5 November 2020
IT4Innovations
Europe/Prague timezone

Precision in ab initio calculations

5 Nov 2020, 11:30
20m
ONLINE (IT4Innovations)

ONLINE

IT4Innovations

Studentská 1B 708 33 Ostrava - Poruba

Speaker

Jiri Klimes (Faculty of Mathematics and Physics, Charles University)

Description

Computer simulations of materials have been immensely useful for understanding processes at the atomic and molecular levels. The ability to 'see' atoms moving during reaction or dynamics is exciting and useful to understand experimental observations. However, one should not forget that a range of approximations is needed to be able to perform such a simulation. For quantum simulation of molecules and materials one usually things first about the accuracy -- approximations of Hamiltonian and wavefunction. Less severe approximations give more accurate results but usually require substantially more computational resources compared to more approximate methods. However, computer simulations also require approximations to numerical set-up, such as introducing a real-space grid or basis functions to represent continuous functions. Using coarser grid or less functions speeds-up the calculations but can lead to a substantial loss of precision, sometimes changing the results qualitatively. I will discuss two examples of issues concerning precision. First, many simulations of materials use pseudopotentials or similar methods, such as the projector-augmented wave (PAW) scheme, to avoid the calculation of tightly-bound core electrons. Typically, different PAW datasets are available for every element, more precise hard or less precise soft allowing much faster calculations. We have analyzed the errors of different PAW datasets by comparing to binding curves of molecules calculated with reference all-electron set-up. Our goal is not only to evaluate the errors but also understand how much are the errors transferable, that is, if a simple scheme could be used to correct the results obtained with the less precise datasets. Second, I will discuss an interesting class of molecular solids, called co-crystals, which are formed by two, or more, components. There are co-crystals with strong hydrogen bonds between the individual molecules. In some cases the hydrogen can transfer from one molecule to the other so that a salt is formed. While one needs accurate methods to correctly describe if the hydrogen will transfer or not, the process is also very sensitive to the precision of the computational set-up. I will show examples where choosing too cheap set-up qualitatively changes the results and how this can be avoided.

Primary author

Jiri Klimes (Faculty of Mathematics and Physics, Charles University)

Co-authors

Dr Sirous Yourdkhani (Charles University) Dr Michal Hušák (University of Chemistry and Technology, Prague)

Presentation materials

There are no materials yet.