Research

M.S. Research — meshless lattice Boltzmann for 3D acoustics

Oregon State University, 2024 · Computer Graphics & Visualization Group

A Highly-Parallel Meshless LBM Leveraging Nonlocal Calculus for 3D Acoustics in Graphics.

This master’s research develops a meshless lattice Boltzmann method whose spatial discretization comes from nonlocal vector calculus rather than a grid: the streaming step becomes a sparse operator on an unstructured point cloud, retaining grid-LBM's embarrassing parallelism while fitting real 3D scenes without meshing. Applied to immersive sound propagation for computer graphics — the first fluid-simulation approach to 3D acoustics in that setting — with validation against analytic solutions, Doppler shift, and double-slit interference, and demonstrations including a ~4M-node cathedral and a city scene with moving sources, all on a single consumer GPU.

The discretization generalizes beyond LBM: the same nonlocal gradient/divergence/Laplacian construction applies to arbitrary PDEs — a thread that continues directly into my current work on kinetic light transport.

Research report (PDF, 17 MB)

Current program

My current research runs on CospanMDAO, a GPU-first differentiable modeling & optimization framework, and spans two applications: a meshless kinetic light-transport renderer (publication in preparation) and Mycelium, a dynamical neural-field architecture exploring stateful, continuous-time computation. The common thread is kinetic transport methods on unstructured point clouds — carrying one discretization philosophy from sound, to light, to neural dynamics.

Publications