Block Lanczos Eigensolver

A C++20 implementation of a Block Lanczos eigensolver with thick restarting scheme for computing eigenvalues and eigenvectors of large sparse matrices.

Design

The library is designed with a flexible backend architecture that allows different multivector implementations to be easily integrated. This enables the use of specialized hardware implementations, such as GPU-accelerated backends, without modifying the core algorithm.

Backends

Currently provides an OpenMP backend for CPUs and a SYCL backend using AdaptiveCpp (other SYCL implementations probably work as well but are not tested). The SYCL backend is experimental.

Building

Requires:

• CMake
• C++20 compatible compiler
• Eigen
• Optionally AdaptiveCpp for the SYCL backend

If only the OpenMP backend should be built:

mkdir build && cd build
cmake ..
make

Run tests:

ctest