finished. (cid:3) If B is infinite-dimensional, for technical reasons the so-called projective tensor product should be used. See Landsman (2017), § Cf. Tychonoff’s theorem. The associated Borel structure is the one defined by the cylinder sets. In what follows, the notion of 1-randomness, originally defined by Martin-L¨of in the settingof constructive measure theory, will be explained through an equivalent definition in termsof Kolmogorov complexity. We assume basic familiarity with the notion of a computablefunction f : N → N , which may be defined through recursion theory or Turing machines.A string is a finite succession of bits (i.e. zeros and ones). The length of a string σ isdenoted by | σ | . The set of all strings of length N is denoted by 2 N , where 2 = { , } , and2 ∗ = [ N ∈ N N (B.1)denotes the set of all strings. The Kolmogorov complexity K ( σ ) of σ ∈ ∗ is defined,roughly speaking, as the length of the shortest computer program that prints σ and thenhalts. We then say, again roughly, that σ is Kolmogorov random if this shortest programcontains all of σ in its code, i.e. if the shortest computable description of σ is σ itself.To make this precise, fix some universal prefix-free Turing machine U , seen as per-forming a computation on input τ (in its prefix-free domain) with output U ( τ ), and define K ( σ ) = min τ ∈ ∗ {| τ | : U ( τ ) = σ } . (B.2)The function K : 2 ∗ → N is uncomputable, but that doesn’t mean it is ill-defined. Thechoice of U affects K ( σ ) up to a σ -independent constant, and to take this dependency intoaccount we state certain results in terms of the “big-O” notation familiar from Analysis. For example, if σ is easily computable, like the first | σ | binary digits of π , then K ( σ ) = O (log | σ | ) , (B.3)with the logarithm in base 2 (as only the length of σ counts). However, a random σ has K ( σ ) = | σ | + O (log | σ | ) . (B.4)We say that σ is c -Kolmogorov random , for some σ -independent constant c ∈ N , if K ( σ ) ≥ | σ | − c. (B.5) For details see Volchan (2002), Terwijn (2016), Diaconis & Skyrms (2018, Chapter 8), and Eagle (2019)for starters, technical surveys by Zvonkin & Levin (1970), Muchnik et al. (1998), Downey et al. (2006),Gr¨unwald & Vit´anyi (2008), and Dasgupta (2011), and books by Calude (2002), Li & Vit´anyi (2008), Nies(2010), and Downey & Hirschfeldt (2010). For history see van Lambalgen (1987, 1996) and Li & Vit´anyi(2008). For physical applications see e.g. Earman (1986), Svozil (1993, 2018), Calude (2004), Wolf (2015),Bendersky et al. (2016, 2017), Senno (2017), Baumeler et al. (2017), and Tadaki (2018, 2019). A Turing machine T is prefix-free if its domain D ( T ) consists of a prefix-free subset of 2 ∗ , i.e., if σ ∈ D ( T ) then στ / ∈ D ( T ) for any σ, τ ∈ ∗ , where στ is the concatenation of σ and τ : if T halts on input σ then it does not halt on either any initial part or any extension of σ . The prefix-free version is onlyneeded to correctly define randomness of sequences in terms of randomness of their initial parts, whichis necessary to satisfy Earman’s Principle : ‘
While idealizations are useful and, perhaps, even essential toprogress in physics, a sound principle of interpretation would seem to be that no effect can be counted asa genuine physical effect if it disappears when the idealizations are removed.’
See Earman (2004), p. 191.For finite strings σ one may work with the plain Kolmogorov complexity C ( σ ), defined as the length (inbits) of the shortest computer program (run on some fixed universal Turing machine U ) that computes σ . Recall that f ( n ) = O ( g ( n )) iff there are constants C and N such that | f ( n ) | ≤ C | g ( n ) | for all n ≥ N . | σ | = N gets large, the overwhelming majorityof strings in 2 N (and hence in 2 ∗ ) is c -random. The following theorem, which might becalled
Chaitin’s first incompleteness theorem , therefore shows that randomness is elusive: Theorem B.1
For any sound mathematical theory T containing enough arithmetic thereis a constant C ∈ N such that T cannot prove any sentence of the form K ( σ ) > C (althoughinfinitely many such sentences are true), and as such T can only prove (Kolmogorov)randomness of finitely many strings (although infinitely many strings are in fact random). The proof is quite complicated in its details but it is based on the existence of a com-putably enumerable (c.e.) list T = ( τ , τ , . . . ) of the theorems of T , and on the factthat after G¨odelian encoding by numbers, theorems of any given grammatical form canbe computably searched for in this list and will eventually be found. In particular, thereexists a program P (running on the universal prefix-free Turing machine U used to define K ( · )) such that P ( n ) halts iff there exists a string σ for which K ( σ ) > n is a theorem of T . If there is such a theorem the output is P ( n ) = σ , where σ appears in the first suchtheorem of the kind (according to the list T ). By definition of K ( · ), this means that K ( σ ) ≤ | P | + | n | . (B.6)Now suppose that no C as in the above statement of the theorem exists. Then there is n ∈ N large enough that n > | P | + | n | and there is a string σ ∈ ∗ such that T proves K ( σ ) > n . Since T is sound this is actually true, which gives a contradiction between K ( σ ) > n > | P | + | n | ; K ( σ ) ≤ | P | + | n | . (B.7)Note that this proof shows that a proof in T of K ( σ ) > n (if true) would also identify σ .As an idealization of a long (binary) string, a (binary) sequence x = x x · · · is aninfinite succession of bits, i.e. x ∈ N , with finite truncations x | N = x · · · x N ∈ N for each N ∈ N . We then call x Levin–Chaitin random if each truncation of x is c -Kolmogorovrandom for some c , that is, if there exists c ∈ N such that K ( x | N ) ≥ N − c for each N ∈ N .Equivalently, a sequence x is Levin–Chaitin random if eventually K ( x | N ) >> N , in thatlim N →∞ ( K ( x | N ) − N ) = ∞ . (B.8)Apart from having the same intuitive pull as Kolmogorov randomness (of strings), thisdefinition gains from the fact that it is equivalent to two other appealing notions of ran-domness, namely patternlessness and unpredictability , both also defined computationally. It is easy to show that least 2 N − N − c +1 + 1 strings σ of length | σ | = N are c -Kolmogorov random. Here “sound” means that all theorems proved by T are true; this is a stronger assumption than consis-tency (in fact only the arithmetic fragment of T needs to be sound). One may think of Peano Arithmetic(PA) or of Zermelo–Fraenkel set theory with the axiom of choice (ZFC). As in G¨odel’s theorems, onealso assumes that T is formalized as an axiomatic-deductive system in which proofs could in principle becarried out mechanically by a computer. The status of the true but unprovable sentences K ( σ ) > C inChaitin’s theorem is similar to that of the sentence G in G¨odel’s original proof of his first incompletenesstheorem, which roughly speaking is an arithmetization of the statement “I cannot be proved in T ”: as-suming soundness and hence consistency of T , one can prove G and K ( σ ) > C in the usual interpretationof the arithmetic fragment of T in the natural numbers N . See Chaitin (1987) for his own presentationand analysis of his incompleteness theorem. Raatikainen (1998) also gives a detailed presentation of thetheorem, including a critique of Chaitin’s ideology. Incidentally, he shows that there even exists a U withrespect to which K ( · ) is defined such that C = 0 in ZFC. See also Franz´en (2005) and G´acs (1989). The following contradiction can be made more dramatic by taking n such that n >> | P | + | n | . See Calude (2002), Theorem 6.38 (attributed to Chaitin) for this equivalence.
16n view of these equivalences we simply call a Levin–Chaitin random sequence . A sequence x ∈ k N is Borel normal in base k if each string σ has frequency k −| σ | in x . Any hope of defining randomness as Borel normality in base 10 is blocked by Champernowne’s number · · · , which is Borel normal but clearly notrandom in any reasonable sense (this is also true in base 2). The decimal expansion of π is also conjectured to be Borel normal in base 10 (with huge numerical support), although π clearly is not random either. However, Borel normality seems a desirable property oftruly random numbers on any good definition, and so we are fortunate to have: Proposition B.2
A 1-random sequence is Borel normal (in base 2, but in fact in anybase) and hence (“monkey typewriter theorem”) contains any finite string infinitely often. Another desirable property comes from the following theorem due to Martin-L¨of, in which P is the 50-50 probability on { , } and P ∞ is the induced probability measure on 2 N : Theorem B.3
With respect to P ∞ almost every outcome sequence x ∈ N is 1-random. This implies that the 1-random sequences form an uncountable subset of 2 N , althoughtopologically this subset is meagre (i.e. Baire first category). Chaitin’s incompletenesstheorem for (finite) strings has the following counterpart for (infinite) sequences:
Theorem B.4 If x ∈ N is 1-random, then ZFC (or any sufficiently comprehensive math-ematical theory T meant in Theorem B.1) can compute only finite many digits of x . This clearly excludes defining a 1-random number by somehow listing its digits, but somecan be described by a formula. One example is Chaitin’s Ω, or more precisely Ω U , whichis the halting probability of some fixed universal prefix-free Turing machine U , given byΩ U := X τ ∈ ∗ | U ( τ ) ↓ −| τ | . (B.9) Any pattern in a sequence x would make it compressible, but one has to define the notion of a patternvery carefully in a computational setting. This was accomplished by Martin-L¨of in 1966, who defineda pattern as a specific kind of probability-zero subset T of 2 N (called a “test”) that can be computablyapproximated by subsets T n ⊂ N of increasingly small probability 2 − n ; if x ∈ T , then x displays somepattern and it is patternless iff x / ∈ T for all such tests. Martin-L¨of’s definition yields what usually called1-randomness, in view of his use of so-called Σ sets. See the textbooks Li & Vit´anyi (2008), Calude (2002),Nies (2009), and Downey & Hirschfeldt (2010) for the equivalences between Levin–Chaitin randomness(incompressibility), Martin-L¨of randomness (patternlessless), and a third notion (unpredictability) thatevolved from the work of von Mieses and Ville, finalized by Schnorr. The name Levin–Chaitin randomness ,taken from Downey et al. (2006), is justified by its independent origin in Levin (1973) and Chaitin (1975). For details and proofs see Calude (2002), Corollary 6.32 in § § To see this, use the measure-theoretic isomorphism between (2 N , Σ K , P ∞ ) and ([0 , , Σ L , dx ), whereΣ K is the “Kolmogorov” σ -algebra generated by the cylinder sets [ σ ] = { x ∈ N | x || σ | = σ } , where σ ∈ ∗ ,and Σ K is the “Lebesgue” σ -algebra generated by the open subsets of [0 , § See Calude (2002), Theorem 6.63. Hence meagre subsets of [0 ,
1] exist with unit Lebesgue measure! More precisely, only finitely many true statements of the form: ‘the n ’th bit x n of x equals its actualvalue’ (i.e. 0 or 1) are provable in T (where a proof in T may be seen as a computation, since one mayalgorithmically search for this proof in a list). See Calude (2002), Theorem 8.7, which is stated for Chaitin’sΩ but whose proof holds for any 1-random sequence. Indeed, as pointed out to the author by Bas Terwijn,even more generally, ZFC (etc.) can only compute finitely many digits of any immune sequence (we saythat a sequence x ∈ N is immune if the corresponding subset S ⊂ N (i.e. 1 S = x ) contains no infinite c.e.subset), and by (for example) Corollary 6.42 in Calude (2002) any 1-random sequence is immune. There exists a U for which not a single digit of Ω U can be known, see Calude (2002), Theorem 8.11. Bell’s theorem and free will theorem
In support of the analysis of hidden variable theories in the main text, this appendixreviews Bell’s (1964) theorem and the free will theorem, streamlining earlier expositions(Cator & Landsman, 2014; Landsman, 2017, Chapter 6) and leaving out proofs and otheradornments. In the specific context of ’t Hooft’s theory (where the measurement settingsare determined by the hidden state) and Bohmian mechanics (where they are not, as in theoriginal formulation of Bell’s theorem and in most hidden variable theories) an advantageof my approach is that both free (uncorrelated) und correlated settings fall within itsscope; the former are distinguished from the latter by an independence assumption. As a warm-up I start with a version of the Kochen–Specker theorem, whose logicalform is very similar to Bell’s (1964) theorem and the free will theorem, as follows:
Theorem C.1
Determinism, qm , non-contextuality, and free choice are contradictory. Of course, this unusual formulation hinges on the precise meaning of these terms. • determinism is the conjunction of the following two assumptions.1. There is a state space X with associated functions A : X → S and L : X → O ,where S is the set of all possible measurement settings Alice can choose from, namelya suitable finite set of orthonormal bases of R (11 well-chosen bases will do to arriveat a contradiction), and O is some set of possible measurement outcomes . Thussome x ∈ X determines both Alice’s setting a = A ( x ) and her outcome α = L ( x ).2. There exists some set Λ and an additional function H : X → Λ such that L = L ( A, H ) , (C.1)in the sense that for each x ∈ X one has L ( x ) = ˆ L ( A ( x ) , H ( x )) for a certain functionˆ L : S × Λ → O . This self-explanatory assumption just states that each measurementoutcome L ( x ) = ˆ L ( a, λ ) is determined by the measurement setting a = A ( x ) andthe “hidden” variable or state λ = H ( x ) of the particle undergoing measurement. • qm fixes O = { (0 , , , (1 , , , (1 , , } , which is a non-probabilistic fact of quan-tum mechanics with overwhelming (though indirect) experimental support. • non-contextuality stipulates that the function ˆ L just introduced take the formˆ L (( ~e , ~e , ~e ) , λ ) = ( ˜ L ( ~e , λ ) , ˜ L ( ~e , λ ) , ˜ L ( ~e , λ )) , (C.2) The original reference for Bell’s theorem is Bell (1964); see further footnote 6, and in the context ofthis appendix also Esfeld (2015) and Sen & Valentini (2020) are relevant. The free will theorem originatesin Heywood & Redhead (1983), followed by Stairs (1983), Brown & Svetlichny (1990), Clifton (1993), and,as name-givers, Conway & Kochen (2009). Both theorems can and have been presented and interpreted inmany different ways, of which we choose the one that is relevant for the general discussion on randomnessin the main body of the paper. This appendix is taken almost verbatim from Landsman (2020). This addresses a problem Bell faced even according to some of his most ardent supporters (Norsen, 2009;Seevinck & Uffink, 2011), namely the tension between the idea that the hidden variables (in the pertinentcausal past) should on the one hand include all ontological information relevant to the experiment, but onthe other hand should leave Alice and Bob free to choose any settings they like. Whatever its ultimatefate, ’t Hooft’s staunch determinism has drawn attention to issues like this, as has the free will theorem. If her setting is a basis ( ~e , ~e , ~e ), Alice measures the quantities ( J ~e , J ~e , J ~e ), where J ~e = h ~J, ~e i i isthe component of the angular momentum operator ~J of a massive spin-1 particle in the direction ~e i . L : S × Λ → { , } that also satisfies ˜ L ( − ~e, λ ) = ˜ L ( ~e, λ ). • free choice finally states that the following function is surjective: A × H : X → S × Λ; x ( A ( x ) , H ( x )) . (C.3)In other words, for each ( a, λ ) ∈ S × Λ there is an x ∈ X for which A ( x ) = a and H ( x ) = λ . This makes A and H “independent” (or: makes a and λ free variables).See Landsman (2017), § Bell’s (1964) theorem and the free will theorem both take a similar generic form, namely:
Theorem C.2
Determinism, qm , local contextuality, and free choice, are contradictory. Once again, I have to explain what these terms exactly mean in the given context. • determinism is a straightforward adaptation of the above meaning to the bipartite“Alice and Bob” setting. Thus we have a state space X with associated functions A : X → S ; B : X → S ; L : X → O R : X → O, (C.4)where S , the set of all possible measurement settings Alice and Bob can each choosefrom, differs a bit between the two theorems: for the free will theorem it is the sameas for the Kochen–Specker theorem above, as is the set O of possible measurementoutcomes, whereas for Bell’s theorem (in which Alice and Bob each measure a 2-levelsystem), S is some finite set of angles (three is enough), and O = { , } . – In the free will case, these functions and the state x ∈ X determine both thesettings a = A ( x ) and b = B ( x ) of a measurement and its outcomes α = L ( x )and β = R ( x ) for Alice on the L eft and for Bob on the R ight, respectively. – All of this is also true in the Bell case, but since his theorem relies on impossiblemeasurement statistics (as opposed to impossible individual outcomes), one inaddition assumes a probability measure µ on X . Furthermore, there exists some set Λ and some function H : X → Λ such that L = L ( A, B, H ); R = R ( A, B, H ) , (C.5)in the sense that for each x ∈ X one has functional relationships L ( x ) = ˆ L ( A ( x ) , B ( x ) , H ( x )); R ( x ) = ˆ R ( A ( x ) , B ( x ) , H ( x )) , (C.6)for certain functions ˆ L : S × S × Λ → O and ˆ R : S × S × Λ → O . Here S = { ( x, y, z ) ∈ R | x + y + z = 1 } is the 2-sphere, seen as the space of unit vectors in R . Eq. (C.2) means that the outcome of Alice’s measurement of J ~e i is independent of the “context”( J ~e , J ~e , J ~e ); she might as well measure J ~e i by itself. The last equation is trivial, since ( J − ~e i ) = ( J ~e i ) . The assumptions imply the existence of a coloring C λ : P → { , } of R , where P ⊂ S consist of allunit vectors contained in all bases in S , and λ “goes along for a free ride”. A coloring of R is a function C : P → { , } such that for any set { e , e , e } in P with e i e j = δ ij and e + e + e = 1 where 1 isthe 3 × e i for which C ( e i ) = 1. Indeed, one finds C λ ( ~e ) = ˜ L ( ~e, λ ). Thekey to the proof of Kochen–Specker is that on a suitable choice of the set S such a coloring cannot exist. The existence of µ is of course predicated on X being a measure space with corresponding σ -algebraof measurable subsets, with respect to which all functions in (C.4) and below are measurable. qm reflects elementary quantum mechanics of correlated 2-level and 3-level quantumsystems for the Bell and the free will cases, respectively, as follows: – In the free will theorem , O = { (0 , , , (1 , , , (1 , , } is the same as forthe Kochen–Specker theorem. In addition perfect correlation obtains: if a =( ~e , ~e , ~e ) is Alice’s orthonormal basis and b = ( ~f , ~f , ~f ) is Bob’s, one has ~e i = ~f j ⇒ ˆ L i ( a, b, z ) = ˆ R j ( a, b, z ) , (C.7)where ˆ L i , ˆ R j : S × S × Λ → { , } are the components of ˆ L and ˆ R , respectively.Finally, if ( a ′ , b ′ ) differs from ( a, b ) by changing the sign of any basis vector,ˆ L ( a ′ , b ′ , λ ) = ˆ L ( a, b, λ ); ˆ R ( a ′ , b ′ , λ ) = ˆ R ( a, b, λ ) . (C.8) – In Bell’s theorem , O = { , } , and the statistics for the experiment is reproducedas conditional joint probabilities given by the measure µ through P ( L = R | A = a, B = b ) = sin ( a − b ) . (C.9) • local contextuality , which replaces and weakens non-contextuality, means that L ( A, B, H ) = L ( A, H ); R ( A, B, H ) = G ( B, H ) . (C.10)In words: Alice’s outcome given λ does not depend on Bob’s setting, and vice versa . • free choice is an independence assumption that looks differently for both theorems: – In the free will theorem it means that each ( a, b, λ ) ∈ S × S × Λ is possible inthat there is an x ∈ X for which A ( x ) = a , B ( x ) = b , and H ( x ) = λ . – In Bell’s theorem , (
A, B, H ) are probabilistically independent relative to µ . This concludes the joint statement of the free will theorem and Bell’s (1964) theorem in theform we need for the main text. The former is proved by reduction to the Kochen–Speckentheorem, whilst the latter follows by reduction to the usual version of Bell’s theorem viathe free choice assumption; see Landsman (1917), Chapter 6 for details.For our purposes these theorems are equivalent, despite subtle differences in theirassumptions. Bell’s theorem is much more robust in that it does not rely on perfectcorrelations (which are hard to realize experimentally), and in addition it requires almostno input from quantum theory. On the other hand, Bell’s theorem uses probability theoryin a highly nontrivial way: like the hidden variable theories it is supposed to exclude itrelies on the possibility of fair sampling of the probability measure µ . The factorizationcondition defining probabilistic independence passes this requirement of fair sampling onto both the hidden variable and the settings, which brings us back to the main text.Different parties may now be identified by the assumption they drop: Copenhagenquantum mechanics rejects determinism, Valentini (2019) rejects the Born rule and hence qm , Bohmians rejects local contextuality, and finally ’t Hooft rejects free choice. However,as we argue in the main text, even the latter two camps do not really have a deterministictheory underneath quantum mechanics because of their need to randomly sample theprobability measure they must use to recover the predictions of quantum mechanics. In Bell’s theorem quantum theory can be replaced by experimental support (Hensen et al. , 2015). As in Kochen–Specker, this is because Alice & Bob measure squares of (spin-1) angular momenta. By definition, this also implies that the pairs (
A, B ), (
A, H ), and (
B, H ) are also independent. eferences [1] Baumeler, A., B´edard, C.A., Brassard, G., Wolf, S. (2017). Kolmogorov amplificationfrom Bell correlation, IEEE International Symposium on Information Theory (ISIT) . doi:10.1109/ISIT.2017.8006790 .[2] Bell, J.S. (1964). On the Einstein–Podolsky–Rosen Paradox, Physics
1, 195–200.[3] Bell, J.S. (1966). On the problem of hidden variables in quantum mechanics,
Reviews ofModern Physics
38, 447–452.[4] Bendersky, A., de la Torre, G., Senno, G., Figueira, S., Ac´ın, A. (2016). Algorithmic pseudo-randomness in quantum setups,
Physical Review Letters
Physical Review Letters
Physical Review
85, 166–179, 180–193.[7] Bohr, N. 91935). Can quantum-mechanical description of physical reality be considered com-plete?,
Physical Review
48, 696–702.[8] Born, M. (1926). Zur Quantenmechanik der Stoßvorg¨ange,
Zeitschrift f¨ur Physik
37, 863–867.[9] Breuer, T. (2001). Von Neumann, G¨odel and quantum incompleteness,
John von Neumannand the Foundations of Quantum Physics , eds. R´edei, M., St¨oltzner, M., pp. 75–82 (Kluwer).[10] Broglie, L. de (1928). La nouvelle dynamique des quanta,
Electrons et Photons: Rapports etDiscussions du Cinqui`eme Conseil de Physique , pp. 105–132 (Gauthier-Villars).[11] Brown, H., Svetlichny, G. (1990). Nonlocality and Gleason’s lemma. Part I. deterministictheories,
Foundations of Physics
20, 1379–1387.[12] Brown, H., Timpson, C.G. (2014). Bell on Bell’s theorem: The changing face of nonlocality, \protect\vrule width0pt\protect\href{http://arxiv.org/abs/1501.03521}{arXiv:1501.03521} .[13] Bub, J. (2011). Is von Neumann’s ‘no hidden variables’ proof silly?,
Deep Beauty: Mathe-matical Innovation and the Search for Underlying Intelligibility in the Quantum World , ed.Halvorson, H., pp. 393–408 (Cambridge University Press).[14] Butterfield, J. (1992). Bell’s theorem: what it takes,
The British Journal for the Philosophyof Science
43, 41–83.[15] Calude, C.S. (2002).
Information and Randomness: An Algorithmic Perspective , 2nd Edition(Springer).[16] Calude, C.S. (2004). Algorithmic randomness, quantum physics, and incompleteness,
LectureNotes in Computer Science
Foundations of Physics
44, 781–791.[18] Clifton, R. (1993). Getting contextual and nonlocal elements-of-reality the easy way,
AmericanJournal of Physics
61, 443–447.[19] Chaitin, G.J. (1975). A theory of program size formally identical to information theory,
Jour-nal of the Association for Computing Machinery
22, 329–340.[20] Chaitin, G.J. (1987).
Information, Randomness & Incompleteness: Papers on AlgorithmicInformation Theory (World Scientific).[21] Conway, J.H., Kochen, S. (2009). The Strong Free Will Theorem,
Notices of the AmericanMathematical Society
56, 226–232.
22] Corry, L. (2004).
David Hilbert and the Axiomatization of Physics (1898–1918): From Grund-lagen der Geometrie to Grundlagen der Physik (Kluwer).[23] Crull, E., Bacciagaluppi, G. (2016).
Grete Hermann-Between Physics and Philosophy (Springer).[24] Dasgupta, A. (2011). Mathematical Foundations of Randomness,
Handbook of the Philosophyof Science. Volume 7: Philosophy of Statistics , eds. Bandyopadhyay, P.S., Forster, M.R., pp.641–710 (North-Holland/Elsevier).[25] Diaconis, P., Skyrms, B. (2018).
Ten Great Ideas About Chance (Princeton University Press).[26] Dieks, D. (2016). Von Neumann’s impossibility proof: Mathematics in the service of rhetorics,
Studies in History and Philosophy of Modern Physics
60, 136–148.[27] Downey, R., Hirschfeldt, D.R. (2010).
Algorithmic Randomness and Complexity (Springer).[28] Downey, R., Hirschfeldt, D.R., Nies, A., Terwijn, S.A. (2006). Calibrating randomness,
Bul-letin of Symbolic Logic
12, 411–491.[29] D¨urr, D., Goldstein, S., Zanghi, N. (1992). Quantum equilibrium and the origin of absoluteuncertainty,
Journal of Statistical Physics
67, 843–907.[30] Eagle, A. (2019). Chance versus Randomness,
The Stanford Encyclopedia of Philosophy ,URL = https://plato.stanford.edu/archives/spr2019/entries/chance-randomness/ .[31] Earman, J. (2004). Curie’s Principle and spontaneous symmetry breaking,
International Stud-ies in Philosophy of Science
18, 173–198.[32] Einstein, A., Podolsky, B., Rosen, N. (1935). Can quantum-mechanical description of physicalreality be considered complete?,
Physical Review
47, 777–780.[33] Esfeld, M. (2015). Bell’s Theorem and the issue of determinism and indeterminism,
Founda-tions of Physics
45, 471–482.[34] Franz´en, T. (2005).
G¨odel’s Theorem: An Incomplete Guide to its Use and Abuse (AK Peters).[35] G´acs, P. (1989). Review of “Algorithmic Information Theory” by Gregory J. Chaitin,
TheJournal of Symbolic Logic
54, 624–627.[36] G¨odel, K. (1931). ¨Uber formal unentscheidbare S¨atze der Principia Mathematica und ver-wandter Systeme I,
Monatshefte f¨ur Mathematik und Physik
38, 173–198.[37] Goldstein, S. (2017). Bohmian Mechanics,
The Stanford Encyclopedia of Philosophy , https://plato.stanford.edu/archives/sum2017/entries/qm-bohm/ .[38] Greenstein, G. (2019). Quantum Strangeness: Wrestling with Bell’s Theorem and the UltimateNature of Reality (The MIT Press).[39] Gr¨unwald, P.D., Vit´anyi, P.M.B. (2008). Algorithmic information theory,
Handbook of thePhilosophy of Information , eds. Adriaans, P., van Benthem, J., pp. 281–320 (Elsevier).[40] Haag, R. (1992).
Local Quantum Physics: Fields, Particles, Algebras (Springer).[41] Hacking, I. (1990).
The Taming of Chance (Cambridge University Press).[42] Hensen, B. et al (2015). Experimental loophole-free violation of a Bell inequality using entan-gled electron spins separated by 1.3 km,
Nature
Foundationsof Physics
13, 481–499.[44] Hilbert, D. (1902). Mathematical Problems. Lecture delivered before the InternationalCongress of Mathematicians at Paris in 1900,
Bulletin of the American Mathematical Society
8, 437–479. Translated from
G¨ottinger Nachrichten , 1900, pp. 253–297.
45] Hooft, G. ’t (2016).
The Cellular Automaton Interpretation of Quantum Mechanics (SpringerOpen). .[46] Kline, M. (1980).
Mathematics: The Loss of Certainty (Oxford University Press).[47] Kochen, S., Specker, E. (1967). The problem of hidden variables in quantum mechanics,
Journal of Mathematics and Mechanics
17, 59–87.[48] Kolmogorov, A.N. (1933).
Grundbegriffe de Wahrscheinlichkeitsrechnung (Springer–Verlag).[49] Lambalgen, M. van (1987).
Random Sequences . PhD Thesis, University of Amsterdam, .[50] Lambalgen, M. van (1996). Randomness and foundations of probability: Von Mises’ axioma-tisation of random sequences,
Statistics, Probability and Game Theory: Papers in Honour ofDavid Blackwell, IMS Lecture Notes–Monograph Series, Vol. 30 , pp. 347–367.[51] Landsman, K. (2017).
Foundations of Quantum Theory: From Classical Concepts to OperatorAlgebras (Springer Open). .[52] Landsman, K. (2020). Randomness? What randomness?,
Foundations of Physics
50, 61–104.[53] Landsman, K. (2021). Quantum theory and functional analysis,
Oxford Handbook of the His-tory of Interpretations and Foundations of Quantum Mechanics , ed. O. Freire, to appear(Oxford University Press).[54] Levin, L.A. (1973). On the notion of a random sequence,
Soviet Mathematics–Doklady
An Introduction to Kolmogorov Complexity and Its Applica-tions. Third Edition (Springer).[56] Mehra, J., Rechenberg, H. (2000).
The Historical Development of Quantum Theory. Vol. 6:The Completion of Quantum Mechanics 1926–1941. Part 1 (Springer).[57] Mises, R. von (1919). Grundlagen der Wahrscheinlichkeitsrechnung,
Mathematische Zeitschrift
5, 52-99.[58] Montague, R. (1974). Deterministic theories,
Formal Philosophy: Selected Papers of RichardMontague , ed. Thomason, R.H., pp. 303–360 (Yale University Press).[59] Moschovakis, J.R. (2016). Iterated definability, lawless sequences and Brouwer’s continuum.
G¨odel’s Disjunction , eds. Horsten, L., Welch, P., pp. 92–107 (Oxford University Press).[60] Muchnik, A.A., Semenov, A.L., Uspensky, V.A. (1998). Mathematical metaphysics of ran-domness,
Theoretical Computer Science
Mathematische Grundlagen der Quantenmechanik (Springer–Verlag). English translation:
Mathematical Foundations of Quantum Mechanics (PrincetonUniversity Press, 1955).[62] Nies, A. (2009).
Computability and Randomness (Oxford University Press).[63] Norsen, T. (2009). Local causality and completeness: Bell vs. Jarrett,
Foundations of Physics
39, 273–294.[64] Pour-El, M.B., & Richards, J.I. (2016).
Computability in Analysis and Physics (CambridgeUniversity Press).[65] Raatikainen, P. (1998). On interpreting Chaitin’s incompleteness theorem,
Journal of Philo-sophical Logic
27, 569–586.[66] R´edei, M., St¨oltzner, eds. (2001).
John von Neumann and the Foundations of QuantumPhysics . Dordrecht: Kluwer.[67] Redhead, M. (1989).
Incompleteness, Nonlocality, and Realism (Oxford University Press).
68] Renn, J., ed. (2007).
The Genesis of General Relativity, Vol. 4 (Springer).[69] Scriven, M. (1957). The present status of determinism in physics,
Journal of Philosophy
Explanation, Prediction, andConfirmation
2, 425–450.[71] Sen, I., Valentini, A. (2020a). Superdeterministic hidden-variables models I: nonequilibriumand signalling, \protect\vrule width0pt\protect\href{http://arxiv.org/abs/2003.11989}{arXiv:2003.11989} ,II: conspiracy, \protect\vrule width0pt\protect\href{http://arxiv.org/abs/2003.12195}{arXiv:2003.12195} .[72] Senno, G. (2017).
A Computer-Theoretic Outlook on Foundations of Quantum Information ,Ph.D. Thesis, Universidad de Buenos Aires.[73] Sieg, W. (2013).
Hilbert’s Programs and Beyond (Oxford University Press).[74] Smith, P. (2013).
An Introduction to G¨odel’s Theorems (Cambridge University Press).[75] Stairs, A. (1983). Quantum logic, realism, and value definiteness,
Philosophy of Science
Randomness & Undecidability in Physics (World Scientific).[77] Svozil, K. (2018).
Physical (A)Causality: determinism, Randomness and Uncaused Events (Springer Open). .[78] Svozil, K., Calude, C.S., Stay, M.A. (2005). From Heisenberg to G¨odel via Chaitin,
Interna-tional Journal of Theoretical Physics
44, 1053–1065.[79] Szangolies, J. (2018). Epistemic horizons and the foundations of quantum mechanics, arXiv: 1805.10668 .[80] Tadaki, K. (2018). A refinement of quantum mechanics by algorithmic randomness, arXiv:1804.10174 .[81] Tadaki, K. (2019).
A Statistical Mechanical Interpretation of Algorithmic Information Theory (Springer).[82] Tait, W. (2005).
The Provenance of Pure Reason: Essays in the Philosophy of Mathematicsand Its History (Oxford University Press).[83] Tapp, C. (2013).
An den Grenzen des Endlichen: Das Hilbertprogramm (Springer).[84] Terwijn, S.A. (2016). The Mathematical Foundations of Randomness,
The Challenge ofChance: A Multidisciplinary Approach from Science and the Humanities , eds. Landsman, K.,van Wolde, E., pp. 49–66 (Springer), .[85] Valentini, A. (2019). Foundations of statistical mechanics and the status of the Born rule inde Broglie-Bohm pilot-wave theory, arXiv:1906.10761 .[86] Volchan, S.B. (2002). What is a random sequence?,
The American Mathematical Monthly
Quantum Information& Computation
1, 1–25.[88] Wiseman, H.M. (2014). The two Bell’s theorems of John Bell,
Journal of Physics A
47, 424001.[89] Wolf, S. (2015). Nonlocality without counterfactual reasoning,
Physical Review A
92, 052102.[90] Yanofsky, N.S. (2013).
The Outer Limits of Reason (The MIT Press).[91] Zach, R. (2001).
Hilbert’s Finitism: Historical, Philosophical, and Metamathematical Perspec-tives , PhD Thesis, University of California at Berkeley.[92] Zvonkin, A.K., Levin, L.A. (1970). The complexity of finite objects and the development ofthe concepts of information and randomness by means of the theory of algorithms,
RussianMathematical Surveys
25, 83–124.25, 83–124.