13-10-2026, 04:39 PM
(This post was last modified: 13-10-2026, 10:40 PM by josemendez.)
(12-10-2026, 03:56 PM)Qriva0 Wrote: Yes, for some reason I missed that link xD
I checked the post and this is a lot of stuff done! It makes sense why it took so long.
Got just one question, is multigrid solver result similar to tether constraits, worse or better?
Overall is the final result almost the same or it introduces completely different problems, plus what is their relation to performence?
I am corious if it solves the same constraint but with bigger gaps between particles or this is something else and is that for all particles or I can select what constraints are solved this way.
Hi!
Multigrid results are better than tethers’. It’s loosely based on HPBD (hierarchical position based dynamics) but in addition to a prolongation pass as in the original paper, it also has a restriction pass for contacts, pin constraints, and some others.
The basic idea is to solve the same constraints with bigger gaps between particles and interpolate the solution to particles in-between (this is called prolongation, in multigrid jargon). However this doesn’t work well if there’s other constraints affecting individual particles, say collisions or attachments, since the coarser levels act as if they didn’t exist: the corrections made by these must be applied to particles in coarser levels as well (restriction) so that they work as intended.
The cost of multigrid is roughly 1 constraint iteration per level. Usually, 3-4 levels are enough and yield results comparable to 40-80 substeps in terms of rigidness. Right now, multigrid can be enabled for distance and stretch/shear constraints on a per-actor basis, and they work with collisions, particle collisions, pins, stitches and pinholes.
Let me know if you want further details,
Kind regards