Georgia Institute of Technology
High-Performance Multiscale Fluid Simulation: GPU-Based Representations, Solvers, and Adaptive Algorithms
Abstract
dc:description.abstractThis dissertation develops high-performance multiscale fluid simulation algorithms for features that span spatial scales far smaller than the simulation domain, such as thin films and fine vortical structures. The first part introduces a codimensional smoothed particle hydrodynamics (SPH) method for thin films and bubbles. By constraining particles to a manifold and evolving thickness via lubrication theory, the method captures Marangoni flows and curvature-driven dynamics at a fraction of volumetric simulation costs. This is complemented by a sub-grid interface tracking method that reconstructs fluid regions as explicit polygons within triangle meshes, achieving second-order geometric accuracy and near-machine precision mass conservation through an area-correction advection scheme. The second part presents a GPU-resident adaptive simulation system for large-scale 3D vortical flows. The system combines an adaptive octree data structure with a matrix-free multigrid-preconditioned conjugate gradient (MGPCG) solver. By employing algebraically consistent coarsening and flux-consistent corrections, the solver ensures robust convergence at T-junctions and irregular boundaries where geometric multigrid typically fails. To preserve turbulent detail, a hybrid flow-map transport scheme is extended to the adaptive grid, transporting impulse variables and their gradients via APIC-style transfers. This allows the particle distribution to serve as an inherent refinement indicator, automatically concentrating resolution near active turbulence. The integrated system achieves effective resolutions of 0.5 billion cells on a single GPU, producing turbulent smoke and complex interactions with detail far exceeding uniform-grid simulations at the same memory budget. Ablation studies confirm that adaptive resolution and hybrid transport are both essential for high-fidelity multiscale results.
Degree
thesis:*- Name thesis:degree_name
- Computer Science, PhD
- Department thesis:degree_department
- Interactive Computing
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Wang, Mengdi
- Advisor dc:contributor.advisor
-
- Zhu, Bo
- Committee members dc:contributor.committeemember
-
- Turk, Gregory
- Ha, Sehoon
- Eftychios Sifakis
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81730