-2

Is it true that there exist a continuous function f that for every algebraic number q , his image f(q) is a rational number?

Thank you for your answers

Adam Hughes
  • 36,777

2 Answers2

2

Yes -- there is even a strictly increasing continuous bijection $\mathbb R\to\mathbb R$ such that the image of the algebraic reals is exactly the rational numbers.

The key to this is that both the algebraic reals and the rational numbers are countable dense total orders without first or last element, and it is a general theorem that every such two sets are order-isomorphic. Since both sets are dense in $\mathbb R$, an isomorphism between them extends uniquely to a continuous function $\mathbb R\to\mathbb R$.

(Even better, every strictly increasing continuous bijection $\mathbb R\to\mathbb R$ can be uniformly approximated by such functions).

1

You can even do it with an entire function.

EDIT: Here's one way to construct an entire function that maps $\mathbb R$ to $\mathbb R$, $\mathbb A$ to $\mathbb Q[i]$, and is one-to-one on $\mathbb R$. Here $\mathbb A$ is the algebraic numbers (not necessarily real) and $\mathbb Q$ the rationals.

Start by fixing an enumeration $a_k$ of $\{z \in \mathbb A: \text{Im}(z) \ge 0\}$. The function will be of the form

$$ f(z) = z + \sum_{k=1}^\infty f_k(z)$$

where $f_j$ are chosen inductively so that, writing $F_j(z) = z + \sum_{k=1}^j f_k(z)$,

  1. $f_j(\overline{z}) = \overline{f_j(z)}$. In particular, $f_j$ maps $\mathbb R$ into $\mathbb R$.
  2. $f_k(a_j) = 0$ for $j < k$. Thus $f(a_j) = F_j(a_j)$.
  3. $F_j(a_j) \in \mathbb Q[i]$.
  4. $|f_j(z)| \le 2^{-j}$ for $|z| < j$. This will imply that the series converges uniformly on compact sets to an entire function.
  5. $|f'_j(z)| \le 2^{-j-1}$ on $\mathbb R$. This will imply that $f'(z) \ge 1/2$, so $f$ maps $\mathbb R$ one-to-one onto $\mathbb R$.

Given $f_k$ for $k < j$, we start with $$ h(z) = e^{-z^2} \prod_{k < j} (z - a_k)(z - \overline{a_k})$$ which satisfies (1) and (2).

If $\alpha_j \in \mathbb R$, then $h(\alpha_j) \in \mathbb R \backslash \{0\}$ and $F_{j-1}(\alpha_j) \in \mathbb R$. Since $h$ is bounded on $\{z: |z| \le j\}$ and $h'$ is bounded on $\mathbb R$, there is $\epsilon > 0$ such that (4) and (5) are true with $f_j (z) = t h(z)$ as long as $|t| < \epsilon$, and (3) will be true for a dense set of $t \in (-\epsilon, \epsilon)$.

If $\alpha_j \notin \mathbb R$, (3) is a bit more complicated. We therefore take $f_j(z) = (t + s z) h(z)$: there will still be $\epsilon > 0$ such that (4) and (5) hold for $(s,t) \in (-\epsilon, \epsilon)^2$, and for a dense set of such $(s,t)$ we will have (3).

Robert Israel
  • 448,999
  • Hmm, I see. Can you also achieve both $f(\mathbb R)=\mathbb R$ and $f(\mathbb A)=\mathbb Q[i]$ with a bijective entire function? – hmakholm left over Monica Apr 27 '16 at 19:00
  • Bijective on $\mathbb R$, I hope you mean. There aren't too many entire functions bijective on $\mathbb C$. – Robert Israel Apr 27 '16 at 19:02
  • Could you elaborate? I'd guess this is an instance of a more general theorem, could you state it? – Wojowu Apr 27 '16 at 19:12
  • @Robert: Hmm, right. So the best we can do for the complex algebraics is probably an entire function $f$ such that $f^{-1}(\mathbb Q[i])\supsetneq \mathbb A$. – hmakholm left over Monica Apr 27 '16 at 19:14
  • I think we can still have $f^{-1}(\mathbb Q[i]) = \mathbb A$ by modifying the construction, ensuring at stage $j$ that any $w$ with $|w| \le j$ for which $F_j(w)$ is of the first $j$ members of $\mathbb Q[i]$ (in a fixed enumeration) is in $\mathbb A$, and making $f_k(w) = 0$ for $k > j$. – Robert Israel Apr 27 '16 at 20:45
  • @RobertIsrael: Hmm, I'm not convinced. If we look at some transcendental $w$, I'll buy that your modification can ensure that $F_j(w)$ eventually avoids every member of $\mathbb Q[i]$. But that won't necessarily prevent $\lim_{j\to\infty} F_j(w)$ from being in $\mathbb Q[i]$ anyway. You can't require $f_k(w)=0$ for every $w$ that is in danger of converging towards $\mathbb Q[i]$ because that's all of them, uncountably many! – hmakholm left over Monica Apr 27 '16 at 21:25
  • If $f(w) = z \in \mathbb Q[i]$ with multiplicity $m$, take $r > 0$ such that $w$ is the only zero of $f - z$ within distance $r$ of $w$. By Rouché's theorem there exists $N$ such that for $n > N$, $F_n - z$ has exactly $m$ zeros (counted by multiplicity) within distance $r$ of $w$. But by construction, such zeros (if $n$ is large enough) are in $\mathbb A$ and will also be zeros of $F_k - z$ for all $k > n$, and therefore of $f - z$: thus they must be $w$, and $w \in \mathbb A$. – Robert Israel Apr 27 '16 at 22:11
  • @RobertIsrael: Still not convinced. What prevents the $m$ zeroes of $F_n-z$ from being $a_j$s with $j>n$, and to keep moving to even later-in-the-enumeration $a_j$s as $n$ grows? A sequence of such $a_j$s would be able to converge to a transcendental $w$. – hmakholm left over Monica Apr 27 '16 at 22:29
  • Further, it seems that the construction can easily be amended to make sure that all of the $F(a_j)$s are different, without losing anything you depend on. But $f$ is entire, definitely not affine, so it cannot be injective. Thus there will be $z_1\ne z_2$ with $f(z_1)=f(z_2)$. But then there is a neighborhood of $f(z_1)$ that is the continuous image of disjoint neighborhoods of $z_1$ and $z_2$, such that a rational point in this neighborhood will be the image of both a point near $z_1$ and a different point near $z_2$. By construction these won't both be algebraic. – hmakholm left over Monica Apr 27 '16 at 22:34
  • I think you misunderstand what I had in mind, but it'll be better to edit the answer rather than trying to clarify with further comments. I don't have time right now, but maybe later tonight... – Robert Israel Apr 28 '16 at 00:27
  • Well, it does get complicated trying to write out the details. But I'm quite confident the result is true. In the amended construction, the value at $a_j$ may be assigned before stage $j$, in which case it won't be reassigned. These values cannot be all different. For a particular $w \in \mathbb Q[i]$ and $R > 0$, the roots of $F_j(z) - w$ with $|z| < R$ will be fixed at some stage, and they can't move after that. – Robert Israel Apr 28 '16 at 23:18