Probing Vibrational Excitation Patterns of Benzene Dimer using Machine Learning Tools

In this letter, another important application of machine learning (ML) techniques is demonstrated in understanding and analyzing vibrational excitation patterns of benzene dimer complex. A total of 15000 trajectories is integrated via chemical dynamics simulations to generate the required data for the present ML applications. This work shows that, when training with the survival probability information of the complex, a Multi Layer Perceptron model can predict the same when all modes are randomly excited. A Gradient Boosting model is used to find out the contribution from modes of different frequency groups to the overall dissociation pattern. Most importantly, a Decision Tree Regression model shows its power in predicting the result obtained from a different excitation energy without using any data from that energy.