[MBDyn-users] Time calculation using KLU

masarati at aero.polimi.it masarati at aero.polimi.it
Wed Jul 21 14:51:19 CEST 2010


>
> Hi,
>
> ..is KLU automatically recognized by MBDyn's build script ?

Yes.

> ..what would be the exact syntax to use the KLU solver ?

linear solver: klu;

> ..does KLU belong to the group of linear solvers (I guess as you mentioned
> it in this context) ?

Yes.

> ..would you recommend to use KLU as the preferred solver ? if yes, then
> why: for robustness, for speed, anything else ?

I'm not recommending anything right now; the documentation gives some
guidelines and recommends to experiment with the specific problem, as
there is a strong dependence of performances (and occasionally robustness)
on the size, sparsity and conditioning of the matrix.  What I (we)
observed is that it is usually faster than Umfpack, and comparable (for
relatively large problems) with Naive in terms of performances.  This is
consistent with the findings of other researchers

Manuel González, Francisco González, Daniel Dopico and Alberto Luaces, "On
the effect of linear algebra implementations in real-time multibody system
dynamics", Computational Mechanics, 41(4), 607-615, 2008
doi:10.1007/s00466-007-0218-2

> I could not find KLU in the list of supported solvers when browsing the
> input documentation at
> http://www.aero.polimi.it/masarati/MBDyn-input/mbdyn-input-1.X-Devel/node26.html#SECTION008111400000000000000

That documentation is not updated.

Cheers, p.



More information about the MBDyn-users mailing list