Optimization Workflow

Normally, the workflow during an optimization is the following:

  1. In the PARROT module, run the setup macro file using MACRO_FILE_OPEN.
  2. Go to ED_EXP with EDIT_EXPERIMENTS, use READ_WORKSPACES to load your experimental data and then execute .
  3. If an equilibrium cannot be computed, then this either means that the start values (site fractions, for example) were inappropriate or that the equilibrium cannot be calculated with the current set of parameters. If the latter, then the equilibrium may have to be excluded from the optimization. However, you should make an effort to exclude the possibility that the reason is just that you do not have the appropriate start values for the equilibrium.
  4. When each equilibrium is either successfully computed or excluded from the calculation, go back to the PARROT module and run the OPTIMIZE_VARIABLES command. Give the number of optimization iterations you want the Thermo‑Calc program to carry out as an argument to the command. This number specifies how many different sets of values for the optimizing variables that PARROT tries to fit with the experimental data. It is recommended that you run 0 iterations first to see the discrepancy between the experimental data and the initial calculated results. The current workspace file is automatically updated with the last set of optimized variables and calculated results.
  5. List and evaluate the result of the optimization using LIST_RESULT. Typically, what is of interest in the results are the relative standard deviation of the optimizing variables and the error values in the list of calculated experimental data results. If the relative standard deviation for an optimizing variable is very large, this means that the variable doesn’t have any effect on the computed results and that the variable should not be used (you can make PARROT ignore the variable by setting its value to 0 with SET_FIX_VARIABLE). If the relative standard deviation is close to zero for a variable, then this might indicate that you need to use more variables. When it comes to errors in the calculated equilibrium results, these can be decreased by lowering the weight that PARROT gives the experiment in the calculation (the more weight, the more the experiment affects the calculated results).
  6. Use OPTIMIZE_VARIABLES or CONTINUE_OPTIMIZATION to carry out a new optimization cycle. Go back to step 5 above and continue to do optimization cycles until you judge that there is a good enough fit between your calculated results and your experimental data. It is recommended that you do not only rely on feedback from LIST_RESULT but also frequently plot various diagrams and compare them visually to some of your experimental data.