Someone explained the second order condition for optimality of a constrained optimization problem here:
So I should take the Hessian, and ZZ = nullspace of the jacobian of all active constraints, and then it is an optimal point if
Z'*Hessian*Z >= 0.
Here is my code:
[xx,~,~,~,~,~,hess] = fmincon(@(XX)-XX(1)^2-XX(2)^2,[0.3,0.3],,,[1,1],1);
The active linear constraint: x1 + x2 = Q, so I thought the Jacobian of the only actice constraint is [1,1].
Matlab's nullspace: ZZ = [-sqrt(2)/2;sqrt(2)/2)];
Z'*Hessian*ZZ is 1, but the point is not a local minimum (only stationary point). I want to proof numerically that it is only a staionary point.
What goes wrong?