Ishna Satyarth: Application of Neural Networks in Quantum Chemistry

Understanding the motion of electrons in a molecule is a big piece of the puzzle to understand the quantum world that these sub-atomic particles live in. This complexity increases several folds when correlation of pairs of electrons are interpreted in the wave function. Current methods of including pair correlation are too expensive to apply to large molecules. Our goal is to reduce the error in the Tensor Hypercontraction approximation, which can reduce the cost of such calculations. Artificial Intelligence and in particular the application of neural networks can prove to be a savior in such situations of many unknown parameters. A neural network works like a collection of brain cells where each neuron is responsible for processing a single piece of information and increasing the number of neurons allows adding complexity in a systematic way. Since there are more than 22 different input variables accounting for the energy error, we will be using a multilayer perceptron model. Using an Artificial Neural Network, we have tried to structure them into a systematic network and tried to decipher how each variable contributes towards the outcome of total energy or bond strength. We are currently working on collecting as many data points to train our Neural Network, so that we can get the most accurate results in the future.