Accelerated Infeasibility Detection of Constrained Optimization and Fixed-Point Iterations

Abstract

As first-order optimization methods become the method of choice for solving large-scale optimization problems, optimization solvers based on first-order algorithms are being built. Such general-purpose solvers must robustly detect infeasible or misspecified problem instances, but the computational complexity of first-order methods for doing so has yet to be formally studied. In this work, we characterize the optimal accelerated rate of infeasibility detection. We show that the standard fixed-point iteration achieves a $\mathcal{O}(1/k^2)$ and $\mathcal{O}(1/k)$ rates, respectively, on the normalized iterates and the fixed-point residual converging to the infimal displacement vector, while the accelerated fixed-point iteration achieves $\mathcal{O}(1/k^2)$ and $\tilde{\mathcal{O}}(1/k^2)$ rates. We then provide a matching complexity lower bound to establish that $\Theta(1/k^2)$ is indeed the optimal accelerated rate.

Cite

Text

Park and Ryu. "Accelerated Infeasibility Detection of Constrained Optimization and Fixed-Point Iterations." International Conference on Machine Learning, 2023.

Markdown

[Park and Ryu. "Accelerated Infeasibility Detection of Constrained Optimization and Fixed-Point Iterations." International Conference on Machine Learning, 2023.](https://mlanthology.org/icml/2023/park2023icml-accelerated/)

BibTeX

@inproceedings{park2023icml-accelerated,
  title     = {{Accelerated Infeasibility Detection of Constrained Optimization and Fixed-Point Iterations}},
  author    = {Park, Jisun and Ryu, Ernest K.},
  booktitle = {International Conference on Machine Learning},
  year      = {2023},
  pages     = {27294-27345},
  volume    = {202},
  url       = {https://mlanthology.org/icml/2023/park2023icml-accelerated/}
}