Implementation and testing of Lanczos-based algorithms for Random-Phase Approximation eigenproblems

Computational Materials Science

Volumn 50, Issue 7, 2011, Pages 2148-2156

  Grüning, Myrta ^a,e   Marini, Andrea ^b,c,d   Gonze, Xavier ^e  

^a UNIVERSITY OF COIMBRA   (Portugal)

^b UNIVERSITY OF ROME TOR VERGATA   (Italy)

^c UNIVERSITY OF THE BASQUE COUNTRY UPV EHU   (Spain)

^d BASQUE FOUNDATION FOR SCIENCE   (Spain)

^e UNIVERSITÉ CATHOLIQUE DE LOUVAIN   (Belgium)

Author keywords

Bethe Salpeter equation; Iterative solvers; Tamm Dancoff approximation; Time dependent density functional theory

Indexed keywords

BETHE-SALPETER EQUATION; COMPUTATIONAL LOADS; EIGEN-PROBLEMS; HERMITIANS; ITERATIVE SOLVERS; LANCZOS; NON-HERMITIAN HAMILTONIANS; OPTICAL ABSORPTION; OPTICAL RESPONSE; RANDOM PHASE APPROXIMATIONS; TAMM-DANCOFF APPROXIMATION; TECHNICAL DETAILS; TIME-DEPENDENT DENSITY FUNCTIONAL THEORY;

APPROXIMATION ALGORITHMS; CARBON NANOTUBES; DENSITY FUNCTIONAL THEORY;

HAMILTONIANS;

EID: 79954423115     PISSN: 09270256     EISSN: None     Source Type: Journal    
DOI: 10.1016/j.commatsci.2011.02.021     Document Type: Article

References (38)

1. A.L. Fetter, and J.D. Walecka Quantum Theory of Many-Particle Systems 2003 Dover p. 565 (Chapter 15)
2. P. Ring, and P. Schuck The Nuclear Many-Body Problem (Theoretical and Mathematical Physics) 1980 Springer p. 244 (Chapters 7 and 8)
3. E. Runge, and E.K.U. Gross Phys. Rev. Lett. 52 1984 997
4. E.E. Salpeter, and H.A. Bethe Phys. Rev. 84 1951 1232
5. RPA can be more efficient than the direct solution especially when one is interested only to a frequency range where the excitations are well separated.
6. G. Onida Rev. Mod. Phys. 74 2002 601
7. Z. Bai, J. Demmel, J. Dongarra, A. Ruhe, H. van der Vorst (Eds.), Templates for the Solution of Algebraic Problems: A Practical Guide, SIAM, Philadelphia, 2000.
8. S. Tretiak, C.M. Isborn, A.M.N. Niklasson, and M. Challacombe J. Chem. Phys. 130 2009 054111
9. M. Casida D.P. Chong, Recent Advances in Density Functional Methods vol. 1 1995 World Scientific Singapore
10. V. Olevano, and L. Reining Phys. Rev. Lett. 86 2001 5962
11. A. Marinopoulos, M. Grüning, Phys. Rev. B., submitted for publication.
12. M. Grüning, A. Marini, and X. Gonze Nano Lett. 9 2009 2820
13. Y. Ma, M. Rohlfing, and C. Molteni Phys. Rev. B 80 2009 241405
14. H.A. van der Vorst Math. Comp. 39 1982 559
15. Note that in the Lanczos approach the energy range includes always the lowest energies. In principle it is possible to choose a different energy range, not necessarily containing the lowest energy, using the modification proposed by G. Grosso, L. Martinelli and G. Pastori Parravini, Phys. Rev. B 51 (1995) 13033-13038.
16. R. Haydock, in: H. Ehrenfest, F. Seitz, D. Turnbull (Eds.), Solid State Phys. 35 (1980) 215, Academic Press.
17. M. Cini Topics and Methods in Condensed Matter Theory from Basic Quantum Mechanics to the Frontiers of Research 2007 Springer Berlin pp. 313-316
18. In particular, at each iteration both the computational cost and the storage are increasing, since all the vectors in the basis are needed to determine the new one.
19. In principle also the Hermitian algorithm can break down. In fact after many iterations, because of the numerical error, the vectors in the Lanczos basis may stop to be orthonormal.
20. B. Walker, A.M. Saitta, R. Gebauer, and S. Baroni Phys. Rev. Lett. 96 2006 113001
21. D. Rocca, R. Gebauer, Y. Saad, and S. Baroni J. Chem. Phys. 128 2008 154105
22. D. Rocca, D. Lu, and G. Galli J. Chem. Phys. 133 2010 164109
23. A. Mostafazadeh J. Math. Phys. 43 2002 205
24. A. Mostafazadeh J. Math. Phys. 43 2002 3944
25. R. Zimmermann Phys. Stat. Sol. 41 1970 23
26. L.X. Benedict, and E.L. Shirley Phys. Rev. B 59 1999 5441
27. For real matrices, as in the case of TD-DFT for finite systems and local exchange-correlation functionals, the cost is reduced further, since one can construct before-hand the two matrix A = R + C and B = R - C, so that only a single matrix-vector multiplication is needed (giving another factor two). Moreover usually B reduces to a diagonal matrix, containing on the diagonal the energy differences of the particle-hole pairs. In this case the algorithm is equivalent to the one proposed by Casida, see Ref. [9].
28. X. Gonze, J.M. Beuken, R. Caracas, F. Detraux, M. Fuchs, G.M. Rignanese, L. Sindic, M. Verstraete, G. Zerah, and F. Jollet Comput. Mater. Sci. 25 2002 478
29. X. Gonze, G.M. Rignanese, M. Verstraete, J.M. Beuken, Y. Pouillon, R. Caracas, F. Jollet, M. Torrent, G. Zerah, and M. Mikami Z. Kristallogr. 220 2005 558
30. X. Gonze, B. Amadon, P.M. Anglade, J.M. Beuken, F. Bottin, P. Boulanger, F. Bruneval, D. Caliste, R. Caracas, and M. Côté Comput. Phys. Commun. 180 2009 2582
31. A. Marini, C. Hogan, M. Grüning, and D. Varsano Comput. Phys. Commun. 180 2009 1392
32. -1 using a gas saturation method (298 K), J. Mol. Spectrosc. 123 (1987) 382.
33. R. Chen, and H. Guo Chem. Phys. Lett. 369 1997 650
34. R. Chen, and H. Guo J. Chem. Phys. 119 2003 5762
35. g.
36. The sum of the asymptotic values tends to twice the particle-hole pair energy cutoff, the difference to the optical gap of the system. See also Ref. [21].
37. We have tested that the convergence behavior does not depend on the particular expression for the error and that the behavior resembles that of the error with respect to the exact diagonalization, at least when no terminator is used.
38. The reason of just 1 eV difference, while the unoccupied orbital space of R′ contains 25 more states, is that KS states in this energy region are in fact very close to each others.