arXiv++ Combinatorics

In this paper we show lower bounds for a certain large class of algorithms solving the Graph Isomorphism problem, even on expander graph instances. Spielman [25] shows an algorithm for isomorphism of strongly regular expander graphs that runs in time exp(O(n^(1/3)) (this bound was recently improved to expf O(n^(1/5) [5]). It has since been an open question to remove the requirement that the graph be strongly regular. Recent algorithmic results show that for many problems the Lasserre hierarchy works surprisingly well when the underlying graph has expansion properties. Moreover, recent work of Atserias and Maneva [3] shows that k rounds of the Lasserre hierarchy is a generalization of the k-dimensional Weisfeiler-Lehman algorithm for Graph Isomorphism. These two facts combined make the Lasserre hierarchy a good candidate for solving graph isomorphism on expander graphs. Our main result rules out this promising direction by showing that even Omega(n) rounds of the Lasserre semidefinite program hierarchy fail to solve the Graph Isomorphism problem even on expander graphs.

We consider the problem of testing if a given function f : F_2^n -> F_2 is close to any degree d polynomial in n variables, also known as the Reed-Muller testing problem. The Gowers norm is based on a natural 2^{d+1}-query test for this property. Alon et al. [AKKLR05] rediscovered this test and showed that it accepts every degree d polynomial with probability 1, while it rejects functions that are Omega(1)-far with probability Omega(1/(d 2^{d})). We give an asymptotically optimal analysis of this test, and show that it rejects functions that are (even only) Omega(2^{-d})-far with Omega(1)-probability (so the rejection probability is a universal constant independent of d and n). This implies a tight relationship between the (d+1)st Gowers norm of a function and its maximal correlation with degree d polynomials, when the correlation is close to 1. Our proof works by induction on n and yields a new analysis of even the classical Blum-Luby-Rubinfeld [BLR93] linearity test, for the setting of functions mapping F_2^n to F_2. The optimality follows from a tighter analysis of counterexamples to the "inverse conjecture for the Gowers norm" constructed by [GT09,LMS08]. Our result has several implications. First, it shows that the Gowers norm test is tolerant, in that it also accepts close codewords. Second, it improves the parameters of an XOR lemma for polynomials given by Viola and Wigderson [VW07]. Third, it implies a "query hierarchy" result for property testing of affine-invariant properties. That is, for every function q(n), it gives an affine-invariant property that is testable with O(q(n))-queries, but not with o(q(n))-queries, complementing an analogous result of [GKNR09] for graph properties.