The steering's constants (IOptions.steer), as KNITRO-Active applies
Algorithm 5.1 (Byrd, Nocedal and Waltz, p. 13 of the 2007 technical
report): a step keeps feasibility = 0.1 of the best linearised
reduction of the violation (their epsilon_1 in (5.13)), the model's
decrease is merit = 0.5 of the penalty times the step's reduction (their
epsilon_2 in (5.14)), and the penalty grows tenfold until both hold;
increases bounds the growth where the objective falls under the
merit's rounding, sixteen tenfold increases, the decimal digits of a
double.
The steering's constants (
IOptions.steer), as KNITRO-Active applies Algorithm 5.1 (Byrd, Nocedal and Waltz, p. 13 of the 2007 technical report): a step keepsfeasibility= 0.1 of the best linearised reduction of the violation (their epsilon_1 in (5.13)), the model's decrease ismerit= 0.5 of the penalty times the step's reduction (their epsilon_2 in (5.14)), and the penalty grows tenfold until both hold;increasesbounds the growth where the objective falls under the merit's rounding, sixteen tenfold increases, the decimal digits of a double.