Page 1 of 1

Adaptive Grid for simulations

Posted: Mon May 23, 2016 10:06 pm
by TimTruett
If you want to do a 3D Polywell simulation, then the video at the link below provides some hints about how to do it. The video explains the use of adaptive grids. Where detail is not important, you can use a large grid size. Where small detail is important, use a small grid.

The adaptive grid changes where the small grid steps are and where the large grid steps are at each time step in order to follow what is happening in the simulation. (That's what makes it "adaptive.")

The video is about Computation Fluid Dynamics (CFD) and combustion in rocket engines, but I think that the ideas are applicable to other problems also.

https://www.youtube.com/watch?v=vYA0f6R5KAI

Re: Adaptive Grid for simulations

Posted: Sun Oct 01, 2017 4:19 pm
by happyjack27
also similar are the fast-multipole method

https://en.wikipedia.org/wiki/Fast_multipole_method

and a variation i wrote based on oct-trees ( https://en.wikipedia.org/wiki/Octree ) with dynamic depth.

https://sourceforge.net/projects/octree ... particles/

they all attempt to reduce the o(n x n) all pairs problem to an approximately o(n x ln(m)) problem (where m is precision)


an unsolved problem with doing this on a massive scale is i/o bandwidth. if you have many computers working on it at once, you need to be able to break up the problem into pieces that are as independant as possible, to minimize i/o between them.

that's part of what i was hoping to move towards with my oct-tree method.