Research map / Probabilistic and scientific modeling / Dynamical systems
Guaranteed Conservation of Momentum for Learning Particle-based Fluid Dynamics
Lukas Prantl, Benjamin Ummenhofer, Vladlen Koltun, Nils Thuerey
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the antisymmetric output layer (ablation) | Base network · WBC-SPH | velocity-distribution accuracy | Improved |
| add the antisymmetric output layer (ablation) | Base network · WBC-SPH | change in momentum: 0.0 -> 1.0 | Improved |
| use multi-scale processing with farthest-point sampling (ablation) | ASCC · WBC-SPH | overall performance | Mixed |
| replace farthest-point sampling with voxelization (ablation) | Multi Scale · WBC-SPH | resource usage | Improved |
| add temporal preprocessing rollouts (ablation) | Voxelize · WBC-SPH | average relative performance: 0.94 -> 1.0 | Improved |
| add input rotation normalization and enhanced boundary processing (ablation) | Preprocess · WBC-SPH | average relative performance: 0.94 -> 1.0 | Improved |
| replace ASCC with regular CConv (ablation) | Ours · Liquid column, Free Fall | RMSE: 19.68 -> 46.36843 | Hurt |
| train with a 20-step rather than 5-step rollout (ablation) | Ours · WaterRamps | EMD: 0.09155 -> 0.06156 | Improved |
| use the proposed method instead of GNS | WaterRamps · WaterRamps | EMD: 24% | Improved |
| use the proposed method instead of GNS | Ours · Droplet collisions, default gravity | EMD: 57% | Improved |
| change gravity to zero at evaluation | GNS · Droplet collisions, zero gravity | EMD: 195% larger error than our method | Hurt |
| train with the momentum-conserving constraint (ablation) | ASCC · WBC-SPH sample-efficiency evaluation | EMD: ca. 6% ... more than 20% | Improved |
Closest papers by meaning
- Space and time continuous physics simulation from partial observations (ICLR 2024 · similarity 0.87)
- Learning 3D Particle-based Simulators from RGB-D Videos (ICLR 2024 · similarity 0.87)
- PARCv2: Physics-aware Recurrent Convolutional Neural Networks for Spatiotemporal Dynamics Modeling (ICML 2024 · similarity 0.86)
- Latent Intuitive Physics: Learning to Transfer Hidden Physics from A 3D Video (ICLR 2024 · similarity 0.86)
- Keep the Momentum: Conservation Laws beyond Euclidean Gradient Flows (ICML 2024 · similarity 0.86)
- Symmetric Basis Convolutions for Learning Lagrangian Fluid Mechanics (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
