[Bf-committers] Patch: Adaptive time step for fluid particles

[Alex Fraser]

----- Original Message -----
> From: "Shaul Kedem" <shaul.kedem at gmail.com>
> No, I mean is it deterministic?

>From one run to the next, yes, you should get the same result.

In terms of being predictable, I did some tests with the TwoPipe demo, in which water drains through a hole (you may need to increase the fluid interaction radius to see this effect). With zero subframes, it takes a long time to drain; with more subframes, it tends to take less time (time in frames, not computation time). However, it starts to converge for subframes > 3 or a Courant target < 0.2, and barely changes beyond subframes = 8 or Courant target = 0.1. So it seems to become more accurate with smaller time steps, but you should rarely need to go beyond those values.

The adaptive subframe code gives a stronger guarantee that your simulation will be stable, but it's not predictive, so for some simulations a constant time step is still more appropriate.

Cheers,
Alex

-- 
Alex Fraser
Software Engineer
The Victorian Partnership for Advanced Computing