AC0 layer switching with separate source and target bounds #
The first restriction in the standard parity lower-bound argument starts from
the width-one literal layer but aims for decision-tree depth t. Later rounds
start and end at depth t. This module therefore separates the incoming
normal-form width sourceBound from the desired outgoing decision-tree depth
targetBound.
For sourceBound <= targetBound, one logical layer advances except with
probability at most
andOrCost(program) * (5 * p * sourceBound)^(targetBound + 1).
Keeping these parameters distinct is quantitatively essential: charging the
first round as though its source width were already t would introduce an
artificial extra factor of t and lose the standard depth exponent. The proof
uses the same exact semantic switching and finite union bounds as the
equal-bound theorem; it performs no circuit search or finite experiment.
Advance the shallow invariant when old layers have a possibly smaller bound than the newly exposed layer. Arbitrary internal NOT gates are handled by the raw one-bound successor theorem.
Compatibility wrapper for the checked input-negation presentation.
A gate represented by source-width normal form fails the target tree-depth bound with the two-parameter switching-lemma estimate.
The next-layer event for one charged gate satisfies the two-parameter bound; gates outside that layer contribute the empty event.
Union bound over the connective gates in the next layer, retaining separate source-width and target-depth parameters.
One random restriction advances the semantic invariant from source bound
sourceBound to target bound targetBound, except with the standard charged
two-parameter switching probability. This raw form permits arbitrary internal
NOT gates.
Compatibility wrapper for the checked input-negation presentation.