Selected-coefficient Fusion for polynomial circuits #
Project a polynomial onto any finite family of selected monomial coefficients. If the selected monomials are distinct and none is a constant or one of the free input variables, their coefficient vectors form a standard basis. Computing all of them therefore needs one multiplication per output, even with arbitrary field constants, subtraction, and cancellation.
Simultaneously extract the coefficients of a selected exponent family.
Equations
- Algebraic.Fusion.Arithmetic.Interaction.Polynomial.coefficientFeature exponent = LinearMap.pi fun (output : I) => MvPolynomial.lcoeff K (exponent output)
Instances For
Selected nonconstant coefficients vanish on scalar polynomials.
Selected coefficients vanish on a variable when no selected exponent is that variable's degree-one exponent.
A selected monomial maps to the corresponding standard basis vector.
Requested monomials for a selected exponent family.
Equations
- Algebraic.Fusion.Arithmetic.Interaction.Polynomial.targets exponent output = (MvPolynomial.monomial (exponent output)) 1
Instances For
The selected-coefficient features of distinct requested monomials are linearly independent.
Polynomial problem whose free inputs are the designated variables. Its dummy target is unused by the multi-output theorem.
Equations
- Algebraic.Fusion.Arithmetic.Interaction.Polynomial.inputProblem inputVariables = { inputCount := n, inputs := fun (input : Fin n) => MvPolynomial.X (inputVariables input), target := 0 }
Instances For
Matrix of selected coefficients: rows are selected exponents and columns are requested outputs.
Equations
- Algebraic.Fusion.Arithmetic.Interaction.Polynomial.coefficientMatrix exponent outputs selected output = (outputs output).coeff (exponent selected)
Instances For
The dimension of the selected-coefficient span of arbitrary requested polynomials is at most the multiplication cost of a circuit producing them.
The selected coefficient-matrix rank lower-bounds multiplication cost.
Selected coefficient-matrix rank also lower-bounds total nonconstant gate cost.
Selected coefficient-matrix rank lower-bounds raw circuit size.
Computing m distinct selected monomials, none constant or already a free
input, requires at least m multiplication gates.
Total nonconstant arithmetic-gate cost is at least the number of selected monomial outputs.
Raw circuit size is at least the number of selected monomial outputs.