Libo Huang, Torsten Hädrich, and Dominik L. Michels have recently presented their research paper for SIGGRAPH 2019, and Libo Huang told us about ferrofluids and their approach to their accurate simulation.
I am Libo Huang, currently a 3rd-year Ph.D. student studying at KAUST in Saudi Arabia. I am originally from China. My current study focuses on efficient and accurate simulation related to computer graphics.
A ferrofluid is a liquid but at the same time magnetizable just like a piece of iron. It was originally produced by NASA scientist so that they can control the liquid using a magnetic field in low-gravity environments. Nowadays it is more popular for its intriguing geometry interacting with a magnetic field. Hence it is used in science exhibition (quite a hot topic for youtube videos). Despite many existing efforts to approximate its geometry and dynamic motions in VFX, I have not yet seen any methods that capture the merging and splitting behavior of spikes of the ferrofluid.
As we know, in VFX many physics rules can be broken to achieve desired effects. In contrast, the strange behavior of ferrofluid is totally physical but difficult to generate in VFX software.
There are several key behaviors of ferrofluid that define its realism:
- Cone-like spikes which existing VFX techniques have achieved
- Spikes repelling each-other which is also achieved by procedural modeling. See:
- Realistic merging and splitting of spikes, currently not seen
- Realistic interaction with a magnetic body. For example the complex behavior here:
Thus, we can restrict a simulation to those purely based on physics principles instead of ad-hoc procedural modeling techniques. The above four points are challenging because we cannot directly set rules on when these spikes form, what their radius is, when they merge, split, etc. We must first establish a physically correct model and find efficient and accurate techniques to convert the equations from the model to computer programs.
There have been successful examples for the static shape of ferrofluid in a simple magnetic environment, they use Finite Element Method (FEM) for all the simulation which is inefficient for handling large dynamic deformation when ferrofluid moves. We augment existing Smooth Particle Hydrodynamics (SPH) method with magnetic forces to simulate the liquid.
Simulation Techniques for Ferrofluids
Previous techniques to simulate ferrofluid shapes can be categorized into two types:
- Methods using meshes and finite element method
- Methods using particles with additional procedural modeling
The first method only gives the final static shapes of a simple magnetic field (e.g. constant), and the evolution of liquid requires lots of re-meshing.
The second method can produce only a bulge instead of lots of spikes in previous simulations.
Our contribution is the new accurate, efficient, robust calculation of magnetic forces that distinguish ferrofluids from normal liquids. We combine this magnetic force into existing SPH fluid simulation with surface tension to get the successful results.
When the ferrofluid is placed in a magnetic field, the fluid particles have additional magnetic forces between them. This is what distinguishes a ferrofluid from a normal fluid. Naive handling of the magnetic force would result in infinitely large forces when two fluid particles are close enough. Our approach utilizes a smooth magnet so that the forces do not go to infinity.
Influence of Different Physical Conditions
The reason why we show the influence of different surface tension and the field strength is to show how these two factors influence the appearance of the ferrofluid spikes. We have no prior knowledge about what shapes it will form at all. We simply let the simulation do the evolution job. The crazy shapes are formed naturally according to the interplay between different physics laws, we cannot directly manipulate the shapes but to tune the physics quantity.
We are the first to build a successful dynamic ferrofluid simulation system purely based on physics laws. The simulator tries to reproduce how the true ferrofluid behaves in reality, given the magnetic field, surface tension, etc. If there is some impossible scene setup in reality that evolves ferrofluid, our system can do that in the virtual world.
However, we must admit that the current version is quite computationally expensive, and the level of details compared to the real video sequence are less abundant. In simple words, our simulation looks cool but not as cool as the true ferrofluid. I believe to achieve a cool visual effect, there are tons of other techniques which do not necessarily have to be purely based on physics principles.
We hope after some refinement, our method can be utilized in the industry during the production of VFX in a typical scene that emphasizes exotic/ alien-like feeling. Additionally, this approach may be used to help design the ferrofluid sculpture.