General

General

Why does the ANSYS Fluent adjoint gradient-based optimizer reports all calculations to be converged but checking the flow residuals of the optimal design shows that only one iteration is calculated for each design point?

    • FAQFAQ
      Participant

      It is possible that after a design change the flow results don’t change much. The residuals may need more than one iteration to rise above the convergence criteria, therefore Fluent detects the solution to be converged immediately. This triggers the optimizer to continue with the next adjoint calculation prematurely. The only way to detect this behavior is to check the transcript very carefully or look at a cas/dat pair that is either saved as optimal solution or by the autosave feature of the optimizer. To avoid this behavior, increase the reporting interval on the ‘run calculation’ task page to a value larger than 1, e.g. 5. This reduces the residual output to the console and transcripts for the flow calculation but it also limits the convergence check to the specified interval. Keywords: ANSYS Fluent; adjoint; gradient-based optimizer; converge; convergence