The (4 - ε) n bound for nondeterministic circuits #
A nondeterministic circuit for f : {0,1}ⁿ → {0,1} is a circuit on n + m
inputs whose accepted inputs x are exactly those with some witness y for
which the output is 1. Witness inputs are read by the wiring graph but carry
no port of the function computed, so the cut-counting lemma applies to the
forgotten network (Network.forget) with the same multigraph. The excess of
the wiring graph is the number of reachable gates minus the number of
reachable inputs, ordinary and witness alike, which is at most the number of
gates minus the number of ordinary inputs read. Hence the same bound holds:
nondeterminism does not reduce the size below (4 - ε) n for any
rectangle-free family with the hypotheses of eventually_lt_size.
The proof organization, the threshold-edge charging, the extractor application, and the graph restoration argument of the deterministic bound follow Ryan Williams's private working note (September 2026); this module only replaces the wiring network by its forgotten version.
The Boolean function computed nondeterministically by a circuit on
n + m inputs: x is accepted when some witness y makes the output 1.
Equations
- Algebraic.Cutwidth.nondetFunction circuit x = decide (∃ (y : Fin m → Bool), circuit.eval Algebraic.Binary.interpretation (Fin.append x y) 0 = true)
Instances For
A circuit on n + m inputs computes f nondeterministically when f x = 1
exactly when some witness makes the output 1.
Equations
- Algebraic.Cutwidth.NondetComputes circuit f = ∀ (x : Cslib.BitString n), f x = true ↔ ∃ (y : Fin m → Bool), circuit.eval Algebraic.Binary.interpretation (Fin.append x y) 0 = true
Instances For
The function computed nondeterministically by a program with m witness
inputs and output gate out.
Equations
- Algebraic.Cutwidth.nondetGateFunction p out x = decide (∃ (y : Fin m → Bool), p.eval Algebraic.Binary.interpretation (Fin.append x y) out = true)
Instances For
The forgotten wiring network computes the nondeterministic gate function.
The nondeterministic version of the circuit-level bound: the ordinary
inputs read are those of the forgotten network, and the excess of the wiring
graph is at most s minus their number.
The fixed-n core for nondeterministic circuits. The hypotheses are
those of lt_size_of_bounds with the number of witness inputs m at most
n, so that the vertex count stays below 16 n.
Nondeterministic circuits. Under the graph-ordering hypothesis and the
hypotheses of eventually_lt_size on the family f n, for every ε > 0 and
all sufficiently large n, every circuit on n + m inputs with m ≤ n that
computes f n nondeterministically has more than (4 - ε) n gates.
Nondeterministic circuits from the pathwidth hypothesis.