SHNITSEL - Surface Hopping Nested Instances Training Set for Excited-state Learning

SHNITSEL-dynamic The Surface Hopping Nested Instances Training Set for Excited-State Learning (SHNITSEL) is a comprehensive data repository designed to support the development and benchmarking of excited-state dynamics methods. SHNITSEL-dynamic contains datasets with comprehensive surface hopping trajectory data of five organic molecules: Alkenes: ethene (A01), propene (A02), 2-butene (A03) Ring structures: 1,3-cyclohexadiene (R02) The data are stored in xarray format using xarray.DataTree objects for efficient handling of multidimensional trajectory ensembles. Metadata such as units, electronic-structure method, charge, and number of electronic states are stored as attributes of the individual datasets within the tree. The datasets contain key electronic-structure quantities for singlet and triplet states, including energies, nuclear forces, dipole moments, transition dipole moments, nonadiabatic couplings, and spin-orbit couplings, computed at the multireference ab initio level. Two data representations are provided: Stacked (#271,700 data points in total): for each molecule, the DataTree contains a single dataset in which all trajectories are stacked along a trajectory dimension. Unstacked (#271,700 data points in total): for each molecule, the DataTree contains one dataset per trajectory, stored as individual leaves. These complementary formats allow users to choose between trajectory-resolved and ensemble-level data representations depending on their analysis needs.