Proofs for balanced padding #
Fixing the fresh coordinate loses at most a factor two in source support. The resulting padding has exact balance and preserves polynomial-time evaluation.
Tail strings in a fixed first-bit fibre of a source.
Equations
- Algebraic.Cutwidth.Extractor.Internal.tailFiber P b = Finset.image Fin.tail ({x ∈ P | x 0 = b})
Instances For
Error strictly below one half guarantees both output values on every large flat-source pair.
A two-sided sumset disperser remains one after balanced padding, with a factor-two loss in support threshold.
Dispersion excludes one-rectangles, without any accepting-density assumption.
Any extraction error strictly below one half yields a rectangle-free balanced extension.
Exactly one fresh bit accepts for each old input, so balanced padding accepts exactly half its input cube.
The list evaluator agrees with the finite Boolean-function padding on every positive input length.
Balanced padding preserves a uniform polynomial-time evaluator.