凸函数的局部最优也是全局最优的证明_强凸函数的最优解是唯一的证明-CSDN博客

本文链接：https://blog.csdn.net/robert_chen1988/article/details/139534616

这个性质早就知道了，但并不太清楚严谨的证明是什么。这也是《Introduction to linear optimization》书中第三章课后题的第一题。这篇博客给出严谨的证明。

Exercise 3.1 (Local minimum of convex functions)

Let $\mathcal{R}^n \rightarrow \mathcal{R}$ be a convex function and let $\in \mathcal{R}^n$ be a convex set. Let $x^\ast$ be an element of $S$ . Suppose that $x^\ast$ is a local optimum for the problem of minimizing $f (x)$ over $S$ ; that is, there exists some $\epsilon>0$ such that $f(x^\ast)\leq f(x)$ for all $x\in S$ for which $x^\ast|\leq \epsilon$ . Prove that $x^\ast$ is globally optimal; that is, $f(x^\ast)\leq f(x)$ for all $x\in S$ .

Solution:

We prove this problem by contradiction (反证法).

Proof.

Suppose that there exists a point $x_0\in S$ in which $f(x_0)<f(x^\ast)$ . Since $f (x)$ is a convex function, for any $\lambda\in [0,1]$ ,

$f(\lambda x^\ast + (1-\lambda)x_0)\leq \lambda f(x^\ast) + (1-\lambda )f(x_0).$

Since $f(x_0)<f(x^\ast)$ ,

$f(\lambda x^\ast + (1-\lambda)x_0)< \lambda f(x^\ast) + (1-\lambda )f(x^\ast)=f(x^\ast).$

We can find a $\lambda$ that makes $\lambda x^\ast + (1-\lambda)x_0$ locates at the neighbourhood of $x^\ast$ , i.e.,
$x^\ast-\epsilon\leq\lambda x^\ast + (1-\lambda)x_0\leq x^\ast+\epsilon;$

i.e., we can make $\lambda$ satisfies:

$-\epsilon\leq(1-\lambda)(x^\ast-x^0)\leq \epsilon;$

i.e., we can make $\lambda$ satisfies:
$1-\lambda\leq \frac{\epsilon}{|x_0-x^\ast|}.$

This $\lambda$ does exist, so there will exists a point in the neigourhood of $x^\ast$ but its function value is smaller than $f(x^\ast)$ , which contradicts that $x^\ast$ is a local minimum.