Linear Algebra Kernel for volesti #76
Description
Project
GeomScale
Summary
This project will introduce a flexible linear algebra kernel system in volesti, allowing users and developers to plug in various external linear algebra libraries such as Eigen, MKL, Spectra, and ARPACK. Using modern C++ traits and concepts, we will decouple volesti’s geometric algorithms from specific dependencies and enable high-performance, extensible, and user-configurable linear algebra backends.
submitter
Vissarion Fysikopoulos
project lead
vissarion
Community benefit
volesti is an open-source library for high dimensional sampling, volume computation, and semi-definite programming. By introducing this kernel abstraction, we expect a 2× speedup in main routines and the ability to handle larger, sparse, and higher-dimensional problems. This will directly benefit users working in computational geometry, optimization, Bayesian modeling, finance and systems biology.
Beyond volesti, other open-source developers will benefit from modular, header-only components that can be integrated easily into their own C++ projects. The new architecture will also make it easier for contributors to extend volesti with additional linear algebra backends in the future (e.g. GPU Linear Algebra), enhancing long-term sustainability and community engagement.
Amount requested (USD)
4000
Execution plan
Milestones:
M1 (Months 1–2): Design and implement the linear algebra kernel abstraction using Eigen as the baseline and add MKL support for core matrix operations and eigenvalue computations.
M2 (Months 3–4): Integrate Spectra for advanced eigenvalue problems and refactor volesti’s main routines to use the new kernel; benchmark, document, and merge into the main repository.
Who (github usernames)
vissarion, TolisChal
They are both very experienced with the volesti codebase (already very active contributors).