[Help-glpk] intermediate solutions in the dual simplex - next question

[HBuesching04]

Hey!

I make a case study, what happes to the intermediate dual solution when 
to an optimally solved problem another constraint is added.
I managed to solve my last issues by myself, by filling the parameters 
LPX_K_ITLIM and LPX_K_DUAL in glpsol and by providing glpsol a 
precalculated basis.
I can follow the intermediate solution in the dual simplex now. 

Sadly the behaviour of the dual Simplex is not as wanted.

|     0:   objval =   5.663113946e+04   infeas =   6.722711631e-01 (6)
|     1:   objval =   5.663114076e+04   infeas =   5.465827705e+01 (6)
...
|     8:   objval =   5.663114611e+04   infeas =   3.930494746e+01 (6)
|     9:   objval =   5.663115258e+04   infeas =   6.765461431e+01 (6)
|    10:   objval =   5.663115299e+04   infeas =   1.865660255e+01 (6)
|    11:   objval =   5.663117154e+04   infeas =   1.635406984e+01 (6)
|    12:   objval =   5.663118213e+04   infeas =   7.475547208e+01 (6)
|    13:   objval =   5.663118335e+04   infeas =   2.172500586e+01 (6)
|    14:   objval =   5.663119723e+04   infeas =   1.358614171e+02 (6)
|    15:   objval =   5.663119723e+04   infeas =   3.800135916e+01 (6)
|    16:   objval =   5.663119723e+04   infeas =   3.887601837e+01 (6)
|    17:   objval =   5.663119723e+04   infeas =   3.757736813e+01 (6)
|    18:   objval =   5.663120303e+04   infeas =   1.185012674e+01 (6)
|    19:   objval =   5.663120303e+04   infeas =   2.685168586e+01 (6)
|    20:   objval =   5.663121747e+04   infeas =   1.130289742e+02 (6)
|    21:   objval =   5.663121747e+04   infeas =   1.126672578e+02 (6)
|    22:   objval =   5.663121747e+04   infeas =   8.943864847e+01 (6)
|    23:   objval =   5.663121943e+04   infeas =   1.350197910e+01 (6)
|    24:   objval =   5.663125431e+04   infeas =   1.531555092e+02 (6)
|    25:   objval =   5.663126030e+04   infeas =   9.709632246e+01 (6)
|    26:   objval =   5.663126030e+04   infeas =   4.197406211e+01 (6)
|    27:   objval =   5.663128490e+04   infeas =   5.419948128e+01 (6)
...
|   164:   objval =   5.663229167e+04   infeas =   1.779560770e-12 (6)

What I had hoped that the intermediate solution tend first to be much 
better then the last one and at the end there should be less 
improvements. I had thought that steep pricing should be what I wanted, 
but I suppose I was wrong.

The problem is sparse and based on a TSP instance. The additional 
constraint was get by setting one variable >= 1.

What is the reason that my hopes are not fulfilled?

1. The aimed behaviour is not possible at all to be realized.
2. With the current implementation in glpk it is not possible.
3. Normally glpk will do it right, it must be a instance-specific 
behaviour.

Do someone know other accessible software, which have the aimed 
behaviour? Is there literature on it, so that I may consider to 
implement it by myself?

I hope it made it clear, what I want.

Best regards
Harald.