Let $f$ be differentiable at $x=a$.

Prove that if $x_n \to a^+$ and $y_n \to a^-$ then:

$$\lim_{n\to \infty} \frac{f(x_n)-f(y_n)}{x_n-y_n}=f'(a).$$

Every option that I think about seems to my very trivial, so I believe that I am doing something wrong. Both numerator and denominator approach zero as $n\to\infty$ as the case of the formal definition of derivative, but it isn't guaranteed that the limits are equal (“$\frac{0}{0}$”).

Any direction?


First remark that the result is easy if you assume that $f$ is derivable with continuous derivative in a neighborhood of $a$. Indeed, by the mean value theorem, you can write $\frac{f(x_n)-f(y_n)}{x_n - y_n}$ as $f'(c_n)$ for some $c_n$ between $y_n$ and $x_n$ and the result follows by continuity of the derivative.

For the general case, I would do the following: write $\frac{f(x_n)-f(y_n)}{x_n - y_n}$ as $\frac{f(x_n) - f(a)}{x_n - a} \frac{x_n - a}{x_n - y_n} + \frac{f(a)-f(y_n)}{a-y_n} \frac{a-y_n}{x_n-y_n}$. Then we would like to compute the limits of these quantities and find that this gives $f'(a)$ by addition and multiplication.

The problem is the following: the quantity $\frac{x_n-a}{x_n - y_n}$ may well have no limit at all. For instance, take $a=0$, $x_n = 1/n$ and $y_n = -1/n$ for odd $n$ and $0$ for even $n$. Then this quantity is alternatively $1$ and $1/2$, hence does not converge.

Here is how I resolve this issue. Write $u_n$ for $\frac{f(x_n)-f(y_n)}{x_n - y_n}$. Then, $u_n$ is bounded since in the above writing, the quantites $\frac{x_n-a}{x_n - y_n}$ and $\frac{a-y_n}{x_n-y_n}$ are between $0$ and $1$. It is a standard fact that a bounded sequence converges if and only if it has a unique adherence value.

By Bolzano-Weierstrass theorem, let $u_{\phi(n)}$ be an extracted sequence that converges to $l_\phi$. By further extracting, we can also assume that the sequences $\frac{x_n-a}{x_n - y_n}$ and $\frac{a-y_n}{x_n-y_n}$ also converge to $l_x$ and $l_y$. Remark that necessarily, $l_x + l_y = 1$ since $\frac{x_n-a}{x_n - y_n} + \frac{a-y_n}{x_n-y_n} = 1$. By the above writing, this shows that $l_u = f'(a)l_x + f'(a)l_y = f'(a)$. Hence, the only adherence value of $u_n$ is $f'(a)$, showing that $f'(a)$ is the limit of $u_n$.

Maybe there is a shortcut to the extracting argument...

Edit: Indeed, there is a shortcut:

Write $u_n$ as $\frac{f(a)-f(y_n)}{a-y_n} + \frac{x_n - a}{x_n - y_n}(\frac{f(x_n) - f(a)}{x_n - a} - \frac{f(a)-f(y_n)}{a-y_n})$. The first term tends to $f'(a)$, both terms in the parenthesis have the same limit and the coefficient before the parenthesis is bounded.

  • $\begingroup$ That's a nice observation on the quantity $\frac{x_n-a}{x_n - y_n}$! I didn't know that! I just thought it was 1. Upvoted! $\endgroup$ – ireallydonknow Jan 6 '14 at 15:35
  • $\begingroup$ Wow, thank you very much. $\endgroup$ – Lior Jan 6 '14 at 16:00

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.