arXiv:quant-ph/0501159v1 27 Jan 2005 oa,hde aibeter about theory variable hidden local, o esr oe hnte s h esrmn settings measurement the use m they an when Alice “one” both a th that measure of Bob probability probabilities the the Hence, expressing outcomes. when various omitted be will part hr eitrrttebnr ausa lmnso mdl 2 1 ‘modulo of that elements such as calculations’ values binary the interpret we where asuetesm tt-rprto of state-preparation same the use ways m ilto ftebudo qain1by 1 Equation of bound the of violation a ucms si sudrto htfrec trial each for that understood is it As th and outcomes. measurements different p the the between all correlations differen of ble estimation the accurate The using an obtaining times thus “1”. many settings, and experiment “0” the labeled repeat are parties which outcomes, possible two and mgn w ate lc n o ( mechanics. Bob quantum and by Alice parties violated two be Imagine can which separated th space-like periments, on two bound between upper correlations of following strength the gives theories classical for uhta nttlteeaefu xeietlstusta c that system: set-ups combined experimental the four to are system, apply there the total of in part their that on such measurements two of out one form itiue system distributed ewl aet elc h hoyo unu ehnc by mechanics futur quantum the of in theory that the inconceivable replace not to is have relevan it will more as we the matter is the aspect of experimental side This many 10]). done o been [1, has nonlocality times (as the experimentally verified that be is can conclusion Nature important the- local more a the as ory, phrased be cannot mechanics quantum of theory projectors | if Φ 0 A A A h anrsl fBl 2 n HH[]i htfrany for that is [6] CHSH and [2] Bell of result main The h lue-on-hmn-ot(HH nqaiy[6] inequality (CHSH) Clauser-Horne-Shimony-Holt The AB and and and ( x i m , y = ∈{ 1 A ∑ m m B , 0 m √ 1 B B , s,freape h nage aro unu bits quantum of pair entangled the example, for use, 1 1 m h olwn nqaiyms hold: must inequality following the , eateto optrSine nvriyo aionaa California of University Science, Computer of Department 2 1 B sdntdsml yProb by simply denoted is } ( x ) eie h atta hsrsl rvsta the that proves result this that fact the Besides . Prob , | o ahmaueeto ahsd hr are there side each on measurement each For . ewrs onain fqatmmcais nonlocality, mechanics, quantum of foundations Keywords: 03.67.Mn 03.67.Hk, 03.65.Ta, 03.65.Ud, numbers: PACS indee a are such correlation allow superstrong not why does reason Nature a perfo that gives believes be Letter one can If computations bit. one distributed with all tha shown that is implies tha It bits stronger are communication. faster-than-light that prohibit inequalities Bell/CHSH of violations y 00 ∈{ ∑ hsLte ok ttecneune fs-ald‘superst so-called of consequences the at looks Letter This i 0 ( , + 1 m } Φ x A | Prob 11 AB + i ahprycnidpnetyper- independently can party Each . m ) + ( m y B n utbesto measurement of set suitable a and 1 x A ≡ ≡ mluil osqecso uesrn Nonlocality Superstrong of Consequences Implausible + x .Qatmmcaisallows mechanics Quantum 0. ( m · m y y B = ) 0 A Φ ≡ , AB m ( x 0 B m A · n h measurements the and ) 2 y 0 A Φ , and ) + ( = AB m √ ≤ 1 h conditional the , 0 A 2 B , , m m 3 htsaea share that ) ≈ , 1 B 1 B A ) = , 3 and ( . 1 41 m ) 1 A . i a Dam van Wim , , B ossi- m ex- (1) eir al- an 0 B d e e e ) f t t omncto complexity communication 5 o ro ht2 that proof a for [5] ilto fteCS nqaiywt orlto emof term correlation a with allow inequality 2 to CHSH seems the Nature of violation why a question the ask Rohrlich Daniel physics. quantum of theory the by of allowed plausibility the at cality look we Letter this blt itiuino qain2i h nypsil solu possible only the is prob 2 the Equation fact, of In distribution with communication. ability party conventional other of use the the to Alic out information prevents transferring still from outcomes Bob the or of randomization the while t ssrasdb orlto au f4 h non-zero The 4. simply of are value theory correlation super-nonlocal this a of Inequal by probabilities nonlocality surpassed thi is the prove 1 where authors ity The toy-theory case. a the constructing be to by surpris- not Perhaps that out turns proof mechanics.” this a quantum ingly, to of closer all be determine would they and we o and nonlocality violation inequality, then quantum CHSH the the so, determine If . together . “. better would a causality for towards step Nature great be of would [2 result understanding a to Such violation 4?” requiremen the of stead the restrict Could . causality . “. relativistic [13]: of themselves ask They limit.) esol osdralpsil iltoso qain1 no 1, Equation “2 of the violations just possible all physics’ consider quantum should ‘nonlocality-without we perspectiv call this could From we well. as which candid nonlocal future be to proper have any will i explanation, theory by local data a empirical out the rules as self and ag results; to experimental have our will matter with theory no succeeding the But formulation, exact irrelevant. the its mechanics making Nature, quantum of of model general nonlocality more or accurate more a hslaside otemxmlyvoaigcreainva correlation violating maximally the to indeed leads This at abr,SnaBraa A91651,USA 93106-5110, CA Barbara, Santa Barbara, Santa t h xsec fmxmlysprtogcorrelated superstrong maximally of existence the t o possible. not d unu ehnc los e ekeog to enough weak yet allows, mechanics quantum n mdwt rva muto omncto,i.e. communication, of amount trivial a with rmed + ognnoa orltos,wihaehypothetical are which correlations’, nonlocal rong nasre fatce 1,1,1] ad oec and Popescu Sandu 14], 13, [12, articles of series a In √ opttoa fe uc’ hntersl nthe in result the then lunch’, ‘free computational ∗ Prob Prob Prob Prob ,btntwt oe Seteatceb oi Cirel’son Boris by article the (See more. with not but 2, hr h olclcreain r togrta those than stronger are correlations nonlocal the where ( ( ( ( m m m m + x 1 A 1 A x A x A , y √ = = = = ∈{ ∑ 2 0 0 1 1 0 , 6≤ , , , , 1 } m m m m Prob ”voaino unu ehnc.In mechanics. quantum of violation 3” 1 B 1 B y B y B + = = = = √ ( 1 0 1 0 side h unu mechanical quantum the indeed is 2 m = ) = ) = ) = ) x A + 1 2 2 2 2 1 1 1 m   y B ≡ if if x · xy xy y = ) uesrn nonlo- superstrong = { ∈ 11 00 4 . , , 01 + , 10 √ } ]in- 2] , tion ree lue (3) (2) ate e, t- e s - f - - t t , 2 if we want to combine a correlation value of 4 with the preser- Any Boolean function f : 0,1 n 0,1 n 0,1 can vation of causality. be expressed as a multi-variable{ polynomial} ×{ with} →{ modulo} two So, if causality is still respected with the superstrong corre- arithmetic (where 1 + 1 = 2 0). This is most easily seen ≡ lations of Equation 2, why does Nature not allow it? Are there by the fact that elementary Boolean operations like AND, OR, any obvious first principles that forbid a violation stronger NOT or ‘equivalence’ can be calculated with addition and mul- than that of quantum mechanics? When trying to answer tiplication over 0,1 : this question in a meaningful way, it is important to remem- { } (x AND y) x y, (x OR y) x + y + x y, ber to ignore everything one knows about quantum mechan- ≡ · ≡ · ics. As explained above, the point is to consider all possi- NOT(x) 1 + x, (x y) 1 + x + y,  ≡ ⇔ ≡ ble (future) physical theories, not just the contemporary one. where the value 1 means “True” and 0 means “False”. Just as Cirel’son’s bound already shows us that quantum mechanics is any Boolean function f : 0,1 n 0,1 can be constructed incompatible with a violation of Equation 1 that goes beyond from those primitives, so{ can }f be→{ constructed} from the ele- “2 √2 3”, hence it is not interesting to derive a contra- + mentary mod2 operations “+”and“ ”. The 2-bit equivalence diction under6≤ assumptions that use features of quantum me- · relation EQ, for example, thus becomes chanics like the superposition principle, linearity, et cetera. Instead, we want to assume nothing else but a violation of EQ(x1x2,y1y2) = (x1 y1)AND(x2 y2) Equation 1 by a value greater than 2 + √2. ⇔ ⇔ (1 + x1 + y1) (1 + x2 + y2). In this Letter we look at the consequences of super- ≡ · strong nonlocality for the theory of communication complex- Furthermore, we can rewrite such polynomials as a finite sum- ity, which describes how much communication is needed to mation of products f (~x,~y) ∑i Pi(~x) Qi(~y), where Pi are poly- n ≡ · n evaluate a distributed function f . More specifically, consider nomials in~x 0,1 and Qi are monomials in~y 0,1 . In n n ∈{n } ∈{ } a Boolean function f : 0,1 0,1 0,1 , which has total there are 2 different monomials Qi(~y)= ∏ j S y j that we as input two n-bit strings{ ~x,~y} ×{0,1 }n.→{ If A possesses} the x- have to consider (one for each subset S 1,...,∈ n ), hence ⊆{ } string and B the y-string, how∈{ many} bits do A and B have to the index i in the summation can be limited to 1 i 2n. ≤ ≤ exchange in order to determined the function value f (~x,~y)? This gives us a way of representing the function f as an inner How to answer this question—which depends on the specific product problem of input size 2n: function and the resources of A and B—is studied in the field 2n of (quantum) communication complexity [11]. For certain f (x1 xn,y1 yn) ∑ Pi(~x) Qi(~y), (4) f it has been shown that quantum entanglement can reduce ··· ··· ≡ i=1 · the amount of classical information that A and B need to ex- with~x,~y 0,1 n. For the example of the 2-bit equality func- change to evaluate f [7], while for other functions the quan- ∈{ } tion EQ this is shown by tum complexity is effectively the same as the classical com- plexity. An example of latter is the Inner Product function EQ(x1x2,y1y2) (1 + x1 + y1) (1 + x2 + y2) n n IPn : 0,1 0,1 0,1 , which is defined by ≡ · { } ×{ } →{ } (1 + x1 + x2 + x1x2) 1 + 1 y1y2 n ≡ · · +(1 + x2) y1 + (1 + x1) y2 IPn(x1 xn,y1 yn) ∑ xi yi. · · ··· ··· ≡ i=1 · 4 ∑ Pi(x1,x2) Qi(y1,y2). Even if we allow A and B to use an unlimited amount of en- ≡ i=1 · tangled qubits, the communication complexity will still be n bits [8] (which is the maximum possible complexity as B can We can view this as an inner product problem because all the always send A all of the n bits of his input ~y, after which A bit values Pi(~x) are known to Alice and all the values Qi(~y) evaluates f (~x,~y) on her side). are known to Bob without the need for any communication Here it will be shown that a maximum violation of the between them. Hence, if A and B are able to compute the n CHSH Inequality 1 (according to the “4 3” of Equation 3) IP function for input sizes of 2 with one bit of communica- leads to a situation where the notion of communication6≤ com- tion, then they are also able to calculate any decision problem f : 0,1 n 0,1 n 0,1 with a single bit of information plexity is vacuous: all distributed decision problems can be { } ×{ } →{ } solved with 100% accuracy with only one bit of communica- exchange. Next we will see that this indeed possible with a tion. (Note that at least one bit needs to be communicated maximum violation of the CHSH inequality. if we want to preserve causality.) To prove our result, we Assume a model of Nature where the probabilities of Equa- first describe a way of expressing all possible distributed func- tions 2 and 3 are applicable, and hence where the correlation tions in a standardformat that coincides with the inner product Prob(mA + mB x y) = 1 problem for two parties. Then we will see how, with super- x y ≡ · strong correlations, the IP problem (and hence all problems) holds for all x,y 0,1 . In such a world Alice and Bob can be solved with the minimal amount of one bit of commu- (with input bits x∈{and y}) can perform two separated mea- nication from Bob to Alice. The results in this Letter were surements on their super-correlated states that yield the out- mentioned earlier in the Ph.D. thesis of the author [9]. comes α and β obeying α + β x y. From this it follows ≡ · 3 that in the case of the inner product function IPN on strings need an exponential amount of prior superstrong nonlocality of length N, Alice and Bob can perform N measurements on (as is indeed sometimes the case), the solution of all possi- N super-correlated particle pairs in order to obtain—without ble distributed functions with a single bit of communication any communication—a collection of bit values αi and βi, with does contradict our experiences that certain computational αi +βi xi yi for every1 i N. The commutativityof addi- tasks are harder than other ones. Similar as in computabil- tion (modulo≡ · two) allows≤ the≤ following regrouping of the bits ity theory, there is a hierarchy of different complexity classes by the two separated sides of the communication protocol: of communication problems [3]. Such hierarchies are at the core of theoretical computer science, and their absence—as N happened here by assuming superstrong correlations—goes IPN(x1 xN ,y1 yN) ∑ xi yi ··· ··· ≡ i=1 · squarely against the worldview and experience of probably all N researchers in the field of complexity theory. ∑ (αi + βi) If we accept the absence of intrinsic complexity as an ar- ≡ i=1 gument against superstrong correlations, then it is natural to N N wonder if we can obtain similar results for correlationsthat are ∑ αi + ∑ βi . ≡ less strong than those of Equation 3. It is tempting to specu- i=1 ! i=1 ! late that a more detailed analysis would reveal that the 2+√2 Alice’s side Bob’s side of quantum mechanics is a critical value that separates trivial from nontrivial communication complexity, which would give | {z β} | {z } Because Bob can construct and add his i values without re- an argument for Cirel’son’s nonlocality bound without refer- quiring any information from Alice, he can therefore compute ring to quantum mechanics. Recent work by Buhrman et al. ∑ β the value b i i by himself and broadcast this single bit to [4] shows that trivial communication complexity can indeed ≡ α Alice. She, on her part, creates the i values and finishes the be achieved with a correlation value strictly less than 4. Be- protocol with the errorless conclusion IP(x,y) b + ∑ αi. ≡ i cause there is a tight connection between the communication We just saw how the IP function has a communicationcom- complexity of distributed functions and the depth of circuits plexity of one bit for every finite input size N in the setting for these problems [11] it also possible to consider the im- of superstrong correlations. Hence, we can apply the reduc- plications of superstrong nonlocality for computational com- tion shown earlier to reach the result that any distributed de- plexity. cision problem f (~x,~y) can be exactly computed with a single bit of communication. Equation 4 tells us that we can rewrite the function f to f (~x,~y) ∑ Pi(~x) Qi(~y). As Bob can com- ≡ i · pute all the Qi values by himself, he and Alice can also re- motely and independently create the αi and βi values such that ∗ Email: [email protected] n αi + βi Pi(x) Qi(y) for all 1 i 2 . After the appropriate [1] A. Aspect, J. Dalibard, and G. Roger, “Experimental test of regrouping≡ of the· sum, Equation≤ 4≤ then becomes Bell’s inequalities using time-varying analyzers”, Physical Re- view Letters, 49:1804–1807, 1982. 2n 2n [2] J.S. Bell, “On the Einstein-Podolsky-Rosen paradox,” Physics, f (x1 xn,y1 yn) ∑ αi + ∑ βi . 1:195–200, 1964. ··· ··· ≡ i=1 ! i=1 ! [3] L. Babai, P.G. Frankl, and J. Simon, “Complexity classes in communication complexity theory”, Proceedings of the Alice’s side Bob’s side 27th IEEE Symposium on Foundations of Computer Science, | {z } | {z } pages 337–347, IEEE Computer Society Press, 1986. It should now be clear that Bob can compute the bit b ∑ βi ≡ i [4] H. Buhrman, R. Cleve, N. Linden, and F. Unger, in preparation. by himself and then communicate it to Alice who, just as for [5] B.S. Cirel’son, “Quantum generalizations of Bell’s inequality”, α the IP function, concludes with f (~x,~y) b + ∑i i. Letter in Mathematical Physics, 4:93–100, 1980. This finishes the proof that with the help≡ of the superstrong [6] J.F. Clauser, M.A. Horne, A. Shimony, and R.A. Holt, “Pro- correlations of Equation 2 any distributed function can be posed experiment to test local hidden-variable theories”, Phys- decided on Alice’s side without error after only one bit of ical Review Letters, 23:880–884, 1969. [7] R. Cleve and H. Buhrman, “Substituting quantum entanglement communication from Bob. It is true that in this protocol the for communication”, Physical Review A, 56(2):1201–1204, amount of resources (the super-correlated states) can grow ex- 1997; arXiv:quant-ph/9704026. ponentially with the input size n but this does not effect the [8] R. Cleve, W.van Dam, M. Nielsen, and A. Tapp, “Quantum En- conclusion that the communication complexity—after the in- tanglement and the Communication Complexity of the Inner puts are distributed—is minimal. Product Function”, Lecture Notes in Computer Science, Vol- We can now rephrase our original question as: why would ume 1509, pp. 71–74, 1998; arXiv:quant-ph/9708019. Nature not allow super-efficient distributed computing? It is [9] W. van Dam, Chapter 9 in “Nonlocality & Communication Complexity”, D.Phil. thesis, University of Oxford, Department not clear if there is a convincing answer to this, as it does not of Physics, 2000 seem to conflict with any physical intuition. However, trivial [10] S.J. Freedman and J.F. Clauser, “Experimental test of local communication complexity does prohibit the existence of an hidden-variable theories”, Physical Review Letters, 28:938– intrinsic ‘complexity’ for distributed tasks. Even though we 941, 1972. 4

[11] E. Kushilevitz and N. Nisan, Communication Complexity, Cam- arXiv:quant-ph/9605004 as “Action and Passion at a Distance: bride University Press, Cambridge, United Kingdom, 1997. An Essay in Honor of Professer Abner Shimony”. [12] S. Popescu and D. Rohrlich, “Quantum nonlocality as an ax- [14] D. Rohrlich and S. Popescu, “Nonlocality as an axiom for quan- iom”, Foundations of Physics, 24(3):379–385, 1994. tum theory”, in A. Mann and M. Revzen, editors, The dilemma [13] S. Popescu and D. Rohrlich, “The relativistic EPR argument”, of Einstein, Podolsky and Rosen, 60 years later: International in R.S. Cohen, M. Horne, and J.J. Stachel, editors, Potential- symposium in honour of Nathan Rosen, volume 12 of Annals of ity, entanglement and passion-at-a-distance: quantum me- the Israel Physical Society, chapter 16, Israel Physical Society, chanical studies for Abner Shimony, Volume Two, volume Haifa, Israel, 1996; arXiv:quant-ph/9508009. 194 of Boston studies in the philosophy of science, Kluwer Academic Publishers, Dordrecht, The Netherlands, 1997;