Electronic Structure Laboratory

Welcome to eslab.ucdavis.edu. Our research focuses on the development of numerical algorithms and high-performance software for electronic structure computations and First-Principles Molecular Dynamics (FPMD) simulations.

High-Performance First-Principles Molecular Dynamics

In order to enable large and accurate simulations of materials properties, we are developing new scalable algorithms for First-Principles Molecular Dynamics (FPMD). Our goal is to efficiently use the power of the largest supercomputers available today to extend the range of applications of FPMD. We develop advanced simulation features such as on-the-fly computation of spectroscopic data and coupling of FPMD simulations with efficient statistical sampling algorithms. Code development is carried out using C++/MPI/OpenMP and targets platforms such as Cray XT4, Cray XE6, IBM BlueGene/P and BlueGene/Q. This project is supported by the US Department of Energy Office of Basic Energy Sciences through grant DE-SC0008938 and is pursued in collaboration with Prof. G. Galli (Dept. of Chemistry, UC Davis) and Dr. E. Schwegler (Quantum Simulations Group, Lawrence Livermore National Laboratory).

ESTEST project

The ESTEST project aims at developing a framework for the automatic verification and validation of electronic structure programs. It allows users of electronic structure codes to archive and compare results obtained with various electronic structure programs (Abinit, Quantum Espresso, Qbox, Exciting and Siesta). ESTEST is available at http://estest.ucdavis.edu. A paper describing ESTEST has appeared in Computational Science and Discovery, available at http://iopscience.iop.org/1749-4699/3/1/015004. A more recent paper describing the distributed network features of ESTEST has appeared in Computer Physics Communications, available at http://dx.doi.org/10.1016/j.cpc.2012.03.016.

Pseudopotential repository project

A pseudopotential repository is available at http://fpmd.ucdavis.edu/potentials/index.htm. The repository contains potentials generated using the method of Hamann, Schluter and Chiang, modified by Vanderbilt, for LDA and PBE exchange-correlation functionals. Potentials translated from the UPF format used in the Quantum Espresso package are also included to facilitate validation and verification.

Qbox project

We develop and support Qbox, a C++/MPI implementation of FPMD for massively parallel computers. Qbox is available in source form under a GPL license. See the Qbox home page.
Release 1.60.0 of Qbox is available. It includes an implementation of stress tensor calculations for the PBE0 and B3LYP hybrid density functionals and several bug fixes.
Qbox implements the plane-wave, pseudopotential electronic structure method and was designed for scalability on thousands of processors. It has been ported to large parallel platforms, including BlueGene/P, BlueGene/Q, Cray XT-5, and a variety of Linux/Intel clusters. It is currently used in projects involving high-pressure simulations of liquids, semiconductor nanostructures, and materials science. Qbox achieved a performance of 207 TFlops on the BlueGene/L computer. The paper Large-Scale Electronic Structure Calculations of High-Z Metals on the BlueGene/L Platform was awarded the 2006 ACM/IEEE Gordon Bell Prize for Peak Performance. The design of Qbox is described in the following architecture paper.

Web tools

We develop XML-based tools to facilitate web-based information exchange for FPMD simulations. Web tools are built to interface to the Qbox code and other post-processing tools, including visualization programs. They conform to the FPMD XML Schema specification ( http://www.quantum-simulation.org)

Algorithm research projects

We are developing specialized parallel linear algebra implementations to accelerate the most time-consuming steps of electronic structure computations. Our work builds on the ScaLAPACK and Elemental parallel libraries. Applications include the calculation of Maximally Localized Wannier Functions (MLWFs) and their relation with algorithms for simultaneous approximate diagonalization of symmetric matrices, and the development of optimal extrapolation algorithms for Born-Oppenheimer FPMD.