24

Problem: Let $f$ be defined for all real $x$, and suppose that

$$|f(x)-f(y)|\le (x-y)^2$$

for all real $x$ and $y$. Prove $f$ is constant.

Source: W. Rudin, Principles of Mathematical Analysis, Chapter 5, exercise 1.

Jonas Meyer
  • 53,602
Potato
  • 40,171
  • 3
    This has been asked before on the site, with roughly the same answers (based on the derivative like Potato's or on the triangular inequality like Frank's). Potato: I wonder if asking questions you answer yourself immediately is part of a Grand Plan of yours. – Did Jun 30 '12 at 06:26
  • 3
    To be honest about my motives: writing up answers is good proofwriting practice, and occasionally someone spots an error I didn't. It's very educational. – Potato Jun 30 '12 at 06:27
  • Is this a declared goal of the site? – Did Jun 30 '12 at 06:30
  • @did Yes. ${}{}{}{}$ – Rudy the Reindeer Jun 30 '12 at 06:31
  • 1
    @MattN. Would you know any reference for the fact that it is? (This in my previous comment refers to proofwriting practice.) – Did Jun 30 '12 at 06:32
  • @did No apparently not. But as I understand it the site is here to help people improve their mathematical skills. And proofwriting is one. – Rudy the Reindeer Jun 30 '12 at 06:36
  • 1
    Somewhat relevant to the above comments: http://meta.math.stackexchange.com/questions/1878/are-please-check-my-proof-type-of-questions-proper – Jonas Meyer Jun 30 '12 at 06:37
  • 2
    @did: Perhaps someone should start a meta thread? Feel free to name me explicitly and link to my post history. I will stop if people want me to. – Potato Jun 30 '12 at 06:37
  • ducks away ${}{}{}$ – Rudy the Reindeer Jun 30 '12 at 06:40
  • @did I eventually find that his latest posts are probably consecutive exercises on Rudin's book. – Yai0Phah Jun 30 '12 at 06:53
  • 1
    Potato: As an aside, I note that in this stream of questions you omit any indication about what you tried, what you know, and so on, which is definitely recommended on the site. @FrankScience: Yes, this is mentioned in the questions. – Did Jun 30 '12 at 06:56
  • 2
    @did But I don't think they're all good questions, also answering one's own question is encouraged – Yai0Phah Jun 30 '12 at 06:59
  • @Frank: yes, answering one's own problem. Thanks for the thread (which does not seem to have precisely the present situation (posting one by one the exercises from a textbook) in mind...). – Did Jun 30 '12 at 07:07
  • @did We are saying the same thing (PS: I find that also in my comment is really although). – Yai0Phah Jun 30 '12 at 07:17

3 Answers3

37

For any $x\in\mathbb{R}$, $$ \begin{align} |f'(x)| &=\lim_{h\to0}\frac{|f(x+h)-f(x)|}{|h|}\\ &\le\lim_{h\to0}\frac{h^2}{|h|}\\ &=0 \end{align} $$ Therefore, $f$ is constant.

robjohn
  • 345,667
  • 1
    But why is $f$ differentiable? –  Nov 07 '18 at 17:22
  • 2
    Because the limit, which is the definition of the derivative, exists (and equals $0$). – robjohn Nov 07 '18 at 17:24
  • 1
    Oh, right. The order things appeared confused me. Thanks. –  Nov 07 '18 at 17:36
  • This is a fantastic example of turning what might be a messy proof into something so elegant, succinct, and clear it's irresistible. I assume this was not your first draft, but rather a rework of a rougher proof. Could you add a bit of how you went from a rougher proof to this? E.g. imagine you started with this, how did you get to your proof? Interestingly, the proof begins not from the given property ($\leq (x-y)^2$), nor from the definition of derivative, but in the middle, with $|f'(x)|$. – SRobertJames Dec 19 '22 at 14:32
  • I’m sorry, but this was the first draft. If it seems polished, it might be because I have done this type of thing many times. – robjohn Dec 19 '22 at 17:17
  • How did you get the idea to start with $|f'|$? It's not a natural starting point, neither what you're given (the inequality) or what you wish to end with (the derivative). – SRobertJames Dec 19 '22 at 20:22
  • 1
    We want to show that $f’$ is close to $0$. The cleanest way to show $f’$ is close to $0$ is to show $|f’|$ is small (this avoids having to show two inequalities). – robjohn Dec 20 '22 at 05:12
23

Here's a proof more elementary.

Let $c=f(0)$, we have to prove that $f(x)=c$ whenever $x\neq0$. Supposing that $n$ is an arbitrary positive integer, we have $$\left|f\left(\frac{m+1}nx\right)-f\left(\frac mnx\right)\right|\le\left(\frac{m+1}nx-\frac mnx\right)^2=\frac{x^2}{n^2}$$ Hence \begin{align*} |f(x)-f(0)| \;&=\;\left|\,\sum_{m=0}^{n-1}\left(f\left(\frac{m+1}nx\right)-f\left(\frac mnx\right)\right)\,\right|\\ &\le\;\sum_{m=0}^{n-1}\,\left|f\left(\frac{m+1}nx\right)-f\left(\frac mnx\right)\right|\\ &\le\;\frac{x^2}n \end{align*} Let $n\to\infty$, we have $|f(x)-f(0)|=0$, thus $f(x)=c$.

Yai0Phah
  • 9,733
14

It suffices to show that $f'(x)=0$ for all $x\in\mathbb{R}$. We see that the given condition implies

$$\left| \frac{f(x)-f(y)}{x-y} \right| \le |x-y|.$$

So in a $\delta$-neighborhood of $x$, the quotient in definition of the derivative is less than $\delta$. So the limit is 0, and we are done.

Potato
  • 40,171