menu_book View Full WaterLily Documentation

OTOM CFD Engine

A high-fidelity Eulerian-Lagrangian hybrid solver based on the WaterLily method.

Interaction

• Pan: Left Mouse Drag
• Zoom: Scroll Wheel
• Spin: Real-time torque interaction

Touch (phone / tablet)

• One finger: orbit (3D), pan (2D), draw in Draw Mode
• Two fingers: pinch to zoom, drag to pan
• Double tap: reset the view
• ✈ button (3D): fly mode — left stick moves, drag to look, ▲▼ rise/sink

Click a scene to load a complete saved setup (settings + drawn obstacle picture).

Train a neural surrogate to replace the physics solver for fast inference. Trained models are good only inside their training distribution — out-of-range inputs can give physically incorrect results.

Analytical & empirical reference values for the active shape, compared against the live simulation. References are textbook correlations valid only for canonical geometry — custom drawn/STL shapes have no closed-form solution.

Llama 3.2 1B — fastest (~0.9 GB) Gemma 2 2B — balanced (~1.4 GB) DeepSeek-R1 Distill 7B — reasoning (~4.3 GB, needs ≥6 GB VRAM) Qwen2.5 3B — stronger (~2.2 GB) Llama 3.2 3B — stronger (~2.2 GB) auto_awesome Explain this scene (local AI)

Hardware test: runs a set of 2D and 3D simulations back to back and reports the real, uncapped frame rate and solver throughput your GPU reaches. Your scene is restored when it finishes.