Saved in:
Bibliographic Details
Main Authors: von Esch, Maximilian Pierer, Völz, Andreas, Graichen, Knut
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.15938
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918510609825792
author von Esch, Maximilian Pierer
Völz, Andreas
Graichen, Knut
author_facet von Esch, Maximilian Pierer
Völz, Andreas
Graichen, Knut
contents Sensitivity-based distributed programming (SBDP) is a decomposition method for solving large-scale nonlinear programs over graph-structured networks. However, its convergence depends on the strength and structure of subsystem coupling. To address this limitation, we propose SBDP+, a distributed optimization scheme based on a structured primal-dual operator design. The method employs first-order sensitivities and primal decomposition to construct low-dimensional local subproblems that are solved in parallel using neighbor-to-neighbor communication. In contrast to SBDP, SBDP+ introduces a novel primal-dual update that ensures convergence under general coupling structures. Specifically, we establish local linear convergence for non-convex problems under standard regularity conditions. Numerical experiments demonstrate the effectiveness of SBDP+ and highlight improved robustness compared to SBDP and representative distributed optimization methods in applications such as statistical learning.
format Preprint
id arxiv_https___arxiv_org_abs_2509_15938
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Enforcing Convergence in Sensitivity-based Distributed Programming via Transformed Primal-Dual Updates
von Esch, Maximilian Pierer
Völz, Andreas
Graichen, Knut
Optimization and Control
Sensitivity-based distributed programming (SBDP) is a decomposition method for solving large-scale nonlinear programs over graph-structured networks. However, its convergence depends on the strength and structure of subsystem coupling. To address this limitation, we propose SBDP+, a distributed optimization scheme based on a structured primal-dual operator design. The method employs first-order sensitivities and primal decomposition to construct low-dimensional local subproblems that are solved in parallel using neighbor-to-neighbor communication. In contrast to SBDP, SBDP+ introduces a novel primal-dual update that ensures convergence under general coupling structures. Specifically, we establish local linear convergence for non-convex problems under standard regularity conditions. Numerical experiments demonstrate the effectiveness of SBDP+ and highlight improved robustness compared to SBDP and representative distributed optimization methods in applications such as statistical learning.
title Enforcing Convergence in Sensitivity-based Distributed Programming via Transformed Primal-Dual Updates
topic Optimization and Control
url https://arxiv.org/abs/2509.15938