0

Consider a real-valued function $f:[a,b]\rightarrow \mathbb{R}$ and consider some point $c\in (a,b)$. Assume that $$f^\prime(c^-):=\lim_{x\nearrow c}\frac{f(x)-f(c)}{x-c}\quad\text{and}\quad f^\prime(c^+):=\lim_{x\searrow c}\frac{f(x)-f(c)}{x-c}$$ both exist. That is, assume that the left-hand and right-hand derivatives of the function exist at the point $c$. Further assume that $f^\prime(c^-)=f^\prime(c^+)$, i.e., that the left-hand and right-hand derivatives at the point $c$ equal. Does this necessarily imply that $f$ is differentiable at $c$, i.e., that $f^\prime(c):=\lim_{x\rightarrow c}\frac{f(x)-f(c)}{x-c}$ exists? Could you provide a proof or reference for this fact? If it doesn't necessarily imply differentiability, are there additional conditions that would imply differentiability?

Intuitively, it seems to be true. But I couldn't prove it. I saw a few related posts on Math Stack Exchange (with one directly asking this question), but for some reason I didn't see a convincing enough answer there.

Edit: There is something that I want to add here and ask additionally. Suppose you know the derivative of $f$ to both the left and right of $c$, but not at $c$. Suppose you can show that $$\lim_{x\nearrow c} f^\prime(x)=\lim_{x\searrow c} f^\prime(x).$$ Can it be concluded from this that the derivative of $f$ at $c$, i.e., $f^\prime(c)$, exists and equals those left-hand and right-hand limits? Furthermore, can it be concluded that $f^\prime$ is continuous at $c$?

Satana
  • 1,259
  • 3
    If a function has left and right limits at the point $c$, then it has a limit at this point if and only if the left and right limits are equal. This applies to $\frac{f(x)-f(c)}{x-c}$ – Gribouillis Nov 03 '24 at 08:38
  • Thanks for the comment. I realized that the original question I asked wasn't what I really wanted to know. I added an edit on the original post. If you have any insights, that would be great. @Gribouillis – Satana Nov 03 '24 at 15:36
  • 1
    OP for your update, this is a standard question and I explain the situation in some detail in this answer of mine. Tldr is you need to assume in addition that $f$ is continuous at $c$; then $f’(c)$ exists and $f’$ is continuous at $c$. Without the extra assumption this is false (see the link… or the now-deleted answer here). – peek-a-boo Nov 03 '24 at 15:42
  • Thank you very much! And I assume that the third (bullet point) condition in that answer of yours holds here because if the right hand and left hand limits exist, then the limit exists? @peek-a-boo – Satana Nov 03 '24 at 15:45
  • 1
    Well… it’s that both limits exist and are equal (by your assumption). It’s just the basic fact I mentioned under one of the answers. – peek-a-boo Nov 03 '24 at 15:46
  • Sorry just to clarify and beat this horse dead. In my update I said let's assume that both the left-hand and right-hand limits of the derivative $f^\prime$ exist at $c$ and are equal. Therefore, we can conclude from this that $\lim_{x\rightarrow c} f^\prime(x)$ exists. From which the third bullet point on your answer from the previous post will hold. Is this sequence of logic correct? I am just confirming that what I stated here as assumptions give me everything I need from that previous answer of yours. I appreciate your answers, thank you! @peek-a-boo – Satana Nov 03 '24 at 16:01
  • 1
    Now what you said in this comment is correct. In your previous comment you were missing the remark that the limits are equal (but you did include the hypothesis in the edited post)… I was just drawing your attention to this detail. – peek-a-boo Nov 03 '24 at 16:04
  • Good catch. I totally missed that I missed that. (Say that ten times.) Thank you sincerely for your attention to detail. @peek-a-boo – Satana Nov 03 '24 at 16:07
  • 1
    The answer to your edit is YES, it is a consequence of the mean value theorem applied to the left and to the right of the point $c$. – Gribouillis Nov 03 '24 at 16:08
  • Thank you @Gribouillis! – Satana Nov 03 '24 at 16:09

2 Answers2

2

The derivative of a function $f$ exists if the right-hand and left-hand derivative of $f$ exist, and both of them are equal.



Indeed, let $(x'_n)_{n\in\mathbb{N}}$ be any sequence in $[a, b]$ convergent to $c$ s.t. $x'_n \neq c$ for all $n \in \mathbb{N}$. Take an arbitrary subsequence $(x_n)_{n\in\mathbb{N}}$ of $(x'_n)$.

Then we can take a further subsequence $(z_n)_{n\in\mathbb{N}}$ s.t. $z_n < c$ for all $n \in \mathbb{N}$, or $z_n > c$ for all $n \in \mathbb{N}$ since $(x_n)$ must have infinitely many $n\in \mathbb{N}$ s.t. $x_n < c$ or have infinitely many $n$ s.t. $x_n > c$.

Since both right-side derivative and left-side derivative exist and they are equal, we have $$\lim_{n\to\infty} \frac{f(z_n)-f(c)}{z_n-c} = f'(c^-)=f'(c^+)$$ for any case.

We showed that for any subsequence $(\frac{f(x_n)-f(c)}{x_n-c})_n$ of $(\frac{f(x'_n)-f(c)}{x'_n-c})_n$ we have a further subsequence $(\frac{f(z_n)-f(c)}{z_n-c})$ convergent to $f'(c^-)=f'(c^+)$. Therefore, since if for any subsequence of a sequence there exists a further subsequence converging then a sequence converges, $$\lim_{n\to\infty} \frac{f(x'_n)-f(c)}{x'_n-c} = f'(c^-) = f'(c^+)$$

Thus $f'(c)$ exists and $f'(c)=f'(c^-)=f'(c^+)$

ShoboDra
  • 184
  • 2
    We really don’t need to go through sequences and subsequences of subsequences though… this just directly falls out of $\epsilon$-$\delta$ that a limit exists if and only if both one-sides limits exist and are equal. – peek-a-boo Nov 03 '24 at 09:17
  • Thank you both for your comments/answers. I added an additional question as an edit in my original post. If possible, I would appreciate insight into that as well! @peek-a-boo – Satana Nov 03 '24 at 15:30
1

Theorem: A function $f$ is differentiable at $c$ if and only if the left and right limits of its difference quotient exist and are equal, i.e., $ f'(c^-) = f'(c^+) = L $ for some real number $ L $.

Proof:

We’ll prove each direction.

Forward Direction: Assume $ f $ is differentiable at $ c $, which means the limit \begin{equation} f'(c) = \lim_{x \to c} \frac{f(x) - f(c)}{x - c} \end{equation} exists. By the sequential criterion for limits, this implies that for any sequence $ x_n \to c $, the sequence of difference quotients \begin{equation} \frac{f(x_n) - f(c)}{x_n - c} \to f'(c). \end{equation} In particular, this holds for sequences $ (x_n) $ approaching $ c $ from above or from below. Thus, both the right-hand limit $ f'(c^+) $ and the left-hand limit $ f'(c^-) $ must exist and equal $ f'(c) $. Therefore, we conclude that $ f'(c^-) = f'(c^+) = f'(c) $.

Reverse Direction: Assume that the left and right limits of the difference quotient exist and are equal; let $ f'(c^+) = f'(c^-) = L $. We’ll show that this implies $ f $ is differentiable at $ c $ with $ f'(c) = L $.

Consider any sequence $ x_n \to c $. We can divide $ (x_n) $ into two subsequences: $ (y_m) $ (where $ y_m > c $ for all $ m $) and $ (z_l) $ (where $ z_l < c $ for all $ l $), though it may happen that one of these subsequences is finite or empty. At least one of them, however, must be infinite, and in any of these scenarios we have that $\forall n, x_n=y_m$ or $x_n=z_m$ with $y_m \rightarrow c^+$ and $z_l \rightarrow c^-$. So, we have either: \begin{equation} \frac{f(x_n) - f(c)}{x_n - c} =\frac{f(y_m) - f(c)}{y_m - c} \rightarrow f'(c^+) = L, \end{equation} or \begin{equation} \frac{f(x_n) - f(c)}{x_n - c} =\frac{f(z_l) - f(c)}{z_l - c} \rightarrow f'(c^-) = L. \end{equation} Thus, regardless of the approach direction, the difference quotient converges to $ L $, meaning that \begin{equation} \lim_{x \to c} \frac{f(x) - f(c)}{x - c} = L. \end{equation} This proves $ f $ is differentiable at $ c $ with $ f'(c) = L $.

Of course, the last part should be made more rigerous using the $\epsilon-N$ convergence of the sequences.

Edit

For a function $ g $ to be continuous at $ c $, we must have $ \lim_{x \rightarrow c} g(x) = g(c) $. The argument above holds if we replace $ f'(x) $ by a function $ g(x) $; that is, the limit exists if and only if $ g(c^+) = g(c^-) $, or alternatively, this means a function is continuous if and only if $ g(c^+) = g(c) = g(c^-) $. This holds true in the case of $ f'(x) $ when $ f $ is differentiable.

abc1455
  • 522
  • 1
    Thank you for your answer! There is one follow-up question that I forgot to ask but just added it as an edit to my original post. Any insights into that? @abc1455 – Satana Nov 03 '24 at 15:30