Solvers & Methods used in the softwares available in the market

#1

The thread is already in development here in here:
http://forums.cgsociety.org/showthread.php?p=6278949#post6278949

There are several competing techniques for liquid simulation with a variety of trade-offs. The most common are Eulerian grid-based methods, smoothed particle hydrodynamics methods, vorticity-based methods, and Lattice Boltzmann methods. These methods originated in the computational fluid dynamics community, and have steadily been adopted by graphics practitioners over the past decade.

In computer graphics, the earliest attempts to solve the Navier-Stokes equations in full 3D came in 1996, by Nick Foster and Dimitris Metaxas. In 1999, Jos Stam published the so-called Stable Fluids method at SIGGRAPH, which exploited a semi-Lagrangian advection technique to provide unconditionally stable behaviour. This allowed for much larger time steps and in general, faster simulations. Blender 3d, uses a stable Lattice Boltzmann method implemented.

Maya uses which method exactly? I know the bases is Navier Stokes. In the market we have these softwares: FlowLine VFX, 3D Aliens, SitniSati, Katachi, Real Flow, Houdini, ILM Zeno and Softimage ICE.

Can someone tell me something about these too? I’m working in solving this question too. I´ll post my studies here when I found some results.
Cheers

#2

Sorry,I’m confused about your question/s.
Are you enquiring what methods Maya fluids use in a Max forum?

#3

There are also hybrid methods - check out the ILM paper on fire from the last siggraph for example.

#4

Sorry,I’m confused about your question/s. Are you enquiring what methods Maya fluids use in a Max forum?

It’s a generalist question. Not Maya only but others like the ones I mention.
I allready answer the question about maya. It uses Navier Stokes but i would like to know if it uses a Eulerian or Lagrangian method. I wanted to know what FumesFx for example uses? Navier Stokes? Eulerian?

There are also hybrid methods - check out the ILM paper on fire from the last siggraph for example.

Do you know the name of the paper?
Is it this: “Directable. High Resolution Simulation og Fire on the GPU”
I haven’t read it but i will.

#5

Assuming Stam’s paper is patented by Alias, Maya’s fluids seems to be a primary implementation of the paper you mentioned earlier (along with dozens of additional stuff of course). Realflow is a SPH simulator. Houdini has two (three actually) solvers: Level set (NS/Eulerian), fluid particles ( SPH) , and additionally semi Eulerian/Lagrangian solver (Sand Solver), which is kind of partly implemented. These are of course rough assumptions based on what you see outside. I’ve never touched Fumefx, but I bet it’s a NS/Eulerian solver judging from a results.

ps Softimage ICE is not a solver, but architecture and environment, there are some solvers avaiable for it, but its design varies (I saw sph one).

#6

Yep, that’s the one.

#7

Thanks a lot SYmek

#8

not sure if this is that relevant to ya, you might have stumbled apon this already…

Masters Thesis:Physical Simulation of Fire and Smoke
David Minor 2007

I belive it say FumeFx uses Navier-Stokes,also talks about some Houdini stuff at the end

#9

Thanks a lot Alexander. This is great. Do you know other papers that can be useful? Or some master thesis like this?

This is a link to my specific research: http://www.vitorteixeira.com/blog/university/dissertation/specific-research/

and my generalist research:
http://www.vitorteixeira.com/blog/university/dissertation/research/

Hope you find this useful to you too.
Cheers

#10

http://nccastaff.bournemouth.ac.uk/jmacey/MastersProjects/MSc09/Claes/index.html

Another one which can give you good insight into Houdini’s implementation.

#11

Thanks a lot

#12

After lots and lots of reading I still with a doubt.
Can someone make me understand, probably I know the answer but I’m not quite understanding after so many names in my head, why Eulerian (good for fluids, bad for solids) grid based solvers are better for fire and smoke and Lagrangian (bad for fluids, good for solids) mesh free, particle based are better for water ?

What type of solver Fume and after burn uses?
Semi-Lagrangian or Eulerian?

#13

^Afterburn is not a fluids system, just a method for rendering volumetric particles

Fume on the other hand uses ‘Conjugate Gradient’ – grid based… not sure where that falls in the semi-Lagrangian or Eulerian categories

#14

“Numerically, the Lagrangian viewpoint corresponds to a particle system, with or without a mesh connecting up the particles, and the Eulerian viewpoint corresponds to using a fixed grid that doesn’t change in space even as the fluid flows thought it.” - Robert Bridson.

Thanks about afterburn, didn’t knew that.
I searched information about it but could fine any specific detail about it.
For example Maya is a semi-lagrangian, solving the Navier Stokes equations for fluid simulation, that is based on the Stable Fluids by Jos Stam.
Really like your blog, amazing work you have in your portfolio section.
Congrats

#15

Thx Mate! I have a blog? Oh yeah… I really need to update that thing! Btw, I think you’ve somewhat inspired me to do more research into fluids now after reading this thread and the other on in the maya forum. I usually just stick to the front end (on fluids), but now I’m really starting to get interested in behind-the-scenes fluids stuff… so thanks! BTW: don’t forget about houdini… it’s got particle fluids, grid based water fluids, grid based gaseous fluids, and even a sand solver hiding in there (don’t ask me!).

Here’s some more fun fluids stuff I haven’t put up on my site yet:

http://fx-td.com/content/misc/flood_corner.mov < houdini water fun

Needed for a top-down shot:
http://fx-td.com/content/misc/Fireball_v001.MOV <fume fun
http://fx-td.com/content/misc/fireball_v005.mov <fume fun (big file)
http://fx-td.com/content/misc/fireball_v005_side2.mov <fume fun (big file)

#16

Dude that’s awesome! Big-ass grids!

#17

mhmmm tasty lookin fluids SoLiTuDe !
Never used houdini, how long does it take to simulate and render fireballs like those ?

#18

Gord lord, those are, excuse the expression… Tits!

Haha, render quick, sim looks like an overnighter on the office’s badass rig :smiley:

#19

Hey, those fireballs were actually Max/Fume… I think the v001 took like 4 to 5 hours (less frames)… the other one maybe 9 hours or so? Something like that anyway… I should pay more attention to that. I was actually quite limited by ram, I had wanted to push the detail higher… I simmed on a 16 gig machine, and most of the render machines were 8 gigs. The workstation I was using to set up / test was actually faster at simming, but only had 12 gigs of ram, so I put it on a slower machine for the extra ram. :shrug: Ahh well, I’m glad with how it turned out.

P.S. Thanks Guys!

Edit: Oh yeah the water sim in houdini took like 6 hours on my not-for-cg laptop… I’d love to play with it more, but need a better workstation for it. (at home)

#20

Those are great tests definitely.
Continue with the topic.
Check out these two threads I’ve started:
http://www.realflowforum.com/view_topic.php?id=5713
http://area.autodesk.com/blogs/duncan/scene_files_from_tech_talk

I think you are right about Houdini you can find some nice papers in the maya thread.
But unfortunately never try it.
I’m still giving my first steps in Maya fluids and fumefx and realflow :buttrock:

and I’ve been asking some advices to Robert Bridson about is Naiad software, quite impressive.
I’m willing to give it a try.
But I’m developing a lot of information about this in my master degree, I can share very useful information with you and who’s interested on it.
Cheers