1 Introduction
Probabilistically Checkable Proofs (PCPs) [BFLS91, FGL96, AS98, ALM98]
are proofs that can be verified by a probabilistic verifier that queries only a few locations of the proof. PCPs have been a powerful tool in the theory of computing, with applications in diverse areas such as hardness of approximation
[FGL96] and delegation of computation [Kil92, Mic00]. A seminal result of [AS98, ALM98], known as the PCP theorem, says that every language decidable by a nondeterministic Turing machine in time
has a PCP system which allows to check if a given input of length is in the language by using random bits and making only queries to the given proof.Recall that in the classical setting of PCPs the two standard requirements are completeness and soundness. Completeness
requires that if a given input is in the language, then there is some proof that convinces the prover with probability
. Soundness requirement states that if the input is not in the language, the prover rejects any proof with some significant probability. In this paper we study PCP systems that are sound against nonsignalling proofs or nonsignalling strategies, i.e., we require the prover to reject any nonsignalling proof with some significant probability.Nonsignalling strategies are a certain restricted class of probabilistic oracles. When such oracle is given a set of queries, the response to the queries is sampled from a distribution such that the answer to each query may depend on all queries. More precisely, a non signalling strategy with locality is a collection , where each is a distribution over (i.e., over functions ), and for any two subsets of size at most , the restrictions of and to are equal as distributions. This setting stands in contrast to the standard notion of a classical proof, where the answer to each query is deterministic. Note that if the locality is the maximum possible, i.e., , then is a distribution over functions, which is (essentially) equivalent to the classical notion of a proof.
We note that one may think about nonsignalling functions, equivalently, as the class of all feasible solutions to the linear program arising from the ’th level relaxation of the SheraliAdams hierarchy [SA90]. This implies that computing the maximum acceptance probability of an nsPCP verifier that uses random bits, where the maximum is taken over all nonsignaling proofs, reduces to a linear program with variables and constraints. In particular, if a language has a PCP verifier that on an input of length uses random bits, and is sound against nonsignaling proofs over an alphabet of constant size, then is decidable in time .
Nonsignaling strategies have been studied in physics since 1980’s [Ras85, KT85, PR94] in order to better understand quantum entanglement. Indeed, these strategies strictly generalize quantum strategies and capture minimal requirements on “nonlocal” correlations that rule out instantaneous communication.
PCP systems that are sound against nonsignalling proofs have recently found numerous applications in theoretical computer science, including schemes for 1round delegation of computation from cryptographic assumptions [KRR13, KRR14], and hardness of approximation for linear programming [KRR16]. However, as opposed to the well studied setting of the classical PCP theorem, where there are many constructions achieving best parameters possible, in the nonsignalling setting many parameters of the known PCP constructions appear to be far from optimal.
One of the most important parameters associated with a nonsignalling proof is the locality parameter, denoted by . Indeed, [KRR13, KRR14] have studied the related notion of multiprover interactive proofs that are sound against nonsignaling strategies (nsMIPs). They have shown that nsPCPs are essentially equivalent to nsMIPs where , the locality of the proof in the nsPCP setting, exactly corresponds to the number of provers in the nsMIP setting.
Despite the importance of the locality parameter, the exact complexity of languages admitting nsPCPs that are sound against nonsignaling proofs is still open for most ’s. Note that as the locality of the proof decreases, there are fewer constraints imposed on the proof, and hence the task of the verifier becomes more challenging. The seminal result of Kalai, Raz, and Rothblum [KRR13, KRR14] showed that every language in has an nsPCP verifier that uses random bits, makes queries to a proof of length , and is sound against nonsignaling proofs. In particular, every language in is captured by a nsMIP with a polynomial time randomized verifier who communicates with nonsignaling provers. For the limitations of nsPCPs, Ito [Ito10] proved that for the corresponding linear program is solvable in , which is tight by the result of [IKM09], and hence the class is captured by PCPs that are sounds against nonsignaling proofs. Much less is known about the power of PCP systems that are sound against nonsignaling proofs for . Recently, Holden and Kalai [HK20] proved that prover nonsignalling proofs with negligible soundness is contained in .
All these results give rise to the following question, raised in [CMS19], asking for the nonsignaling analogue of the PCP theorem.
Question 1.1.
Is it true that every language in has an nsPCP verifier that uses random bits, makes queries to the proof, and is sound against nonsignalling functions?
Motivated by this problem, Chiesa et. al [CMS19, CMS20] started a systematic study of nonsignalling PCPs. They proposed studying the classical (algebraic) PCP constructions and their building blocks (which are very well understood in the classical setting), and adapting each of the building blocks to the nonsignaling setting. In particular, focusing on the PCP construction of [ALM98] they made an appropriate definition of linear nonsignalling functions and analyzed the linearity test of [BLR93] against nonsignalling strategies [CMS20]. Then, building on the linearity test, they proved in [CMS19] that the classical exponential length query PCP of [ALM98] is sound against nonsignalling proofs. We emphasize, that even for exponential length nsPCPs (corresponding to nsPCPs with randomness), there are no known constructions that are sound against nonsignaling proofs. Given this state of affairs, it is natural to ask the following question, that is simpler than creftype 1.1
Question 1.2.
Is it true that every language in has an nsPCP verifier that uses random bits, makes queries to the proof, and is sound against nonsignalling functions?
One must be careful with the precise formulation of creftype 1.2. Note that if the verifier uses more than random bits, the runtime spent on reading the randomness is more than , which is the time complexity of the problem. To recover a nontrivial question, we require the verifier to be input oblivious. That is, in order to decide whether an instance belongs to the given language , the verifier generates the queries based only on the length of the input and its randomness (but not the input itself), and then rules according to an time decision predicate (where the predicate does depend on ). Indeed, the [ALM98] verifier studied in [CMS19] is input oblivious.
In this work we build on the work of [CMS19] and provide a positive answer to creftype 1.2 assuming a certain geometric hypothesis. Specifically, we construct an input oblivious nsPCP verifier for any language that uses random bits, makes queries to a given proof, and is sound against nonsignalling functions, with two caveats.

The first is that the alphabet of the nsPCP system is , instead of the binary alphabet employed by [CMS19, KRR14, ALM98]. Still, this means that the verifier reads a total of bits from the proof, which makes our result nontrivial. Also, recall that in the classical setting, we have the alphabet reduction technique using proof composition, and it is plausible that we can apply similar ideas also in the nonsignaling setting. Indeed, proof composition is an important building block in the classical PCP literature, and we believe it will also be an important step toward resolving creftype 1.1.

The second caveat is that our result depends on a certain quantitative geometric hypothesis about proximity between almost nonsignaling proofs and exactly nonsignaling proofs. Equivalently, the hypothesis says that every feasible solution for the noisy SheraliAdams LP is close (in some precise, rather weak, sense) to a feasible solution for the (exact) SheraliAdams LP. See creftype 2 for details, and the discussion in Appendix A.
Our work follows the general philosophy of [CMS19, CMS20], who proposed building modular analogues of tools and techniques from the classical PCP literature. A classical tool used in the construction of PCPs is parallel repetition [Raz98, Hol09]. In the classical setting of 2query PCP, parallel repetition is used to reduce the soundness error. In this work we use parallel repetition for nonsignalling proofs to reduce the locality to , while the soundness stays in the “highprobability acceptance regime”. In addition to parallel repetition, we study additional tools from the PCP literature. Specifically, we use the modular approach that is typical for the classical setting. Specifically, we show first that the parallel repetition of the [ALM98] verifier is sound against “nicely structured” proofs. Then, we use linearity test and direct product test, and claim that proofs that satisfy both tests with high probability must be nicely structured, and hence we essentially reduce the analysis to the structured case.
Another interesting feature of our proof is the reduction from the parallel repetition of the [ALM98] verifier to the nonrepeated [ALM98] verifier. Specifically, we show that if for some input , the parallel repetition of the [ALM98] verifier accepts a proof with high probability, and the proof is “nicely structured”, then it is possible to “flatten” the repeated proof into a proof over the binary alphabet, that satisfies the (nonrepeated) [ALM98] verifier with high probability. Therefore, by applying the result of [CMS19] about the soundness of the [ALM98] verifier, we conclude that the input is in the language.
1.1 Informal statement of the result
Below we discuss the main result of the paper. Our result depends on an hypothesis about approximating almost nonsignaling functions using exactly nonsignaling functions.
Hypothesis 1 (Informal).
Any almost linear, almost nonsignaling function can be well approximated by some nonsignaling function of slightly lower locality.
Equivalently, any solution to the noisy SheraliAdams LP can be well approximated by a solution to the (exact) SheraliAdams LP of slightly lower level in the hierarchy.
The exact formulation of the hypothesis relies on the precise definitions of nonsignaling and almost nonsignaling functions (or, equivalently, the related notions of noisy SheraliAdams LP), as well as the appropriate definitions of distance. For the formal statement of the hypothesis see creftype 2 following the required definitions in Section 2.
We are now ready to state our main theorem.
Theorem 1 (Main theorem  informal).
Assuming creftype 1 every language has an input oblivious nsPCP verifier that on an input of length uses random bits, makes queries to proofs over the alphabet , and is sound against nonsignaling proofs. The query sampler runs in time , and the decision predicate runs in time .
To the best our knowledge, this is the first result that constructs a PCP system that is sound against nonsignaling proofs with constant locality.
1.2 Roadmap
The rest of the paper is organized as follows. In Section 2 we formally define the notions that we utilize throughout this work, and use them to formally state our hypothesis and the main theorem in Section 3. In Section 4 we recall the ALMSS verifier, and define our variant of its parallel repetition. In Section 5 we provide an overview of the soundness proof. In Section 6 we prove soundness of our verifier against structured proofs. In Section 7 we discuss our local testing and selfcorrection, which enables us to reduce soundness against general proofs to soundness against structured proofs. Finally, in Section 8 we prove the main result.
2 Preliminaries
2.1 Probabilistically Checkable Proofs
We start with the definition of Probabilistically Checkable Proofs (PCPs). Recall that a classical PCP verifier for a language is given an input , and an oracle access to a proof. The verifier reads the input, uses randomness, queries the proof in a small number of coordinates, and based on the answers to the queries decides whether to accept or reject. Completeness requires that if , then there exists a proof that makes the verifier always accept. Soundness requires that if , then for any proof the verifier will reject with high probability.
In the nonsignaling setting, a nonsignaling PCP verifier is a verifier, whose soundness is further required to hold against any nonsignaling proof of prescribed locality. More precisely, an nsPCP verifier for a language gets an input and an oracle access to a nonsignaling function . The verifier reads the input , uses random bits to decide on a subset on which is queried. Then, based on the answer it decides to accept or reject.
Definition 2.1.
A nsPCP verifier for a language is a randomized algorithm that gets an input and oracle access to a nonsignaling proof . The verifier uses randomness to decide on a subset of size , and queries on . Then, based on the answer it decides to accept or reject. We say that has perfect completeness and soundness error against nonsignaling proofs if the following holds.
 Completeness:

For all there exists a (classical) proof such that .
 Soundness:

If , then for all nonsignaling proofs it holds that .
We say that verifier is input oblivious if the choice of the query set depends only on the input length , the randomness of the verifier, but is independent of .
Remark 2.2.
Note that in the nonsignaling setting the locality parameter upper bounds the number of queries made by the verifier, and it is possible that the actual predicate used by the verifier depends on significantly less than coordinates of the proof. For example, [CMS19] proved that the 11queries verifier of [ALM98] is sound against nonsignaling proofs, and it is not known whether the verifier is sound against nonsignaling proofs, or even nonsignaling proofs.
2.2 Parallel repetition
In the classical setting a proof is assumed to be a string, or equivalently, a static function committed by the prover. A parallel repetition of a proof is a mapping that allows accessing locations of the (supposed) proof by making only 1 query to a (longer) proof over a larger alphabet. That is, the intended proof corresponds to some “base” proof defined as . Analogously, given a verifier , a repeated verifier which is denoted by , runs parallel independent instances of and accepts if and only if all instances accept.
The original motivation for using parallel repetition was to reduce the soundness error of a proof system, while keeping the number of queries fixed. In the classical setting, if the repeated proof is indeed a parallel repetition of some base proof , then it is not hard to see that the soundness error of is exponentially smaller than the soundness error of . The soundness analysis of the repeated proof need not be based on this comparison to the soundness error of the base proof, and analyzing such proofs in both classical and nonsignalling settings has been a subject of a long line of research [Ver96, Raz98, Hol09, DS14a, BG15, LW16, HY19].
In this work, we use parallel repetition to improve the minimum locality parameter of nonsignaling proofs required for the soundness of the verifier, rather than its soundness error. Next, we formally define nonsignaling proofs, and some properties of such proofs that we will need in the paper.
2.3 Nonsignaling functions
In this work we consider PCP verifiers that are sound against nonsignaling proofs. Below, we formally define the notion of nonsignaling functions, and introduce some notation we will use in the paper. Throughout the paper we will use terms nonsignaling function, nonsignaling proof, and nonsignaling strategy interchangeably.
Definition 2.3.
Fix a domain , an alphabet , and a parameter . A nonsignaling function is a collection , where each is a distribution over assignments , such that for every two subsets each of size at most , the marginal distributions of and restricted to are equal.
Unlike a classic function, we can use a nonsignaling function only once in the sense that one has to present the set of at most queries all at once. In other words, it is not possible to use the nonsignaling function adaptively.
Remark 2.4.
Throughout the paper we will consider nonsignaling functions of two types:

functions over the domain for some and alphabet ;

functions over the domain and alphabet for some parameters .
Next, we define a relaxed notion of nonsignaling functions, that allows the marginal distributions induced by different query sets to be only statistically close rather equal on the intersection. This relaxation arises in our analysis. It has also appeared naturally in other works in this area, especially in cryptographic applications [ABOR00, DLN04, KRR13, KRR14].
Definition 2.5.
Fix a domain , an alphabet , and parameters and . A nonsignaling function over a domain and an alphabet , is a collection , where each is a distribution over assignments , such that for every two subsets each of size at most , the marginal distributions of and restricted to are close with respect to total variation distance, i.e.,
In particular, a nonsignalingfunction coincides with the definition of nonsignaling function from Definition 2.3.
Next we define nonsignaling and almost nonsignaling counterpart of parallel repeated functions.
Definition 2.6.
Fix a domain , an alphabet , and parameters . A repeated nonsignaling function is an nonsignaling function . Namely, a repeated nonsignaling function is a collection , where each is a distribution over assignments , such that for every two subsets each of size at most , the marginal distributions of and restricted to are close with respect to total variation distance.
We will also need the definition of distance between nonsignaling or almost nonsignaling functions.
Definition 2.7 (Statistical distance).
Let be two nonsignaling or almost nonsignaling functions with locality . For the distance between and is defined as
where is the total variation distance between and .
We say that and are close in the distance if , and say that they are far otherwise.
2.4 Permutation folded repeated nonsignaling functions
Folding is a technique used to impose some structure on the given proof without really making extra queries. The idea of using folded proofs was first introduced by [BGS98]. We formally define the permutation folding property, and then explain why we can impose this property without making extra queries.
Definition 2.8.
Let be a
values vector, and let
be a permutation of the indices . Define to be the vector obtained from by permuting the coordinates according to .Let be a repeated nonsignaling function. is said to be permutation folded or permutation invariant if for any with , for any for some permutations , and for any it holds that
Observation 2.9.
It is important to note that we can fold any given repeated nonsignaling function by partitioning into equivalence classes, where and belong to the same class if for some permutation .
We defined the folding of , denoted by as follows. For any query to , let be a uniformly random permutation, and define the distribution of as the distribution of .
It is easy to see that is indeed nonsignaling and permutation folded. Furthermore, note that if is permutation folded, then .
2.5 Linear nonsignaling functions
In this part, we define linear repeated nonsignaling functions. Linear nonsignaling boolean functions have been studied in [CMS20, CMS19], and played a key role in the proving that the PCP verifier of [ALM98] is sound against nonsignaling proofs. We also use such structured nonsignaling proofs in this paper. See Section 4 for details.
Definition 2.10 (Linear repeated functions).
Let be a repeated nonsignaling function. We say that is linear if for all , and defined by the coordinatewise addition modulo 2, and for all containing of size at most , it holds that
Remark 2.11.
Note that in the degenerate case of if a (nonrepeated) nonsignaling function satisfies the linearity condition in Definition 2.10 then for all , i.e., satisfies the linearity test of [BLR93] with probability 1. Nonsignaling functions satisfying this property have been the subject of work on linearity testing in the nonsignaling setting [CMS20].
Next we extend Definition 2.10 by introducing the notion of an almost linear repeated nonsignalling function.
Definition 2.12 (Almost linear repeated functions).
Let be a repeated nonsignaling function. We say that is linear if for all , and defined by the coordinatewise addition modulo 2, and for all containing of size at most , it holds that
We will allow ourselves to use the informal term almost linear, when referring to a nonsignaling function that is linear for some small .
2.6 Consistent repeated nonsignaling functions
In this part, we define the notion of consistency for repeated nonsignaling function.
Definition 2.13 (Consistent repeated functions).
Let be a repeated nonsignaling function. We say that is consistent, if for any it holds that
Similarly to the almost linear property, we define the relaxed notion of almost consistent nonsignalling function.
Definition 2.14 (Almost consistent repeated functions).
Let be a repeated nonsignaling function. We say that is consistent, if for any
We will allow ourselves to use the informal term almost consistent, when referring to a nonsignaling function that is consistent for some small .
Claim 2.15.
Let be a repeated nonsignaling function for , and suppose that is consistent. Fix and let . Then, for any event it holds that
Proof.
Note that
where the last inequality is by the assumption that is consistent. By symmetry, we also get the inequality in the other direction, and the claim follows. ∎
We observe that for and (almost) linearity implies (almost) consistency. Specifically, we prove the following claim.
Claim 2.16.
Let be a repeated nonsignaling function, and suppose that (i) is linear, and (ii) for all . Then, is consistent when treated as a nonsignaling function.
Proof.
Let be a set of queries of size . Let , and let . We show below that
Consider the set of queries , where . In particular, for all . By the assumption of the claim we get that . Therefore, using the assumption that is linear it follows that
Therefore, is consistent, as required. ∎
2.7 Flattening of a nsPCP
Below we define the flattening operation, which transforms a given repeated proof into a nonrepeated proof in the natural way. Namely, given a query set to the nonrepeated proof, we create a vector containing all the elements of , query the repeated proof on , and respond according to the received answer.
Definition 2.17.
Let be a nonsignaling repeated proof. Define the flattening of , denoted by as follows. For a query set of size , define a vector whose first entries are and the rest are set arbitrarily, query on the single query , and let the distribution of be
Claim 2.18.
Let be a nonsignaling function that is permutation folded and consistent for . Then is a nonsignaling function.
Furthermore, fix a query for , a query set of size for , also let such that are distinct and for all . Then, the distribution of and are close in total variation distance.
Proof.
To prove that is nonsignaling function let be two sets of queries, and suppose . We want to show that for any event it holds that
(1) 
Define as in Definition 2.17, let be permutations such that for all it holds that . By nonsignaling and permutation invariance of , if we query it on we have:
Then, by creftype 2.15 we get the following:
which proves Eq. 1. Therefore, is a nonsignaling function.
Next we prove the second part of the claim. Given , define as in Definition 2.17, and consider the query set to . Since is permutation folded, we may assume that for all . Therefore, for any we have:
which is upper bounded by by creftype 2.15. This complets the proof of creftype 2.18 ∎
The following claim is follows rather immediately from creftype 2.18 above.
Claim 2.19.
Let , and let be a nonsignaling function that is permutation folded, linear, and consistent. Then is a (nonrepeated) nonsignaling linear function.
Proof.
By applying creftype 2.18 on , we get that is a nonsignaling function. Next we prove that is linear. Fix , and let be a query set for such that . We want to prove that
(2) 
Let be as in Definition 2.17. By the permutation folding property of we may assume that the first three coordinates of are . That is , and .
By definition of we have . Consider now the vectors , , and . Since is linear, we get that . Since is consistent, it follows that
as required. ∎
3 Main result
In this section we formally state the main result of the paper. In order to describe the result we need to first state the hypothesis conditioned on which our main theorem holds.
Hypothesis 2.
Fix integers and , and let . For any almost nonsignaling function that is linear there exists a nonsignaling function such that , where for some positive absolute constant , and is some function that depends only on such that as .
Remark 3.1.
We make two remarks regarding the hypothesis.

A statement analogous to creftype 2 has been proven in [CMS20], showing that there exist a nonsignaling function such that . The multiplicative factor of is too large, which makes it insufficient for our applications.

For our applications, we need a much weaker version of creftype 2. We elaborate more on the hypothesis in Appendix A.
For a computable function we denote by the complexity class of all languages having a uniform family of boolean circuits of maximum fanin 2 with AND, OR, and NOT gates, such that has at most wires for all .^{1}^{1}1Note that our complexity measure for the size of a circuit is the number of wires, (and not the number of gates, which is more standard) as this measure directly affects the complexity of the PCP construction. However, for circuits with bounded fanin, the two quantities are equal up to a multiplicative constant factor.
Theorem 2 (Main theorem).
Assuming creftype 2 every language has an input oblivious nsPCP verifier that on input of length uses random bits, makes queries to proofs over the alphabet , and is sound against nonsignaling proofs. The query sampler runs in time , and the decision predicate runs in time . That is,
It is clear that creftype 1 follows from creftype 2 since .
4 The PCP construction
In this section we formally describe our PCP construction. In one sentence, the PCP verifier gets a permutation invariant proof , runs on it linearity test, direct product test, and the parallel repetition of the ALMSS verifier, and accepts if and only if all tests accepts.
We start by recalling the setting of the PCP verifier of [ALM98] (the “linear ALMSS verifier”). Let be a language, and let be a uniform family of boolean circuits with wires that decides . That is, for all inputs of length it holds that if and only if .
For a given length let be the circuit corresponding to the computation on inputs of length . The computation of on the input is viewed as a system of constraints over boolean variables , where are quadratic polynomials (each involving at most three variables in ) and are boolean constants. Each variable represents the value of one of the wires of during the computation on the input . In particular, the first variables, , correspond to the input wires, and the variable corresponds to the output wire. The constraints are of three types:
 Input consistency:

For every , and .
 Gate consistency:

For every ,

If the wire represented by the variable is an output of an AND gate , where the inputs to are given by , then and .

If the wire represented by the variable is an output of an OR gate , where the inputs to are given by , then and .

If the wire represented by the variable is an output of a NOT gate , where the input to is given by , then and .

 Accepting output:

and .
We overload notation, and use to also denote the upper triangular matrix in with
Comments
There are no comments yet.