I am looking for FORTRAN source code for two related problems. Both
involve unconstrained optimization. The applications involve finding
the parameter values that maximize the posterior density function (in
Bayesian Statistics). For the applications I’m working on, there are
no closed form expressions for the parameter estimates.
1. FORTRAN source code that implements a Newton’s method where the
Hessian can be computed in closed form. I am looking for a routine
that either modifies the step length using for example the Wolfe
condition, or implements trust-region based approach.
2. FORTRAN source code that would implement a method such as the
Limited-Memory BFGS or some other comparable method for large-scale
unconstrained optimization problems.
It would be very useful to obtain the source code for these methods so
I can incorporate into larger applications I am writing.
none - 22 Mar 2009 21:55 GMT
> I am looking for FORTRAN source code for two related problems. Both
> involve unconstrained optimization. The applications involve finding
[quoted text clipped - 13 lines]
> It would be very useful to obtain the source code for these methods so
> I can incorporate into larger applications I am writing.
Always worth looking at Netlib (www.netlib.org), TOMS
(www.netlib.org/toms) and the Decision Tree for Optimisation
(http://plato.asu.edu/guide.html). Whether these have what you want I
don't know, but you might get a better response if you satte that you have
tried these & not found the relevant codes.
MichaelHill - 29 Mar 2009 20:53 GMT
> I am looking for FORTRAN source code for two related problems. Both
> involve unconstrained optimization. The applications involve finding
[quoted text clipped - 13 lines]
> It would be very useful to obtain the source code for these methods so
> I can incorporate into larger applications I am writing.
The book "Numerical Methods and Software" by Kahaner, Moler, and Nash
came with a disk that included FORTRAN source code for many programs.
In fact, one of the authors, Nash, has also posted source code on his
web site at George Mason University:
http://iris.gmu.edu/~snash/nash/software/NMS__single/ch09.f
FMIN is for one dimensional problems.
UNCMIN is for multi-dimensional problems. It uses a quasi-Newton
method.
Hopefully this helps a bit. These are from the single precision
section; he also lists double-precision routines.