On the Metric Subregularity for Cyclic Relaxed Douglas-Rachford Operators via Spectral Theory and Differentiability

Người báo cáo: Mai Ngọc Hoàng Anh

Date: 21 May 2026 (Thursday)

Venue: Room 507, A6, Institute of Mathematics

Time: 9:30 - 11:00

Abstract: We study metric subregularity of the residual mapping $I-T$ associated with a cyclic relaxed Douglas--Rachford ($\CDR_\lambda$) operator $T$ in a finite dimension.

First, for linear operators we provide a spectral characterization of the linear subregularity modulus in terms of the smallest singular value of $I-T$ restricted to $(\Fix T)^\perp$.

Second, for nonlinear operators generated by smooth manifolds we combine differentiability of the projector (Lewis--Malick) with linearization to reduce local behavior of $\CDR_\lambda$ to a cyclic composition of relaxed Douglas--Rachford mappings on tangent subspaces. In the consistent case, tangent-space transversality forces $\Fix DT(\bar x)=\{0\}$ and yields linear metric subregularity. In the inconsistent case, local linearization is performed along a fixed cycle of shadow points.  Each block derivative of the cyclic relaxed Douglas--Rachford operator acts on the tangent spaces of the active sets, and the fixed space of the full derivative is characterized via an intersection of the kernels of the corresponding shape operators. This yields an exact formula for $\Fix DT(\bar x)$ and allows us to deduce linear metric subregularity under active-tangent transversality.

Finally, we formulate metric subregularity with nonlinear gauge and show that analytic fixed-point mappings satisfy a H\"older-type error bound via a \L ojasiewicz inequality, yielding nonlinear-gauge subregularity even when linear subregularity fails.

This is joint work with David Russell Luke and Thi Lan Dinh, Institute for Numerical and Applied Mathematics, University of Göttingen.

  Hoạt động tuần
Xuất bản mới
Adam Czornik, Đoàn Thái Sơn, Nguyễn Thị Thu Sương, Pole Placement Theorem for Linear Measurable Time-Varying Control Systems with Single Input, SIAM Journal on Control and Optimization, Vol. 64, Iss. 4 (2026)
Đinh Nho Hào, Maxim Shishlenin, Van Ba Cong, Stable Numerical Solution to Multi-dimensional Nonlinear Inverse Heat Conduction Problems via Artificial Neural Networks, Lobachevskii Journal of Mathematics, Volume 47, pages 1213–1232 (2026)
Nguyễn Trung Thành, Gianluca Barone, Dat Tran, Đinh Nho Hào, All-at-once proximal alternating minimization method for an inverse medium scattering problem, Journal of Computational Physics, Article: 115187 Volume: Volume 565 (2026)