A little bit of help with CONOPT Termination message 101

[Jose Gómez de la Varga]

Good day,

 

My name is José Gómez and I am a researcher currently using the CONOPT solver with a license for AMPL. I am trying to understand a termination message the solver is outputting, but I find little to no further information than the one in the manual Using CONOPT with AMPL. I have tried writing directly to the developer (ARKI Consulting & Development at in...@arki.dk) but the recipient address was rejected as the domain was not found. My problem is with the termination message that has the value 101, whose message only says Bogus “Termination by solver” (CONOPT bug). I could not find any further explication as to why this error happens or what are its implications. I hope you could shed a little bit of light in this matter, or redirect my message to anyone who can.  

 

Thanks in advance, any help would be appreciated.

JGV

[AMPL Google Group]

Dear José,

Curiously, in...@arki.dk is the address given at http://www.conopt.com/Contact.htm. I am checking to see what the correct address might be.

Also, can you send us files that we could use to reproduce this problem with CONOPT on our computers? Then we can determine where in the AMPL-CONOPT interface the problem is happening, and can provide that information to ARKI. In addition to the files, send us the version information produced by these two AMPL commands:

option version;
shell "conopt -v";

(If you want to keep your files private, you can mail them as attachments to sup...@ampl.com.)

--
Robert Fourer
am...@googlegroups.com

{#HS:1767510923-108326#}

[José Gómez de la Varga]

Dear Robert,

Sorry for the late response, I was preparing the files to send it to you when I found the extra information I was looking for. Activating the option outlev=3, I could link the termination message I asked about with an alternative termination message for an optimal solution. It appears in fact in the manual Using CONOPT with AMPL, but as there was no connection with the the termination message 101, I wrongly thought it different. What I have found out  is, in case other people could find it interesting, that the solver will raise a warning with the termination message 101 Bogus “Termination by solver” (CONOPT bug) when this optimal solution is reached (extracted from the manual before):

  ** Optimal solution. Convergence too slow. The change in objective has been less than xx.xx for xx consecutive iterations. 

CONOPT stopped with a solution that seems optimal. The solution process is stopped because of slow progress. The largest component of the reduced gradient is greater than the optimality tolerance rtredg, but less than rtredg multiplied by a scaling factor equal to the largest Jacobian element divided by 100. Again, the model must have large derivatives for this message to appear so it is advisable to scale the model.   

Thanks anyway!

[AMPL Google Group]

Thanks for posting this useful information.

--
Robert Fourer
am...@googlegroups.com

{#HS:1767510923-108326#}

On Wed, Feb 2, 2022 at 2:40 PM UTC, AMPL Modeling Language <am...@googlegroups.com> wrote:

Dear Robert,

Sorry for the late response, I was preparing the files to send it to you when I found the extra information I was looking for. Activating the option outlev=3, I could link the termination message I asked about with an alternative termination message for an optimal solution. It appears in fact in the manual Using CONOPT with AMPL, but as there was no connection with the the termination message 101, I wrongly thought it different. What I have found out is, in case other people could find it interesting, that the solver will raise a warning with the termination message 101 Bogus "Termination by solver" (CONOPT bug) when this optimal solution is reached (extracted from the manual before):

** Optimal solution. Convergence too slow. The change in objective has been less than xx.xx for xx consecutive iterations.

CONOPT stopped with a solution that seems optimal. The solution process is stopped because of slow progress. The largest component of the reduced gradient is greater than the optimality tolerance rtredg, but less than rtredg multiplied by a scaling factor equal to the largest Jacobian element divided by 100. Again, the model must have large derivatives for this message to appear so it is advisable to scale the model.

Thanks anyway!

On Wed, Jan 26, 2022 at 12:49 AM UTC, AMPL Google Group <am...@googlegroups.com> wrote:

Dear José,

Curiously, in...@arki.dk is the address given at http://www.conopt.com/Contact.htm. I am checking to see what the correct address might be.

Also, can you send us files that we could use to reproduce this problem with CONOPT on our computers? Then we can determine where in the AMPL-CONOPT interface the problem is happening, and can provide that information to ARKI. In addition to the files, send us the version information produced by these two AMPL commands:

option version;
shell "conopt -v";

(If you want to keep your files private, you can mail them as attachments to sup...@ampl.com.)

--
Robert Fourer
am...@googlegroups.com