Skip to main content

How to write Max[] constraint in model?

Answered

Comments

8 comments

  • Official comment
    Simranjit Kaur
    • Gurobi Staff
    This post is more than two years old. Some information may not be up to date. For current information, please check the Gurobi Documentation or Knowledge Base. If you need more help, please create a new post in the community forum, or try Gurobot, our chatbot interface offering instant, expert-level support.
  • Ismail Gokay Dogan
    • Gurobi-versary
    • First Comment

    Hi Haoxing,

    I could not fully understand the constraint that you are trying to generate, but have you tried using max_() function? You can find the documentation here:

    https://www.gurobi.com/documentation/current/refman/py_max_.html

    In your case I believe you can represent your constraint like (0.5 Z + max_((Z-2), constant = 0 ) + ... + max_((Z-20), constant = 0)

    0
  • Jaromił Najman
    • Gurobi Staff

    To add \(\max\) constraints in Gurobi, you will have to introduce auxiliary variables for each \(\max\) term. This means that your expression becomes

    \[\begin{align*}
    (0.5 \cdot Z &+ a_1 + a_2 + \dots + a_{10})/Z \\
    a_1 &= \max \{0, b_1\} \\
    b_1 &= Z - 2 \quad(*)\\
    a_2 &= \max \{0, b_2\}\\
    b_2 &= Z - 4 \quad(*)\\
    \dots\\
    a_{10} &= \max \{0, b_{10}\}\\
    b_{10} &= Z - 20 \quad(*)\\
    \end{align*}\]

    Note that you might need to adjust the lower bounds of the \(b_i\) variables, because the default lower bound for variables in Gurobi is \(0\). Once you have implemented all auxiliary variables \(a_i, b_i\) and all \((*)\) equality constraints, you can use the addGenConstrMax method to implement the remaining \(\max\) constraints.

    For the division by \(Z\), please refer to the Knowledge Base article How do I divide by a variable in Gurobi?

    Best regards, 
    Jaromił

    0
  • HAOXING OUYANG
    • Gurobi-versary
    • First Comment
    • First Question

    Hi Jaromił,

    Thanks for your reply. I followed your instructions to add Max constraints, it worked.

    But I am facing the problem with the division constraint. 

    According to your reply, I refer to the Knowledge Base article How do I divide by a variable in Gurobi?

    The formula I wrote above is just a simplified version. In my model, I created an Integer variable: sumZ[i][k], and I want to divide by sumZ[i][k], so I have to calculate its reciprocal: reverseSumZ[i][k]. 

    Firstly, I creating these tow varialbes: 

    GRBVar sumZ[][] = new GRBVar[instance.nodesNum][instance.vehiclesNum];
    GRBVar reverseSumZ[][] = new GRBVar[instance.nodesNum][instance.vehiclesNum];

    Secondly, I built their bounds and types: 

    for (int k = 0; k < instance.vehiclesNum; k++) {
    for (int i = 0; i < instance.nodesNum; i++) {
    sumZ[i][k] = model.addVar(0, GRB.INFINITY, 0.0, GRB.INTEGER, "sumZ_" + i + "_" + k);
    reverseSumZ[i][k] = model.addVar(0, 1, 0.0, GRB.CONTINUOUS, "reverseSumZ_" + i + "_" + k);
    }
    }

    Thirdly, I created the constraint: 

    //reverseSumZ[i][k] * sumZ[i][k] = 1
    for (int k = 0; k < instance.vehiclesNum; k++) {
    for (int i = 1; i < instance.nodesNum - 1; i++) {
    GRBQuadExpr expr = new GRBQuadExpr();
    expr.addTerm(1, reverseSumZ[i][k], sumZ[i][k]);
    model.addQConstr(expr, GRB.EQUAL, 1, "internalVar_adverse_" + i + "_" + k);
    }
    }

    The range for i is from 1 to instance.nodesNum -1, to make sure that the sumZ[i][k] is not equal to zero.

    But running the program will prompt that the model has no solution. And I ensure that the model only has no solution after adding this constraint.

    Can you help me answer this?

    Best regards,

    Haoxing 

    0
  • HAOXING OUYANG
    • Gurobi-versary
    • First Comment
    • First Question

    Hi Jaromił,

    I would like to add is:

    Integer variable sumZ[i][k] >=0, only when i∈N \ {0,n+1} will sumZ[i][k] ≠ 0. 

     

    0
  • Jaromił Najman
    • Gurobi Staff

    Hi Haoxing,

    But running the program will prompt that the model has no solution. And I ensure that the model only has no solution after adding this constraint.

    Is your model declared infeasible? Did you try to determine why it is infeasible, see How do I determine why my model is infeasible? and How do I use 'compute IIS' to find a subset of constraints that are causing model infeasibility? The 0 in your reverse sum might be an issue, because this would mean that you allow division by 0.

    Integer variable sumZ[i][k] >=0, only when i∈N \ {0,n+1} will sumZ[i][k] ≠ 0. 

    I am not sure what you mean. Do you mean to add constraint sumZ[i][k] >=0 for all i∈N \ {0,n+1}? Or do you mean to add sumZ[i][k] >=0 only when sumZ[i][k] ≠ 0?

    Best regards, 
    Jaromił

    0
  • HAOXING OUYANG
    • Gurobi-versary
    • First Comment
    • First Question

    Hi Jaromił,
    Thanks for your reply. I am sorry for my confusious  answered.
    In my model, sumZ[i][k] is >=0(sumZ[i][k] = 0 ← when i == 0 && i == instances.nodesNum), and I want to calculate its count backwards, so I construct a constraintin:

    //reverseSumZ[i][k] * sumZ[i][k] = 1
    for (int k = 0; k < instance.vehiclesNum; k++) {
        for (int i = 1; i < instance.nodesNum - 1; i++) {
            GRBQuadExpr expr = new GRBQuadExpr();
            expr.addTerm(1, reverseSumZ[i][k], sumZ[i][k]);
            model.addQConstr(expr, GRB.EQUAL, 1, "internalVar_adverse_" + i + "_" + k);
        }
    }

    In the constraint I set the range for i between 1 and instamce.nodesNum-1, to avoid divide by 0. 
    I think this is reasonable in logical, but the model is infeasible, that I consider the situation of dividing by 0 has not been ruled out yet.

    0
  • Jaromił Najman
    • Gurobi Staff

    I think this is reasonable in logical, but the model is infeasible, that I consider the situation of dividing by 0 has not been ruled out yet.

    In this case, you should try to follow the steps in  How do I determine why my model is infeasible? and How do I use 'compute IIS' to find a subset of constraints that are causing model infeasibility? With that you should be able to find the source of infeasibility for your model.

    0

Post is closed for comments.