Dual characterizations of three distance functions

MaRDI QIDQ6620710zbMATH OpenFDO

Publication date 17 October 2024

Published in Journal of Convex Analysis (Search for Journal in Brave)

distance function farthest distance function Fenchel conjugate Fenchel subdifferential oriented distance function

Convexity of real functions of several variables, generalizations (26B25) Geometry and structure of normed linear spaces (46B20) Applications of functional analysis in optimization, convex analysis, mathematical programming, economics (46N10) Convex functions and convex programs in convex geometry (52A41)

Review text

Let $(X,\left\Vert \cdot\right\Vert )$ be a real normed space and $(X^{\ast },\left\Vert \cdot\right\Vert _{\ast})$ be its dual space. The distance function to a closed convex set $C\subset X$ is the function $d_{C} :X\rightarrow\mathbb{R}$ defined by $d_{C}(x)=\inf\left\{ \left\Vert x-c\right\Vert :c\in C\right\} ,$ while the oriented distance function associated to $C$ is defined via the formula $\Delta_{C}=d_{C}-d_{X\backslash C}.$ When $C$ is convex and compact, one can also associate to $C$ the farthest distance function $ F_{C}^{\left\Vert \cdot\right\Vert },$ defined by $F_{C}^{\left\Vert \cdot\right\Vert }(x)=\max\left\{ \left\Vert x-c\right\Vert :c\in C\right\} .$ All these three functions are convex and the aim of the present paper is to characterize them in terms of Fenchel conjugates and Fenchel subdifferentials. Given a lower semicontinuous proper convex function $f:X\rightarrow\mathbb{R\cup\{\infty\}}$, we attach to it the function $\eta_{f}:X^{\ast}\rightarrow\mathbb{R\cup\{\infty\}},$ defined by $\eta_{f}(0)=0$ and $\eta_{f}(x^{\ast})=\left\Vert x^{\ast}\right\Vert _{\ast }f^{\ast}\left( x^{\ast}/\left\Vert x^{\ast}\right\Vert \right) $ if $x^{\ast}\neq0;$ here $f^{\ast}$ denotes the Fenchel conjugate of $f$. According to Corollary 2.4, for any function $f:X\rightarrow\mathbb{R}$, the following statements are equivalent: (a)~there exists a nonempty closed convex set $C\subset X$ such that $f=d_{C};$ (b)~$\partial f(x)\cap B^{\ast} \neq\emptyset$ for every $x\in X$ and $\eta_{f}(x^{\ast})=f^{\ast}(x^{\ast})$ for all $x^{\ast}\in B^{\ast}.$ Here $B^{\ast}=\{ x^{\ast}:x^{\ast} \in X^*$, $\left\Vert x^{\ast}\right\Vert _{\ast}\leq1\} .$ In the context of Hilbert spaces, Theorem~3.1 asserts that the functions $\Delta_{C}$ are precisely the functions $f:X\rightarrow\mathbb{R}$ such that $\partial f(x)\cap S^{\ast}\neq\emptyset$ for every $x\in X$ and $\eta_{f}$ is convex and lower semicontinuous. Here $S^{\ast}=\{ x^{\ast}:x^{\ast} \in X^*$, $\left\Vert x^{\ast}\right\Vert _{\ast}=1\} $. A characterization of the functions $F_{C}^{\left\Vert \cdot\right\Vert }$ is provided by Theorem~4.1 in the case where $X$ is the space $\mathbb{R}^{N}$ endowed with a norm $\left\Vert \cdot\right\Vert $ less than or equal to the Euclidean norm: For $f:\mathbb{R}^{N}\rightarrow\mathbb{R}$, the following statements are equivalent: (a)~there exists a nonempty compact convex set $C\subset\mathbb{R}^{N}$ such that $f=F_{C}^{\left\Vert \cdot\right\Vert };$ (b)~$\partial f(x)\cap S^{\ast}\neq\emptyset$ for every $x\in\mathbb{R}^{N}$ and $\eta_{f}$ is finite-valued and concave.

Recommendations

Cites work

Cited in

(1)

A characterization of sets in ${\mathbb R}^2$ with DC distance function

This page was built for publication: Dual characterizations of three distance functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6620710)