

Stabilizing the European power grid after a fault
A 900 MW generation loss in Greece sends a frequency-deviation wave across the continent. Differentiable predictive control drives every node back to nominal in real time.
PhD Student · Johns Hopkins University
I build neural operators and differentiable controllers for partial differential equations — fast, accurate surrogates for turbulence, fracture, phase fields and biomechanics, scaled on national-lab supercomputers.
Research
Selected projects, with real outputs from the papers. Choose a project to see it in motion.
A 900 MW generation loss in Greece sends a frequency-deviation wave across the continent. Differentiable predictive control drives every node back to nominal in real time.
Neural-operator policies for multi-agent control of incompressible flow, invariant to the number of actuators and sensors. The pipeline behind 1st place at the global Tesseract hackathon.
Differentiable predictive control on a time-integrated DeepONet tracks a target profile with distributed actuators. The surrogate matches the solver to a median relative \(L_2\) error of 1.11%.
Dendritic solidification and other moving-boundary problems, together with agentic workflows for materials science, from my 2026 internship at the LLNL Data Science Institute.
Speeds up crack-propagation simulation without losing accuracy, with a reinforcement-learning agent that steers the crack. Built with FEniCSx.
Multimodal neural operators for TBI biomechanics, aimed at personalized protective gear.
Distributed CUDA + MPI training of high-dimensional operators: 83% faster, two Director's Discretionary Awards.
A multimodal RAG assistant for Statics & Mechanics, in production on Streamlit.
Gradient boosting with differential evolution for engineered systems. 3rd place, NASA & DNV challenge.
Live Lab
A vibrating membrane integrated with explicit finite differences, drawn as a 3D wireframe.
A uniform mixture separates into two phases and coarsens, minimizing interface energy.
Stam's stable-fluids solver: semi-Lagrangian advection and a pressure projection that keeps the flow divergence-free.
A heated rod tracks a moving target \(u^{*}\) (dashed) with distributed actuators. A simple version of my differentiable predictive control work.
Publications
CINOC: Cardinality-Invariant Neural Operator Policies for Scalable PDE Control
Zanotta P., Sarkar D.R., Zheng H., Goswami S., Drgoňa J.
Learning to Control PDEs with Differentiable Predictive Control and Time-Integrated Neural Operators
Sarkar D.R., Goswami S. · arXiv:2511.08992
Learning Hidden Physics and System Parameters with Deep Operator Networks
Sarkar D.R., Kag V., Pal B., Goswami S. · 456:118926
Multimodal Neural Operators for Real-Time Biomechanical Modelling of Traumatic Brain Injury
Agarwal A., Sarkar D.R., Goswami S. · 109398
Learning Generalizable Neural Operators for Inverse Problems
Thorpe A.J., Tretiakov S., Sarkar D.R., Kumar K., Topcu U. · arXiv:2512.18120
Adaptive Interface-PINNs for Inverse Problems: Determining Material Properties for Heterogeneous Systems
Sarkar D.R., Annavarapu C., Roy P. · 249:104373
Uncertainty-Aware Optimization in Engineered Systems via Gradient Boosting and Differential Evolution
Sarkar D.R., Basu S., Manuel L., Goswami S.
Luo Y., Sarkar D.R., Sangree R.H., Goswami S. · arXiv:2604.06179
Adaptive Interface-PINNs (AdaI-PINNs) for Transient Diffusion: Forward and Inverse Problems in Heterogeneous Media
Roy S., Sarkar D.R., Annavarapu C., Roy P., Lecampion B., Valiveti D.M. · 244:104305
Interface Physics-Informed Neural Networks (I-PINNs) to Solve Inverse Problems in Heterogeneous Materials
Sarkar D.R., Annavarapu C., Roy P. · Machine Learning for Computational Science and Engineering
Full record on Google Scholar →
About
Civil engineer by training, now building machine learning for the equations underneath the structures.


