The maximal monotonicity of the subdifferentials of convex functions: Simons' problem (Q5961468)

Let \(E\) be a Banach space and \(f:E\to\mathbb{R}\cup\{+\infty\}\) be a convex lower semicontinuous function. Then it is well known that the subdifferential \(\partial f\) of \(f\) is a maximal monotone mapping. Hence, for a point \((q,q^*)\in E\times E^*\) with \[ \langle z^*-q^*,z-q\rangle\geq 0\qquad\forall(z,z^*)\in\text{graph }\partial f \] one can state that also \((q,q^*)\in\text{graph }\partial f\). In 1994, Simons presented two generalizations of this result showing the so-called \(Q-q^*\) and \(q-Q^*\) theorems: \(Q-q^*\):\hskip17mm If \(Q\subseteq E\) is weakly compact convex, \(q^*\in E^*\) and \hskip31mm \(\forall(z,z^*)\in\text{graph }\partial f\;\exists q\in Q:\langle z^*-q^*,z-q\rangle\geq 0\) \hskip31mm then \((Q\times q^*)\cap\text{graph }\partial f\neq\emptyset\). \(q-Q^*\):\hskip17mm If \(Q^*\subseteq E^*\) is weak\(^*\) compact convex, \(q\in E\) and \hskip31mm \(\forall(z,z^*)\in\text{graph }\partial f\;\exists q^*\in Q^*:\langle z^*-q^*,z-q\rangle\geq 0\) \hskip31mm then \((q\times Q^*)\cap\text{graph }\partial f\neq\emptyset\). In the paper, the author deals with an natural extension of both implications, the so-called \(Q-Q^*\) hypothesis: \(Q-Q^*\):\hskip17mm If \(Q\subseteq E\), \(Q^*\subseteq E^*\) be weakly/weak\(^*\) compact convex and \hskip31mm \(\forall(z,z^*)\in\text{graph }\partial f\;\exists q\in Q\), \(q^*\in Q^*:\langle z^*-q^*,z-q\rangle\geq 0\) \hskip31mm then (?) \((Q\times Q^*)\cap\text{graph }\partial f\neq\emptyset\). First, he gives an affirmative answer for the one-dimensional case, i.e., when \(E=\mathbb{R}\). Then for the case \(E=\mathbb{R}^2\) a simple counter example is presented, where \(Q\) and \(Q^*\) are unit squares and \(f\) is a specially constructed convex \(C^1\) function. If, however, an additional orthogonality condition for the sets \(Q\) and \(Q^*\) is assumed then the author shows that the hypothesis is true, even for higher-dimensional spaces. Finally, it is pointed out that in two-dimensional spaces this orthogonality condition and the \(Q-Q^*\) hypothesis are equivalent.