I build theoretical frameworks and computational strategies for problems where signals hide in noise and structure must be inferred from incomplete data. Analytical solutions when they exist, large-scale simulations when they don't — and increasingly, machine learning that respects the underlying physics. Neural networks gain power when you encode the right constraints. The most interesting problems are the ones where brute force fails and elegance is required.
Reconstructing 3D configurations from sparse 2D measurements using physics-informed constraints.
Large-scale simulations — billions of particles on massively parallel architectures.
Closed-form solutions for stochastic systems, polymer physics, and statistical mechanics.
Dynamical arrest, glass transitions, and the signatures of regime change.
Topology, entanglement, and collective dynamics in confined and knotted systems.
130,000+ cores, distributed algorithms, petabyte-scale data pipelines.
I built polymer models and ran molecular dynamics simulations to reconstruct the 3D chromosomal environment where Xist RNA spreads. The modeling showed how chromatin architecture constrains the spreading — basically, the physical landscape determines the biology.
I developed a method to recover full spatiotemporal dynamics — 3D structure plus how it evolves — from Hi-C contact frequency matrices. These are sparse, noisy measurements of which parts of the genome touch each other. Polymer physics plus constrained optimization lets you pull continuous structure out of discrete contacts.
The first billion-atom biomolecular simulation. One billion particles, Newtonian dynamics, 130,000+ cores on Trinity. We had to figure out how to distribute force calculations, manage state across thousands of nodes, and make sense of petabytes of trajectory data.
I derived the analytical conditions for when a polymer system arrests — transitions from flowing to stuck. Turns out connectivity alone determines the critical point. It happens exactly where constraint counting says rigidity should emerge.
How do you get directional motion from random thermal noise? Break the symmetry. I worked out the analytics and simulations for Brownian ratchets in confined geometries — the same mechanism molecular motors use.
My path hasn't been linear. I trained as a physicist at Cambridge, spent three years at Los Alamos in the Center for Nonlinear Studies, and am now a Principal Investigator at Harvard and Mass General. Along the way: Brownian ratchets, viral proteins, chromosome dynamics.
What connects it all is the problem of finding structure when noise dominates. I go for analytical solutions when they exist — there's something satisfying about an exact result. When that's not possible, I build simulations big enough that the structure can emerge on its own.
The problems I like best sit right at the edge of tractable. The ones where the right representation, the right constraints, the right level of description turns something that looks impossible into something you can actually solve.
Python, C++, CUDA, MPI. Molecular dynamics, statistical mechanics, stochastic processes, polymer physics, HPC, signal processing.
Harvard University & Mass General Hospital
Los Alamos National Laboratory
University of Cambridge
University of Edinburgh
Always interested in hard problems.