AutoLyap

A Python package for automated Lyapunov-based convergence analyses of first-order optimization and inclusion methods.

PyPI version PyPI downloads GitHub stars Paper


Overview

AutoLyap streamlines the process of constructing and verifying Lyapunov analyses by formulating them as semidefinite programs (SDPs). It supports a broad class of structured optimization and inclusion problems, providing computer-assisted proofs of linear or sublinear convergence rates for many well‑known algorithms.

A typical workflow:

  1. Choose the class of optimization/inclusion problems.

  2. Choose the first-order method to analyze.

  3. Choose the type of Lyapunov analysis to search for or verify (which implies a convergence or performance conclusion).

AutoLyap builds the underlying SDP and solves it through configurable backend solvers.

Quick start

For installation instructions and first end-to-end workflows, see Quick start.

Cite this project

If AutoLyap contributes to your research or software, please cite [UDGT+26].

@misc{upadhyaya2026autolyap,
  author = {Upadhyaya, Manu and Das Gupta, Shuvomoy and Taylor, Adrien B. and Banert, Sebastian and Giselsson, Pontus},
  title = {The {AutoLyap} software suite for computer-assisted {L}yapunov analyses of first-order methods},
  year = {2026},
  archivePrefix = {arXiv},
  eprint = {2506.24076},
  primaryClass = {math.OC},
}
[UDGT+26]

Manu Upadhyaya, Shuvomoy Das Gupta, Adrien B. Taylor, Sebastian Banert, and Pontus Giselsson. The AutoLyap software suite for computer-assisted Lyapunov analyses of first-order methods. 2026. arXiv:2506.24076.

Other computer-assisted methodologies

PEPit is a computer-assisted performance estimation framework that targets worst-case analyses of first-order methods through SDP formulations. AutoLyap is complementary: it focuses on Lyapunov analyses and automates the corresponding SDP formulations. In practice, PEPit is a strong choice for tight bounds, while AutoLyap is tailored to Lyapunov-based proofs and scalable analysis patterns.