Let $\emptyset\ne A\subset\mathbb{R}^n$ be open and let $f \in C^1 (A, \mathbb{R})$ be a function. Let $\emptyset\ne K\subset A$ be compact and convex. I want to prove that $f$ is Lipschitz on $K$; that is, prove there exists a constant $c > 0$ such that $| f ( x ) - f( y ) | \leq c \, \| x - y \|, \forall x , y \in K$.

My approach:

Let $x,y\in K$ be two arbitrary points. Then, since $K$ is convex, the line segment between $x$ and $y$, i.e. $Co(x,y)$, is in $K$. Thus, by MVT, there exists a vector $b\in Co(x,y)$ such that

$$\text{(*) } |f(x)-f(y)|=|\langle x-y, (\nabla f)(b)\rangle|\le \|x-y\|\|(\nabla f)(b)\|$$

Now, since $K$ is compact, $f$ takes a maximum and a minimum values on $K$, so that $\exists b'\in K$ such that $\|(\nabla f)(b') \|\ge \|(\nabla f)(b) \|$, for all $b\in K$. Let $c\in \mathbb{R}$, $c:=(\nabla f)(b')$, then

$$|f(x)-f(y)|\le\|x-y\|\|(\nabla f)(b)\|\le\|x-y\|\|(\nabla f)(b')\|=c\|x-y\|$$

This implies that $f$ is Lipschitz on $K$ for all $x,y\in K$.

Please let me know if you think my proof is correct or not very much? I'm somewhat concerned about the part with the gradient - how exactly is the maximality of the norm of the gradient related to the EVT, that is to $f$ taking maximum and minimum values? As far as I can tell, the maximum norm of the gradient exists because that is the direction to the maximum (or minimum) point of $f$.

  • $\begingroup$ What is $a$? It seems undefined to me . . . $\endgroup$ – Robert Lewis Jun 28 '17 at 0:24
  • $\begingroup$ Sorry, it was supposed to be $y$. Fixed. $\endgroup$ – sequence Jun 28 '17 at 0:26
  • $\begingroup$ I guess is because $||f'||$ is a real function on a compact set? $\endgroup$ – Li Chun Min Jun 28 '17 at 0:41
  • $\begingroup$ @LiChunMin Can you please clarify your question? $\endgroup$ – sequence Jun 28 '17 at 0:47
  • 1
    $\begingroup$ Looks good to me. You do not need the EVT. Since $\nabla (f) $ is continuous, the real-valued function $\|\nabla (f)\|$ is continuous .So $T=\{\| \nabla f(x)\|: x\in K\}$ is compact (because $K$ is compact) , so $T$ is a bounded real subset....So $\|f(x)-f(y)\|\leq \|x-y\|\cdot \sup T$ for all $x,y\in K$. $\endgroup$ – DanielWainfleet Jun 28 '17 at 2:48

Here's how I would do it:

for $x, y \in K$, let $\gamma(t):[0, 1] \to K$ be the line segment

$\gamma(t) = x + t(y - x); \tag{1}$


$\gamma(0) = x, \tag{2}$

$\gamma(1) = x +(y - x) = y, \tag{3}$


$\gamma'(t) = y - x; \tag{4}$

furthermore, since $K$ is convex, $\gamma(t)$ lies entirely within $K$, hence also in $A$.

Now, for $x, y \in K$, we have:

$\Vert f(y) - f(x) \Vert = \Vert \displaystyle \int_0^1 \dfrac{d(f(\gamma(t))}{dt}dt \Vert = \Vert \displaystyle \int_0^1 \nabla f(\gamma(t)) \cdot \gamma'(t) dt \Vert$ $\le \displaystyle \int_0^1 \Vert \nabla f(\gamma(t)) \cdot \gamma'(t) \Vert dt \le \displaystyle \int_0^1 \Vert \nabla f(\gamma(t)) \Vert \Vert \gamma'(t) \Vert dt.\tag{5}$

Since $K$ is compact and $\nabla f \in C^0(A, \Bbb R)$, so hence $\nabla f \in C^0(K, \Bbb R)$, $\Vert \nabla f \Vert$ is bounded by some $B$ on $K$, hence

$\displaystyle \int_0^1 \Vert \nabla f(\gamma(t)) \Vert \Vert \gamma'(t) \Vert dt \le \displaystyle \int_0^1 B \Vert \gamma'(t) \Vert dt; \tag{6}$

using (4),

$\displaystyle \int_0^1 B \Vert \gamma'(t) \Vert dt = \displaystyle \int_0^1 B \Vert y - x \Vert dt = B \Vert y - x \Vert; \tag{7}$

bringing together (5), (6), and (7) yields

$\Vert f(y) - f(x) \Vert \le B \Vert y - x \Vert, \tag{8}$

that is, $f(x)$ is Lipschitz continuous on $K$.

  • $\begingroup$ Thank you, this is very elegant. Do you think my proof could possibly be correct? $\endgroup$ – sequence Jun 28 '17 at 1:32
  • 1
    $\begingroup$ Yes, your method seems fine. Note it agrees with mine. $\endgroup$ – Robert Lewis Jun 28 '17 at 1:39
  • 1
    $\begingroup$ "Since $K$ is compact and $\nabla f \in C^0(A, \Bbb R)$, so hence $\nabla f \in C^0(K, \Bbb R)$, $\Vert \nabla f \Vert$ is bounded by some $B$ on $K$.." Is there a proof for that or is it trivial? $\endgroup$ – abuchay May 14 '18 at 14:25
  • 1
    $\begingroup$ @abuchay: Well, $\nabla f \in C^0(K)$ since $\nabla f \in C^0(A)$ and $K \subset A$. $\Vert \nabla f \Vert \in C^0(K)$ since $\Vert \cdot \Vert$ is itself continuous in its argument. Then the boundedness follows since continuous functions on compacta are bounded, which is standard result in elementary topology. $\endgroup$ – Robert Lewis May 14 '18 at 15:40

The convexity of $K$ is not needed. Suppose the conclusion fails. Then for each $m\in \mathbb N,$ there exist $y_m,x_m \in K$ such that

$$\tag 1 |f(y_m)- f(x_m)| > m|y_m-x_m|.$$

Because $K$ is compact, we can find a subsequence $m_k$ such that the sequences $y_{m_k},x_{m_k}$ converge to points $y,x\in K$ respectively.

Suppose $y\ne x.$ Then $|y-x| > 0.$ For large $k$ we then have

$$|f(y_{m_k})- f(x_{m_k})| > m_k|y_{m_k}-x_{m_k}|> m_k(|y-x|/2)\to \infty$$

This implies $f$ is not bounded on $K.$ But $f$ is continuous on $A,$ hence is continuous on $K,$ hence $f$ is bounded on $K$ by compactness. This contradiction shows $y=x.$

Because $A$ is open we can choose $r>0$ such that $B(x,r)\subset A.$ For large $k$ we then have $y_{m_k},x_{m_k}$ in the compact convex set $\overline {B(x,r/2}) \subset A.$ By $(1),$ your highlighted mean value inequality then fails in this last set, contradiction. Therefore $(1)$ cannot hold, proving the result.

  • $\begingroup$ I dont understand why $m_k(|y-x|/2)\to \infty$. Why does it tend to infinity? If $m_k$ is a subsequence of $y_{mk}$ then $m_k$ also converges to $y$. Then, how can $m_k$ be a subsequence of $x_{mk}$ too? $\endgroup$ – John Keeper Oct 24 '18 at 9:46
  • $\begingroup$ @JohnKeeper $m_k$ is a subsequence of $1,2,\dots $ hence $\to \infty$ $\endgroup$ – zhw. Oct 24 '18 at 21:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.