Multivariate degree bounds for polynomial coefficients #
In Polynomial (MvPolynomial ι R), the outer variable is a Taylor variable z, and the
coefficient variables are formal Xᵢ. If coefficient j of every substituted input has
multivariate total degree at most j, the same is true after polynomial evaluation.
This is the algebraic closure behind equation (3.9), p24 of OpenAI, The Exponential Time Hypothesis is False. The coefficient ring can be any commutative ring. No Taylor reconstruction, truncation or quotient identification is used.
Main results #
MvPolynomial.totalDegree_coeff_eval₂_le: polynomial substitution preserves the coefficient degree bound.MvPolynomial.totalDegree_X_le: every formal variableX ihas total degree at most1.MvPolynomial.totalDegree_coeff_X_mul_map_le: a scalar polynomial with zero constant term can multiply a formal variable while preserving the bound.
If every input's coefficient j has total degree at most j in the formal variables,
then every coefficient of the substituted polynomial satisfies the same bound. Constants of
R are embedded as constants in both polynomial layers.
Every formal variable X i has total degree at most 1, without requiring Nontrivial R.
Every formal-variable occurrence in this correction carries a positive power of the outer variable: the scalar polynomial's constant coefficient must be zero. Higher outer powers are allowed, and no nontrivial-ring hypothesis is needed.