On a whim, I decided to see how hard it would be to make a bigfloat
version of fmin_cobyla. The conversion of the code itself was
straightforward, requiring only a few declaration tweaks and replacing
some f2cl-lib functions like fsqrt, dmax1, and dmin1 with the generic
bigfloat equivalents, sqrt, max, man. The interface required a bit more
work, keeping track of whether we wanted bigfloat objects or maxima
bfloat objects.
This all works out, and bf_fmin_coblya works, except that only iprint =
0 works. The result is pretty slow, some 50-100 times slower than
fmin_cobyla, even with fpprec = 16.
But I guess if you need a bigfloat constrained minimum, this will do it
if you wait long enough.
I might check this in, in case someone needs a bigfloat version.
Ray