Skip to content

FSI/cylinderFlap_2D/OpenFOAM-deal.II gives IQN error #94

Closed

Description

@MakisH

With preCICE v2.1.0 (release candidate, 295d27), either something changed in behavior, or some check became stricter:

Timestep 1 @ 0.002s
_______________________________________________________________________________________
    SOLVER STEP     |  LIN_IT   LIN_RES    RES_NORM   RES_ABS      U_NORM     U_ABS 
_______________________________________________________________________________________
  0  CST  ASM  SLV  |       1  0.000e+00  1.000e+00  1.115e+02    1.000e+00  7.714e-05  
  1  CST  ASM  SLV  |       1  0.000e+00  2.800e-04  3.122e-02    6.721e-04  5.184e-08  
  2  CST  ASM  SLV  |       1  0.000e+00  8.246e-08  9.193e-06    2.823e-06  2.178e-10  
  3  CST  ASM  SLV  |       1  0.000e+00  4.042e-10  4.506e-08    2.300e-08  1.774e-12  
  4  CST  ASM  CONVERGED! 
_______________________________________________________________________________________
Relative errors:
Displacement:	2.300e-08
Residual: 	3.427e-12
v / V_0:	7.020e-03 / 7.020e-03
---[precice]  relative convergence measure: relative two-norm diff = 5.50865e-11, limit = 5e-05, normalization = 19916.5, conv = true
---[precice]  relative convergence measure: relative two-norm diff = 1, limit = 5e-05, normalization = 4.46315e-05, conv = false
---[precice] ERROR:  Attempting to add a zero vector to the quasi-Newton V matrix. This means that the residual in two consecutive iterations is identical. There is probably something wrong in your adapter. Maybe you always write the same (or only incremented) data or you call advance without providing  new data first.

The FSI/flapPerp_2D/OpenFOAM-deal.II is not affected.

Activity

  1. changed the title [-]FSI/cylinderFlap/OpenFOAM-deal.II gives IQN error[/-] [+]FSI/cylinderFlap_2D/OpenFOAM-deal.II gives IQN error[/+] on Jul 21, 2020
  2. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    This is due to a new check introduced in precice/precice#787

  3. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    I rebuild preCICE yesterday in the evening and it worked with the deal.II adapter (in 2D).
    I think this is a problem of the initial ramp, where not very much is happening and we just write the same data due to nothing is happening in the first time step. I get also a warning in the beginning, but not an error. In 3d, it might be heavier, since the out of plane direction writes always zero data.

  4. MakisH commented on Jul 21, 2020

    @MakisH
    MemberAuthor

    So, what do we do then?

  5. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    Remains the error, when you remove the ramp?

    Options are

    1. raise a warning instead of an error in preCICE
    2. remove the 3d deal.II examples
    3. modify the ramp
  6. MakisH commented on Jul 21, 2020

    @MakisH
    MemberAuthor

    Could you please try again with the release candidate?

    This error I get with the 2D case, so I don't understand how removing the 3D examples would help.

  7. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    This error I get with the 2D case, so I don't understand how removing the 3D examples would help.

    My bad, i will check it again, though I'm working with the parabolic profile and groovy.

  8. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    This looks really seriously, I can reproduce it. It fails in the first iteration (for some strange reason not with groovyBC, but with the original one): The problem is (my best guess), that the displacement doesn't converge according to the prescribed limits, but the initial data (zero) is passed to the solid solver again due to reloading.

    Quick workaround/fix: Switch to a serial-implicitscheme

  9. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    The problem is the missing data entry in the acceleration

            <acceleration:IQN-ILS>
                <data name="Displacement" mesh="Solid_mesh"/>
    +           <data name="Stress" mesh="Fluid-Mesh-Centers"/>
                <preconditioner type="residual-sum"/>
                <filter type="QR1" limit="1e-6"/>
                <initial-relaxation value="0.1"/>
                <max-used-iterations value="50"/>
                <time-windows-reused value="10"/>
            </acceleration:IQN-ILS>

    However, then, dealii destroys the SolverInterface early in the second timestep

    ---[precice]  Implicitly finalizing in destructor
    

    I guess due to convergence problems?

    All in all, I think the preCICE error message here is meaningful.

  10. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    Hm, I'm currently running parallel-implicit without acceleration in both data sets, should not cause problems.

    So yes, looking at the solver data, the error is right, deal.II receives all the time the same data leading to all the time the same output..

    I guess due to convergence problems?

    This should happend only, when the simulation is terminated abnormal.

  11. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    We used to have a check for this:

    https://github.com/precice/precice/blob/b2b33b1efba24bdbdee22f1ed3b1901cd3552fab/src/cplscheme/ParallelCouplingScheme.cpp#L54-L55

    But we removed it at some point as it does not always make sense.

    For the nonlinear dealii solver it also crashes with serial-implicit for me (have not tested the linear one yet).

  12. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    Yes this is related to #77 . It runs properly with a better config.

    In your case, the problem should be different from this though.

  13. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    linear works

  14. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    Suggestion: let's remove this tutorial from the documentation until we have #77 ?

  15. davidscn commented on Jul 21, 2020

    @davidscn
    Member

    linear works

    parallel or serial?

    I think we can not resolve this in a few hours without looking more in detail what's going wrong, so yes.

  16. uekerman commented on Jul 21, 2020

    @uekerman
    Member

    parallel or serial?

    with the linear solver both coupling schemes work

  17. MakisH commented on Jul 21, 2020

    @MakisH
    MemberAuthor

    I ran all the cases (flap/turek, 2D/3D, and linear/nonlinear) in serial. Here are some results: precice/precice#835 (comment)

  18. added a commit that references this issue on Nov 7, 2020
    ef1072b
  19. IshaanDesai commented on Apr 14, 2021

    @IshaanDesai
    Member

    Resolved in #116

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions