Proposition C.1 — Gaussian Gram matrix moment
OpenSketchAttr.LOO.Supplement.proposition_C_1gaussian-matricesmomentsp2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let and be integers, and let have independent standard real Gaussian entries. The squared Frobenius norm of its Gram matrix is integrable and
The exact moment controls the normalization of the hard matrix distribution used in the diagonal-estimation lower bound. The empty-row case has zero Gram matrix and both sides equal zero.
Differences from the source:
- encoding: the finite expectation of the squared Frobenius norm → integrability is asserted together with the integral equality, using the published tensor Frobenius norm; reason: the explicit integrability clause prevents Lean from assigning a default integral value
- encoding: an r by n real Gaussian matrix with positive ambient dimension → Fin r by Fin n matrices, n at least one, with the product of standard real Gaussian laws; reason: finite coordinates and the explicit law express the same independent entries
Preamble
import Mathlib import Definitions.Def_SketchAttr_LOO_Supplement_Quadratic open MeasureTheory
Formal statement
namespace SketchAttr.LOO.Supplement
theorem proposition_C_1 (r n : ℕ) (hn : 1 ≤ n) :
Integrable (fun G : Matrix (Fin r) (Fin n) ℝ => matrixFrobNorm (gram G) ^ 2)
(realGaussianMatrix r n) ∧
(∫ G : Matrix (Fin r) (Fin n) ℝ, matrixFrobNorm (gram G) ^ 2 ∂realGaussianMatrix r n) =
(n : ℝ) * r * (r + n + 1) := by sorry
end SketchAttr.LOO.Supplement
Source
Chris Liaw, Aranyak Mehta, Grigoris Velegkas, Felix Zhou, Sketching Language Models for Context Attribution, STOC 2027 submission #9 (October 2026), PDF sha256 86af2569e45c55b1009e7453d4f53047b5820d44868a7f528bdb6fed1a2ef0e2, p. 29, Proposition C.1