Differentiable physics
jaxphys
I worked with Sushaan Kandukoori on jaxphys, a differentiable physics library built on JAX. It covers classical mechanics, electromagnetics, quantum mechanics, optics, and statistical mechanics, and uses automatic differentiation to solve inverse problems.
Differentiable simulation
The library includes symplectic integrators, a finite-difference time-domain Maxwell solver with perfectly matched layer boundaries, a split-operator Schrödinger solver, and GPU-accelerated Ising Monte Carlo.
Inverse problems
A Lagrangian defines the equations of motion automatically. Differentiating through the simulation lets an optimizer work backward from an observed outcome to model parameters.
Collaboration
Built with Sushaan Kandukoori in June–July 2026. The project was previously listed as neurosim; its repository and current name are jaxphys.