Theoretical analysis uncovers near-n^(5/2) gate lower bounds for depth-two threshold circuits in complexity class E^NP, demonstrating near-optimal boundaries for circuit size.
Building on Chen, Tal, and Wang’s n^(5/2−ε) lower bounds for depth-two threshold circuits, this paper gives an explicit polylogarithmic approach to the exponent-5/2 boundary. It constructs a single language in E^NP that, at every sufficiently large input length N, requires more than Ω(N^(5/2)/log^(9/2) N) gates for SYM∘THR circuits and Ω(N^(5/2)/log^9 N) gates for THR∘THR circuits, even with arbitrary real weights. The proof develops exact column-counting algorithms after logarithmic live-variable restrictions, combining truncated-binomial lifting and threshold noise stability in the symmetric-top case with exact-equation decompositions, small-prime sketches, radix traces, and expander-walk amplification in the threshold-top case. A uniform refuter construction places both bounds on the same language and extends them from an addressed subsequence to every sufficiently large length.
No takes yet. Share an insight, caveat, or question.
Dev Nag (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: