The First Variation of the Entropy Along a Gradient Perturbation of the Identity
lemmaProbabilitylem:entropy-first-variation-euclidean-2026aPushing a measure of finite entropy forward by the identity plus a small multiple of the gradient of a test function keeps the entropy finite, changes it by minus the mean of log det of the perturbed Hessian, and the derivative at zero is minus the mean of the Laplacian.
In the setting of Wasserstein Spaces, Random Vectors, Vector Fields and Symmetric Matrices in Every Dimension: Standing Notation, with finite entropy, the entropy and the set as in that definition, and with the test functions , their gradient maps and Laplacians as fixed there. For a test function , are its second partial derivatives and its Hessian matrix at , as in Differential Calculus and Convexity on Euclidean Open Sets: Standing Notation §derivatives; for , is the map . The determinant is written , the natural logarithm , and is the identity matrix; and are the constants of Determinants of Positive Definite Matrices: Positivity, the Bound , Bounds under Pinching, and the Expansion of §expansion, and the natural number is read in where a real number is required. Differentiability at of a real function on an open interval is that of Derivative at an Interior Point. Let , let , and let be a nonnegative real number with for all and .
1. (Push-forward)¶ Let satisfy . Then for every , the map is Borel, the function is Borel and bounded, , and
2. (Expansion)¶ Let satisfy and . Then
3. (First variation)¶ There is a positive real number such that for every , and the function
is differentiable at with derivative .
Loading…
Prerequisites
No prerequisites tracked.
Dependents
No dependents yet.
Dependent proofs
No dependent proofs yet.
No relations recorded yet.