Problem with large effective mass and SI convergence

[jimali]

Hi,

i have some trouble getting my SI to converge fast enough when two bodies with large masses collide on both sides of a body with a small mass.
The way i see it, when we solve the impulse locally the one large mass don't see the "real" effective mass in the contact eg. the small mass plus the large mass. It only sees the small mass which results in a small change in the large bodys velocity.
I've looked a bit on how Eran Guendelman use shock propagation but it seems that this is only usefull when one of the colliding bodies are resting on a fixed object (infinite effective mass). Instead i'm thinking of calculating the "real" effective mass seen from the contact point and use this to make some form of shock propagation every n'th iteration in the SI.
I'm not sure wether this is even posible so if anybody has comments or knows how to solve this problem please reply.

cheers,
Jimmy

[Dirk Gregorius]

This is a known limitation of the iterative solver. Basically you should make sure that your mass ratios doesn't exceed 1 : 10 (as a rule of thumb).

[jimali]

Hi Dirk,

thx for the quick reply. Reducing the mass ratio is not an option in my simulation.

Do you think it is posible to make colliding objects sense the effective mass of other nearby colliding objects? and if it is posible, should this not accelerate the convergence in large mass ratio collisions?

[Erin Catto]

You can use a conjugate gradient solver (Google GPCG).

You can also form the associated LCP problem and use a pivoting solver like Lemke's method.

[Dirk Gregorius]

I thing the ODE has PCG method. So you might look at this as reference.

[Antonio Martini]

http://siconos.gforge.inria.fr/index.html

Siconos seems to have many lcp solvers:

http://siconos.gforge.inria.fr/Numerics ... be1847b8c0
http://siconos.gforge.inria.fr/Numerics ... ource.html

papers you may find interesting
http://mtrenouf.googlepages.com/Renouf.Alart2004.pdf
http://mtrenouf.googlepages.com/Renouf.Acary.ea2005.pdf

cheers,
Antonio

[jimali]

thx guys,

i will read up on the links and references..