There is a constraint “P=Q./R”, where P and Q are vector{variableRef}, Q is array{Float64,2}
JuMP model is as follow:
using JuMP,Gurobi,LinearAlgebra
model=Model(Gurobi.Optimizer)
Q=[1,2,3]
@variable(model, P[1:3])
@variable(model, R[1:3])
@expression(model, expr1[i=1:n],Q[i]/R[i])
@constraint(model, con1[i=1:3], P[i]== expr1[i])
Error: / is not defined?!
How could I solve it using @expression not @NLexpression and with Gurobi solver?
Your constraint is in fact nonlinear, so you should use @NLconstraint as suggested by the error message.
If the constraint@constraint(model, con1[i=1:3], P[i]== expr1[i]) is written as:
@constraint(model, con1[i=1:3], R[i]*P[i]== Q[i])
@constraint(model, con2[i=1:3],R[i]>0)
@constraint(model, con3[i=1:3],R[i]<0)
May equivalent to the original constraint?
You can use R[i] * P[i]. This is still nonlinear, but the special case of quadratic expressions is supported by @constraint.
The other two constraints using < and > contradict each other and can not be satisfied simultaneously. Further, strict inequalities are not supported by solvers such as Gurobi.
Ok. Thank you for you answer!
I still have some question. For example
using JuMP, Gurobi
m=Model(Gurobi.Optimizer)
@variables m begin
fp
fq
a
v
end
@NLconstraint(m, exp(fp) <= 40)
@NLconstraint(m, exp(1/v) <= 40)
@constraint(m,fq+a <= 40)
optimize!(m)
Can’t set optimizer_attribute or other setting to solve the problem? I saw Gurobi’s official website that it can solve NLP.
Thank you for your help.
Gurobi doesn’t support arbitrary nonlinear. It is restricted to quadratic programs.
OK, I see. I need to change the nonlinear model to a linear one.
Thank you for your answer again.
You can also use a nonlinear optimizer like Ipopt (https://github.com/jump-dev/Ipopt.jl).
Nice! It is also a good method.