Extremal properties of central half-spaces for product measures (Q5937217)

The author deals with the isoperimetric and the shift problem for subsets of measure \({1\over 2}\) in product spaces. The paper is divided into two technically independent parts. Both contain results on the extremality of canonical half-spaces for product measures. The proofs follow the same pattern: a sharp comparison with the Gaussian measure is established for a measure \(\mu\) on a Riemannian manifold \(M\), and this comparison is then extended to the product measure \(\mu^{\otimes n}\) on \(M^n\), \(n\geq 1\), which finally shows the extremality of subsets of the form \(A\times M^{n-1}\). In the first part isoperimetric and shift functionals are compared; here the tensorization devices allowing to go to product measures, are Bobkov-type functional forms of the geometric inequalities. Applications to several probability spaces are given: even log-concave probabilities on the real line, uniform probabilities on Euclidean spheres and on Euclidean balls. In the second part, a method of \textit{J. D. Vaaler} [Pac. J. Math. 83, 543-553 (1979; Zbl 0465.52011)] is used. This time one compares the values of measures on symmetric convex sets, while the tensorizing device is a result of \textit{M. Kanter} [Trans. Am. Math. Soc. 229, 65-85 (1977; Zbl 0379.60015)] about the peaked order on unimodal measures. This method is applied to unimodal probabilities on the real line. This second part is completed by an extension of a theorem of \textit{M. Meyer} and \textit{A. Pajor} [J. Funct. Anal. 80, No. 1, 109-123 (1988; Zbl 0667.46004)] about the volume of section on \(l_p\)-sums of finite-dimensional normed spaces. The two approaches give quite similar results. Nevertheless, they are efficient in very different settings. The first one is convenient for the general isoperimetric problem on manifolds. The second requires ond a linear setting but it can be applied to non log-concave measure, where the first one would fail.