1. Influence of the choice of projection manifolds in the CASPT2 implementation
- Author
-
Takeshi Yanai, Jakub Chalupský, Per-Åke Malmqvist, Roland Lindh, Masaaki Saitow, and Yuki Kurashige
- Subjects
Source code ,010304 chemical physics ,Degree (graph theory) ,Computer science ,media_common.quotation_subject ,Suite ,Biophysics ,Electronic structure ,010402 general chemistry ,Condensed Matter Physics ,01 natural sciences ,Projection (linear algebra) ,0104 chemical sciences ,0103 physical sciences ,Code (cryptography) ,Perturbation theory (quantum mechanics) ,Complete active space ,Physical and Theoretical Chemistry ,Molecular Biology ,Algorithm ,media_common - Abstract
The Complete Active Space Second-Order Perturbation Theory (CASPT2) is well-established as a high-accuracy electronic structure method. It was originally implemented in the early 1990s to an efficient computer code in the molcas program suite, and this implementation has been extensively used as a standard tool. Here, we report a comparison of it against two independent computer-aided implementations of the CASPT2 method, revealing that the CASPT2 energies provided by the original code of molcas (version 8 or earlier) are inconsistent with the predictions of the lately developed computer-aided implementations. It is shown that this error is associated with the projections of the first-order equation onto the fully internally contracted multireference bases which are partially inconsistent between the left- and right-hand sides. The degree of the errors is assessed by performing illustrative CASPT2 calculations. The errors in total CASPT2 energies are demonstrated to be negligible relative to chemical accuracy in many cases, while there is a difficult case where they may substantially alter chemical description. The incorporation of the consistent projections into molcas has been carried out, which is available in the version 8 sp1. (Less)
- Published
- 2016