Experiences with participation in crystal structure prediction test - CSP7

M. Hušák, F. Fňukal

University of Chemistry and Technology, Prague, Technická 5, 166 28 Praha 6 – Dejvice

husakm@vscht.cz

Introduction

Blind crystal structure prediction (CSP) test is a periodic event organized by Cambridge Crystallographic Data Centre. Structure prediction typically consists from 3 phases:

1) Building a 3D molecule model from its connectivity information

2) Generating a set of suggested structures

3) Ranking of the suggested structures

This year, it was optional to participate only in the phases 3). We had participated in ranking of suggested structures of following molecules:

Figure 1. Structures from CSP7 test, XXVII (100 suggestions), XXXI (100 suggestions), XXXIII (500 suggestions)

Methods

Based on preliminary benchmarks we had used for the ranking energy calculation ground on DFT theory. We had used rSCAN functional combined with MBD dispersion correction. The data were prepared in BIOVIA Materials Studio software. Computation was performed on Karolina supercomputer at TU Ostrava. CASTEP 20.11 software was used. The computational power required for the calculations was approximately 2 000 000 core/hours.

Results

Based on the energy ranking, we had identified following structures from the input data sets as the one probably existing in reality:  XXVII_structure_59, XXXI_structure_59, XXXIII_structure_452. The correct results of CSP7 were not disclosed up to now so we can not confirm any success.

During the data processing we were faced with limitation of commercial software used for CIF data transfer to DFT software acceptable format as well as with issues related to manage multiple jobs on a supercomputer. Based on the experiences we had chosen to develop our own software for DFT calculation handling.

Development of the CheckCIF-DFT code

The CheckCIF-DFT code is software targeted primary for transformation of CIF files to format acceptable by DFT software. The code just supports CASTEP and Quantum Espresso formats. The program can manage external DFT software run as well as analyze results. It can manage computation of multiple structures collection suitable for structure prediction. Calculation setup can be defined in a form of user-editable templates. The code was tested on structure ranking of the XXXI CSP7 target. Calculations on an 8 core PC (PBE functional with Grimme dispersion correction, Quantum Espresso) take 5 months. The results suggested the same structure (XXXI_structure_59) as the previous calculation. The code can be utilized for crystal structure solution verification, low-quality structure solution improvement as well as for light atom position clarification.