A sharp Blaschke-Santaló inequality for \(\alpha \)-concave functions (Q464247): Difference between revisions

In [Mathematika 51, No. 1--2, 33--48 (2004; Zbl 1121.52021)], \textit{S. Artstein-Avidan} et al. got a functional extension of the classical Blaschke-Santaló inequality: for each log-concave function \(f:{\mathbb R}^n\longrightarrow[0,\infty)\) (integrable with positive integral), there exists a translation \(\widetilde{f}(x)=f(x-a)\) such that \[ \int\widetilde{f}\cdot\int\widetilde{f}^*\leq (2\pi)^n, \] where \(\widetilde{f}^*=e^{-\mathcal{L}(-\log f)}\) and \(\mathcal{L}\) is the Legendre transform; choosing \(a\) to minimize the product on the left, equality holds if and only if \(f\) is (proportional to) the Gaussian distribution. The even case of this result was already proved by \textit{K. Ball} [Isometric problems in \(\ell_p\) and sections of convex sets. Cambridge: University of Cambridge (PhD Thesis) (1987)]. A more general notion than the one of log-concave function is that of \(\alpha\)-concave function: a function \(f:{\mathbb R}^n\longrightarrow[0,\infty)\) is \(\alpha\)-concave (for short, \(f\in\text{C}_{\alpha}({\mathbb R}^n)\)), \(-\infty\leq\alpha\leq\infty\), if \(f\) is supported on some convex set \(\Omega\) and \(f\bigl(\lambda x+(1-\lambda)y\bigr)\geq\bigl[\lambda f(x)^{\alpha}+(1-\lambda)f(y)^{\alpha}\bigr]^{1/\alpha}\) for every \(x,y\in\Omega\) and \(\lambda\in[0,1]\); log-concave functions arise when \(\alpha=0\) (understanding the definition in the limiting sense). In the paper under review, the author studies Blaschke-Santaló inequalities for \(\alpha\)-concave functions when \(\alpha<0\) (the case \(\alpha\geq 0\) was already solved in the above mentioned paper). He shows the reason why a known result by \textit{S. G. Bobkov} [Probab. Theory Relat. Fields 147, No. 1--2, 303--332 (2010; Zbl 1247.46011)], which provides an upper bound for the product \(\int f\cdot\int f^*\) when \(f\in\text{C}_{\alpha}({\mathbb R}^n)\) is integrable with positive integral and \(-1/n<\alpha\leq 0\), is not sharp; and even more, why there is no finite upper bound if \(\alpha\leq -1/n\): the point is that the transform \(*\) is not the proper extension of polarity to be used in the functional Blaschke-Santaló inequality. Thus, in order to get a sharp Blaschke-Santaló inequality, the author defines a new transform on \(\text{C}_{\alpha}({\mathbb R}^n)\), which is called \(\sharp\)-transform: \[ f^{\sharp}(x)=\inf\left\{\frac{1}{f(y)\bigl(1-\alpha\langle x,y\rangle\bigr)^{-1/\alpha}}:y\in{\mathbb R}^n,\,f(y)>0,\,\langle x,y\rangle>\frac{1}{\alpha}\right\}. \] He studies several interesting properties of this new transform (\(f^{\sharp}\geq f^*\), if \(f\) is even then \(f^{\sharp}\) is even as well, it is order reversing, it is a duality transform on its image and is neither injective nor surjective on the full class of \(\alpha\)-concave functions), and proves the following sharp functional Blaschke-Santaló inequality: if \(f\in\text{C}_{\alpha}({\mathbb R}^n)\), \(-1/n<\alpha\leq 0\), is integrable with positive integral, then \[ \int f\cdot\int f^{\sharp}\leq\left(\int H_{\alpha}\right)^2, \] with equality if and only of \(f=H_{\alpha}\) (up to a linear transformation). Here \(H_{\alpha}(x)=\bigl(1-\alpha|x|^2\bigr)^{1/(2\alpha)}\).