a redundant constraint speeds up an MIP model, why?

Hi All,

y,s are binary

B(i,nn) is a non-negative integer; 0<= B(i,nn) <=BigM
BigM is constant positive integer

the equations below are part of a multiobjective mip problem


min y(i,nn)

B(i,nn) = sum( S(ijr) )

B(i,nn) <= BigM*y(i,nn)

y(i,nn) <= B(i,nn)

the last constraint is redundant since the problem is to minimize y
(and y is binary in the first place)
but it speeds up my model if I add it, why?

The speed was twofold increase; I have not tried other instances, but
rerun the same instance several times with a faster solve using that
last constraint. Any hints?

Thanks,

-Fawzi Sdudah

Fawzi Sdudah wrote:

>
This objective term and its constraints are the simplest and the
latest added parts in a bit hairy model.
With this objective and its associated constraints turned off, the
model solves in a short time. As long as they are turned on with the
previous solution fed as an initial solution, the solve time rockets.
However, subsequent runs of the whole stuff (including the new turned
on set ) do not show that sudden behavior of solve time surge.
>
If that analysis in the previous post is true and considering what
just mentioned above, then most likely that redundant constraint
accounts for the speed up.
>

I agree -- but probably as a result of an interaction with something
else in the "hairy" part of the model.

/Paul