>
> No, the failure was the n_points() call in assemble_ldm().  The
> quadrature rule in that function isn't getting initialized properly.
Thanks, I will check that part, maybe some typos there lead to the error.
> This is a libstdc++ generated error; to get a stack trace there the
> best way is to run from within gdb and type "where" after it crashes.
>
OK. let me try the gdb mode and find out the problem.
>
>
> Yes, for one element.  A dense matrix for the whole problem would have
> 1000000x1000000 entries and would require around 8,000GB of memory.
>
Ohh, this will be very expensive, there are totally three sparse matrix 
including
the M, C, K, and one sparse vector F. If the desktop has 8G memory, the 
each
matrix should be less than (8G/4)/8= 25e7 components, which means 15811 
nodes
totally and 25 nodes per dimension. So the mesh density is limited at small
scale. For the coupling system involving more variables, the model would be
smaller, it makes me a little bit frustrated.

>
> Yes, but difficult.  I wouldn't attempt it yet.  Using ex8 as a
> template for your own application isn't exactly easy, but it's a lot
> easier than using newmark_system.* as a template for your own
> integrator would be.
Yes, I believe so. Is that possible to develop a 2nd order time solver 
depending
on the Densematrix? Should any parallelism or
adaptive issue be considered? The time solver system may inherit
LinearImplicitSystem or NewmarkSystem class. Beside that, any other source
codes should be reviewed such as PETSc, System class, ImplicitSystem,
Explicit System, etc. Or any reference would you recommend to extend the
LibMesh at time solver.

Cheers!

Liang



------------------------------------------------------------------------------
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to