A Casual Tour Around a Circuit Complexity Bound

I will discuss the recent proof that the complexity class NEXP (nondeterministic exponential time) lacks nonuniform ACC circuits of polynomial size. The proof will be described from the perspective of someone trying to discover it.

💡 Research Summary

The paper presents a detailed exposition of the breakthrough result that the nondeterministic exponential‑time class NEXP does not have polynomial‑size non‑uniform ACC⁰ circuits. The authors frame the discussion as a narrative of discovery, walking the reader through the two main technical pillars that underpin the proof.

First, they describe a sub‑exponential‑time algorithm for the satisfiability problem on ACC⁰ circuits (ACC‑SAT). By converting any ACC⁰ circuit into a “convolution normal form” and then applying Fourier analysis together with low‑degree polynomial approximations, the algorithm exploits the periodic nature of MODₘ gates to reduce the effective degree of the representing polynomial. This reduction enables a SAT solver that runs in time 2^{n^{o(1)}}, dramatically faster than the naïve 2^{n} bound.

Second, the paper explains how this algorithm is leveraged through the hardness‑versus‑randomness paradigm to obtain a circuit lower bound. Using the known equivalence NEXP = MIP, an NEXP language can be expressed as a multi‑prover interactive proof whose verifier can be simulated by an ACC⁰ circuit of polynomial size. If NEXP were contained in non‑uniform ACC⁰, then the verifier’s circuit would exist for every input length. The newly developed ACC‑SAT algorithm would then allow us to decide the verifier’s acceptance condition in sub‑exponential time, yielding a deterministic algorithm for every NEXP problem running in 2^{n^{o(1)}} time. This contradicts the established separation NEXP ⊄ SUBEXP, thereby proving NEXP ⊄ ACC⁰.

The authors also discuss the broader implications of the technique. The ACC‑SAT algorithm itself is a valuable tool that may be adapted to other circuit classes, and the hardness‑versus‑randomness connection illustrated here suggests a general template: an efficient algorithm for a restricted circuit class can translate into a lower bound against that class for powerful complexity classes. The paper surveys prior work on ACC⁰ lower bounds, highlights how this result improves upon earlier sub‑exponential barriers, and outlines possible extensions to richer circuit families such as ACC⁰