Hi all. I know that this paper ( http://www.matthiasmueller.info/
) has been around a bit, and has been discused on this forum previously but i have a few queries about it and im hoping some of you may be able to shed some light.
(This question may render the other questions invalid)
1) There has been mention on this forum that the original paper released had some errors in but they were corrected in a more recent release of the paper. Does anyone know what these errors were? So i know i am using the paper with the correct maths in it.
2) Has anyone else who has implimented this paper had issues where the angular momentum is NOT conserved during the solving of the constraints? My tests show that the linear momentum is conserved to near perfection but the angular momentum often is not.
Take for example the pressure constraint... If you consider a unit cube (8 vertices, 6 quad faces which break into 12 tris); If pressure is applied to the cube you would expect the vertices to all move away in a direction the same as if you cast a ray from the centre of the cube to the particle positions. You would therefore expect the direction of the gradient to be in the same direction as the vertex normal of particle, however if you consider that the gradient is built up from the faces that each particle is connected to, the gradient is actually pulled towards the faces of the cube who contribute the most. In my tests i have found that this causes angular momentum to be added when the overpressure changes. Visually and mathematically i can see why this occurs, also if each particle was connected to the same number of tri faces then i believe this problem would not arise. However, the paper does not seem to mention this regularity in faces and particle connectivity as a requirement. Has anyone else had similar results or can see the error in my results?
3) I noticed that Bullet supports pressure in its soft body system through forces applied to each particle, is the reason for this at all related to the problems i have had in question 2? (The constraint based method offered in the paper would seem to provide a stiffer response if required)
Thanks in advance.....
P.S The paper is a great paper for anyone has not read it. Like all white papers though, a companion paper would be handy that explains the problems and *missing* bits