Newton’s Method

To derive newton’s method, we simply have to find the optimum point from second order Taylor series expansion of $f(x)$ $$\begin{aligned} x_{k+1} &= x_k - [H(x_k)]^{-1}\nabla f(x_k)^\intercal \end{aligned}$$ Derivation: From a point $x_k$, we want to compute the best possible move $x_k+s$ to minimise $f$. Using taylor series expansion, we have $$f(x_k+s) = f(x_k) + s\nabla f(x_k) + \frac{s^2}{2!} H(x_k) = g(s)$$

$$\begin{aligned} 0 &= \nabla_s g(s) = \nabla f(x_k) + s H(x_k) \\ s &= - H(x_k)^{-1} {\nabla f(x_k)}^\intercal \end{aligned}$$




tags: optimisation