Floating functions (Q2421892)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Floating functions
scientific article

    Statements

    Floating functions (English)
    0 references
    0 references
    0 references
    0 references
    18 June 2019
    0 references
    Let $K$ be a convex body in $\mathbb{R}^n$, which here means that $K$ is compact, convex and has non-empty interior. For $0<\delta< \operatorname{vol}_n(K)$, the \textit{(convex) floating body} $K_\delta$ of $K$ is defined as the intersection of all halfspaces $H^+$ whose defining hyperplanes $H$ cut off a cap of volume $\delta$ from $K$. That is \[K_\delta = \bigcap_{\{H: \operatorname{vol}_n(H^-\cap K) \leq \delta\}} H^+.\] If one takes the right-derivative of the volume of the floating body as a function in $\delta$, then one obtains (up to a constant) the affine surface area of $K$, that is \[ \lim_{\delta \to 0^+} \frac{\operatorname{vol}_n(K)-\operatorname{vol}_n(K_\delta)}{\delta^{\frac{2}{n+1}}}=c_n \operatorname{as}(K) \] where $c_n$ is a constant that only depends on the dimension $n$ and $\operatorname{as}(K)$ denotes the \textit{affine surface area of $K$}. If $\mu_K$ is the usual surface area measure of $\partial K$ and $\kappa_K(z)$ denotes the (generalized) Gauss curvature of $K$ at $z$, then \[ \operatorname{as}(K)=\int_{\partial K} (\kappa_K(z))^{\frac{1}{n+1}}\;\mathrm{d} \mu_K(z). \] In particular, the affine surface area vanishes for polytopes. The relation between floating bodies and affine surface area was first discovered by \textit{W. Blaschke} [Vorlesung über Differentialgeometrie II: Affine Differentialgeometrie. Berlin: Springer (1923; Zbl 02599108)] for three-dimensional convex bodies with analytic boundary and was later established for all convex bodies in general dimension by the second and third author [Math. Scand. 66, 275--290 (1990; Zbl 0739.52008)]. In the present paper, floating functions for convex and log-concave functions are introduced and studied. Let $\psi:\mathbb{R}^n\to\mathbb{R}$ denote a convex function such that $0<\int_{\mathbb{R}^n} e^{-\psi(x)} \;\mathrm{d} x < \infty$ or equivalently $\lim_{|x|\to\infty} \psi(x)=\infty$. The epigraph of such a function $\psi$, \[ \operatorname{epi}(\psi)=\{(x,y)\in \mathbb{R}^n\times \mathbb{R} : \psi(x) \leq y\}, \] is a closed convex set in $\mathbb{R}^{n+1}$. For sufficiently small $\delta>0$, the authors now define the floating set $(\operatorname{epi}(\psi))_{\delta}$ as \[ (\operatorname{epi}(\psi))_{\delta}=\bigcap_{\{H:\operatorname{vol}_{n+1}(H^-\cap \operatorname{epi}(\psi))\leq \delta \}} H^+. \] The \textit{floating function} $\psi_\delta$ of $\psi$ is now defined as the unique convex function $\psi_\delta:\mathbb{R}^n\to\mathbb{R}$ with the property that $(\operatorname{epi}(\psi))_{\delta}=\operatorname{epi}(\psi_\delta)$. Furthermore, if $f(x)=e^{-\psi(x)}$ is a log-concave function, where $\psi$ is as above, then $f_\delta(x)=e^{-\psi_\delta(x)}$ is called the floating function of $f$. In their main result, the authors show that \[ \lim_{\delta\to 0^+} \frac{\int_{\mathbb{R}^n}\left(e^{-\psi(x)}-e^{-\psi_{\delta}(x)}\right)\; \mathrm{d} x }{\delta^{\frac{2}{n+2}}}=c_{n+1} \int_{\mathbb{R}^n} (\det(\nabla^2 \psi(x)))^{\frac{1}{n+2}} e^{-\psi(x)} \; \mathrm{d} x, \] where $c_{n+1}$ is as above. Comparing this with the classical results for convex bodies, the \textit{affine surface area} of a log-concave function $f=e^{-\psi}$ is defined as \[ \operatorname{as}(f)=\int_{\mathbb{R}^n} (\det(\nabla^2 \psi(x)))^{\frac{1}{n+2}} e^{-\psi(x)}\; \mathrm{d} x. \] Moreover, the authors show that the affine surface area of a log-concave function has similar properties as the affine surface area of a convex body. In particular, affine invariance properties and the valuation property as well as the fact that $\operatorname{as}(f)$ vanishes if $\psi$ is piecewise affine.
    0 references
    convex functions
    0 references
    log-concave functions
    0 references
    floating bodies
    0 references
    affine surface area
    0 references
    floating functions
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references