MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $f:R\rightarrow R$. If there exists the finite limit $$\lim_{(x,y) \rightarrow a \atop x\neq y} \frac{f((y)-f(x)}{y-x}$$ then obviously there is a finite derivative $f'(a)$ and is equal this limit.

What about similar problem for higher order divided differences?

May is it true that existence of finite $$\lim_{(x_0,...,x_n)\rightarrow (a,...,a) \atop x_i \neq a} [x_0,...,x_n;f]$$ implies existence of $f^{(n)}(a)$?

If not is there connetion between high order divided differences and derivatives?

share|cite|improve this question
Seems too easy for this forum. Anyway, what exactly is the limit you are looking for? Check this limit for simple functions not twice differentiable, like $x\sqrt{x^2}$, or $x^2\sin(1/x)$... – Feldmann Denis Aug 24 '12 at 14:27
up vote 1 down vote accepted

The answer is yes:

Assume we have a $\delta$ such that for $|x_i-a|<\delta$, we have $|[x_0,...,x_n;f]-\lim| < \epsilon$. Assume WLOG that $\lim = 0$ (by subtracting off a polynomial of degree $n$). Then if $y_0, ..., y_{n-1}$ and $z_0, ..., z_{n-1}$ are in the $\delta$-ball around $a$, we can show

$|[y_0,...,y_{n-1};f]-[z_0,...,z_{n-1};f]| < \epsilon\sum_{i=0}^{n-1}|y_i-z_i|$

by inducting on the number of $i$s such that $y_i\ne z_i$. Thus $\lim_{y_0,...,y_{n-1}\rightarrow y}[y_0,...,y_{n-1};f]$ exists, and by induction on $n$ is equal to $f^{(n-1)}(y)$. Letting the $y_i$s in the above approach $y$, and letting the $z_i$s approach $z$, we see

$|f^{(n-1)}(y)-f^{(n-1)}(z)| \le n\epsilon|y-z|$,

so $\left|\frac{f^{(n-1)}(y)-f^{(n-1)}(z)}{y-z}\right| \le n\epsilon$ for any $y,z$ in a $\delta$-ball around $a$. Taking $\epsilon$ to $0$, we see that $f^{(n)}(a)$ exists and is equal to $0$.

share|cite|improve this answer

In general the answer is no. For instance, if f is any odd function, then $$\lim_{h\to 0}\frac{f(h)+f(-h)-2f(0)}{h^2}=0,$$ without any assumptions on differentiability of f. So it certainly does not follow that f''(0) exists.

share|cite|improve this answer
Thanks, but condition which is assumed (case $n=2$) is that there exists a finite: $$\lim_{(x_0,x_1,x_2)\rightarrow a \atop x_i \neq x_j} \frac{ \frac{f(x_2)-f(x_1)}{x_2-x_1}-\frac{f(x_1)-f(x_0)}{x_1-x_0}}{x_2-x_0} $$ |It seems be stronger limits $\lim_{h\rightarrow 0} \frac{f(x+h)+f(x-h)-2f(x)}{h^2}$. – Jan Aug 24 '12 at 18:49

This question is discussed in some detail here. (Derivative approximation by finite differences, Dave Eberly)

share|cite|improve this answer
The document linked in Igor's answer seems to tacitly assume analyticity of the function. For example, equation (2) says that $F'(x)$ differs from $(F(x+h)-F(x))/h$ by $O(h)$, which fails at $x=0$ for the function $F(x)=x^{5/3}$. (It's possible that the OP also had analytic functions in mind.) – Andreas Blass Aug 24 '12 at 16:15

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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