The Sequential Quadratic Programming (SQP) Algorithm Given a solution estimate xk, and a small step d, an arbitrary numerical optimization problem can be approximated as follow: f(xk+d)=f(xk)+[▽f(xk)] T*d + 1/2*(dT)[▽2f(xk)]*d+.... h(xk+d)=h(xk)+[▽h(xk)]T*d + 1/2*(dT)[▽2h(xk)]*d+.... = 0 g(xk+d)=g(xk)+[▽g(xk)]T*d + 1/2*(dT)[▽2g(xk)]*d+.... >= 0 where x=[x1,x2,…xk]T, d=[d1,d2,…dk]T Form the linearly-constrained/quadratic minimization problem: Minimize: f(xk)+[▽f(xk)]T*d + 1/2*(dT)[▽2f(xk)]*d Subject to: h(xk)+[▽h(xk)]T*d = 0; g(xk)+[▽g(xk)]T*d >=0; In the SQP loop, the approximate QP should be a convex Quadratic Programming, in which the matrix Q = ▽2f(xk) should be positive semidefinite, Q ≥ 0. Actually, the Q is the Hessian matrix of the function f(x) at the point xk.

http://blog.csdn.net/zc02051126/article/details/8588644 中用到的资源

量子状态编程第2版，英文原档翻印，英文差的朋友可以参考！ 好书不多说了！你懂的！