3
$\begingroup$

The function needs to fit the following requirements: Function $f:\mathbb{Q} \to \mathbb{Q}$ where $\vert{f(x)-f(y)}\vert$ $\le$ $1/2$$\vert x-y \vert$ $\forall x,y \in \mathbb{Q}$ and $f(x')\neq x'$ $\forall x'\in \mathbb{Q}$

  • 0
    It seems that $x'$ is in $A,$ but you have $f$ defined on rationals. So should $f$ be defined actually on $A$?2017-01-18
  • 0
    What did you try?2017-01-18
  • 0
    In order to make it so that no number is mapped on to itself could you make it such that the function multiples by an irrational number but I just need a function that fits the inequality2017-01-18
  • 0
    Could the function map all numbers to $\sqrt{2}$ for example2017-01-18
  • 0
    The answer depends on $A$. For example, if $A = \{0\}$ then no such function exists. On the other hand, if there is an irrational $a\in A$, then $f(x) = a$ satisfies the requirement (like Bradley Hill suggested).2017-01-18
  • 0
    I've changed the question so that it must go from $\mathbb{Q} \mapsto \mathbb{Q}$ I can't think of an example of this2017-01-18
  • 1
    Hint: if $|f(x)-f(y)| \leq 1/2 |x-y|$ then $f$ is contractive. Yet you say $f$ has no rational fixed point. That means that you will need the limit of fixed point iterations of $f$ to converge to an irrational number (because it is impossible for it to fail to converge altogether, in view of the Banach fixed point theorem). Do you know any fixed point iteration methods which converge to an irrational number?2017-01-18
  • 0
    You should use `\to` instead of `\mapsto`.2017-01-18
  • 0
    @Enmar Changed.2017-01-18
  • 0
    @Ian I can't think of any immediately2017-01-18
  • 0
    Do you know a fixed point iteration technique for computing $\sqrt{2}$?2017-01-18
  • 0
    Not that I can think of2017-01-18
  • 0
    @Ian I've tried searching for it but I can't really find anything2017-01-18
  • 0
    ...OK, well, here's Newton's method for computing $\sqrt{2}$: $f(x)=x/2 + 1/x$. This is not contractive everywhere (the derivative is $1/2-1/x^2$, which is larger than $1$ in absolute value near $x=0$), nor is it even defined at $x=0$, so that alone is not a solution to your problem. But one would expect that modifying the function near $x=0$, or maybe on $(-\infty,a]$ for some $a<\sqrt{2}$, would result in a solution to your problem...2017-01-18
  • 0
    So i can set $f(0)=0$ and $f(x)=-x/2-1/x$ for $x \in (-\infty,-\sqrt{2})$ but I can't figure out what to do in the intervals $[-\sqrt{2},0)$ and $(0,\sqrt{2}]$2017-01-18
  • 0
    Whatever you're going to do is going to have to have a "slope" of at most $1/2$ at every point. So for example you could take $f(x)=x/2+1/x$ on $[a,\infty)$ where $a<\sqrt{2},-1/2 \leq 1/2-1/a^2 \leq 1/2$, and $a$ is rational. Then extend $f$ to $(-\infty,a)$ from there, for example by defining it to be $f(a)-(x-a)/2$. Then the function is globally contractive on $\mathbb{R}$, so it has just one fixed point, which we know by construction is not rational.2017-01-18
  • 0
    So would $f(x)$ =$x/2+1/x$ for $x\in[1,\infty]$ , $f(1)-(x-1)/2$ for $x\in(-\infty,1)$2017-01-18
  • 0
    That was what I came up with, yes. It's worth plotting this function, alongside $g(x)=x$, to see what's going on.2017-01-18
  • 0
    Okay thank you, sorry you had to effectively walk me through it all, and yeah I will do2017-01-18

1 Answers 1

1

OP already found an answer, but I will formalize the reasoning here.

Let $f\colon\Bbb Q\to\Bbb Q$ such that $$|f(x)-f(y)|\leq \frac 12|x-y|,\ \forall x,y\in\Bbb Q.\tag{1}$$

Since $\Bbb R$ is metric space completion of $\Bbb Q$ and $f$ is uniformly continuous, there is unique uniformly continuous function $\bar f\colon \Bbb R\to\Bbb R$ such that $\bar f(x) = f(x),\ \forall x\in\Bbb Q$. From continuity it directly follows that for any $x\in\Bbb R$ and any sequence $(x_n)$ in $\Bbb Q$ converging to $x$ in $\Bbb R$, we have $$\bar f(x) = \lim_nf(x_n)$$ and thus the relation $(1)$ is true for $\bar f$ on whole $\Bbb R$ as well. Hence, it is not loss of generality to assume that $f\colon\Bbb R\to\Bbb R$ in the first place, with $f(\Bbb Q)\subseteq \Bbb Q$.

By Banach fixed point theorem, there exists unique fixed point $a$ of $f$.

Let $g(x) = x - f(x)$. Note that $a$ is a fixed point of $f$ if and only if $g(a) = 0$, and that $f(\Bbb Q)\subseteq \Bbb Q$ if and only if $g(\Bbb Q)\subseteq \Bbb Q$. Since you are looking for $f$ such that $f(x)\neq x,\ \forall x\in\Bbb Q$, the unique fixed point of $f$ must be irrational, i.e., $g$ has unique irrational zero.

We can now rewrite $(1)$ as follows:

\begin{align} |f(x)-f(y)|\leq \frac 12|x-y| &\iff |g(x)-g(y)-(x-y)|\leq\frac 12|x-y|\\ &\iff \left|\frac{g(x)-g(y)}{x-y}-1\right|\leq\frac 12\\ &\iff \frac 12\leq \frac{g(x)-g(y)}{x-y}\leq \frac 32\tag{2} \end{align}

Relation $(2)$ can be used to derive simple sufficient condition on $g$ to get $f$ with desired properties:

Let $g\colon\Bbb R\to\Bbb R$ be differentiable such that $\frac 12\leq g'(x)\leq \frac 32, \forall x\in\Bbb R,$ and let $g$ have irrational zero. Then $f(x)=x-g(x)$ (restricted on $\Bbb Q$) satisfies condition $(1)$ and has no rational fixed point. If $g(\Bbb Q)\subseteq\Bbb Q$, then $f(\Bbb Q)\subseteq\Bbb Q$ as well.

Proof. From the above discussion, all that remains to be proven is that $\frac 12\leq g'(x)\leq \frac 32$ implies $(2)$. Assume that there exist some $x,y\in\Bbb R$ such that $$\frac{g(x)-g(y)}{x-y}\not\in\left[\frac 12,\frac 32\right].$$ By the mean value theorem, there exists some $c\in\Bbb R$ such that $g'(c)\not\in \left[\frac 12,\frac 32\right]$. Contradiction. $\tag*{$\square$}$

Using this, you should be able to come up with as many examples as you'd like, but let me just discuss the one Ian suggested in the comments. It comes from Newton's method for numerical approximation of root of $x^2-2$. So, we could let $g(x) = \frac{x^2-2}{2x}$, but there are some troubles with it. First of all, it is not defined at $0$. Secondly, $g'(x) = \frac 12 + \frac 1{x^2}$ and $g'(x)\in\left[ \frac 12,\frac 32\right]$ if and only if $x^2\geq 1$. This leads to considering changing definition of $g$ on $(-\infty,1)$. We want to do it in simplest way possible so that $g$ remains differntiable and $g'(x)\in\left[ \frac 12,\frac 32\right]$. Since $g(1) = -\frac 12$ and $g'(1)=\frac 32$, we consider function $x\mapsto\frac 32 x - 2$, i.e. we can define

$$g(x) =\left\{ \begin{array}{c l} \frac{x^2-2}{2x}, &x\geq 1\\ \frac 32 x-2, &x<1 \end{array}\right.$$ which is differentiable and satisfies $g'(x)\in\left[ \frac 12,\frac 32\right]$ on whole $\Bbb R$. This gives your $f$.