rademacher_sampled_matrix_noncommutative_khintchine_diagonal_gram_bound
Provedkhintchinematrix-completionschatten
Noncommutative (Lust-Picquard / Buchholz) Khintchine inequality specialised to the fixed sampled coordinate Rademacher series (Candes-Recht 2009, Section 6.1, Lemma 6.1, p.24). Let and be the symmetrised sampled matrix with . There is an absolute constant such that, in the window with and ,
Because the Gram sums are diagonal, the two operator-Gram Schatten norms are exactly sampledRowGramSchatten and sampledColumnGramSchatten (the norms of the diagonal entries and ). This is the genuine noncommutative-Khintchine wall: the left-hand side involves the singular values of itself for arbitrary (possibly odd) integer , which is not reducible to scalar Khintchine/Rosenthal. The constant works (Lust-Picquard 1986 / Buchholz 2001).
Preamble
import Definitions.Def_matrix_completion_gram_schatten import Mathlib.Analysis.SpecialFunctions.Pow.Real import Mathlib.Analysis.SpecialFunctions.Log.Basic open MatrixCompletion
Formal statement
theorem rademacher_sampled_matrix_noncommutative_khintchine_diagonal_gram_bound :
∃ C : ℝ, 0 < C ∧
∀ (β : ℝ), 2 < β →
∀ (n₁ n₂ m q : ℕ)
(Omega : Finset (Fin n₁ × Fin n₂))
(X : Matrix (Fin n₁) (Fin n₂) ℝ),
1 ≤ q →
(q : ℝ) ≥ β * Real.log (↑(max n₁ n₂)) →
rademacherExpectation
(fun eps =>
schattenNorm (q : ℝ)
(rademacherSampledMatrix Omega eps
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ))) X) ^ q) ≤
(C * Real.sqrt (q : ℝ) *
max (sampledRowGramSchatten Omega
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ))) X (q : ℝ))
(sampledColumnGramSchatten Omega
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ))) X (q : ℝ))) ^ q := by
sorrySource
Candes, Recht, Exact matrix completion via convex optimization, arXiv:0805.4471, Section 6.1, Lemma 6.1, p.24.