I'm constructing a continuous function $f$ which is well defined on $\mathbb R$ and its derivative $g$ does not exists on a countable infinite points. This kind of functions is easy to find. For example, a continuous function consisting a series of connected line segments. I think the $g$ is actually continuous on its domain, if its domain is restricted on where it exists (correct me if I am wrong). Here its domain is a dense set. Then I am wondering if we can find an example or a counterexample which succeed or fail the L'Hospital rule, of which the numerator and denominator are continuous on $\mathbb R$, and their derivative exists almost everywhere.


I guess I should be more specific. I want the points where the derivative does not exist are concentrated around the point where L' Hospital will be used. For example, L' Hospital rule will be used at $x=0$ and the numerator and denominator's derivatives do not exist on $\{x=\frac{1}{n}\}$


I note that in common L'Hospital rule statement and proof, it is assumed that in $\lim\frac{f(x)}{g(x)}$, $f$ and $g$ are not only continuous on an interval (neighborhood of $x_0$, where the rule will be used), but also differentiable on the interval everywhere. See here. I am wondering if this condition is too strong. I think the derivatives of $f$ and $g$ are not necessary to exist everywhere on the interval but only need the $\lim \frac{f'}{g'}$ to exist, to apply L'Hospital rule. In other words, can we relax the condition by allowing $f'$ and $g'$ to exist outside a smaller, in some sense (e.g. measure zero ?), subset of the interval.

  • $\begingroup$ Re Update 2: As discussed below this ultimately depends on obtaining a monotonicity theorem of the form "$f$ has some property and $f'(x)>0$ almost everywhere, therefore $f$ is nondecreasing." For more than you might care to know about monotonicity theorems of this type see Chapter 11 of Andy Bruckner's monograph Differentiation of Real Functions. $\endgroup$ – B. S. Thomson Dec 31 '15 at 17:16

Maybe, perhaps, this is what you are thinking of? I will pose as another (related) problem.

Problem. Suppose $f$ is continuous in some neighborhood $(x_0-\delta,x_0+\delta)$ of a point $x_0$ and there is a "small" set $N$ so that $f'(x)$ exists for each $x\not\in N$ and moreover $$ \alpha < f'(x) < \beta$$ for each $x\in (x_0-\delta,x_0+\delta)\setminus N$.

Can we conclude that, for all $x\in (x_0-\delta,x_0+\delta) $ $$ \alpha \leq \frac{f(x)- f(x_0)}{x-x_0} \leq \beta?$$

If so then you could use this result for your L'Hopital's rule. For example if $f(x_0)=0$ and $g(x_0)=0$ then $$ \lim_{x\to x_0} \frac{f(x)}{g(x)} = \lim_{x\to x_o, x\not\in N} \frac{f'(x)}{g'(x)} $$ would be true under the right hypotheses. You need (i) $N$ to be suitably small (i.e., it answers the problem above) (ii) you need $f$ and $g$ differentiable outside $N$ and (iii) you need
both limits to exist: $$\lim_{x\to x_o, x\not\in N} {f'(x)} $$ and $$ \lim_{x\to x_o, x\not\in N} {g'(x)} \not= 0 $$

So really your fundamental problem here is a monotonicity problem. The answer is that for continuous $f$ and a countable set $N$ the answer to the problem posed is "yes". For continuous $f$ and a measure zero set $N$ the claim might be false. For absolutely continuous $f$ and a measure zero set the answer is again "yes."

I hope this isn't too brief or cryptic for your purposes.

  • $\begingroup$ Thanks for this original response. I am interested the second last paragraph in your answer. Could you refer me to some more detailed source about monotonicity problem? And I guess I could make my purpose more clear by updating the question. $\endgroup$ – Hua Dec 31 '15 at 5:19
  • $\begingroup$ Work on this problem (or ask for help here): If $f:[a,b]\to\mathbb R$ is continuous and $f'(x)>0$ for all $x$ with at most countably many exceptions then $f$ is nondecreasing. $\endgroup$ – B. S. Thomson Dec 31 '15 at 17:08

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.