Convex analysis can be helpful for the asymptotic analysis of monotone operators (Q959952): Difference between revisions

The authors use asymptotic representations for (maximal) monotone multimaps in order to avoid the use of Moreau--Yosida approximations. Let \(X\) be a reflexive Banach space and let \(M\) be a maximal monotone multimap from \(X\) to \(X^*\). A function \(f:X\times X^*\to\mathbb{R}\cup\{\infty\}\) is called a representative of \(M\) if \(f\) is convex and lower semicontinuous and if \(f(x,x^*)\geq\langle x,x^*\rangle\) and \(f^*(x,x^*):=\sup\{(\langle x,w^*\rangle+ \langle w,x^*\rangle -f(w,w^*)\mid (w,w^*)\in X\times X^*\}\geq\langle x,x^*\rangle \) and if \(\text{graph}(M)=\{(x,x^*)\in X\times X^*\mid f(x,x^*)=\langle x,x^*\rangle\}\). Denote by \(\sigma\) the function on \(X\) defined by \(\sigma(x):=\sup\{\langle x,x^*\rangle \mid x^*\in\operatorname {Im}M\}\) and denote by \(M_\infty\) the subdifferential of \(\sigma\). Let \(f_\infty^M\) be the asymptotic function of \(M\) (this notion is never explicitly defined in the article, but a good guess would be that it is a representative of \(M_\infty\)). Assume that, whenever \((t_n)\) is a real sequence converging to \(\infty\), \((u_n)\) is a sequence in the unit sphere of \(X\), and \((u_n^*)\) is a bounded sequence in \(X^*\) such that \((t_nu_n,u_n^*)\in\text{graph}(M)\), we have \((u_n)\) has a cluster point. The authors prove that there is an \(x\in X\) such that \(0\in M(x)\) provided that, in addition, \(\{v\in X\mid F_\infty^M(v)\leq0\}\) is a linear subspace of \(X\).