Coverage fusion for finite monomial supports #
This is the support-theoretic counterpart of the rank certificate. Semantic values are finite sets of monomials, addition is union, and a charged term contributes a prescribed finite support. A witness is one monomial in the target support. Union preserves non-membership, while a term fails precisely on the target monomials it covers.
If every allowed term covers at most r target monomials, every circuit whose
union is the target uses at least ceil(target.card / r) charged terms. The
local coverage estimate can come from separated monomials, rectangle bounds,
or other monotone-support arguments.
Compatibility name for the reusable finite-support carrier.
Instances For
Construct a target support from charged terms, with no free inputs.
Equations
Instances For
Fusion model whose observations say that a target monomial remains uncovered.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Algebraic.Fusion.SumOfTerms.Coverage.witnessFintype target termSupport = id inferInstance
Union preserves non-membership of every target monomial.
A term preserves a witness exactly when it does not cover that monomial.
Target witnesses covered by one dictionary term.
Equations
- Algebraic.Fusion.SumOfTerms.Coverage.coveredWitnesses target termSupport term = {witness ∈ target.attach | ↑witness ∈ (termSupport term).monomials}
Instances For
A local bound on how many target monomials one term can cover.
- capacity : ℕ
Maximum target coverage of one term.
The combinatorial coverage estimate for every allowed term.
Instances For
Coverage bounds compile to the generic finite-witness failure bound.
Equations
- bound.failureBound = { capacity := bound.capacity, failure_card_le := ⋯ }
Instances For
A positive coverage capacity gives the expected circuit lower bound.
The separated case: each term covers at most one target monomial.