MATRIX lab's research is centered around developing mathematical techniques and novel scalable computational methodologies that are aimed at pushing the boundaries of the current predictive capabilities of design of materials, opening up the possibility of addressing a broad range of scientifically and technologically important problems that have been out of reach so far. Research at MATRIX lab is highly interdisciplinary and combines ideas from quantum mechanics, materials science, chemistry, mechanics of solids, adaptive finite-element methods, numerical linear algebra, machine learning and heavy dose of high performance computing (HPC) .
Recent disruptive advancements in parallel computing architectures require a paradigm shift in the development and numerical implementation of computational methods to exploit extreme levels of parallelism offered by these machines. If one can exploit the capabilities of these machines, fast, scalable and high-fidelity materials simulations are possible than before. Our research group is attempting to bridge this gap by (a) developing mathematical techniques, scalable computational methods and hardware-aware implementation strategies that can enhance the predictive abilities of materials design employing quantum mechanical theories, (b) developing light-weight machine learning frameworks for accelerated materials discovery, (c) leveraging these abilities to address challenging material modeling problems which can provide deeper insights into various aspects of material properties at the nano-scale, thereby informing higher-scale models for accurate prediction of material properties at the macroscopic scale.
MATRIX lab is involved in developing the open-source code DFT-FE, a massively parallel finite-element-based open-source code for material modelling using density functional theory (DFT). The current development efforts span our research group at IISc and the Computational Materials Physics Group at the University of Michigan, USA. DFT-FE was nominated as a 2019 ACM Gordon Bell Prize Finalist and also the workhorse behind the 2023 ACM Gordon Bell Prize, the highest prize in high-performance computing applications, and the first time, a research group from India has been a recipient of this prestigious prize as a part of an international team.
Fast, accurate and scalable real-space finite-element-based computational methodologies for quantum modelling of materials on emerging architectures.
Hardware-aware algorithms focused towards extreme computing (MPI+CUDA) for the solution of large-scale nonlinear eigenvalue problems and linear systems of equations with applications towards quantum modelling of materials.
Machine learning frameworks towards accelerating materials discovery.
Algorithms leveraging quantum computing paradigm for large sparse-matrix eigenvalue problems.
Collaborate with application experts to address large-scale material modelling problems leveraging the methods developed in our group. A few examples include --
(i) Ab initio understanding of solid electrode-electrolyte interfaces for efficient battery materials; Firsi principles material modeling for energy-storage materials.
(ii) Electrochemical CO2 reduction pathways in the presence of supported doped nanoparticles.
(iii) Electronic properties of nanoparticles with charged dopants, twisted bi-layers of TMDs.
(iv) Atomistic modelling of shear transformation zones in bulk-metallic glasses (BMGs) -- a multiscale modeling approach.
(v) Multi-scale frameworks for chemical kinetic modelling with combustion applications.
University of Michigan, Ann Arbor, USA (Mechanical Engineering and Materials Engineering Departments)
Indian Institute of Technology, Mumbai (Mechanical Engineering Department)
Indian Institute of Science, Bangalore (Materials Engineering, Chemical Engineering, SSCU Departments)
Research Institute for Sustainable Energy (RISE), TCG Crest, Kolkata, India
Indo-Korea Science and Technology Center (IKST), (Bangalore and South Korea)
NVIDIA, India
Argonne National Lab, USA