Optimization Difficulty: Advanced

Newton's Method

Newton's method uses second-order information (the Hessian) to achieve faster local convergence. It is foundational for interior-point methods and scientific computing.

Key Points

  • Newton's method solves $\nabla f(x) = 0$ using local quadratic approximations.
  • For strongly convex functions with Lipschitz Hessian, convergence is quadratic near the optimum.
  • Quasi-Newton methods (L-BFGS) approximate the Hessian to reduce cost.

Formulas

Newton update
$$x_{k+1} = x_k - [\nabla^2 f(x_k)]^{-1} \nabla f(x_k)$$
Quadratic model
$$m_k(p) = f(x_k) + \nabla f(x_k)^\top p + \frac{1}{2} p^\top \nabla^2 f(x_k) p$$

Code Example

import numpy as np

def f(x):
    return x**2 + 4*x + 4

def grad_f(x):
    return 2*x + 4

def hess_f(x):
    return 2.0

x = 0.0
for _ in range(5):
    x -= grad_f(x) / hess_f(x)
print(x)  # -2 (exact in one step)

Tags

  • second-order-methods
  • hessian

References

  • Numerical Optimization
    Jorge Nocedal and Stephen J. Wright · Springer · source
  • Convex Optimization
    Stephen Boyd and Lieven Vandenberghe · Cambridge University Press · source

Knowledge Graph