Originally posted by: iturkmen

Hi,

I want to have a branch and bound procedure which always select the largest fractional value of the variable while selecting a variable to any branch. The basic algorithm should be like this:

1.solve LP relaxation, and find the largest fractional value of decision variables.

2.Set this variable to 1 and solve the LP relaxation. 

3. Find the largest fractional value of unselect decision variables.

4.Set this variable to 1 and solve the LP relaxation.  

5. Do this until find a integer solution. 

Is there any function(callback etc.) that I can do this procedure? or should I write a heuristic code for this? 

Thanks.

It is unclear whether you mean to do this at each node -- solve the LP relaxation, temporarily fix variables until you find a feasible solution, check whether it is a new incumbent, then unfix them and branch normally -- or whether you mean to replace the branch and bound algorithm with this (so that each time you fix a variable at 1, you create and move to a new node in the tree). If the former, you can do it with a heuristic callback. If the latter, you can use a branch callback to find examine the LP solution, select a variable and fix it to either 1 or 0 (creating two child nodes), and set the branching direction parameter (BrDir) to up (1) so that the child with the variable fixed to 1 is processed before its sibling.

Paul

Originally posted by: iturkmen

Thanks for your help Paul. I am sorry for the unclear explanation for my problem. I think I need to use heuristic callback function and this function will decide the variable in each node. I am working on heuristic call back function right now.