[Ipopt] problem about the starting point of IPOPT

[Stefan Vigerske]

Hi,

On 12/19/2013 08:39 AM, header intact wrote:
> Hi, everybody:
>       Now I am learning the IPOPT software, Now I have too questions. The first is:
> 1)    for the HS0701 example which is included in the IPOPT software, the result is (1.000000,4.7429963,3.82114998,1.37940829),however,
>     if I set the starting point with values(1.000000,4.7429963,3.82114998,1.37940829), It still has 5 iterations. Is this normal?
>    The iteration information is as follows:
> iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
>     0 1.7159545e+001 2.01e-002 5.20e-001   0.0 0.00e+000    -  0.00e+000 0.00e+000   0
>     1 1.7097682e+001 5.44e-004 4.08e-003  -1.8 1.22e-001    -  9.91e-001 1.00e+000h  1
>     2 1.7016042e+001 1.41e-003 1.09e-003  -3.1 1.26e-001    -  9.97e-001 1.00e+000f  1
>     3 1.7013987e+001 1.03e-004 1.71e-005  -5.1 2.92e-003    -  1.00e+000 1.00e+000h  1
>     4 1.7014017e+001 2.38e-008 1.35e-008 -11.0 3.70e-005    -  1.00e+000 1.00e+000h  1
> Number of Iterations....: 4

This is
1.) because Ipopt moves the starting point away from the bounds. You can 
modify this with the bound_frac and bound_push options:
    http://www.coin-or.org/Ipopt/documentation/node44.html
2.) you do not provide an optimal dual solution but only a primal point, 
so Ipopt needs to calculate these too.

>    2) The second question is that I want to know the difference of scaled an unscaled  as below. sometimes the two methos have different objectives.
>
>                                     (scaled)                                  (unscaled)
> Objective...............:  1.7014017140224134e+001   1.7014017140224134e+001
> Dual infeasibility......:  2.2928101314633036e-014   2.2928101314633036e-014
> Constraint violation....:  2.1316282072803006e-014   2.1316282072803006e-014
> Complementarity.........:  1.0023967333275279e-011   1.0023967333275279e-011
> Overall NLP error.......:  1.0023967333275279e-011   1.0023967333275279e-011

Ipopt by default solves a scaled version of the problem.
It shows you some key values w.r.t. the scaled and the unscaled version 
of the problem.

Stefan

>
> Thanks for the help!
> yours:
>       Kaiyan Pan.
>
>
>
>
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt
>