Analysing the correlation between the water's OH stretching band and its hydrogen bonding configurations by machine learning

In this work, we combined ab initio calculations and machine learning to analyse the relationship between the different hydrogen bonding configurations and the OH stretching band of water. IR wavenumbers and intensities of water clusters of various sizes were theoretically calculated to form a database for machine learning. An artificial neural network model was then trained to predict the water molecules’ stretching mode vibrational energy, and an importance analysis of the model was performed. The importance analysis of the model reveals that the strength of the donor hydrogen bond has a larger effect on the vibrational energy than the acceptor hydrogen bond, and the vibrational energy of the symmetric and asymmetric stretching modes depends on the strength of the donor hydrogen bond on different OH arms. Based on the importance analysis results, we also discussed the origin behind the differences in the wavenumbers between each hydrogen bonding configuration and showed that the same principle of hydrogen cooperativity in water clusters can be extended to bulk water.