Error evaluating objective function

[fgallagher27]

I have a pyomo model using ipopt to solve and an objective function that sums over some indexes. Within the sum, it calculates (a * b * c) ** d.

a, b, c are all positive (a,b are pyomo variables with Domain=pyo.PositiveReals and an additional constraint a,b >= 1e-6. c is a positive parameter.

d is a constant parameter currently set to 1.5. I do not want to have to set this to be a positive integer as there is no theoretical justification for this in my context.

For some parameter values and therefore model guesses, ipopt fails to solve and throws the error Error evaluating objective function: can't evaluate pow(-3.17431e-06,1.5). I have interpreted this as an error being triggered because a negative value to the power of 1.5 is undefined.

Given all the elements of my objective function have to be positive, how has this evaluation occurred and is there a numerical tolerance option I can use to avoid this?

Thanks very much.

[svigerske]

Ipopt relaxes given variable bounds by a small amount (option bound_relax_factor), but this should not move a lower bound of 1e-6 below 0. By default, it should just move to 1e-6 - 1e-14.

So I cannot say why it evaluates at -3e-6. You could increase the print_level to see what iterate Ipopt is currently looking at, e.g., the values of a and b.

If you didn't add a,b >= 1e-6 as bounds, but really as inequality constraints, then change this to variable bounds.

[fgallagher27]

Thanks for the reply! With the print level is there a way to see the values of the failed iteration - with print_level = 8 I only seem to be able to see all the successful iterations, not the failed one with its guesses?

[svigerske]

I thought it would print also trial points.

If you are able to set option halt_on_ampl_error, then that may be useful to stop at the right point.