We show that the maximum success probability of players sharing quantum entanglement in a two-player game with classical questions of logarithmic length and classical answers of constant length is NP-hard to approximate to within constant factors. As a corollary, the inclusion $\mathrm{NEXP}\subseteq\mathrm{MIP}^*$, first shown in [IV12] with three provers, holds with two provers only. The proof is based on a simpler, improved analysis of the low-degree test Raz and Safra (STOC'97) against two entangled provers.
We show that the maximum success probability of players sharing quantum entanglement in a two-player game with classical questions of logarithmic length and classical answers of constant length is NP-hard to approximate to within constant factors. As a corollary, the inclusion NEXP⊆MIP∗, first shown in [IV12] with three provers, holds with two provers only. The proof is based on a simpler, improved analysis of the low-degree test Raz and Safra (STOC'97) against two entangled provers.
We show that for any ε>0 the problem of finding a factor (2−ε) approximation to the entangled value of a three-player XOR game is NP-hard. Equivalently, the problem of approximating the largest possible quantum violation of a tripartite Bell correlation inequality to within any multiplicative constant is NP-hard. These results are the first constant-factor hardness of approximation results for entangled games or quantum violations of Bell inequalities shown under the sole assumption that P=NP. They can be thought of as an extension of H\aastad's optimal hardness of approximation results for MAX-E3-LIN2 [J. ACM, 48 (2001), pp. 798--859] to the entangled-player setting. The key technical component of our work is a soundness analysis of a plane-vs-point low-degree test against entangled players. This extends and simplifies the analysis of the multilinearity test by Ito and Vidick [Proceedings of the 53rd FOCS, IEEE, Piscataway, NJ, 2012, pp. 243--252]. Our results demonstrate the possibility of efficient reductions between entangled-player games and our techniques may lead to further hardness of approximation results.
We show that for any eps>0 the problem of finding a factor (2-eps) approximation to the entangled value of a three-player XOR game is NP-hard. Equivalently, the problem of approximating the largest possible quantum violation of a tripartite Bell correlation inequality to within any multiplicative constant is NP-hard. These results are the first constant-factor hardness of approximation results for entangled games or quantum violations of Bell inequalities shown under the sole assumption that P \neq NP. They can be thought of as an extension of Hastad's optimal hardness of approximation results for MAX-E3-LIN2 (JACM'01) to the entangled-player setting. The key technical component of our work is a soundness analysis of a point-vs-plane low-degree test against entangled players. This extends and simplifies the analysis of the multilinearity test by Ito and Vidick (FOCS'12). Our results demonstrate the possibility for efficient reductions between entangled-player games and our techniques may lead to further hardness of approximation results.
The complexity of free games with two or more classical players was essentially settled by Aaronson, Impagliazzo, and Moshkovitz (CCC’14). In the quantum world, there are two complexity classes that can be considered quantum analogues of classical free games: (1) AM*, the multiprover interactive proof class corresponding to free games with entangled players, and, somewhat less obviously, (2) BellQMA(2), the class of quantum Merlin-Arthur proof systems with two unentangled Merlins, whose proof states are separately measured by Arthur. In this work, we make significant progress towards a tight characterization of both of these classes. (1) We show a BellQMA(2) protocol for 3SAT on n variables, where the total amount of communication is Õ(√n). This answers an open question of Chen and Drucker (2010) and also shows, conditional on ETH, that the algorithm of Brandão, Christandl and Yard (STOC’11) for optimizing over separable states is tight up to logarithmic factors. (2) We show that AM* with nprovers = 2, question length O(1), and answer-length log(n) is equal to RE, i.e. that free entangled games with constant-sized questions are as powerful as general entangled games. (In contrast, Aaronson, Impagliazzo and Moshkovitz show that classical free games are much weaker than general classical games.) We show this using a question “hyper-compression” theorem that iteratively applies the introspection technique of Ji et al. (2020). Our result is a significant improvement over the headline result of Ji et al., whose MIP* protocol for the halting problem has (n)-sized questions and answers. (3) By the same techniques, we obtain a zero-gap AM* protocol for a Π2 complete language with constant-size questions and almost logarithmically (O(logn · log* n)) large answers, improving on the headline result of Mousavi, Nezhadi and Yuen (STOC’22). (4) Using a connection to the nonuniform complexity of the halting problem we show that any MIP* protocol for RE requires Ω(logn) bits of communication. It follows that our results in item 3 are optimal up to an O(log* n) factor, and that the gapless compression theorems of Mousavi, Nezhadi and Yuen are asymptotically optimal. We conjecture that these bounds can be saturated in the gapped case as well.
We show that given an explicit description of a multiplayer game, with a classical verifier and a constant number of players, it is QMA-hard, under randomized reductions, to distinguish between the cases when the players have a strategy using entanglement that succeeds with probability 1 in the game, or when no such strategy succeeds with probability larger than 1/2. This proves the “games quantum PCP conjecture” of Fitzsimons and the second author (ITCS'15), albeit under randomized reductions. The core component in our reduction is a construction of a family of two-player games for testing n-qubit maximally entangled states. For any integer n ≥ 2, we give such a game in which questions from the verifier are O(log n) bits long, and answers are poly(loglogn) bits long. We show that for any constant ε ≥ 0, any strategy that succeeds with probability at least 1 - ε in the test must use a state that is within distance δ(ε) = O(εc) from a state that is locally equivalent to a maximally entangled state on n qubits, for some universal constant c > 0. The construction is based on the classical plane-vs-point test for multivariate low-degree polynomials of Raz and Safra (STOC'97). We extend the classical test to the quantum regime by executing independent copies of the test in the generalized Pauli X and Z bases over Fq, where q is a sufficiently large prime power, and combine the two through a test for the Pauli twisted commutation relations. Our main complexity-theoretic result is obtained by combining this family of games with techniques from the classical PCP literature. More specifically, we use constructions of PCPs of proximity introduced by Ben-Sasson et al. (CCC'05), and crucially rely on a linear property of such PCPs. Another consequence of our results is a deterministic reduction from the games quantum PCP conjecture to a suitable formulation of the constraint satisfaction quantum PCP conjecture.
Quantum entanglement is known to provide a strong advantage in many two-party distributed tasks. We investigate the question of how much entanglement is needed to reach optimal performance. For the first time we show that there exists a purely classical scenario for which no finite amount of entanglement suffices. To this end we introduce a simple two-party nonlocal game H, inspired by Lucien Hardy's paradox. In our game each player has only two possible questions and can provide bit strings of any finite length as answer. We exhibit a sequence of strategies which use entangled states in increasing dimension d and succeed with probability 1−O(d−c) for some c≥0.13. On the other hand, we show that any strategy using an entangled state of local dimension d has success probability at most 1−Ω(d−2). In addition, we show that any strategy restricted to producing answers in a set of cardinality at most d has success probability at most 1−Ω(d−2). Finally, we generalize our construction to derive similar results starting from any game G with two questions per player and finite answers sets in which quantum strategies have an advantage.
Forty years later, a group of computer scientists, Richard Cleve, Peter Høyer, Ben Toner and John Watrous [1], provided a computer science perspective on Bell inequalities in the framework of multi-player one-round games. In a multi-player one-round game, a referee samples and asks questions to multiple non-communicating players and checks the validity of the answers according to some rule. The classical value of the game is the maximum success probability with which the players can pass the referee's check. Multi-player one-round games form a well-studied topic in theoretical computer science and have connections to constraint satisfaction problems, interactive proof systems, probabilistically checkable proof (PCP) and the celebrated PCP theorem. Cleve et al. reformulated Bell inequalities as multi-player one-round games such that the players can achieve a higher success probability, called the quantum value, when they measure a shared entangled state to obtain the answers. They also pointed out that shared entanglement may cause the so-called soundness problem in quantum interactive proof systems, an intriguing phenomenon that leads to a series of investigations on the limitation and power of entanglement in multi-player games. It is well known that it is NP-hard to approximate the classical value of a game to inverse polynomial precision where NP stands for nondeterministic polynomial time. Just how hard is it to approximate the quantum value of a game? We do not know the answer after a decade of research efforts, mainly because the players measure on an arbitrarily high dimensional quantum system. Because of the soundness problem pointed out by Cleve et al., we did not even know how to establish that the quantum value is NP-hard until Kempe et al. finally proved it in 2007 [2]. Very recently, the author established that the quantum value of a game is NEXP-hard to approximate [3,4] where NEXP stands for nondeterministic exponential time. This can be seen as a complexity-theoretic Bell inequality. The standard Bell inequality tells us that the classical and quantum values can be different for a fixed game; our result instead establishes that the computational complexities for approximating the classical and quantum values are different, since NP ≠ NEXP. The complexity of Bell inequalities or the quantum values of multi-player one-round games has deep connections to quantum physics and theoretical computer science as we have already discussed. It is also related to the Connes embedding conjecture in mathematics. Using a group theory approach, Slofstra proved that the problem of determining whether a game has quantum value 1 or not is undecidable [5]. Is the quantum value approximation problem also undecidable or can we establish any complexity upper bound? Only time will tell.