Hello, I'm a new user of Elmer.
I want to use "Stress Solver" to find the displacement, but if the number of elements exceeds 4k, the solver does no more converge and displays the following error message:
Error occurred in umf4num: -1.0000000000000000
It would be very nice if somebody could take a look at the case and give some help elements for fixing this problem.
Thanks in advance.
-emB
Linear elasticity
Linear elasticity
- Attachments
-
- case.sif
- (2.44 KiB) Downloaded 54 times
-
- comsol 4.67k.mphtxt
- (696.14 KiB) Downloaded 49 times
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Linear elasticity
That is the error umfpack throws when it runs out of memory. You may have to switch to an indirect method.
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Linear elasticity
I was able to solve it using mumps instead of umfpack, ran out of memory on 12 GB machine using umfpack.
Re: Linear elasticity
Hi kevinarden,
Thank you for the reply! But if I had to use Elmer software in windows, there's any other good way?
I've tried the iterative method, but the convergence takes too long to work.
I wonder why only "StressSolver" should lower the number of element when other solvers are convergence smoothly in more number of element.
-emB
Thank you for the reply! But if I had to use Elmer software in windows, there's any other good way?
I've tried the iterative method, but the convergence takes too long to work.
I wonder why only "StressSolver" should lower the number of element when other solvers are convergence smoothly in more number of element.
-emB
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Linear elasticity
Hi kevinarden,
I'm still triyng to compile MUMPS. I think it's not that easy:(
And i used umfpack on 128GB machine, it was still out of memory. But I don't think it's not enough.
Does Elmer have any options for memory control?
I'm still triyng to compile MUMPS. I think it's not that easy:(
And i used umfpack on 128GB machine, it was still out of memory. But I don't think it's not enough.
Does Elmer have any options for memory control?
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Linear elasticity
I think the problem should fit in 128 Gig. Have you tried banded?
You can use task manage, performance tab to monitor memory usage during the run, to see if it really is a memory error.
You can use task manage, performance tab to monitor memory usage during the run, to see if it really is a memory error.
-
- Posts: 2301
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Linear elasticity
FYI umfpak is an added on module to Elmer it is note developed by the Elmer team
https://en.wikipedia.org/wiki/UMFPACK
https://scicomp.stackexchange.com/quest ... imitations
https://en.wikipedia.org/wiki/UMFPACK
https://scicomp.stackexchange.com/quest ... imitations
-
- Site Admin
- Posts: 4825
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Linear elasticity
Hi,
As Kevin said the memory consumption of direct solvers is beyond the control of Elmer. The finite element matrix topology defines the memory usage. 3D & vector valued problems are most difficult since the fill ratio of the multifrontal methods becomes quite large.
You could try iterative methods, for example BiCGStabl or GCR with block diagonal prec., BILU2 etc. By theory the block preconditioners work often well but there are quite many bells and whistles. You could look at test cases "BlockLinElast3*" for inspiration. There are many iterative methods and just making test with some of them is not conclusive.
In parallel you have additional robust options, like BoomerAMG preconditioner of Hypre package.
What operating system are you running under?
-Peter
As Kevin said the memory consumption of direct solvers is beyond the control of Elmer. The finite element matrix topology defines the memory usage. 3D & vector valued problems are most difficult since the fill ratio of the multifrontal methods becomes quite large.
You could try iterative methods, for example BiCGStabl or GCR with block diagonal prec., BILU2 etc. By theory the block preconditioners work often well but there are quite many bells and whistles. You could look at test cases "BlockLinElast3*" for inspiration. There are many iterative methods and just making test with some of them is not conclusive.
In parallel you have additional robust options, like BoomerAMG preconditioner of Hypre package.
What operating system are you running under?
-Peter
Re: Linear elasticity
Hi,
Thank you for your reply and sorry for my late response.
I tried iterative methods ( BiCGStabl, GCR, Block solver etc), But it didn't converge or got incorrect results.
When I used banded, I didn't get a satisfactory result either.
I'm running under Windows(8GB machine).
Is there any way to run Hyper package under Windows?
p.s. I apologize for my bad English
Thank you for your reply and sorry for my late response.
I tried iterative methods ( BiCGStabl, GCR, Block solver etc), But it didn't converge or got incorrect results.
When I used banded, I didn't get a satisfactory result either.
I'm running under Windows(8GB machine).
Is there any way to run Hyper package under Windows?
p.s. I apologize for my bad English