1
$\begingroup$

If $g:\mathbb R\to\mathbb R$ is differentiable, how can we show that $$h(z):=\min\left(1,e^{g(z)}\right)\;\;\;\text{for }z\in\mathbb R$$ is also differentiable, except at a countable number of points, with derivative given Lebesgue almost everywhere by the function $$z\ni\mathbb R\mapsto\begin{cases}g'(z)e^{g(z)}&\text{, if }g(z)<0\\0&\text{, if }g(z)\ge0\end{cases}?\tag1$$

It's clear to me that $\min:\mathbb R^2\to\mathbb R$ is partially differentiable in both arguments, except on the diagonal $\Delta:=\left\{(x,x):x\in\mathbb R\right\}$ (which clearly has Lebesgue measure $0$). However, it's not clear to me why $h$ is differentiable except on a countable set.

Remark 1: I'm aware of Rademacher's theorem, which yields that $\mathbb R\ni x\mapsto\min(1,e^x)$ is differentiable Lebesgue almost everywhere (noting that this function is Lipschitz continuous). But this doesn't immediately yield that the derivative of $h$ is as claimed and it doesn't yield that the null set is countable.

Remark 2: We may note that the Lipschitz continuity implies absolutely continuity. So, maybe this is related to this Wikipedia entry: https://en.wikipedia.org/wiki/Absolute_continuity#Equivalent_definitions. But again, this result seems only to yield Lebesgue almost everywhere differentiability, but not the concrete shape of the derivative.

EDIT: We know that $$\frac{\rm d}{{\rm d}x}\min(1,e^x)=\left.\begin{cases}e^x&\text{, if }x<0\\0&\text{, if }x>0\end{cases}\right\}\;\;\;\text{for all }x\in\mathbb R.\tag2$$ So, the problematic set seems to be $\left\{x\in\mathbb R:g(x)=0\right\}$, which doesn't need to be countable ... The claim can be found here on page 12.

$\endgroup$
6
$\begingroup$

Yes, it is true. Since $2\min(a,b)=a+b-|b-a|$ for positive numbers $a,b$, and $F(z)=e^{g(z)}-1$ is differentiable, it suffices to prove that $|F|$ is differentiable except at a countable set of points. Namely, this exceptional set is a set of isolated zeroes of $F$ (or, better to say, a subset of the set of isolated zeroes). Indeed, if $F(x)\ne 0$, of course $|F|$ is differentiable at $x$. If $F(x)=0$ and $F'(x)=0$, then $(|F|)'(x)=0$, since $|F(x+t)|-|F(x)|=|0+o(t)|-|0|=o(t)$ for small $t$. All points which remain are isolated zeroes of $h$, there are at most countably many of them.

$\endgroup$
  • $\begingroup$ Thank you for your answer. Let me note that the symbol $h$ is already used in the question. So, it is a bit unfortunate that you use it for $z\mapsto e^{g(z)}-1$. $\endgroup$ – 0xbadf00d May 1 at 13:40
  • $\begingroup$ indeed, now I fixed $\endgroup$ – Fedor Petrov May 1 at 14:11
  • $\begingroup$ You have shown that $$N:=\left\{a\in\mathbb R:f(a)=0\text{ and }f'(a)\ne0\right\}$$ is countable and $h$ is differentiable at $a$ with $$h'(a)=\begin{cases}e^{g(a)}g'(a)&\text{, if }g(a)<0\\0&\text{, if }g(a)>0\\0&\text{, if }g(a)=0\text{ and }g'(a)=0\end{cases}$$ for all $a\in\mathbb R\setminus N$. What can we say about the second derivative? It's clear that $$h''(a)=\begin{cases}e^{g(a)}\left(\left|g'(a)\right|^2+g'(a)\right)&\text{, if }g(a)<0\\0&\text{, if }g(a)>0\end{cases}$$ for all $a\in\mathbb R$ though. $\endgroup$ – 0xbadf00d May 1 at 15:14
  • $\begingroup$ I've asked for that separately here: math.stackexchange.com/q/3209891/47771. It would be enough for me if the set on which $|h|$ is twice differentiable has Lebesgue measure $0$. $\endgroup$ – 0xbadf00d May 1 at 19:21

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.