Proximity for sums of composite functions (Q536271)

Let \(C\) be a nonempty closed convex subset of a real Hilbert space \(H\) with scalar product \(\langle\cdot,\cdot\rangle\) and associated norm \(\|\cdot\|\). A best approximation to a point \(z\in H\) in \(C\) is the point \(P_cz\) that satisfies \(\| P_c z-z\|= \min_{x\in C}\| x-z\|\). The mapping \(P_c: H\to C\) is called best approximation operator or projector onto \(C\). If \(i_c\) is the indication function of \(C\), i.e., \[ i_c: x\mapsto\begin{cases} 0\quad &\text{if }x\in C,\\ +\infty\quad &\text{if }x\not\in C,\end{cases} \] then \(P_cz\) is solution to the minimization problem to minimize \(i_c(x)+{1\over 2}\| x-z\|^2\), \(x\in H\). Let \(\Gamma_0(H)\) be the class of lower-semicontinuous convex functions \(f: H\to\mathbb{R}\cup\{+\infty\}\) such that \(\text{dom}f= \{x\in H: f(x)<+\infty\}\neq\varnothing\). \textit{J. J. Moreau} [C. R. Acad. Sci., Paris 255, 2897--2899 (1962; Zbl 0118.10502)] observed that, for every function \(f\in\Gamma_0(H)\), the proximal minimization problem possesses a unique solution \(\text{prox}_fz\). The resulting proximity operator \(\text{prox}_f: H\to H\) extends the notion of best approximation operator for a convex set. The objective of this paper is to propose a splitting algorithm to compute proximity operators in the case when \(f\) can be decomposed as a sum of composite functions and to investigate its asymptotic behaviour. Applications to best approximation and image recovery are also described.