Modular implementation of linear- and cubic-scaling orbital minimization methods

26 Apr 2023

SIESTA was originally designed to support simulations that scale linearly with the number of atoms, but in practice the linearly-scaling solvers have not been widely used: most calculations so far have used the default cubic-scaling solvers.

We have recently implemented an improved, more reliable version of linear-scaling solvers in SIESTA, as reported in the following open-access publication:

Irina V. Lebedeva, Alberto García, Emilio Artacho, and Pablo Ordejón: Modular implementation of the linear- and cubic-scaling orbital minimization methods in electronic structure codes using atomic orbitals, Royal Society Open Science 10, 230063 (2023) Open Access logo.]

In particular, this work is based on the libOMM library, which implements orbital minimization methods (OMM) – unconstrained minimization methods of suitable energy functionals – as part of the Electronic Structure Library (ESL). We have created and integrated into SIESTA an interface to libOMM, and we have added linear-scaling solvers to libOMM, which already provided cubic-scaling solvers. This also involved improvements to the MatrixSwitch library used by libOMM. MatrixSwitch ultimately leverages the ScaLAPACK and DBCSR libraries for linear algebra methods with dense and sparse matrices, respectively.

This modularization is yet another step towards making SIESTA more sustainable from the software engineering point of view. More importantly, this work will underpin future efficient simulations of larger atomistic systems thanks to the improved linear-scaling methods, thus increasing the breadth of the applications of SIESTA.

This new functionality is scheduled to become available in SIESTA as part of the 6.0 release. A preview is already available on the rel-MaX-2 showcase branch, for more information see the Guide to SIESTA versions.