Lie group integrators and efficient integration of gradient flow

29 Jul 2021, 22:00
15m
Oral presentation Algorithms (including Machine Learning, Quantum Computing, Tensor Networks) Algorithms (including Machine Learning, Quantum Computing, Tensor Networks)

Speaker

Oleksiy Bazavov (Michigan State University (US))

Description

It has been recently shown how explicit low-storage Lie group integrators can be built from classical low-storage methods of Williamson's type. We discuss a one-parameter family of three-stage third-order methods and show how a coefficient scheme can be chosen specifically for optimal integration of the gradient flow. We also illustrate how two low-storage fourth-order integrators can be used as Lie group methods and compare the performance of various integrators. The low-storage schemes of this type can be easily implemented in the existing lattice codes.

Authors

Oleksiy Bazavov (Michigan State University (US)) Thomas Chuna (Michigan State University)

Presentation materials