Function whose image of every open interval is $(-\infty,\infty)$

Question

How to find a function from reals to reals such that the image of every open interval is the whole of R?

Is there one which maps rationals to rationals?

Answer 1

Though this can be done explicitly with enough cleverness (for example with the Conway base 13 function), I rather like the following choice-based argument, which requires almost no thought once you're familiar with transfinite recursion.

Consider the set consisting of all ordered triples of reals $(a,b,c)$ with $a<b$. This has the cardinality of the continuum. By the axiom of choice, it is well-orderable; pick a well-ordering of smallest possible order type, so that every triple has less than continuum-many predecessors.

Now we build our function $f$ recursively along this well-order, fixing the value of $f$ at one new point at each step. At the step corresponding to $(a,b,c)$, we want to ensure that there exists a point $x$ in the open interval $(a,b)$ such that $f(x)=c$. Since we've only fixed $f$ at less-than-continuum-many points, and $(a,b)$ has cardinality continuum, we can choose an $x$ in $(a,b)$ such that $f(x)$ is not yet fixed, and fix $f(x)$ to be $c$.

This recursion gives us a partial function from $\mathbb{R}$ to $\mathbb{R}$ that already satisfies our requirements. We can make it total by just setting $f(x)$ to be $0$ (say) wherever $f(x)$ is not yet defined.

If we additionally want $f$ to map rationals to rationals, we can simply set $f(x)=0$ for every rational $x$ before commencing the recursion.

Answer 2

With a minor modification, Qiaochu’s deleted answer works fine. Let $\mathfrak{C}=\left(\Bbb R/\Bbb Q\right)\setminus\{\Bbb Q\}$, and let $h:\mathfrak{C}\to\Bbb R$ be a bijection. Each coset $C\in\mathfrak{C}$ is dense in $\Bbb R$, so $C\cap(a,b)\ne\varnothing$ for each open interval $(a,b)$ in $\Bbb R$ and $C\in\mathfrak{C}$. Now define

$$f:\Bbb R\to\Bbb R:x\mapsto\begin{cases} h(x+\Bbb Q),&\text{if }x\in\Bbb R\setminus\Bbb Q\\ x,&\text{if }x\in\Bbb Q\;; \end{cases}$$

then $f[I]=\Bbb R$ for every non-empty open interval $I$ in $\Bbb R$, and $f[\Bbb Q]=\Bbb Q$.

A slightly different approach is to think of $\Bbb R$ as a vector space over $\Bbb Q$ and let $B$ be a basis for $\Bbb Q$. Without loss of generality assume that $1\in B$. $|B|=|\Bbb R|$, so there is a bijection $h:B\setminus\{1\}\to\Bbb R$. Define

$$f:\Bbb R\to\Bbb R:x\mapsto\begin{cases} h(b),&\text{if }x\in b\Bbb Q\setminus\{0\}\text{ for some }b\in B\setminus\{1\}\\ 0,&\text{otherwise}\;; \end{cases}$$

$b\Bbb Q\cap c\Bbb Q=\{0\}$ whenever $b,c\in B$ with $b\ne c$, so $f$ is well-defined. For each $b\in B$ the set $b\Bbb Q$ is dense in $\Bbb R$, so $f[I]=\Bbb R$ for every non-empty open interval $I$ in $\Bbb R$. Finally, $f[\Bbb Q]=\{0\}\subseteq\Bbb Q$. If it’s desired to make $f[\Bbb Q]=\Bbb Q$, just modify the definition of $f$ slightly:

$$f:\Bbb R\to\Bbb R:x\mapsto\begin{cases} h(b),&\text{if }x\in b\Bbb Q\setminus\{0\}\text{ for some }b\in B\setminus\{1\}\\ x,&\text{if }x\in\Bbb Q\\ 0,&\text{otherwise}\;. \end{cases}$$

Of course the same basic idea works with any family of $2^\omega$ pairwise disjoint dense subsets of $\Bbb R$.

(But I find transfinite recursion the most natural approach.)

Answer 3

First construct a nowhere-dense perfect set (i.e. a compact set with no isolated points) $E \subset (0,1)$ that contains no rationals. This is not hard to do by a modification of the construction of the Cantor set. Note that $E$ has the cardinality of the continuum, and in fact it's not hard to construct a function $f_E$ from $E$ onto $\mathbb R$.

Now enumerate the intervals with rational endpoints as $(a_i, b_i)$. We can take sets $E_i \subset (a_i, b_i)$ that are pairwise disjoint, of the form $E_i = s_i + t_i E = \{s_i + t_i e: e \in E\}$ where $s_i$ and $t_i$ are rational and $t_i > 0$: in fact to choose $s_i$ and $t_i$ it suffices to have $[s_i, s_1 + t_i] \subset (a_i, b_i)$ disjoint from $\bigcup_{j=1}^{i-1} E_i$, which can be done since the latter is nowhere dense. Now for $x \in E_i$ take $f(x) = f_E((x - s_i)/t_i)$. Thus $f$ maps $E_i$ onto $\mathbb R$, and since every open interval contains some $E_i$ the image of such an interval is all of $\mathbb R$. Now define $f$ however you want on the complement of $\bigcup_i E_i$ (in particular you could take $f$ to be rational on the rationals).