Neural network potentials for the phosphoester bond formation between phosphate and methanol in water

Neural Network Potentials for H2PO4 and HPO4, as well as the training sets used to train them. 

The energies are in eV, the forces in eV/angstrom, the xyz coordinates in angstrom, and the box size in angstrom. 

For H2PO4: 

Number of structures = 386457

Number of atoms = 397

- coord_H2PO4.npy: npy array containing the xyz coordinates. Array size = (number of structures, number of atoms * 3)

- box_H2PO4.npy: npy array containing the box sizes. Array size = (number of structures, 9)

- force_H2PO4.npy: npy array containing the atomic forces. Array size = (number of structures, number of atoms * 3)

- energy_H2PO4.npy: npy array containing the full box potential energy. Array size = (number of structures)

- type_H2PO4.npy: npy array containing the atom types. Array size = (number of atoms)

- type_H2PO4.txt: text file containing the correspondance between the type index, as given in type_H2PO4.npy and the atom name

- graph_1_000_compressed_H2PO4.pb: neural network potential to propagate trajectories.

The nomenclature is the same for HPO4.

Number of structures = 227881

Number of atoms = 396