[Ipopt] ERROR 20 IN ROUTINE g2dHesRowStruct: the NL instruction stack not empty

[Stefan Vigerske]

Hi,

the error comes from the GAMS hessian evaluation routines.
   (and I think you stumbled about it some time ago before?)

One workaround could be to enable Ipopts hessian approximation.

But updating to a newer GAMS version would be best.
GAMS >= 23.4 uses a different code for computing the hessian, so the old 
one from 23.3 is not maintained anymore.
Since GAMS 23.7, Ipopt is also linked against the multithread version of 
the MKL, maybe that can help to bring some speedup.

Stefan

> I have a curious problem I do not understand. I have a NLP model written
> in GAMS (v. 23.3.3) which I try to solve with Ipopt (v. 3.8).
>
> Thereby I get the following error message:
>
> ************************************************************************
> --- Starting compilation
> --- tmp_S5iQo3(137150) 19 Mb
> --- Starting execution: elapsed 0:00:01.036
> --- tmp_S5iQo3(92420) 20 Mb
> --- Generating NLP model HSsHM_V10_model
> --- tmp_S5iQo3(92422) 31 Mb
> --- 12,388 rows 22,262 columns 52,928 non-zeroes
> --- 176,655 nl-code 18,489 nl-non-zeroes
> --- tmp_S5iQo3(92422) 29 Mb
> --- Executing IPOPTD: elapsed 0:00:05.601
> --- tmp_S5iQo3(92422) 29 Mb
>
> GAMS/IpoptD (Ipopt Library 3.8)
> written by A. Waechter
> ERROR 20 IN ROUTINE g2dHesRowStruct: the NL instruction stack not empty
> as expected
> --- Restarting execution
> --- tmp_S5iQo3(92422) 16 Mb
> --- Reading solution for model HSsHM_V10_model
> --- tmp_S5iQo3(137149) 16 Mb
> --- Putfile nlp_results
> /home/euklid/willert/work/egt/lamatto/tmp_S5iQo3.dat
> *** Status: Normal completion
> --- Job tmp_S5iQo3 Stop 09/21/11 10:04:34 elapsed 0:00:09.785
> ************************************************************************
>
> Where is the problem? And how can I resolve it?
>
> Tanks in advance,
> Martin Schmidt
>