AMPL silently crashing with large model

[Jacob JKW]

After feeding my nonlinear model to AMPL using 'model foo.mod;' and
then typing "solve;" AMPL immediately exits (crashes) to the command
prompt with absolutely no error message. of any kind.

The issue seems to be somehow related to the length of the problem. I
can get the problem to the point where it just works (by editing out
part of the problem text), and then when I append a " + 1" the problem
fails.

The problem only has 14 variables with no constraints (I took them
out), but it is rather long -- 3,796,374 bytes for the objective
function.

I'm having the same problem with both IPOPT and Minos, although the
issue does not seem to be with the solver (AMPL crashes *before* the
solver message (e.g., "Ipopt 3.2.3:") appears.

Any help would be greatly appreciated.

JACOB.

[Robert Fourer]

Dear Jacob,

Try setting "option gentimes 1" to get more detailed output concerning AMPL's
progress. (For further explanation of this option see the discussion in
question 3.6 at www.ampl.com/FAQ.)

Any crash is by definition a bug, though. So unless your problem magically
goes away, you should send your files to in...@ampl.com so that they can be run
on the latest version of AMPL and a bug fix can be made if necessary.

Bob Fourer
4...@ampl.com

[Jacob JKW]

Thanks very much for the reply, Bob.

On Apr 6, 3:47 pm, "Robert Fourer" <4...@ampl.com> wrote:
> Try setting "option gentimes 1" to get more detailed output concerning AMPL's
> progress. (For further explanation of this option see the discussion in
> question 3.6 atwww.ampl.com/FAQ.)

I just have that a try. I also added 'option times 1;' and 'option
show_stats 1;' for good measure.

Output as follows:
C:\amplcml>ampl.exe
ampl: model foo.mod;
# incremental total
#phase seconds memory memory
#execute 0.015625 206868 206872
### foo.mod:4(44) option ...
#execute 0 0 206872
### foo.mod:5(63) option ...
#execute 0 0 206872
### foo.mod:16421(2847148) solve ...
C:\amplcml>

> Any crash is by definition a bug, though. So unless your problem magically

> goes away, you should send your files to i...@ampl.com so that they can be run

> on the latest version of AMPL and a bug fix can be made if necessary.

Sounds about right. I just want to make sure the problem's not my own.
Wouldn't be the first time I've screwed something up.

A simplified version of my objective function is a linear combination
of logarithms with each log term looking something like this:

1.39628860198736e-005 * log( 1 + -1 * x1 + -1 * x2 + -1 * x3 + -1 * x4
+ -1 * x5 + -1 * x6 + -1 * x7 + -1 * x8 + -1 * x9 + -1 * x10 + -1 *
x11 + -1 * x12 + -1 * x13 + -1 * x14 )

Within each log term every combination of the 14 x1 ... x14 decision
variables with coefficients of +1 and -1 are summed together (2^14
16,384 terms). So one term would be as above, another term would be
identical but with a +1 as coefficient to x1, etc.