The following are some general remarks regarding the use of the optimizer and the Ansys Polyflow solver.

In the optimizer, the cost function is of the order of magnitude of 100. You should therefore note that when you specify an accuracy of it is with respect to a value of 100 (that is, the real accuracy is ).

For an optimization task, the accuracy of the solvers will automatically be set to . Such a high level of accuracy is necessary, because the accuracy of the solvers has a major effect on the behavior of the optimizer.

If the flow problem is nonlinear, it is recommended that you perform an initial calculation (perhaps using evolution) in order to obtain a first solution. The optimization can then start from this solution. Be sure to save this new result file with a different name than the original result file.

An optimization task is not compatible with evolution or transient tasks. The incompatibility with evolution should not be understood as a limitation. If the flow problem is so complex that it requires evolution to obtain a first solution (for example, due to shear rate dependence on the viscosity, viscous heating, or temperature-dependent properties), evolution may still be implemented to calculate an initial solution prior to optimization. You can then start the optimization task using the results file obtained at the end of such an evolution simulation. Note that starting from the converged initial solution does not guarantee the convergence of the subsequent line search algorithm. Similar to the standard evolution scheme, if the solution is converging, the incremental increase of is also increased until the final solution is reached. If the solution is diverging, then the incremental increase of is reduced to allow the convergence of the otherwise nonlinear problem. These subiterations are terminated if either the target value of is reached or the incremental change of reaches a value below the minimum allowable value. The minimum allowable value is hardcoded as 1% of the total difference between the initial and target values of .

Because the convergence criteria is relatively small (10^-6), it is recommended that you use the Newton-Raphson scheme during optimization. This scheme yields better sensitivities, and usually the new design variable values are close to the previous values (so the solver should converge easily). If the solver does not happen to converge, the optimizer will reduce the variations of the design variables.

During optimization, if the element distortion check is activated and a bad element is detected, the optimizer will reduce the variations of the design variables.