rudelson_selection_symmetrized_tensor_khintchine_dense
ProvedSymmetrized tensor non-commutative Khintchine bound (dense regime) = Steps 1 and 2 of the proof of Theorem 1 in Rudelson 1999 (J. Funct. Anal. 164). There is a constant such that, under the dense density hypothesis and the coordinate-radius bound , the expected tangent-sampling deviation satisfies with , . The proof is symmetrization (ghost copy) followed by the matrix Lust-Picquard / non-commutative Khintchine inequality applied to the rank-one tensor operators , , using the spectral identity so that .
Preamble
import Definitions.Def_matrix_completion_tangent open MatrixCompletion open scoped Classical BigOperators
Formal statement
theorem rudelson_selection_symmetrized_tensor_khintchine_dense :
∃ Csym : ℝ, 0 < Csym ∧
∀ (β : ℝ), 2 < β →
∀ (n₁ n₂ r m : ℕ) (M : Matrix (Fin n₁) (Fin n₂) ℝ)
(S : SVD M r) (R : ℝ),
0 < n₁ → 0 < n₂ → 0 < r → m ≤ n₁ * n₂ →
0 ≤ R →
(m : ℝ) ≥ β * (↑(max n₁ n₂)) * (r : ℝ) * Real.log (↑(max n₁ n₂)) →
(∀ i : Fin n₁, ∀ j : Fin n₂,
frobeniusNorm (tangentProjection S (coordinateMatrix i j)) ≤ R) →
bernoulliExpectation ((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))
(fun Omega =>
tangentSamplingDeviation Omega S
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))) ≤
Csym *
(Real.sqrt
(Real.log (↑(max n₁ n₂)) /
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))) * R)
* Real.sqrt
(bernoulliExpectation ((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))
(fun Omega =>
tangentSamplingDeviation Omega S
((m : ℝ) / ((n₁ : ℝ) * (n₂ : ℝ)))) + 1) := by sorrySource
Rudelson, J. Funct. Anal. 164 (1999) 60-72, Theorem 1 proof Steps 1-2 (symmetrization p.3 + Lust-Picquard tensor lemma p.4); Lust-Picquard/Pisier non-commutative Khintchine; van Handel/Tropp matrix concentration. Candes-Recht 2009 arXiv:0805.4471 Section 4.2 eq(4.9) p.18.