12 november 2020 13:00 Zoom
The research centre AASS arranges a seminar with Danny Thonig, Örebro University.
Model hamiltonians in materials theory are an established way to map out energy variations from first-principles methods into a parametrized form . This has the advantage that typical computational simulation scales improve from femtoseconds and hundrates of atoms in first-principles methods up to nanoseconds and billions of atoms in simulations with parametrized model hamiltonians. Here, the parameters in, for instance, linearized model hamiltonians are often be calculated from ab-initio with satisfying, but not always good agreement to the experiment  due to limits in the ab-initio theory. But also limit of such parametrization itself is strongly debatable , precisely because information about the first derivatives of the energy (atomic forces and fields) are sufficient to characterize thermal equilibrium and the dynamics in materials. To overcome the above mentioned issues - inconsistence to experiment and of established linearized model hamiltonians - more and more machine learning techniques are applied . These applications are e.g. "fitting" the parameters in the linearized model hamiltonian directly to experiment, determining phase transitions as well as critical phenomena , predicting thermal ground states with a drastic reduction of the autocorrelation compared other established methods [5,6], or designing non-linearized model hamiltonians for multi-scale approaches .
In my talk, I want to present different examples from literature and own research for the application of machine learning in materials theory, in particular for the Heisenberg model in magnetic systems, harmonic approximation for atomic crystal lattice vibrations, and the tight binding model for the electron ground state.
 O. Eriksson et al., "Atomistic Spin- dynamics - Foundations and Applications" (2017), Oxford Univ. Press  A. Szilva et al., Phys. Rev. B 96, 144413  E. A. Bedolla-Montiel et al., J. Phys.: Cond. Matt. (2020) (accepted)  A. A. Shirinyan et al., Phys. Rev. B 99, 041108(R) (2019)  H. Y. Kwon et al., Phys. Rev. B 99, 024423 (2019)  A. Kovacs et al., J. Mag. Mag. Mat. 491, 165548 (2019)  B. Mortazavi et al., Mater. Horiz., 7, 2359-2367 (2020)