This zip file contains code for calculating the anharmonic infrared (IR) spectrum of polycyclic aromatic hydrocarbons (PAHs) using a machine learning-based molecular dynamics (MLMD) approach. This zip file also includes the spectral data for the 1704 theoretically-calculated and 49 experimentally-tested PAHs as mentioned in the paper. The MLMD approach employs two distinct machine learning (ML) models: Neural Network Force Field (NNFF) [doi:10.1021/acs.jcim.1c01380]: Used to construct the potential energy surface. Electron Passing Neural Network (EPNN) [doi:10.1021/acs.jcim.0c01071]: Used to predict the molecular dipole moment. To compute the anharmonic IR spectrum, molecular dynamics (MD) simulations are performed to obtain atomic configurations (trajectories) during molecular vibrations. These configurations are generated using atomic forces predicted by the NNFF model. The resulting atomic trajectories are then used by the EPNN model to calculate the dipole moments at each time step. The dipole time-autocorrelation function is subjected to a Fourier transform to derive the IR absorption cross-section intensity. Both the NNFF and EPNN models are pre-trained and ready for immediate use, no additional ML training is required.

Peer reviewed