Khot–Vishnoi Bell Game: Structure & Violation

• The Khot–Vishnoi Bell game is an explicit two-player nonlocal game defined on the Boolean cube using cosets of a Hadamard subgroup, derived from Unique Games.
• It contrasts classical strategies, with a winning probability of order 1/n, against entangled strategies achieving probabilities on the order of 1/(log n)².
• Its spectral and Fourier analytic structure not only facilitates tight Bell-violation estimates but also motivates methods for input compression and parallel repetition.

Searching arXiv for the cited Khot–Vishnoi Bell-game papers and related sources. Searching for "Khot-Vishnoi Bell game". The Khot–Vishnoi Bell game is an explicit two-player nonlocal game derived from the Khot–Vishnoi integrality-gap framework for Unique Games. In its standard formulation, the game is parameterized by an integer $n$ that is a power of $2$ and a noise parameter $\eta\in[0,1/2]$. It is played on cosets of the Hadamard subgroup of the Boolean cube $(\{0,1\}^n,\oplus)$, with Alice receiving a uniformly random coset and Bob receiving a noisy translate of that coset. Each player outputs an element of the received coset, and they win when the xor of the outputs equals the sampled noise string. The game is significant because it realizes a near-linear separation between entangled and classical performance: for a suitable choice of $\eta$, the entangled winning probability is of order $1/(\log n)^2$, while the classical winning probability is of order $1/n$, yielding Bell violation $\Omega(n/(\log n)^2)$ (Buhrman et al., 2010).

1. Definition and ambient structure

The standard Khot–Vishnoi game, denoted $\mathrm{KV}_n$, is defined on the Boolean cube/group

$(\{0,1\}^n,\oplus),$

with bitwise addition mod $2$0, where $2$1 is assumed to be a power of $2$2 (Buhrman et al., 2010).

A central ingredient is the Hadamard subgroup

$2$3

where

$2$4

This subgroup has size $2$5 and partitions $2$6 into

$2$7

cosets, each of cardinality $2$8 (Buhrman et al., 2010). The question set for each player is precisely the set of these cosets, so the number of possible inputs to each player is $2$9, while the number of possible outputs per player is $\eta\in[0,1/2]$0 (Buhrman et al., 2010).

The input distribution is specified by sampling $\eta\in[0,1/2]$1 uniformly and a noise string $\eta\in[0,1/2]$2 whose bits are independently equal to $\eta\in[0,1/2]$3 with probability $\eta\in[0,1/2]$4. Alice receives

$\eta\in[0,1/2]$5

and Bob receives

$\eta\in[0,1/2]$6

Alice must output an element $\eta\in[0,1/2]$7, Bob an element $\eta\in[0,1/2]$8, and they win iff

$\eta\in[0,1/2]$9

The formulation is therefore a two-player nonlocal game closely related to a Unique Games instance rather than an XOR game (Buhrman et al., 2010).

The source also notes a technical nuance: for a fixed pair of cosets $(\{0,1\}^n,\oplus)$0, there are $(\{0,1\}^n,\oplus)$1 possible strings $(\{0,1\}^n,\oplus)$2, so the standard presentation is technically a game with a randomized predicate. It is remarked that one can modify it to a deterministic-predicate game, but that modification is not needed for the main results (Buhrman et al., 2010).

2. Unique Games origin and Khot–Vishnoi construction

The game is directly based on the Khot–Vishnoi integrality-gap instance for Unique Games (Buhrman et al., 2010). In the Unique Games framework, an instance consists of a constraint graph $(\{0,1\}^n,\oplus)$3, an alphabet of size $(\{0,1\}^n,\oplus)$4, variables $(\{0,1\}^n,\oplus)$5, and edge permutations $(\{0,1\}^n,\oplus)$6, with an edge $(\{0,1\}^n,\oplus)$7 satisfied when

$(\{0,1\}^n,\oplus)$8

The spectral formulation represents the instance as $(\{0,1\}^n,\oplus)$9, where $\eta$0 is the adjacency matrix of the label-extended graph (Kolla, 2011).

For the Khot–Vishnoi instance, the construction is expressed both in $\eta$1 and $\eta$2 language. In the $\eta$3 formulation, one sets

$\eta$4

with addition mod $\eta$5, identifies $\eta$6 with the Hadamard code on $\eta$7 bits, and views the constraint graph $\eta$8 as a Cayley graph of the quotient group

$\eta$9

Its vertices are the cosets of $1/(\log n)^2$0, hence $1/(\log n)^2$1 in number, matching the input set of the Bell game (Kolla, 2011).

The associated label-extended graph $1/(\log n)^2$2 has vertex set $1/(\log n)^2$3 in the $1/(\log n)^2$4 model and is described as an “$1/(\log n)^2$5-perturbed” hypercube/noise graph: the weight between $1/(\log n)^2$6 and $1/(\log n)^2$7 is proportional to the probability of reaching $1/(\log n)^2$8 from $1/(\log n)^2$9 by flipping each bit independently with probability $1/n$0 (Kolla, 2011). This hypercube-noise structure is one of the reasons the Khot–Vishnoi instance is unusually transparent to Fourier analysis.

The relation between the Unique Games instance and the Bell game is explicit in the literature summarized here: the Khot–Vishnoi Bell game is built from the same Unique Games instance, and deterministic classical strategies correspond to labelings, with shared randomness only convexifying over them. This suggests that certification of low satisfiability for the Unique Games instance transfers directly to certification of low classical winning probability for the associated nonlocal game (Kolla, 2011).

3. Classical value and its analysis

The key classical theorem for the Khot–Vishnoi game states that for every $1/n$1 that is a power of $1/n$2, and every $1/n$3, every classical strategy for $1/n$4 has winning probability at most

$1/n$5

This is the main upper bound on the local value $1/n$6 (Buhrman et al., 2010).

The proof reduces without loss of generality to deterministic strategies. One defines indicator functions

$1/n$7

by declaring $1/n$8 iff Alice, on coset $1/n$9, outputs the element $\Omega(n/(\log n)^2)$0, and similarly for Bob. Since each player selects exactly one representative from each coset,

$\Omega(n/(\log n)^2)$1

The winning probability can then be rewritten as a noisy correlation: $\Omega(n/(\log n)^2)$2 Introducing the Boolean noise operator $\Omega(n/(\log n)^2)$3, one has

$\Omega(n/(\log n)^2)$4

Applying Cauchy–Schwarz and the Bonami–Beckner inequality yields the bound

$\Omega(n/(\log n)^2)$5

The same derivation underlies later repetition results and is one of the standard analytic signatures of the game (Buhrman et al., 2010).

A related perspective comes from the Khot–Vishnoi Unique Games instance itself. The summary of the spectral-algorithm literature quotes the original Khot–Vishnoi facts that every labeling satisfies at most a

$\Omega(n/(\log n)^2)$6

fraction of the total edge weight, while the standard SDP relaxation has objective value greater than

$\Omega(n/(\log n)^2)$7

This sharp discrepancy is the integrality-gap phenomenon that motivates the Bell-game construction (Kolla, 2011). In Bell-game language, it means that the true classical value is very small even though the standard SDP relaxation appears nearly satisfiable. The paper summarizing the spectral method does not make a quantum-value statement, so no entangled-value bound should be attributed to it; its contribution is instead a certification of low classical optimum for the underlying Khot–Vishnoi object (Kolla, 2011).

4. Entangled strategy and Bell violation

The entangled lower bound is equally explicit. For every $\Omega(n/(\log n)^2)$8 that is a power of $\Omega(n/(\log n)^2)$9, and every $\mathrm{KV}_n$0, there exists an entangled strategy that wins $\mathrm{KV}_n$1 with probability at least

$\mathrm{KV}_n$2

using a maximally entangled state of local dimension $\mathrm{KV}_n$3 (Buhrman et al., 2010).

The strategy associates to each $\mathrm{KV}_n$4 the vector

$\mathrm{KV}_n$5

These are unit vectors satisfying

$\mathrm{KV}_n$6

A crucial structural fact is that the $\mathrm{KV}_n$7 vectors $\mathrm{KV}_n$8, as $\mathrm{KV}_n$9 ranges over a coset of $(\{0,1\}^n,\oplus),$0, form an orthonormal basis of $(\{0,1\}^n,\oplus),$1 (Buhrman et al., 2010). Every question therefore determines an orthonormal basis, and the local measurements are simply projective measurements in those bases.

For the maximally entangled state of local dimension $(\{0,1\}^n,\oplus),$2, the joint output probability obeys

$(\{0,1\}^n,\oplus),$3

If $(\{0,1\}^n,\oplus),$4, then success requires $(\{0,1\}^n,\oplus),$5, and the winning probability for fixed inputs becomes

$(\{0,1\}^n,\oplus),$6

Averaging over the Bernoulli noise gives

$(\{0,1\}^n,\oplus),$7

by Jensen’s inequality (Buhrman et al., 2010).

The standard parameter choice is

$(\{0,1\}^n,\oplus),$8

Then

$(\{0,1\}^n,\oplus),$9

so the entangled winning probability is

$2$00

whereas the classical value becomes

$2$01

up to lower-order factors. Consequently,

$2$02

This is the game’s canonical Bell-violation estimate (Buhrman et al., 2010).

The same source emphasizes why the construction is regarded as near-optimal. Known upper bounds show that with local entanglement dimension $2$03, Bell violation is at most $2$04, and with at most $2$05 outputs per player, Bell violation is at most $2$06. Since the Khot–Vishnoi game has both local dimension $2$07 and $2$08 outputs per player, the factor $2$09 is within a polylogarithmic factor of the optimal $2$10 (Buhrman et al., 2010).

5. Spectral and harmonic-analytic structure

The Khot–Vishnoi construction is unusually amenable to spectral analysis. In the label-extended graph $2$11, the eigenvectors are the characters of the group $2$12, and the eigenvalue corresponding to $2$13 is

$2$14

where $2$15 is the Hamming weight of $2$16, with multiplicity

$2$17

Equivalently, the operator is exactly the Boolean noise operator, up to scaling by $2$18 (Kolla, 2011).

This Fourier/noise-operator structure explains why the Khot–Vishnoi instance can be hard for SDP yet tractable for full-spectrum methods. The spectral algorithm literature defines

$2$19

where $2$20 for the Khot–Vishnoi instance. Choosing $2$21, the relevant Hamming weights satisfy $2$22, and the dimension of the near-top eigenspace is bounded by

$2$23

where $2$24 is the size of the lifted hypercube (Kolla, 2011). This yields a quasi-polynomial-time algorithm that certifies high unsatisfiability of the Khot–Vishnoi instance even though the standard SDP relaxation fails (Kolla, 2011).

For Bell-game purposes, the important consequence is not a new entangled-value estimate but a structural one: the underlying classical game can be analyzed through the full low-degree Fourier spectrum of the hypercube. This suggests that the Khot–Vishnoi Bell game sits at a particularly fertile intersection of nonlocal games, Fourier analysis on $2$25, and Unique Games integrality gaps.

A comparison with general Bell-game background is useful here. A later primer on Bell games discusses CHSH-type two-player games, local hidden-variable models, classical versus quantum winning probabilities, and sequential statistical formulations, but it explicitly does not mention the Khot–Vishnoi Bell game, Khot–Vishnoi construction, Unique Games, or integrality gaps (Gill, 2021). Its relevance is therefore only as generic Bell-game background, not as a source on Khot–Vishnoi specifics (Gill, 2021).

6. Variants, reductions, and parallel repetition

Subsequent work has focused not only on the original separation but also on the game’s combinatorial cost, especially its exponential number of inputs. The standard Khot–Vishnoi game has

$2$26

inputs per player and $2$27 outputs per player (Buhrman et al., 2010). This input complexity motivated work on reducing the number of questions while keeping essentially the same violation.

A vector-valued version of Schechtman’s empirical method was used to reduce the number of inputs in a nonlocal game while preserving the quotient $2$28 of the quantum over classical bias. Applied to the Khot–Vishnoi game, this yields another game with polynomially many questions,

$2$29

and quantum-over-classical bias

$2$30

The resulting reduced game preserves the key asymptotic violation scale of the original Khot–Vishnoi construction while replacing exponentially many questions by polynomially many (Junge et al., 2016).

The reduction, however, involves tradeoffs. The original Khot–Vishnoi game is explicit and has positive coefficients in the operator-space formulation, whereas the reduced game is probabilistic or non-explicit, and positivity is not directly preserved before the final Bell-functional-to-game conversion (Junge et al., 2016). This suggests that the original construction remains conceptually cleaner even when later work succeeds in compressing its input complexity.

Another development concerns parallel repetition. A 2026 multipartite nonlocality paper proves an asymptotic perfect parallel repetition theorem for the local value of the Khot–Vishnoi game. Writing $2$31 for the single-copy local bound, the theorem states that for the $2$32-repetition of the Khot–Vishnoi game,

$2$33

in the asymptotic limit of infinitely large inputs (Miethlinger et al., 18 Mar 2026). The proof follows the same noise-operator and hypercontractive pattern as the single-copy classical-value argument: the repeated local win probability is rewritten as a noisy correlation over $2$34, bounded by Bonami–Beckner–Gross hypercontractivity, and shown to equal

$2$35

the $2$36-th power of the single-copy bound (Miethlinger et al., 18 Mar 2026).

This repetition theorem is used there to control biseparable bounds in network lifts of bipartite games and thereby certify genuine multipartite Bell nonlocality. The Khot–Vishnoi game is thus not only a benchmark bipartite game; it also functions as a modular component in multipartite constructions because its local value has an unusually clean repeated-game behavior (Miethlinger et al., 18 Mar 2026).

7. Comparisons, misconceptions, and related games

The Khot–Vishnoi Bell game is sometimes conflated with other basis-based nonlocal games, but the distinctions matter. One comparison made explicitly in the literature is with independent-bases games. Both kinds of games associate each input with an orthonormal basis and admit natural entangled strategies using those bases, yet their matching geometry differs: in the Khot–Vishnoi game, the two bases for a given input pair are aligned, and valid outputs are matching elements of the two bases; in the independent-bases setting, the two bases are independent random bases, and players are rewarded for outputting vectors with large positive inner product (Regev, 2011).

This distinction also clarifies that the Khot–Vishnoi game is not an XOR game (Buhrman et al., 2010). It is much closer to a unique game or label-extended game built from cosets (Buhrman et al., 2010). A related misconception is that all strong Bell-violation results of comparable scale concern the same operational figure of merit. Some papers emphasize ratios of biases, while the Khot–Vishnoi game is notable for yielding a strong separation directly in terms of winning probabilities for a positive Bell functional [(Buhrman et al., 2010); (Regev, 2011)]. This difference in normalization is substantive rather than merely notational.

Another useful comparison concerns the role of maximally entangled states. The standard Khot–Vishnoi entangled strategy uses a maximally entangled state of local dimension $2$37 (Buhrman et al., 2010). By contrast, some later coset-based games inspired by the same hypercube/Hadamard geometry were designed to show that non-maximally entangled states can outperform all maximally entangled-state strategies. Those constructions are explicitly described as hybrids of Junge–Palazuelos and Khot–Vishnoi ideas rather than restatements of the Khot–Vishnoi game itself (Regev, 2011).

The game’s significance therefore lies in a precise combination of properties: it is explicit; it is built from the Khot–Vishnoi Unique Games integrality-gap instance; its classical value is tiny; its entangled value is substantially larger; its violation is near-optimal in terms of outputs and local dimension; and its underlying operator is exactly the Boolean noise operator on the hypercube [(Buhrman et al., 2010); (Kolla, 2011)]. A plausible implication is that few nonlocal games expose as clearly the shared structure between approximation hardness, Fourier analysis, semidefinite relaxations, and Bell nonlocality.

8. Significance and continuing role

Within the Bell-inequality literature, the Khot–Vishnoi Bell game occupies a distinctive place because it imports a canonical integrality-gap instance from theoretical computer science into quantum information in a way that preserves explicitness and yields a quantitatively strong separation (Buhrman et al., 2010). The original result already showed that $2$38-dimensional entanglement allows the game to be won with probability $2$39, while the best winning probability without entanglement is $2$40, giving a near-linear ratio that is almost optimal in terms of local dimension and number of possible outputs (Buhrman et al., 2010).

Later work did not displace this role so much as refine different aspects of it. Spectral methods showed that the same Khot–Vishnoi instance has a fully analyzable low-degree Fourier spectrum and can be certified highly unsatisfiable in quasi-polynomial time despite the failure of the standard Unique Games SDP (Kolla, 2011). Input-compression methods showed that the exponential number of questions is not essential for preserving the asymptotic bias ratio, though explicitness and positivity become more delicate (Junge et al., 2016). Parallel-repetition results showed that the local value behaves multiplicatively in repeated play, which makes the game particularly well suited for network and multipartite extensions (Miethlinger et al., 18 Mar 2026).

For these reasons, the Khot–Vishnoi Bell game remains a central example of a nonlocal game whose importance is not exhausted by a single asymptotic inequality. It serves simultaneously as an explicit Bell-violation construction, a translation of a Unique Games integrality gap into nonlocal-game language, a testbed for harmonic analysis on the Boolean cube, and a reusable component in later activation and network-nonlocality arguments [(Buhrman et al., 2010); (Kolla, 2011); (Miethlinger et al., 18 Mar 2026)].