The sum and chain rules for maximal monotone operators (Q1005114): Difference between revisions

For maximal monotone multivalued operators \(A,B\subset X\times X^*\), \(M\subset Y\times Y^*\) and a linear bounded operator \(L:X\to Y\), where \(X,Y\) may be nonreflexive, the maximality of the operators \(A+B\) and \(L^*ML\) is studied. The notion of a representable map is successfully used. A multivalued operator \(A\subset Z=X\times X^*\) is called representable if there is a proper convex lower semicontinuous function \(h:Z\to {\mathbb R}\cup \{\infty\}\) such that \(h(z)\geq p(z)\) for every \(z\in Z\), where \(p(x,x^*)=x^*(x)\) is the duality map, and \(z\in A\) iff \(h(z)=p(z)\). It is proved (Theorem 3.4) that an operator \(A\) is maximal monotone iff \(A\) is representable and \(h_A\geq p\) in \(Z\), where \(h_A\) is the Fitzpatrick function \(h_A(z)=\sup_{a\in A}\{z\cdot a -p(a)\}\). The author solves the maximality problem for representable operators under an additional assumption on a position of \(0\) with respect to domains of the operators.