Handling constant in objective for max problems v0.18

[GillesBareilles]

Hi there,

I'm having some trouble when solving maximisation problems with a non null constant with JuMP; here is a small example, along with the output.

using JuMP, SCS

function run_SDPpb(cst, optsense)
    # m = Model(solver=MosekSolver(LOG=0))
    m = Model(solver=SCSSolver(verbose=0))

    @variable(m, Xs1[1:3,1:3], SDP)
    @variable(m, Xs2[1:1,1:1], SDP)
    @variable(m, Xs3[1:1,1:1], SDP)

    @constraint(m, Xs1[1,1] == 1)
    @constraint(m, Xs2[1,1] + Xs1[2,2] + Xs1[3,3] == 4)
    @constraint(m, Xs3[1,1] - Xs1[2,1] - Xs1[3,1] == 0)

    # Find upper bound
    @objective(m, optsense, -Xs1[1,3]+cst)
    solve(m)
    return getobjectivevalue(m)
end

@show run_SDPpb(0., :Min)
@show run_SDPpb(0.5, :Min)

@show run_SDPpb(0., :Max)
@show run_SDPpb(0.5, :Max)

Which gives the following output:

run_SDPpb(0.0, :Min) = -2.000000252312453
run_SDPpb(0.5, :Min) = -1.5000002523124532
run_SDPpb(0.0, :Max) = 1.4141624244050313
run_SDPpb(0.5, :Max) = 0.9141624244050313

When solving maximization problems max (f(x) + c), it seems value returned by getobjectivevalue is max(f(x)) - c.

More precisely, JuMP seems to only pass minimzation problems to solvers (at least for SDP), so the objective should be max (f(x) + c) = - min (-f(x)-c) = (-min -f(x) ) + c, while it seems to be (-min -f(x) + c) = (-min -f(x) ) - c.

I am working with

• JuMP 0.18.2
• MathProgBase 0.7.2

[mlubin]

Yes, this looks like a bug. Thanks for reporting!

[mlubin]

Fixed on the release-0.18 branch by #1727. The bug does not apply to master.