Skip to content

Refinement moves your layer thicknesses to minimize the merit function. It never adds or removes layers — only their thicknesses change. This is the everyday workhorse: fast, robust, and the natural finishing pass after any synthesis run or hand edit.

It minimizes the operand set you defined in the Merit Function Editor, using the method you pick from a dropdown. The default is SQP, which treats a minimum thickness as a hard limit, but you can switch to several other local and global methods. The variables that move follow the surface mode set in the Design Editor (front, back, both, or symmetric), and every method works in every mode.

For a description of each method and guidance on when to choose it, see Optimization Methods.

Method — the optimizer to run: SQP, Damped Least Squares, Conjugate Gradient, Newton, Newton-CG, DLS multi-start, Differential Evolution, Simulated Annealing, or Try all — keep best. Your choice is remembered across designs.

Max iter — the most steps the optimizer may take. The run still stops early when it converges, so this is just a cap. It defaults to a sensible budget for the selected method.

Restarts (N) — for DLS multi-start only: how many randomly perturbed starts to run before keeping the best (typically 4–8, or more for a tough surface).

Perturbation — for DLS multi-start only: how much to jitter each layer’s thickness (as a percentage) at the start of every restart.

The surface mode and merit-evaluation mode in effect are shown as badges on the window.

While the run is live you see the spectrum update and a running merit-function readout (current value, best so far, and the starting value). The MF trend chart plots the best merit found against iteration on a logarithmic axis — a steadily falling curve that flattens out means the method has settled into a minimum.

Reset returns to the thicknesses you had before the run, and a single undo covers the whole run. The History panel keeps a snapshot of each run so you can jump back to any earlier result.

A single-start run (N = 1) is usually enough to re-settle a hand-edited stack. After a synthesis pass, a multi-start run helps confirm you have reached the true local minimum rather than an improved-but-not-bottom point. Minimum and maximum thickness limits are honoured here (they are deliberately relaxed during Needle and Gradual Evolution synthesis).

  • K. Levenberg, Quart. Appl. Math. 2, 164 (1944); D. Marquardt, J. SIAM 11, 431 (1963).
  • J. Nocedal & S. Wright, Numerical Optimization, 2nd ed.
  • H. A. Macleod, Thin-Film Optical Filters, 5th ed., §13.