Interface between SchNetPack and SHARC

SHARC-Interface

…

Figure 1: Overview of the calculations conducted in a single iteration from time \(t\) to \(t+\Delta{t}\)) within a surface hopping simulation while connecting SchNetPack 2.0 (a machine learning model) with SHARC 3.0 through SPaiNN (SPaiNNulator). These stages encompass: (1) the anticipation of energies (in matrix \(H_{ji}\)), gradients (\(-\nabla_{\mathbf{R}}E_j\)), and couplings (\(\mathbf{C}_{ji}\)) for the configuration (\(\mathbf{R}, Z\)) at time-point \(t\) via the ML model; (2) the adjustment of atomic positions \(\mathbf{R}\); (3) the prognostication of energies, forces, and couplings for the revised configuration via the ML model; (4) the progression of electronic coefficients \(c_j\); (5) the computation of hopping probabilities \(h\) followed by a stochastic selection of the new active state \(j\); and (6) the computation of the gradient for the newly designated active state, set for employment in the subsequent iteration. For more details see 10.1002/9781119417774.ch16.