Starting solution for LP problem

Answered

Paul van Kessel

• January 28, 2020 09:28

Hello,

Currently I modelling a manpower assignment problem. In the initial run I don't provide any starting solution. Afterwards, I model a disruption by makings a few changes to the model. I would like to use the optimal solution of the previous model as a starting point for the new model. Since there are multiple equivalent solutions, there are some random differences between the two model solutions which have nothing to do with the disruption that I provide. 

For my LP problem, Gurobi doesn’t seem to use the initial solution that I provide and solves the problem by itself. Only when I change the variables types in the objective function from continuous to integer, Gurobi starts to use the initial solution that I provide. My guess is that Gurobi only accepts initial solutions if it applies branch & bound.

Is it also possible to provide an initial solution to an LP problem instead of an (M)ILP problem?

I can get the problem to work as an (M)ILP problem since I only need two digits for my continuous variables. So multiplying my decision variables by a 100 and change them from continuous to integer will do the trick.

But the downside of using an MILP is that the computational time slightly increases.

0

• 

  Silke Horn

  Gurobi Staff

  • January 29, 2020 10:51
  • Edited

  MIP starts or the variables' Start attribute are only available for MIPs. They are ignored for LPs.

  For LPs, you can supply warm-start information by using VBasis/CBasis or PStart/DStart.

  0

Please sign in to leave a comment.