Say we have a function $f: \mathbb{R}^n \to \mathbb{R}^m$ which has differentiable, and its total derivative at some point $x \in \mathbb{R}^n$ is given by $Df(x)$. What does it mean for $|Df|(x)$ to be bounded. Obviously, in one dimension we can view the derivative as a scalar valued function, so we have an idea of what boundedness is. How do we understand boundedness in higher dimensions?

My point being: how can we understand (by looking at the total derivative) when a multivariable function is Lipschitz continuous?

  • $\begingroup$ One guess that came to my mind is the boundedness of each entry of $Df(x)$ - without context of how this was used, it might be tough. I've never encountered this convention. $\endgroup$ – yoshi Jun 7 at 2:09
  • $\begingroup$ I've edited a typo, if that helps. $\endgroup$ – gtoques Jun 7 at 2:11

For convenience, denote $V = \mathbb{R}^n$, and $W = \mathbb{R}^m$, and let $\mathcal{L}(V,W)$ be the vector space of linear transformations from $V$ into $W$. For each $x \in V$, $Df(x)$ is a linear transformation from $V$ into $W$; or said differently, $Df(x) \in \mathcal{L}(V,W)$. On $\mathcal{L}(V,W)$, we can define a norm via the rule \begin{equation} \lVert T \rVert = \sup \{ \lVert T(\xi) \rVert: \lVert \xi\rVert \leq 1 \}. \end{equation} (There are 3 different norms in the equation above, all of which I've denoted by $\lVert \rVert$, so hopefully you know which norm is acting on which space).

So, one possible notion of boundedness of the derivative is the following: there is an open ball $B \subset V$, and a number $M > 0$, such that for all $x \in B$, we have $\lVert Df(x) \rVert \leq M$. Under this assumption, you can prove via the mean-value theorem$^1$ that $f$ is Lipschitz continuous on $B$.

One special case is the following: suppose all the partial derivatives $\partial_jf_i$ are continuous. Then, $Df: V \to \mathcal{L}(V,W)$ is continuous (see the book linked below for a proof of this). Then, given a compact convex set $K \subset V$ (such as a closed ball or a closed rectangle), since $\lVert Df(\cdot) \rVert$ is continuous, by the extreme value theorem, this function attains a maximum value $M$ on $K$. Then, you can apply the MVT as in the previous paragraph to conclude $f$ is Lipschitz on $K$.

Of course, you can generalize these to situations with weaker hypotheses etc, but this relatively simple formulation (or a minor variant of it) has served me well for many applications

[1.] For a statement and proof of the MVT, see Loomis and Sternberg's Advanced Calculus, Chapter 3, Theorem 7.4.


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.