Skip to main content

What is root relaxation in MILP models?

Answered

Comments

4 comments

  • Mario Ruthmair
    • Gurobi Staff

    Hi,

    For a MILP, the root relaxation is the LP relaxation at the root node of the branch-and-bound tree, i.e., the first LP relaxation that is solved after presolving.

    Best regards,
    Mario

    0
  • Tobia Marcucci
    • Gurobi-versary
    • First Comment

    Hello,

    What about a convex MIQP? I am solving an MIQP and that line of the log says

    Root relaxation: objective 1.406271e-02, 2275 iterations, 0.07 seconds (0.08 work units)

    However, if I set all the binary variables to be continuous instead, I get a convex program whose optimal value is  29.3.

    On a similar note, the convex relaxation takes 0.23 seconds to solve, while the MIQP log after 8 seconds still says

        Nodes    |    Current Node    |     Objective Bounds      |     Work
    Expl Unexpl |  Obj  Depth IntInf | Incumbent    BestBd   Gap | It/Node Time
    0     0   29.13175    0  129          -   29.13175      -     -    8s

    So it seems that it has not started the branch and bound process yet, but is still working on solving the convex relaxation. Any suggestions to speed up the solution of this MIQP?

    Thank you very much,

    Tobia

    0
  • Mario Ruthmair
    • Gurobi Staff

    Hi Tobia,

    In most cases, the quadratic objective terms in an MIQP are linearized to convert it to an MILP. This can weaken the bound obtained from the continuous relaxation. This is probably what you observe in your run.

    You can control what is done with the quadratic terms with the parameters PreQLinearize and PreMIQCPForm (if Q-constraints are present).

    Mario

    1
  • Tobia Marcucci
    • Gurobi-versary
    • First Comment

    Hi Mario,

    Thank you for the quick response. Setting PreMIQCPForm to 1 or 2 solves the issue.

    Tobia

    0

Please sign in to leave a comment.