Stolz-Cesàro theorem is usually stated to be the discrete version of L'Hospital's rule. I was merely wondering whether one of these theorems could be used to prove the other (I couldn't find any proof that does this online).


I found a proof of L 'Hopital's rule that makes use of Stolz-Cesaro for the case $g(x)\rightarrow \infty$.

I've noticed recently that in the hypothesis of L'Hopital's rule it is often assumed that $f$ and $g$ need be defined and differentiable at the point $p$ which is being approached in the limit (this is done in the proof I mentioned and in baby Rudin). Yet the theorem still holds even when $f$ and $g$ are not defined nor differentiable at $p$, so I will rewrite the proof with care so as to prove the theorem in the more general case.

Theorem: Let $f$ and $g$ be real functions differentiable in the open interval $I$ except possibly at the point $p\in I$. Assume that $g'(x)\neq 0$ for every $x\in I-\{p\}$ and that $$\lim_{x\rightarrow p}\frac{f'(x)}{g'(x)}$$ exists and equals $L$. If either

i) $\lim_{x\rightarrow p}g(x)=\pm \infty$, or

ii) $\lim{x\rightarrow p}f(x)=g(x)=0$,

then $\lim_{x\rightarrow p}f(x)/g(x)$ exists and equals $L$.


i) Suppose $\lim_{x\rightarrow p}g(x)=\infty$. Consider all points $x\in I$ less than $p$, since $g'(x)\neq 0$ we have that $g'$ is either positive or negative, yet $\lim_{x\rightarrow p}g(x)=\infty$ implies it can only be positive. We may also assume $g(x)\neq 0$ by making $I$ smaller if necessary.

It suffices to show that for any monotone sequence $x_n\rightarrow p$ we get $f(x_n)/g(x_n)\rightarrow L$, so let $x_n$ be such a sequence. Regardless of whether it is approaching $p$ from the right or the left, we have that $g(x_{n+1})>g(x_n)$. Stolz-Cesaro gives

$$\liminf \frac{f(x_{n+1})-f(x_n)}{g(x_{n+1})-g(x_n)}\leq \liminf \frac{f(x_n)}{g(x_n)}\leq \limsup \frac{f(x_n)}{g(x_n)}\leq \limsup \frac{f(x_{n+1})-f(x_n)}{g(x_{n+1})-g(x_n)}$$

By Cauchy's Mean Value Theorem there exists $y_n$ between $x_n$ and $x_{n+1}$ such that $$\frac{f(x_{n+1})-f(x_n)}{g(x_{n+1})-g(x_n)}=\frac{f'(y_n)}{g'(y_n)}$$

noting $\lim_{n\rightarrow \infty}f'(y_n)/g'(y_n)=L$ gives the desired result.

ii) Let $\epsilon >0$. Then there exists a neighborhood $U=(u,p)$ of $p$ such that $x\in U$ implies $$\Big| \frac{f'(x)}{g'(x)} -L\Big| <\epsilon$$

Let $y_n\rightarrow p$ be a monotone sequence in $U$. By Cauchy's Mean Value Theorem there is a sequence $z_n\in (x,y_n)$, so that $z_n\rightarrow p$, such that $$\frac{f(x)-f(y_n)}{g(x)-g(y_n)}=\frac{f'(z_n)}{g'(z_n)}\in (L-\epsilon,L+\epsilon)$$ therefore $\lim_{n\rightarrow \infty}\frac{f(x)-f(y_n)}{g(x)-g(y_n)}=\frac{f(x)}{g(x)}\in [L-\epsilon,L+\epsilon]$ for any $x\in U$, so $\lim_{x\rightarrow p^-}f(x)/g(x)=L$. The other direction is similar.


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.