English

Subspace-Invariant AC$^0$ Formulas

Logic in Computer Science 2023-06-22 v4 Computational Complexity

Abstract

We consider the action of a linear subspace UU of {0,1}n\{0,1\}^n on the set of AC0^0 formulas with inputs labeled by literals in the set {X1,X1,,Xn,Xn}\{X_1,\overline X_1,\dots,X_n,\overline X_n\}, where an element uUu \in U acts on formulas by transposing the iith pair of literals for all i[n]i \in [n] such that ui=1u_i=1. A formula is {\em UU-invariant} if it is fixed by this action. For example, there is a well-known recursive construction of depth d+1d+1 formulas of size O(n2dn1/d)O(n{\cdot}2^{dn^{1/d}}) computing the nn-variable PARITY function; these formulas are easily seen to be PP-invariant where PP is the subspace of even-weight elements of {0,1}n\{0,1\}^n. In this paper we establish a nearly matching 2d(n1/d1)2^{d(n^{1/d}-1)} lower bound on the PP-invariant depth d+1d+1 formula size of PARITY. Quantitatively this improves the best known Ω(2184d(n1/d1))\Omega(2^{\frac{1}{84}d(n^{1/d}-1)}) lower bound for {\em unrestricted} depth d+1d+1 formulas, while avoiding the use of the switching lemma. More generally, for any linear subspaces UVU \subset V, we show that if a Boolean function is UU-invariant and non-constant over VV, then its UU-invariant depth d+1d+1 formula size is at least 2d(m1/d1)2^{d(m^{1/d}-1)} where mm is the minimum Hamming weight of a vector in UVU^\bot \setminus V^\bot.

Keywords

Cite

@article{arxiv.1806.04831,
  title  = {Subspace-Invariant AC$^0$ Formulas},
  author = {Benjamin Rossman},
  journal= {arXiv preprint arXiv:1806.04831},
  year   = {2023}
}
R2 v1 2026-06-23T02:28:07.992Z