Answers are grounded in this paper and cite their sources.
For the problem of finding zeroes of maximally monotone operators, we introduce a general scheme allowing both symmetric and asymmetric metrics in the proximal point algorithms. Our approach subsumes various special cases of proximal point methods in the literature. Moreover, it brings new insights into the relationship between the hybrid proximal methods with bounded relative-error approximations and inexact proximal methods with error-summability-type conditions. In particular, introducing asymmetric metrics made it possible to cast summable errors in the framework of relative ones. Global convergence is established under standard assumptions. In the process, a certain generalization of Féjer-monotonicity is presented to obtain the local linear convergence rate under the calmness condition on the operator (which does not require uniqueness of solution). As a proof-of-concept, we show some computational advantages of the introduced asymmetric scheme, deriving from it a regularized diagonal Newton method for smooth nonlinear equations.
No takes yet. Share an insight, caveat, or question.
Buscaglia et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: