9
$\begingroup$

I'm curious about some properties of the powers of 5 $$5^2=25,\quad5^3=125,\quad 5^4=625,\quad 5^5=3125,\quad ...$$ Is it true that at least $50$% of the digits in the decimal representation of $5^n$ are non-zero? This seems pretty modest since assuming each digit will be equally likely, only about $10$% of the digits will be zero on average. The first zero occurs at $5^8=390625$ and the power with the largest percentage of zeros seems to be $$5^{45}=28421709430404007434844970703125$$ in which $\approx 22$% of the digits are zero. I checked up to $5^{1000}$.

The difficulty is the statement seems so obvious from a probabilistic perspective, yet I can't pin down any definite theorems!

Obviously the first and the last 3 digits will always be non-zero so we have at least 4 non-zero digits. But I'm hoping to prove some properties about $5^n$ in general which require the non-zero digits to be at least linear in $n$. So $50$% would be more than enough. Really any probability $\epsilon>0$ will do -- bigger the better though.

Maybe analyze $\langle 5 \rangle^\times$ in $\mathbb{Z}/10^k\mathbb{Z}$? Could probability theory produce the bound in question?

$\endgroup$
  • 3
    $\begingroup$ I suspect this is too hard for today's mathematics. It's probably not even known that for all sufficiently large values of $n$ there's at least one zero in the decimal expression for $5^n$. $\endgroup$ – Gerry Myerson Jul 1 at 2:41
  • 1
    $\begingroup$ This question has some information about the appearance of digits in powers. These seem like hard questions! math.stackexchange.com/questions/116026/… $\endgroup$ – Steve Kass Jul 11 at 0:24
  • $\begingroup$ recursive $5^n=25+100\cdot {5^{n-2}-1\over 4}\quad n>1$ $\endgroup$ – Roddy MacPhee Aug 12 at 20:23
7
$\begingroup$

A good picture is worth a thousand words. It seems that the number of zeroes is growing in an almost linear fashion with a fair amount of "noise" around the meadian line. Here are the results for a number of zeroes in $5^n$ up to $n=10,000$

enter image description here

Not much will change if you extend the range to $n=20,000$:

enter image description here

Linear fit gives the following approximation:

$$n_{zero}=0.0699383 n-0.606536$$

...which means that the number of zeroes is roughly around 7%.

Mathematica provides the following regression analysis:

$$\begin{array}{l|llll} \text{} & \text{Estimate} & \text{Standard Error} & \text{t-Statistic} & \text{P-Value} \\ \hline 1 & -0.606536 & 0.354669 & -1.71015 & 0.0872544 \\ x & 0.0699383 & 0.0000307141 & 2277.07 & 5.874\cdot 10^{-24155} \\ \end{array}$$

If you try to fit the data with a parabola you get the following approximation:

$$n_{zero}=0.0931693 + 0.0697284 n+1.0494\cdot 10^{-8}n^2$$

In the given range, the quadratic term is almost neglectable which supports the conjecture that the relation between the exponent and the number of zeroes is approximately linear.

EDIT: Fun fact: $5^{58}$ has not a single zero.

EDIT 2: Mathematica code to play with:

countZeros[n_] := Module[
   {m, cnt, d},
   m = n;
   cnt = 0;
   While[m > 0,
    d = Mod[m, 10];
    If[d == 0, cnt++];
    m = Quotient[m, 10];
    ];
   Return[cnt];
   ];

analyzeExponents[limit_] := Module[
   {i, data, exp},
   i = 0;
   exp = 1;
   data = {};
   While[i <= limit,
    i++;
    exp *= 5;
    AppendTo[data, {i, countZeros[exp]}];
    If[Mod[i, 100] == 0, Print["Reached i=", i]];
    ];
   Return[data];
   ];

ListPlot[analyzeExponents[10000]]
$\endgroup$
  • $\begingroup$ (+1) Would you mind sharing the code from Mathematica? $\endgroup$ – TheSimpliFire Jul 1 at 8:10
  • 2
    $\begingroup$ There will be about $n\log_{10}5=0.7n$ digits, so that would make it about ten percent of the digits. The first one and last three won't be, but that is a small effect. $\endgroup$ – Empy2 Jul 1 at 8:14
  • $\begingroup$ @TheSimpliFire I have updated the answer with the code. $\endgroup$ – Oldboy Jul 1 at 9:25
  • 2
    $\begingroup$ These calculations confirm our intuitions, but of course they prove nothing. $\endgroup$ – Gerry Myerson Jul 1 at 10:38
  • 1
    $\begingroup$ Not only this proves nothing, but it's barely related with the OP's goal: the question is not about the mean or approximate average number of zeroes, but about bounding the worst case. That is, is $z_n$ is the number of zeroes, we want to find some $\alpha$ such that $z_n/n \le \alpha $ for all $n$. In particular we want to know if $\alpha=0.5$ works . $\endgroup$ – leonbloy Jul 1 at 12:17
1
$\begingroup$

Turns out one good proof direction is hopeless -- but still informative.

Suppose we could bound the number of consecutive zeros that appear in $5^n$. So some statement like "No more than 4 consecutive zeros occur in the decimal expansion of $5^n$". That would give us our result since it would follow that at $1$ in every $5$ digits -- so at least $20$% -- are nonzero.

This turns out not to be the case. One can find arbitrarily long sequences of zeros in $5^n$. In fact, it is claimed that there exist arbitrarily long sequences of zeros in $5^{m+2^m+2}$ for sufficiently large $m$. For example, $$5^{2+2^2+2}=5^8=390625$$ $$5^{5+2^5+2}=5^{39}=\ ...30078125$$ $$5^{8+2^8+2}=5^{266}=\ ...10009765625$$ $$5^{12+2^{12}+2}=5^{4110}=\ ...100006103515625$$ $$5^{15+2^{15}+2}=5^{32785}=\ ...700000762939453125$$ $$5^{18+2^{18}+2}=5^{262164}=\ ...900000095367431640625$$ One can achieve this result by analyzing the $k$th digit of $5^n$ as $n$ varies. For example if we let $$5^n=\sum_{k=0}^\infty a_n(k)10^k$$ where each $a_n(k)$ is one of $0,...,9$. Denote the sequence of the $k$th digit as $$S_k=\{a_n(k)\}_{n=0}^\infty.$$ One can show that for all $k\ge1$ that $S_k$ begins a repeated cycle of length $2^{k-1}$ at $n=k+1$. The proof I found of this was tedious so I'll neglect to include it here. The important bit is this though: the first $\lfloor k\log_5 10\rfloor$ digits of $S_k$ must be zero due to the fact that $10^n$ grows faster than $5^n$. It follows that the repeated cycle of $S_k$, call it $C_k$, contains an increasingly longer prefix of zeros. Example, $$C_2=\{1,6\}$$ $$C_3=\{0,3,5,8\}$$ $$C_5=\{0,0,3,9,7,8,1,7,5,5,8,4,2,3,6,2\}$$ $$C_7=\{0,0,0,4,4,2,0,1,...\}$$ The values $n=m+2^m+2$ are simply the powers for which the prefixes of these cycles all line up again.

Some interesting corollaries also emerged from the proof:

$\bullet$ The distribution of digits in each $S_k$ tends to perfect equalibrium -- so $10$% of each digit -- as $k$ grows.

$\bullet$ Each $S_k$ obeys a sort of mirror image law in which there are equally many $0$'s and $5$'s, $1$'s and $6$'s, $2$'s and $7$'s, etc.

$\endgroup$

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.