sampled_row_gram_schatten_antitone_exp
Provedℓ_q-Gram-vector antitone (row). For a sampling probability 0 ≤ p, real exponents 1 ≤ q ≤ q', and any real matrix X, the per-row sampled Gram-Schatten ℓ_q quantity decreases in the exponent: sampledRowGramSchatten Ω p X q' ≤ sampledRowGramSchatten Ω p X q. Here sampledRowGramSchatten Ω p X q = (∑_i (p⁻¹ √(∑_{(i,j)∈Ω} X_{ij}²))^q)^{1/q} is the ℓ_q norm of the per-row vector p⁻¹ √(rowEnergy_i). This is the power-mean / ℓ_p-monotonicity inequality (Hardy–Littlewood–Pólya, Inequalities §2.10) applied to that nonnegative vector — the Gram-side analogue of the Schatten antitone schatten_norm_antitone_exp. It is the (d) analytic step of the Candès–Recht 2009 §6.1 matrix-Khintchine general-q reduction: the RHS Gram term rowGS(2n) for the even integer 2n ∈ [q, q+2) is dominated by rowGS(q). Proof: normalize by A = (∑σ^q)^{1/q}; the normalized entries t_i = σ_i/A ≤ 1 give t_i^{q'} ≤ t_i^q, hence ∑ t_i^{q'} ≤ 1, which rearranges to the claim (A = 0 handled separately). The vector σ_i = p⁻¹ √(rowEnergy_i) is nonnegative since 0 ≤ p.
import Definitions.Def_matrix_completion_gram_schatten open MatrixCompletion
theorem sampled_row_gram_schatten_antitone_exp {n1 n2 : Nat} (Omega : Finset (Fin n1 × Fin n2)) (p : Real) (hp : 0 ≤ p) (X : MatrixCompletion.RealMatrix n1 n2) (q q' : Real) (hq : 1 ≤ q) (hqq' : q ≤ q') : MatrixCompletion.sampledRowGramSchatten Omega p X q' ≤ MatrixCompletion.sampledRowGramSchatten Omega p X q := by sorry