Ryan Goldade has received the 2021 Alain Fournier Dissertation Award for his thesis titled “Efficient liquid animation: New discretizations for spatially adaptive liquid viscosity and reduced-model two-phase bubbles and inviscid liquids.”
This national award, conferred by the Canadian Human Computer Communication Society, recognizes an outstanding doctoral dissertation in computer graphics completed at a Canadian university. The annual award is named in honour of Alain Fournier, a researcher who promoted excellence in the theory and application of computer graphics.
Ryan will receive a plaque, an announcement as this year’s dissertation awardee in the Graphics Interface proceedings, and a prize of $1,000. He has also been invited to present his research at Graphics Interface 2022.
Ryan was a PhD student in the Cheriton School of Computer Science’s Computer Graphics Lab from 2014 to 2021, and advised by Professor Christopher Batty.
“I was lucky to have Ryan as the first PhD student to join my group at Waterloo,” Professor Batty said. “His commitment to excellence and infectious enthusiasm for research made him a pleasure to advise. The fact that his research was published in two papers at SIGGRAPH, the top academic computer graphics conference, and incorporated into SideFX’s award-winning visual effects software, Houdini, speaks to the quality and impact of his contributions. I am thrilled to see him receive this well-deserved recognition.”
Ryan is currently a research scientist at Meta Reality Labs in Zurich, working at the intersection of simulation, graphics, computer vision, and machine learning.
Adaptive Viscous Liquids, presented at SIGGRAPH 2019
Ryan’s doctoral research addressed several challenges in computer animation of fluids. He first explored simulating highly viscous liquids efficiently, a fundamental phenomenon that arises in many animation applications from simulating honey coiling and chocolate melting to paint dripping and lava flowing.
Ryan’s experience as a part-time 3D developer at SideFX Software revealed that the viscous liquid tools in Houdini were underused by animators because they were glacially slow. He accelerated them by adopting a natural approach — use of spatially adaptive octree grids. Because the interior of viscous liquids tends to be smooth, the domain can be discretized using large grid cells on the interior for efficiency and fine cells near the surface to capture intricate details.
The so-called T-junction structures occurring at resolution changes in octrees make deriving a computationally efficient and mathematically convergent formulation highly nontrivial. Ryan’s work showed that instead of using standard finite difference or finite element methods, expressing the partial differential equations for viscosity in an octree-based variational finite difference form leads to a simulator with a range of practical benefits, most importantly that the linear solve times are reduced by up to an order of magnitude.
Ryan’s method was shipped as a key new feature in Houdini v18.5. The underlying variational octree finite difference technique he introduced is highly generalizable, and could potentially be applied to enable efficient octree adaptivity for a wide range of solvers, including solid elasticity, granular flows, stream function solvers, Stokes flows, and more.
Efficient Bubbles and Two-Phase Flows, presented at SIGGRAPH 2020
The second efficiency challenge Ryan tackled is that of air bubbles and two-phase flows. These effects are often even more prevalent than viscous liquids, as simply pouring a glass of water leads to many air bubbles being captured. The classic approach to such multiphase phenomena is to simulate both the liquid and air as individual fluids having different densities and interacting across the liquid surface.
Unfortunately, this approach can be extremely expensive because the air region is often large, and the 1000:1 density ratio of liquid to air leads to numerically ill-conditioned problems. Moreover, the air is not the material the animator ultimately renders and displays. Its effects are seen indirectly through the motion of the liquid.
In Ryan’s dissertation, the bubble effect is achieved efficiently by observing that the key visual feature of air bubbles is that they preserve their net volume. The key insight of Ryan’s work is that this property can instead be expressed as a constraint on the liquid surfaces enclosing each bubble, without ever explicitly solving for the air region. Doing so adds little overhead to a standard liquid simulator, meaning that artists can now add bubble effects to their scenes essentially for free. This feature shipped as “air incompressibility” in Houdini v16.5.