[Ipopt] Constrains in Ipopt

[Teodosi Geninski]

Hi,

I am having a problem with constrains voilation in Ipopt.
For example if I have
0 < x + y < 1
in order to get a solution that takes into account those constrains I should pay attention to bound_relax_factor (for example 1e-8 as default), constr_viol_tol (for example 1e-4 as default) and acceptable_constr_viol_tol (for example 1e-2 as default).
My questions are:
1) Which bounds are honored when comparing constrain violation max-norm to constr_viol_tol? The original bounds or the relaxed bounds? Because with bounds relaxation we can never reach beyond 1 * bound_relax_factor for a single bound and hence for the max-norm if the original bounds are used.
2) What happenes with the equality constraints? Does Ipopt treat them the way they are put in Ipopt - as two inequalities and 1) from above?
3) Do the variables bounds are treated the same way (as in 1))?

If it is the way I understand it then if I have
g[0] = x+y, subject to 0 < g[0] < 2
I have to set acceptable_constr_viol_tol = constr_viol_tol and g_L[0] = 0 + constr_viol_tol and g_U[1] = 2 - constr_viol_tol and it doesn't really matter what is the value of constr_viol_tol.
And with an equation constrains there is not much of a choice but to choose bound_relax_factor as low as you can (because 1 * bound_relax_factor is as far as it can go), i.e. the default value 1e-8;

Teodosi.

-----------------------------------------------------------------
Купи сега Armani, GUCCI, Miss Sixty и други с до 70% отстъпка!
http://fashiondays.bg