Basically I thought about a kind of modulo 2 equivalence class for real numbers, if that makes sense. With that, and noting that for each number $y \in [1,2)$, the numbers $2y$ and $y/2 $ are not in $ [1,2)$, I have defined an arbitrary function on the interval $[1,2)$ and repeated it (stretched or shrunk in a way that maintains the property) along the real axis. I have found that, if $h$ and $g$ are arbitrary functions from $\mathbb{R}$ to $\mathbb{R}$, then an $f$ constructed as follows satisfies the request. $\lfloor \cdot \rfloor$ represents the floor function.

\begin{equation*} f(x) = \begin{cases} 2^{\lfloor\log_2 |x|\rfloor} \cdot h(|x| \cdot 2^{-\lfloor\log_2 |x|-1\rfloor} )& x >0 \\ 2^{\lfloor\log_2 |x|\rfloor} \cdot g(|x| \cdot 2^{-\lfloor\log_2 |x|-1\rfloor} ) & x <0\\ 0 & x =0 \end{cases} \end{equation*}

This works because if $x>0$, then: $$f(2x) = 2^{\lfloor\log_2 (2x)\rfloor} \cdot h(2x \cdot 2^{-\lfloor\log_2 (2x)-1\rfloor} ) = 2^{1 + \lfloor\log_2 (x)\rfloor} \cdot h(2x \cdot 2^{- 1 -\lfloor\log_2 (x)-1\rfloor} ) = 2f(x) \, .$$

Similarly, this happens when $x<0$, and by defining $f(0)=0$ we also get that the property is satisfied for $x=0$.

My question is this: how can it be said that these are all the possible functions that satisfy the request? And also, if $h$ and $g$ are not the same linear function, can $f$ be differentiable at $x=0$? It seems strange to me that a function that oscillates infinitely many times near the origin would be differentiable there. Can it be that, if neither $g$ nor $h$ are linear, $f$ is $C^\infty$ at every $x\neq 0$? My candidate for that is $g(x)=h(x)=\exp\left[-\left(\frac{1}{|x-2|} + \frac{1}{|x-4|} \right)\right]$ . Lastly, is there a prettier way to express $f$ ?

Thanks for reading :)


It's easy to prove that what you wrote is the only correct form (and I don't think it can get much prettier). In fact every f such that $f(2x)=2f(x)$ of course induces an $h:[1,2) \to \mathbb{R}$ by restriction; then for every $x >0$ there exists a unique $k \in \mathbb{Z} $ such that $2^k x \in [1,2)$ and in particular $f(x)=2^{-k} f( 2^k x) = 2^{-k}h(2^kx)$ as you described.

regarding the differentiability in $0$ one has, for every $x \in [1,2)$ that $x_k=x \cdot 2^{-k}$ is a sequence tending to $0$. But then, by the properties of $f$, if it is differentiable in $0$ we would have

$$f'(0) = \lim_{k \to \infty} \frac {f(x_k)-f(0)}{x_k-0} = \frac {h(x)}{x}$$

and the same should be true for $g$; in particular $f$ is differentiable in $0$ iff $h(x)=ax$ and $g(x)=ax$ for the same $a \in \mathbb{R}$, which will be the derivative of $f$ in $0$.

You can have that the function is $C^{\infty}$ outside $0$, it is sufficient to take any $C^{\infty}$ function $t$ which is $0$ outside $[0,1]$, for example $t(x)=e^{- 1/|x(1-x)|}$, similarly as you suggested...

then you can consider $h(x)=g(x)=x+ t(2*(x-4/3))$ (in such a way that it gets modified only on $[4/3,11/6]$, and there should be no problems at $1$ and $2$, but you can take also $h(x)=x+t(x-1)$ it works fine).

In conclusion, in fact everything you said is true!


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.