[ODE] Re: Questions about QuickStep
Nguyen Binh
ngbinh at gmail.com
Wed Aug 25 10:17:24 MST 2004
Hi Cristian!
<snip>
(1) What IS an iterative solver? I thought there were only two
approaches for solving contacts: either you solve them one by one
(which is bad), or you solve them all at once. I don't understand what
is the difference between what you say above at b) and the classic
non-iterative LCP solver.
</snip>
The difference between iterative and classic "big matrix" method is
the difference between, say, LU decomposition method and Gaus-Seidel
iterative method. Basically, on LU decomposition, you have to form the
n*n matrix, which takes O(n*n) time and space but on iterative method,
you just don't need to do so because it only requires vector-vector
multiplication. I suggest you read some book on Linear Algebra for
details on this.
<snip>
(2) I asked the developers of Meqon (www.meqon.com) what approach they
used and they told me their solver is ierative and doesn't use a
"bigmatrix" like the A matrix in the LCP solver. This is totally
confusing to me. They don't solve one at the time, and they don't use
a matrix. How can this be? The demo on their site is amazing, the best
I've seen anywhere. You can see the reply here:
http://meqon.com/forum/viewtopic.php?p=138#138
So, does an iterative solver use a matrix or not?
</snip>
I really don't know what they are doing internally(Sure, if yes I
might setup a firm like them :) ) but I'd guess they use a special
iterative method. I knowed they had developed this algorithm for a
long time, maybe this is their first demo:
http://imp.centercom.se/files/win32/imp-lq.zip
I think they intelligently choose the order of constraints and maybe
they batch some constraints to solve at once,etc...
I really feel impressed by their cloth and vehicle.
Nguyen Binh
www.glassegg.com
----------------------------------------------------------------------------------------------
...currently searching for a graduate school in physical simulation...
----------------------------------------------------------------------------------------------
More information about the ODE
mailing list