Warning: Linear solver failed. Reverting to built-in solver.

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Warning: Linear solver failed. Reverting to built-in solver.

randress
With the size of the model (see below), I was not surprised at the long solve time.  However, I did not understand the message "Linear solver failed. Reverting to build-in solver".

Here are the solver settings. I did not intentionally change from default.  Perhaps I should have :-)   I note there are several solvers to choose from. Which one is built-in?  

Also, I note that the initial conditions specify "Expression" rather than "Computed solution".  Was this my mistake?



I have included the Command log and the model Matlab and script files.

-Randal

======== begin Command Log ========
Grid Statistics:
 
Grid Generation Done.
       number of grid points: 171780
        number of grid cells: 981743
        grid cell min volume: 0.0012
       grid cell mean volume: 0.0081
        grid cell max volume: 0.0238
       grid cell min quality: 0.2459
      grid cell mean quality: 0.7938
        number of boundaries: 14
        number of subdomains: 2
    time for grid generation: 256.8838
 
 
--------------------------------------------------
          Solver Statistics (solvestat)
--------------------------------------------------
 Number of dependent variables
--------------------------------------------------
 V : 171780
--------------------------------------------------
 Simulation timings
--------------------------------------------------
 t_asm(A) :          3.1s \   0%
 t_asm(f) :          0.0s \   0%
 t_sparse :          6.6s \   0%
 t_bdr    :          1.9s \   0%
 t_lc/mv  :          0.0s \   0%
 t_solve  :       2424.4s \  97%
 t_tot    :       2489.2        
--------------------------------------------------
 
 [Warning: Linear solver  failed. Reverting to built-in solver.]
======== end Command Log =========

box20x20x20m&sphere1m-solved.fes

box20x20x20m&sphere1m-solved.m
Reply | Threaded
Open this post in threaded view
|

Re: Warning: Linear solver failed. Reverting to built-in solver.

Precise Simulation
Administrator
I could run it without any issue (although it took a while as the mesh has over a million cells) so the error might be due to your system running out of memory causing the selected linear sparse matrix solver to fail (mumps by default), the program then can try again with the built-in solver (Matlab's backslash which most often is running Tim Davis's SuiteSparse/Umfpack).


randress wrote
With the size of the model (see below), I was not surprised at the long solve time.  However, I did not understand the message "Linear solver failed. Reverting to build-in solver".

Here are the solver settings. I did not intentionally change from default.  Perhaps I should have :-)   I note there are several solvers to choose from. Which one is built-in?  

Also, I note that the initial conditions specify "Expression" rather than "Computed solution".  Was this my mistake?

I have included the Command log and the model Matlab and script files.
 [Warning: Linear solver  failed. Reverting to built-in solver.]
======== end Command Log =========

box20x20x20m&sphere1m-solved.fes

box20x20x20m&sphere1m-solved.m