JuMP with Ipopt - Error "The equality constraints contain an invalid number"

Specific Domains Optimization (Mathematical)
1

Hi there,

I am trying to solve the following system of nonlinear equations

      using JuMP, Ipopt
      c = Model(with_optimizer(Ipopt.Optimizer,max_cpu_time=500.0))
     @variable(c,x)
     @variable(c,y)
     @NLconstraint(c,(1/x)*(2+(0.5)/(x-y)-y^2) == 0) 
     @NLconstraint(c,(1/(x^2))*(-2*y -(0.5)*(y/(x-y))+(1/3)*y^3+0.5*log(x-y)) == 0)
     @NLconstraint(c, x >= y + 0.001) 
     optimize!(c)

but I get an error message, “The equality constraints contain an invalid number”. I cannot understand what is being referred to.

Any hint would be most appreciated, thanks

2

x is allowed to be 0, so you have a divide by zero error with 1/x.

Ipopt assumes that your model is differentiable over the domain.

3

Hi there,

thanks for that. I did think about that: in fact, adding one more constraint

                    @NLconstraint(c, x >=  0.001)

does not improve things much, what I am missing? Ipopt assumes the model is differentiable everywhere, but can this be relaxed via NL constraints?
Will read some more documentation.

Thanks again

4

See Some JuMP variables values are out of bounds - #7 by odow

1 Like
5

Got it, working beatifully now, thanks a lot.

6

In fact, IPOPT assumes everything is differentiable everywhere, not just on the feasible set. It’s an infeasible method.