Title: High-Performance Multiscale Fluid Simulation: GPU-Based Representations, Solvers, and Adaptive Algorithms

Mengdi Wang
Ph.D. Student in Computer Science
School of Interactive Computing
Georgia Institute of Technology

Date: Monday, Oct 27, 2025
Time: 11:00–12:00 PM
Location: CODA Building, Room C0908 “Home Park”

Committee:

Dr. Bo Zhu (Advisor) – School of Interactive Computing, Georgia Institute of Technology
Dr. Sehoon Ha – School of Interactive Computing, Georgia Institute of Technology
Dr. Greg Turk – School of Interactive Computing, Georgia Institute of Technology

Abstract:
Fluid phenomena involve intricate interactions across a vast range of spatial and temporal scales, posing persistent challenges for accurate and efficient numerical simulation. This dissertation investigates these multiscale challenges from complementary perspectives of representation, advection, data organization, and solver design. The first study develops a codimensional particle formulation for thin-film dynamics, coupling in-plane flow with surface deformation to model surface-tension-driven behaviors such as rupture and coalescence. The second introduces a compact, volume-conserving interface-tracking method on unstructured meshes that resolves sub-cell geometries and maintains mass conservation in complex multiphase flows. The third presents an adaptive hybrid flow-map scheme that combines grid-based and particle-based transport to preserve vortical details and reduce numerical dissipation in large-scale fluid motion. Together, these studies advance the algorithmic and computational foundations for simulating complex fluids across scales. Building on these developments, ongoing work on a matrix-free algebraic multigrid Poisson solver for adaptive octree grids extends this exploration toward a high-performance GPU-based simulation framework that unifies efficient computation with geometric and physical accuracy.