non-linear programming - constraints

[Dupuis]

Hello,

I'm trying to solve a non-linear minimisation problems where elements to
determine are part of a matrix: M = [a b; b c], with a and c >0. The
objective function requires that M be positive definite, i.e. ac - b^2 >=0.
I introduced an inequality constraint in sqp, yet during the search in
phi_L1, x sometimes is not acceptable with respect to the inequality
constraint. Is there a mean to strictly enforce the inequality constraint to
be valid ? Another way would be to take beff = sign(b)*(min(abs(b),
sqrt(a*c)) but I fear to introduce discontinuities in the inequality
function. Any hint ?
Regards

Pascal 
-- 
View this message in context: 
http://www.nabble.com/non-linear-programming---constraints-tp25398227p25398227.html
Sent from the Octave - General mailing list archive at Nabble.com.