Please note that this seminar has been cancelled
Ryan
Goldade,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Liquid simulations for computer animation often avoid simulating the air phase to reduce computational costs and ensure good conditioning of the linear systems required to enforce incompressibility. However, this free surface assumption leads to an inability to realistically treat bubbles: submerged gaps in the liquid are interpreted as empty voids that immediately collapse.
To address this shortcoming, I propose an efficient, practical, and conceptually simple approach to augment free surface flows with negligible density bubbles. This method adds a new constraint to each disconnected air region that guarantees zero net flux across its entire surface, and requires neither simulating both phases. Implementation of the method requires only minor modifications to the pressure solve of a standard grid-based fluid solver, and yields linear systems that remain sparse and symmetric positive definite. Additionally, I propose an extension of the surface-based bubble constraint for non-negligible densities where both the proposed constraint and previous monolithic pressure models fail to capture dense bubbles that remain suspended or sink in an immersed liquid. Finally, because the surrounding liquid represents a significant number of degrees-of-freedom in the resulting linear system and is therefore the bottleneck in the residual reduction of the linear solver, I propose an efficient application-specific preconditioner to realize the performance of the simplified bubble models.