CFD Utilities Files

The optlib collection includes general purpose 1-D zero-finding and minim-
ization utilities, the n-dimensional unconstrained optimizers QNMDIF[2]
(quasi-Newton) and MNA (modified Newton algorithm), both originally by
Gill, Murray, et al.  For constrained optimization at NASA Ames Research
Center, the NPOPT variant of NPSOL has long been employed, but these
packages require a license from Stanford University and thus cannot be
offered here, although some such applications are included.

Frameworks for new applications of [sequences of] unconstrained and
constrained optimization may be found under QNMDRIVER2, QNMDRIVER3, and
NPOPT_DRIVER.

Reverse-communication forms of the early derivative-free FMIN and ZEROIN
methods by Richard Brent are recommended for efficient 1-D minima and
zero-finding.  Here, the utility returns for some action at each iteration
(such as evaluating the [objective] function at the current x, or terminat-
ing the iteration).  The function can thus be as simple or as complex as
the application calls for, as opposed to being of the form FUN (X) with
all other needed variables accessed inconveniently via a common block (or
module, these days).

For looking at maxima and minima in 1-D datasets, see the MAXMIN driver.

Compilation warning:  qnm_solution.f90 & qnm_solution3.f90 both use
the module in qnm_module.f90 and hence don't belong in f90optlib.a
because they should all be compiled into the application program directory.

cendif.f          3-pt. df/dxi, d2f/dxi2 estimates; option to improve "h"i
cubic.f           Roots of a cubic equation with real coefficients
fmin77.f          Robert Kennelly's initial refinement of FMIN; try FMINRC
fminrc2.f         Same as FMINRC with a distinct name for nested minimizns.
fminrc.f          1-D minimizn.; no-derivative method; reverse communicatn.
hooke_jeeves.f90  N-D minimizn.; non-derivative; variable bounds (only)
mna.f             Modified Newton Algorithm, 1st/2nd derivs., unconstrained
objeps.f          Function precision estimation via difference tables
qnmdif2.f         N-D quasi-Newton minimzn., 2/3-pt. grads., unconstrained
qnmdif.f          Earlier Gill/Murray package, not fully argument-driven
qnm_module.f90    Replaces common block originally employed by QNMDIF2
qnm_solution.f90  Subroutine form of the essence of QNMDRIVER2 framework
qnm_solution3.f90 Subr. form of QNMDRIVER3 framework (seq. of minimizns.)
zerorc.f          1-D zero finder, no-derivative method, reverse communicn.
zeros.f           (Crude) estimation of all zeros (if any) in a fn. array