Use this URL to cite or link to this record in EThOS:
Title: Evaluating many-electron molecular integrals for quantum chemistry
Author: Womack, James Christopher
ISNI:       0000 0004 5923 6835
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
The evaluation of molecular integrals is a vital but computationally expensive part of electronic structure calculations. This computational expense is particularly problematic for the explicitly correlated methods, in which complicated and numerous integrals over more than two-electrons must be evaluated. The successful R12/F12 methods overcome this difficulty by decomposition of these many-electron integrals by means of approximate resolutions of the identity (RIs). To obtain accurate results with this approach, however, requires large auxiliary basis sets with high angular momentum functions. To address this issue, we present a new RI-free variant of MP2-F12 theory, which uses density fitting to approximate three-electron integrals, rather than RIs. This approach demonstrates improved convergence of calculated energies with respect to the size and maximum angular momentum of the auxiliary basis set compared to the standard RI-based approach. For the systems on which the method was tested, relatively small auxiliary basis sets were sufficient to reduce errors in the correlation energy to less than a millihartree. The software implementation of the three-electron integral types needed in the new MP2-F12 variant proved to be extremely time-consuming. This difficulty inspired us to develop "Intception", a code generator which generates code for molecular integral evaluation. Intception is capable of automatically implementing code for evaluating a wide range of molecular integral types, using a general theoretical framework based on Obara-Saika-type recurrence relations [1] . To flexibly express integral definitions for use in Intception, a new domain-specific language was created. Testing revealed that the generated code evaluated integrals to a high numerical accuracy and on a reasonable timescale, though somewhat slower than existing optimized implementations. A detailed analysis of the performance of the generated code was undertaken, which suggested some possible routes to improving the efficiency of the code.
Supervisor: Not available Sponsor: Engineering and Physical Sciences Research Council
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available