Expected squared Frobenius norm of a scaled Gaussian matrix:
ProvedGaussianMatrix.frobenius_second_momentFix matrices and and let be a standard Gaussian matrix. Then
This second-moment identity is the first of the Gaussian moment bounds (Halko–Martinsson–Tropp Proposition 10.1, eq. (10.1); Tropp–Webber Lemma B.1). It is the identity that produces the factor in the expected Frobenius error of the randomized SVD, and with , it says .
Formalization Note. The expectation is a Lebesgue integral against ; the integrand is a polynomial in the entries of and hence integrable.
import Definitions.Def_GaussianMatrix_basic open MeasureTheory ProbabilityTheory open scoped Matrix
namespace GaussianMatrix
theorem frobenius_second_moment {a p m n : ℕ} (S : Matrix (Fin a) (Fin p) ℝ)
(T : Matrix (Fin m) (Fin n) ℝ) :
∫ G, frobSq (S * Matrix.of G * T) ∂(gaussianMatrix p m) = frobSq S * frobSq T := by sorry
end GaussianMatrixRead-back
What the Lean code literally says, in plain math · claude-fable-5-1
Let be natural numbers (implicit; any may be ), and let and be arbitrary real matrices (no hypotheses on them). Let be the law of a array with independent entries. Writing for the squared Frobenius norm of a real matrix (the empty sum is ), the statement asserts
where is the matrix product (with viewed as a matrix). The left-hand side is Lean's Bochner integral of the real-valued function against ; by Lean's convention this integral is defined to be if the integrand is not integrable, and the statement does not separately assert integrability (the integrand is a polynomial of degree in the Gaussian entries). Edge cases: if or then is an empty matrix, the integrand is identically , and or is ; if or the array space is a single point (the integral is the value at that point), , and or respectively.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.