A unified family of multiblock ADMM schemes by means of graph splitting
Abstract
This paper introduces a novel unified family of multiblock ADMM schemes by combining Gabay duality with a graph-based splitting methodology. This framework allows us to recover existing ADMM algorithms as particular instances, as well as to generate new ones by varying the underlying network topology. We focus specifically on the case of the complete graph, a structure of significant practical interest due to its dense connectivity, for which we provide a fully explicit and directly implementable scheme. We further derive an equivalent reformulation of the proposed family in which the algorithm is expressed in terms of a reduced set of dual variables, which clarifies its connection to the classical ADMM framework. Numerical experiments are conducted to compare the performance of different graph choices, showing particularly favorable results for the complete graph.
Related
- Domain preserving splitting schemes for a class of SPDEs driven by a standard Brownian motion
- Finite Hardy Phase Theory for Power Means: An exact analytic shooting family through the Carleman parameter
- A unified family of counterexamples to Batyrev's non-negativity conjecture on stringy Hodge numbers
- U-Examples: A Unified Family of Worst-Case Instances for Optimization Algorithms
- MBAnalysis: An R package for multiblock data analysis
BibTeX
@misc{campoy2026unified,
title = {{A unified family of multiblock ADMM schemes by means of graph splitting}},
author = {Rubén Campoy and Scott B. Lindstrom and César López-Pastor},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.36278},
eprint = {2610.36278},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.36278},
}
RIS
TY - UNPB
TI - A unified family of multiblock ADMM schemes by means of graph splitting
AU - Rubén Campoy
AU - Scott B. Lindstrom
AU - César López-Pastor
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2610.36278
UR - https://arxiv.org/abs/2610.36278
AB - This paper introduces a novel unified family of multiblock ADMM schemes by combining Gabay duality with a graph-based splitting methodology. This framework allows us to recover existing ADMM algorithms as particular instances, as well as to generate new ones by varying the underlying network topology. We focus specifically on the case of the complete graph, a structure of significant practical interest due to its dense connectivity, for which we provide a fully explicit and directly implementable scheme. We further derive an equivalent reformulation of the proposed family in which the algorithm is expressed in terms of a reduced set of dual variables, which clarifies its connection to the classical ADMM framework. Numerical experiments are conducted to compare the performance of different graph choices, showing particularly favorable results for the complete graph.
ER -