The frontier lower bound with gates of unbounded fan-in #
Main theorem (eventually_lt_mul_card_of_ledger). Fix r ≥ 2 and assume the layout
hypothesis LayoutBound (r + 1) A. Let S_n ⊆ U^n be dense K_n-rectangle-free sets with
log K_n = o(n), as in Cslib.Circuits.Frontier.eventually_lt_mul_innerSize, and let
β(n) = o(n). For every ε > 0 and all large n, every circuit deciding S_n whose gates have
fan-in at most r, except for special gates of any fan-in that have a ledger in a monoid M
with log |M| ≤ β(n), has s ordinary inner gates (inner gates that are not special) with
(r - 1) s > (1 + 1/A - ε) n.
Constants and special gates are free; the ledger contributes only to the sublinear error. So the
coefficient of the frontier method survives a sublinear amount of unbounded fan-in. For instance,
k special gates that aggregate in a fixed finite commutative monoid T, such as unbounded AND,
OR, parity, or counts modulo a fixed m, have a ledger with log |M| = k log |T|
(Cslib.Circuits.Frontier.Ledger.ofAggregates), so k = o(n) such gates are allowed
(eventually_lt_mul_card_of_aggregates). With the layout hypothesis of coefficient one
(Cslib.Circuits.Frontier.layoutBound_one) the bounds are unconditional:
(r - 1) s > (2 - ε) n (the module
Complexity.Computability.Circuit.Frontier.Ledger.Unconditional).
Proof. Erase the special gates and lay out the live network (the graph of the erased program
with every live wire as an output), one connected component after another
(Cslib.Circuits.Frontier.LayoutBound.exists_layout_componentwise). Sweep the program along this
layout with the ledger in the messages (Cslib.Circuits.Frontier.Ledger.sweep). At each step the
transitions are bounded both by the width of the step and by the number of inputs read in the
component being processed. The frontier counting lemma then finds a component W that reads
almost all n inputs and whose frontier is wide, up to 2 log |M| and the usual error terms
(Cslib.Circuits.Frontier.Ledger.exists_ncard_le). The cycle rank of W is at most
(r - 1) s + 1 minus the number of inputs it reads
(Cslib.Circuits.Frontier.Compiler.cycleRankOn_add_ncard_readIn_le), so W alone has the
demand and the supply of a connected graph, and the comparison of demand and supply
(Cslib.Circuits.Frontier.eventually_lt_of_frontier, with the ledger in its sublinear slack)
finishes the proof.
Main results #
Cslib.Circuits.Frontier.eventually_lt_mul_card_of_ledger: the frontier lower bound with a ledger, conditional onLayoutBound.Cslib.Circuits.Frontier.eventually_lt_mul_card_of_aggregates: the frontier lower bound with aggregate gates, conditional onLayoutBound.
The frontier lower bound with a ledger. Let r ≥ 2 and assume the layout hypothesis for
maximum degree r + 1 with coefficient A > 0. Let S_n ⊆ U^n be K_n-rectangle-free sets over
a finite alphabet U with at least two symbols, with log K_n = o(n) and
n log |U| - log |S_n| = o(n), and let β(n) = o(n). Then for every ε > 0 and all large n,
every circuit deciding S_n, over any basis and interpretation on U, whose gates have fan-in at
most r except for special gates of any fan-in with a ledger in a finite commutative monoid M
with log |M| ≤ β(n), has s ordinary inner gates, the inner gates that are not special, with
(r - 1) s > (1 + 1/A - ε) n.
The frontier lower bound with aggregate gates. Under the hypotheses of
eventually_lt_mul_card_of_ledger, for every ε > 0 and all large n, every circuit deciding
S_n, over any basis and interpretation on U, whose gates have fan-in at most r except for
k special gates of any fan-in that aggregate in a finite commutative monoid T, with
k log |T| ≤ β(n), has s ordinary inner gates with (r - 1) s > (1 + 1/A - ε) n. For instance,
the special gates may compute unbounded AND, OR, parity, or counts modulo a fixed m.