2
$\begingroup$

Hello,

If $f:\mathbb{R} \to \mathbb{R}$ a differentiable function, it is very easy to find its Lipschitz constant. Is there any way to extend this to functions $f: \mathbb{R} \to \mathbb{R}^n$ (or similar)?

$\endgroup$
2
  • 1
    $\begingroup$ You probably mean "continuously differentiable" (i.e. differentiable with continuous first derivative) and "local Lipschitz constants" (see Wikipedia page on Lipschitz for more details). In which case, the answer is "Yes". $\endgroup$ Jan 12, 2010 at 10:33
  • 1
    $\begingroup$ It is not true that any differentiable function $f: \mathbb{R} \rightarrow \mathbb{R}$ is Lipschitz: $f(x) = x^2$ is a counterexample. A continuously differentiable function $f: [a,b] \rightarrow \mathbb{R}$ is Lipschitz, because its derivative is bounded. $\endgroup$ Jan 12, 2010 at 10:34

3 Answers 3

2
$\begingroup$

Let $f = (f_1,\ldots,f_n): [a,b] \rightarrow \mathbb{R}^n$ be a continuously differentiable function. (See the comments above for an explanation as to why the hypotheses have been strengthened.)

For $1 \leq i \leq n$, let

$L_i = \max_{x \in [a,b]} |f_i'(x)|$,

so that, by the Mean Value Theorem, for $x,y \in [a,b]$,

$|f_i(x)-f_i(y)| = |f_i'(c)||x-y| \leq L_i |x-y|$.

Then, taking the standard Euclidean norm on $\mathbb{R}^n$,

$|f(x)-f(y)|^2 = \sum_{i=1}^n |f_i(x)-f_i(y)|^2 \leq (\sum_{i=1}^n L_i^2) \ |x-y|^2$,

so

$|f(x)-f(y)| \leq \sqrt{(\sum_{i=1}^n L_i^2)} \ |x-y|$.

Thus we can take

$L = \sqrt{\sum_{i=1}^n L_i^2}$.

Since all norms on $\mathbb{R}^n$ are equivalent -- i.e., differ at most by a multiplicative constant -- the choice of norm on $\mathbb{R}^n$ will change the expression of the Lipschitz constant $L$ in terms of the Lipschitz constants $L_i$ of the components, but not whether $f$ is Lipschitz.

$\endgroup$
1
  • $\begingroup$ In my answer above, I interepreted "its Lipschitz constant" as "an explicit constant which makes the function Lipschitz", not the least such constant. Off-hand it seems that the one-dimensional constants $L_i$ I gave are sharp but that $L$ I gave need not be. This makes me think that Leonid's answer is better than mine... $\endgroup$ Jan 12, 2010 at 16:50
2
$\begingroup$

In fact a statement similar to what was described by Pete Clark is true for all normed vector spaces: Let $X$ and $Y$ be normed vector spaces. A (total) differentiable function $f:X\to Y$ is Lipschitz iff its derivative is bounded. Every upper bound for the differential is a Lipschitz constant.

One direction follows from the mean value theorem: $\|f(x)-f(y)\|\leq \|Df(\xi)\|\cdot\|x-y\|$ for some $\xi$ on the straight line from $x$ to $y$.

The other follows immediately from $Df(x)=\lim_{h\to 0}\frac{f(x)-f(x+h)}{\|h\|}$

$\endgroup$
0
$\begingroup$

Another nice way to do this computation (which generally gives more precise information) is to use the formula

$ f(x + h) - f(x) = [ \int_0^1 Df(x + th) dt ] h$

Also, the formula for $Df(x)$ in Hahn's post above is not correct.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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