5
$\begingroup$

A somewhat unintuitive result of real analysis is that decimal expansions are not unique. For example, $$0.99999...=1.$$ So it can be gathered that every real number has at least one base-$10$ decimal expansion, sometimes even two. But is two the maximum? Are there any real numbers with three different base-$10$ decimal expansions?

Intuitively, I would think not, but I have close to no idea how to prove it other than knowing that the easiest method would be a proof by contraction.

It may help to restrict the problem by considering the expansions of numbers only in $[0,1]$. That is because if $a\in[0,1]$ has more than two base-$10$ decimal expansions, then so does $a+x$ for all $x\in\Bbb R$. And likewise, if $x\in\Bbb R\setminus [0,1]$ has more than two base-$10$ expansions, it can be written as $x=\mathrm{sgn}(x)(\lfloor x\rfloor+a)$, where $a\in[0,1)$, and by necessity $a$ has more than two base-$10$ expansions.


To be clear, I should define what I mean by base-$10$ expansion. Let $N\in[0,1]$. Then a decimal expansion of $N$ is a sequence $\delta=(\delta_0,\delta_1,\delta_2,...)$ of integers $0\le \delta_i\le 9$ such that $$\sigma(\delta):=\sum_{i\ge0}\frac{\delta_i}{10^i}=N.$$ Furthermore, let $\mathcal U=\{(a_0,a_1,a_2,...):0\le a_i\le 9,\, a_i\in\Bbb Z\}$ and let $$D_N=\{\delta\in\mathcal U :\sigma(\delta)=N\}.$$ Lastly, let $$\mathcal C_k=\{N\in[0,1]:\#(D_N)\ge k\},\qquad k\in\Bbb N$$ where $\#(S)$ is the number of elements in the set $S$.

So, is $\mathcal C_k$ empty for $k>2$?

$\endgroup$
4
  • $\begingroup$ The only way to get two is if the digits eventually stabilize to zero. You seem to know that if you're in such a case that there are indeed two. If they don't stabilize to zero, then let $\delta_n$ be the first digit that two representations differ. Then sum over $\Sigma_n^\infty 10^{-k}(\delta_k - \delta_k')$ and this will be less than one (but not zero) so mod $1$ they'll be different numbers. For a number that eventually stabilizes (say at decimal $n$), multiply by $10^n$ and just use a limit argument from either side on your interval $[0,1]$. $\endgroup$ Feb 9 at 20:36
  • $\begingroup$ Would $1.000000000.....$ with the "last" digit of $1$ after an infinite numbers of $0$ would qualify as a different expansion? $\endgroup$
    – Andrei
    Feb 9 at 20:36
  • $\begingroup$ @Andrei Does it fit the given definition of decimal expansion? If so, yes, if not, no. $\endgroup$
    – user239203
    Feb 9 at 20:40
  • $\begingroup$ @Andrei yes it would count, as $\delta=(1,0,0,0,...)$ does indeed satisfy $$\sigma(\delta)=1$$ $\endgroup$
    – clathratus
    Feb 9 at 20:43
7
$\begingroup$

$\cal C_k$ is indeed empty for $k > 2$.


Suppose that a number has two distinct decimal expansions $\delta = (\delta_0, \delta_1, \delta_2, \ldots)$ and $\delta' = (\delta_0', \delta'_1, \delta'_2, \ldots)$. Then, we must have

$$0 = \sum_{i \ge 0}\dfrac{\delta_i - \delta'_i}{10^i}.$$

Wlog, we may assume that $\delta_0 \neq \delta'_0$. (There must be some smallest $i$ for which they are unequal. By multiplying by a suitable power of $10$, we may assume that it is $i = 0$.) Furthermore, we may assume that $\delta_0 > \delta'_0$.

Thus, we have $$1 \le \delta_0 - \delta'_0 = \sum_{i \ge 1}\dfrac{\delta'_i - \delta_i}{10^i}.$$

Taking absolute value on all sides and using triangle inequality for series, we see that

$$1 \le \delta_0 - \delta'_0 \le \sum_{i \ge 1}\dfrac{|\delta'_i - \delta_i|}{10^i} \le \sum_{i \ge 1}\dfrac{9}{10^i} = 1.$$

Thus, we have equality throughout. Note that this means that $|\delta'_i - \delta_i| = 9$ for all $i$. In fact, we see that $\delta'_i - \delta_i$ must have the same sign for all $i$. This sign must, of course, be positive.

Thus, we have $\delta_0 = \delta'_0 + 1$ and $\delta_i = 0$ for all $i \ge 1$, $\delta'_i = 9$ for all $i \ge 1$.

(All the manipulations above were justified as the series converged absolutely.)


The above shows that if a number has two decimal expansions, it must actually just be a variant of $1 = 0.\bar{9}$. In particular, there are at most two decimal expansions.

$\endgroup$
7
  • $\begingroup$ So, $\mathcal C_{3}$ is empty because there are at most $2$ ways to use the $1=0.\bar{9}$ trick? $\endgroup$
    – clathratus
    Feb 9 at 20:48
  • 1
    $\begingroup$ Yes. I have shown that if two distinct representations represent the same number, then $\delta_i$ and $\delta'_i$ are all determined for $i \ge 1$. Moreover, $\delta_0$ and $\delta'_0$ will be fixed by looking at $\lfloor x \rfloor$. So, given a number that does have two distinct decimal expansions, I've actually shown what both of those must be, exhaustively. (In particular, there's no third possibility.) $\endgroup$ Feb 9 at 20:51
  • 1
    $\begingroup$ Ah I see. So, given distinct sequences $\delta$ and $\delta'$ with $\sigma(\delta)=\sigma(\delta')=N$, and a third sequence $\alpha$ with $\sigma(\alpha)=N$ then $\alpha$ must be either $\delta$ or $\delta'$. Very clever! +1 $\endgroup$
    – clathratus
    Feb 9 at 20:55
  • $\begingroup$ What about for any base? $\endgroup$
    – AMDG
    Jul 22 at 17:02
  • $\begingroup$ @AMDG: (I'm assuming that your base is still an integer.) Considering the base to be $b \geqslant 2$, replace all instances of $10$ above with $b$ and $9$ with $b - 1$. The analogous result follows. (Note that $$\sum_{i \geqslant 1} \frac{b - 1}{b^i} = 1$$ still holds.) $\endgroup$ Jul 22 at 17:23

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.