Final PhD Defence of Mr. Yagyank Srivastava, scheduled for tomorrow, April 25, from 2:00 PM to 3:00 PM in the ME Auditorium, Mechanical Engineering Department. 

Title: Accelerated Lattice Thermal Conductivity Prediction Using Machine Learning 
Date & Time: April 25, 2:00–3:00 PM 
Venue: ME Auditorium, Department of Mechanical Engineering 

Examination Committee: 

External Examiner: Prof. Umesh V Waghmare, Theoretical Science Unit, JNCASR.

Chairperson: Prof. Aftab Alam, Department of Physics, IIT Bombay. 

Internal Examiner: Prof. Soham Mujumdar, Mechanical Engineering Department, IIT Bombay. 

Supervisor: Prof. Ankit Jain, Mechanical Engineering Department, IIT Bombay. 

Abstract: 
The study of lattice thermal transport connects atomic vibrations to key engineering challenges in thermal management for electronics, energy materials, and thermoelectric applications. Thermal transport calculations typically involve Density Functional Theory (DFT), lattice dynamics, and the solution of the Boltzmann Transport Equation (BTE). However, this workflow is both computationally expensive and time-consuming, particularly for materials with large unit cells or low symmetry. This work aims to complement these calculations using machine learning (ML). A diverse dataset comprising BTE-driven thermal conductivity values for 235 ternary semiconductors, spanning 43 chemical species and 32 space groups, was created using ab initio-driven interatomic force constants. The performance of existing state-of-the-art ML models, along with a newly proposed graph neural network-based model capable of learning material fingerprints directly from data, was tested for predicting thermal conductivity. The performance of this end-to-end ML approach was found to be suitable for material screening purposes. Furthermore, instead of relying solely on the end-to-end approach, two major computational bottlenecks, DFT-based local potential energy surface (PES) mapping and BTE-based scattering rate calculations, were individually accelerated using ML. These ML-assisted approaches are complementary and can be integrated within the same workflow. The accuracy of thermal conductivity prediction was preserved within 10% of the reference values. Overall, by incorporating ML-assisted lattice dynamics and BTE calculations, a significant 10× reduction in computational cost was achieved across the thermal conductivity prediction workflow. 

About the Candidate: 
Yagyank is a Ph.D. researcher in Mechanical Engineering at IIT Bombay, working towards understanding and optimising thermal transport in semiconductors and complex materials. Yagyank is passionate about bridging physics-based models with data-driven techniques. His current research focuses on combining machine-learned interatomic potentials with first-principles methods to compute harmonic and higher-order interatomic force constants (IFCs), and to accelerate Boltzmann Transport Equation (BTE) calculations. 

He is a direct Ph.D. candidate, having completed his undergraduate studies in Mechanical Engineering from SMVDU, J&K. Before joining academia, he worked at Tata Consultancy Services (TCS) for two years as a CAD engineer. At IIT Bombay, he has also contributed as a Research Assistant at the Makerspace facility, supporting innovation and hands-on prototyping. 

Outside of research, Yagyank has been actively involved in institute's cultural activities. He has written and performed scripts for speaking events, including stand-up acts and anchoring. He also served as the convenor of the Speaking Arts and Comedy Club in the PG cultural council of IITB. He considers himself an optimistic and light-hearted person, with a cautious approach and a strong sense of presence and awareness, both in and outside the lab.