Let $n$ be an integer greater than $1$. The notation $f^n$ is notoriously ambiguous: it means either the $n$-th iterate of $f$ or its $n$-th power.

I was wondering when the two interpretations are in fact the same. In other words, if we write $f^n(x)$ for $f(f(\dotsb f(x) \dotsb))$ and $f(x)^n$ for $f(x) \cdot f(x) \dotsb f(x)$:

For which functions $f \colon I \to \mathbb R$ is $f^n(x) = f(x)^n$ for all $x \in I$?

So far I have been able to show that:

  • The constant functions $f(x) = 0$ and $f(x) = 1$ satisfy the condition for all $n$ and $f(x) = -1$ satisfies the condition for all odd $n$. Also, if $f$ satisfies the condition for $n$, then the only fixed points of $f$ can be either $0$, $1$, or if $n$ is odd also $-1$.
  • The squaring function $f(x) = x^2$ satisfies the condition for $n = 2$ and is essentially the only non-constant differentiable function to do so. Indeed, $$f(f(x)) = f(x)^2 \implies f'(f(x)) f'(x) = 2 f(x) f'(x)$$ so if we assume $f'(x) \neq 0$ and let $y = f(x)$, we have that $$f'(y) = 2 y \implies f(y) = y^2 + c$$ and furthermore $$(y^2 + c)^2 + c = (y^2 + c)^2 \implies c = 0.$$ (Of course we could then consider also, e.g., $f(x) = x^2$ for $x \ge 0$ and $f(x) = 0$ for $x < 0$.)
  • More generally, $f(x) = x^\sqrt[n-1]{n}$ satisfies the condition for any $n > 2$. These functions are only defined on $\mathbb R^+$: this is why I chose an interval $I$ instead of all $\mathbb R$ as the domain, but I actually don't care if a solution is defined on any other non-trivial subset.

Are there any other solutions? If not, how can we prove so?

Remark: The question can be generalized to $n \in \mathbb Z$ if we assume $f$ to be invertible and denote by $f^{-n}$ either the $n$-th iterate of its compositional inverse or the $n$-th power of its multiplicative inverse (the cases $n = 0, 1$ are trivial). But this seems to be an even harder problem. For the case $n = -1$ see this question.

  • 2
    $\begingroup$ Isn't $f^n(x)$ sometimes used for the nth derivative ? $\endgroup$ – herb steinberg Aug 3 '18 at 23:44
  • 5
    $\begingroup$ That would usually be with parentheses: $f^{(n)}(x)$. $\endgroup$ – Zachary Aug 4 '18 at 0:59
  • 1
    $\begingroup$ I'll mention an identity. If we assume $f$ is invertible, then $f^{n-1}(y)=y^n$. Letting $y=z^p$ and exponentiating both sides by $q$ gives $$[f^{n-1}(z^p)]^q=z^{npq}=z^{nqp}=[f^{n-1}(z^q)]^p$$. $\endgroup$ – user254433 Aug 4 '18 at 1:52

Something non-constant and discontinuous everywhere:

$$f(x) = \left\{\begin{array}{ll} x^2 & \text{if } x \text{ is an integer} \\ \lceil x \rceil & \text{if } x \text{ is rational but not an integer}\\ \lfloor x \rfloor & \text{if } x \text{ is irrational} \end{array} \right.$$

seems to satisfy $f(f(x))=f(x)\cdot f(x)$

  • $\begingroup$ should it be $f(x)$ not $f(m)$? $\endgroup$ – user254433 Aug 4 '18 at 0:54
  • $\begingroup$ @user254433 Yes it should - and now does - thank you $\endgroup$ – Henry Aug 4 '18 at 7:26
  • 4
    $\begingroup$ I like this example because together with user254433's comment it suggests a way to make a new function $g$ out of a known invertible one $f$. First we need to find a subset $A$ such that $f(A) \subseteq A$. Then we need to find a subset $B$ disjoint from $A$ and a function $g \colon B \to A$. Finally we extend $g$ to $A \cup B$ by letting $g(x) = f(x)$ for $x \in A$. It follows that $g^n(x) = g^{n-1}(g(x)) = f^{n-1}(g(x)) = g(x)^n$ for all $x \in A \cup B$. $\endgroup$ – Luca Bressan Aug 4 '18 at 8:08

If for all $x$, $f(\cdots f(f(x)))=(f(x))^n$ then $f(\cdots f(y))=(y)^n$, and the function $f$ is a functional $n-1^{th}$ root of the $n^{th}$ power function.

In particular, for $n=2$, $f$ is the square function. For $n=3$, a solution is $f(x)=x^{\sqrt 3}$ and more generally


  • $\begingroup$ I don't see how this provides an answer to the question, since I had already pointed out the solution $f(x) = x^{\sqrt[n-1]{n}}$ in my attempt. $\endgroup$ – Luca Bressan Aug 14 '18 at 11:08
  • $\begingroup$ @LucaBressan: I am turning your question to a more classical one, that of the functional root. $\endgroup$ – Yves Daoust Aug 14 '18 at 11:58
  • $\begingroup$ This only shows that $f^{n-1}(y)=y^n$ for $y\in f(X)$, $X$ the domain of $f$. The relation may not hold on all of $X$, for example the zero function satsfies OP's condition on all $\mathbb R$ but only satisfies your condition on $\{0\}$. $\endgroup$ – Jack M Aug 15 '18 at 20:51

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.