Positive Semidefinite Hessian Matrix
ProvedFunctionalInequalities.Analysis.BakryEmery.hessianMatrixIn_posSemidef_readablefunctional-inequalitieskls-conjecturekls-modularkls-readablelog-concave
This theorem converts pointwise nonnegativity of a Hessian bilinear form into positive semidefiniteness of its matrix in an orthonormal basis.
Let be a real inner-product space, let be an orthonormal basis indexed by , and let . At , assume that is twice continuously differentiable and that
Then the matrix hessianMatrixIn φ b x is positive semidefinite:
This is a pointwise linear-algebraic bridge between the Hessian form and its coordinate matrix.
Original Lean declaration header
Within namespace FunctionalInequalities.Analysis.BakryEmery. Shown in its original module context. Proof omitted.
theorem hessianMatrixIn_posSemidef {m : ℕ} (b : OrthonormalBasis (Fin m) ℝ E)
{φ : E → ℝ} {x : E} (hφ : ContDiffAt ℝ 2 φ x)
(hH : ∀ v : E, 0 ≤ bilinearIteratedFDerivTwo ℝ φ x v v) :
(hessianMatrixIn φ b x).PosSemidef
Module navigation
Module: Analysis/BakryEmery/MongeAmpere.
Source excerpt, lines 219–222 · Frozen source 408915f.
Preamble
import Definitions.Def_KLSModular_FunctionalInequalities_Analysis_BakryEmery_MongeAmpere
variable {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [FiniteDimensional ℝ E]
omit [FiniteDimensional ℝ E]
Formal statement
theorem FunctionalInequalities.Analysis.BakryEmery.hessianMatrixIn_posSemidef_readable {m : ℕ} (b : OrthonormalBasis (Fin m) ℝ E)
{φ : E → ℝ} {x : E} (hφ : ContDiffAt ℝ 2 φ x)
(hH : ∀ v : E, 0 ≤ bilinearIteratedFDerivTwo ℝ φ x v v) :
(hessianMatrixIn φ b x).PosSemidef := by sorry
Source
Lean source: https://github.com/human-ai-mathematics/kls-lean/blob/408915f/formalization/FunctionalInequalities/Analysis/BakryEmery/MongeAmpere.lean#L217-L241 (repository human-ai-mathematics/kls-lean, commit 408915f, module `FunctionalInequalities.Analysis.BakryEmery.MongeAmpere`, declaration `FunctionalInequalities.Analysis.BakryEmery.hessianMatrixIn_posSemidef`; the repository is private at upload time).