Impossible deduced bounds ...

[Daniel Fokum]

Hello,

When I try to solve the attached model I get a message that states:

presolve, variable S[1,6,1,3]:
    impossible deduced bounds: lower = 0, upper = -3

However, S is a binary variable. I have executed a sequence of expand commands as shown below to determine why the bounds are impossible, unfortunately I am not able to resolve this error.

ampl: expand S[1,6,1,3];
Coefficients of S[1,6,1,3]:
    VisDtct[6]               0.9
    VisFlsA[6]               0.002
    VisTimely[6]             0.96
    TxRng2[1,6,1,3,1,2]     52
    costMetric           12586.8

ampl: expand VisDtct[6];
subject to VisDtct[6]:
    0.9*S[1,6,1,3] + 0.93*S[2,6,1,3] + 0.97*S[3,6,1,3] + 0.905*S[4,6,1,3]
     + 0.915*S[5,6,1,3] + 0.925*S[6,6,1,3] + 0.895*S[7,6,1,3] >= 0.89;

ampl: expand VisFlsA[6];
subject to VisFlsA[6]:
    0.002*S[1,6,1,3] + 0.001*S[2,6,1,3] + 0.001*S[3,6,1,3] +
    0.003*S[4,6,1,3] + 0.002*S[5,6,1,3] + 0.002*S[6,6,1,3] +
    0.002*S[7,6,1,3] <= 0.005;

ampl: expand VisTimely[6];
subject to VisTimely[6]:
    0.96*S[1,6,1,3] + 0.98*S[2,6,1,3] + 0.97*S[3,6,1,3] + 0.955*S[4,6,1,3]
     + 0.965*S[5,6,1,3] + 0.95*S[6,6,1,3] + 0.96*S[7,6,1,3] >= 0.91;

ampl: expand TxRng2[1,6,1,3,1,2];
subject to TxRng2[1,6,1,3,1,2]:
    52*S[1,6,1,3] >= 16.165;

Thanks in advance for any pointers.

Daniel

[Paul]

>  SnsrAssignRMtd.mod
> 8KViewDownload
>
>  SnsrAssignRMtd7.dat
> 3KViewDownload

The LP relaxation is feasible, but with the binary restrictions the
model is infeasible. If you are convinced that the model should have
a feasible binary solution, about the only suggestion I can make is to
concoct one (it does not need a nice objective value, it just has to
be feasible), plug the values in for S, and see which constraints are
violated.

Another possibility, assuming you have the CPLEX interactive
optimizer, would be to write the model out from AMPL to an MPS file,
read it into CPLEX, and see if the CPLEX conflict refiner can point
you to a set of constraints that collectively make the problem
infeasible when S is binary.

/Paul

[Robert Fourer]

Notice that the TxRng2 constraints are like

 

   subject to TxRng2[2,6,1,3,1,2]:

      57*S[2,6,1,3] >= 16.165;

 

Since the S-variables are binary, these force variables S[2,6,1,3], S[3,6,1,3], ..., S[7,6,1,3] to be 1.  Setting them all to 1 in the constraint

 

   subject to VisFlsA[6]:

      0.002*S[1,6,1,3] + 0.001*S[2,6,1,3] + 0.001*S[3,6,1,3] +

      0.003*S[4,6,1,3] + 0.002*S[5,6,1,3] + 0.002*S[6,6,1,3] +

      0.002*S[7,6,1,3] <= 0.005;

 

and simplifying gives S[1,6,1,3] <= -3, hence the message about the impossible deduced bound "upper = -3".  There is not feasible solution possible since also as a binary variable S[1,6,1,3] >= 0.

 

Bob Fourer

4...@ampl.com

 

 

---