Convergence Issues in Parallel Field Inversion with Periodic Boundary Conditions

[lnhang0618]

Dear Professor He,

In our previous discussion, there was mention of convergence issues with the periodic boundary conditions in Field Inversion during parallelization. I experimented with version 3.11 and it seems that the problem still exists. What is the main cause of this issue?"

[friedenhe]

If I remember correctly, the problem is in the accuracy of the adjoint derivatives. The flow should still converge. We are still not quite sure what caused the issue...

[lnhang0618]

Yes, the adjoint derivatives are not correct when parallel, which causes issues in my attempt to directly train the neural network using dJdbeta.

[friedenhe]

Please run the periodic hill case in serial.

[uttamcadambi07]

What if we wanted to run another case that has periodic boundary conditions that is not the periodic hill case? For example we are interested in doing infinite span wing simulations and want to initialise the spanwise boundaries as periodic. It is impossible to run it in serial (we have some 2million cells) so can we still run it successfully in parallel? please advice

[friedenhe]

DAFoam's adjoint accuracy is good in parallel for other cases with periodic boundary conditions (e.g., propellers), it just somehow does not work for the periodic hill case. I suggest you run your case in parallel and test its accuracy. I may work without a problem, so there is no need to worry about it.