Prediction of interaction energy for rare gas dimers using machine learning approaches

In our present work, we applied Machine Learning approaches to predict potential energy profiles for rare gas dimers as well as for the H-2 molecule. We designed an Artificial Neural Network (ANN) model with one and two layers, with two to eight neurons in each layer, to predict potential energy values. We compared the ANN predicted energy values with the ab initio data and we found an excellent agreement between the actual and predicted values. The root mean squared deviation (RMSD) values for the test data are found to be 0.10, 0.22, 0.03 and 0.47 cm(-1) for He-2, Ne-2, Kr-2 and Ar-2, respectively. Further, we observed that the ANN method is able to fit the potential energy profile for weak van der Waals dimers as well as covalently bound molecules.

Automated summary

In this study, we used Machine Learning techniques to predict potential energy profiles for rare gas dimers and the H-2 molecule. We developed an Artificial Neural Network (ANN) model with one to two layers and two to eight neurons per layer to make these predictions. Comparisons between the ANN predicted energy values and the ab initio data showed excellent agreement. The root mean squared deviation (RMSD) values for the test data were determined to be 0.10, 0.22, 0.03, and 0.47 cm(-1) for He-2, Ne-2, Kr-2, and Ar-2, respectively. Furthermore, the ANN method successfully fit the potential energy profiles for both weak van der Waals dimers and covalently bound molecules.