[Paper Review] Efficient Optimal Control of Smoke using Spacetime Multigrid

Zherong Pan, Dinesh Manocha|arXiv (Cornell University)|Aug 3, 2016

Computer Graphics and Visualization Techniques20 references3 citations

TL;DR

This paper presents a spacetime multigrid optimization method for efficient keyframe-based control of 2D and 3D smoke simulations. By solving the Karush-Kuhn-Tucker (KKT) conditions of a PDE-constrained optimization problem using a novel spacetime full approximation scheme (STFAS), the method reduces computation time by over an order of magnitude compared to prior gradient-based approaches, enabling high-resolution control of long fluid animations on standard desktop hardware.

ABSTRACT

We present a novel algorithm to control the physically-based animation of smoke. Given a set of keyframe smoke shapes, we compute a dense sequence of control force fields that can drive the smoke shape to match several keyframes at certain time instances. Our approach formulates this control problem as a PDE constrained spacetime optimization and computes locally optimal control forces as the stationary point of the Karush-Kuhn-Tucker conditions. In order to reduce the high complexity of multiple passes of fluid resimulation, we utilize the coherence between consecutive fluid simulation passes and update our solution using a novel spacetime full approximation scheme (STFAS). We demonstrate the benefits of our approach by computing accurate solutions on 2D and 3D benchmarks. In practice, we observe more than an order of magnitude improvement over prior methods.

Motivation & Objective

• To address the high computational cost of keyframe-based optimal control in 2D and 3D smoke simulations, where control force sequences span high-dimensional spaces.
• To reduce the need for repeated fluid simulations in iterative optimization by exploiting temporal and spatial coherence in the solution process.
• To develop a scalable, multigrid-based solver that maintains both primal and dual variables to avoid solving Navier-Stokes equations from scratch in each iteration.
• To enable accurate, smooth, and physically plausible smoke animations that match user-specified keyframe shapes at designated timesteps.
• To provide a framework that balances keyframe fidelity and control force magnitude through a regularization parameter, supporting artistic and special effects applications.

Proposed method

• Formulates smoke control as a PDE-constrained spacetime optimization problem, minimizing both keyframe deviation and control force magnitude.
• Solves the optimization by computing the stationary point of the Karush-Kuhn-Tucker (KKT) conditions, maintaining both primal (velocity) and dual (Lagrangian multiplier) variables.
• Introduces a spacetime full approximation scheme (STFAS), a nonlinear multigrid solver that uses a novel spacetime smoothing operator for efficient convergence.
• Employs a fixed-point iteration over the KKT system, enabling convergence independent of grid resolution and number of timesteps.
• Uses trilinear prolongation and restriction operators for multigrid transfer, compatible with existing boundary-aware multigrid techniques.
• Leverages the structure of the Navier-Stokes equations on staggered grids to compute derivatives of ghost forces with respect to velocity fields efficiently.

Experimental results

Research questions

• RQ1 Can a spacetime multigrid method significantly reduce the computational cost of optimal control in 2D and 3D smoke simulations?
• RQ2 How can the KKT conditions of a PDE-constrained optimization problem be solved efficiently without repeated fluid simulations?
• RQ3 Can a full approximation scheme (FAS) be adapted to spacetime problems to achieve resolution-independent convergence?
• RQ4 To what extent can maintaining dual variables (Lagrangian multipliers) reduce the number of fluid simulations required per optimization iteration?
• RQ5 How does the proposed method compare in speed and accuracy to gradient-based optimizers that require full fluid solves per iteration?

Key findings

• The proposed spacetime multigrid method achieves over a 10x speedup compared to prior gradient-based optimal control methods on standard desktop hardware.
• The method reduces optimization time to approximately 30 minutes for a 128² × 40 timestep 2D smoke animation, achieving 17× speedup over conventional optimizers.
• Convergence of the STFAS solver is independent of grid resolution and number of timesteps, enabling scalability to high-resolution and long-duration simulations.
• The method maintains high accuracy in matching keyframe shapes while producing smooth, physically plausible control forces with minimal magnitude.
• The approach supports user control over the trade-off between keyframe fidelity and control force energy via a regularization parameter.
• The framework is extensible to include force templates (e.g., vortex forces) and can be accelerated via parallelization or receding horizon control strategies.