[Ipopt] 3.7.0 Restoration phase in the restoration phase failed.

[Andreas Waechter]

Hi Stefan,

I think I fixed the problem (only for square problems, right?).  The order 
of the testing of the convergence checks was incorrect.  I committed my 
change to trunk.  Could you please verfiy that this is now doing also the 
right thing for you?  This chance already fixes the issue that Ali Baharev 
reported for the two test cases he put online (Thanks, Ali, for making 
the test cases available).

Regards,

Andreas


On Wed, 30 Sep 2009, Stefan Vigerske wrote:

> Hi,
>
>> I experience similar behaviour. Several nlp's that solved fine with
>> Ipopt 3.6 give restoration failed trouble with Ipopt 3.7.
>>
>> If I revert chgset 1485 (IpRestoConvCheck.cpp), it seems to work better.
>>
>> Any clue why sometimes orig_theta_max should not be set to 0, even
>> though kappa_resto_ is 0? (whatever this means ;-))
>
> I just see that I confused chgset's 1484 and 1485.
> So the patch that works for me is to remove lines 163-165 from
> IpRestoConvCheck.cpp:
>
> --- Ipopt/src/Algorithm/IpRestoConvCheck.cpp    (revision 1572)
> +++ Ipopt/src/Algorithm/IpRestoConvCheck.cpp    (working copy)
> @@ -160,9 +160,6 @@
>     Number orig_inf_pr_max = Max(kappa_resto_*orig_curr_inf_pr,
>                                  Min(orig_ip_data->tol(),
>                                      orig_constr_viol_tol_));
> -    if (kappa_resto_ == 0.) {
> -      orig_inf_pr_max = 0.;
> -    }
>
>     if (first_resto_iter_) {
>       Jnlst().Printf(J_DETAILED, J_MAIN,
>
>
> Stefan
>
>
>>
>> Best,
>> Stefan
>>
>> Ali Baharev wrote:
>>> Dear Developers,
>>>
>>> With the 3.6.1 version, i could successfully solve my feasibility
>>> problems.
>>>
>>> http://reliablecomputing.eu/ipopt/extractive_22.nl
>>>
>>> http://reliablecomputing.eu/ipopt/extractive_30.nl
>>>
>>> With the 3.7.0 i get Restoration Failed.
>>>
>>> Should i stick with 3.6.1? What has changed in 3.7.0 that causes this?
>>> How can i solve them with the new version?
>>>
>>> Many thanks,
>>>
>>> Ali
>>>
>>> _______________________________________________
>>> Ipopt mailing list
>>> Ipopt at list.coin-or.org
>>> http://list.coin-or.org/mailman/listinfo/ipopt
>>>
>>
>>
>
>
> -- 
> Stefan Vigerske
> Humboldt University Berlin, Numerical Mathematics
> http://www.math.hu-berlin.de/~stefan
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>
>