Research

My research focus lies at the intersection of computer science, numerical methods and material science. The overarching goal of my research is to develop computational algorithms and data-driven methods that can leverage the exascale supercomputers of tomorrow to perform ab-initio simulations at larger lengths and longer time scales, pivotal for addressing intricate challenges in energy storage device design. I also focus on developing multi-scale modelling strategies leveraging these methods to inform higher-scale models with useful data for accurate prediction of properties at continuum length scale. In particular my PhD research focus is centered around:

1. Contributing and maintaining DFT-FE a massively parallel real-space code for first principles modelling of materials using finite-element discretisation. 

2. Subspace projection approaches for large-scale chemical bonding analysis: This work provided an important direction towards extracting chemical bonding information using projected population analysis approach from large-scale real-space density functional theory (DFT) calculations on materials systems involving thousands of atoms with generic boundary conditions for the first time ever.  Large-scale chemical bonding analysis using the developed methods can play a crucial role in a variety of technologically important areas — materials design for hydrogen storage,  solid-state battery materials design,  efficient energy materials design for carbon capture etc., For more details about the method, read here(JCTC)

3. Real-space finite-element based methods for Projector-Augmented Wave (PAW) formalism in DFT:  Currently working on developing a local real-space formulation amenable for finite-element discretization of DFT to incorporate PAW formalism within the computational framework of DFT-FE code. The higher regularity associated with the PAW-DFT wavefunctions can lead to a drastic reduction of the degrees of freedom required for achieving chemical accuracy in material properties, thereby enabling even larger length-scales and longer time-scales in DFT calculations. Efficient computational strategies and scalable implementation procedures are being developed to reduce the computational cost of solving the underlying generalized Kohn-Sham eigenvalue problem.

4. Data driven approaches for accelerated Materials discovery: Robust on-the-fly active learning approaches are being developed within Graph Neural Network frameworks for deploying  machine-learning models trained using DFT data during the course of molecular dynamics simulations

5. Leveraging the developed numerical methodologies to address challenging problems in large-scale material modelling such as

   • Gaining first principles insights into migration properties of poly-crystalline electrolytes, grain boundary dependence on ion migration, and the growth of Li dendrites in solid-state electrolytes interface in the presence of an external electric field using ab-initio calculations

   • Understanding the electrochemical reaction pathway of CO2 reduction: to design and develop highly efficient and selective catalysts for CO2 reduction to more useful C1/C2 based products.