This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Almost Optimal Time Lower Bound for Approximating Parameterized Clique, CSP, and More, under ETH

Venkatesan Guruswami Simons Institute for the Theory of Computing, and Departments of EECS and Mathematics, UC Berkeley. Email: [email protected]. Research supported in part by NSF grants CCF-2228287 and CCF-2211972 and a Simons Investigator award.    Bingkai Lin State Key Laboratory for Novel Software Technology, Nanjing University. Email: [email protected]    Xuandi Ren Department of EECS, UC Berkeley. Email: [email protected]. Supported in part by NSF grant CCF-2228287.    Yican Sun School of Computer Science, Peking University. Email: [email protected]    Kewen Wu Department of EECS, UC Berkeley. Email: [email protected]. Supported by a Sloan Research Fellowship and NSF CAREER Award CCF-2145474.
Abstract

The Parameterized Inapproximability Hypothesis (PIH), which is an analog of the PCP theorem in parameterized complexity, asserts that, there is a constant ε>0\varepsilon>0 such that for any computable function f:f:\mathbb{N}\to\mathbb{N}, no f(k)nO(1)f(k)\cdot n^{O(1)}-time algorithm can, on input a kk-variable CSP instance with domain size nn, find an assignment satisfying 1ε1-\varepsilon fraction of the constraints. A recent work by Guruswami, Lin, Ren, Sun, and Wu (STOC’24) established PIH under the Exponential Time Hypothesis (ETH).

In this work, we improve the quantitative aspects of PIH and prove (under ETH) that approximating sparse parameterized CSPs within a constant factor requires nk1o(1)n^{k^{1-o(1)}} time. This immediately implies that, assuming ETH, finding a (k/2)(k/2)-clique in an nn-vertex graph with a kk-clique requires nk1o(1)n^{k^{1-o(1)}} time. We also prove almost optimal time lower bounds for approximating kk-ExactCover and Max kk-Coverage.

Our proof follows the blueprint of the previous work to identify a "vector-structured" ETH-hard CSP whose satisfiability can be checked via an appropriate form of "parallel" PCP. Using further ideas in the reduction, we guarantee additional structures for constraints in the CSP. We then leverage this to design a parallel PCP of almost linear size based on Reed-Muller codes and derandomized low degree testing.

1 Introduction

One of the goals of complexity theory is to pinpoint the asymptotically optimal time (or other resource) needed to solve basic computational problems or classes of problems. The theory of NP-completeness attacks this at a coarse level, but modern complexity theory also has tools to give more fine-grained information on computational complexity.

A common setting for fine-grained understanding of computational hardness is considering parameterized problems. Under this setting, each instance is attached with an additional parameter kk indicating some specific quantities (e.g., the optimum or the treewidth). We treat kk as some super constant that is much smaller than the instance size nn and consider the existence or absence of algorithms with running time depends both on nn and kk (e.g., with running time 22knO(1)2^{2^{k}}n^{O(1)}, or nkn^{\sqrt{k}}). The hardness of parameterized problems is studied under the realm of parameterized complexity theory [17]. It is a central challenge to figure out the minimal time (depending on both nn and kk) to solve prototypical parameterized problems.

A representative example is the kk-Clique problem parameterized by the optimum kk, which is one of the most fundamental problems in parameterized complexity theory: given an nn-vertex graph as input, determine if it has a clique of size kk. The naive brute force algorithm takes roughly nkn^{k} time. Using fast matrix multiplication, there are better algorithms that take nωk/3n^{\omega k/3} time, where ω\omega is the matrix multiplication exponent. On the hardness side, it is known that no algorithm can decide kk-Clique within running time f(k)no(k)f(k)n^{o(k)}, for any computable function f(k)f(k), under the widely-considered Exponential Time Hypothesis (ETH) [14], which states that algorithms cannot solve 3SAT formulas on nn variables within running time 2o(n)2^{o(n)}. The optimal running time for kk-clique is therefore pinpointed to be nΩ(k)n^{\Omega(k)}, assuming ETH.

Can one design a faster algorithm if one settles for approximating kk-Clique? For example, what if we only want to find a clique of size k/2k/2 in a graph that is promised to have a kk-clique.111This can also be formulated as a “gap” decision problem of distinguishing graphs with a kk-clique from those which do not even have a (k/2)(k/2)-clique.

It was shown that such an approximation to kk-Clique still requires the tightest f(k)nΩ(k)f(k)n^{\Omega(k)} time [11], under the very strong assumption Gap-ETH. The Gap-ETH postulates an exponential time lower bound of approximating Max 3-SAT within a constant ratio. The constant gap baked into the assumption is then transformed into a constant gap for approximating kk-Clique. Though Gap-ETH has been proved under particular strengthening of ETH (smooth version of ETH) [5], a theoretically more satisfactory result is to obtain the hardness of approximating kk-Clique under the original ETH. Under ETH, weaker time lower bounds were known for constant-factor approximations of kk-Clique: f(k)nΩ(logk6)f(k)n^{\Omega\left(\sqrt[6]{\log k}\right)} in [31], which was later improved to f(k)nΩ(logk)f(k)n^{\Omega(\log k)} in [32, 13] and f(k)nkΩ(1/loglogk)f(k)n^{k^{\Omega(1/\log\log k)}} in [33].222There is another line of work on improving the inapproximability factor of kk-Clique under the minimal hypothesis W[1]\neqFPT: constant factor in [31], and the improved ko(1)k^{o(1)} in [28, 13]. However, all these approaches cannot obtain lower bounds better than f(k)nΩ(k)f(k)n^{\Omega(\sqrt{k})} due to the coding theoretic barriers [30].

However, this paper significantly improves this lower bound, assuming only the original ETH.

Theorem 1.1.

Assume ETH. For any constant ε>0\varepsilon>0 and any computable function f(k)f(k), any algorithm that approximates kk-Clique within an ε\varepsilon ratio must take runtime f(k)nk1o(1)f(k)n^{k^{1-o(1)}}.

To prove the theorem above, we follow a similar idea of proving the NP-hardness of approximating cliques, which relies on the NP-hardness of constant approximating CSP (a.k.a., the PCP theorem) and a subsequent FGLSS reduction [18]. In the parameterized setting, we can apply an analogous reduction.

  • We first establish a near-optimal lower bound for approximating (sparse) “parameterized CSPs” with kk variables, O(k)O(k) constraints, an alphabet of size nn, and some constant inapproximability factor.

  • Then Theorem 1.1 follows immediately by the FGLSS reduction [18] and an expander-based gap-amplification procedure [3, 32].

The first step is equivalent to establishing a quantitative version of the Parameterized Inapproximability Hypothesis (PIH) [34], which plays the role of the PCP theorem in the parameterized complexity theory. The first quantitative version of PIH was established under Gap-ETH [11], with the lower bound f(k)nΩ(logk)f(k)n^{\Omega(\log k)}. A recent improvement proved PIH under ETH [21], with a weaker lower bound f(k)nΩ(loglogk)f(k)n^{\Omega\left(\sqrt{\log\log k}\right)}. However, both results are too weak to establish Theorem 1.1. In this paper, we make a significant quantitative improvement to the reduction in [21], obtaining our main theorem stated below.

Theorem 1.2 (Informal version of  Theorem 4.1).

Assume ETH. For some constant ε(0,1)\varepsilon\in(0,1), for any computable function f(k)f(k), every algorithm that takes as input a satisfiable parameterized 2CSP instance with kk variables, O(k)O(k) constraints and size-nn alphabet finds an assignment satisfying 1ε1-\varepsilon fraction of the constraints, must take f(k)nk1o(1)f(k)n^{k^{1-o(1)}} time.

Combining with the parallel repetition in projection games [38], we can immediately boost the soundness to any constant ε>0\varepsilon>0 with lower bound f(k)nkΩ(1/log(1/ε))f(k)n^{k^{\Omega\left(1/\log(1/\varepsilon)\right)}}.

Corollary 1.3.

Assume ETH. For any computable function f(k)f(k) and any constant ε(0,1)\varepsilon\in(0,1), every algorithm that takes as input a satisfiable parameterized 2CSP instance with kk variables, O(k)O(k) constraints and size-nn alphabet finds an assignment satisfying 1ε1-\varepsilon fraction of the constraints, must take f(k)nkΩ(1/log(1/ε))f(k)n^{k^{\Omega\left(1/\log(1/\varepsilon)\right)}} time.

By the discussion above, Theorem 1.1 follows immediately by combining Theorem 1.2, the FGLSS reduction, and the expander-based gap-amplification procedure [3, 32]. Moreover, using Theorem 1.2 as the foundation, we can obtain, via reductions, strong inapproximability results for other fundamental parameterized problems. Below, we list two application highlights and refer interested readers to [21] for more detailed discussions.

Application Highlight: kk-ExactCover

kk-ExactCover (a.k.a., kk-UniqueSetCover) is one of the canonical problems in the parameterized world. It is a weaker version of the kk-SetCover problem. For the ρ\rho-approximation version of kk-ExactCover with ρ1\rho\geq 1, denoted by (k,ρk)(k,\rho\cdot k)-ExactCover, the instance consists of a universe UU and a collection 𝒮\mathcal{S} of subsets of UU, with a goal to distinguish the following two cases.

  • There exists kk disjoint subsets from 𝒮\mathcal{S} whose union equals the whole universe UU.

  • The union of any ρk\rho\cdot k subsets of 𝒮\mathcal{S} is a proper subset of UU.

Here, the parameter is the optimum kk. We remark that the additional disjointness requirement in the completeness part makes (k,ρk)(k,\rho\cdot k)-ExactCover an excellent intermediate problem for proving the hardness of other problems [2, 35].

On the algorithmic side, the (k,ρk)(k,\rho\cdot k)-ExactCover has a brute-force |𝒮|Ω(k)|\mathcal{S}|^{\Omega(k)}-time algorithm. However, no |𝒮|o(k)|\mathcal{S}|^{o(k)}-time algorithms are known. Thus, it is natural to consider whether we can establish a matching |𝒮|Ω(k)|\mathcal{S}|^{\Omega(k)} lower bound. Our work almost establishes a lower bound for (k,ρk)(k,\rho\cdot k)-ExactCover, under ETH, for some constant ρ\rho. Previously, this was only known under the Gap-ETH assumption [35].

Theorem 1.4.

Assume ETH. There exists some constant ρ1\rho\geq 1, such that for any computable function f(k)f(k), any algorithm deciding (k,ρk)(k,\rho\cdot k)-ExactCover must take runtime f(k)|𝒮|k1o(1)f(k)|\mathcal{S}|^{k^{1-o(1)}}.

To prove the theorem above, We note that the previous work [22] achieves a parameter-preserving reduction from PIH to (k,ρk)(k,\rho\cdot k)-ExactCover for any constant ρ\rho, by imitating the beautiful reduction of Feige [16].333In fact, a weaker variant of PIH, named Average Baby PIH over rectangular constraints, suffices.

Therefore, Theorem 1.4 follows by combining the reduction of [22] with our Theorem 1.2. Applying Corollary 1.3, we can boost the ratio ρ\rho to any constant with lower bound f(k)|𝒮|kΩ(1/log(ρ))f(k)|\mathcal{S}|^{k^{\Omega(1/{\log(\rho)})}}.

Proposition 1.5.

Assume ETH. For any constant ρ1\rho\geq 1 and computable function f(k)f(k), any algorithm deciding (k,ρk)(k,\rho\cdot k)-ExactCover must take runtime f(k)|𝒮|kΩ(1/log(ρ))f(k)|\mathcal{S}|^{k^{\Omega(1/{\log(\rho)})}}.

Application Highlight: Max kk-Coverage

The Max kk-Coverage is the maximization variant of the kk-SetCover problem. For the ρ\rho-approximation version of Max kk-Coverage with ρ<1\rho<1, denoted by Max (ρ,k)(\rho,k)-Coverage, the instances are the same as kk-ExactCover above, but the goal changes to distinguish the following two case:

  • There exists kk subsets from 𝒮\mathcal{S} whose union equals 𝒰\mathcal{U}.

  • Any kk subsets from 𝒮\mathcal{S} has the union size at most ρ|𝒰|\rho\cdot|\mathcal{U}|.

Max (ρ,k)(\rho,k)-Coverage has been widely studied in previous literature. There exists a simple greedy algorithm solving Max (11e,k)(1-\frac{1}{e},k)-Coverage within polynomial runtime [23].

On the hardness side, a celebrated result of Feige [16] showed the NP-hardness of Max (11e+ε,k)(1-\frac{1}{e}+\varepsilon,k)-Coverage for any ε>0\varepsilon>0, thus proving a tight inapproximability result.

In the parameterized world, one can solve Max kk-Coverage in |𝒮|k|\mathcal{S}|^{k} time by brute force enumeration. On the other hand, Cohen-Addad, Gupta, Kumar, Lee, and Li [10] showed that assuming Gap-ETH, Max (11e+ε,k)(1-\frac{1}{e}+\varepsilon,k)-Coverage requires f(k)|𝒮|kpoly(1/ε)f(k)|\mathcal{S}|^{k^{{\rm poly}(1/\varepsilon)}} runtime. Manurangsi [35] further improved this lower bound to the tightest f(k)|𝒮|Ω(k)f(k)|\mathcal{S}|^{\Omega(k)} under Gap-ETH.

Our work implies an almost-optimal time lower bound for Max (ρ,k)(\rho,k)-Coverage under ETH for some constant ρ\rho.

Theorem 1.6.

Assume ETH. There exists some constant ρ(0,1)\rho\in(0,1), such that for any computable function f(k)f(k), any algorithm deciding Max (ρ,k)(\rho,k)-Coverage must take runtime f(k)|𝒮|k1o(1)f(k)|\mathcal{S}|^{k^{1-o(1)}}.

Theorem 1.6 follows from our Theorem 1.2 and the analysis in [35, Sections 9.1 and 9.2] that accomplishes a gap-preserving reduction from kk-CSP to Max kk-Coverage. Applying Corollary 1.3, we can boost the ratio to the tightest 11e+ε1-\frac{1}{e}+\varepsilon with lower bound f(k)|𝒮|kεf(k)|\mathcal{S}|^{k^{\varepsilon^{\prime}}}.

Proposition 1.7.

Assume ETH. for any constant ε(0,1)\varepsilon\in(0,1) and computable function f(k)f(k), there exists a constant ε=ε(ε)\varepsilon^{\prime}=\varepsilon^{\prime}(\varepsilon), any algorithm deciding Max (11e+ε,k)(1-\frac{1}{e}+\varepsilon,k)-Coverage must take runtime f(k)|𝒮|kεf(k)|\mathcal{S}|^{k^{\varepsilon^{\prime}}}.

New PCP Characterizations

An interesting byproduct of Theorem 1.2 is a new PCP theorem for 3SAT as follows.

Theorem 1.8 (Informal Version of Theorem 4.2).

For any parameter knk\ll n, 3SAT has a constant-query PCP verifier with alphabet size |Σ|=2n/k1o(1)|\Sigma|=2^{n/k^{1-o(1)}}, runtime poly(|Σ|,n){\rm poly}(|\Sigma|,n), and logk+O(logkloglogk)\log k+O\left(\sqrt{\log k}\log\log k\right) random coins, which has perfect completeness and soundness 12\frac{1}{2}.

Theorem 1.8 generalizes the classic PCP theorem and gives a smooth trade-off between the proof length and the alphabet size, connecting parameterized complexity and the classical complexity theory.

Paper Organization

In Section 2, we provide an overview for our proof of Theorem 1.2 and discuss future works. In Section 3, we formalize necessary notation and concepts. The formal structure for proving Theorem 1.2 is presented in Section 4, with most technical statements deferred to other sections: the reduction from ETH-hard problems to special vector-valued CSPs is provided in Section 5, and the PCPP verifier for a helper language is constructed in Section 6. In Appendix A, we give details of derandomized parallel low degree tests which is a key component for the PCPP construction.

2 Technical Overview

In this part, we provide general ideas of proving Theorem 1.2. Due to the equivalence between the existence of PCP systems and the inapproximability of constraint satisfaction problems [15, 1], Theorem 1.8 directly follows from Theorem 1.2.

We follow the spirit of [21] to prove Theorem 1.2. The proof framework is as follows.

  • First, we reduce an ETH-hard problem to a CSP problem with specific structures.

  • Then, leveraging the special structures, we construct a probabilistic checkable proof of proximity (PCPP) verifier to check whether the encoding of some given solution satisfies all constraints. Theorem 1.2 follows by converting the PCP verifier into CSP instances [1, 15].

For the first step, [21] reduces 3SAT to vector-valued CSPs (VecCSPs for short), whose variables take values in a vector space 𝔽4t\mathbb{F}_{4}^{t}. In addition, each constraint is either a coordinate-wise parallel constraint or a linear constraint.

  • A parallel constraint (over variables xx and yy) is defined by a sub-constraint Πsub:𝔽4×𝔽4{0,1}\Pi^{sub}:\mathbb{F}_{4}\times\mathbb{F}_{4}\to\{0,1\} and a subset of coordinate Q[t]Q\subseteq[t]. It checks whether Πsub(xi,yi)=1\Pi^{sub}(x_{i},y_{i})=1 for every coordinate iQi\in Q.

  • A linear constraint enforces that two vector-valued variables satisfy a linear equation specified by a matrix M𝔽4t×tM\in\mathbb{F}_{4}^{t\times t}, i.e., y=Mxy=Mx.

Then, in the second step, [21] encodes the solution by parallel Walsh-Hadamard code and constructs a PCPP verifier with double-exponential proof length, resulting in an f(k)nΩ(loglogk)f(k)n^{\Omega(\sqrt{\log\log k})} lower bound for ε\varepsilon-Gap kk-Variable CSP.

2.1 More Refined Vector Structure

Unfortunately, the vector structure used in [21] is far from being enough for obtaining an almost-optimal lower bound due to the following reasons.

  • For parallel constraints, VecCSP sets up the sub-constraints over a subset of the coordinates. There might be 2|E𝗉|2^{|E_{\sf p}|} (where |E𝗉||E_{\sf p}| is the number of parallel constraints) different sub-CSP instances over all coordinates, each of which requires an individual PCPP verifier. To simultaneously check the satisfiability of all these sub-instances, they tuple these verifiers into a giant verifier, resulting in an exponential blowup of the proof length. Hence, an almost linear proof size is impossible.

  • For linear constraints, VecCSP defined in  [21] allow them to be over arbitrary pairs of variables. They introduce auxiliary variables for all pairs of variables and their corresponding linear constraints to check such unstructured constraints. The number of auxiliary variables is |V||E𝗅||V|\cdot|E_{\sf l}|, where |V|,|E𝗅||V|,|E_{\sf l}| are the numbers of variables and linear constraints, respectively. This means that the proof length is at least quadratic, making an almost linear proof size impossible.

  • Furthermore, the VecCSP instance in [21] has parameters |V|=O(k2),|E|=O(k2)|V|=O(k^{2}),|E|=O(k^{2}), which is a starting point with already a quantitative loss for any subsequent constructions.

The analysis above urges us to mine more vector structures and devise new reductions with smaller parameter blowups. In this work, we further engineer VecCSP and obtain special vector-valued CSPs (SVecCSP for short) with three more features.

  • First, SVecCSP partitions the variables into two disjoint parts {x1,xk}˙{y1,yk}\{x_{1},\ldots x_{k}\}\dot{\cup}\{y_{1},\ldots y_{k}\}.

  • Second, for parallel constraints, SVecCSP sets up the sub-constraint on all coordinates, which implies a unified sub-CSP instance among all coordinates. As a result, we avoid the tuple procedure, enabling a highly succinct proof.

  • Third, for linear constraints, SVecCSP only sets up linear constraints over xix_{i} and yiy_{i}, with the same index ii. After encoding xx and yy by the parallel Reed-Muller code, we can leverage this alignment and introduce auxiliary proofs, which is also a codeword of the parallel Reed-Mulle code, to check the validity of linear constraints efficiently, with an almost-linear blowup.

In addition, to decrease the parameter blowup, we apply the reduction in [36, 29] to obtain ETH-hard sparse parameterized CSP instances. Then, we imitate the reduction in [21] to obtain a sparse VecCSP instance with |V|=O(k),|E|=O(k)|V|=O(k),|E|=O(k), which ultimately reduces to SVecCSP, with an almost optimal lower bound, by properly duplicating variables and relocating constraints.

2.2 Applying The Parallel Reed-Muller Code

After establishing the almost optimal runtime lower bound for SVecCSP, we need to design a PCPP verifier certifying the encoding of a given assignment satisfies all constraints. However, the previous work [21] designs a PCPP verifier that requires a proof with a double-exponential length, which is far from obtaining Theorem 1.2.

This paper applies the parallel Reed-Muller (RM) encoding with an almost-linear codeword length. Recall that the variables in the SVecCSP instance are divided into two disjoint parts {x1,xk}˙{y1,yk}\{x_{1},\ldots x_{k}\}\dot{\cup}\{y_{1},\ldots y_{k}\}, the input proof of the PCPP verifier consists of x^\widehat{x} and y^\widehat{y}, which are supposed to be the parallel RM encoding of the assignment over xx and yy respectively. The verification procedure is demonstrated as follows.

First, to ensure that x^\widehat{x} and y^\widehat{y} are indeed codewords of the parallel RM code, we apply parallel low degree testing. The standard low degree testing causes a quadratic blowup in the proof length. To ensure an almost-linear proof length, we use the parallel version of the derandomized low degree testing [9] (Appendix A).

Second, we check whether x^y^\widehat{x}\circ\widehat{y} satisfies all parallel constraints. Since a unified sub-CSP instance exists among all coordinates, as long as we have constructed a PCPP verifier for this sub-CSP, we can simulate it in parallel to check all coordinates at the same time, with no blowup in the proof length (as opposed to an exponential blowup in [21]). Our sub-CSP has an alphabet of constant size. Thus, we can apply the existing approach (see, e.g., [8, Theorem 3.3]) in a black box way to construct an almost-linear PCPP verifier for all parallel constraints in our sub-CSP.

Finally, we check whether x^y^\widehat{x}\circ\widehat{y} satisfies all linear constraints. Recall that SVecCSP only sets up linear constraints over xix_{i} and yiy_{i}, with the same index ii. Denote MiM_{i} as the matrix for the index ii, i.e., the ii-th linear constraint is yi=Mixiy_{i}=M_{i}x_{i}. We introduce an auxiliary proof z^\widehat{z}, satisfying

z^=y^M^x^\widehat{z}=\widehat{y}-\widehat{M}\widehat{x}

where M^\widehat{M} is the parallel RM encoding for matrices {M1,M2,,Mk}\{M_{1},M_{2},\dots,M_{k}\}. Then, all systematic parts of z^\widehat{z}, i.e., the codeword entries corresponding to zi:=yiMixiz_{i}:=y_{i}-M_{i}x_{i} for i[k]i\in[k], should be 0\vec{0}. The key observation is that if x^\widehat{x} and y^\widehat{y} are codewords of parallel RM code, then so does z^\widehat{z}. Hence, we can apply parallel derandomized low degree testing for z^\widehat{z} and apply another PCPP, as in the parallel part, to check whether all systematic parts of z^\widehat{z} are 0\vec{0}. Finally, we check whether z^\widehat{z} satisfies the equation above by simply querying a random index, for which the soundness and completeness are guaranteed by Schwartz-Zippel lemma [40]. In this way, we have a highly efficient PCPP verifier that checks all linear constraints with an almost linear proof length.

The overall PCPP verifier is the combination of the verifiers for parallel constraints and for linear constraints. A more detailed framework of our proof is in Section 4.

2.3 Future Works

Our work gives almost optimal time lower bounds for the approximation version of many canonical parameterized problems under ETH, including kk-Clique, kk-ExactCover, kk-Variable CSPs, and Max kk-Coverage.

Technically, we prove the almost optimal time lower bound for constant-gap kk-Variable CSPs. Using this result as the cornerstone, we obtain the inapproximability for other problems by existing reductions. One open question is whether almost optimal time lower bounds for other problems also follow from this result, e.g., kk-Balanced Biclique [11, 20].

The second question is to obtain the (actual) optimal f(k)nΩ(k)f(k)n^{\Omega(k)} time lower bounds for constant-gap kk-Variable CSPs. This problem can be seen as the parameterized extension of the long-standing linear-size PCP conjecture [7]. In the non-parameterized world, the state-of-the-art PCP theorem with the shortest proof length is due to [15] with quasilinear proof length. It is also interesting if this optimal bound (i.e., the parameterized extension of the linear-size PCP conjecture) can be established assuming the existence of linear-size PCP.

For more interesting directions, please refer to [21].

3 Preliminaries

For a positive integer nn, we use [n][n] to denote the set {1,2,,n}\left\{1,2,\ldots,n\right\}. We use log\log to denote the logarithm with base 22. For a prime power q=pcq=p^{c} where pp is a prime and c1c\geq 1 is an integer, we use 𝔽q\mathbb{F}_{q} to denote the finite field of order pcp^{c} and characteristic 𝖼𝗁𝖺𝗋(𝔽)=p\mathsf{char}(\mathbb{F})=p.

For an event \mathcal{E}, 𝟙\mathbbm{1}_{\mathcal{E}} is defined the indicator function, which equals 11 if \mathcal{E} happens and 0 otherwise. For a finite set SS\neq\emptyset, we use xSx\sim S to denote a uniformly random element from SS.

For disjoint sets SS and TT, we use S˙TS\dot{\cup}T to denote their union while emphasizing ST=S\cap T=\emptyset.

Asymptotics

Throughout the paper, we use O(),Θ(),Ω()O(\cdot),\Theta(\cdot),\Omega(\cdot) to hide absolute constants that do not depend on any other parameter. We also use 𝗉𝗈𝗅𝗒()\mathsf{poly}(\cdot) to denote some implicit polynomial in terms of the parameters within, e.g., 𝗉𝗈𝗅𝗒(f,g)\mathsf{poly}(f,g) is upper bounded by (f2+g2+C)C(f^{2}+g^{2}+C)^{C} for some absolute constant C0C\geq 0.

3.1 Constraint Satisfaction Problem

In this paper, we only focus on constraint satisfaction problems (CSPs) of arity two. Formally, a CSP instance GG is a quadruple (V,E,Σ,{Πe}eE)(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}), where:

  • VV is for the set of variables.

  • EE is for the set of constraints. Each constraint e={ue,ve}Ee=\left\{u_{e},v_{e}\right\}\in E connects two distinct variables ue,veVu_{e},v_{e}\in V.

    The constraint graph is the undirected graph on vertices VV with edges EE. Note that we allow multiple constraints between the same pair of variables; thus, the constraint graph may have parallel edges.

  • Σ\Sigma is for the alphabet of each variable in VV. For convenience, we sometimes have different alphabets for different variables, and we will view them as a subset of a grand alphabet with some natural embedding.

  • {Πe}eE\{\Pi_{e}\}_{e\in E} is the set of constraint validity functions. Given a constraint eEe\in E, the function Πe:Σ×Σ{0,1}\Pi_{e}\colon\Sigma\times\Sigma\to\{0,1\} checks whether the constraint ee between ueu_{e} and vev_{e} is satisfied.

We use |G|=(|V|+|E|)|Σ||G|=(|V|+|E|)\cdot|\Sigma| to denote the size of a CSP instance GG.

Assignment and Satisfiability Value

An assignment is a function σ:VΣ\sigma\colon V\to\Sigma that assigns each variable a value in the alphabet. We use

val(G,σ)=1|E|eEΠe(σ(ue),σ(ve)){\rm val}(G,\sigma)=\frac{1}{|E|}\sum_{e\in E}\Pi_{e}(\sigma(u_{e}),\sigma(v_{e}))

to denote the satisfiability value for an assignment σ\sigma. The satisfiability value for GG is val(G)=maxσ:VΣval(G,σ){\rm val}(G)=\max_{\sigma\colon V\to\Sigma}{\rm val}(G,\sigma). We say that an assignment σ\sigma is a solution if val(G,σ)=1{\rm val}(G,\sigma)=1, and GG is satisfiable iff GG has a solution. When the context is clear, we omit σ\sigma in the description of a constraint, i.e., Πe(ue,ve)\Pi_{e}(u_{e},v_{e}) stands for Π(σ(ue),σ(ve))\Pi(\sigma(u_{e}),\sigma(v_{e})).

Boolean Circuits

Throughout the paper, we consider standard Boolean circuits with AND/OR gate of fan-in two and fan-out two, and NOT gate with fan-out two.

Exponential Time Hypothesis (ETH)

Exponential Time Hypothesis (ETH), first proposed by Impagliazzo and Paturi [24], is a famous strengthening of the 𝖯𝖭𝖯\mathsf{P}\neq\mathsf{NP} hypothesis and that has since found numerous applications in modern complexity theory, especially in fine-grained complexity.

The ETH postulates that the general 3SAT problem has no sub-exponential algorithm. In this paper, we use the ETH-based hardness of the 4-Regular 3-Coloring problem.

Definition 3.1 (4-Regular 3-Coloring).

A 2CSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) is an instance of 4-Regular 3-Coloring if (1) Σ=𝔽4\Sigma=\mathbb{F}_{4}, (2) the constraint graph is 44-regular, and (3) each Πe\Pi_{e} checks whether the two endpoints of ee are assigned with different colors from Λ\Lambda, where Λ𝔽4\Lambda\subset\mathbb{F}_{4} has size three and is fixed in advance.

We remark that usually 3-Coloring is defined directly with a ternary alphabet Λ\Lambda. Here for simplicity of later reductions, we assume the alphabet is 𝔽4\mathbb{F}_{4}. This is without loss of generality since the coloring constraint is also upgraded to check additionally whether the colors are from Λ\Lambda.

Theorem 3.2 (ETH Lower Bound for 4-Regular 3-Coloring [12]).

Assuming ETH, no algorithm can decide 4-Regular 3-Coloring in 2o(|V|)2^{o(|V|)} time.

3.2 Parameterized Complexity Theory

In parameterized complexity theory, we consider a promise language Lyes˙LnoL_{\rm yes}\dot{\cup}L_{\rm no} equipped with a computable function κ\kappa, which returns a parameter κ(x)\kappa(x)\in\mathbb{N} for every input instance xx. We use (Lyes˙Lno,κ)(L_{\rm yes}\dot{\cup}L_{\rm no},\kappa) to denote a parameterized language. We think of κ(x)\kappa(x) as a growing parameter that is much smaller than the instance size |x||x|.

A parameterized promise language (Lyes˙Lno,κ)(L_{\rm yes}\dot{\cup}L_{\rm no},\kappa) is fixed parameter tractable (FPT) if there is an algorithm such that for every input (x,κ(x))(Lyes˙Lno,κ)(x,\kappa(x))\in(L_{\rm yes}\dot{\cup}L_{\rm no},\kappa), it decides whether xLyesx\in L_{\rm yes} in f(κ(x))|x|O(1)f(\kappa(x))\cdot|x|^{O(1)} time for some computable function ff.

An FPT reduction from (L𝗒𝖾𝗌L𝗇𝗈,κ)(L_{\sf yes}\cup L_{\sf no},\kappa) to (L𝗒𝖾𝗌L𝗇𝗈,κ)(L^{\prime}_{\sf yes}\cup L^{\prime}_{\sf no},\kappa^{\prime}) is an algorithm 𝒜\mathcal{A} which, on every input (x,κ(x))(x,\kappa(x)) outputs another instance (x,κ(x))(x^{\prime},\kappa^{\prime}(x^{\prime})) such that:

  • Completeness. If xL𝗒𝖾𝗌x\in L_{\sf yes}, then xL𝗒𝖾𝗌x^{\prime}\in L^{\prime}_{\sf yes}.

  • Soundness. If xL𝗇𝗈x\in L_{\sf no}, then xL𝗇𝗈x^{\prime}\in L^{\prime}_{\sf no}.

  • FPT. There exist universal computable functions ff and gg such that |κ(x)|g(κ(x))|\kappa^{\prime}(x^{\prime})|\leq g(\kappa(x)) and the runtime of 𝒜\mathcal{A} is bounded by f(κ(x))|x|O(1)f(\kappa(x))\cdot|x|^{O(1)}.

We refer to [17] for the backgrounds on fixed parameter tractability and FPT reductions.

ε\varepsilon-Gap kk-Variable CSP

We mainly focus on the gap version of the parameterized CSP problem. Formally, an ε\varepsilon-Gap kk-Variable CSP problem is the following parameterized promise language (Lyes˙Lno,κ)(L_{\rm yes}\dot{\cup}L_{\rm no},\kappa).

  • LyesL_{\rm yes} consists of all CSPs GG with val(G)=1{\rm val}(G)=1.

  • LnoL_{\rm no} consists of all CSPs GG with val(G)<1ε{\rm val}(G)<1-\varepsilon.

  • κ(G)\kappa(G) equals the number of variables in GG.

In other words, we need to decide whether a given CSP instance (G,|V|)(G,|V|) with kk variables satisfies val(G)=1{\rm val}(G)=1 or val(G)<1ε{\rm val}(G)<1-\varepsilon.

Parameterized Inapproximability Hypothesis (PIH)

Parameterized Inapproximability Hypothesis (PIH) is a folklore conjecture generalizing the celebrated PCP theorem to the parameterized complexity. It was first rigorously formulated in [34]. Below, we present a slight reformulation, asserting fixed parameter intractability (rather than W[1]W[1]-hardness specifically) of gap CSP.

Hypothesis 3.3 (PIH).

For an absolute constant 0<ε<10<\varepsilon<1, no FPT algorithm can decide ε\varepsilon-Gap kk-Variable CSP.

3.3 Parallel Reed-Muller Code

Word

We say xx is a word (a.k.a., vector) with alphabet Σ\Sigma if xx is a string of finite length and each entry is an element from Σ\Sigma; and Σ\Sigma^{*} contains all words with alphabet Σ\Sigma. Assume xx has length mm. For each I[m]I\subseteq[m], we use xIx_{I} to denote the sub-string of xx on entries in II. When I={i}I=\left\{i\right\} is a singleton set, we simply use xix_{i} to denote x{i}x_{\left\{i\right\}}. For a word xx over a vector alphabet Σt\Sigma^{t}, for each entry xix_{i} and j[t]j\in[t], we define xi[j]x_{i}[j] as the jj-th coordinate of xix_{i}. We define x[j]x[j] as x1[j]x2[j]xm[j]x_{1}[j]\circ x_{2}[j]\circ\cdots\circ x_{m}[j], which is a word over Σ\Sigma.

Let yy be another word. We use xyx\circ y to denote the concatenation of xx and yy. If yy is also of length mm, we define Δ(x,y):=𝐏𝐫i[m][xiyi]\Delta(x,y):=\operatorname*{\mathbf{Pr}}_{i\in[m]}[x_{i}\neq y_{i}] as their relative Hamming distance. For a set SS of words, if Δ(x,z)δ\Delta(x,z)\geq\delta holds for every zSz\in S, we say xx is δ\delta-far from SS; otherwise we say xx is δ\delta-close to SS. In particular, if S=S=\emptyset, then xx is 11-far from SS. Below, we recall the notion of error correcting codes.

Definition 3.4 (Error Correcting Code (ECC)).

An error correcting code is the image of the encoding map C:Σ1kΣ2KC\colon\Sigma_{1}^{k}\to\Sigma_{2}^{K} with message length kk and codeword length KK. We say that the ECC has a relative distance δ\delta if Δ(C(x),C(y))δ\Delta(C(x),C(y))\geq\delta holds for any distinct x,yΣ1kx,y\in\Sigma_{1}^{k}. We use δ(C)\delta(C) to denote the relative distance of the image map of CC and use Im(C)\rm{Im}(C) to denote the codewords of CC.

Reed-Muller Code

We use the parallel Reed-Muller (RM) code to construct PCPPs. This parallel operation is called interleaving in coding theory. Below, we present the formal definition.

For an mm-variate parallel-output function f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t}, we denote f[1],,f[t]:𝔽m𝔽f[1],\ldots,f[t]\colon\mathbb{F}^{m}\to\mathbb{F} as its single-output components, i.e., f(x)=(f[1](x),,f[t](x))f(x)=(f[1](x),\ldots,f[t](x)) We say ff is of parallel degree-dd if f[1],,f[t]f[1],\dots,f[t] are degree-dd polynomials, where a polynomial is degree-dd if all monomials with (total) degree larger than dd have zero coefficients.

If |𝔽|>d|\mathbb{F}|>d and |𝔽|m(m+dd)|\mathbb{F}|^{m}\geq\binom{m+d}{d}, by a dimension argument, there exist (m+dd)\binom{m+d}{d} distinct points (a.k.a., interpolation set) {ξ1,,ξ(m+dd)}𝔽m\{\xi_{1},\dots,\xi_{\binom{m+d}{d}}\}\subseteq\mathbb{F}^{m} whose values a=(a1,,a(m+dd))(𝔽t)(m+dd)a=(a_{1},\ldots,a_{\binom{m+d}{d}})\in(\mathbb{F}^{t})^{\binom{m+d}{d}} can uniquely determine the polynomial faf_{a} of parallel degree dd.

Definition 3.5 (Parallel RM Code).

Assume |𝔽|>d|\mathbb{F}|>d and |𝔽|m(m+dd)|\mathbb{F}|^{m}\geq\binom{m+d}{d}. Let {ξ1,,ξ(m+dd)}\{\xi_{1},\dots,\xi_{\binom{m+d}{d}}\} be the set above. The (𝔽,m,d,t)(\mathbb{F},m,d,t)-parallel RM code is the image of the following encoding map:

𝖱𝖬𝔽,m,d,t:(𝔽t)(m+dd)(𝔽t)|𝔽|m,{{\sf RM}}^{\mathbb{F},m,d,t}:\left(\mathbb{F}^{t}\right)^{\binom{m+d}{d}}\to\left(\mathbb{F}^{t}\right)^{|\mathbb{F}|^{m}},

where for each a=(a1,,a(m+dd))(𝔽t)(m+dd)a=(a_{1},\dots,a_{\binom{m+d}{d}})\in(\mathbb{F}^{t})^{\binom{m+d}{d}}, the encoding 𝖱𝖬𝔽,m,d,t(a){{\sf RM}}^{\mathbb{F},m,d,t}(a) is the truth table of faf_{a} over the whole space 𝔽m\mathbb{F}^{m}.

In addition, aa is the systematic part of the parallel RM encoding, which can be read off directly: for each j(m+dd)j\in\binom{m+d}{d}, aja_{j} equals the entry indexed by ξj\xi_{j} in the codeword.

By Schwartz-Zippel lemma444Schwartz-Zippel lemma is usually stated only for single-output polynomials, but it naturally generalizes to the parallel case., the relative distance of (𝔽,m,d,t)(\mathbb{F},m,d,t)-parallel RM code is

δ(𝖱𝖬𝔽,m,d,t)=1d|𝔽|.\delta({{\sf RM}}^{\mathbb{F},m,d,t})=1-\frac{d}{|\mathbb{F}|}.

Furthermore, there is an efficient codeword testing procedure (Theorem 3.6) for 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}, the proof of which is in Appendix A.

Theorem 3.6 (Codeword Testing).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. Let Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1} be the set of univariate degree-dd polynomials over 𝔽\mathbb{F}. There exists an efficient verifier 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} with the following properties.

  • The input of 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} is TπT\circ\pi, where T(𝔽t)|𝔽|mT\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} is supposed to be a codeword of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} and π(Σt)|𝔽|m(mlog|𝔽|)O(1)\pi\in(\Sigma^{t})^{|\mathbb{F}|^{m}\cdot(m\log|\mathbb{F}|)^{O(1)}} is the auxiliary proof.

  • 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} tosses mlog|𝔽|+O(loglog|𝔽|+logm)m\log|\mathbb{F}|+O\left(\log\log|\mathbb{F}|+\log m\right) unbiased coins and makes 22 queries on TπT\circ\pi.

  • If TIm(𝖱𝖬𝔽,m,d,t)T\in\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), then there exists some π\pi such that 𝒫𝗅𝖽𝗍(Tπ)\mathcal{P}_{\sf ldt}(T\circ\pi) always accepts.

  • If TT is δ\delta-far from Im(𝖱𝖬𝔽,m,d,t)\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), then 𝐏𝐫[𝒫𝗅𝖽𝗍(Tπ)rejects]240δ\operatorname*{\mathbf{Pr}}[\mathcal{P}_{\sf ldt}(T\circ\pi)\ \text{rejects}]\geq 2^{-40}\delta for any π\pi.

Given a Boolean circuit C:{0,1}k{0,1}C:\{0,1\}^{k}\to\{0,1\}, we say a word x({0,1}t)kx\in(\{0,1\}^{t})^{k} parallel satisfies CC iff C(x[j])=1C(x[j])=1 holds for every coordinate j[t]j\in[t].

Extracting from the verifier in Theorem 3.6, we obtain a circuit of small size that describes the codeword testing procedure, the proof of which is also in Appendix A.

Theorem 3.7.

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. There exists a Boolean circuit C𝗅𝖽𝗍C_{\mathsf{ldt}} of size |𝔽|m𝗉𝗈𝗅𝗒|𝔽||\mathbb{F}|^{m}\mathsf{poly}|\mathbb{F}| for T(𝔽t)|𝔽|mT\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}}, where we encode 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, such that TT is codeword of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} iff TT parallel satisfies C𝗅𝖽𝗍C_{\mathsf{ldt}}.

3.4 Pair Language and Probabilistic Checkable Proof of Proximity

We will perform various satisfiability testing for assignments of Boolean circuits where both are given as input. This motivates the notion of pair language, where the input is naturally divided into two parts corresponding to circuits and assignments respectively.

Pair Language

Formally, LL is a pair language over the alphabet Σ𝗑\Sigma_{\mathsf{x}} and Σ𝗒\Sigma_{\mathsf{y}} if all words of LL are in the form (x,y)(x,y) where xΣ𝗑,yΣ𝗒x\in\Sigma_{\mathsf{x}}^{*},y\in\Sigma_{\mathsf{y}}^{*}. For each xΣ𝗑x\in\Sigma_{\mathsf{x}}^{*}, we define L(x)={yΣ𝗒|(x,y)L}L(x)=\{y\in\Sigma_{\mathsf{y}}^{*}|(x,y)\in L\} as the restriction of LL on xx.

Probabilistic checkable proof of proximity (PCPP)

PCPP provides robust testing of satisfiability for pair languages.

Definition 3.8 (PCPP).

Given a pair language LL with alphabet Σ\Sigma, a (q,r,δ,ε,Σ2)(q,r,\delta,\varepsilon,\Sigma_{2})-PCPP verifier 𝒫\mathcal{P} takes as input a pair of words (x,y)(x,y) and an auxiliary proof π\pi with alphabet Σ2\Sigma_{2}, such that:

  1. (T1)

    The verifier 𝒫\mathcal{P} reads all bits of xx, tosses at most r(|x|)r(|x|) many unbiased coins, makes at most q(|x|)q(|x|) queries on yy and π\pi, and then decides to accept or reject within runtime (|x|+|Σ|+|Σ2|)O(1)(|x|+|\Sigma|+|\Sigma_{2}|)^{O(1)}.

    We use IrI_{r} to denote the query positions on yπy\circ\pi under randomness rr, and use 𝒫(x,y,r,a)\mathcal{P}(x,y,r,a) to indicate the behavior (i.e., accept or reject) of 𝒫\mathcal{P} under randomness rr and query answer aa.

  2. (T2)

    If (x,y)L(x,y)\in L, then there exists some π\pi such that 𝒫\mathcal{P} always accepts.

  3. (T3)

    If yy is δ\delta-far from L(x)L(x), then 𝒫\mathcal{P} rejects with probability at least ε\varepsilon for every π\pi.

We say 𝒫\mathcal{P} is a (q,r,ε,Σ2)(q,r,\varepsilon,\Sigma_{2})-PCP verifier if it is a (q,r,1,ε,Σ2)(q,r,1,\varepsilon,\Sigma_{2})-PCPP verifier.

Remark 3.9 (Proof Length).

We can always upper bound |π||\pi| by 2r(|x|)q(|x|)2^{r(|x|)}q(|x|), for which reason we do not put an additional parameter in Definition 3.8.

From PCPP to CSP

PCPPs are tightly connected with CSPs. The following standard reduction establishes the connection.

Definition 3.10 (From PCPP to CSP).

Given a (q,r,δ,ε,Σ2)(q,r,\delta,\varepsilon,\Sigma_{2})-PCPP verifier 𝒫\mathcal{P} for the pair language L={(x,y)|x,yΣ}L=\{(x,y)|x,y\in\Sigma^{*}\}. We define a CSP instance G=(V,E,Σ,{Πe}eE)G^{\prime}=(V^{\prime},E^{\prime},\Sigma^{\prime},\{\Pi_{e}^{\prime}\}_{e\in E^{\prime}}), where V=V𝗒˙Vπ˙VpcppV^{\prime}=V_{\mathsf{y}}\dot{\cup}V_{\pi}\dot{\cup}V_{{\rm pcpp}} and Σ=ΣΣ2\Sigma^{\prime}=\Sigma\cup\Sigma_{2}, by the following steps:

  • First, we treat each position of yy (resp., π\pi) as a single variable in V𝗒V_{\mathsf{y}} and (resp., VπV_{\pi}) with alphabet Σ\Sigma (resp., Σ2\Sigma_{2}). Note that |Vy|=|y||V_{y}|=|y| and |Vπ|2r(|x|)q(|x|)|V_{\pi}|\leq 2^{r(|x|)}q(|x|) by Remark 3.9.

  • Then, for each choice of random coins r{0,1}r(|x|)r\in\{0,1\}^{r(|x|)}, let SrS_{r} be the set of query positions over yπy\circ\pi under randomness rr; and we add a variable zrz_{r} to VpcppV_{\rm pcpp} whose alphabet is (ΣΣ2)|Sr|(\Sigma\cup\Sigma_{2})^{|S_{r}|}, i.e., all possible configurations of the query result. Note that |Vpcpp|2r(|x|)|V_{\rm pcpp}|\leq 2^{r(|x|)}.

  • Finally, we add constraints between zrz_{r} and every query position iSri\in S_{r}. The constraint checks whether zrz_{r} is an accepting configuration, and the assignment of the position ii is consistent with the assignment of zrz_{r}.

By construction, the completeness and soundness are preserved up to a factor of qq under this reduction, where the loss comes from splitting qq queries into qq consistency checks. In addition, the reduction from 𝒫\mathcal{P} to GG^{\prime} is an FPT reduction.

Fact 3.11.

The reduction described in Definition 3.10 is an FPT reduction. Recall that 𝒫\mathcal{P} is a (q,r,δ,ε,Σ2)(q,r,\delta,\varepsilon,\Sigma_{2})-PCPP verifier for the pair language LL over the alphabet Σ\Sigma. We have the following properties for GG^{\prime}:

  • Alphabet. The alphabet of GG^{\prime} is (ΣΣ2)q\left(\Sigma\cup\Sigma_{2}\right)^{q}.

  • Parameter Blowup. The number of varialbes |V||y|+2r(|x|)q(|x|)+2r(|x|)|V^{\prime}|\leq|y|+2^{r(|x|)}q(|x|)+2^{r(|x|)}. The number of constraints |E|=O(Vq(|x|))|E^{\prime}|=O(V^{\prime}\cdot q(|x|)).

  • Completeness. If (x,y)L(x,y)\in L, then there exists a solution σ\sigma^{\prime} of GG^{\prime} assigning yy to V𝗒V_{\mathsf{y}}.

  • Soundness. If yy is δ\delta-far from L(x)L(x), then any assignment σ\sigma^{\prime} assigning yy to V𝗒V_{\mathsf{y}} satisfies at most 1εq1-\frac{\varepsilon}{q} fraction of the constraints in GG^{\prime}. Note that if L(x)=L(x)=\emptyset, then val(G)1εq{\rm val}(G^{\prime})\leq 1-\frac{\varepsilon}{q}.

Circuit Value Problem (CktVal)

CktVal is a standard pair language widely used in the classical PCP theorems (see e.g., [8]). It directly checks if a binary string is a solution to a Boolean circuit.

Definition 3.12 (CktVal).

CktVal is a pair language over {0,1}\{0,1\} consisting of all words in the form of w=(C,z)w=(C,z), where CC is a Boolean predicate with |z||z| input bits and zz is a binary string that satisfies CC. We define the input length |w||w| to be the size of CC plus |z||z|.

We quote the following almost-linear PCPP result for CktVal, which follows from [8, Theorem 3.3] by setting tt to be a constant sufficiently large.

Theorem 3.13 (PCPP for CktVal, [8]).

There exists an absolute constant 0<δ21000<\delta_{\star}\leq 2^{-100} such that CktVal has

an (O(1),log|w|+O(log0.1|w|),δ,12,{0,1})\left(O(1),\log|w|+O(\log^{0.1}|w|),\delta_{\star},\frac{1}{2},\{0,1\}\right)-PCPP verifier 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt},

where |w||w| is the input length of CktVal.

Remark 3.14.

In the parallel setting, we want to check if a circuit CC is satisfied on tt different inputs z1,z2,,ztz_{1},z_{2},\ldots,z_{t}. Note that according to the Definition 3.8, given (C,zi)(C,z_{i}) and an auxiliary proof πi\pi_{i} for every i[t]i\in[t], the PCPP reads the whole CC and then queries ziz_{i} and πi\pi_{i}. In other words, the query locations depend only on CC. So, when applied in parallel, the PCPP queries the same locations for different i[t]i\in[t].

4 Proof of the Main Theorems

This section is devoted to giving a landscape of the proof of Theorem 1.2 and Theorem 1.8. To depict a clear picture, we relegate the proof of technical lemmas in subsequent sections. Below, we first present the formal statement of Theorem 1.2 and Theorem 1.8.

Theorem 4.1 (Formal Version of Theorem 1.2).

Assume ETH. Then for some constant ε(0,1)\varepsilon\in(0,1) and any computable function f(K)f(K), no algorithm can take as input a 2CSP instance Λ\Lambda with KK variables, O(K)O(K) constraints and alphabet [N][N], distinguish between the following two cases in f(K)NK/2ω(logKloglogK)f(K)N^{K/2^{\omega({\sqrt{\log K}\log\log K})}} time:

  • Λ\Lambda is satisfiable;

  • any assignment satisfies at most 1ε1-\varepsilon fraction of the constraints in Λ\Lambda.

Theorem 4.2 (Formal Version of Theorem 1.8).

For any integer knk\ll n sufficiently large, 3SAT has a constant-query PCP verifier of alphabet size |Σ|=exp(nk2O(logk))|\Sigma|=\exp\!\left(\frac{n}{k}\cdot 2^{O(\sqrt{\log k})}\right), runtime poly(|Σ|,n){\rm poly}(|\Sigma|,n), and logk+O(logkloglogk)\log k+O\left(\sqrt{\log k}\log\log k\right) random coins, which has perfect completeness and soundness 12\frac{1}{2}.

The proof of Theorem 4.1 follows a similar idea in the previous work [21].

  • First, we reduce from 4-Regular 3-Coloring, which has the lower bound 2Ω(|V|)2^{\Omega(|V|)} under ETH, to a CSP with specific vector structures, termed as the special vector-valued CSP (SVecCSP for short). This step is somewhat similar to the previous approach [21], in the sense that both approaches engineer non-parameterized CSP problems into parameterized CSPs with vector structure. However, the vector structure in [21] is not enough for obtaining the almost-optimal lower bound for kk-Variable CSP. This paper obtains a much more refined structure.

  • Second, we design a PCPP verifier for SVecCSPs. To obtain an almost-optimal lower bound, we encode the solution of SVecCSPs via Reed-Muller code with an almost-linear blowup. Below, we present the details of our proof.

4.1 Reduction from 4-Regular 3-Coloring to SVecCSP

In this section, we define special vector-valued CSP (SVecCSP for short). The notion of vector-valued CSPs was first defined in [21], and here we consider a better-structured variant.

Definition 4.3 (SVecCSP).

A CSP instance G=(V,E=E𝗉˙E𝗅,Σ,{Πe}eE)G=(V,E=E_{\mathsf{p}}\dot{\cup}E_{\mathsf{l}},\Sigma,\{\Pi_{e}\}_{e\in E}) is an SVecCSP if the following properties hold.

  1. (S1)

    V={x1,,xk}˙{y1,,yk}V=\{x_{1},\ldots,x_{k}\}\dot{\cup}\{y_{1},\ldots,y_{k}\}.

  2. (S2)

    Σ=𝔽t\Sigma=\mathbb{F}^{t} is a tt-dimensional vector space over a finite field 𝔽\mathbb{F} with 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2.

  3. (S3)

    For each constraint e={u,v}E𝗉e=\{u,v\}\in E_{\mathsf{p}} where u=(u1,,ut)u=(u_{1},\dots,u_{t}) and v=(v1,,vt)v=(v_{1},\ldots,v_{t}) are two variables in VV, there is a sub-constraint Πesub:𝔽×𝔽{0,1}\Pi_{e}^{sub}:\mathbb{F}\times\mathbb{F}\to\{0,1\} such that Πe(u,v)\Pi_{e}(u,v) checks Πesub\Pi_{e}^{sub} on all coordinates, i.e.,

    Πe(u,v)=i[t]Πesub(ui,vi).\Pi_{e}(u,v)=\bigwedge_{i\in[t]}\Pi_{e}^{sub}(u_{i},v_{i}).
  4. (S4)

    E𝗅={{xi,yi}i[k]}E_{\mathsf{l}}=\{\{x_{i},y_{i}\}_{i\in[k]}\}. For each i[k]i\in[k], there exists a matrix Mi𝔽t×tM_{i}\in\mathbb{F}^{t\times t} and a constraint Πi(xi,yi)\Pi_{i}(x_{i},y_{i}) check if yi=Mixiy_{i}=M_{i}x_{i}, i.e.,

    Πi(xi,yi)=𝟙yi=Mixi.\Pi_{i}(x_{i},y_{i})=\mathbbm{1}_{y_{i}=M_{i}x_{i}}.

The following theorem establishes the hardness of SVecCSP.

Theorem 4.4.

There is a reduction algorithm that holds the following. Given as input an integer 6kn6\leq k\leq n and an nn-variable 4-Regular 3-Coloring instance Γ\Gamma, it produces an SVecCSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) where:

  1. (R1)

    Variables and Constraints. |V|=O(k)|V|=O(k) and |E|=O(k)|E|=O(k).

  2. (R2)

    Runtime. The reduction runs in time 𝗉𝗈𝗅𝗒(n,2nlogk/k)\mathsf{poly}(n,2^{n\log k/k}).

  3. (R3)

    Alphabet. Σ=𝔽4t\Sigma=\mathbb{F}_{4}^{t} where t=O(nlogkk)t=O\left(\frac{n\log k}{k}\right).

  4. (R4)

    Satisfiability. GG is satisfiable iff Γ\Gamma is satisfiable.

We defer the proof of Theorem 4.4 to Section 5. Compared with vector-valued CSPs (VecCSP) used in the previous work [21, Definition 3.3], SVecCSP offers more structural properties.

  • For parallel constraints, the previous work [21] sets up the sub-constraint over a subset of the coordinates. By construction, there might be 2|E𝗉|2^{|E_{\sf p}|} different sub-CSP instances over all coordinates. Each of the 2|E𝗉|2^{|E_{\sf p}|} sub-CSPs requires an individual PCPP verifier. To simultaneously check the satisfiability of all these sub-instances, they tuple these verifiers into a giant verifier, resulting in an exponential blowup of the proof length. Hence, an almost-linear size of proof is impossible.

    In contrast, SVecCSP sets up the sub-constraint on all coordinates, which implies a unified sub-CSP instance among all coordinates. As a result, we avoid the tuple procedure, enabling a highly succinct proof.

  • For linear constraints, VecCSP defined in  [21] allow them to be over arbitrary pairs of variables. They introduce auxiliary variables for all pairs of variables and their corresponding linear constraints to check such unstructured constraints. The number of auxiliary variables is |V||E𝗅||V|\cdot|E_{\sf l}|, which means that the proof length is at least quadratic, making an almost linear size of proof impossible.

    However, SVecCSP only sets up linear constraints over xix_{i} and yiy_{i}, with the same index ii. After encoding xx and yy by the parallel Reed-Muller code, we can leverage this alignment and introduce auxiliary proofs, which is also a codeword of the parallel Reed-Mulle code, to check the validity of linear constraints efficiently, with an almost-linear blowup.

Though SVecCSP has a more refined structure than VecCSP, we can obtain it from VecCSP by properly duplicating variables and relocating constraints. However, the VecCSP instance in [21] has parameters |V|=O(k2),|E|=O(k2)|V|=O(k^{2}),|E|=O(k^{2}), which are too large to obtain Theorem 4.4. To get around this, we combine results from [36, 29] and the reduction in [21] to obtain a sparse VecCSP instance with |V|=O(k),|E|=O(k)|V|=O(k),|E|=O(k), which ultimately leads to Theorem 4.4.

4.2 Reduction from SVecCSPs to ε\varepsilon-Gap kk-Variable CSPs

With the ETH-hardness of SVecCSP by Theorem 4.4, the remaining work is to reduce SVecCSP to ε\varepsilon-Gap kk-Variable CSP. To this end, we follow the same idea of previous works [4, 21].

  1. (L1)

    First, we encode the solution using an error correcting code.

  2. (L2)

    Then, we design a constant-query PCPP verifier to check whether the given proof is the codeword of some solution with the aid of auxiliary proofs.

  3. (L3)

    Finally, we obtain Theorem 4.1 by converting the PCPP verifier into an instance of ε\varepsilon-Gap kk-Variable CSP by Fact 3.11.

Remark 4.5.

In our actual construction of Item (L2), parallel constraints and linear constraints will be processed separately, building on a (supposedly) same solution. Hence we need to encode the solution using error correcting codes in Item (L1) and design PCPP verifier (instead of PCP verifier) for Item (L2) on top of the shared encoding of the solution.

For the first step (Item (L1)), we need error correcting codes with almost-linear length blowup. In detail, we choose the parallel Reed-Muller code (see Definition 3.5) with suitable choice of parameters. This motivates us to define the following pair language SVecCSP Satisfiability (SVSat).

Definition 4.6 (SVSat).

(𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat is a pair language consisting of w=(G,(x^,y^))w=(G,(\widehat{x},\widehat{y})) where:

  • G=(V={x1,,xk}˙{y1,,yk},E=E𝗉˙E𝗅,Σ=t,{Πe}eE)G=(V=\{x_{1},\ldots,x_{k}\}\dot{\cup}\{y_{1},\ldots,y_{k}\},E=E_{\mathsf{p}}\dot{\cup}E_{\mathsf{l}},\Sigma=\mathbb{H}^{t},\{\Pi_{e}\}_{e\in E}) is an SVecCSP instance. We require that k(m+dd)k\leq\binom{m+d}{d} and \mathbb{H} is a subfield of 𝔽\mathbb{F}.

  • x^,y^\widehat{x},\widehat{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}. Suppose x^=𝖱𝖬𝔽,m,d,t(σx)\widehat{x}={{\sf RM}}^{\mathbb{F},m,d,t}(\sigma_{x}) and y^=𝖱𝖬𝔽,m,d,t(σy)\widehat{y}={{\sf RM}}^{\mathbb{F},m,d,t}(\sigma_{y}) for some σx,σy(𝔽t)(m+dd)\sigma_{x},\sigma_{y}\in(\mathbb{F}^{t})^{\binom{m+d}{d}}. Define assignment σ:V𝔽t\sigma:V\to\mathbb{F}^{t} by

    σ(v):={σx(i)v=xi,σy(i)v=yi.\sigma(v):=\begin{cases}\sigma_{x}(i)&v=x_{i},\\ \sigma_{y}(i)&v=y_{i}.\end{cases}

    We further require that σ\sigma is a solution of GG, which implicitly demands σ(v)t\sigma(v)\in\mathbb{H}^{t} for vVv\in V.

Remark 4.7.

Since (m+dd)k\binom{m+d}{d}\geq k, the index ii in defining σ\sigma is at most the length of σx,σy\sigma_{x},\sigma_{y} and thus σ\sigma is well defined in Definition 4.6. The parameters 𝔽,m,d,t\mathbb{F},m,d,t come from the parameters of the parallel Reed-Muller code.

In this pair language, GG is the starting SVecCSP instance in our reduction, and (x^,y^)(\widehat{x},\widehat{y}) serves as the encoding of a solution. We have the following connection between (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat and the satisfiability of SVecCSP.

Fact 4.8.

Let G=(V={x1,,xk}˙{y1,,yk},E=E𝗉˙E𝗅,Σ=t,{Πe}eE)G=(V=\{x_{1},\ldots,x_{k}\}\dot{\cup}\{y_{1},\ldots,y_{k}\},E=E_{\mathsf{p}}\dot{\cup}E_{\mathsf{l}},\Sigma=\mathbb{H}^{t},\{\Pi_{e}\}_{e\in E}) be an SVecCSP instance. Assume k(m+dd)k\leq\binom{m+d}{d} and \mathbb{H} is a subfield of 𝔽\mathbb{F}. Then, GG is satisfiable iff there exists (x^,y^)(\widehat{x},\widehat{y}) such that (G,(x^,y^))(𝔽,m,d,t)(G,(\widehat{x},\widehat{y}))\in(\mathbb{F},m,d,t)-SVSat.

Remark 4.9 (Encoding Choice).

One may wonder the necessity for encoding x^\widehat{x} and y^\widehat{y} separately, as opposed to encoding them jointly as xy^\widehat{x\circ y}. This is due to the bipartite structure of the linear constraints E𝗅E_{\mathsf{l}} (see Item (S4)) that we will need to ensure proximity for the encoding on both xx and yy. This will be clear in Item (T3) and Lemma 4.23.

For the second step (Item (L2)), we need to construct a PCPP verifier for (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat. Formally, we construct a PCPP verifier with the following parameters.

Theorem 4.10.

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{12md,2101mlog|𝔽|}|\mathbb{F}|\geq\max\{12md,2^{101}m\log|\mathbb{F}|\}. Then for any δ[0,1]\delta\in[0,1], (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat has

and (O(1),log(|E|+|𝔽|m)+O(log0.1(|E|+|𝔽|m)+log|𝔽|),δ,Ω(δ),Σt)\left(O(1),\log(|E|+|\mathbb{F}|^{m})+O\left(\log^{0.1}(|E|+|\mathbb{F}|^{m})+\log|\mathbb{F}|\right),\delta,\Omega(\delta),\Sigma^{t}\right)-PCPP verifier 𝒫\mathcal{P},

where |E||E| is the number of constraints in (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat and Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1}.

Fix some (G,(x^,y^))(G,(\widehat{x},\widehat{y})) supposed to be in (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat. We use σ\sigma to denote the assignment recovered (if succeeded) from x^\widehat{x} and y^\widehat{y}. Our goal is to verify whether σ\sigma is the solution of GG. To this end, we will handle parallel and linear constraints in Subsection 4.2.1 and Subsection 4.2.3 separately, then combine them together in Subsection 4.2.3 to prove Theorem 4.10.

4.2.1 Verification of Parallel Constraints

We first consider verifying whether σ\sigma satisfies all parallel constraints by probing the indices of (x^,y^)(\widehat{x},\widehat{y}). Since 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} is a systematic code (Definition 3.5), we can recover the value of an index in the assignment σ\sigma directly from probing an index of x^y^\widehat{x}\circ\widehat{y}. In addition, recall that parallel constraints set up the same sub-constraints over all coordinates (Item (S3)), we can build up a unified Boolean circuit C𝗉:{0,1}2|𝔽|mlog|𝔽|{0,1}C_{\mathsf{p}}:\{0,1\}^{2|\mathbb{F}|^{m}\log|\mathbb{F}|}\to\{0,1\} that checks whether the systematic part of x^\widehat{x} and y^\widehat{y} satisfies the sub-constraints in a single coordinate. In detail, the circuit C𝗉C_{\mathsf{p}} is defined as follows.

Definition 4.11 (The Circuit C𝗉C_{\mathsf{p}}).

Let x~,y~\widetilde{x},\widetilde{y} be words of length |𝔽|m|\mathbb{F}|^{m} over alphabet555For ease of presentation, the circuit’s input is described as alphabet 𝔽\mathbb{F}. We take the trivial conversion from an element of 𝔽\mathbb{F} into a binary string of length log|𝔽|\log|\mathbb{F}|. 𝔽\mathbb{F}. Assume x~,y~\widetilde{x},\widetilde{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}. The circuit C𝗉C_{\mathsf{p}} executes the following.

  • C𝗉C_{\mathsf{p}} recovers the messages σx~,σy~\sigma_{\widetilde{x}},\sigma_{\widetilde{y}} from the systematic part of x~,y~\widetilde{x},\widetilde{y} respectively.

  • After that, C𝗉C_{\mathsf{p}} checks whether the assignment σ~\widetilde{\sigma} specified by σx~,σy~\sigma_{\widetilde{x}},\sigma_{\widetilde{y}} has the correct subfield entries and satisfies all constraints in the E𝗉E_{\mathsf{p}} part of GG, at the single-coordinate level. Specifically, σ~\widetilde{\sigma} is the assignment of VV defined by

    σ(v)={σx~(i)v=xi,σy~(i)v=yi.\sigma(v)=\begin{cases}\sigma_{\widetilde{x}}(i)&v=x_{i},\\ \sigma_{\widetilde{y}}(i)&v=y_{i}.\end{cases}

    For every vVv\in V, C𝗉C_{\mathsf{p}} checks whether σ(v)\sigma(v)\in\mathbb{H}; and for every constraint e={u,v}Epe=\{u,v\}\in E_{\textsf{p}}, C𝗉C_{\mathsf{p}} checks whether Πesub(σ(u),σ(v))=1\Pi_{e}^{sub}(\sigma(u),\sigma(v))=1.

The size of circuit C𝗉C_{\sf p} is bounded by (|𝔽|m+k+|E𝗉|)𝗉𝗈𝗅𝗒|𝔽|=(|𝔽|m+|E|)𝗉𝗈𝗅𝗒|𝔽|(|\mathbb{F}|^{m}+k+|E_{\mathsf{p}}|)\cdot\mathsf{poly}|\mathbb{F}|=(|\mathbb{F}|^{m}+|E|)\cdot\mathsf{poly}|\mathbb{F}|.

Double Test Problem (DoubleTest)

In light of Definition 4.11, (x^,y^)(\widehat{x},\widehat{y}) satisfies all parallel constraints iff x^,y^\widehat{x},\widehat{y} are correct codewords and C𝗉(x^[i]y^[i])=1C_{\mathsf{p}}(\widehat{x}[i]\circ\widehat{y}[i])=1 for each i[t]i\in[t]. This motivates us to consider the following pair language DoubleTest related to CktVal.

Definition 4.12 (DoubleTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2. (𝔽,m,d,t)(\mathbb{F},m,d,t)-DoubleTest is a pair language over Σ𝗑={0,1},Σ𝗒=𝔽t\Sigma_{\mathsf{x}}=\{0,1\},\Sigma_{\mathsf{y}}=\mathbb{F}^{t} consisting of w=(C,T1T2)w=(C,T_{1}\circ T_{2}) where

  • CC is a Boolean circuit with 2|𝔽|mlog|𝔽|2|\mathbb{F}|^{m}\log|\mathbb{F}| input bits and T1,T2(𝔽t)|𝔽|mT_{1},T_{2}\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t};

  • if we view 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, T1T2T_{1}\circ T_{2} parallel satisfies CC.

We define the input length |w||w| to be the size of CC plus 2|𝔽|mlog|𝔽|2|\mathbb{F}|^{m}\log|\mathbb{F}|. Note that the dimension tt is reflected on the alphabet, not on the length.

In short, DoubleTest extends CktVal by allowing the assignment to have more dimensions (i.e., tt), allowing the assignment to be partitioned into two parts (i.e., T1,T2T_{1},T_{2}), and assuming each part is encoded by parallel RM code.

Given Definition 4.12 and Definition 4.11, we have the following statement. Note that the assumption that x~,y~\widetilde{x},\widetilde{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} in Definition 4.11 is guaranteed in Definition 4.12.

Fact 4.13.

Parallel constraints E𝗉E_{\mathsf{p}} in GG are satisfied iff (C𝗉,x^y^)(𝔽,m,d,t)(C_{\mathsf{p}},\widehat{x}\circ\widehat{y})\in(\mathbb{F},m,d,t)-DoubleTest.

Thus, to verify that σ\sigma satisfies all parallel constraints, if suffices to construct a PCPP verifier for DoubleTest. The verifier is formally stated as follows and will be proved in Section 6.

Theorem 4.14 (PCPP for DoubleTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. For any δ[0,1]\delta\in[0,1], (𝔽,d,m,t)(\mathbb{F},d,m,t)-DoubleTest has

an (O(1),log|w|+O(log0.1|w|+log|𝔽|),δ,Ω(δ),Σt)\left(O(1),\log|w|+O\left(\log^{0.1}|w|+\log|\mathbb{F}|\right),\delta,\Omega(\delta),\Sigma^{t}\right)-PCPP verifer 𝒫𝖽𝗍\mathcal{P}_{\sf dt},

where |w||w| is the input length of (𝔽,d,m,t)(\mathbb{F},d,m,t)-DoubleTest and Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1}.

4.2.2 Verification of Linear Constraints

We then turn to verifying whether σ\sigma satisfies all linear constraints E𝗅E_{\mathsf{l}}. Recall from Item (S4) that all linear constraints are set up between xix_{i} and yiy_{i}, with the constraint that yi=Mixiy_{i}=M_{i}x_{i}. By defining auxiliary variables zi:=yiMixiz_{i}:=y_{i}-M_{i}x_{i}, it suffices to check whether zi0tz_{i}\equiv\vec{0}_{t} for every i[k]i\in[k].

Thus, we add the auxiliary proof z^(𝔽t)|𝔽|m\widehat{z}\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}}, which is supposed to fulfill the following condition

z^(p)=y^(p)M^(p)x^(p)holds for all p𝔽m,\widehat{z}(p)=\widehat{y}(p)-\widehat{M}(p)\widehat{x}(p)\quad\text{holds for all $p\in\mathbb{F}^{m}$,} (1)

where M^(𝔽t×t)|𝔽|m\widehat{M}\in(\mathbb{F}^{t\times t})^{|\mathbb{F}|^{m}} is the parallel RM encoding of (M1,,Mk,0t×t,,0t×t)(𝔽t×t)(m+dd)(M_{1},\ldots,M_{k},0^{t\times t},\ldots,0^{t\times t})\in(\mathbb{F}^{t\times t})^{\binom{m+d}{d}}. Here we extend Definition 3.5 for matrix values: the value of ξi\xi_{i} is MiM_{i} for iki\leq k and is 0t×t0^{t\times t} for i>ki>k. Equivalently, for each matrix coordinate (i,j)[t]×[t](i,j)\in[t]\times[t], M^[i,j]𝔽|𝔽|m\widehat{M}[i,j]\in\mathbb{F}^{|\mathbb{F}|^{m}} is the RM encoding of (M1[i,j],,Mk[i,j],0,,0)(M_{1}[i,j],\ldots,M_{k}[i,j],0,\ldots,0). We remark that entries of M^\widehat{M} can be efficiently computed on the fly by demand and are not included as a proof for the PCPP verifier.

Based on the discussion above, we obtain the following fact.

Fact 4.15.

Linear constraints E𝗅E_{\mathsf{l}} in GG are satisfied iff x^,y^\widehat{x},\widehat{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} and the systematic part of z^\widehat{z}, defined by Equation 1, are all 0t0^{t}.

Recall that x^,y^\widehat{x},\widehat{y} being correct codewords are guaranteed by the analysis of parallel constraints above. Hence we focus on testing the systematic part of z^\widehat{z}, which amount to z^\widehat{z} parallel satisfying the following circuit.

Definition 4.16 (The Circuit C𝗅C_{\mathsf{l}}).

The circuit C𝗅C_{\mathsf{l}} receives as input a word z~\widetilde{z} of length |𝔽|m|\mathbb{F}|^{m} over alphabet 𝔽\mathbb{F}. It checks if z~(ξi)=0\widetilde{z}(\xi_{i})=0 holds for all i[k]i\in[k], where we recall that ξi\xi_{i} from Definition 3.5.

The size of the circuit C𝗅C_{\mathsf{l}} is bounded by (|𝔽|m+k)𝗉𝗈𝗅𝗒|𝔽|(|\mathbb{F}|^{m}+k)\cdot\mathsf{poly}|\mathbb{F}|.

We will use a variant of DoubleTest, denoted SingleTest, on C𝗅C_{\mathsf{l}} and z^\widehat{z}. But before that, we give a degree bound for z^\widehat{z}.

Claim 4.17.

If x^,y^\widehat{x},\widehat{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}, then the z^\widehat{z} defined by Equation 1 is a codeword of RM𝔽,m,2d,t{\rm RM}^{\mathbb{F},m,2d,t}.

Proof.

Expanding the matrix multiplication, for each coordinate i[t]i\in[t], we have

z^[i](p)=y^[i](p)j[t]M^[i,j](p)x^[j](p)for all p𝔽m,\widehat{z}[i](p)=\widehat{y}[i](p)-\sum_{j\in[t]}\widehat{M}[i,j](p)\cdot\widehat{x}[j](p)\quad\text{for all $p\in\mathbb{F}^{m}$,}

where M^[i,j](p)\widehat{M}[i,j](p) is the (i,j)(i,j)-th entry of M^(p)\widehat{M}(p). Since y^[i],M^[i,j],x^[j]\widehat{y}[i],\widehat{M}[i,j],\widehat{x}[j] are truth tables of degree-dd polynomials, z^[j]\widehat{z}[j] is the truth table of a degree-2d2d polynomial, which means z^Im(RM𝔽,m,2d,t)\widehat{z}\in\mathrm{Im}(\textsf{RM}^{\mathbb{F},m,2d,t}). ∎

Single Test Problem (SingleTest)

At this point, checking Fact 4.15 can be safely handled by the following SingleTest with proper degree conditions.

Definition 4.18 (SingleTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2. (𝔽,m,d,t)(\mathbb{F},m,d,t)-SingleTest is a pair language over Σ𝗑={0,1},Σ𝗒=𝔽t\Sigma_{\mathsf{x}}=\{0,1\},\Sigma_{\mathsf{y}}=\mathbb{F}^{t} consisting of w=(C,T1)w=(C,T_{1}) where

  • CC is a Boolean circuit with 2|𝔽|mlog|𝔽|2|\mathbb{F}|^{m}\log|\mathbb{F}| input bits and T1(𝔽t)|𝔽|mT_{1}\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} is a codeword of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t};

  • if we view 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, T1T_{1} parallel satisfies CC.

We define the input length |w||w| to be the size of CC plus |𝔽|mlog|𝔽||\mathbb{F}|^{m}\log|\mathbb{F}|.

In comparison, SingleTest simply removes the second table T2T_{2} from DoubleTest. Combining Claim 4.17 and Fact 4.15, we have the following result.

Fact 4.19.

Linear constraints E𝗅E_{\mathsf{l}} in GG are satisfied iff x^,y^\widehat{x},\widehat{y} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}, z^\widehat{z} satisfies Equation 1, and (C𝗅,z^)(𝔽,m,2d,t)(C_{\mathsf{l}},\widehat{z})\in(\mathbb{F},m,2d,t)-SingleTest.

Analogous to Theorem 4.14, we also have an efficient PCPP verifier for SingleTest as follows, which will also be proved in Section 6.

Theorem 4.20 (PCPP for SingleTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. For any δ[0,1]\delta\in[0,1], (𝔽,d,m,t)(\mathbb{F},d,m,t)-SingleTest has

an (O(1),log|w|+O(log0.1|w|+log|𝔽|),δ,Ω(δ),Σt)\left(O(1),\log|w|+O\left(\log^{0.1}|w|+\log|\mathbb{F}|\right),\delta,\Omega(\delta),\Sigma^{t}\right)-PCPP verifer 𝒫𝗌𝗍\mathcal{P}_{\sf st},

where |w||w| is the input length of (𝔽,d,m,t)(\mathbb{F},d,m,t)-SingleTest and Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1}.

Finally, we briefly sketch how to check if z^\widehat{z} satisfies Equation 1, as needed in Fact 4.19. This will be done by randomly picking an index pp and checking whether Equation 1 holds on that pp. The soundness will be analyzed by Schwartz-Zippel Lemma. This will be formalized in Subsection 4.2.3.

4.2.3 The Whole Construction and Analysis

Based on the above discussion, we are now ready to construct the PCPP verifier 𝒫\mathcal{P} for SVSat. We invoke Theorem 4.14 to obtain PCPP verifiers 𝒫𝖽𝗍\mathcal{P}_{\sf dt} and 𝒫𝗌𝗍\mathcal{P}_{\sf st} for (𝔽,m,d,t)(\mathbb{F},m,d,t)-DoubleTest and (𝔽,m,2d,t)(\mathbb{F},m,2d,t)-SingleTest respectively.

Recall that the input of (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat is (G,(x^,y^))(G,(\widehat{x},\widehat{y})). The auxiliary proof consists of z^\widehat{z}, π1\pi_{1}, and π2\pi_{2}, where

  • z^\widehat{z} is supposed to be a codeword in RM𝔽,m,2d,t\textsf{RM}^{\mathbb{F},m,2d,t} and z^(p)=y^(p)M^(p)x^(p)\widehat{z}(p)=\widehat{y}(p)-\widehat{M}(p)\widehat{x}(p) for all p𝔽mp\in\mathbb{F}^{m};

  • π1\pi_{1} is supposed to be the auxiliary proof to convince 𝒫𝖽𝗍\mathcal{P}_{\sf dt} that (C𝗉,x^y^)(C_{\mathsf{p}},\widehat{x}\circ\widehat{y}) belongs to the pair language (𝔽,m,d,t)(\mathbb{F},m,d,t)-DoubleTest.

  • π2\pi_{2} is supposed to be the auxiliary proof to convince 𝒫𝗌𝗍\mathcal{P}_{\sf st} that (C𝗅,z^)(C_{\mathsf{l}},\widehat{z}) belongs to the pair language (𝔽,m,2d,t)(\mathbb{F},m,2d,t)-SingleTest.

The verifier 𝒫\mathcal{P} performs one of the following three tests with equal probability.

  1. (T1)

    Feed the pair of words (C𝗉,x^y^)(C_{\mathsf{p}},\widehat{x}\circ\widehat{y}) and the auxiliary proof π1\pi_{1} into 𝒫𝖽𝗍\mathcal{P}_{\sf dt}. Reject if 𝒫𝖽𝗍\mathcal{P}_{\sf dt} rejects.

  2. (T2)

    Feed the pair of words (C𝗅,z^)(C_{\mathsf{l}},\widehat{z}) and the auxiliary proof π2\pi_{2} into 𝒫𝗌𝗍\mathcal{P}_{\sf st}. Reject if 𝒫𝗌𝗍\mathcal{P}_{\sf st} rejects.

  3. (T3)

    Generate a random point p𝔽mp\in\mathbb{F}^{m}, reject if z^(p)y^(p)M^(p)x^(p)\widehat{z}(p)\neq\widehat{y}(p)-\widehat{M}(p)\widehat{x}(p).

At this point, we are ready to analyze the PCPP verifier 𝒫\mathcal{P}. In particular, Theorem 4.10 follows from the combination of the following Lemma 4.21, Lemma 4.22, and Lemma 4.23.

Lemma 4.21 (Parameters).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{12md,2101mlog|𝔽|}|\mathbb{F}|\geq\max\left\{12md,2^{101}m\log|\mathbb{F}|\right\}. Then 𝒫\mathcal{P} tosses

log(|E|+|𝔽|m)+O(log0.1(|E|+|𝔽|m)+log|𝔽|).\log(|E|+|\mathbb{F}|^{m})+O\left(\log^{0.1}(|E|+|\mathbb{F}|^{m})+\log|\mathbb{F}|\right).

unbiased coins and makes O(1)O(1) queries. The alphabet of the auxiliary proof is Σt\Sigma^{t} where Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1}.

Proof.

By Theorem 4.14 and Theorem 4.20, both 𝒫𝖽𝗍\mathcal{P}_{\sf dt} and 𝒫𝗌𝗍\mathcal{P}_{\sf st} make constant queries. Also in Item (T3), 𝒫\mathcal{P} makes 3 queries on x^,y^,z^\widehat{x},\widehat{y},\widehat{z}. Thus 𝒫\mathcal{P} makes O(1)O(1) total queries.

By Definitions 4.16 and 4.11, the input lengths of 𝒫𝖽𝗍\mathcal{P}_{\sf dt} and 𝒫𝗌𝗍\mathcal{P}_{\sf st} in Item (T1) and Item (T2) are

{|w1|=(|𝔽|m+|E|)𝗉𝗈𝗅𝗒|𝔽|+2|𝔽|mlog|𝔽|(|E|+|𝔽|m)𝗉𝗈𝗅𝗒|𝔽|,|w2|=(|𝔽|m+k)𝗉𝗈𝗅𝗒|𝔽|+|𝔽|mlog|𝔽|(|E|+|𝔽|m)𝗉𝗈𝗅𝗒|𝔽|,\left\{\begin{aligned} |w_{1}|&=(|\mathbb{F}|^{m}+|E|)\cdot\mathsf{poly}|\mathbb{F}|+2|\mathbb{F}|^{m}\log|\mathbb{F}|&\leq(|E|+|\mathbb{F}|^{m})\mathsf{poly}|\mathbb{F}|,\\ |w_{2}|&=(|\mathbb{F}|^{m}+k)\cdot\mathsf{poly}|\mathbb{F}|+|\mathbb{F}|^{m}\log|\mathbb{F}|&\leq(|E|+|\mathbb{F}|^{m})\mathsf{poly}|\mathbb{F}|,\end{aligned}\right.

respectively. Putting this into Theorem 4.14 and Theorem 4.20, the number of unbiased coins used in 𝒫𝖽𝗍\mathcal{P}_{\sf dt} and 𝒫𝗌𝗍\mathcal{P}_{\sf st} is

log(|E|+|𝔽|m)+O(log0.1(|E|+|𝔽|m)+log|𝔽|).\log(|E|+|\mathbb{F}|^{m})+O\left(\log^{0.1}(|E|+|\mathbb{F}|^{m})+\log|\mathbb{F}|\right).

In Item (T3), 𝒫\mathcal{P} tosses mlog|𝔽|m\log|\mathbb{F}| coins. Since 𝒫\mathcal{P} only executes one of the three tests, the randomness is bounded by their maximum.

The auxiliary proofs π1,π2\pi_{1},\pi_{2} have alphabet Σ\Sigma by Theorem 4.14 and Theorem 4.20. The alphabet of z^\widehat{z} is 𝔽t\mathbb{F}^{t}, which can also be embedded into the larger Σt\Sigma^{t}. ∎

Lemma 4.22 (Completeness).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{12md,2101mlog|𝔽|}|\mathbb{F}|\geq\max\left\{12md,2^{101}m\log|\mathbb{F}|\right\}. Suppose x^=𝖱𝖬𝔽,m,d,t(σx),y^=𝖱𝖬𝔽,m,d,t(σy)\widehat{x}={{\sf RM}}^{\mathbb{F},m,d,t}(\sigma_{x}),\widehat{y}={{\sf RM}}^{\mathbb{F},m,d,t}(\sigma_{y}), and the assignment σ\sigma given by σx\sigma_{x} and σy\sigma_{y} (recall Definition 4.6) is a solution to GG. Then there exist z^,π1,π2\widehat{z},\pi_{1},\pi_{2} which 𝒫\mathcal{P} accepts with probability 11.

Proof.

By Definition 4.11, (x^,y^)(\widehat{x},\widehat{y}) parallel satisfies the circuit C𝗉C_{\mathsf{p}}. Thus (C𝗉,x^y^)(𝔽,m,d,t)(C_{\mathsf{p}},\widehat{x}\circ\widehat{y})\in(\mathbb{F},m,d,t)-DoubleTest by Fact 4.13. Hence there exists an auxiliary proof π1\pi_{1} which makes 𝒫𝖽𝗍\mathcal{P}_{\sf dt} accepts with probability 1. Item (T1) therefore always passes.

For each p𝔽mp\in\mathbb{F}^{m}, define z^(p)=y^(p)M^(p)x^(p)\widehat{z}(p)=\widehat{y}(p)-\widehat{M}(p)\widehat{x}(p). By Claim 4.17, z^Im(RM𝔽,m,2d,t)\widehat{z}\in\mathrm{Im}(\textsf{RM}^{\mathbb{F},m,2d,t}). Let σx,σy,σz\sigma_{x},\sigma_{y},\sigma_{z} be the messages x^,y^,z^\widehat{x},\widehat{y},\widehat{z} encodes respectively. Note that (σx,σy)(\sigma_{x},\sigma_{y}) satisfies the linear constraints E𝗅E_{\mathsf{l}} in GG, i.e., σy(i)Miσx(i)=0t\sigma_{y}(i)-M_{i}\sigma_{x}(i)=0^{t} for all i[k]i\in[k]. Hence all i[k]i\in[k], we have

σz(i)=z^(ξi)=y^(ξi)M^(ξi)x^(ξi)=σy(i)Miσx(i)=0t,\sigma_{z}(i)=\widehat{z}(\xi_{i})=\widehat{y}(\xi_{i})-\widehat{M}(\xi_{i})\widehat{x}(\xi_{i})=\sigma_{y}(i)-M_{i}\sigma_{x}(i)=0^{t},

where {ξ1,,ξ(m+dd)}\{\xi_{1},\ldots,\xi_{\binom{m+d}{d}}\} are the distinct points defining the encoding of parallel RM code (see Definition 3.5). Therefore, z^\widehat{z} parallel satisfies C𝗅C_{\mathsf{l}}. By Fact 4.19, there exists an auxiliary proof π2\pi_{2}, which makes 𝒫𝗌𝗍\mathcal{P}_{\sf st} accepts with probability 1, and Item (T2) always passes.

Finally Item (T3) always passes due to the definition of z^\widehat{z}. This completes the proof. ∎

Lemma 4.23 (Soundness).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{12md,2101mlog|𝔽|}|\mathbb{F}|\geq\max\left\{12md,2^{101}m\log|\mathbb{F}|\right\}. Let δ[0,1]\delta\in[0,1] be arbitrary. If (x^,y^)(\widehat{x},\widehat{y}) is δ\delta-far from satisfying, i.e., δ\delta-far from the restriction of (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat on GG, then 𝒫\mathcal{P} rejects with probability Ω(δ)\Omega(\delta).

Proof.

Let κ1\kappa\geq 1 be a large constant, the specific value of which depends on the hidden constants in Theorem 4.14 and Theorem 4.20. By modifying the hidden constant in Ω()\Omega(\cdot) here and noticing that δ\delta-far implies δ\delta^{\prime}-far for any δδ\delta^{\prime}\leq\delta, we safely assume δ1/κ2\delta\leq 1/\kappa^{2}.

Fix arbitrary (x^,y^)(\widehat{x},\widehat{y}) that is δ\delta-far from satisfying. Assume that 𝒫\mathcal{P} rejects with probability at most κδ\kappa\cdot\delta, since otherwise the statement already holds. Then each of the tests Item (T1), Item (T2), and Item (T3) reject with probability at most 3κδ3\kappa\cdot\delta. By choosing κ\kappa sufficiently large and according to soundness guarantee of 𝒫𝖽𝗍\mathcal{P}_{\sf dt} and 𝒫𝗌𝗍\mathcal{P}_{\sf st} in Theorem 4.14 and Theorem 4.20, we know

  1. 1.

    (x^,y^)(\widehat{x},\widehat{y}) is δ\delta-close to (x¯,y¯)(\overline{x},\overline{y}), which is a pair of codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} that parallel satisfies C𝗉C_{\mathsf{p}}.

    Since x^,y^\widehat{x},\widehat{y} have the same length, this alse implies that x^\widehat{x} is 2δ2\delta-close to x¯\overline{x} and y^\widehat{y} is 2δ2\delta-close to y¯\overline{y}.

  2. 2.

    z^\widehat{z} is δ\delta-close to z¯\overline{z}, which is a codeword of RM𝔽,m,2d,t\textsf{RM}^{\mathbb{F},m,2d,t} that parallel satisfies C𝗅C_{\mathsf{l}}.

We aim to show that (G,(x¯,y¯))(𝔽,m,d,t)(G,(\overline{x},\overline{y}))\in(\mathbb{F},m,d,t)-SVSat, which contradicts to the assumption that (x^,y^)(\widehat{x},\widehat{y}) is δ\delta-far from satisfying and completes the proof.

Let σx¯\sigma_{\overline{x}}, σy¯\sigma_{\overline{y}}, σz¯\sigma_{\overline{z}} be the messages of x¯,y¯,z¯\overline{x},\overline{y},\overline{z} respectively. It now suffices to prove the assignment σ\sigma given by

σ(v)={σx¯(i)v=xiσy¯(i)v=yi\sigma(v)=\begin{cases}\sigma_{\overline{x}}(i)&v=x_{i}\\ \sigma_{\overline{y}}(i)&v=y_{i}\end{cases}

satisfies all constraints in E𝗉˙E𝗅E_{\mathsf{p}}\dot{\cup}E_{\mathsf{l}}.

By Item 1 and Fact 4.13, σ\sigma satisfies all constraints in E𝗉E_{\mathsf{p}}. To analyze constraints in E𝗅E_{\mathsf{l}}, we first prove that z¯=y¯M^x¯\overline{z}=\overline{y}-\widehat{M}\overline{x} in accordance with Equation 1. Assume this is false for some entry p𝔽mp\in\mathbb{F}^{m}, i.e.,

z¯(p)y¯(p)M^(p)x¯(p).\overline{z}(p)\neq\overline{y}(p)-\widehat{M}(p)\overline{x}(p). (2)

Note that x¯,y¯,M^\overline{x},\overline{y},\widehat{M} are all of parallel degree-dd and z¯\overline{z} is of parallel degree-2d2d. Then by Schwartz-Zippel lemma, Equation 2 actually happens for at least 12d|𝔽|1-\frac{2d}{|\mathbb{F}|} fraction of points p𝔽mp\in\mathbb{F}^{m}. Now recall the test in Item (T3), which checks precisely the above for a random p𝔽mp\sim\mathbb{F}^{m} with x¯,y¯,z¯\overline{x},\overline{y},\overline{z} replaced by x^,y^,z^\widehat{x},\widehat{y},\widehat{z}. By Items 1 and 2 and a union bound, with probability at least 15δ2d|𝔽|1-5\delta-\frac{2d}{|\mathbb{F}|}, on this random pp we have x^(p)=x¯(p),y^(p)=y¯(p),z^(p)=z¯(p)\widehat{x}(p)=\overline{x}(p),\widehat{y}(p)=\overline{y}(p),\widehat{z}(p)=\overline{z}(p) and Equation 2 happens, which makes Item (T3) reject. By our assumption on |𝔽||\mathbb{F}| and δ1/κ2\delta\leq 1/\kappa^{2} with κ\kappa sufficiently large, this rejection probability is at least 0.9>3κδ0.9>3\kappa\cdot\delta and contradicts to our assumption on the rejection probability of Item (T3). In short, Equation 2 can never happen.

Finally we are ready to show that constraints in E𝗅E_{\mathsf{l}} are satisfied by σ\sigma. By Item 2 and Fact 4.19, σz¯(i)=0t\sigma_{\overline{z}}(i)=0^{t} holds for all i[k]i\in[k], and thus

σy¯(i)Miσx¯(i)=y¯(ξi)M^(ξi)x¯(ξi)=z¯(ξi)=σz¯(i)=0.\sigma_{\overline{y}}(i)-M_{i}\sigma_{\overline{x}}(i)=\overline{y}(\xi_{i})-\widehat{M}(\xi_{i})\overline{x}(\xi_{i})=\overline{z}(\xi_{i})=\sigma_{\overline{z}}(i)=0.

Therefore, all constraints in E𝗅E_{\mathsf{l}} are also satisfied. This completes the whole soundness proof. ∎

4.3 Putting Everything Together

Now, we are ready to prove the main theorems.

Proof of Theorem 4.1.

We start with an arbitrary nn-variable 4-Regular 3-Coloring instance Γ\Gamma, which prohibits algorithms of runtime 2o(n)2^{o(n)} in the worst case by Theorem 3.2 assuming ETH. By Theorem 4.4, we obtain an SVecCSP instance G=(V,E,𝔽4t,{Πe}eE)G=(V,E,\mathbb{F}_{4}^{t},\{\Pi_{e}\}_{e\in E}) in time 𝗉𝗈𝗅𝗒(n,2nlogk/k)\mathsf{poly}(n,2^{n\log k/k}) which preserves the satisfiability of Γ\Gamma. In addition, t=O(nlogkk)t=O\left(\frac{n\log k}{k}\right) and |V|=O(k),|E|=O(k)|V|=O(k),|E|=O(k).

Let mm and dd be integers to be chosen later satisfying

k(m+dd).k\leq\binom{m+d}{d}. (3)

Let 𝔽\mathbb{F} be a field of characteristic two which contains 𝔽4\mathbb{F}_{4} as a subfield and satisfies

|𝔽|max{12md,2101mlog|𝔽|}.|\mathbb{F}|\geq\max\left\{12md,2^{101}m\log|\mathbb{F}|\right\}. (4)

By Fact 4.8, GG is satisfiable iff there exists x^,y^(𝔽t)|𝔽|m\widehat{x},\widehat{y}\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} such that (G,(x^,y^))(𝔽,m,d,t)(G,(\widehat{x},\widehat{y}))\in(\mathbb{F},m,d,t)-SVSat. Then we construct the PCPP verifier 𝒫\mathcal{P} for (𝔽,m,d,t)(\mathbb{F},m,d,t)-SVSat from Theorem 4.10 with δ=1\delta=1 to obtain a PCP verifier (recall Definition 3.8) 𝒫\mathcal{P}^{\prime} for the satisfiability of GG.

The query complexity, completeness, alphabet, and randomness of 𝒫\mathcal{P}^{\prime} follow from those of 𝒫\mathcal{P} in Theorem 4.10; and the soundness of 𝒫\mathcal{P}^{\prime} is the (unspecified) constant soundness parameter by setting δ=1\delta=1 in Theorem 4.10. In particular,

the alphabet size is |Σ|t=|𝔽|(d+1)t=|𝔽|O(dnlogk/k)\text{the alphabet size is }|\Sigma|^{t}=|\mathbb{F}|^{(d+1)\cdot t}=|\mathbb{F}|^{O(dn\log k/k)}

and

the randomness is log(k+|𝔽|m)+O(log0.1(k+|𝔽|m)+log|𝔽|) coins.\text{the randomness is }\log(k+|\mathbb{F}|^{m})+O\left(\log^{0.1}(k+|\mathbb{F}|^{m})+\log|\mathbb{F}|\right)\text{ coins.}

Then we apply Fact 3.11 and obtain a 2CSP instance Λ\Lambda preserving the satisfiability of GG (and thus Γ\Gamma) where

  • the size of the alphabet of Λ\Lambda is

    N=|𝔽|O(dnlogk/k),N=|\mathbb{F}|^{O(dn\log k/k)},
  • the number of variables in Λ\Lambda is at most

    K=2|𝔽|m+2log(k+|𝔽|m)+O(log0.1(k+|𝔽|m)+log|𝔽|)O(1)=(k+|𝔽|m)𝗉𝗈𝗅𝗒(|𝔽|,2log0.1(k+|𝔽|m)),K=2|\mathbb{F}|^{m}+2^{\log(k+|\mathbb{F}|^{m})+O\left(\log^{0.1}(k+|\mathbb{F}|^{m})+\log|\mathbb{F}|\right)}\cdot O(1)=(k+|\mathbb{F}|^{m})\cdot\mathsf{poly}\left(|\mathbb{F}|,2^{\log^{0.1}(k+|\mathbb{F}|^{m})}\right),
  • the number of constraints in Λ\Lambda is a constant multiple of the number of variables in Λ\Lambda.

Finally we optimize the choice of m,d,|𝔽|m,d,|\mathbb{F}|. Assume logk\log k is a perfect square and is sufficiently large and set

|𝔽|=21000logk2logk,m=logk,d=230logk2logk.|\mathbb{F}|=2^{1000}\log k\cdot 2^{\sqrt{\log k}},\quad m=\sqrt{\log k},\quad d=2^{30}\sqrt{\log k}\cdot 2^{\sqrt{\log k}}.

Then

(m+dd)(dm)m=(2302logk)logkk,\binom{m+d}{d}\geq\left(\frac{d}{m}\right)^{m}=\left(2^{30}\cdot 2^{\sqrt{\log k}}\right)^{\sqrt{\log k}}\geq k,

which is consistent with Equation 3. We also have

12md=12230logk2logk|𝔽|and\displaystyle 12md=12\cdot 2^{30}\cdot\log k\cdot 2^{\sqrt{\log k}}\leq|\mathbb{F}|\quad\text{and}
2101mlog|𝔽|2101logk(logk+loglogk+1000)|𝔽|,\displaystyle 2^{101}m\log|\mathbb{F}|\leq 2^{101}\sqrt{\log k}\left(\sqrt{\log k}+\log\log k+1000\right)\leq|\mathbb{F}|,

which is consistent with Equation 4. Moreover, Λ\Lambda has alphabet size NN and the number of variables KK as follows.

N=(logk2logk)O(nlog1.5k2logkk)andK=k2O(logkloglogk)N=\left(\log k\cdot 2^{\sqrt{\log k}}\right)^{O\left(\frac{n\cdot\log^{1.5}k\cdot 2^{\sqrt{\log k}}}{k}\right)}\quad\text{and}\quad K=k\cdot 2^{O(\sqrt{\log k}\log\log k)}

where NN can be furthered simplified to

N=(2n/k)2O(logk),N=\left(2^{n/k}\right)^{2^{O(\sqrt{\log k})}},

Since Γ\Gamma has no 2o(n)2^{o(n)}-time algorithm by assumption, Λ\Lambda has the lower bound f(K)NK/2ω(logKloglogK)f(K)\cdot N^{K/2^{\omega({\sqrt{\log K}\log\log K})}}-time for any computable function ff as claimed.

The PCP statement Theorem 4.2 follows directly from the proof above.

Proof of Theorem 4.2.

From any instance of 3SAT, there is a linear-size reduction to an instance of 4-Regular 3-Coloring [36]. Therefore we only need to construct the desired PCP for 4-Regular 3-Coloring. This follows directly from the verifier 𝒫\mathcal{P}^{\prime} in the proof of Theorem 4.1. In particular, we stick to the parameter choice there and obtain a PCP verifier with alphabet size

2n2O(logk)/k2^{n\cdot 2^{O(\sqrt{\log k})}/k}

and randomness

logk+O(logkloglogk).\log k+O\left(\sqrt{\log k}\log\log k\right).

The implicit constant soundness of 𝒫\mathcal{P}^{\prime} can be boosted to 12\frac{1}{2} by a constant number of randomness-efficient query repetitions (see e.g., [8, Lemma 2.11]). ∎

5 From 4-Regular 3-Coloring to SVecCSP

The goal of this section is to reduce 4-Regular 3-Coloring, which is known to be ETH-hard Theorem 3.2, to SVecCSP.

Theorem (Theorem 4.4 Restated).

There is a reduction algorithm such that the following holds. Given as input an integer 6kn6\leq k\leq n and an nn-variable 4-Regular 3-Coloring instance Γ\Gamma, it produces an SVecCSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) where:

  1. (R1)

    Variables and Constraints. |V|=O(k)|V|=O(k) and |E|=O(k)|E|=O(k).

  2. (R2)

    Runtime. The reduction runs in time 𝗉𝗈𝗅𝗒(n,2nlogk/k)\mathsf{poly}(n,2^{n\log k/k}).

  3. (R3)

    Alphabet. Σ=𝔽4t\Sigma=\mathbb{F}_{4}^{t} where t=O(nlogkk)t=O\left(\frac{n\log k}{k}\right).

  4. (R4)

    Satisfiability. GG is satisfiable iff Γ\Gamma is satisfiable.

The reduction starts by grouping vertices into supernodes, which take vector values. Then the constraints between supernodes correspond to (possibly multiple) constraints in the original instance. To make sure the new instance has small size, we need a grouping method (Lemma 5.1) that produces as few supernodes and constraints as possible. Then we make duplicates of variables and rearrangements of their coordinates to make sure parallel constraints are scattered properly (Proposition 5.4). Finally we make more duplicates to ensure that linear constraints form a matching and parallel constraints are applied on all coordinates (Proposition 5.5).

5.1 From 4-Regular 3-Coloring to VecCSP

The following Lemma 5.1 serves the purpose of the grouping method and can be seen as a parameterized version of the sparsification lemma [24, 25] in classical computation complexity.

Lemma 5.1 ([36, 29]).

There is an algorithm 𝒜\mathcal{A} such that the following holds. 𝒜\mathcal{A} takes as input a 2CSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) and an integer 6k|V|6\leq k\leq|V|, outputs a 2CSP instance G=(V,E,Σt,{Πe}eE)G^{\prime}=(V^{\prime},E^{\prime},\Sigma^{t},\{\Pi^{\prime}_{e}\}_{e\in E^{\prime}}) in time 𝗉𝗈𝗅𝗒(|V|,|Σ|t)\mathsf{poly}\left(|V|,|\Sigma|^{t}\right) where |V|=k|V^{\prime}|=k and tO((|V|+|E|)logkk)t\leq O\left((|V|+|E|)\cdot\frac{\log k}{k}\right), such that GG is satisfiable iff GG^{\prime} is satisfiable. In addition, the constraint graph of GG^{\prime} is a 3-regular graph.

In the actual construction, each xVx\in V^{\prime} corresponds to a subset S(x)VS(x)\subseteq V of size tt and takes values in ΣS(x)\Sigma^{S(x)} as assignments to variables in S(x)S(x). For each e={x,y}Ee=\left\{x,y\right\}\in E^{\prime}, the constraint Πe\Pi_{e} is the conjunction of the following:

  1. 1.

    equality constraints on common variables of S(x)S(x) and S(y)S(y);

  2. 2.

    constraints across666The construction in [29] ensures that each original constraint (u,v)E(u,v)\in E is covered by some new constraint (x,y)E(x,y)\in E^{\prime} in the sense that uS(x),vS(y)u\in S(x),v\in S(y). This means that we only need to check the cross constraints in GG^{\prime} and omit the ones purely inside S(x)S(x) or inside S(y)S(y). S(x)S(x) and S(y)S(y) in GG.

Remark 5.2.

Note that since the constraint graph of GG^{\prime} is 3-regular, the total number of constraints in GG^{\prime} is linear in |V||V^{\prime}|. In addition, tt only incurs an extra logk=ko(1)\log k=k^{o(1)} blowup over the information theoretic limit (|V|+|E|)/k(|V|+|E|)/k. This is crucial for us to get almost tight hardness. Indeed, naive approaches (e.g., the reduction in [21]) will incur a polynomial loss. We also remark that it is an open problem whether the extra logk\log k can be further removed, which, if true, implies a precise parameterized analog of the sparsification lemma and has many applications. See [36] for detailed discussions.

Given Lemma 5.1, we first obtain a vector-valued CSP (VecCSP) instance (Proposition 5.4), which we will soon convert into SVecCSP (Proposition 5.5).

Definition 5.3 (VecCSP, [21]).

A CSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) is a VecCSP if the following properties hold.

  • Σ=𝔽t\Sigma=\mathbb{F}^{t} is a tt-dimensional vector space over a finite field 𝔽\mathbb{F} with 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2.

  • For each constraint e={u,v}Ee=\{u,v\}\in E where u=(u1,,ut)u=(u_{1},\dots,u_{t}) and v=(v1,,vt)v=(v_{1},\ldots,v_{t}) are two variables in VV, the constraint validity function Πe\Pi_{e} is classified as one of the following cases:

    • Linear. There exists a matrix Me𝔽t×tM_{e}\in\mathbb{F}^{t\times t} such that

      Πe(u,v)=𝟙u=Mev.\Pi_{e}(u,v)=\mathbbm{1}_{u=M_{e}v}.
    • Parallel. There exists a sub-constraint Πesub:𝔽×𝔽{0,1}\Pi_{e}^{sub}:\mathbb{F}\times\mathbb{F}\to\{0,1\} and a subset of coordinates Qe[t]Q_{e}\subseteq[t] such that Πe\Pi_{e} checks Πesub\Pi_{e}^{sub} for every coordinate in QeQ_{e}, i.e.,

      Πe(u,v)=iQeΠesub(ui,vi).\Pi_{e}(u,v)=\bigwedge_{i\in Q_{e}}\Pi_{e}^{sub}(u_{i},v_{i}).
  • Each variable is related to at most one parallel constraint.

Note that SVecCSP, the special case of VecCSP we use, additionally enforces linear constraints to be a matching and enforces parallel constraints to operate on all coordinates (i.e., Qe=[t]Q_{e}=[t]).

Proposition 5.4 (VecCSP Intermediate Instance).

There is a reduction algorithm such that the following holds. Given as input an integer 6kn6\leq k\leq n and an nn-variable 4-regular 3-Coloring instance Γ\Gamma, it produces an VecCSP instance G=(V,E,Σ,{Πe}eE)G=(V,E,\Sigma,\{\Pi_{e}\}_{e\in E}) where:

  • Variables and Constraints. |V|=O(k)|V|=O(k) and |E|=O(k)|E|=O(k).

  • Runtime. The reduction runs in time 𝗉𝗈𝗅𝗒(n,2nlogk/k)\mathsf{poly}(n,2^{n\log k/k}).

  • Alphabet. Σ=𝔽4t\Sigma=\mathbb{F}_{4}^{t} where t=O(nlogkk)t=O\left(\frac{n\log k}{k}\right).

  • Satisfiability. GG is satisfiable iff Γ\Gamma is satisfiable.

Proof.

We first plug the 4-Regular 3-Coloring instance Γ\Gamma from Theorem 3.2 into Lemma 5.1 and obtain a 2CSP instance G=(V,E,𝔽4t,{Πe}eE)G^{\prime}=(V^{\prime},E^{\prime},\mathbb{F}_{4}^{t},\{\Pi^{\prime}_{e}\}_{e\in E^{\prime}}). By the construction in Lemma 5.1, each xVx\in V^{\prime} corresponds to a set S(x)S(x) of tt variables in Γ\Gamma. We fix an arbitrary order in S(x)S(x) and use x[i]x[i] to denote the ii-th variable in S(x)S(x). From the fact that Γ\Gamma is 4-regular and the construction in Lemma 5.1, the produced 2CSP instance GG^{\prime} has the following properties:

  • for each {x,y}E\left\{x,y\right\}\in E^{\prime} and i[t]i\in[t], there are at most five sub-constraints between x[i]x[i] and {y[j]:j[t]}\{y[j]:j\in[t]\}: at most one equality check (Item 1 of Lemma 5.1) and at most four coloring checks (Item 2 of Lemma 5.1);

  • the constraint graph of GG^{\prime} is 33-regular;

  • |V|=k|V^{\prime}|=k, |E|=O(k)|E^{\prime}|=O(k), t=O(nlogkk)t=O\left(\frac{n\log k}{k}\right), and the runtime is 𝗉𝗈𝗅𝗒(n,k,4t)=𝗉𝗈𝗅𝗒(n,2nlogk/k)\mathsf{poly}(n,k,4^{t})=\mathsf{poly}(n,2^{n\log k/k});

  • GG^{\prime} is satisfiable iff Γ\Gamma is satisfiable.

Refer to caption
Figure 1: An example of GG^{\prime} and G′′G^{\prime\prime} and the permutation to parallelize sub-constraints.

Thus, we duplicate each xVx\in V^{\prime} into constant many copies, and distribute the sub-constraints in GG^{\prime} onto different copies. This produces another 2CSP instance G′′=(V′′,E′′,𝔽4t,{Πe′′}eE′′)G^{\prime\prime}=(V^{\prime\prime},E^{\prime\prime},\mathbb{F}_{4}^{t},\{\Pi_{e}^{\prime\prime}\}_{e\in E^{\prime\prime}}) where

  • for each e={u,v}E′′e=\left\{u,v\right\}\in E^{\prime\prime}, the constraint Πe′′\Pi^{\prime\prime}_{e} has exactly one type of sub-constraint (i.e., equality or coloring), which forms a partial (non-parallel) matching across the coordinates (i.e., S(u),S(v)S(u),S(v)) of u,vu,v;

    (Note that the consistency checks among duplicates will be added later.)

  • each variable in V′′V^{\prime\prime} is related to exactly one constraint;

  • |V′′|=O(|V|)=O(k)|V^{\prime\prime}|=O(|V^{\prime}|)=O(k) and |E′′|=|E|=O(k)|E^{\prime\prime}|=|E^{\prime}|=O(k).

The above procedure is efficient: we only need to perform matching decompositions for each {x,y}E\left\{x,y\right\}\in E^{\prime} separately for equality checks and coloring checks.

Before adding the consistency checks among duplicates, we first permute coordinates of each variable in V′′V^{\prime\prime} to parallelize the partial matchings. This is possible since each variable in V′′V^{\prime\prime} is related to exactly one constraint in G′′G^{\prime\prime}. For a fixed xVx\in V^{\prime}, let x1,,xmV′′x_{1},\ldots,x_{m}\in V^{\prime\prime} be the duplicates of xVx\in V^{\prime}. After the permutation, we add linear constraints between xix_{i} and xi+1x_{i+1} for 1i<m1\leq i<m to check whether they are consistent (i.e., the correct permuted copies of each other). See Figure 1 for a streamlined presentation.

The construction of the VecCSP instance G=(V,E,Σ=𝔽4t,{Πe}eE)G=(V,E,\Sigma=\mathbb{F}_{4}^{t},\{\Pi_{e}\}_{e\in E}) is completed after the permutation and adding the consistency checks among duplicated. The satisfiability is naturally preserved, |V|=|V′′|=O(k)|V|=|V^{\prime\prime}|=O(k), and |E||E′′|+|V′′|=O(k)|E|\leq|E^{\prime\prime}|+|V^{\prime\prime}|=O(k). In terms of Definition 5.3, the consistency checks are linear constraints and the constraints in G′′G^{\prime\prime} after permutation are parallel constraints. ∎

5.2 From VecCSP to SVecCSP

Given Proposition 5.4, Theorem 4.4 follows directly by the following general reduction from VecCSP to SVecCSP.

Proposition 5.5 (VecCSP to SVecCSP).

There is a reduction algorithm such that the following holds. Given as input a VecCSP instance G=(V,E,𝔽t,{Πe}eE)G=(V,E,\mathbb{F}^{t},\{\Pi_{e}\}_{e\in E}), it produces an SVecCSP instance G=(V,E,𝔽t,{Πe}eE)G^{\prime}=(V^{\prime},E^{\prime},\mathbb{F}^{t},\{\Pi^{\prime}_{e}\}_{e\in E^{\prime}}) where:

  • Variables and Constraints. |V|=O(|V|+|E|)|V^{\prime}|=O(|V|+|E|) and |E|=O(|V|+|E|)|E^{\prime}|=O(|V|+|E|).

  • Runtime. The reduction runs in time 𝗉𝗈𝗅𝗒(|V|,|E|,|𝔽|t)\mathsf{poly}(|V|,|E|,|\mathbb{F}|^{t}).

  • Satisfiability. GG^{\prime} is satisfiable iff GG is satisfiable.

Proof.

Now that we get a VecCSP instance GG, we show how to modify it to to obtain an SVecCSP instance GG^{\prime} that satisfies properties Item (S3) and Item (S4). The construction consists of two steps and see Figure 2 for a streamlined presentation.

  • First, from GG, get another VecCSP instance G^\widehat{G} which satisfies Item (S3).

  • Next, based on G^\widehat{G}, build a SVecCSP instance GG^{\prime} which in addition satisfies Item (S4).

To satisfy Item (S3), we split each variable xx in VV into a parallel variable x𝗉x^{\mathsf{p}} and a linear variable x𝗅x^{\mathsf{l}} in G^\widehat{G} for parallel and linear constraints separately. Then we construct the constraints E^\widehat{E} in G^\widehat{G}.

  • For each linear constraint e={x,y}Ee=\{x,y\}\in E, we add the same linear constraint on {x𝗅,y𝗅}\{x^{\mathsf{l}},y^{\mathsf{l}}\} in E^\widehat{E}.

  • For each parallel constraint e={x,y}Ee=\{x,y\}\in E with sub-constraint Πesub\Pi_{e}^{sub} and subset of coordinates QeQ_{e}, we add a parallel constraint on {x𝗉,y𝗉}\{x^{\mathsf{p}},y^{\mathsf{p}}\} in E^\widehat{E}, which has Πesub\Pi_{e}^{sub} applied on all coordinates.

  • We need to additionally check in G^\widehat{G} the partial equality between x𝗉x^{\mathsf{p}} and x𝗅x^{\mathsf{l}} only on the subset of coordinates QeQ_{e}. Since each variable xx is related to at most one parallel constraint as guaranteed in Definition 5.3, this additional check is well-defined.

    This check can be written as MQex𝗉=MQex𝗅M_{Q_{e}}x^{\mathsf{p}}=M_{Q_{e}}x^{\mathsf{l}}, where MQeM_{Q_{e}} is a matrix which projects on coordinates inside QeQ_{e} and zeroing out coordinates outside QeQ_{e}. To have the matrix only on one side, we introduce an additional variable x𝖺x^{\sf a} in G^\widehat{G}, and add two linear constraints x𝖺=MQex𝗉x^{\sf a}=M_{Q_{e}}x^{\mathsf{p}} and x𝖺=MQex𝗅x^{\sf a}=M_{Q_{e}}x^{\mathsf{l}} to E^\widehat{E}.

We first show that the above construction preserves the satisfiability as follows.

  • Given a solution σ\sigma for the original GG, assign σ(x)\sigma(x) to x𝗅x^{\mathsf{l}} and x𝖺x^{\sf a}, which satisfies all linear constraints in E^\widehat{E}. Then assign σ(x)\sigma(x) to x𝗉x^{\mathsf{p}} on the subset QeQ_{e} of coordinates, which satisfies all the partial equality checks in E^\widehat{E}. Finally, assign arbitrary solution777Technically it is possible that Πesub\Pi_{e}^{sub} is not satisfiable. If Qe=Q_{e}=\emptyset, then we can simply replace Πesub\Pi_{e}^{sub} by any satisfiable sub-constraint. If otherwise QeQ_{e}\neq\emptyset, then the original GG is not satisfiable and G^\widehat{G} is also not satisfiable. Therefore the construction still works. of Πesub\Pi_{e}^{sub} to x𝗉x^{\mathsf{p}} on coordinates outside QeQ_{e}, which satisfies all the parallel constraints in E^\widehat{E}.

  • Given a solution σ\sigma^{\prime} of G^\widehat{G}, assign σ(x𝗅)\sigma^{\prime}(x^{\mathsf{l}}) to every xx in GG, which satisfies all linear constraints in EE. Since σ(x𝗉)\sigma^{\prime}(x^{\mathsf{p}}) satisfies the parallel constraint in GG^{\prime} for all coordinates and the partial equality check guarantees consistency between σ(x𝗅)\sigma^{\prime}(x^{\mathsf{l}}) and σ(x𝗉)\sigma^{\prime}(x^{\mathsf{p}}) on the coordinates QeQ_{e}, the corresponding parallel constraints in EE are satisfied as well.

Moreover, the variable set V^xV{x𝗉,x𝗅,x𝖺}\widehat{V}\subseteq\bigcup_{x\in V}\left\{x^{\mathsf{p}},x^{\mathsf{l}},x^{\sf a}\right\} has size |V^|=O(|V|)|\widehat{V}|=O(|V|) and the constraint set E^\widehat{E} has size |E^||E|+2|V|=O(|V|+|E|)|\widehat{E}|\leq|E|+2\cdot|V|=O(|V|+|E|).

Now we construct GG^{\prime} from G^\widehat{G} to satisfy Item (S4). The final variable set of GG^{\prime} will be V=X˙YV^{\prime}=X\dot{\cup}Y, which is constructed along with the constraint set E=E𝗅˙E𝗉E^{\prime}=E^{\prime}_{\mathsf{l}}\dot{\cup}E^{\prime}_{\mathsf{p}} as follows.

  • Initialize X,YX,Y as disjoint copies of V^\widehat{V} and initialize E𝗅=E𝗉=E^{\prime}_{\mathsf{l}}=E^{\prime}_{\mathsf{p}}=\emptyset. For a variable uV^u\in\widehat{V}, we denote as xu,yux_{u},y_{u} its XX-copy and YY-copy in VV^{\prime}, respectively.

  • For each uV^u\in\widehat{V}, we add an equality, which is a linear constraint with the identity matrix, in E𝗅E^{\prime}_{\mathsf{l}} between xux_{u} and yuy_{u}.

    Note that this is consistent with Item (S4).

  • Then for each parallel constraint e={x𝗉,y𝗉}E^e=\left\{x^{\mathsf{p}},y^{\mathsf{p}}\right\}\in\widehat{E}, we add the same constraint in E𝗉E^{\prime}_{\mathsf{p}} on the XX-copies of x𝗉x^{\mathsf{p}} and y𝗉y^{\mathsf{p}}.

  • Finally for each linear constraint888In particular, e={u,v}e=\left\{u,v\right\} can be {x𝗅,y𝗅}\left\{x^{\mathsf{l}},y^{\mathsf{l}}\right\} or {x𝗅,x𝖺}\left\{x^{\mathsf{l}},x^{\sf a}\right\} or {x𝗉,x𝖺}\left\{x^{\mathsf{p}},x^{\sf a}\right\}. e={u,v}E^e=\{u,v\}\in\widehat{E} that checks u=Mevu=M_{e}v, we add new variables xex_{e} to XX and yey_{e} to YY. Then we impose a linear constraint ye=Mexey_{e}=M_{e}x_{e} between them in E𝗅E^{\prime}_{\mathsf{l}}, which is consistent with Item (S4). We further add two equality constraints, which are identified as parallel constraints in E𝗉E^{\prime}_{\mathsf{p}}, between xex_{e} and xux_{u}, as well as between yey_{e} and xvx_{v}.

The construction of GG^{\prime} preserves the satisfiability of G^\widehat{G} as all the duplicates in GG^{\prime} of variables in G^\widehat{G} are connected by identity constraints. Moreover, |X|=|Y||V^|+|E^|=O(|V|+|E|)|X|=|Y|\leq|\widehat{V}|+|\widehat{E}|=O(|V|+|E|) and |E||V^|+|E^|+3|E^|=O(|V|+|E|)|E^{\prime}|\leq|\widehat{V}|+|\widehat{E}|+3\cdot|\widehat{E}|=O(|V|+|E|) as desired. ∎

Refer to caption
Figure 2: An illustration of the reduction from GG to G^\widehat{G}, and from G^\widehat{G} to GG^{\prime}.

6 PCPP for Multi-Test Problem

In this section we prove Theorem 4.14 for DoubleTest and Theorem 4.20 for SingleTest. For convenience, we define the following multi-test problem (MultiTest) which generalizes DoubleTest and SingleTest in a straightforward fashion.

Definition 6.1 (MultiTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2. (𝔽,m,d,t,u)(\mathbb{F},m,d,t,u)-MultiTest is a pair language over Σ𝗑={0,1},Σ𝗒=𝔽t\Sigma_{\mathsf{x}}=\{0,1\},\Sigma_{\mathsf{y}}=\mathbb{F}^{t} consisting of all words in the form of w=(C,T1Tu)w=(C,T_{1}\circ\cdots\circ T_{u}), where

  • CC is a Boolean circuit with u|𝔽|mlog|𝔽|u\cdot|\mathbb{F}|^{m}\log|\mathbb{F}| input bits and T1,,Tu(𝔽t)|𝔽|mT_{1},\dots,T_{u}\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} are codewords of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t};

  • if we view 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, T1TuT_{1}\circ\cdots\circ T_{u} parallel satisfies CC.

We define the input length |w||w| to be the size of CC plus u|𝔽|mlog|𝔽|u\cdot|\mathbb{F}|^{m}\log|\mathbb{F}|.

By setting u=1u=1 or u=2u=2, we immediately obtain (𝔽,m,d,t)(\mathbb{F},m,d,t)-SingleTest or (𝔽,m,d,t)(\mathbb{F},m,d,t)-DoubleTest. Hence Theorem 4.20 and Theorem 4.14 follows directly from the following result for MultiTest.

Theorem 6.2 (PCPP for MultiTest).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. Assume u250u\leq 2^{50} is a positive integer. Then for any δ[0,1]\delta\in[0,1], (𝔽,d,m,t,u)(\mathbb{F},d,m,t,u)-MultiTest has

an (O(1),log|w|+O(log0.1|w|+log|𝔽|),δ,Ω(δ),Σt)\left(O(1),\log|w|+O\left(\log^{0.1}|w|+\log|\mathbb{F}|\right),\delta,\Omega(\delta),\Sigma^{t}\right)-PCPP verifer 𝒫𝗆𝗍\mathcal{P}_{\sf mt},

where |w||w| is the input length of (𝔽,d,m,t,u)(\mathbb{F},d,m,t,u)-MultiTest and Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1}.

Proof sketch

Let (C,T1Tu)(C,T_{1}\circ\cdots\circ T_{u}) be an input for (𝔽,d,m,t,u)(\mathbb{F},d,m,t,u)-MultiTest. Our goal is to check whether T1TuT_{1}\circ\cdots\circ T_{u} is δ\delta-close to some T1TuT_{1}^{*}\circ\cdots\circ T_{u}^{*}\in MultiTest(C)(C), i.e., the restriction of the pair language MultiTest on CC. In other words, the following two conditions hold:

  1. (C1)

    for each j[u]j\in[u], Tj𝔽|𝔽|mT_{j}^{*}\in\mathbb{F}^{|\mathbb{F}|^{m}} is the truth table of a polynomial of parallel degree dd;

  2. (C2)

    T1TuT_{1}^{*}\circ\cdots\circ T_{u}^{*}, viewed as a word in {0,1}u|𝔽|mlog|𝔽|\{0,1\}^{u|\mathbb{F}|^{m}\log|\mathbb{F}|}, parallel satisfies the Boolean circuit CC.

To guarantee Item (C1), we use the PCPP verifier 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} from the codeword testing of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} (see Theorem 3.6). Given T1TuT_{1}^{*}\circ\cdots\circ T_{u}^{*} satisfying Item (C1), we aim to test that it also satisfies Item (C2). That is, for each fixed i[t]i\in[t], T1[i]Tu[i]T_{1}^{*}[i]\circ\cdots\circ T_{u}^{*}[i] satisfies the Boolean circuit CC.

To this end, we will use the PCPP verifier 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} of CktVal (see Theorem 3.13). This alone, however, is not sufficient as 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} cannot rule out the case that changing o(1)o(1) fraction of entries in T1[i]Tu[i]T_{1}^{*}[i]\circ\cdots\circ T_{u}^{*}[i] satisfying the circuit CC. To fix this issue, we have to exploit the fact that each Tj[i]T_{j}^{*}[i] is supposed to be the truth table of a degree-dd polynomial, which, by Schwartz-Zippel lemma and the fact that uu is a constant, forbids such attacks. As a result, we need to incorporate the codeword testing circuit (see Theorem 3.7) to enforce the low degree condition.

Unfortunately, this still does not work due to a subtle alphabet mismatch: the codeword testing works over 𝔽\mathbb{F} but Item (C2) needs to flat 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}. Therefore, the distance guaranteed by the low degree condition can be dilated by a worst-case factor of log|𝔽|=ω(1)\log|\mathbb{F}|=\omega(1) after converting 𝔽\mathbb{F} to {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, for which reason the mentioned attack can still be carried out. To address this issue, we employ a standard approach to lift the conversion of 𝔽\mathbb{F} via error correcting codes [8, 4, 15]. More formally, after flattening 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, we take it through an error correcting code with constant rate and distance, which produces a codeword in {0,1}O(log|𝔽|)\{0,1\}^{O(\log|\mathbb{F}|)} and, more importantly, has constant relative distance against other codewords.

In summary, to handle Item (C2), we need to use 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} to check the validity of the combination of (1) the original circuit CC, (2) the codeword testing procedure, and (3) the error correcting lifting. This is parallel for each coordinate and is presented in Subsection 6.1. Then in Subsection 6.2, we put together the argument for Item (C1) and prove Theorem 6.2.

Remark 6.3.

One may wonder the necessity of using a separate codeword testing for Item (C1), as we anyway need to use it for Item (C2). The difference lies in the proximity: the former uses Theorem 3.6 which guarantees the parallel proximity (i.e., including all the dimensions [t][t]), whereas the latter uses Theorem 3.7 which only implies coordinate-wise proximity (i.e., individually for each coordinate i[t]i\in[t]).

Without the former, we can only get an Ω(tδ)\Omega(t\cdot\delta) final proximity via a union bound. Upgrading the latter needs to generalize the construction of 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} to the parallel setting, which arguably requires more work. Hence we stick to our current presentation for simplicity.

6.1 Single-Coordinate Checking Circuit

As sketched above, we construct another Boolean circuit CC^{\prime} for Item (C2), which augments the original circuit CC with an alphabet lifting via error correcting code and a low-degree check. Later, this single-coordinate checking circuit will be applied in parallel for every coordinate.

Lifted Flattening of 𝔽\mathbb{F}

We will use the following standard error correcting code to achieve such an alphabet lifting.

Proposition 6.4 (Binary ECC with Constant Rate and Distance [27]).

For every n1n\geq 1, there exists an efficiently computable error correcting code with the encoding map ECCn:{0,1}n{0,1}7n{\rm ECC}_{n}:\{0,1\}^{n}\to\{0,1\}^{7n} of relative distance δ(ECCn)0.01\delta({\rm ECC_{n}})\geq 0.01.

Let Enc:𝔽{0,1}7log|𝔽|\mathrm{Enc}:\mathbb{F}\to\{0,1\}^{7\log|\mathbb{F}|} to be ECClog|𝔽|{\rm ECC}_{\log|\mathbb{F}|} composed with the natural flattening of 𝔽\mathbb{F} into {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}. Let Dec\mathrm{Dec} be the corresponding decoding function.

Later putting back into the construction of 𝒫𝗆𝗍\mathcal{P}_{\sf mt}, we will apply Enc\mathrm{Enc} to every entry of T1[i]Tu[i]𝔽u|𝔽|mT_{1}[i]\circ\cdots\circ T_{u}[i]\in\mathbb{F}^{u|\mathbb{F}|^{m}} for every coordinate i[t]i\in[t]. For convenience, we define the parallel encoding map Enc:𝔽|𝔽|m{0,1}7|𝔽|mlog|𝔽|\mathrm{Enc}^{\odot}:\mathbb{F}^{|\mathbb{F}|^{m}}\to\{0,1\}^{7|\mathbb{F}|^{m}\log|\mathbb{F}|} as

Enc(z)(k,)=Enc(zk)for z𝔽|𝔽|m and k[|𝔽|m],\mathrm{Enc}^{\odot}(z)(k,\cdot)=\mathrm{Enc}(z_{k})\quad\text{for $z\in\mathbb{F}^{|\mathbb{F}|^{m}}$ and $k\in[|\mathbb{F}|^{m}]$,} (5)

where we view Enc(z){0,1}7|𝔽|log|𝔽|\mathrm{Enc}^{\odot}(z)\in\{0,1\}^{7|\mathbb{F}|\log|\mathbb{F}|} as a function [|𝔽|m]×[7log|𝔽|]{0,1}[|\mathbb{F}|^{m}]\times[7\log|\mathbb{F}|]\to\{0,1\} and the index kk points to the lifted flattening of the kk-th entry of zz. Given the definition of Enc\mathrm{Enc}^{\odot}, the lifted flattening of T1[i]Tu[i]T_{1}[i]\circ\cdots\circ T_{u}[i] can be simply represented as Enc(T1[i])Enc(Tu[i])\mathrm{Enc}^{\odot}(T_{1}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}[i]).

The Construction of CC^{\prime}

Given the lifted flattening of alphabet, we now present the construction. The circuit CC^{\prime} takes as input y1yuy_{1}\circ\cdots\circ y_{u} where each yj{0,1}7|𝔽|mlog|𝔽|y_{j}\in\{0,1\}^{7|\mathbb{F}|^{m}\log|\mathbb{F}|} is supposed to be Enc(y^j)\mathrm{Enc}^{\odot}(\widehat{y}_{j}) for some y^j𝔽|𝔽|m\widehat{y}_{j}\in\mathbb{F}^{|\mathbb{F}|^{m}}. Note that y^j\widehat{y}_{j} will be Tj[i]T_{j}[i], rolling over all coordinates i[t]i\in[t]. The circuit CC^{\prime} check the following three things in order:

  1. (S1)

    Check if each yjy_{j} is a codeword of Enc\mathrm{Enc}^{\odot} and, if so, compute each y^j\widehat{y}_{j} and view it as a binary string via the standard flattening of 𝔽\mathbb{F}.

    This requires to decode a total number of u|𝔽|mu\cdot|\mathbb{F}|^{m} words in {0,1}7log|𝔽|\{0,1\}^{7\log|\mathbb{F}|}, which can be has circuit size u|𝔽|m𝗉𝗈𝗅𝗒𝗅𝗈𝗀|𝔽||𝔽|m𝗉𝗈𝗅𝗒|𝔽|u|\mathbb{F}|^{m}\cdot\mathsf{polylog}|\mathbb{F}|\leq|\mathbb{F}|^{m}\mathsf{poly}|\mathbb{F}|.

  2. (S2)

    Check if each yi^\widehat{y_{i}} satisfies C𝗅𝖽𝗍C_{\mathsf{ldt}} from Theorem 3.7, i.e., y^i\widehat{y}_{i} is the truth table of a degree-dd polynomial.

    This has size |𝔽|m𝗉𝗈𝗅𝗒|𝔽||\mathbb{F}|^{m}\mathsf{poly}|\mathbb{F}| by Theorem 3.7.

  3. (S3)

    Check whether y1^yu^\widehat{y_{1}}\circ\cdots\circ\widehat{y_{u}} satisfies the original circuit CC.

    This has size precisely the size of CC.

Now we list properties of CC^{\prime}.

Fact 6.5 (Satisfiability).

If Enc(y1^)Enc(yu^)\mathrm{Enc}^{\odot}(\widehat{y_{1}})\circ\cdots\circ\mathrm{Enc}^{\odot}(\widehat{y_{u}}) satisfies the circuit CC^{\prime}, then y1^yu^\widehat{y_{1}}\circ\cdots\circ\widehat{y_{u}} satisfies the circuit CC.

Fact 6.6 (Size).

Recall from Definition 6.1 that |w||w| is the input length of (𝔽,m,d,t)(\mathbb{F},m,d,t)-MultiTest, which equals the size of CC plus u|𝔽|mlog|𝔽|u\cdot|\mathbb{F}|^{m}\log|\mathbb{F}|. The size of the circuit CC^{\prime} is at most |w|𝗉𝗈𝗅𝗒|𝔽||w|\cdot\mathsf{poly}|\mathbb{F}| and the input of CC^{\prime} has length O(|w|)O(|w|).

Claim 6.7 (Distance).

If y1yuy_{1}\circ\cdots\circ y_{u} passes Items (S1) and (S2) but not Item (S3), then it is 2602^{-60}-far from solutions of CC^{\prime}.

Proof.

Let z1zuz_{1}\circ\cdots\circ z_{u} be an arbitrary solution of CC^{\prime}, which means it passes Items (S1), (S2) and (S3). Let z^1,,z^u\widehat{z}_{1},\ldots,\widehat{z}_{u} be the decoding outcome from Item (S1). Then there exists j[u]j\in[u] such that y^jz^j\widehat{y}_{j}\neq\widehat{z}_{j}. Since they both pass Item (S2), y^j\widehat{y}_{j} and z^j\widehat{z}_{j}, viewed as an element from 𝔽|𝔽|m\mathbb{F}^{|\mathbb{F}|^{m}}, correspond to truth tables of distinct degree-dd polynomials. Hence Δ(y^j,z^j)1d|𝔽|>12\Delta(\widehat{y}_{j},\widehat{z}_{j})\geq 1-\frac{d}{|\mathbb{F}|}>\frac{1}{2} by Schwartz-Zippel lemma and our assumption on |𝔽||\mathbb{F}|. Recall from Equation 5 that Enc\mathrm{Enc}^{\odot} applies Enc\mathrm{Enc} entrywise. Since yj=Enc(y^j)y_{j}=\mathrm{Enc}^{\odot}(\widehat{y}_{j}) and zj=Enc(z^j)z_{j}=\mathrm{Enc}^{\odot}(\widehat{z}_{j}), we now have Δ(yj,zj)>0.0112\Delta(y_{j},z_{j})>0.01\cdot\frac{1}{2} by Proposition 6.4. Since u250u\leq 2^{50}, we have

Δ(y1yu,z1zu)Δ(yj,zj)u>260\Delta(y_{1}\circ\cdots\circ y_{u},z_{1}\circ\cdots\circ z_{u})\geq\frac{\Delta(y_{j},z_{j})}{u}>2^{-60}

as desired. ∎

6.2 Combining Codeword Testing

Now we are ready to combine the codeword testing for Item (C1) and the single-coordinate checking circuit CC^{\prime} for Item (C2) to prove Theorem 6.2.

Proof of Theorem 6.2.

The auxiliary proof for 𝒫𝗆𝗍\mathcal{P}_{\sf mt} consists of two parts.

  • The first part is π𝗅𝖽𝗍,1,,π𝗅𝖽𝗍,u\pi_{\mathsf{ldt},1},\dots,\pi_{\mathsf{ldt},u}. Each π𝗅𝖽𝗍,i\pi_{\mathsf{ldt},i} has alphabet Σt\Sigma^{t} and is constructed by Theorem 3.6, which is supposed to be the auxillary proof for codeword testing of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} on TiT_{i}.

  • The second part is denoted by π𝖼𝗄𝗍\pi_{\mathsf{ckt}}, which has the alphabet {0,1}t\{0,1\}^{t} naturally embedded into Σt\Sigma^{t}. For each coordinate i[t]i\in[t], π𝖼𝗄𝗍[i]\pi_{\mathsf{ckt}}[i] is constructed by Theorem 3.13 for 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} to check if Enc(T1[i])Enc(Tu[i])\mathrm{Enc}^{\odot}(T_{1}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}[i]) satisfies the circuit CC^{\prime}.

Testing Procedure of 𝒫𝗆𝗍\mathcal{P}_{\sf mt}

Now we describe the testing procedure. 𝒫𝗆𝗍\mathcal{P}_{\sf mt} executes one of the following two tests with equal probability.

  • For each i[u]i\in[u], 𝒫𝗆𝗍\mathcal{P}_{\sf mt} invokes 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} to run the codeword testing for 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} on Tiπ𝗅𝖽𝗍,iT_{i}\circ\pi_{\mathsf{ldt},i}. This checks whether TiIm(𝖱𝖬𝔽,m,d,t)T_{i}\in\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}).

  • 𝒫𝗆𝗍\mathcal{P}_{\sf mt} parallel simulates 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} to test if Enc(T1[i])Enc(Tu[i])π𝖼𝗄𝗍[i]\mathrm{Enc}^{\odot}(T_{1}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}[i])\circ\pi_{\mathsf{ckt}}[i] satisfies CC^{\prime} for all coordinates i[t]i\in[t].

    In detail, for each coordinate i[t]i\in[t], 𝒫𝗆𝗍\mathcal{P}_{\sf mt} tosses random coins as 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} does, and probes entries of π𝖼𝗄𝗍[i]\pi_{\mathsf{ckt}}[i] if needed. Whenever 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} needs to probe some bit of Enc(Tj[i])\mathrm{Enc}^{\odot}(T_{j}[i]), 𝒫𝗆𝗍\mathcal{P}_{\sf mt} queries the corresponding entry (i.e., the index kk in Equation 5) of Tj[i]T_{j}[i], performs the lifted flattening of 𝔽\mathbb{F} for that entry, and obtains the desired bit.

    We emphasize that the randomness used to simulate 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} is the same for all coordinates. Therefore, the queries by 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} are simulated in parallel for all coordinates i[t]i\in[t], as the query locations are uniquely determined by the randomness.

Parameters of 𝒫𝗆𝗍\mathcal{P}_{\sf mt}

Since uu is a constant and both 𝒫𝗅𝖽𝗍,𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ldt},\mathcal{P}_{\sf ckt} have constant queries (see Theorem 3.6 and Theorem 3.13), 𝒫𝗆𝗍\mathcal{P}_{\sf mt} makes constant queries.

Recall that the input length |w||w| equals the size of CC plus u|𝔽|mlog|𝔽|u|\mathbb{F}|^{m}\log|\mathbb{F}|. By Theorem 3.6, the first part tosses

mlog|𝔽|+O(loglog|𝔽|+logm)log|w|+O(logm)log|w|+O(log|𝔽|)m\log|\mathbb{F}|+O(\log\log|\mathbb{F}|+\log m)\leq\log|w|+O(\log m)\leq\log|w|+O(\log|\mathbb{F}|)

coins, where we used the assumption on |𝔽||\mathbb{F}|. By Theorem 3.13 and Fact 6.6, the second part tosses

log(|w|𝗉𝗈𝗅𝗒|𝔽|)+O(log0.1(|w|𝗉𝗈𝗅𝗒|𝔽|))log|w|+O(log0.1|w|+log|𝔽|)\log(|w|\cdot\mathsf{poly}|\mathbb{F}|)+O\left(\log^{0.1}(|w|\cdot\mathsf{poly}|\mathbb{F}|)\right)\leq\log|w|+O\left(\log^{0.1}|w|+\log|\mathbb{F}|\right)

coins. Since we only execute one of them, the number of random coins is the maximum of the above two as desired.

Completeness and Soundness

The completeness is straightforward by the completeness of 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} (see Theorem 3.6) and 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} (see Theorem 3.13) and the construction of CC^{\prime} (see Subsection 6.1). We focus on the soundness analysis: assuming that T1TuT_{1}\circ\cdots\circ T_{u} is δ\delta-far from MultiTest(C)(C) (i.e., we do not have Items (C1) and (C2)), 𝒫𝗆𝗍\mathcal{P}_{\sf mt} rejects with probability Ω(δ)\Omega(\delta). By modifying the hidden constant in Ω()\Omega(\cdot) and noticing that δ\delta-far implies δ\delta^{\prime}-far for any δδ\delta^{\prime}\leq\delta, we additionally assume δ2100\delta\leq 2^{-100}.

Assume towards contradiction that the above soundness statement is false. We first show that each TjT_{j} is close to being parallel degree-dd. This comes from the following Fact 6.8, which can be deduced directly from Theorem 3.6.

Fact 6.8.

If for some j[u]j\in[u], TjT_{j} is δ\delta-far from Im(𝖱𝖬𝔽,m,d,t)\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), then 𝒫𝗆𝗍\mathcal{P}_{\sf mt} rejects with probability Ω(δ)\Omega(\delta).

Now we assume each TjT_{j} is δ\delta-close to some TjIm(𝖱𝖬𝔽,m,d,t)T_{j}^{*}\in\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), which corresponds to Item (C1). Next, we show T1TuT_{1}^{*}\circ\cdots\circ T_{u}^{*} parallel satisfies CC, which corresponds to Item (C2). By Fact 6.5, it suffices to show for each coordinate i[t]i\in[t] that Enc(T1[i])Enc(Tu[i])\mathrm{Enc}^{\odot}(T_{1}^{*}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}^{*}[i]) satisfies CC^{\prime}, which is precisely the following Claim 6.9.

Claim 6.9.

For each coordinate i[t]i\in[t], Enc(T1[i])Enc(Tu[i])\mathrm{Enc}^{\odot}(T_{1}^{*}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}^{*}[i]) satisfies Items (S1), (S2) and (S3).

Given Claim 6.9, we arrive at a contradiction and complete the soundness analysis. ∎

Finally we prove Claim 6.9.

Proof of Claim 6.9.

Note that Item (S1) is already satisfied. Since each TjIm(𝖱𝖬𝔽,m,d,t)T_{j}^{*}\in\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), each Tj[i]T_{j}^{*}[i] is the truth table of a degree-dd polynomial, which means Item (S2) is also satisfied. If Item (S3) is false for some i[t]i\in[t], then by Claim 6.7, Enc(T1[i])Enc(Tu[i])\mathrm{Enc}^{\odot}(T_{1}^{*}[i])\circ\cdots\circ\mathrm{Enc}^{\odot}(T_{u}^{*}[i]) is 2602^{-60}-far from any solution of CC^{\prime}. Observe that

Δ(Enc(T1[i])Tu[i],Enc(T1[i])Tu[i])\displaystyle\Delta(\mathrm{Enc}^{\odot}(T_{1}[i])\circ\cdots\circ T_{u}[i],\mathrm{Enc}^{\odot}(T_{1}^{*}[i])\circ\cdots\circ T_{u}^{*}[i]) Δ(Enc(T1)Tu,Enc(T1)Tu)\displaystyle\leq\Delta(\mathrm{Enc}^{\odot}(T_{1})\circ\cdots\circ T_{u},\mathrm{Enc}^{\odot}(T_{1}^{*})\circ\cdots\circ T_{u}^{*})
maxj[u]Δ(Enc(Tj),Enc(Tj))\displaystyle\leq\max_{j\in[u]}\Delta(\mathrm{Enc}^{\odot}(T_{j}),\mathrm{Enc}^{\odot}(T_{j}^{*})) (since Δ\Delta is relative distance)
δ.\displaystyle\leq\delta. (by the choice of TjT_{j}^{*})

Since δ2100\delta\leq 2^{-100}, we know that Enc(T1[i])Tu[i]\mathrm{Enc}^{\odot}(T_{1}[i])\circ\cdots\circ T_{u}[i] is 2702^{-70}-far from solutions of CC^{\prime}. By Theorem 3.13, this means 𝒫𝗆𝗍\mathcal{P}_{\sf mt}, which executes 𝒫𝖼𝗄𝗍\mathcal{P}_{\sf ckt} with half probability, rejects with probability 1212=Ω(δ)\frac{1}{2}\cdot\frac{1}{2}=\Omega(\delta). Recall that we assumed that 𝒫𝗆𝗍\mathcal{P}_{\sf mt} does not reject with probability Ω(δ)\Omega(\delta), which is a contradiction. Thus Item (S3) should also be satisfied and this completes the proof of Claim 6.9. ∎

Acknowledgement

We thank Eli Ben-Sasson for clarifying questions regarding [9] and thank Karthik C.S. for pointing out the application to Max kk-Coverage (Theorem 1.6).

References

  • AB [09] Sanjeev Arora and Boaz Barak. Computational Complexity: A Modern Approach. Cambridge University Press, USA, 1st edition, 2009.
  • ABSS [97] Sanjeev Arora, László Babai, Jacques Stern, and Z. Sweedyk. The hardness of approximate optima in lattices, codes, and systems of linear equations. J. Comput. Syst. Sci., 54(2):317–331, 1997.
  • AFWZ [95] Noga Alon, Uriel Feige, Avi Wigderson, and David Zuckerman. Derandomized graph products. Comput. Complex., 5(1):60–75, 1995.
  • ALM+ [01] Sanjeev Arora, Carsten Lund, Rajeev Motwani, Madhu Sudan, and Mario Szegedy. Proof verification and the hardness of approximation problems. Journal of the ACM (JACM), 45:501–555, 09 2001.
  • App [17] Benny Applebaum. Exponentially-hard gap-csp and local prg via local hardcore functions. In 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS), pages 836–847, 2017.
  • AR [94] Noga Alon and Yuval Roichman. Random Cayley graphs and expanders. Random Structures & Algorithms, 5(2):271–284, 1994.
  • BEKP [13] Edouard Bonnet, Bruno Escoffier, Eun Jung Kim, and Vangelis Th. Paschos. On subexponential and fpt-time inapproximability. In Gregory Gutin and Stefan Szeider, editors, Parameterized and Exact Computation, pages 54–65, Cham, 2013. Springer International Publishing.
  • BGH+ [06] Eli Ben-Sasson, Oded Goldreich, Prahladh Harsha, Madhu Sudan, and Salil Vadhan. Robust PCPs of proximity, shorter PCPs, and applications to coding. SIAM Journal on Computing, 36(4):889–974, 2006.
  • BSSVW [03] Eli Ben-Sasson, Madhu Sudan, Salil Vadhan, and Avi Wigderson. Randomness-efficient low degree tests and short pcps via epsilon-biased sets. In Proceedings of the Thirty-Fifth Annual ACM Symposium on Theory of Computing, STOC ’03, page 612–621, New York, NY, USA, 2003. Association for Computing Machinery.
  • CAGK+ [19] Vincent Cohen-Addad, Anupam Gupta, Amit Kumar, Euiwoong Lee, and Jason Li. Tight FPT Approximations for k-Median and k-Means. In Christel Baier, Ioannis Chatzigiannakis, Paola Flocchini, and Stefano Leonardi, editors, 46th International Colloquium on Automata, Languages, and Programming (ICALP 2019), volume 132 of Leibniz International Proceedings in Informatics (LIPIcs), pages 42:1–42:14, Dagstuhl, Germany, 2019. Schloss Dagstuhl – Leibniz-Zentrum für Informatik.
  • CCK+ [17] Parinya Chalermsook, Marek Cygan, Guy Kortsarz, Bundit Laekhanukit, Pasin Manurangsi, Danupon Nanongkai, and Luca Trevisan. From gap-ETH to FPT-inapproximability: Clique, dominating set, and more. In Chris Umans, editor, 58th IEEE Annual Symposium on Foundations of Computer Science, FOCS 2017, Berkeley, CA, USA, October 15-17, 2017, pages 743–754. IEEE Computer Society, 2017.
  • CFG+ [16] Marek Cygan, Fedor V. Fomin, Alexander Golovnev, Alexander S. Kulikov, Ivan Mihajlin, Jakub Pachocki, and Arkadiusz Socala. Tight bounds for graph homomorphism and subgraph isomorphism. In Robert Krauthgamer, editor, Proceedings of the Twenty-Seventh Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2016, Arlington, VA, USA, January 10-12, 2016, pages 1643–1649. SIAM, 2016.
  • CFLL [23] Yijia Chen, Yi Feng, Bundit Laekhanukit, and Yanlin Liu. Simple combinatorial construction of the ko(1){}^{\mbox{o(1)}}-lower bound for approximating the parameterized k-clique. CoRR, abs/2304.07516, 2023.
  • CHKX [06] Jianer Chen, Xiuzhen Huang, Iyad A. Kanj, and Ge Xia. Strong computational lower bounds via parameterized complexity. J. Comput. Syst. Sci., 72(8):1346–1367, 2006.
  • Din [07] Irit Dinur. The PCP theorem by gap amplification. J. ACM, 54(3):12–es, jun 2007.
  • Fei [98] Uriel Feige. A threshold of lnn\ln n for approximating set cover. Journal of the ACM (JACM), 45(4):634–652, 1998.
  • FG [06] Jörg Flum and Martin Grohe. Parameterized Complexity Theory. Texts in Theoretical Computer Science. An EATCS Series. Springer, 2006.
  • FGL+ [96] Uriel Feige, Shafi Goldwasser, László Lovász, Shmuel Safra, and Mario Szegedy. Interactive proofs and the hardness of approximating cliques. J. ACM, 43(2):268–292, 1996.
  • FS [95] Katalin Friedl and Madhu Sudan. Some improvements to total degree tests. In Proceedings Third Israel Symposium on the Theory of Computing and Systems, pages 190–198. IEEE, 1995.
  • FSLM [20] Andreas Emil Feldmann, Karthik C. S., Euiwoong Lee, and Pasin Manurangsi. A survey on approximation in parameterized complexity: Hardness and algorithms. Algorithms, 13(6), 2020.
  • GLR+ [23] Venkatesan Guruswami, Bingkai Lin, Xuandi Ren, Yican Sun, and Kewen Wu. Parameterized inapproximability hypothesis under ETH. CoRR, abs/2311.16587, 2023.
  • GRS [23] Venkatesan Guruswami, Xuandi Ren, and Sai Sandeep. Baby PIH: Parameterized inapproximability of Min CSP. arXiv preprint arXiv:2310.16344, 2023.
  • Hoc [97] Dorit S. Hochbaum. Approximation algorithms for NP-hard problems. SIGACT News, 28(2):40–52, jun 1997.
  • IP [01] Russell Impagliazzo and Ramamohan Paturi. On the complexity of k-SAT. Journal of Computer and System Sciences, 62:367–375, 2001.
  • IPZ [01] Russell Impagliazzo, Ramamohan Paturi, and Francis Zane. Which problems have strongly exponential complexity? Journal of Computer and System Sciences, 63(4):512–530, 2001.
  • JM [21] Akhil Jalan and Dana Moshkovitz. Near-optimal Cayley expanders for abelian groups. In 41st IARCS Annual Conference on Foundations of Software Technology and Theoretical Computer Science, 2021.
  • Jus [72] Jørn Justesen. Class of constructive asymptotically good algebraic codes. IEEE Trans. Inf. Theory, 18:652–656, 1972.
  • KK [22] Karthik C. S. and Subhash Khot. Almost polynomial factor inapproximability for parameterized k-clique. In Shachar Lovett, editor, 37th Computational Complexity Conference, CCC 2022, July 20-23, 2022, Philadelphia, PA, USA, volume 234 of LIPIcs, pages 6:1–6:21. Schloss Dagstuhl - Leibniz-Zentrum für Informatik, 2022.
  • KMPS [23] Karthik C. S., Dániel Marx, Marcin Pilipczuk, and Uéverton S. Souza. Conditional lower bounds for sparse parameterized 2-CSP: A streamlined proof. CoRR, abs/2311.05913, 2023.
  • KT [00] Jonathan Katz and Luca Trevisan. On the efficiency of local decoding procedures for error-correcting codes. In Proceedings of the Thirty-Second Annual ACM Symposium on Theory of Computing, STOC ’00, page 80–86, New York, NY, USA, 2000. Association for Computing Machinery.
  • Lin [21] Bingkai Lin. Constant approximating kk-clique is W[1]-hard. In Samir Khuller and Virginia Vassilevska Williams, editors, STOC ’21: 53rd Annual ACM SIGACT Symposium on Theory of Computing, Virtual Event, Italy, June 21-25, 2021, pages 1749–1756. ACM, 2021.
  • LRSW [22] Bingkai Lin, Xuandi Ren, Yican Sun, and Xiuhan Wang. On lower bounds of approximating parameterized k-clique. In Mikolaj Bojanczyk, Emanuela Merelli, and David P. Woodruff, editors, 49th International Colloquium on Automata, Languages, and Programming, ICALP 2022, July 4-8, 2022, Paris, France, volume 229 of LIPIcs, pages 90:1–90:18. Schloss Dagstuhl - Leibniz-Zentrum für Informatik, 2022.
  • LRSW [23] Bingkai Lin, Xuandi Ren, Yican Sun, and Xiuhan Wang. Improved hardness of approximating k-clique under ETH. In 64th IEEE Annual Symposium on Foundations of Computer Science, FOCS 2023, Santa Cruz, CA, USA, November 6-9, 2023, pages 285–306. IEEE, 2023.
  • LRSZ [20] Daniel Lokshtanov, M. S. Ramanujan, Saket Saurabh, and Meirav Zehavi. Parameterized complexity and approximability of directed odd cycle transversal. In Shuchi Chawla, editor, Proceedings of the 2020 ACM-SIAM Symposium on Discrete Algorithms, SODA 2020, Salt Lake City, UT, USA, January 5-8, 2020, pages 2181–2200. SIAM, 2020.
  • Man [20] Pasin Manurangsi. Tight running time lower bounds for strong inapproximability of maximum kk-coverage, unique set cover and related problems (via tt-wise agreement testing theorem). In Proceedings of the Fourteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pages 62–81. SIAM, 2020.
  • Mar [10] Dániel Marx. Can you beat treewidth? Theory Comput., 6(1):85–112, 2010.
  • PS [94] Alexander Polishchuk and Daniel A Spielman. Nearly-linear size holographic proofs. In Proceedings of the twenty-sixth annual ACM symposium on Theory of computing, pages 194–203, 1994.
  • Rao [11] Anup Rao. Parallel repetition in projection games and a concentration bound. SIAM Journal on Computing, 40(6):1871–1891, 2011.
  • RS [96] Ronitt Rubinfeld and Madhu Sudan. Robust characterizations of polynomials with applications to program testing. SIAM Journal on Computing, 25(2):252–271, 1996.
  • Sch [80] J. T. Schwartz. Fast probabilistic algorithms for verification of polynomial identities. J. ACM, 27(4):701–717, oct 1980.

Appendix A Derandomized Parallel Low Degree Test

In this section, we design a derandomized parallel low degree test to prove Theorem 3.6 and Theorem 3.7. This is obtained by combining the derandomized low degree test [9, Theorem 4.1] with the parallel low degree test [33, Lemma 7.7], where the latter builds on [19]. While the combination is standard, we decide to expand the proof sketch in [9] to a full proof for completeness.

Theorem (Theorem 3.6 Restated).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. Let Σ=𝔽d+1\Sigma=\mathbb{F}^{d+1} be the set of univariate degree-dd polynomials over 𝔽\mathbb{F}. There exists an efficient verifier 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} with the following properties.

  • The input of 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} is TπT\circ\pi, where T(𝔽t)|𝔽|mT\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}} is supposed to be a codeword of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} and π(Σt)|𝔽|m(mlog|𝔽|)O(1)\pi\in(\Sigma^{t})^{|\mathbb{F}|^{m}\cdot(m\log|\mathbb{F}|)^{O(1)}} is the auxiliary proof.

  • 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} tosses mlog|𝔽|+O(loglog|𝔽|+logm)m\log|\mathbb{F}|+O\left(\log\log|\mathbb{F}|+\log m\right) unbiased coins and makes 22 queries on TπT\circ\pi.

  • If TIm(𝖱𝖬𝔽,m,d,t)T\in\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), then there exists some π\pi such that 𝒫𝗅𝖽𝗍(Tπ)\mathcal{P}_{\sf ldt}(T\circ\pi) always accepts.

  • If TT is δ\delta-far from Im(𝖱𝖬𝔽,m,d,t)\mathrm{Im}({{\sf RM}}^{\mathbb{F},m,d,t}), then 𝐏𝐫[𝒫𝗅𝖽𝗍(Tπ)rejects]240δ\operatorname*{\mathbf{Pr}}[\mathcal{P}_{\sf ldt}(T\circ\pi)\ \text{rejects}]\geq 2^{-40}\delta for any π\pi.

Theorem (Theorem 3.7 Restated).

Assume 𝖼𝗁𝖺𝗋(𝔽)=2\mathsf{char}(\mathbb{F})=2 and |𝔽|max{6md,2100mlog|𝔽|}|\mathbb{F}|\geq\max\left\{6md,2^{100}m\log|\mathbb{F}|\right\}. There exists a Boolean circuit C𝗅𝖽𝗍C_{\mathsf{ldt}} of size |𝔽|m𝗉𝗈𝗅𝗒|𝔽||\mathbb{F}|^{m}\mathsf{poly}|\mathbb{F}| for T(𝔽t)|𝔽|mT\in(\mathbb{F}^{t})^{|\mathbb{F}|^{m}}, where we encode 𝔽\mathbb{F} as {0,1}log|𝔽|\{0,1\}^{\log|\mathbb{F}|}, such that TT is codeword of 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} iff TT parallel satisfies C𝗅𝖽𝗍C_{\mathsf{ldt}}.

A.1 Extra Notation

We first set up some necessary notation.

Parallel Low Degree Polynoimal

We first recall the notion of parallel polynomial from Section 3. Let 𝔽\mathbb{F} be a finite field. For a parallel-output function f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t}, we denote f[1],,f[t]:𝔽m𝔽f[1],\ldots,f[t]\colon\mathbb{F}^{m}\to\mathbb{F} as its single-output components, i.e., f(x)=(f[1](x),,f[t](x))f(x)=(f[1](x),\ldots,f[t](x)). We aim to test if f[1],,f[t]f[1],\ldots,f[t] are consistent with degree-dd polynomials on a large common set of inputs. Formally, we say ff is δ\delta-close to parallel degree-dd iff ff is δ\delta-close to 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t}; and we say ff is parallel degree-dd if δ=0\delta=0.

A simple union bound shows that if each f[i]f[i] is δ\delta-close to degree-dd (e.g., from the standard low degree test), then f=(f[1],,f[t])f=(f[1],\ldots,f[t]) is tδt\delta-close to parallel degree-dd. However for our purposes, such a loss is not affordable since tt is typically large. Therefore we need to open-box the specific low degree test and show that it implies consistency on 1δ1-\delta common fraction of inputs simultaneously for all f[i]f[i].

Parallel Low Degree Test

For x,y𝔽mx,y\in\mathbb{F}^{m}, the line crossing xx in direction yy is the set

x,y:={x+ty:t𝔽}.\ell_{x,y}:=\left\{x+t\cdot y\colon t\in\mathbb{F}\right\}.

Note that if y=0my=0^{m} then x,y={x}\ell_{x,y}=\left\{x\right\}, otherwise x,y\ell_{x,y} has |𝔽||\mathbb{F}| points. Let 𝕃\mathbb{L} be the set of lines in 𝔽m\mathbb{F}^{m}. For a parallel function f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t} and any 𝕃\ell\in\mathbb{L}, we use f|:𝔽tf|_{\ell}\colon\ell\to\mathbb{F}^{t} to denote the restriction of ff on the line \ell.

Let 𝔽(d,t)\mathbb{F}^{(d,t)} be the set of univariate parallel degree-dd polynomials, i.e.,

𝔽(d,t):={h:𝔽𝔽t:hi is a univariate degree-d polynomial for each i[t]}.\mathbb{F}^{(d,t)}:=\left\{h\colon\mathbb{F}\to\mathbb{F}^{t}\colon h_{i}\text{ is a univariate degree-$d$ polynomial for each }i\in[t]\right\}.

The standard low degree test 𝖫𝖣𝖳𝖾𝗌𝗍f,g\mathsf{LDTest}^{f,g} [39], translated to the parallel setting [33], is given oracle access to f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t} and g:𝕃𝔽(d,t)g\colon\mathbb{L}\to\mathbb{F}^{(d,t)}, which later tests their consistency999The term consistency has been called correlated agreement, e.g., in recent literature of proximity gaps for RS codes and related literature in FRI protocol/SNARKs.. For each line 𝕃\ell\in\mathbb{L}, we denote g()𝔽(d,t)g(\ell)\in\mathbb{F}^{(d,t)} as its parallel line polynomial, which we interpret as a univariate parallel degree-dd polynomial mapping elements in \ell to 𝔽t\mathbb{F}^{t}. Thus for a line \ell and a point zz\in\ell, g()(z)𝔽tg(\ell)(z)\in\mathbb{F}^{t} is well defined. We say ff agrees with g()g(\ell) on zz if f(z)=g()(z)f(z)=g(\ell)(z).

For each f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t} and dd\in\mathbb{N}, we use f𝕃:𝕃𝔽(d,t)f_{\mathbb{L}}\colon\mathbb{L}\to\mathbb{F}^{(d,t)} to denote the restriction of ff on each line to its closest parallel degree-dd polynomial. That is, for each line \ell, we set f𝕃()f_{\mathbb{L}}(\ell) to be the parallel degree-dd polynomial closest to f|f|_{\ell}, where we break tie arbitrarily. Note that we will always assume the degree to be tested is dd, and thus we omit dd in defining f𝕃f_{\mathbb{L}} for simplicity.

In the completeness case (i.e., ff is indeed parallel degree-dd), we can pick f𝕃()=f|f_{\mathbb{L}}(\ell)=f|_{\ell} which is also parallel degree-dd. The parallel low degree test 𝖫𝖣𝖳𝖾𝗌𝗍f,g\mathsf{LDTest}^{f,g} explores the reverse direction: independently select x,y𝔽mx,y\sim\mathbb{F}^{m} and accept iff f(x)=g(x,y)(x)f(x)=g(\ell_{x,y})(x), i.e., ff agrees with the parallel line polynomial g(x,y)g(\ell_{x,y}) on point xx. The parallel low degree test [33] shows that if 𝖫𝖣𝖳𝖾𝗌𝗍f,g\mathsf{LDTest}^{f,g} accepts with probability 1δ1-\delta, then ff is O(δ)O(\delta)-close to parallel degree-dd. Note that this bound does not depend on tt.

Derandomized Parallel Low Degree Test

Now we introduce the derandomized version of the parallel low degree test. Following [9], this simply replaces the uniformly random direction yy by a pseudorandom ySy\sim S for a much smaller set S𝔽mS\subseteq\mathbb{F}^{m}. Hence the number of randomness drops from |𝔽|2m|\mathbb{F}|^{2m} to |S||𝔽|m|S|\cdot|\mathbb{F}|^{m}.

Definition A.1 (Derandomized Parallel Low Degree Test).

Let S𝔽mS\subseteq\mathbb{F}^{m} and f:𝔽m𝔽t,g:𝕃𝔽(d,t)f\colon\mathbb{F}^{m}\to\mathbb{F}^{t},g\colon\mathbb{L}\to\mathbb{F}^{(d,t)}.101010Technically, gg only needs to be defined on lines whose directions are in SS. We choose to assume gg is defined over all lines 𝕃\mathbb{L} for simplicity. The derandomized parallel low degree test 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}^{f,g}_{S} is executed as follows: independently select x𝔽mx\sim\mathbb{F}^{m} and ySy\sim S, then accept iff f(x)=g(x,y)(x)f(x)=g(\ell_{x,y})(x).

Later the set SS is chosen to be a λ\lambda-biased set as in [9].

Definition A.2 (λ\lambda-Biased Set).

S𝔽mS\subseteq\mathbb{F}^{m} is a λ\lambda-biased set iff

  • SS is symmetric, i.e., if ySy\in S then yS-y\in S;111111In some literature this symmetric assumption is not imposed. The parameter λ\lambda in that case is comparable with the one here by a multiplicative factor of 22.

  • |𝔼yS[χ(y)]|λ\left|\operatorname*{\mathbb{E}}_{y\sim S}\left[\chi(y)\right]\right|\leq\lambda holds for any non-trivial homomorphism121212This homomorphism is usually referred as character. It is trivial if it maps everything to 11. An illustrative example is when 𝔽=𝔽2\mathbb{F}=\mathbb{F}_{2} and χ\chi is a parity function. χ:𝔽mμp\chi\colon\mathbb{F}^{m}\to\mu_{p}, where μ\mu is the multiplicative group of pp-th unit root and pp is the characteristic of 𝔽\mathbb{F}.

In a graph theoretical reformulation, SS is λ\lambda-biased iff the graph GSG_{S} is an (undirected) expander graph with expansion factor 1λ1-\lambda, where the vertex set of GSG_{S} is 𝔽m\mathbb{F}^{m} and x,yGSx,y\in G_{S} is connected iff xySx-y\in S (see e.g., [26]). For our purposes, we quote the following results derived directly from the expanding property of biased sets.

Lemma A.3 ([9, Lemma 4.3]).

Suppose S𝔽mS\subseteq\mathbb{F}^{m} is λ\lambda-biased. Then for any B𝔽mB\subseteq\mathbb{F}^{m} of density μ=|B||𝔽|m\mu=\frac{|B|}{|\mathbb{F}|^{m}} and any ε>0\varepsilon>0, we have

𝐏𝐫x𝔽m,yS[||x,yB||x,y|μ|>ε](1|𝔽|+λ)με2.\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim S}\left[\left|\frac{\left|\ell_{x,y}\cap B\right|}{\left|\ell_{x,y}\right|}-\mu\right|>\varepsilon\right]\leq\left(\frac{1}{|\mathbb{F}|}+\lambda\right)\cdot\frac{\mu}{\varepsilon^{2}}.

Finally we remark that λ\lambda-biased sets exist for |S|=Ω(mlog|𝔽|λ2)|S|=\Omega\left(\frac{m\log|\mathbb{F}|}{\lambda^{2}}\right) [6] and efficient explicit constructions of sizes 𝗉𝗈𝗅𝗒(m,log|𝔽|,1λ)\mathsf{poly}(m,\log|\mathbb{F}|,\frac{1}{\lambda}) are also obtained.

Fact A.4 (See e.g., [26]).

For any finite field 𝔽\mathbb{F}, positive integer mm, and parameter λ(0,1]\lambda\in(0,1], a λ\lambda-biased set of size 𝗉𝗈𝗅𝗒(m,log|𝔽|,1λ)\mathsf{poly}(m,\log|\mathbb{F}|,\frac{1}{\lambda}) can be constructed efficiently in time 𝗉𝗈𝗅𝗒(m,|𝔽|,1λ)\mathsf{poly}(m,|\mathbb{F}|,\frac{1}{\lambda}).

Augmented Derandomized Parallel Low Degree Test

Unfortunately the derandomized parallel low degree test is not sufficient to guarantee the exact quantitative degree condition, as the directions SS have fairly limited possibilities. More concretely, even if 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}_{S}^{f,g} succeeds with probability 11, it is not guaranteed that ff is parallel degree-dd. To compensate the missing directions, we need to augment Definition A.1 with an additional test that checks the consistency of ff and gg on a purely random direction from the origin [9].

Definition A.5 (Augmented Derandomized Parallel Low Degree Test).

Let S𝔽mS\subseteq\mathbb{F}^{m} and f:𝔽m𝔽t,g:𝕃𝔽(d,t)f\colon\mathbb{F}^{m}\to\mathbb{F}^{t},g\colon\mathbb{L}\to\mathbb{F}^{(d,t)}. The augmented derandomized parallel low degree test 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g} is executed as follows: with equal probability we perform one of the following two tests:

  • Independently select x𝔽mx\sim\mathbb{F}^{m} and ySy\sim S, then accept iff f(x)=g(x,y)(x)f(x)=g(\ell_{x,y})(x).

  • Select z𝔽mz\sim\mathbb{F}^{m}, then accept iff f(z)=g(0m,z)(z)f(z)=g(\ell_{0^{m},z})(z).

The first test in 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g} is simply 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}_{S}^{f,g}, for which we will later show that it guarantees that ff is close to parallel degree-mdmd. Then the second test allows us to further bring the degree down to dd.

A.2 Codeword Testing

Theorem 3.6 and Theorem 3.7 follow directly from the following result, combined with the explicit constructions for biased sets.

Theorem A.6.

Assume |𝔽|6md|\mathbb{F}|\geq 6md, |𝔽|2100mlog|𝔽||\mathbb{F}|\geq 2^{100}\cdot m\log|\mathbb{F}|, and λ12100mlog|𝔽|\lambda\leq\frac{1}{2^{100}\cdot m\log|\mathbb{F}|}. Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set. If

𝐏𝐫[𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g accepts]1δ,\operatorname*{\mathbf{Pr}}\left[\mathsf{AugLDTest}_{S}^{f,g}\text{ accepts}\right]\geq 1-\delta,

then ff is 240δ2^{40}\delta-close to parallel degree-dd.

Proof of Theorem 3.6.

We first note that |𝔽|m>(m+dd)|\mathbb{F}|^{m}>\binom{m+d}{d} and |𝔽|>d|\mathbb{F}|>d are satisfied assuming the conditions on |𝔽||\mathbb{F}| in Theorem 3.6, thus 𝖱𝖬𝔽,m,d,t{{\sf RM}}^{\mathbb{F},m,d,t} is well defined.

Then we instantiate Theorem A.6 with λ=12100mlog|𝔽|\lambda=\frac{1}{2^{100}\cdot m\log|\mathbb{F}|} and the λ\lambda-biased set SS by Fact A.4. The construction of SS is efficient in time 𝗉𝗈𝗅𝗒(|𝔽|,m)\mathsf{poly}(|\mathbb{F}|,m) and SS has size (mlog|𝔽|)O(1)(m\log|\mathbb{F}|)^{O(1)}. Then we define 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} as 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g} where T=fT=f and π\pi is defined to be the entries of gg that can be possibly queried. Recall Definition A.5. Then we have

|π||𝔽|m|S|+|𝔽|m=|𝔽|m(mlog|𝔽|)O(1).|\pi|\leq|\mathbb{F}|^{m}\cdot|S|+|\mathbb{F}|^{m}=|\mathbb{F}|^{m}\cdot(m\log|\mathbb{F}|)^{O(1)}.

In addition, by merging the randomness of the two tests in 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g}, 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} tosses

1+log(|𝔽|m|S|)=mlog|𝔽|+O(loglog|𝔽|+logm)1+\log\left(|\mathbb{F}|^{m}|S|\right)=m\log|\mathbb{F}|+O\left(\log\log|\mathbb{F}|+\log m\right)

total coins. The completeness is obvious and the soundness follows from Theorem A.6. ∎

Proof of Theorem 3.7.

Now we turn to Theorem 3.7. By Theorem 3.6, it suffices to implement 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} purely on ff. In addition, since 𝒫𝗅𝖽𝗍=𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathcal{P}_{\sf ldt}=\mathsf{AugLDTest}_{S}^{f,g} and by Definition A.5, it performs the same check in parallel for each coordinate i[t]i\in[t]. This means that we only need to instantiate 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt} for a single coordinate (or equivalently, think of t=1t=1) to design the circuit C𝗅𝖽𝗍C_{\mathsf{ldt}}.

To get rid of the extra proof gg, we simply set g=f𝕃g=f_{\mathbb{L}}. Then, whenever we need information about entries in gg (i.e., a line polynomial), we can probe the entries along the line in ff to compute it. We remark that this is inefficient in terms of the query complexity, but it is still efficient in terms of the circuit complexity.

Now we describe the construction of C𝗅𝖽𝗍C_{\mathsf{ldt}} for a fixed coordinate i[t]i\in[t]. Based on the coin toss of 𝒫𝗅𝖽𝗍\mathcal{P}_{\sf ldt}, it checks the consistency of an 𝔽\mathbb{F}-valued point (i.e., f[i](x)f[i](x) or f[i](z)f[i](z)) with an evaluation point (i.e., xx or zz) of a degree-dd line polynomial over 𝔽\mathbb{F} (i.e., fx,y[i]f_{\ell_{x,y}}[i] or f0m,z[i]f_{\ell_{0^{m},z}}[i]). To implement it as a circuit, we take the conjunction of all the sub-circuit outcome from coin toss possibilities, where each sub-circuit performs the following computation.

  • It first interpolates the line polynomial using entries of f[i]f[i] along the line, and checks if this line polynomial is degree-dd.

    This can be efficiently done with 𝗉𝗈𝗅𝗒|𝔽|\mathsf{poly}|\mathbb{F}| gates.

  • Then it evaluates the value of the desired point of the line polynomial, and checks if it is the same as the one directly obtained from f[i]f[i].

    This also requires 𝗉𝗈𝗅𝗒|𝔽|\mathsf{poly}|\mathbb{F}| gates only.

The correctness of C𝗅𝖽𝗍C_{\mathsf{ldt}} follows directly from Theorem 3.6 and our choice of g=f𝕃g=f_{\mathbb{L}}. Since the number of coin toss possibilities is |𝔽|m𝗉𝗈𝗅𝗒(m,log|𝔽|)|\mathbb{F}|^{m}\mathsf{poly}(m,\log|\mathbb{F}|), by our assumption on |𝔽||\mathbb{F}|, the size of C𝗅𝖽𝗍C_{\mathsf{ldt}} is |𝔽|m𝗉𝗈𝗅𝗒|𝔽||\mathbb{F}|^{m}\mathsf{poly}|\mathbb{F}| as claimed. ∎

To prove the statement about 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍\mathsf{AugLDTest}, we need to analyze 𝖫𝖣𝖳𝖾𝗌𝗍\mathsf{LDTest} first, which guarantees a weaker degree bound.

Theorem A.7.

Assume |𝔽|3d|\mathbb{F}|\geq 3d, |𝔽|2100mlog|𝔽||\mathbb{F}|\geq 2^{100}\cdot m\log|\mathbb{F}|, and λ12100mlog|𝔽|\lambda\leq\frac{1}{2^{100}\cdot m\log|\mathbb{F}|}. Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set. If

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Sf,g accepts]1δ,\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S}^{f,g}\text{ accepts}\right]\geq 1-\delta,

then ff is 230δ2^{30}\delta-close to parallel degree-mdmd.

Assuming Theorem A.7, we conclude the proof of Theorem A.6.

Proof of Theorem A.6.

First we assume δ240\delta\leq 2^{-40} since otherwise 240δ12^{40}\delta\geq 1 and the statement trivially holds. Recall Definition A.5 that 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g} executes 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}_{S}^{f,g} with probability 1/21/2. Hence 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}_{S}^{f,g} must accept with probability at least 12δ1-2\delta. By Theorem A.7, this means that ff is 231δ2^{31}\delta-close to a parallel degree-mdmd polynomial ff^{\prime}. It suffices to show that ff^{\prime} is acutally parallel degree-dd.

Assume towards contradiction that ff^{\prime} is not parallel degree-dd. Now we consider the second half of 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g}, which checks if f(z)=g(0m,z)(z)f(z)=g(\ell_{0^{m},z})(z) for z𝔽mz\sim\mathbb{F}^{m}. Then we have

𝐏𝐫z𝔽m[f(z)=g(0m,z)(z)]\displaystyle\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f(z)=g(\ell_{0^{m},z})(z)\right] 𝐏𝐫z𝔽m[f(z)f(z)]+𝐏𝐫z𝔽m[f(z)=g(0m,z)(z)]\displaystyle\leq\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f(z)\neq f^{\prime}(z)\right]+\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f^{\prime}(z)=g(\ell_{0^{m},z})(z)\right]
231δ+𝐏𝐫z𝔽m[f(z)=g(0m,z)(z)].\displaystyle\leq 2^{31}\delta+\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f^{\prime}(z)=g(\ell_{0^{m},z})(z)\right]. (6)

To analyze Equation 6, we consider the following quantity:

𝐏𝐫w𝔽m,i𝔽[f(iw)=g(0m,w)(iw)].\operatorname*{\mathbf{Pr}}_{w\sim\mathbb{F}^{m},i\sim\mathbb{F}}\left[f^{\prime}(i\cdot w)=g(\ell_{0^{m},w})(i\cdot w)\right]. (7)

On the one hand, conditioned on i0i\neq 0, we have 0m,w=0m,iw\ell_{0^{m},w}=\ell_{0^{m},i\cdot w} and iwi\cdot w being uniform in 𝔽m\mathbb{F}^{m}. Therefore we can relate Equation 6 with Equation 7:

Equation 7(11|𝔽|)𝐏𝐫z𝔽m[f(z)=g(0m,z)(z)].\lx@cref{creftypecap~refnum}{eq:thm:derand_augpldt_detail_2}\geq\left(1-\frac{1}{|\mathbb{F}|}\right)\cdot\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f^{\prime}(z)=g(\ell_{0^{m},z})(z)\right]. (8)

On the other hand, conditioned on ww, f(iw)f^{\prime}(i\cdot w) is a univariate parallel degree-mdmd polynomial in ii. Let d<dmdd<d^{\prime}\leq md be the parallel degree of ff^{\prime}. Then the coefficient of idi^{d^{\prime}} in f(iw)f^{\prime}(i\cdot w) is a non-zero parallel degree-dd^{\prime} polynomial. By Schwartz–Zippel lemma, it vanishes on at most d|𝔽|\frac{d^{\prime}}{|\mathbb{F}|} fraction of choices of ww. For each ww that the top coefficient of idi^{d^{\prime}} does not vanish, by Schwartz–Zippel lemma, f(iw)f^{\prime}(i\cdot w) agrees with g(0m,w)(iw)g(\ell_{0^{m},w})(i\cdot w) on at most dd^{\prime} choices of ii since g(0m,w)g(\ell_{0^{m},w}) is parallel degree-dd and d<dd<d^{\prime}. This means

Equation 7 𝐏𝐫w𝔽m[coeff of id in f(iw) vanishes]+𝐏𝐫w𝔽m,i𝔽[f(iw)=g(0m,w)(iw)|not vanish]\displaystyle\leq\operatorname*{\mathbf{Pr}}_{w\sim\mathbb{F}^{m}}\left[\text{coeff of }i^{d^{\prime}}\text{ in }f^{\prime}(i\cdot w)\text{ vanishes}\right]+\operatorname*{\mathbf{Pr}}_{w\sim\mathbb{F}^{m},i\sim\mathbb{F}}\left[f^{\prime}(i\cdot w)=g(\ell_{0^{m},w})(i\cdot w)\,\middle|\,\text{not vanish}\right]
d|𝔽|+d|𝔽|2md|𝔽|.\displaystyle\leq\frac{d^{\prime}}{|\mathbb{F}|}+\frac{d^{\prime}}{|\mathbb{F}|}\leq\frac{2md}{|\mathbb{F}|}. (since dmdd^{\prime}\leq md)

Combining this with Equation 8 and Equation 6, we have

𝐏𝐫z𝔽m[f(z)=g(0m,z)(z)]\displaystyle\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f(z)=g(\ell_{0^{m},z})(z)\right] 231δ+2md|𝔽|112,\displaystyle\leq 2^{31}\delta+\frac{2md}{|\mathbb{F}|-1}\leq\frac{1}{2},

where we used the fact that δ240\delta\leq 2^{-40} and |𝔽|6md|\mathbb{F}|\geq 6md. Since this is half of the actual 𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{AugLDTest}_{S}^{f,g}, it means

𝐏𝐫[𝖠𝗎𝗀𝖫𝖣𝖳𝖾𝗌𝗍Sf,g accepts]12+1212<1δ,\operatorname*{\mathbf{Pr}}\left[\mathsf{AugLDTest}_{S}^{f,g}\text{ accepts}\right]\leq\frac{1}{2}+\frac{1}{2}\cdot\frac{1}{2}<1-\delta,

which is a contradiction.

In conclusion, ff^{\prime} must be parallel degree-dd. ∎

Our Theorem A.7 is the parallel version of [9, Theorem 4.1]. Its proof follows from iteratively applying the following lemma, which is the parallel version of [9, Lemma 4.4].

Lemma A.8.

Assume |𝔽|3d|\mathbb{F}|\geq 3d. Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set and TST\subseteq S of size |T||S|/2|T|\geq|S|/2. Let f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t}. If

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 accepts]1δ,\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{T}^{f,f_{\mathbb{L}}}\text{ accepts}\right]\geq 1-\delta,

then for any 2δγ2202\delta\leq\gamma\leq 2^{-20}, there exists f:𝔽m𝔽tf^{\prime}\colon\mathbb{F}^{m}\to\mathbb{F}^{t} and TTT^{\prime}\subseteq T with the following properties:

  1. 1.

    |T|(1δγ)|T||T|2|T^{\prime}|\geq\left(1-\frac{\delta}{\gamma}\right)|T|\geq\frac{|T|}{2}.

  2. 2.

    Δ(f,f)4δ\Delta(f^{\prime},f)\leq 4\delta.

  3. 3.

    𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 accepts]1240γ(1|𝔽|+λ)\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}^{f^{\prime},f_{\mathbb{L}}}_{T^{\prime}}\text{ accepts}\right]\geq 1-2^{40}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right).

Assuming Lemma A.8, we first conclude Theorem A.7.

Proof of Theorem A.7.

First we assume δ230\delta\leq 2^{-30} since otherwise 230δ12^{30}\delta\geq 1 and the statement trivially holds. Next, we can also assume without loss of generality that g=f𝕃g=f_{\mathbb{L}}. This is because, for each possible131313A line is possible in 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}_{S}^{f,g} if its direction lies in SS. line 𝕃\ell\in\mathbb{L}, 𝖫𝖣𝖳𝖾𝗌𝗍Sf,g\mathsf{LDTest}^{f,g}_{S} conditioned on this line checks a uniformly random point xx\in\ell whether f(x)=g()(x)f(x)=g(\ell)(x). Since g()g(\ell) is parallel degree-dd, the success probability maximized when g()=f𝕃()g(\ell)=f_{\mathbb{L}}(\ell).

We will repeatedly apply Lemma A.8 to bring the soundness gap down to |𝔽|2m\ll|\mathbb{F}|^{-2m}, at which point it is actually zero by granularity. Then we use the following characterization of parallel low degree polynomials similar to [39] to arrive at an actual parallel degree-mdmd polynomial.

Theorem A.9.

Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set and TST\subseteq S of size |T|>1+λ2|S||T|>\frac{1+\lambda}{2}\cdot|S|. Then f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t} is parallel degree-mdmd if f|x,yf|_{\ell_{x,y}} is parallel degree-dd for every x𝔽m,yTx\in\mathbb{F}^{m},y\in T.

Lemma A.8 will be proved in Subsection A.4. Now we focus on reducing the soundness gap.

The δ1260mlog|𝔽|\delta\geq\frac{1}{2^{60}\cdot m\log|\mathbb{F}|} Case

In this case, we first perform a pre-processing round to bring down the soundness gap. By Lemma A.8 with T=ST=S and γ=220\gamma=2^{-20}, we have S1SS_{1}\subseteq S and f(1):𝔽m𝔽tf^{(1)}\colon\mathbb{F}^{m}\to\mathbb{F}^{t} with the following property:

  1. 1.

    |S1|(1210)|S||S_{1}|\geq\left(1-2^{-10}\right)\cdot|S|, since δ230\delta\leq 2^{-30}.

  2. 2.

    Δ(f(1),f)4δ\Delta(f^{(1)},f)\leq 4\delta.

  3. 3.

    𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍S1f(1),f𝕃 accepts]11260mlog|𝔽|\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}^{f^{(1)},f_{\mathbb{L}}}_{S_{1}}\text{ accepts}\right]\geq 1-\frac{1}{2^{60}\cdot m\log|\mathbb{F}|}, since |𝔽|2100mlog|𝔽||\mathbb{F}|\geq 2^{100}\cdot m\log|\mathbb{F}| and λ12100mlog|𝔽|\lambda\leq\frac{1}{2^{100}\cdot m\log|\mathbb{F}|}.

Now define

δi=2i250mlog|𝔽|andγi=δi210mlog|𝔽|=2i240.\delta_{i}=\frac{2^{-i}}{2^{50}\cdot m\log|\mathbb{F}|}\quad\text{and}\quad\gamma_{i}=\delta_{i}\cdot 2^{10}\cdot m\log|\mathbb{F}|=\frac{2^{-i}}{2^{40}}.

For each i=1,,2mlog|𝔽|i=1,\ldots,2\left\lceil m\log|\mathbb{F}|\right\rceil, we apply Lemma A.8 on δi,Si,f(i)\delta_{i},S_{i},f^{(i)} and obtain δi+1,Si+1,f(i+1)\delta_{i+1},S_{i+1},f^{(i+1)}. To show the correctness of this process, we verify by induction on ii that the conditions in Lemma A.8 are satisfied, i.e.,141414In Lemma A.8 we only require |Si||S|/2|S_{i}|\geq|S|/2. Here we strengthen it for convenience of Theorem A.9.

|Si|0.9|S|and𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Sif(i),f𝕃]1δi,|S_{i}|\geq 0.9|S|\quad\text{and}\quad\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S_{i}}^{f^{(i)},f_{\mathbb{L}}}\right]\geq 1-\delta_{i}, (9)

where we omit 2δiγi2202\delta_{i}\leq\gamma_{i}\leq 2^{-20} since it holds by the definition of δi,γi\delta_{i},\gamma_{i}.

The base case i=1i=1 is valid by Item 1 and Item 3. For the inductive cases i2i\geq 2, we first observe that the first condition in Equation 9 follows from the following calculation:

|Si|\displaystyle|S_{i}| (1δi1γi1)|Si1|=(11210mlog|𝔽|)|Si1|\displaystyle\geq\left(1-\frac{\delta_{i-1}}{\gamma_{i-1}}\right)|S_{i-1}|=\left(1-\frac{1}{2^{10}\cdot m\log|\mathbb{F}|}\right)|S_{i-1}| (by Lemma A.8)
(11210mlog|𝔽|)i1|S1|\displaystyle\geq\cdots\geq\left(1-\frac{1}{2^{10}\cdot m\log|\mathbb{F}|}\right)^{i-1}|S_{1}| (by Lemma A.8 iteratively)
(11210mlog|𝔽|)i1(1210)|S|\displaystyle\geq\left(1-\frac{1}{2^{10}\cdot m\log|\mathbb{F}|}\right)^{i-1}\cdot\left(1-2^{-10}\right)|S| (by Item 1)
0.9|S|.\displaystyle\geq 0.9|S|. (since i2mlog|𝔽|i\leq 2\left\lceil m\log|\mathbb{F}|\right\rceil)

The second condition in Equation 9 follows from last round of Lemma A.8, which establishes that

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Sif(i),f𝕃 accepts]\displaystyle\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S_{i}}^{f^{(i)},f_{\mathbb{L}}}\text{ accepts}\right] 1240γi1(1|𝔽|+λ)=12i+1(1|𝔽|+λ)\displaystyle\geq 1-2^{40}\cdot\gamma_{i-1}\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)=1-2^{-i+1}\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)
12i+1212100mlog|𝔽|1δi.\displaystyle\geq 1-2^{-i+1}\cdot 2\cdot\frac{1}{2^{100}\cdot m\log|\mathbb{F}|}\geq 1-\delta_{i}.

Let k=2mlog|𝔽|k=2\left\lceil m\log|\mathbb{F}|\right\rceil. Then the above analysis shows

|Sk|0.9|S|>1+λ2|S|and𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Skf(k),f𝕃 accepts]12k250mlog|𝔽|>1|𝔽|2m.|S_{k}|\geq 0.9|S|>\frac{1+\lambda}{2}\cdot|S|\quad\text{and}\quad\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S_{k}}^{f^{(k)},f_{\mathbb{L}}}\text{ accepts}\right]\geq 1-\frac{2^{-k}}{2^{50}\cdot m\log|\mathbb{F}|}>1-|\mathbb{F}|^{-2m}.

Since 𝖫𝖣𝖳𝖾𝗌𝗍Skf(k),f𝕃\mathsf{LDTest}^{f^{(k)},f_{\mathbb{L}}}_{S_{k}} samples x𝔽mx\sim\mathbb{F}^{m} and ySk𝔽my\sim S_{k}\subseteq\mathbb{F}^{m} then perform a deterministic check whether f(k)(x)=f𝕃(x,y)(x)f^{(k)}(x)=f_{\mathbb{L}}(\ell_{x,y})(x), its accepting probability is an integer multiple of 1|𝔽|m|Sk||𝔽|2m\frac{1}{|\mathbb{F}|^{m}\cdot|S_{k}|}\geq|\mathbb{F}|^{-2m}. Therefore, we actually have 𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Skf(k),f𝕃 accepts]=1\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S_{k}}^{f^{(k)},f_{\mathbb{L}}}\text{ accepts}\right]=1, which means that f(k)f^{(k)} is parallel degree-dd on x,y\ell_{x,y} for all x𝔽m,ySkx\in\mathbb{F}^{m},y\in S_{k}. By Theorem A.9, this means that f(k)f^{(k)} is parallel degree-mdmd. In addition, its distance from the original ff is

Δ(f(k),f)\displaystyle\Delta(f^{(k)},f) Δ(f(1),f)+i=1k1Δ(f(i+1),f(i))4δ+i=1k1Δ(f(i+1),f(i))\displaystyle\leq\Delta(f^{(1)},f)+\sum_{i=1}^{k-1}\Delta(f^{(i+1)},f^{(i)})\leq 4\delta+\sum_{i=1}^{k-1}\Delta(f^{(i+1)},f^{(i)}) (by Item 2)
4δ+i=1k14δi=4δ+i=1k142i250mlog|𝔽|\displaystyle\leq 4\delta+\sum_{i=1}^{k-1}4\cdot\delta_{i}=4\delta+\sum_{i=1}^{k-1}\frac{4\cdot 2^{-i}}{2^{50}\cdot m\log|\mathbb{F}|} (by Lemma A.8)
230δ.\displaystyle\leq 2^{30}\delta. (since δ1260mlog|𝔽|\delta\geq\frac{1}{2^{60}\cdot m\log|\mathbb{F}|})
The δ<1260mlog|𝔽|\delta<\frac{1}{2^{60}\cdot m\log|\mathbb{F}|} Case

In this case, the analysis is even simpler as we do not need pre-processing. Let S1=SS_{1}=S and f(1)=ff^{(1)}=f. Define

δi=2i+1δandγi=δi210mlog|𝔽|=2i+11δmlog|𝔽|.\delta_{i}=2^{-i+1}\cdot\delta\quad\text{and}\quad\gamma_{i}=\delta_{i}\cdot 2^{10}\cdot m\log|\mathbb{F}|=2^{-i+11}\cdot\delta\cdot m\log|\mathbb{F}|.

We also apply Lemma A.8 for each i=1,,2mlog|𝔽|i=1,\ldots,2\left\lceil m\log|\mathbb{F}|\right\rceil on δi,Si,f(i)\delta_{i},S_{i},f^{(i)} to obtain δi+1,Si+1,f(i+1)\delta_{i+1},S_{i+1},f^{(i+1)}. The conditions in Lemma A.8 along this process can be verified in the similar fashion. Here we only highlight the difference: for the second condition, we have

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Sif(i),f𝕃 accepts]\displaystyle\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{S_{i}}^{f^{(i)},f_{\mathbb{L}}}\text{ accepts}\right] 1240γi1(1|𝔽|+λ)=12i+51δmlog|𝔽|(1|𝔽|+λ)\displaystyle\geq 1-2^{40}\cdot\gamma_{i-1}\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)=1-2^{-i+51}\cdot\delta\cdot m\log|\mathbb{F}|\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)
12i48δ1δi,\displaystyle\geq 1-2^{-i-48}\cdot\delta\geq 1-\delta_{i},

and for the third condition, we have

γi=2i+11δmlog|𝔽|2i+11260220.\displaystyle\gamma_{i}=2^{-i+11}\cdot\delta\cdot m\log|\mathbb{F}|\leq\frac{2^{-i+11}}{2^{60}}\leq 2^{-20}.

Then similarly, we set k=2mlog|𝔽|k=2\left\lceil m\log|\mathbb{F}|\right\rceil and obtain f(k)f^{(k)} as a parallel degree-mdmd polynomial. Moreover, we have

Δ(f(k),f)\displaystyle\Delta(f^{(k)},f) i=1k1Δ(f(i+1),f(i))i=1k14δi\displaystyle\leq\sum_{i=1}^{k-1}\Delta(f^{(i+1)},f^{(i)})\leq\sum_{i=1}^{k-1}4\cdot\delta_{i} (since f(1)=ff^{(1)}=f and by Lemma A.8)
=i=1k142i+1δ230δ,\displaystyle=\sum_{i=1}^{k-1}4\cdot 2^{-i+1}\cdot\delta\leq 2^{30}\delta,

which completes the proof. ∎

A.3 One Round of Correction

This section is devoted to proving Lemma A.8, which follows the sketch outlined in [9].

Lemma (Lemma A.8 Restated).

Assume |𝔽|3d|\mathbb{F}|\geq 3d. Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set and TST\subseteq S of size |T||S|/2|T|\geq|S|/2. Let f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t}. If

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 accepts]1δ,\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{T}^{f,f_{\mathbb{L}}}\text{ accepts}\right]\geq 1-\delta,

then for any 2δγ2202\delta\leq\gamma\leq 2^{-20}, there exists f:𝔽m𝔽tf^{\prime}\colon\mathbb{F}^{m}\to\mathbb{F}^{t} and TTT^{\prime}\subseteq T with the following properties:

  1. 1.

    |T|(1δγ)|T||T|2|T^{\prime}|\geq\left(1-\frac{\delta}{\gamma}\right)|T|\geq\frac{|T|}{2}.

  2. 2.

    Δ(f,f)4δ\Delta(f^{\prime},f)\leq 4\delta.

  3. 3.

    𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 accepts]1240γ(1|𝔽|+λ)\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}^{f^{\prime},f_{\mathbb{L}}}_{T^{\prime}}\text{ accepts}\right]\geq 1-2^{40}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right).

Proof.

Let TTT^{\prime}\subseteq T be the set of directions yTy\in T such that for at least 1γ1-\gamma fraction of x𝔽mx\in\mathbb{F}^{m}, ff agrees with the parallel line polynomial x,y\ell_{x,y} on xx. That is,

T={yT:|{x𝔽m:f(x)=f𝕃(x,y)(x)}|(1γ)|𝔽|m}.T^{\prime}=\left\{y\in T\colon|\left\{x\in\mathbb{F}^{m}\colon f(x)=f_{\mathbb{L}}(\ell_{x,y})(x)\right\}|\geq(1-\gamma)\cdot|\mathbb{F}|^{m}\right\}. (10)

Then

1δ\displaystyle 1-\delta 𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 accepts]\displaystyle\leq\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{T}^{f,f_{\mathbb{L}}}\text{ accepts}\right] (by assumption)
=𝐏𝐫x𝔽m,yT[f(x)=f𝕃(x,y)(x)]\displaystyle=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T}\left[f(x)=f_{\mathbb{L}}(\ell_{x,y})(x)\right] (by Definition A.1)
=|T||T|𝐏𝐫x𝔽m,yT[f(x)=f𝕃(x,y)(x)|yT]+|TT||T|𝐏𝐫x𝔽m,yT[f(x)=f𝕃(x,y)(x)|yT]\displaystyle=\frac{|T^{\prime}|}{|T|}\!\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T}\left[f(x)=f_{\mathbb{L}}(\ell_{x,y})(x)\,\middle|\,y\in T^{\prime}\right]+\frac{|T\setminus T^{\prime}|}{|T|}\!\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T}\left[f(x)=f_{\mathbb{L}}(\ell_{x,y})(x)\,\middle|\,y\notin T^{\prime}\right]
|T||T|+(1|T||T|)(1γ),\displaystyle\leq\frac{|T^{\prime}|}{|T|}+\left(1-\frac{|T^{\prime}|}{|T|}\right)\cdot\left(1-\gamma\right), (by Equation 10)

which implies Item 1 by rearranging.

To construct f:𝔽m𝔽tf^{\prime}\colon\mathbb{F}^{m}\to\mathbb{F}^{t}, for each x𝔽mx\in\mathbb{F}^{m} we define f(x)𝔽tf^{\prime}(x)\in\mathbb{F}^{t} to be the most common value of f𝕃(x,y)(x)f_{\mathbb{L}}(\ell_{x,y})(x) over yTy\in T^{\prime} where we break tie arbitrarily. Now we verify Item 2. Let

B={x𝔽m:𝐏𝐫yT[f(x)f𝕃(x,y)(x)]12}.B=\left\{x\in\mathbb{F}^{m}\colon\operatorname*{\mathbf{Pr}}_{y\sim T^{\prime}}\left[f(x)\neq f_{\mathbb{L}}(\ell_{x,y})(x)\right]\geq\frac{1}{2}\right\}.

By the definition of ff^{\prime}, we know that f(x)=f(x)f^{\prime}(x)=f(x) holds for any xBx\notin B. Hence Item 2 reduces to showing 𝐏𝐫x𝔽m[xB]4δ\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m}}[x\in B]\leq 4\delta as follows:

𝐏𝐫x𝔽m[xB]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m}}[x\in B] =𝐏𝐫x𝔽m[𝐏𝐫yT[f(x)f𝕃(x,y)(x)]12]\displaystyle=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m}}\left[\operatorname*{\mathbf{Pr}}_{y\sim T^{\prime}}\left[f(x)\neq f_{\mathbb{L}}(\ell_{x,y})(x)\right]\geq\frac{1}{2}\right]
2𝐏𝐫x𝔽m,yT[f(x)f𝕃(x,y)(x)]\displaystyle\leq 2\cdot\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T^{\prime}}\left[f(x)\neq f_{\mathbb{L}}(\ell_{x,y})(x)\right] (by Markov’s inequality)
2|T||T|𝐏𝐫x𝔽m,yT[f(x)f𝕃(x,y)(x)]\displaystyle\leq 2\cdot\frac{|T|}{|T^{\prime}|}\cdot\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T}\left[f(x)\neq f_{\mathbb{L}}(\ell_{x,y})(x)\right] (since TTT^{\prime}\subseteq T)
=2|T||T|𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 rejects]2|T||T|δ\displaystyle=2\cdot\frac{|T|}{|T^{\prime}|}\cdot\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}_{T}^{f,f_{\mathbb{L}}}\text{ rejects}\right]\leq 2\cdot\frac{|T|}{|T^{\prime}|}\cdot\delta (by assumption)
4δ.\displaystyle\leq 4\delta. (by Item 1)

To prove Item 3, we first get rid of ff^{\prime}. For a fixed x𝔽mx\in\mathbb{F}^{m} and each b𝔽tb\in\mathbb{F}^{t}, let UbTU_{b}\subseteq T^{\prime} be the set of directions yy such that f𝕃(x,y)(x)=bf_{\mathbb{L}}(\ell_{x,y})(x)=b. Denote b=f(x)b^{*}=f^{\prime}(x), which is defined to be the most common value of f𝕃(x,y)(x)f_{\mathbb{L}}(\ell_{x,y})(x) over yTy\in T^{\prime}. Thus |Ub||Ub||U_{b^{*}}|\geq|U_{b}| for any b𝔽tb\in\mathbb{F}^{t}. As a result, we have

𝐏𝐫yT[f(x)=f𝕃(x,y)(x)]\displaystyle\operatorname*{\mathbf{Pr}}_{y\sim T^{\prime}}\left[f^{\prime}(x)=f_{\mathbb{L}}(\ell_{x,y})(x)\right] =|Ub||T|=|Ub||T|b|Ub||T|\displaystyle=\frac{|U_{b^{*}}|}{|T^{\prime}|}=\frac{|U_{b^{*}}|}{|T^{\prime}|}\cdot\sum_{b}\frac{|U_{b}|}{|T^{\prime}|} (since UbU_{b}’s form a partition of TT^{\prime})
b(|Ub||T|)2\displaystyle\geq\sum_{b}\left(\frac{|U_{b}|}{|T^{\prime}|}\right)^{2} (since |Ub||Ub||U_{b^{*}}|\geq|U_{b}| for all bb)
=𝐏𝐫y1,y2T[f𝕃(x,y1)(x)=f𝕃(x,y2)(x)].\displaystyle=\operatorname*{\mathbf{Pr}}_{y_{1},y_{2}\sim T^{\prime}}\left[f_{\mathbb{L}}(\ell_{x,y_{1}})(x)=f_{\mathbb{L}}(\ell_{x,y_{2}})(x)\right].

Now taking the negation and the expectation over random xx, we have

𝐏𝐫[𝖫𝖣𝖳𝖾𝗌𝗍Tf,f𝕃 rejects]\displaystyle\operatorname*{\mathbf{Pr}}\left[\mathsf{LDTest}^{f^{\prime},f_{\mathbb{L}}}_{T^{\prime}}\text{ rejects}\right] =𝐏𝐫x𝔽m,yT[f(x)f𝕃(x,y)(x)]\displaystyle=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y\sim T^{\prime}}\left[f^{\prime}(x)\neq f_{\mathbb{L}}(\ell_{x,y})(x)\right]
𝐏𝐫x𝔽m,y1,y2T[f𝕃(x,y1)(x)f𝕃(x,y2)(x)].\displaystyle\leq\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1},y_{2}\sim T^{\prime}}\left[f_{\mathbb{L}}(\ell_{x,y_{1}})(x)\neq f_{\mathbb{L}}(\ell_{x,y_{2}})(x)\right]. (11)

To upper bound Equation 11, we will use bivariate testing theorems [37] similar to [33, 19]. The idea is, as sketched in [9], to use Lemma A.3 to show the following claim.

Claim A.10.

With probability at least

1240γ(1|𝔽|+λ)1-2^{40}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)

over x,y1,y2x,y_{1},y_{2}, we have

  1. (a)

    𝐏𝐫i𝔽[f𝕃(x+iy1,y2)(x+iy1)f(x+iy1)]1200\operatorname*{\mathbf{Pr}}_{i\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1})\neq f(x+i\cdot y_{1})\right]\leq\frac{1}{200}.

  2. (b)

    𝐏𝐫j𝔽[f𝕃(x+jy2,y1)(x+jy2)f(x+jy2)]1200\operatorname*{\mathbf{Pr}}_{j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+j\cdot y_{2},y_{1}})(x+j\cdot y_{2})\neq f(x+j\cdot y_{2})\right]\leq\frac{1}{200}.

  3. (c)

    𝐏𝐫i,j𝔽[f𝕃(x+iy1,y2)(x+iy1+jy2)f(x+iy1+jy2)]1200\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1}+j\cdot y_{2})\neq f(x+i\cdot y_{1}+j\cdot y_{2})\right]\leq\frac{1}{200}.

  4. (d)

    𝐏𝐫i,j𝔽[f𝕃(x+jy2,y1)(x+iy1+jy2)f(x+iy1+jy2)]1200\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+j\cdot y_{2},y_{1}})(x+i\cdot y_{1}+j\cdot y_{2})\neq f(x+i\cdot y_{1}+j\cdot y_{2})\right]\leq\frac{1}{200}.

Claim A.10 intuitively says that ff is almost parallel degree-dd along the line x,y1\ell_{x,y_{1}} (Item (a)), the line x,y2\ell_{x,y_{2}} (Item (b)), and the plane spanned by x,y1,x,y2\ell_{x,y_{1}},\ell_{x,y_{2}} (Items (c) and (d)). For simplicity, define

R(i,j)=f𝕃(x+iy1,y2)(x+iy1+jy2)andC(i,j)=f𝕃(x+jy2,y1)(x+iy1+jy2).R(i,j)=f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1}+j\cdot y_{2})\quad\text{and}\quad C(i,j)=f_{\mathbb{L}}(\ell_{x+j\cdot y_{2},y_{1}})(x+i\cdot y_{1}+j\cdot y_{2}). (12)

Given Claim A.10 and Equation 11, it suffices to show

R(0,0)=C(0,0)assuming Items (a)(b)(c) and (d).R(0,0)=C(0,0)\quad\text{assuming \lx@cref{creftypepluralcap~refnum}{itm:bivariate_1}, \lx@cref{refnum}{itm:bivariate_2}, \lx@cref{refnum}{itm:bivariate_3} and~\lx@cref{refnum}{itm:bivariate_4}.} (13)

Observe that R:𝔽×𝔽𝔽tR\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F}^{t} is a parallel bivariate polynomial. Recall that f𝕃f_{\mathbb{L}} is the restriction of ff on each line of its closest parallel degree-dd polynomial. Thus for each fixed i𝔽i\in\mathbb{F}, R(i,j)R(i,j) is parallel degree-dd in the variable jj. Let R1,,Rt:𝔽×𝔽𝔽R_{1},\ldots,R_{t}\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F} be the single-output components of RR. Then for each r[t]r\in[t] and in Rr(i,j)R_{r}(i,j), the variable ii has degree at most |𝔽|1|\mathbb{F}|-1 and the variable jj has degree at most dd. We say RR is of parallel degree (|𝔽|1,d)(|\mathbb{F}|-1,d) for shorthand. Similarly, C(i,j)C(i,j) is of parallel degree (d,|𝔽|1)(d,|\mathbb{F}|-1).

Combining Items (c) and (d), we have

𝐏𝐫i,j𝔽[R(i,j)C(i,j)]1100.\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[R(i,j)\neq C(i,j)\right]\leq\frac{1}{100}. (14)

We will use the following lemma to zero out the inconsistent entries of R(i,j)R(i,j) and C(i,j)C(i,j).

Lemma A.11 ([37, Lemma 3]).

Let Z𝔽×𝔽Z\subseteq\mathbb{F}\times\mathbb{F} be arbitrary. There exists a non-zero bivariate polynomial E:𝔽×𝔽𝔽E\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F} of degree (|Z|,|Z|)\left(\left\lfloor\sqrt{|Z|}\right\rfloor,\left\lfloor\sqrt{|Z|}\right\rfloor\right) such that E(i,j)=0E(i,j)=0 for all (i,j)Z(i,j)\in Z.

By setting ZZ to be the set of (i,j)(i,j) with R(i,j)C(i,j)R(i,j)\neq C(i,j), we have |Z||𝔽|2/100|Z|\leq|\mathbb{F}|^{2}/100 by Equation 14. Thus by Lemma A.11, there is a non-zero bivariate polynomial EE of degree at most (|𝔽|10,|𝔽|10)\left(\frac{|\mathbb{F}|}{10},\frac{|\mathbb{F}|}{10}\right) such that E(i,j)R(i,j)=E(i,j)C(i,j)E(i,j)R(i,j)=E(i,j)C(i,j) holds for all i,j𝔽i,j\in\mathbb{F}. Note that RR and CC are polynomials with tt outputs and EE has single output. The product E(i,j)R(i,j)E(i,j)R(i,j) produces a vector of length tt with entries of R(i,j)R(i,j) scaled by E(i,j)E(i,j); same for E(i,j)C(i,j)E(i,j)C(i,j).

Then we use the following lemma to show that RR and CC are close to a parallel bivariate polynomial of parallel degree (d,d)(d,d).

Lemma A.12 ([37, Lemma 8]).

Let E,P:𝔽×𝔽𝔽E,P\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F} be bivariate polynomials of degree (b,a)(b,a) and (b+d,a+d)(b+d,a+d) respectively. Assume n>min{2b+2d,2a+2d}n>\min\left\{2b+2d,2a+2d\right\}. Assume further there exist distinct i1,,ini_{1},\ldots,i_{n} such that E(ik,)E(i_{k},\cdot) divides P(ik,)P(i_{k},\cdot) for all k[n]k\in[n], and distinct j1,,jnj_{1},\ldots,j_{n} such that E(,jk)E(\cdot,j_{k}) divides P(,jk)P(\cdot,j_{k}) for all k[n]k\in[n]. Then E(,)E(\cdot,\cdot) divides P(,)P(\cdot,\cdot), i.e., there exists a bivariate polynomial Q:𝔽×𝔽𝔽Q\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F} of degree (d,d)(d,d) such that E(i,j)Q(i,j)=P(i,j)E(i,j)Q(i,j)=P(i,j) holds for all i,j𝔽i,j\in\mathbb{F}.

Fix an arbitrary r[t]r\in[t]. For each i,j𝔽i,j\in\mathbb{F}, define Pr(i,j)=E(i,j)Rr(i,j)P_{r}(i,j)=E(i,j)R_{r}(i,j), which also equals E(i,j)Cr(i,j)E(i,j)C_{r}(i,j) since E(i,j)R(i,j)=E(i,j)C(i,j)E(i,j)R(i,j)=E(i,j)C(i,j). Since EE is non-zero and has degree (|𝔽|10,|𝔽|10)\left(\frac{|\mathbb{F}|}{10},\frac{|\mathbb{F}|}{10}\right), there are at least 9|𝔽|10\frac{9|\mathbb{F}|}{10} many distinct i𝔽i\in\mathbb{F} such that E(i,)E(i,\cdot) is an univariate polynomial not identically zero, for which E(i,)E(i,\cdot) divides Pr(i,)P_{r}(i,\cdot); same for E(,j)E(\cdot,j) and Pr(,j)P_{r}(\cdot,j). By Lemma A.12 with a=b=|𝔽|10a=b=\frac{|\mathbb{F}|}{10} and n=9|𝔽|10n=\frac{9|\mathbb{F}|}{10}, there exists a bivariate polynomial Qr:𝔽×𝔽𝔽Q_{r}\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F} with parallel degree (d,d)(d,d) such that E(i,j)Qr(i,j)=Pr(i,j)=E(i,j)Rr(i,j)=E(i,j)Cr(i,j)E(i,j)Q_{r}(i,j)=P_{r}(i,j)=E(i,j)R_{r}(i,j)=E(i,j)C_{r}(i,j) holds for all i,j𝔽i,j\in\mathbb{F}, where we used the fact that |𝔽|3d|\mathbb{F}|\geq 3d implies |𝔽|>10d7|\mathbb{F}|>\frac{10d}{7}.151515If d1d\geq 1, then 3d>10d73d>\frac{10d}{7}; otherwise 10d7=0<1|𝔽|\frac{10d}{7}=0<1\leq|\mathbb{F}|.

Let Q:𝔽×𝔽𝔽tQ\colon\mathbb{F}\times\mathbb{F}\to\mathbb{F}^{t} be the parallel-output bivariate polynomial with single-output components Q1,,QrQ_{1},\ldots,Q_{r} obtained above. Then Q(i,)=R(i,)Q(i,\cdot)=R(i,\cdot) holds as long as E(i,)E(i,\cdot) is not identically zero, which has at least 9|𝔽|10\frac{9|\mathbb{F}|}{10} possibilities out of |𝔽||\mathbb{F}| total choices off ii. Recall Item (a). For at least 9101200>45\frac{9}{10}-\frac{1}{200}>\frac{4}{5} fraction of ii’s, we have f(x+iy1)=R(i,0)=Q(i,0)f(x+i\cdot y_{1})=R(i,0)=Q(i,0). Note that the distance between any two distinct (parallel) degree-dd polynomial is at least 1d|𝔽|1-\frac{d}{|\mathbb{F}|}, which is at least 23<215\frac{2}{3}<2\cdot\frac{1}{5} since |𝔽|3d|\mathbb{F}|\geq 3d. In addition, Q(,0)Q(\cdot,0) is parallel degree-dd. Thus Q(,0)Q(\cdot,0) is the closest parallel degree-dd polynomial of f|x,y1f|_{\ell_{x,y_{1}}}, i.e., Q(i,0)=f𝕃(x,y1)(x+iy1)=C(i,0)Q(i,0)=f_{\mathbb{L}}(\ell_{x,y_{1}})(x+i\cdot y_{1})=C(i,0) holds for all i𝔽i\in\mathbb{F} where we recall Equation 12. In particular, this means Q(0,0)=C(0,0)Q(0,0)=C(0,0).

Similarly, using Item (b), we can show Q(0,0)=R(0,0)Q(0,0)=R(0,0), which verifies Equation 13 and thus Item 3, and completes the proof of Lemma A.8. ∎

Finally it remains to prove Claim A.10.

Proof of Claim A.10.

We show that each item holds with probability at least 1230γ(1|𝔽|+λ)1-2^{30}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right), and then Claim A.10 follows from a union bound.

We first consider Item (a) and the analysis for Item (b) is almost identical. For each ySy\in S, define

By={x𝔽m:f𝕃(x,y)(x)f(x)}.B_{y}=\left\{x\in\mathbb{F}^{m}\colon f_{\mathbb{L}}(\ell_{x,y})(x)\neq f(x)\right\}.

Then for any fixed y2Ty_{2}\in T^{\prime}, we have |By2|γ|𝔽|m|B_{y_{2}}|\leq\gamma\cdot|\mathbb{F}|^{m} by Equation 10, and, moreover,

𝐏𝐫x𝔽m,y1T[Item (a) does not hold]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\text{\lx@cref{creftypecap~refnum}{itm:bivariate_1} does not hold}\right]
=\displaystyle= 𝐏𝐫x𝔽m,y1T[𝐏𝐫i𝔽[f𝕃(x+iy1,y2)(x+iy1)f(x+iy1)]>1200]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\operatorname*{\mathbf{Pr}}_{i\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1})\neq f(x+i\cdot y_{1})\right]>\frac{1}{200}\right]
=\displaystyle= 𝐏𝐫x𝔽m,y1T[𝐏𝐫i𝔽[x+iy1By2]>1200]=𝐏𝐫x𝔽m,y1T[|x,y1By2||x,y1|>1200]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\operatorname*{\mathbf{Pr}}_{i\sim\mathbb{F}}\left[x+i\cdot y_{1}\in B_{y_{2}}\right]>\frac{1}{200}\right]=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\frac{|\ell_{x,y_{1}}\cap B_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{1}{200}\right]
\displaystyle\leq 𝐏𝐫x𝔽m,y1T[|x,y1By2||x,y1|>|By2||𝔽|m+1400]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\frac{|\ell_{x,y_{1}}\cap B_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}+\frac{1}{400}\right] (since |By2||𝔽|mγ\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}\leq\gamma and γ220\gamma\leq 2^{-20} by assumption)
\displaystyle\leq |S||T|𝐏𝐫x𝔽m,y1S[|x,y1By2||x,y1|>|By2||𝔽|m+1400]\displaystyle\frac{|S|}{|T^{\prime}|}\cdot\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim S}\left[\frac{|\ell_{x,y_{1}}\cap B_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}+\frac{1}{400}\right] (since TTST^{\prime}\subseteq T\subseteq S)
\displaystyle\leq 4𝐏𝐫x𝔽m,y1S[|x,y1By2||x,y1|>|By2||𝔽|m+1400]\displaystyle 4\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim S}\left[\frac{|\ell_{x,y_{1}}\cap B_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}+\frac{1}{400}\right] (by assumption and Item 1)
\displaystyle\leq 220(1|𝔽|+λ)|By2||𝔽|m\displaystyle 2^{20}\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)\cdot\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}} (by Lemma A.3)
\displaystyle\leq 220γ(1|𝔽|+λ).\displaystyle 2^{20}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right). (since |By2||𝔽|mγ\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}\leq\gamma)

By taking the expectation over y2y_{2}, Item (a) holds with the desired probability.

Now we consider Item (c) and the analysis for Item (d) is almost identical. For each ySy\in S, define

B¯y={x𝔽m:𝐏𝐫j𝔽[f𝕃(x,y)(x+jy)f(x+jy)]>210}.\overline{B}_{y}=\left\{x\in\mathbb{F}^{m}\colon\operatorname*{\mathbf{Pr}}_{j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x,y})(x+j\cdot y)\neq f(x+j\cdot y)\right]>2^{-10}\right\}. (15)

For each y2Sy_{2}\in S, we have

|B¯y2||𝔽|m\displaystyle\frac{|\overline{B}_{y_{2}}|}{|\mathbb{F}|^{m}} =𝐏𝐫x𝔽m[𝐏𝐫j𝔽[f𝕃(x,y2)(x+jy2)f(x+jy2)]>210]\displaystyle=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m}}\left[\operatorname*{\mathbf{Pr}}_{j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x,y_{2}})(x+j\cdot y_{2})\neq f(x+j\cdot y_{2})\right]>2^{-10}\right]
210𝐏𝐫x𝔽m,j𝔽[f𝕃(x,y2)(x+jy2)f(x+jy2)]\displaystyle\leq 2^{10}\cdot\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x,y_{2}})(x+j\cdot y_{2})\neq f(x+j\cdot y_{2})\right] (by Markov’s inequality)
=210𝐏𝐫x𝔽m,j𝔽[f𝕃(x+jy2,y2)(x+jy2)f(x+jy2)]\displaystyle=2^{10}\cdot\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+j\cdot y_{2},y_{2}})(x+j\cdot y_{2})\neq f(x+j\cdot y_{2})\right] (since x,y2=x+jy2,y2\ell_{x,y_{2}}=\ell_{x+j\cdot y_{2},y_{2}})
=210𝐏𝐫z𝔽m[f𝕃(z,y2)(z)f(z)]=210|By2||𝔽|m\displaystyle=2^{10}\cdot\operatorname*{\mathbf{Pr}}_{z\sim\mathbb{F}^{m}}\left[f_{\mathbb{L}}(\ell_{z,y_{2}})(z)\neq f(z)\right]=2^{10}\cdot\frac{|B_{y_{2}}|}{|\mathbb{F}|^{m}}
210γ,\displaystyle\leq 2^{10}\cdot\gamma, (16)

and

𝐏𝐫x𝔽m,y1T[Item (c) does not hold]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\text{\lx@cref{creftypecap~refnum}{itm:bivariate_3} does not hold}\right]
=\displaystyle= 𝐏𝐫x𝔽m,y1T[𝐏𝐫i,j𝔽[f𝕃(x+iy1,y2)(x+iy1+jy2)f(x+iy1+jy2)]>1200]\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1}+j\cdot y_{2})\neq f(x+i\cdot y_{1}+j\cdot y_{2})\right]>\frac{1}{200}\right]
\displaystyle\leq 𝐏𝐫x𝔽m,y1T[𝐏𝐫i𝔽[x+iy1B¯y2]>1300],\displaystyle\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\operatorname*{\mathbf{Pr}}_{i\sim\mathbb{F}}\left[x+i\cdot y_{1}\in\overline{B}_{y_{2}}\right]>\frac{1}{300}\right], (17)

where we use the following reasoning for the last inequality: if the event inside bracket does not happen, then

𝐏𝐫i,j𝔽[f𝕃(x+iy1,y2)(x+iy1+jy2)f(x+iy1+jy2)]\displaystyle\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1}+j\cdot y_{2})\neq f(x+i\cdot y_{1}+j\cdot y_{2})\right]
\displaystyle\leq 1300+𝐏𝐫i,j𝔽[f𝕃(x+iy1,y2)(x+iy1+jy2)f(x+iy1+jy2)|x+iy1B¯y2]\displaystyle\frac{1}{300}+\operatorname*{\mathbf{Pr}}_{i,j\sim\mathbb{F}}\left[f_{\mathbb{L}}(\ell_{x+i\cdot y_{1},y_{2}})(x+i\cdot y_{1}+j\cdot y_{2})\neq f(x+i\cdot y_{1}+j\cdot y_{2})\,\middle|\,x+i\cdot y_{1}\notin\overline{B}_{y_{2}}\right]
\displaystyle\leq 1300+210<1200.\displaystyle\frac{1}{300}+2^{-10}<\frac{1}{200}. (by Equation 15)

Then similar to the analysis for Item (a), we continue upper bounding Equation 17 as follows:

RHS of Equation 17 =𝐏𝐫x𝔽m,y1T[|x,y1B¯y2||x,y1|>1300]\displaystyle=\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\frac{|\ell_{x,y_{1}}\cap\overline{B}_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{1}{300}\right]
𝐏𝐫x𝔽m,y1T[|x,y1B¯y2||x,y1|>|B¯y2||𝔽|m+1400]\displaystyle\leq\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim T^{\prime}}\left[\frac{|\ell_{x,y_{1}}\cap\overline{B}_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{|\overline{B}_{y_{2}}|}{|\mathbb{F}|^{m}}+\frac{1}{400}\right] (by Equation 16 and γ220\gamma\leq 2^{-20})
4𝐏𝐫x𝔽m,y1S[|x,y1B¯y2||x,y1|>|B¯y2||𝔽|m+1400]\displaystyle\leq 4\operatorname*{\mathbf{Pr}}_{x\sim\mathbb{F}^{m},y_{1}\sim S}\left[\frac{|\ell_{x,y_{1}}\cap\overline{B}_{y_{2}}|}{|\ell_{x,y_{1}}|}>\frac{|\overline{B}_{y_{2}}|}{|\mathbb{F}|^{m}}+\frac{1}{400}\right] (since TST^{\prime}\subseteq S and |T||S|/4|T^{\prime}|\geq|S|/4)
220(1|𝔽|+λ)|B¯y2||𝔽|m230γ(1|𝔽|+λ).\displaystyle\leq 2^{20}\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right)\cdot\frac{|\overline{B}_{y_{2}}|}{|\mathbb{F}|^{m}}\leq 2^{30}\cdot\gamma\cdot\left(\frac{1}{|\mathbb{F}|}+\lambda\right). (by Lemma A.3 and Equation 16)

Therefore Item (c) holds with the desired probabilit by taking the expectation over y1y_{1}. ∎

A.4 Derandomized Characterizations of Parallel Low Degree Polynomials

In this part, we extend the characterization of low degree polynomials in [39] to the derandomized setting, where we only consider lines with directions generated by a (large subset of) small-biased set.

We use superscript \top to denote vector and matrix transpose. For two vectors u,v𝔽du,v\in\mathbb{F}^{d}, we use u,v\left\langle u,v\right\rangle to denote their inner product which equals uvu^{\top}v (or vuv^{\top}u).

Theorem (Theorem A.9 Restated).

Let S𝔽mS\subseteq\mathbb{F}^{m} be a λ\lambda-biased set and TST\subseteq S of size |T|>1+λ2|S||T|>\frac{1+\lambda}{2}\cdot|S|. Then f:𝔽m𝔽tf\colon\mathbb{F}^{m}\to\mathbb{F}^{t} is parallel degree-mdmd if f|x,yf|_{\ell_{x,y}} is parallel degree-dd for every x𝔽m,yTx\in\mathbb{F}^{m},y\in T.

Proof.

Assume without loss of generality that t=1t=1, since we can apply the analysis individually for each single-output component f[1],,f[t]f[1],\ldots,f[t].

We first prove that TT has the full rank mm. Assume towards contradiction that TT has rank at most m1m-1. Then there exists a non-zero vector z𝔽mz\in\mathbb{F}^{m} such that zy=0z^{\top}y=0 holds for all yTy\in T. In light of Definition A.2, we construct a non-trivial homomorphism χ:𝔽mμp\chi\colon\mathbb{F}^{m}\to\mu_{p} to derive a contradiction, where pp is the characteristic of 𝔽\mathbb{F}. Let ξ:𝔽μp\xi\colon\mathbb{F}\to\mu_{p} be a non-trivial (group) homomorphism, where we view 𝔽\mathbb{F} as an additive group. We define for each x𝔽mx\in\mathbb{F}^{m} that

χ(x)=ξ(xz),\chi(x)=\xi(x^{\top}z),

which is a non-trivial homomorphism since ξ\xi is a non-trivial homomorphism and z0mz\neq 0^{m}. Note that for all yTy\in T, we have

χ(y)=ξ(yz)=ξ(0)=1.\chi(y)=\xi(y^{\top}z)=\xi(0)=1.

Thus

|𝔼yS[χ(y)]|\displaystyle\left|\operatorname*{\mathbb{E}}_{y\sim S}\left[\chi(y)\right]\right| =|𝔼yS[χ(y)(𝟙yT+𝟙yT)]|=|𝔼yS[𝟙yT+χ(y)𝟙yT]|\displaystyle=\left|\operatorname*{\mathbb{E}}_{y\sim S}\left[\chi(y)\cdot\left(\mathbbm{1}_{y\in T}+\mathbbm{1}_{y\notin T}\right)\right]\right|=\left|\operatorname*{\mathbb{E}}_{y\sim S}\left[\mathbbm{1}_{y\in T}+\chi(y)\cdot\mathbbm{1}_{y\notin T}\right]\right|
𝔼yS[𝟙yT]𝔼yS[𝟙yT]=|T||S||S||T||S|\displaystyle\geq\operatorname*{\mathbb{E}}_{y\sim S}\left[\mathbbm{1}_{y\in T}\right]-\operatorname*{\mathbb{E}}_{y\sim S}\left[\mathbbm{1}_{y\notin T}\right]=\frac{|T|}{|S|}-\frac{|S|-|T|}{|S|}
>λ,\displaystyle>\lambda, (since |T|>1+λ2|S||T|>\frac{1+\lambda}{2}\cdot|S|)

which contradicts Definition A.2.

Now we fix a set of mm linearly independent directions y1,,ymTy_{1},\ldots,y_{m}\in T. Then we can interpolate ff using degree-dd polynomials f|x,y1,,f|x,ymf|_{\ell_{x,y_{1}}},\ldots,f|_{\ell_{x,y_{m}}} for x𝔽mx\in\mathbb{F}^{m}. For concreteness, we apply the invertible linear transform on 𝔽m\mathbb{F}^{m} such that y1,,ymy_{1},\ldots,y_{m} map to axis parallel directions e1,,eme_{1},\ldots,e_{m}, where ei=(0,,0i1,1,0,,0mi)e_{i}=(\underbrace{0,\ldots,0}_{i-1},1,\underbrace{0,\ldots,0}_{m-i}). Let ff^{\prime} be the polynomial after this transform, which shares the same degree with ff since the transform is invertible and linear. Since f|x,y1,,f|x,ymf|_{\ell_{x,y_{1}}},\ldots,f|_{\ell_{x,y_{m}}} are all degree-dd, we know that ff^{\prime} is degree-dd along the axis parallel lines. Then by polynomial interpolation (see e.g., [39, Lemma 28]), ff^{\prime} has degree at most mdmd, which in turn means that ff has degree at most mdmd as claimed. ∎