doi:10.1016/j.cpc.2004.12.003
Copyright © 2004 Elsevier B.V. All rights reserved.
Efficient computation of the coupling matrix in time-dependent density functional theory
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Emmanuel Lorin de la Grandmaisona, Shivaraju B. Gowdab, Yousef Saadb,
,
, Murilo L. Tiagoc and James R. Chelikowskyc
aDépartement de Mathématiques, Université d'Orsay, 91405 Orsay (France) and Centre de Mathématiques et de Leurs Applications, Ecole Normale Supérieure de Cachan, 94235 Cachan, France
bDepartment of Computer Science and Engineering, Institute for the Theory of Advanced Materials in Information Technology, Digital Technology Center, University of Minnesota, Minneapolis, MN 55455, USA
cDepartment of Chemical Engineering and Materials Science, Institute for the Theory of Advanced Materials in Information Technology, Digital Technology Center, University of Minnesota, Minneapolis, MN 55455, USA
Received 3 August 2004;
revised 2 December 2004;
accepted 7 December 2004.
Available online 12 February 2005.
Abstract
We present an efficient implementation of the computation of the coupling matrix arising in time-dependent density functional theory. The two important aspects involved, solution of Poisson's equation and the assembly of the coupling matrix, are investigated in detail and proper approximations are used. Poisson's equation is solved in the reciprocal space and bounded support of the wave functions are exploited in the numerical integration. Experiments show the new implementation is more efficient by an order of magnitude when compared with a standard real-space code. The method is tested to compute optical spectra of realistic systems with hundreds of atoms from first principles. Details of the formalism and implementation are provided and comparisons with a standard real-space code are reported.
Fig. 1. Flowchart for finding oscillator strengths using TDDFT.
Fig. 2. Construction of the coupling matrix.
Fig. 3. Cut-off of Coulomb interaction between periodic images of a supercell.
Fig. 4. Long-range part and short range part of the density.
Fig. 5. Parallelization of the construction of the coupling matrix for 4 processors.
Fig. 6. Optical absorption spectra for SiH4.
Fig. 7. Optical absorption spectra for Si5H12.
Fig. 8. Optical absorption spectra for Si34H36.
Table 1.
Comparison of the computational complexity for several fast Poisson solvers. The complexity for Gaussian elimination assumes a 3D mesh leading to a banded matrix of band N2/3

Table 2.
Solution of the Poisson's equation

Table 3.
Comparison of wall-clock runtime of the parallel TDLDA code (not including the generation of Kohn–Sham eigenvalues and wave functions) using Fourier space and Real space for the Si34H36 test case running on 8 processors

Work supported by NSF grants ITR-0082094, DMR-0325218, by DOE under Grants DE-FG02-03ER25585, DE-FG02-03ER15491, and by the Minnesota Supercomputing Institute.

Corresponding author.