A parallel implementation of the recently developed local pair natural orbital coupled electron pair approximation (LPNO-CEPA/n, n = Version 1, 2, or 3) and the corresponding LPNO coupled cluster method with single- and double excitations (LPNO-CCSD) is described. A detailed analysis alongside pseudocode is presented for the most important computational steps. The scaling with respect to the number of processors is reasonable and speedups of about 10 with 14 processors have been found in benchmark calculations (wall-clock time). The most important factor limiting the efficiency of the scaling with respect to the number of processors is probably the limited bandwidth of the presently prevailing multicore machines. The parallel LPNO methods were applied to study weak intermolecular interactions. Initially, the well-established S22 set of molecules was studied. The mean absolute error resulting from the use of the LPNO-CEPA/1 method relative to the most recent CCSD(T) reference data is found to be 0.24 kcal/mol. Thus, LPNO-CEPA/1 holds great promise for the efficient ab initio treatment of weak intermolecular interactions. In order to demonstrate the applicability of the methods to real systems, a two-dimensional potential energy surface for a trimer of 2,4-dihydroxy-3-acetyl-6-methyl acetophenone [C11H12O4] (81 atoms, 1296 basis functions, 133 single points) has been calculated with the LPNO-CEPA/1 method. In this system, a clear distinction can be made between hydrogen bonding and π-π interactions. The global minimum on the PES obtained from the calculations agrees excellently with the experimentally determined crystal structure. By contrast, popular density functional methods show no discernible minimum.

Download full-text PDF

Source
http://dx.doi.org/10.1021/ct100445sDOI Listing

Publication Analysis

Top Keywords

local pair
8
pair natural
8
natural orbital
8
orbital coupled
8
coupled cluster
8
scaling respect
8
respect number
8
number processors
8
weak intermolecular
8
intermolecular interactions
8

Similar Publications

Want AI Summaries of new PubMed Abstracts delivered to your In-box?

Enter search terms and have AI summaries delivered each week - change queries or unsubscribe any time!