v4.31.0
v4.31.0
This release upgrades SLT to Lean and Mathlib v4.31.0 and upstreams the local matrix concentration development.
Highlights
- Pins
lean-toolchain,lakefile.lean, andlake-manifest.jsonto Lean 4.31.0 and Mathlib v4.31.0. - Adds
SLT.MatrixInfra, covering singular values, SVD and spectral decompositions, Courant-Fischer min-max principles, Eckart-Young-Mirsky, Weyl perturbation, Davis-Kahan, Loewner/matrix calculus, and deterministic Lieb concavity for HDP Theorem 5.4.8. - Adds
SLT.HansonWright, deriving Hanson-Wright MGF certificates from independent sub-Gaussian or bounded coordinates and proving HDP-style Hanson-Wright tail bounds. - Adds
SLT.RMTrandom-matrix results: rectangular and symmetric sub-Gaussian operator-norm bounds, expectation forms, lower bounds, two-sided singular-value bounds for isotropic sub-Gaussian rows, the random-matrix Lieb step, and the finite matrix Bernstein inequality. - Adds
SLT.TDudley, a truncated Dudley entropy bound using subset nets, local/global oscillation controls, and the existing metric-entropy API. - Extends
SLT.SubGaussianwith scalarIsSubGaussian, ψ₂-scale infrastructure, Bernstein-style CGF tail bounds, finite-supremum integrability lemmas, and compatibility with MathlibHasSubgaussianMGF. - Updates existing Gaussian, Poincare, Dudley, and least-squares files for Lean 4.31 and Mathlib API changes while preserving the prior development.
Verification
LEAN_NUM_THREADS=$(nproc) lake buildcompleted successfully.- Scanned
SLT/forsorry,admit,axiom, andnative_decide; none were found.