[DuMuX] "Matrix for composition of phases could not be solved."

[Dmitry Pavlov]

Hello,

I am a bit puzzled about this error:
[solve:/home/dpavlov/DUMUX/dumux/dumux/material/constraintsolvers/misciblemultiphasecomposition.hh:182]: 
Matrix for composition of phases could not be solved.

I have two phase (wetting + nonwetting), three component (water + oil + 
soluble) problem. I am using the box method. The error arises even if 
the concentration of the soluble is zero in both initial and boundary 
conditions. It arises with non-zero concentration as well. I am not sure 
I understand what it means. It happens with analytic Jacobian but not 
with numeric, though the two Jacobians are close and the resulting 
deltaU are close. There is constant influx boundary condition and 
Dirichlet condition implemented as neumann() on the other boundary.

Here is the output with some debug printing of most simple (2x1 grid) case.

Analytic Jacobian, first step:

deltaU  = [
-8.186022503050234e+05 1.513202330364747e-11 0.000000000000000e+00
-3.593011248700532e+05 1.739221259891074e-23 0.000000000000000e+00
1.000000000000004e+05 1.100997636307169e-33 0.000000000000000e+00
-8.186022503050242e+05 1.513202330364747e-11 0.000000000000000e+00
-3.593011248700537e+05 1.739221259882897e-23 0.000000000000000e+00
1.000000000000004e+05 6.193670380440475e-34 0.000000000000000e+00]

Update: x^(k+1) = x^k - deltax^krank 0 caught an exception while 
assembling:NumericalProblem 
[solve:/home/dpavlov/DUMUX/dumux/dumux/material/constraintsolvers/misciblemultiphasecomposition.hh:182]: 
Matrix for composition of phases could not be solved.
Throwing NumericalProblem for trying with smaller timestep.
Newton: Caught exception: "NumericalProblem 
[assemble_:/home/dpavlov/DUMUX/dumux/dumux/assembly/fvassembler.hh:421]: 
A process did not succeed in linearizing the system"
Newton solver did not converge with dt = 0.001 seconds. Retrying with 
time step of 0.0005 seconds

Numeric Jacobian, first step

deltaU  = [
-8.186005351489465e+05 1.513202513724707e-11 0.000000000000000e+00
-3.593004585059684e+05 1.557296475841733e-23 0.000000000000000e+00
1.000026453219807e+05 3.232270669529124e-24 0.000000000000000e+00
-8.186005600992423e+05 1.513202513724701e-11 0.000000000000000e+00
-3.593004843413417e+05 1.532724499151271e-23 0.000000000000000e+00
1.000034579193550e+05 3.723737820455500e-24 0.000000000000000e+00]
result = 1

Update: x^(k+1) = x^k - deltax^k
Newton iteration  1 done, residual = 3.317e-06


Best regards,

Dmitry


_______________________________________________
Dumux mailing list
Dumux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux