Hi We are planning a new RPG, which intruduces some inprecedented gameplay features. One among them is using physically correct water. It's able to flow, interact with other object, etc. This allows to do quite interesting things: fill pits, then turn water into ice and move on it; or fill pits during rain; or make a whole in wall with strong water flow; and many more... But it requires physically based water simulation. So, have anybody tried to implement such thing, or it's possible, that someone have thought about it or has interesting papers, links? What can you recommend, suggest? Any information on topic would be useful. PS: I hope, that my poor English doesn't hide my thoughts ;)

That's quite the task you have ahead of you.

A quick google for particle based fluid simulation yields this paper (.pdf), which looks decent.

This paper (again, a .pdf) may also be useful.

ok, i only know one concept:

the basic idea is, to divide the water into cells, each cell has some attributes like pressure. so every cell looks in their neighbourhood, what pressure they have. so high pressure move to low pressure and so on.

most games only simulate the graphic of water (approximations), so what you are planning needs a lot of cpu power.
(thats why fluid dynamics are often calculatet on clusters)

You have three methods:

1. Full 3D simulation using grid methods. This will allow for splashing etc, but is difficult to get robust (e.g. just conserving fluid quantity) and likely to be slow.

2. Full 3D sumulation using particle methods. You can trade off quality with performance with this, and will handle splashing etc. It's slow for large volumes (you could probably handle 2000-10000 particles, depending on your CPU budgets - remember the more particles the more rendering time too).

3. 2D surface (heightfield) simulation. Really cheap (in comparison to the 3D methods above) and suitable for medium-large expanses of water. Also relatively easy to render. However, it won't handle splashing/pouring etc. You can add some visual effects with very simple (non-physical) particle effects on the surface.

My inclination would be to experiment with a combination of 2 and 3. E.g. if you have a buck of water and an empty bath - initially the water in the bucket is simulated using a heightfield method (quite tricky, since the boundaries/depth will change as the bucket moves, but doable). Then when the bucket is tipped enough, particles get spawned to simulate the flowing water, and the parameters of the heightfield water in the bucket get changed (to conserve mass). Eventually the bucket becomes empty, and that heightfield simulation gets deleted. When the flowing water hits the bath this enables a heightfield simulation there, and somehow(!) the particles convert to the heightfield parameters - to conserve mass. The particles hitting the bath/water surface will stay as particles until the stop splashing.

Rendering the flowing/particle fluid could be done using a surface reconstruction (marching cubes) or maybe a billboard approach would look good enough.

I think that the old PS3 ducks in a bath demo from a year or so ago (actually, where they used eye-toy or something to scoop water up from a tub into a small container, and pour it out again, IIRC) probably used this combined method - but that's just a guess.

Edit: I have a 2D SPH demo, including source code. Also, the Ageia SDK supports liquid simulation (including interaction with rigid bodies), and maybe that would be fast enough, if your game required their PPU board...