The integral of a nonnegative function is at most the relative entropy plus the logarithm of its exponential moment under the reference measure; relative to a diagonal Gaussian this bounds diagonal quadratic moments, and the second moment, linearly by the relative entropy with explicit constants.
In the settings of The Real Numbers: Standing Notation and Background and The Intrinsic Calculus on the Wasserstein Space: Standing Notation, finite relative entropy and are those of that definition on ; Borel and integrable are as in Probability Measures on Euclidean Space and Random Vectors: Standing Notation §measures, a nonnegative Borel function being integrated as a -valued map; is the exponential function and the natural logarithm.
1. (Entropy inequality) Let with of finite relative entropy with respect to , and let be Borel with integrable with respect to . Then is integrable with respect to , is a positive real number, and
2. (Diagonal quadratic moments) Let be a variance vector, with greatest variance , let be the diagonal Gaussian measure with variances , let with for every , and let be positive with for every . Then for every of finite relative entropy with respect to the function is integrable with respect to and
3. (Second moment) Let and be as in clause 2. Every of finite relative entropy with respect to belongs to , and its second moment satisfies
Loading…
No relations recorded yet.