Home 21090308 Forums Numerical Method Optimization SQP – “cannot perform quadratic programming” Reply To: SQP – “cannot perform quadratic programming”

#2912
haksunli
Keymaster

My email is consistent with the javadoc description.

epsilon1 is for the whole SQP solver

The way to determine a stop on the algorithm is:

d is the difference between the new and old solutions.

epsilon2 is for the QP solver, QPPrimalActiveSetMinimizer.

The algorithms to find a feasible solution can be found here
LinearGreaterThanConstraints.getFeasibleInitialPoint.

* “Jorge Nocedal, Stephen Wright, “p. 473,” Numerical Optimization.”
* “Andreas Antoniou, Wu-Sheng Lu, “Eq 11.25, Quadratic and Convex Programming,” Practical
* Optimization: Algorithms and Engineering Applications.”
* http://www.mathworks.com/help/toolbox/optim/ug/brnox7l.html (initialization)

* Given a collection of linear greater-than-or-equal-to constraints as well as a collection of
* equality constraints,
* find a feasible initial point that satisfy the constraints.
* This implementation solves eq. 11.25 in the reference.
* The first (n-1) entries consist of a feasible initial point.
* The last entry is the single point perturbation.

So, the last entry needs to be bigger than epsilon2.