24

Is a Lipschitz function differentiable?

I have been wondering whether or not this property applies to all functions.

I do not need a formal proof, just the concept behind it.
Let $f: [a,b] \to [c,d]$ be a continuous function (What is more - it is uniformly continuous!) And let's assusme that it's also Lipschitz continuous on this interval.

Does this set of assumptions imply that $f$ is differentiable on $(a,b)$?

Guy Fsone
  • 25,237
Aemilius
  • 3,759
  • 6
    It only implies it is differentiable almost everywhere. – Ian Jan 17 '18 at 22:13
  • 7
    "Let $f: [a,b] \to [c,d]$ be a continuous function (What is more - it is uniformly continuous!) And let's assusme that it's also Lipschitz continuous on this integral." You should be aware that these assumptions are highly redundant. Lipschitz implies all preceding assumptions – zhw. Jan 17 '18 at 22:42
  • 2
    Lipschitz continuous does not imply differentiability. In fact, we can think of a function being Lipschitz continuous as being in between continuous and differentiable, since of course Lipschitz continuous implies continuous. If a function is differentiable then it will satisfy the mean value theorem, which is very similar to the condition to be Lipschitz continuous. – Jürgen Sukumaran Jan 18 '18 at 09:43

3 Answers3

39

It is not always true indeed, good counterexample could be $x\mapsto |x-a|$. But rather, we have

Theorem: Radamacher theorem says every Lipschitz function is almost everywhere differentiable

Fine a nice proof of this theorem here: An Elementary Proof of Rademacher's Theorem - James Murphy or Here using distribution theory

Guy Fsone
  • 25,237
  • 2
    those do not include a proof of the Rademacher theorem in dimension 1. Terrence Tao has a reasonably self-contained proof of this and a couple other related here: https://terrytao.wordpress.com/2010/10/16/245a-notes-5-differentiation-theorems/#more-4290 – Calvin Khor Feb 02 '18 at 23:08
  • @CalvinKhor I don't think those proof exclude the one dimensional case. Or am I wrong . one the proof use distributional derivative – Guy Fsone Feb 02 '18 at 23:49
  • Well I may have missed it but both links start with "in dimension one its because Lipschitz implies absolutely continuous/bounded variation so the result is true", which isn't wrong but I doubt someone who would ask this question on MSE would know this result for AC or BV functions – Calvin Khor Feb 03 '18 at 00:00
  • 1
    yes many proofs of Rademacher in R^n rely on the fact having been established on R. To prove it on R, one of the the underlying truths is that a continuous and increasing function is differentiable almost everywhere. In fact an increasing function is differentiable almost everywhere, but this isn't needed for Rademacher's theorem. – SBK May 15 '20 at 15:13
20

The function $$x \mapsto \left|x\right|$$ is Lipschitz-continuous (with $k=1$) but not differentiable at $0$.

AccidentalFourierTransform
  • 2,662
Atmos
  • 7,455
6

No, it does not imply $f$ is differentiable.

Try $f(x) = |x|$ as an example!

Caleb Stanford
  • 47,093