Proxtk

Proxtk, short for Proximal Toolkit, is an open-source software library designed to facilitate the development and benchmarking of proximal algorithms for convex optimization. It provides a modular framework for problems of the form min f(x) + g(x), where f and g are convex, possibly non-smooth.

The core consists of a solver engine, a library of proximal operators, and a problem-definition API. Prox operators support closed-form expressions and numerical approximations; solvers include proximal gradient methods, accelerated variants like FISTA, and operator-splitting methods such as ADMM and Douglas-Rachford. The design emphasizes extensibility, allowing users to add custom operators, constraints, and termination criteria. It supports multi-dimensional arrays, sparse matrices, and optional GPU acceleration.

Proxtk is intended for applications in signal processing, image reconstruction, compressed sensing, and machine learning, where regularization terms are expressed via proximal operators. It includes utilities for benchmarking, convergence diagnostics, and reproducible experiments, and provides bindings for commonly used languages such as Python and C++.

Development is community-driven, with ongoing contributions and documentation. Proxtk is used in academic research and education to illustrate proximal methods and compare algorithmic variants.

Related topics include proximal operator, proximal gradient method, ADMM, and convex optimization.