Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 44: Developement of Calculation Methods II

MM 44.2: Talk

Wednesday, March 20, 2024, 16:00–16:15, C 264

Machine-learning interatomic potentials with beyond-DFT accuracy: application to covalent-organic frameworks — •Yuji Ikeda, Axel Forslund, and Blazej Grabowski — University of Stuttgart, Stuttgart, Germany

Covalent-organic frameworks (COFs) are nanoporous crystalline materials formed by strong covalent bonds of organic secondary building units composed mostly of light elements like C, N, O, H, etc. Most COFs are quasi-two-dimensional materials with layers interacting with van der Waals (vdW) forces. It is fascinating to investigate COFs using machine-learning interatomic potentials (MLIPs) because of their capability to access, e.g., long time- and length-scales in molecular-dynamics (MD) simulations. To simulate vdW materials, MLIPs should be trained by data including the vdW interaction. Such data are typically prepared with vdW-DFT functionals. These vdW-DFT functionals are however essentially semi-empirical in the sense that their parameters are fitted to show agreement with experiments, and hence they imply concerns about transferability. Our solution is to consider training data obtained from post-Hartree-Fock (HF) methods such as the coupled-cluster (CC) methods, which are non-empirical and have beyond-DFT accuracy. Using MLIPs trained on the beyond-DFT data, we demonstrate the calculation of structural properties of COFs.

Keywords: Machine-learning interatomic potentials (MLIPs); Covalent-organic frameworks (COFs)