Inverse-exponential correlation bounds and extremely rigid matrices from a new derandomized XOR lemma
Inverse-exponential correlation bounds and extremely rigid matrices from a new derandomized XOR lemma
复制标题
来自新的去随机 XOR 引理的反指数相关界限和极其严格的矩阵
DOI:
--
复制
发表时间:
2021
期刊:
影响因子:
--
通讯作者:
Xin Lyu
中科院分区:
文献类型:
--
作者:
Lijie Chen;Xin Lyu
In this work we prove that there is a function f ∈ E NP such that, for every sufficiently large n and d = √n/logn, fn (f restricted to n-bit inputs) cannot be (1/2 + 2−d)-approximated by F2-polynomials of degree d. We also observe that a minor improvement (e.g., improving d to n1/2+ε for any ε > 0) over our result would imply E NP cannot be computed by depth-3 AC0-circuits of 2n1/2 + ε size, which is a notoriously hard open question in complexity theory. Using the same proof techniques, we are also able to construct extremely rigid matrices over F2 in P NP. More specifically, we show that for every constant ε ∈ (0,1), there is a P NP algorithm which on input 1n outputs an n× n F2-matrix Hn satisfying RHn(2log1 − ε n) ≥ (1/2 − exp(−log2/3 · ε n) ) · n2, for every sufficiently large n. This improves the recent P NP constructions of rigid matrices in [Alman and Chen, FOCS 2019] and [Bhangale et al., FOCS 2020], which only give Ω(n2) rigidity. The key ingredient in the proof of our new results is a new derandomized XOR lemma based on approximate linear sums, which roughly says that given an n-input function f which cannot be 0.99-approximated by certain linear sum of s many functions in F within ℓ1-distance, one can construct a new function Ampf with O(n) input bits, which cannot be (1/2+sΩ(1))-approximated by F-functions. Taking F to be a function collection containing low-degree F2-polynomials or low-rank F2-matrices, our results are then obtained by first using the algorithmic method to construct a function which is weakly hard against linear sums of F in the above sense, and then applying the derandomized XOR lemma to f. We obtain our new derandomized XOR lemma by giving a generalization of the famous hardcore lemma by Impagliazzo. Our generalization in some sense constructs a non-Boolean hardcore of a weakly hard function f with respect to F-functions, from the weak inapproximability of f by any linear sum of F with bounded ℓp-norm. This generalization recovers the original hardcore lemma by considering the ℓ∞-norm. Surprisingly, when we switch to the ℓ1-norm, we immediately rediscover Levin’s proof of Yao’s XOR Lemma. That is, these first two proofs of Yao’s XOR Lemma can be unified with our new perspective. For proving the correlation bounds, our new derandomized XOR lemma indeed works with the ℓ4/3-norm.
登录
查看更多内容
DOI:
10.1109/focs.2018.00093
发表时间:
2018
期刊:
59th Annual IEEE Symposium on Foundations of Computer Science (FOCS 2018
影响因子:
--
作者:
Forbes, Michael A.;Kelley, Zander
通讯作者:
Kelley, Zander
DOI:
--
发表时间:
2019
期刊:
(ITCS
影响因子:
--
作者:
Chattopadhyay, Eshan;Hatami, Pooya;Lovett, Shachar;Tal, Avishay
通讯作者:
Tal, Avishay
DOI:
10.1007/978-3-030-79416-3_11
发表时间:
2021
期刊:
2021
影响因子:
--
作者:
Huang, Xuangui;Viola, Emanuele
通讯作者:
Viola, Emanuele
DOI:
10.1109/focs46700.2020.00009
发表时间:
2020
期刊:
2020 IEEE 61st Annual Symposium on Foundations of Computer Science (FOCS
影响因子:
--
作者:
Chen, Lijie;Lyu, Xin;Williams, R. Ryan
通讯作者:
Williams, R. Ryan
DOI:
10.1145/3357713.3384242
发表时间:
2020
期刊:
52nd Annual ACM Symposium on Theory of Computing (STOC
影响因子:
--
作者:
Chattopadhyay, Eshan;Hatami, Pooya;Hosseini, Kaave;Lovett, Shachar;Zuckerman, David
通讯作者:
Zuckerman, David