SNOPT Changelog
Sun May 11 18:35:33 EDT 1997
Tweak things to reflect changes to SNOPT since 19970318.
Fri Jul 11 02:06:00 EDT 1997
Trivial tweak to snopt.c; bring xsum0.out up to date (it was wrong).
Fri Oct 17 01:17:15 EDT 1997
Report numbers of nonlinear objective and constraint evaluations
(when positive), and the numbers of gradient (and Jacobian) evaluations.
Give default 99999999 to the iteration limit “iterations”.
This stuff still pertains to SNOPT 5.1.2; a driver for 5.3 is coming.
Thu Oct 23 21:07:31 EDT 1997
Update to version 5.3.
Fri Oct 24 11:34:08 EDT 1997
Fix glitch with “snopt -v”, which erroneously mentioned “March 1997”.
Thu Oct 30 11:52:40 EST 1997
Make outlev > 1 work as described by “snopt -?”.
Thu Nov 6 09:01:26 EST 1997
snopt.c: tweak for -DKR_headers; snoptver.c0 not changed.
Fri Nov 7 12:45:25 EST 1997
snopt.c: fix typo in snopt -=; invoke xknowne (added last night
to amplsolver.a) rather than xknown.
Wed Nov 12 17:01:39 EST 1997
README makefile: tweak to reflect libamplsnopt.a that can now be
created by the makefile in Philip Gill’s snopt.tar.gz.
Fri Nov 14 07:47:44 EST 1997
snopt.c snoptver.c0: correct calling seq. of funcon and funobj (the
change should be invisible, as it involves arguments that are never
referenced); change MINOS to SNOPT in the output of “snopt -?”.
Mon Nov 24 08:45:44 EST 1997
snopt.c snoptver.c0: supply problem name (based on stub), as it
appears in output generated by outlev >= 2.
Tue Feb 3 13:28:01 EST 1998
snopt.c snoptver.c0: recognize return (not documented in the SNOPT
source) for “Primal feasible solution; could not satisfy dual
feasibility”.
Sun Apr 19 21:40:42 EDT 1998
makefile: omit $S/funcadd0.o; on most systems, $S/amplsolver.a now
provides a recommended default funcadd: see ampl/solvers/funclink/README
for details.
Tue May 26 23:11:20 EDT 1998
Update to SNOPT 5.3-2.
Fri Jun 5 15:04:15 EDT 1998
Adjust returned status values so nonbasic slacks for equality
constraints will have .status value “equ”.
Tue Jul 28 22:36:04 EDT 1998
README: add table of possible solve_result_num values.
Wed Oct 14 00:31:19 EDT 1998
snopt.c: updated to fix a bug with problems having a nonlinear
objective and some purely linear constraint variables that do not
appear in the objective. Too much of the gradient was zeroed in
this case, possibly corrupting memory. The fix involves an update of
the AMPL/solver interface library (amplsolver.a or amplsolv.lib), which
must be recompiled.
Thu Nov 19 16:07:42 EST 1998
snopt.c: correct an off-by-one bug in the numbers of objective and
constraint evaluations reported in the solution message: the numbers
were high by one.
snoptver.c0: reflect version 5.3-4.
Mon Nov 23 15:18:58 EST 1998
snopt.c: change default superbasics_limit (which must be known too
soon for the usual default computation by the SNOPT routines) to
min(500, nnL+1), where nnL is the number of nonlinear variables.
makefile: pout out CFLAGS comments in $S/makefile.
Tue Nov 24 17:21:22 EST 1998
snopt.c: call snMem() rather than s8Mem() to get memory estimates.
This changes should be invisible, but makes the code a bit simpler and
more readable.
Sat Feb 26 15:25:26 EST 2000
snopt.c: remove assignment “Plinfy = Infinity” to keep SNOPT’s
scaling routine from generating NaNs (under rare conditions).
Mon Feb 28 13:22:45 EST 2000
snopt.c: recant yesterday’s change; something else was amiss in the
example that led to the change.
Thu May 11 12:58:46 EDT 2000
snopt.c: zero memory passed to sninit_().
Wed May 23 13:38:49 EDT 2001
snopt.c: Adjust so outlev=n for n > 1 will report SNOPT’s function
evaluation times by default (unless timing_=0 is specified). On some
systems, these times are less accurate than those produced when
ftimes=1 is specified.
snopt.c: updated for version 6.1-1; the previous snopt.c (for
version 5.3.5) is now snopt535.c.
README updated.
makefile renamed makefile.u; new makefile.vc and subdirectory src
(with its own Unix and Microsoft makefile variants).
Wed Jun 20 16:52:59 EDT 2001
Minor tweaks… snopt.c: fix glitch on line 942 (change = to ==) and
add #ifdef Finite_Plinfy for machines with buggy Infinity arithmetic.
Add src/makefile.fu.
Add student snopt binaries to subdirectories of /ampl/student.
Wed Apr 10 14:02:46 EDT 2002
snopt.c: upon SIGINT (which may be caused on some systems by typing
control-C or Del), return with solve_result_num = 401 and give
solve_message “SIGINT treated as iteration limit”.
Tue Apr 30 18:35:10 EDT 2002
snopt.c: show less version information by default: rather than,
“SNOPT 6.1-1(4)(Jun 2001):”, print just “SNOPT 6.1-1:”. More detail
is still available with the command-line invocation “snopt -v” or
with “version=1” in $snopt_options.
Mon Oct 22 01:20:56 MDT 2007
Update to SNOPT version 7.2-5. Add simple configure and
configurehere scripts.
Thu Nov 29 02:06:53 MST 2007
snopt.c, README: when the objective cannot be evaluated initially,
report this (with new solve_result_num value 515) rather than that
the constraints cannot be evaluated at the starting point.
snopt_.c: update for SNOPT 7.*.
Sun Apr 6 22:26:59 MDT 2008
snopt.c: adjust get_Title to ensure Oinfo.version is null-terminated.
Wed Jun 1 13:17:17 MDT 2011
snopt.c: fix an (apparently so-far harmless) out-of-bounds array reference
on QP problems.
20140313
snopt.c: new keyword objrep: Whether to replace
with
minimize obj: f(x)
when variable v appears linearly in exactly one constraint of the form
s.t. c: v >= f(x);
or
s.t. c: v == f(x);
Possible objrep values:
0 = no
1 = yes for v >= f(x) (default)
2 = yes for v == f(x)
3 = yes in both cases
snopt.c and README: new solve_result_num value 530 for
“Indefinite Hessian”, which is possible with the default qpcheck = 1
when the objective is quadratic and the constraints are linear.
20150525
snopt.c: fix a bug that caused keyword meminc to be ignored, and
arrange for “snopt stub meminc=? …” to show the default meminc value
(which is 20*(M + N), where M = number of constraints and N = number of
variables).
20181101
snopt.c: fix a glitch with a starting point where a derivative cannot be evaluated.
Instead of saying, e.g.,
SNOPT 7.5-1.2 : Bug: "scream" called.
-11111 iterations
say
SNOPT 7.5-1.2 : Derivative evaluation error.
0 iterations, infeasibility sum 0
snopt.c, README.snopt: add some solve_result_num values for
unlikely errors:
517 Bad "nn=filename" assignment
561 Bug: "scream" called
562 Evaluation error
563 Derivative evaluation error
564 Bug: surprise setjmp() return
565 Floating-point error