Just recently I read up something about involutions (functions $f: A \rightarrow A$ such that $f(f(x))=x$, for all $x$ in the domain of $f$), and was wondering how many (if there is a small set of general functions) such involutions exist for $A = \mathbb{R}$, or maybe $A = \mathbb{R} - S$, where S is some set of points that would make the involution work if they were left out of $A$. In general I'm interested in real functions that are involutions, continuous or otherwise.

There were some examples I found here, but any more would be certainly very interesting!

  • 6
    $\begingroup$ Do you want to require any special properties of the involution? Otherwise there are lots: an involution on a set is the same thing as a partition of that set into blocks of size $1$ and $2$. So I think for an infinite set $A$ there are $2^A$ involutions. $\endgroup$ – Qiaochu Yuan Jun 21 '11 at 4:45
  • 3
    $\begingroup$ Your title asks for involutions of "real functions"; your body asks for involutions of the real numbers. Could you harmonize title and body? $\endgroup$ – Arturo Magidin Jun 21 '11 at 4:52
  • $\begingroup$ @Arturo: Do you mean he should ask about harmonic functions which are involutions? :-P $\endgroup$ – Asaf Karagila Jun 21 '11 at 6:16
  • $\begingroup$ @Asaf: Well, one could ask about involutions on the set $\mathbb{R}^{\mathbb{R}}$; when I read the title, I thought that's where it was going. $\endgroup$ – Arturo Magidin Jun 21 '11 at 6:19
  • $\begingroup$ @Qiaochu: I wonder, why did you tag this under set-theory? (I'm not sure where it fits anyway, and the question isn't very clear as pointed out by you and by Arturo.) $\endgroup$ – Asaf Karagila Jun 21 '11 at 10:58

If we do not put any conditions on $f$, there will be far too many involutions. Divide the reals in any way into a disjoint union of $1$-element sets and/or $2$-element sets.

For any such subdivision $\mathbb{U}$, if $\{a\}$ is a singleton set in the subdivision, let $f(a)=a$. If $\{a,b\}$ is a doubleton set in the subdivision, let $f(a)=b$ and $f(b)=a$. Then $f$ is an involution. Any subdivision determines an involution, and conversely every involution determines a subdivision.

It follows that there are $2^\mathfrak{c}$ involutions, "just as many," from the point of view of cardinality, as the number of functions from $\mathbb{R}$ to $\mathbb{R}$.


If you assume continuity, you still have as many involutions as continuous functions from $\mathbb R$ to itself, namely cardinality $c$. For if $f$ is any strictly decreasing continuous function from $[0, \infty)$ onto $(-\infty, 0]$, you can extend $f$ to $(-\infty, 0)$ by $f(t) = x$ where $f(x) = t$, and you get a continuous involution.

  • $\begingroup$ How about differentiable involutions? analytic involutions? $\endgroup$ – Gerry Myerson Jun 21 '11 at 5:28
  • $\begingroup$ Wait, the cardinality of continuous functions from $\mathbb R$ to itself is the same as the cardinality of $\mathbb R$? $\endgroup$ – Rahul Jun 21 '11 at 5:40
  • 5
    $\begingroup$ You get $\mathfrak{c}$ analytic involutions from the examples $x\mapsto a-x$. @Rahul: Continuous functions are determined by their values on a countable dense set, so there are at most $\mathfrak{c}^{\aleph_0}=\mathfrak{c}$ of them. $\endgroup$ – Jonas Meyer Jun 21 '11 at 6:12

Any function withich is symmetric w.r.t.the 45°-axis has $f(x)=f^{-1}(x)$ and so $f(f(x))=x$. Visually, obvious examples are $f(x):=\pm x$ and $f(x):=\frac{1}{x}$.

Moreover, if we take such a function and shift it along the 45°-axis, the property doesn't get lost. Hence, take any function for which $f(f(x))=x$ and define $F_{(f,d)}(x):=f(x-d)+d$, then


This explains why $\frac{1}{x-d}+d$ works. Also visually, it's clear that the property doesn't get lost if you mirror the function along the axis. Define $G_f(x):=-f(-x)$, then


This "generating new functions via shift" busines reminded me of the modular group. I determined it's nontrivial involutive elements and they are $\frac{ax+b}{cx-a}$. Shifting them along the axis by $s$ gives a lot of example solutions: $\frac{b-cs(s+x)+a(2s+x)}{c(s+x)-a}$.


Here are a couple of examples that might have some appeal.

  1. The link has $f(x)=a+(x-a)^{-1}$, but in fact $f(x)=a+b(x-a)^{-1}$ works for any real $a$ and $b$ (of course one must remove $x=a$ from the domain).

  2. Let $g$ be defined on $\lbrace0,1,\dots,9\rbrace$ by $g(9)=9$, $g(d)=8-d$ for $d\ne9$. Then get $f(x)$ by applying $g$ to each of the digits in the decimal expansion of $x$. E.g., $\pi=3.14159265\dots$, so $f(\pi)=5.74739623\dots$.

These are, of course, special cases of the construction given by user6312, but maybe their concreteness and simplicity justify their separate mention.

It further appears that the equation $f(f(x))=x$ is known as Babbage's functional equation. There's a nice paper about it by J F Ritt, On certain real solutions of Babbage's functional equation, Annals of Math 17 (1916) 113-122, doi: 10.2307/2007270.


The set of continuous involutions on $\mathbb R$ has a fairly easy to describe structure. Let $f$ be a continuous involution on $\mathbb R$. Then:

  1. Any point is either fixed, or of period $2$. The set of fixed points is closed, so the set of periodic points is open.
  2. If $x$ is of period $2$, there is a fixed point between $x$ and $f(x)$.
  3. If $x$ is of period $2$, there is exactly one fixed point between $x$ and $f(x)$. Let $a$ be the nearest fixed point to $x$ between $x$ and $f(x)$. Then as $x$ approaches $a$, $f(x)$ must approach $a$, and can never be equal to a fixed point, since by our choice of $a$, $x$ is never itself fixed as it approaches $a$. This is impossible if there's another fixed point between $x$ and $f(x)$.
  4. If $f$ has at least two fixed points, then suppose there is a non-fixed point in between them. Take the largest open interval of non-fixed points around that point. Its boundaries are fixed. By continuity and (2), near the left edge of that interval, $f(x)$ must be outside of the interval on the left, and near the right edge, $f(x)$ must be outside on the right. As $x$ runs through the interval, $f(x)$ can't touch any fixed points, so this violates the intermediate value theorem.
  5. If on the other hand, any two fixed points of $f$ have no non-fixed points between them, then the set of fixed points forms a single closed interval. If this interval isn't all of $\mathbb R$, then (3) is impossible to satisfy for any non-fixed $x$.

Thus $f$ either fixes $\mathbb R$ or has exactly one fixed point, dividing the real line into two halves. By (2), the left half must map onto the right half and vice-versa. For simplicity let's say the fixed point is $0$. Then $f$ can/must be equal to any/some strictly decreasing continuous function with $f(0)=0$ on $\mathbb R^+$, and to its inverse on $\mathbb R^-$. For instance we can take $f(x)=-2x$ for positive $x$ and $f(x)=-\frac12x$ for negative $x$.


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.