In this paper, the Orlicz addition of measures is proposed and an interpretation of the f -divergence is provided based on a linear Orlicz addition of two measures. Fundamental inequalities, such as, a dual functional Orlicz-Brunn-Minkowski inequality, are established. We also investigate an optimization problem for the f -divergence and establish functional affine isoperimetric inequalities for the dual functional Orlicz affine and geominimal surface areas of measures.
Index Termsaffine isoperimetric inequality, affine surface area, Brunn-Minkowski theory, dual Brunn-Minkowski theory, the f -divergence, geominimal surface area, optimization problem for the f -divergence.with equality if and only if p = q almost everywhere with respect to Q. When f is strictly concave with f (1) = 0, one gets similar results with "≥" replaced by "≤". From this viewpoint, the f -divergence can be used to distinguish two measures. Without doubt, the f -divergence plays fundamental roles in, such as, image analysis, information September 11, 2018 DRAFT 2 theory, pattern matching and statistical learning (see [5], [11], [24], [27], [34]), where the measure of difference between measures is required. Moreover, in general, the f -divergence is arguably better than the L p distance. Recent development in convex geometry has witnessed the strong connections between the f -divergence and convex geometry. For instance, it has been proved that the L p affine surface area [7], [32], [37], a central notion in convex geometry, is related to the Renyi entropy [38]; while the general affine surface area [28], [30] is associated to the f -divergence [9]. Note that these affine surface areas are valuations; and valuations are the key ingredients for the Dehn's solution of Hilbert third problem. Moreover, under certain conditions (such as, semicontinuity), it has been proved that these affine surface areas can be used to uniquely characterize all valuations which remain unchanged under linear transforms with determinant ±1 (see e.g. [23], [29], [30]). On the other hand, as showed in the Subsection V-A, the affine and geominimal surface areas (see e.g. [32], [35], [40], [41]) can be translated to an optimization problem for the f -divergence. This observation leads us to investigate the dual functional Orlicz affine and geominimal surface areas for measures, which are invariant under linear transforms with determinant ±1. The Brunn-Minkowsi inequality is arguably one of the most important inequalities in convex geometry. It can be used to prove, for instance, the celebrated Minkowski's and isoperimetric inequalities. (Note that the isoperimetric problem has a history over 1000 years). See the excellent survey [18] for more details. On the other hand, the dual Brunn-Minkowski inequality and dual Minkowski inequality are crucial for the solutions of the famous Busemann-Petty problem (see e.g., [17], [22], [31], [43]). The Brunn-Minkowsi inequality and its dual have been extended to the Orlicz theory in [19], [20], [39], [44].This paper is dedicated to...