Error while using sparsesolver or UMFPACK

General Discussion
1

Hello I am trying to solve a system using the Edge element and the variation form involves complex numbers .
Some lines of the code :
fespace Th( meshName , Edge13d )
…
…
…
Th < complex > [Ex,Ey,Ez] ;
…
…
…
solve EQ([Ex,Ey,Ez],[Vx,Vy,Vz],solver = UMFPACK)
= int3d(Team7)(Nu * Curl(Ex,Ey,Ez)’ * Curl(Vx,Vy,Vz) )
+ int3d(Team7)([Jx, Jy, Jz]’ * [Vx, Vy, Vz])
- int3d(Team7)([fx,fy,fz]’ * [Vx,Vy,Vz])
+ on(19,18,21,23,20,22,Ex =0 ,Ey = 0,Ez =0);//

The output I am getting is :
Error umpfack umfpack_zi_numeric status -1
dosolver UMFPACK complex/int 1 0
Error umfpack_di_solve status -3

I would like to ask what does this message imply .

2

You are solving the indefinite Maxwell equation. Your problem is singular, thus UMFPACK fails.

  1. are you sure you meant int3d(Team7)([Jx, Jy, Jz]' * [Vx, Vy, Vz]) and not int3d(Team7)([Ex, Ey, Ez]' * [Vx, Vy, Vz])?
  2. if yes, use a better direct solver, like MUMPS, using load "MUMPS"
  3. you can also use AMS, it’s only currently interfaced for lowest-order Nédélec FE right now but I can add higher orders as well, cf. maxwell-3d-PETSc.edp.
1 Like
3

Dear Sir/Madam .

I really adore and appreciate your effort for getting back to me.Thank a ton ! !

  1. Yes I was naively wrong with the variation formulation :

The weak form of the time harmonic Electromagnetic wave equation that I want to solve reads:

f * v dx = (mu^-1).(curl E) * (curl v) dx + k^2. (E * v) dx
f = -i * omega * J
k^2 = (i * sigma * omega) - (eps * omega^2)

I wrote the variation form which reads :

varf vEMwaveEqn([Ex,Ey,Ez],[Vx,Vy,Vz])
= int3d(Team7)(Nu * Curl(Ex,Ey,Ez)’ * Curl(Vx,Vy,Vz) )

  • int3d(Team7)([Ex, Ey, Ez]’ * [Vx, Vy, Vz])
  • int3d(Team7)([fx,fy,fz]’ * [Vx,Vy,Vz])
  • on(4,Ex =0 ,Ey = 0,Ez =0);

//…
matrix EmWaveEqn = vEMwaveEqn(FeEdge,FeEdge,solver = GMRES) ;
complex[int] RHS = vEMwaveEqn(0,FeEdge) ;

I solved the values of :
FeEdge [Jx,Jy,Jz]; // Edge elements has 3 components !

[Jx,Jy,Jz] = Grad(u) * sigma; //*(region == 1) ;

FE1 k2 ; // FE1 is P2 element

FeEdge[Ex,Ey,Ez] ; // FeEdge Edge13d element .

After using iterative solvers, or direct solvers my job is getting “Killed” after few minutes in the terminal .
So I am trying to get some possible answers to the problem.

  1. load "MUMPS" I think it is a plugin and I am trying to install along with PETSc and link it to Freefem++
    3. macro def(i)[i, i#y, i#z]// EOM . Can you please clarify the meaning of" # "in Freefem++.

I wanted to write you after I ran your code but I am having some issues with the PETSc installation due to 2 times installation of MPI (I will fix it soon !)
I would be grateful if you could get back to me.

Regards
Saptarshi Mandal
Bergische University of Wuppertal
Germany

4

Having two MPI installations is your way to disaster, PETSc or no PETSc. I’d suggest you make a clean install so that you can have all the tools FreeFEM and PETSc have to offer, see the instructions here. Regarding your question about macro, please read the documentation.

5

Dear Sir
Thank you for the suggestion to use the MUMPS solver.In my above mentioned problem MUMPS worked where as other

direct solvers did not.I am wondering why.

6

Hi, I have the same problem and would like to ask you how to install the direct solver MUMPS and SuperLU on Freefem++. Your reply would be greatly appreciated.