Originally posted by: NashVio

Dear all,

I'am working on an optimization problem using C++  and have trouble using multiple mipstarts,

I have 10 different mipStarts. Each of them is feasible, and if I use only one of the every thing is fine.

But if I add all mipstarts I receive
"1 of 10 MIP starts provided solutions."

even though all of them are feasible and i expected "10 of 10..."

- I added the starts using     cplex.addMIPStart(varsarray, valsarray);

- Using cplex.writeMIPStarts("mipStart", 0, 10); shows me that all starts are in the cplex object.

Where did I go wrong? Why is there a difference if I use them indivual compared to using the alltogether?

Thank you very much in advance!

Tobi

The order of the MIP starts matters.

If the first MIP start provides the best objective function value then the other MIP starts will not be considered "providing a solution", even if they are feasible. If you sort your MIP starts by objective value (so that the best start is the last one) then you should see "10 of 10 MIP starts provided a solution".

Originally posted by: NashVio

Thank you for the fast answer.

The problem is the first MIP start does not provide the best solution.  Most of the later  are better.

That is odd. What are the effort levels you have set for the starts? Are these MIP starts with all binary/integer variables fixed or are these only partial MIP starts?

Can you share your model and the MIP starts? If not in public then maybe with daniel(dot)junglas(at)de(dot)ibm(dot)com?

Originally posted by: NashVio

It have been only partil MIP start. I change this, so that now all variables are fixed.  The I also set the effort levels to  "MIPStartSolveFixed", know it works out.

Thank you for your hints :)

Originally posted by: VictorAS

Hello.

I had a similar problem, but now I understood the issue: the order of the MIP starts matters if one wants to see all starts reported by CPLEX.

Another doubt is: I supplied CPLEX with 30 definitely feasible MIP starts. The CPLEX log picture is attached. As you see, it says "29 of 30 MIP starts... ". I was confused with that, then I realized that two of my solutions among the 30 have identical objective value. May I conclude that CPLEX deems them as only one?

Thanks!

Yes, the order matters. A MIP start is only considered to provide a feasible solution if the solution found from it is feasible and better than the current incumbent (which may have come from a previous MIP start).

Originally posted by: VictorAS

Great! Thanks, Daniel!