Speaker
Description
Accurate prediction of physical properties is necessary when developing high-value materials such as pharmaceuticals. At the same time, methods needed to achieve required accuracy, such as periodic density functional theory (DFT), remain commonly used despite being computationally demanding. Given that interest in modelling even larger and more complex systems has only continued to grow, these factors have led to the development of cruder yet faster algorithms such as density functional tight binding (DFTB). For instance, its implementation in DFTB+ is reported to improve runtimes up to two orders of magnitude by exploiting parallelization and matrix algebra [1]. Still, approximations inherent in these semi-empirical techniques can cause properties to be predicted inaccurately.
This work explores the performance of DFTB-D4 for predicting thermodynamic properties of two common pharmaceutical precursors, benzophenone and sulfamerazine, within quasi-harmonic approximation (QHA). As DFTB-only QHA is burdened by unacceptable errors, we look at whether matching DFTB outputs with DFT signals improves DFTB-only predictions [2]. In this composite DFT/DFTB QHA framework, costly constant-volume DFT calculations are run only at the equilibrium volume, and the volume-dependence of the crystal’s static and dynamic degrees of freedom is obtained instead by running DFTB at various fixed volumes. Comparisons for α- and β-benzophenone are made to describe whether discrepancies observed between individual QHA levels relate to polymorphism. Finally, the utility of the hybrid approach is demonstrated through the stable Pna21 polymorph of sulfamerazine, where a volume-dependent description of DFT energies can be too costly to obtain otherwise.
[1] Hourahine, B., et al., (2020), J. Chem. Phys., 152, 124101.
[2] Ludík, J., et al., (2024), J. Chem. Theory. Comput., 20, 2858.
Financial support from Czech Science Foundation under the project No. 23-05476M and from COST action CA22107 (BEST-CSP) is acknowledged. Computational resources were supplied under the project e-INFRA CZ (ID:90254) provided by the Ministry of Education, Youth and Sports of the Czech Republic. R. G. acknowledges funding from the grant of specific university research No. A1_FCHI_2024_001.
