12
$\begingroup$

Let $f:\mathbb{R}\to\mathbb{R}$ be a function such that $f(f(x))=f(x)+x, \forall x\in\mathbb{R}$. Find all such functions $f$.

Clearly, $f$ is an "one-to-one function". I have tried setting where $x\rightarrow f(x)$ so: $f(f(f(x)))=f(f(x))+f(x)=2f(x)+x$.

Also, from the original relation, we get that: $f(f(f(x)))=f(f(x)+x)$.

Equating the RHS of the last two relations we get that: $f(f(x)+x)=2f(x)+x$.

Probably to exploit the 1-1 we need a relation of the form $f(f(x))=f(\cdots)$.

I don't know how to carry this further... Any hint?

$\endgroup$
  • 3
    $\begingroup$ @Tintarn Proof for 1-1: Let $x_1,x_2\in\mathbb{R}$ such that $f(x_1)=f(x_2)$. Then, $f(f(x_1))=f(f(x_2))\Rightarrow f(x_1)+x_1=f(x_2)+x_2\Rightarrow x_1=x_2$. Since for all $x_1,x_2\in\mathbb{R}$ we have that: $f(x_1)=f(x_2)\Leftrightarrow x_1=x_2$, f is one-to-one. $\endgroup$ – Jason Aug 28 '15 at 11:40
  • 3
    $\begingroup$ I don't think that the given condition is enough to provide only "nice" solutions. But clearly, $f(x)=\varphi x$ and $f(x)=-\frac{1}{\varphi}x$ are solutions. $\endgroup$ – Redundant Aunt Aug 28 '15 at 11:47
  • 4
    $\begingroup$ @Tintarn Here it is said that a "one-to-one function" is injective. Anyway, I clarified the Original Post. $\endgroup$ – Jason Aug 28 '15 at 11:48
  • 4
    $\begingroup$ @Tintarn Isn't the normal terminology that "one-to-one" implies injectivity and "onto" implies surjectivity? $\endgroup$ – Daniel R Aug 28 '15 at 12:12
  • 3
    $\begingroup$ @Tintarn: See this question, for example. $\endgroup$ – Cameron Buie Aug 28 '15 at 12:17
4
$\begingroup$

Suppose $f$ is continuous and bijective... I claim either $f(x) = \phi x$ for all $x$, or $f(x) =-x/\phi$ for all $x$. Here $\phi=(1+\sqrt{5})/2$.

We use the Fibonacci numbers $F_n$: $$ F_{n+2}=F_{n+1}+F_n,\qquad F_0=0,\qquad F_1=1 . \tag{1}$$ With $F_{-n}=(-1)^{n+1}F_n$, the recurrence holds for all integers, positive and negative. And of course $$ F_n = \frac{\phi^n+(-\phi)^{-n}}{\sqrt{5}} \tag{2}$$

We assume $f$ is continuous and bijective. And satisfies $$ f(f(x)) = f(x) + x \tag{3}$$ Proved in another answer: $f(0)=0$. So $f$ is either increasing or decreasing. Let's consider the increasing case. So $f$ sends positive to positive and negative to negative. I claim $f(x) = \phi x$ for all $x$. Let $a > 0$, $f(a)=b$. (The case $a<0$ is below.) We will show $b = \phi a$.

Starting with $f(a)=b$, we get by induction using (1) and (3): $$ f(F_{k-1}a+F_{k}b)=F_{k}a+F_{k+1}b \tag{4}$$ for all $k \ge 0$. And applying (1), (3) backward, we get (4) for negative $k$ also. (We assumed $f$ is bijective, so we can go backward in (3).)

Since this is the increasing case, we have $b=f(a)>0$ since $a>0$. For $k>0$ consider the negative values $-k$: $$ f(F_{-k-1}a+F_{-k}b)=F_{-k}a+F_{-k+1}b \\ f\big((-1)^kF_{k+1}a+(-1)^{k+1}F_{k}b\big) =(-1)^{k+1}F_{k}a+(-1)^kF_{k-1}b $$ What is the sign of $v_{-k}=(-1)^kF_{k+1}a+(-1)^{k+1}F_{k}b$ ? If $b/a > \phi$,then by (2) the sign is $(-1)^{k+1}$ for large $k$. So $f$ sends positive to negative and negative to positive for these values $v_{-k}$. Similarly, if $b/a<\phi$, then the sign of $v_{-k}$ is $(-1)^k$ for large $k$, and again $f$ sends positive to negative and negative to positive for these values $v_{-k}$. Since this is impossible, we conclude $b/a=\phi$.

For case $a<0$, we have $b<0$. Again we get negative to positive and positive to negative for $v_{-k}$ with $k$ large.

Now consider the decreasing case. I claim $f(x) = -x/\phi$ for all $x$. We still get (4). But now $f$ is supposed to send positive to negative and negative to positive. This time we get our contradictions by considering $v_k$ for large positive $k$ in (4).

$\endgroup$
  • 1
    $\begingroup$ Pfft, continuous is lame. ;) $\endgroup$ – Hagen von Eitzen Aug 28 '15 at 17:10
  • 2
    $\begingroup$ There's no need to assume surjectivity. Injectivity follows readily and your $(4)$ for $a,b>0$ shows (using continuity) that $[0,\infty)$ is in the image. As $x\mapsto-f(-x)$ also has the given property, we conclude that also $(-\infty,0]$ is in the image. (This also shows we need only consider $a>0$) $\endgroup$ – Hagen von Eitzen Aug 28 '15 at 17:32
  • $\begingroup$ OK, how to prove surjective in the decreasing case? $\endgroup$ – GEdgar Aug 28 '15 at 20:58
7
$\begingroup$

Claim: $f(0)=0$.

Assume $f(0) = k\neq 0 $, then $f(k)=f(k)+k$, a contradiction.

Claim: If $f(x)$ is an analytic function, $f(x)=\phi x$ or $f(x)=-x/\phi$.

If $f$ is an analytic function, it has a unique polynomial expansion with zero constant term, i.e. $f(x) = a_1x+a_2x^2 +\cdots$. Then by comparing powers we get that: $$a_1=\phi, -1/\phi$$ And all other powers are zero.

If $f(x)$ is not analytic, I'm not sure what else can be said about it.

$\endgroup$
  • $\begingroup$ Great thought! Can you elaborate on "comparing powers"? $\endgroup$ – Jason Aug 28 '15 at 12:57
  • 2
    $\begingroup$ @Jason - Just write out $f=a_0 + a_1 x + a_2 x^2 + \ldots$, and compare powers of $f(f(x)-f(x)-x=0$ which all should be zero. $\endgroup$ – nbubis Aug 28 '15 at 13:06
  • $\begingroup$ If $f(0)=k,$ then $f(f(0))=f(k)=f(0)+0=k$. How did you get your first line? This implies $f(0)=f(k)$ or $k=0$ by injectivity. $\endgroup$ – snulty Aug 29 '15 at 12:22
  • $\begingroup$ Knowing that f is continuous and bijective thus monotone, can we prove that f is analytic? $\endgroup$ – Jason Sep 7 '15 at 9:24
7
$\begingroup$

For each number $x\in[1,\phi)$, let either $$f(x)=\phi x,f(-x)=-\phi x\text{ or else }\\f(x)=-x/\phi,f(-x)=x/\phi$$ The choice is independent for each $x$ in that interval. Then, for all other numbers, $f(x\phi^k)=f(x)\phi^k$.
For each ladder $x\phi^{-2},x\phi^{-1},x,x\phi,x\phi^2,...$, $f$ consistently shifts numbers up the scale, or down the scale.

$\endgroup$
  • $\begingroup$ That is not all the functions, correct? $\endgroup$ – nbubis Aug 28 '15 at 13:48
  • $\begingroup$ I don't know. There may be a way to define $f(x)$ arbitrarily over an interval, then use the function definition to extend it over $\mathbb{R}$. $\endgroup$ – Empy2 Aug 28 '15 at 13:51
  • $\begingroup$ Is there another way to break the reals into disjoint sequences $a_k\phi^n+b_k(-\phi)^{-n}$? $\endgroup$ – Empy2 Aug 28 '15 at 14:11
2
$\begingroup$

Pick a Hamel basis $(v_i)_{i\in I}$ for $\mathbb R$ over $\mathbb Q$. Group the basis vectors in pairs $(v_i,v_j)$ and on $v_i\mathbb Q+v_j\mathbb Q$ let $f$ be given (with respect to the basis $(v_1,v_2)$) by the matrix $A=\begin{pmatrix}0&1\\1&1\end{pmatrix}$. As $A^2=A+1$, $f$ has the desired property. In other words: Find $J\subset I$ with $|J|=|I\setminus J|$ and a bijection $h\colon J\to I\setminus J$ and then let $f(\sum_{i\in I} q_iv_i)=\sum_{i\in J}((q_i+q_{h(i)})v_{h(i)}+q_{h(i)}v_i)$ (those are sums, not series!)

$\endgroup$
  • 1
    $\begingroup$ These should be all $\mathbb Q$-linear solutions. My solution set overlaps with @Michael's, but neither solution set is contained in the other. $\endgroup$ – Hagen von Eitzen Aug 28 '15 at 14:45

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.