Proximal approximation of firmly nonexpansive maps with small derivative asymmetry
We study uniform approximation of a firmly nonexpansive map by the proximal map of a single convex function. Let $H$ be a real Hilbert space, let $R>0$, and suppose that $T$ is continuously Fr\'echet differentiable on a neighbourhood of the closed ball $B_{\sqrt{2}R}$, firmly nonexpansive on that ball, and satisfies $\sup_{B_{\sqrt{2}R}}\|DT-DT^*\|\le\varepsilon$. We prove that a finite continuous convex function $\varphi$ exists with $\sup_{B_R}\|T-\operatorname{prox}_{\varphi}\|\le R\sqrt{2\varepsilon}$. The estimate is independent of dimension and requires no bound on derivatives of the skew part. The proof converts the differential defect into approximate convex interpolation inequalities, then applies an elementary form of established approximate-subdifferential integration. An affine proximal approximation gives the complementary bound $R/2$. For a planar rotation perturbation with $0\le\varepsilon\le1$, the exact approximation error is $\varepsilon R/2$; thus this example does not establish optimality of the square-root rate.
Authors
- Sam Vaseghi (ORCID: https://orcid.org/0009-0006-4105-1949)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.5281/zenodo.24157351
- Primary Topic
- Optimization and Variational Analysis
- Type
- preprint