IMPROVING 3N CIRCUITCOMPLEXITY LOWER BOUNDS

While it can be easily shown by counting that almost all Boolean predicates of n variables have circuit size Omega(2(n)/n), we have no example of an NP function requiring even a superlinear number of gates. Moreover, only modest linear lower bounds are known. Until recently, the strongest known lower bound was 3n-o(n) presented by Blum in 1984. In 2011, Demenkov and Kulikov presented a much simpler proof of the same lower bound, but for a more complicated function -an affine disperser for sublinear dimension. Informally, this is a function that is resistant to any n-o(n)affine substitutions. In 2011, Ben-Sasson and Kopparty gave an explicit construction of such a function. The proof of the lower bound basically goes by showing that for any circuit there exists an affine hyperplane where the function complexity decreases at least by three gates. In this paper, we prove the following two extensions.1. A (3+1/86)n-o(n) lower bound for the circuit size of an affine disperser for sublinear dimension. The proof is based on the gate elimination technique extended with the following three ideas: (i) generalizing the computational model by allowing circuits to contain cycles, this in turn allows us to perform affine substitutions, (ii) a carefully chosen circuit complexity measure to track the progress of the gate elimination process, and (iii) quadratic substitutions that may be viewed as delayed affine substitutions.2. A much simpler proof of a stronger lower bound of 3.11n for a quadratic disperser. Informally, a quadratic disperser is resistant to sufficiently many substitutions of the form x <- p, where p is a polynomial of degree at most two. Currently, there are no constructions of quadratic dispersers in NP (although there are constructions over large fields, and constructions with weaker parameters over GF(2)). The key ingredient of this proof is the induction on the size of the underlying quadratic variety instead of the number of variables as in the previously known proofs.