The first variation of the total mass of log-concave functions and related inequalities (Q2437426)

The main motivation of the paper relays on the surprising role of log-concave functions in convex geometry. A function \(f:\mathbb R^n\to\mathbb R\) is called log-concave if it is represented in the form \(f(x)=\exp(-u(x))\), where \(u\) is convex (and may take \(+\infty\) as value). The importance of log-concave functions in convex geometry has become clear after the discovery of the Prékopa-Leindler inequality. This result, in a proper setting, reduces to the Brunn-Minkowski inequality, and hence reflects the analogue behavior of log-concave functions and convex bodies. According to this phenomenon, the paper is devoted to the geometry of log-concave functions, proposing functional counterparts of fundamental inequalities of convex geometry. The starting point of the investigations is Minkowki's first inequality and the fact, that its upper part has an elegant integral representation. Roughly speaking, this upper part is a kind of directional derivative of the volume function while the point and the direction are convex bodies. In order to place the classical setup into the context of log-concave functions, a natural and elegant idea is followed. First, the volume of convex bodies is replaced by the total mass of log-concave functions. Second, two internal operations (a sum and a multiplication by positive scalars) are introduced that endow the class of log-concave functions with a suitable algebraic structure. These operations are given in terms of Fenchel conjugation and make it possible to define a suitable directional derivative. It turns out that the directional derivative of the total mass of log-concave functions exists under a mild regularity assumption. In some special cases, the directional derivative is computed explicitly. One of these cases makes it possible to define the entropy of a log-concave function. Moreover, under some restrictions (for example, when the domain is a smooth strictly convex body and the functions have further properties), integral representations can also be presented for the directional derivative of the generalized total mass functional. Using the representations, several classical inequalities are extended to this setting. The authors derive the functional form of the isoperimetric inequality, and a family of logarithmic-type Sobolev inequalities with respect to log-concave probability measures. Finally, a functional version of the classical Minkowski's problem is proposed and some partial results are proved toward its solution. Itself the approach requires a very careful analysis and advanced tools. The results obtained generalize several recent developments, replacing them into an other perspective. The introduction of the paper, besides the excellent motivation, makes easy-to-follow the forthcomings, even for those who are not experts in the field.