074103-16 Thøgersen et al. J. Chem. Phys. 123, 074103 2005 TABLE VI. The gradient norm g=4SDF−FDS in the first six iterations of the cadmium complex calculations seen in Fig. 9. DIIS DIIS-TRRH TRSCF Iteration E KS g E KS g E KS g 1 −5597.0 7.8 −5597.0 7.8 −5597.0 7.8 2 −5502.3 14.9 −5598.4 7.2 −5598.3 7.1 3 −5602.1 9.7 −5600.3 8.5 −5603.7 9.3 4 −5628.5 2.1 −5599.9 7.7 −5611.1 9.1 5 −5627.4 3.5 −5599.9 7.8 −5616.8 7.7 6 −5628.8 0.8 −5600.2 8.1 −5622.7 7.5 conv no conv conv In the previous implementation of the TRSCF algorithm, the focus was on the optimization of the Hartree–Fock energy. As the Kohn–Sham energy is nonquadratic in the density matrix, the local DSM energy model has been generalized and is now expanded about the current-density matrix D 0 in the subspace of the density matrices D i of the previous iterations. To satisfy the idempotency condition, the energy model function is parametrized in terms of a purified averaged density matrix. The local energy function is correct to second order in D i −D 0 and can be set up solely in terms of the density matrices and Kohn–Sham matrices of the previous iterations. In the Hartree–Fock theory, the new local energy model is identical to the one previously used in TRSCF optimizations. The EDIIS function is discussed in the context of the proposed model. In the Hartree–Fock theory, the EDIIS function is obtained from our proposed energy function by neglecting terms that result from the purification of the density matrix; the EDIIS function therefore does not reproduce the Hartree–Fock gradient at the expansion point. In the DFT, the EDIIS function is inappropriate for other reasons as well. A rederivation of the original DIIS algorithm is also performed to understand when it can safely be applied. In particular, it is shown that the DIIS method may be viewed as a quasi-Newton method, thus explaining its fast local convergence. In the global region, its behavior is less predictable, although we note that its gradient-norm minimization mechanism usually allows it to recover safely from sudden, large increases in the total energy brought on by the Roothaan– Hall iterations. The TRSCF scheme is tested both in a computationally demanding, robust line-search implementation TRSCF-LS, and in our standard implementation, where only the Fock/ Kohn–Sham matrices of previous iterations are used. Our test calculations indicate not only that the TRSCF-LS method is a highly stable and robust method, but also that the standard TRSCF implementation converges rapidly in most cases, with little degradation relative to the TRSCF-LS scheme. Relative to these schemes, the DIIS method is somewhat more erratic since it makes no use of Hessian information and therefore cannot predict reliably what directions will reduce the total energy. For example, in situations where the energy changes in the course of the iterations but the gradient does not, the DIIS algorithm is unable to identify the density matrix with the lowest energy and may diverge. Nevertheless, the DIIS method handles most optimizations amazingly well, which is particularly impressive in view of its very simplicity; never has so few lines of code done so much good for so many calculations. In general, however, it is outperformed by the TRSCF method, which introduces Hessian information at little extra cost, and is well founded in the global as well as local regions of the optimization. The current formulation of TRSCF requires a few diagonalizations in each TRRH step, and to obtain linear scaling these diagonalizations should be avoided. An even more efficient algorithm may be obtained if the Roothaan–Hall and DSM steps are integrated in such a manner that the information from the previous density matrices are directly used in the Roothaan–Hall optimization step. Work along these lines is in progress. ACKNOWLEDGMENTS We thank Peter Taylor, Ditte Jørgensen, and Stephan Sauer for providing some of the test examples. This work has been supported by the Danish Natural Research Council. We also acknowledge support from the Danish Center for Scientific Computing DCSC. 1 C. C. J. Roothaan, Rev. Mod. Phys. 23, 691951. 2 G. G. Hall, Proc. R. Soc. London A205, 541 1951. 3 P. Pulay, Chem. Phys. Lett. 73, 393 1980; J. Comput. Chem. 3, 556 1982. 4 K. N. Kudin, G. E. Scuseria, and E. Cancès, J. Chem. Phys. 116, 8255 2002. 5 G. Karlström, Chem. Phys. Lett. 67, 348 1979. 6 L. Thøgersen, J. Olsen, D. Yeager, P. Jørgensen, P. Sałek, and T. Helgaker, J. Chem. Phys. 121, 162004. 7 R. Fletcher, Practical Methods of Optimization, 2nd ed. Wiley, New York, 1987. 8 V. R. Saunders and I. H. Hillier, Int. J. Quantum Chem. 7, 6991973. 9 J. B. Francisco, J. M. Martínez, and L. Martínez, J. Chem. Phys. 121, 22 2004. 10 W. Koch and M. C. Holthausen, A Chemist’s Guide to Density Functional Theory Wiley-VCH, Weinheim, 2000. 11 T. Helgaker, P. Jørgensen, and J. Olsen, Molecular Electronic Structure Theory Wiley & Son, ltd., Chichester, 2000. Downloaded 23 Aug 2005 to 130.225.22.254. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
074103-17 Trust-region self-consistent-field J. Chem. Phys. 123, 074103 2005 12 R. Seeger and J. A. Pople, J. Chem. Phys. 65, 265 1976. 13 G. B. Bacskay, Chem. Phys. 61, 385 1981; J. Phys. France 35, 639 1982. 14 P. Jørgensen, P. Swanstrøm, and D. Yeager, J. Chem. Phys. 78, 347 1983. 15 R. McWeeny, Rev. Mod. Phys. 32, 335 1960. 16 X. P. Li, W. Nunes, and D. Vanderbilt, Phys. Rev. B 47, 10891 1993. 17 J. M. Millam and G. E. Scuseria, J. Chem. Phys. 106, 5569 1997. 18 C. Ochsenfeld and M. Head-Gordon, Chem. Phys. Lett. 270, 3391997. 19 X. Li, J. M. Millam, G. E. Scuseria, M. J. Frisch, and H. B. Schlegel, J. Chem. Phys. 119, 7651 2003. 20 T. Helgaker, H. J. Jensen, P. Jørgensen et al., DALTON, a molecular electronic structure program, Release 2.0, 2004; http://www.kjemi.uio.no/ software/dalton 21 A. Schäfer, H. Horn, and R. Ahlrichs, J. Chem. Phys. 97, 2571 1992. 22 T. H. Dunning, J. Chem. Phys. 90, 10071989. Downloaded 23 Aug 2005 to 130.225.22.254. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
- Page 1:
PhD Thesis Optimization of Densitie
- Page 5 and 6:
Contents Preface ..................
- Page 7:
Summary............................
- Page 11:
List of Publications This thesis in
- Page 14 and 15:
Part 1 Improving Self-consistent Fi
- Page 16 and 17:
Part 1 Improving Self-consistent Fi
- Page 18 and 19:
Part 1 Improving Self-consistent Fi
- Page 20 and 21:
Part 1 Improving Self-consistent Fi
- Page 22 and 23:
Part 1 Improving Self-consistent Fi
- Page 24 and 25:
Part 1 Improving Self-consistent Fi
- Page 26 and 27:
Part 1 Improving Self-consistent Fi
- Page 28 and 29:
Part 1 Improving Self-consistent Fi
- Page 30 and 31:
Part 1 Improving Self-consistent Fi
- Page 32 and 33:
Part 1 Improving Self-consistent Fi
- Page 34 and 35:
Part 1 Improving Self-consistent Fi
- Page 36 and 37:
Part 1 Improving Self-consistent Fi
- Page 38 and 39:
Part 1 Improving Self-consistent Fi
- Page 40 and 41:
Part 1 Improving Self-consistent Fi
- Page 42 and 43:
Part 1 Improving Self-consistent Fi
- Page 44 and 45:
Part 1 Improving Self-consistent Fi
- Page 46 and 47:
Part 1 Improving Self-consistent Fi
- Page 48 and 49:
Part 1 Improving Self-consistent Fi
- Page 50 and 51:
Part 1 Improving Self-consistent Fi
- Page 52 and 53:
Part 1 Improving Self-consistent Fi
- Page 54 and 55:
Part 1 Improving Self-consistent Fi
- Page 56 and 57:
Part 1 Improving Self-consistent Fi
- Page 58 and 59:
Part 1 Improving Self-consistent Fi
- Page 60 and 61:
Part 1 Improving Self-consistent Fi
- Page 62 and 63:
Part 1 Improving Self-consistent Fi
- Page 64 and 65:
Part 1 Improving Self-consistent Fi
- Page 66 and 67:
Part 1 Improving Self-consistent Fi
- Page 68 and 69:
Part 1 Improving Self-consistent Fi
- Page 70 and 71:
Part 1 Improving Self-consistent Fi
- Page 72 and 73:
Part 2 Atomic Orbital Based Respons
- Page 74 and 75:
Part 2 Atomic Orbital Based Respons
- Page 76 and 77:
Part 2 Atomic Orbital Based Respons
- Page 78 and 79:
Part 2 Atomic Orbital Based Respons
- Page 80 and 81:
Part 2 Atomic Orbital Based Respons
- Page 82 and 83:
Part 2 Atomic Orbital Based Respons
- Page 84 and 85:
Part 2 Atomic Orbital Based Respons
- Page 86 and 87:
Part 2 Atomic Orbital Based Respons
- Page 88 and 89:
Part 2 Atomic Orbital Based Respons
- Page 90 and 91:
Part 3 Benchmarking for Radicals ar
- Page 92 and 93:
Part 3 Benchmarking for Radicals le
- Page 94 and 95: Part 3 Benchmarking for Radicals As
- Page 96 and 97: Part 3 Benchmarking for Radicals R
- Page 99: Summary The developments in compute
- Page 103: Appendix A The Derivatives of the D
- Page 106 and 107: Thus, the density element D µν is
- Page 109 and 110: References 1 2 3 4 5 6 7 8 9 C. C.
- Page 111: 65 T. Helgaker and P. Jørgensen, T
- Page 115 and 116: JOURNAL OF CHEMICAL PHYSICS VOLUME
- Page 117 and 118: 18 J. Chem. Phys., Vol. 121, No. 1,
- Page 119 and 120: 20 J. Chem. Phys., Vol. 121, No. 1,
- Page 121 and 122: 22 J. Chem. Phys., Vol. 121, No. 1,
- Page 123 and 124: 24 J. Chem. Phys., Vol. 121, No. 1,
- Page 125 and 126: 26 J. Chem. Phys., Vol. 121, No. 1,
- Page 127: Part 1 The Trust-region Self-consis
- Page 130 and 131: 074103-2 Thøgersen et al. J. Chem.
- Page 132 and 133: 074103-4 Thøgersen et al. J. Chem.
- Page 134 and 135: 074103-6 Thøgersen et al. J. Chem.
- Page 136 and 137: 074103-8 Thøgersen et al. J. Chem.
- Page 138 and 139: 074103-10 Thøgersen et al. J. Chem
- Page 140 and 141: 074103-12 Thøgersen et al. J. Chem
- Page 142 and 143: 074103-14 Thøgersen et al. J. Chem
- Page 147: Part 3 A Coupled Cluster and Full C
- Page 150 and 151: L. Thøgersen, J. Olsen / Chemical
- Page 152 and 153: L. Thøgersen, J. Olsen / Chemical
- Page 154 and 155: L. Thøgersen, J. Olsen / Chemical
- Page 156 and 157: L. Thøgersen, J. Olsen / Chemical
- Page 159 and 160: 3030 J. Phys. Chem. A 2004, 108, 30
- Page 161 and 162: 3032 J. Phys. Chem. A, Vol. 108, No
- Page 163: 3034 J. Phys. Chem. A, Vol. 108, No