Dear Bruno, thanks for your advice. It actually helped to find another error related to a rank one deficient linear system due to a missing pressure boundary condition.
However, this did not help to debug my actual issue. I solved the non-linear PDE system on coarser meshes and realized that the results coincide. On refined mesh, the differences/residuals became larger. The large residuals are caused by the initialization of some component of the solution. I modified the Newton iteration by resetting this component of solution after the first iteration. This resolved the problem. So I guess my test case / algorithmen was simply not robust enough. Anyway lesson learned. Best wishes, Sebastian [email protected] schrieb am Freitag, 1. April 2022 um 14:55:43 UTC+2: > Sebastian, > > You should get the same result with Docker and on your machine. I have had > a few cases where the result was wrong in docker but correct on my machine. > This was always related to a variable that was not initialized. For some > reason, on my machine the variable would be set to zero but in docker the > value was random. It should be easy to find out where the problem is by > comparing intermediate results from docker with the results from your > machine. > > Best, > > Bruno > > On Friday, April 1, 2022 at 6:40:52 AM UTC-4 SebG wrote: > >> Dear everyone, >> >> I am using GitHub actions and the deal.II docker container to run some >> tests of my applications. The tests solve small problems and the output to >> std::cout is compared with the reference output. However, there is a >> problem with one test, where the L2 norm of a vector becomes approx. >> 1.0e+03. >> >> The norm computed on my machine (Ubuntu 21.04 and deal.II 9.3.0) and on >> the docker container dealii:v9.3.0-focal differ by 4.0e-1 such that the >> test on the docker container fails. I guess that the difference is due to >> floating-point arithmetic because it becomes smaller when a coarser mesh is >> used. >> >> Did you have a similar experience regarding the docker container? Do you >> think that my test setup is simply not suitable if the results are not >> reproducible on another machine? >> >> Best regards, >> Sebastian >> > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/8286ba19-c7fb-409d-8d0f-fa99fa862cc3n%40googlegroups.com.
