Thuerey Group

In addition, the new release supports surface tension forces, e.g., for simulating small droplets, and a viscosity solve for thicker materials and physically more accurate simulations. The fast multigrid solver is another highlight. It allows for efficient calculations of large-scale effects.

Here’s an incomplete feature list:

• multigrid-preconditioned solver
• Eulerian simulation using MAC Grids, PCG pressure solver and MacCormack advection
• Flexible particle systems
• FLIP simulations for liquids
• Surface mesh tracking
• Free surface simulations with levelsets, fast marching
• Wavelet and surface turbulence
• K-epsilon turbulence modeling and synthesis
• Maya and Blender export for rendering
• tensorflow couping via numpy arrays

Btw., mantaflow has been used in numerous publications! Among others:

• Data-Driven Synthesis of Smoke Flows with Convolutional-Neural-Network-based Feature Descriptors, SIGGRAPH 2017
• Interpolations of Smoke and Liquid Simulation, Transactions on Graphics 2016
• Narrow-band FLIP, Eurographics 2016
• Surface Turbulence for Particle-Based Liquid Simulations, SIGGRAPH Asia 2015
• Connecting Forward and Inverse Problems in Fluids, SIGGRAPH 2014
• Liquid Surface Tracking with Error Compensation, SIGGRAPH 2013
• Turbulent fluids: Course, SIGGRAPH 2013
• Lagrangian Vortex Sheets for Fluid Animation, SIGGRAPH 2012
• Physics-Inspired Topology Changes for Thin Fluid Features, SIGGRAPH 2010
• A Multiscale Approach to Mesh-based Surface Tension flows, SIGGRAPH 2010

Fig. 1: A few images of a controlled smoke simulation using the PD-guiding feature of mantaflow.

We just added two new publications. One of them focuses on robust decompositions of fluid flows into vortex filaments, and it will be presented at SCA 2017 in Los Angeles soon. You can find all detail on the accompanying page, and you can see a preview below.

The other work targets generative neural networks (full details here). In this case the network actually generates a dense space-time deformation field to capture spaces of liquid behavior. We have also created a proof-of-concept Android app (Lukas Prandtl was the main driving force behind this one), which you can download to try it out yourself. The app evaluates the full trained neural network every time you tap to synthesize the corresponding deformation field. This is especially tough in our setting, as the network employs 4D deconvolution layers that need to be evaluated on the mobile device.