Gamma-ratio bound
ProvedGaussianMatrix.tw_gamma_ratio_boundam-gmdeterministicgamma-functiongautschi-inequalitylog-convexityspecial-functions
Let be integers. Then
where is Euler's Gamma function.
Equivalently, . This is the deterministic step from Tropp–Webber's (B.5) to (B.6): it replaces the density constant of Edelman's bound by an elementary one. The proof writes , bounds each factor by using Wendel–Gautschi (a consequence of log-convexity of ), and then uses AM–GM in the paired form to get .
Formalization Note. Real.Gamma; the powers are real powers with positive bases. Both sides are positive under the hypotheses.
Preamble
import Definitions.Def_GaussianMatrix_basic open MeasureTheory ProbabilityTheory open scoped Matrix
Formal statement
namespace GaussianMatrix
theorem tw_gamma_ratio_bound {r k : ℕ} (hr : 1 ≤ r) (hrk : r ≤ k) :
2 ^ (((k : ℝ) - r - 1) / 2) * Real.Gamma (((k : ℝ) + 1) / 2)
/ (Real.Gamma ((r : ℝ) / 2) * Real.Gamma ((k : ℝ) - r + 1))
≤ (((k : ℝ) + r) / 2) ^ (((k : ℝ) - r + 1) / 2) / (2 * Real.Gamma ((k : ℝ) - r + 1)) := by
sorry
end GaussianMatrixSource
J. A. Tropp and R. J. Webber, arXiv:2306.12418, Appendix B, proof of Lemma B.3, the display between (B.5) and (B.6). It uses Γ(z+1/2) ≤ z^{1/2} Γ(z) (J. G. Wendel, Note on the gamma function, Amer. Math. Monthly 55 (1948) 563–564; TW ref. [59]) and (r+i)(k−i) ≤ ((k+r)/2)².