[Raytrace] Some Optimiser strategies

[Peter John Smith]

I really do not know how useful this is.

There are so many diverse applications with different programs with
different capabilities. Some will simply think this is simplistic - yet
others may be lost.  For that I apologise.

Quite some thought should be given to the best way to constrain important
parameters - yet allow many appropriate degrees of freedom during
optimisation.

I have found that with Zemax, convergence can be quite sensible  and orderly
with a large number of variables - as long as constraints and variables are
well chosen.

If it is easy and fast to experiment, do some trials.

A lot of care must be taken when making spacings variable - one has to keep
a tight reign.
The real fools mate one will discover early on.  Since better performance
usually results as F:NO > infinity, if there are not sufficient restraints,
the Focal length may > infinity as spacings change to suite.  Some ways to
constrain this are to specify an effective focal length or a key large
spacing or possibly the ROC of a primary mirror.  This happens to everyone.
If this happens, just undo the change and consider more appropriate
restraints.

Since the thicknesses of lenses can become mathematically negative w.r.t.
the light direction and still make sense to an optimiser, it is usually best
to make these constants.  (Note - if the light direction is in the negative
sense, they should have a negative spacing.  )  .  Towards the end of a
session, it may be appropriate to make lens thicknesses variables for short
runs just to see whether they want to increase or decrease.  When this has
been established, change them again to constants of better values.  But keep
a watch to see they remain sensible

A very powerful technique is to tie two spacings togeather - ie make one
dependant on another.
For example, It is possible to make the distance from primary to cass sec a
variable and then force the distance from sec to image to be, say,  -1.1
times this value.  This is done by picking up a surface and supplying a
multiplier.

Another example could be applied to the DK11 design I talked about earlier.
If the spacing from primary to sec is a variable, then the distance from sec
to corrector set to -0.95 of this value, (ie just in front of the primary)
and finally the spacing from last corrector surface to image set to 200 mm,
the design is not overconstrained yet will not run away.

When constraints on the R of surfaces are needed, it may be more useful to
make one surface dependent on another rather than a constant.  For
example,it is difficult for an equiconvex  or equiconcave lens to run amok
during optimising.  To constrain a lens to be equiconvex simply pick up the
R of the previous surface and modify it by using a multiplier of -1.  Then,
as the design progresses, remove other variables but then make both R values
of this lens independent variables and reoptimise.

All this sounds messy.  But remember than any optical design is a compromise
anyway.  Often far more of a compromise than we realize.

Peter Smith.