Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

We have a function:

$f(n)$ = number of occurrences of the digit $1$ in the numbers from $0$ to $n$.

Example: $f(12) = 5$

It is obvious that $f(1)=1.$

Question: Which is the next number for which $f(n) = n$?

share|cite|improve this question
11111111111111111111111111111111111287981 – The Chaz 2.0 Jun 24 '11 at 21:57
Can I write a program to solve it? – usul Jun 24 '11 at 21:58
@S4M: count the 1 in 12 as well. – Isaac Jun 24 '11 at 22:05
@Mark: The number in my first comment was meant to stir the pot; I don't think there is another solution. – The Chaz 2.0 Jun 24 '11 at 22:16
Wouldn't potential solutions to this be more likely/frequent, say, in base 2? ;-) – amWhy Jun 24 '11 at 23:22
up vote 8 down vote accepted

So I dug up my notes on this problem (for those interested this is project euler problem # 156) There is an analytical form for the solution to this problem (but alas, even the analytical form is too slow to solve the big problem, which requires finding ALL numbers that satisfy this condition for ALL digits 1-9). I will state my result first (to avoid hiding it in the text) and provide explanation for it after.


Define a list representation of a number $n$ to be (following a notation I borrowed from continued fractions)

$$n = r_k*10^k + r_{k-1}*10^{k-1} + ... + r_0*10^0 \equiv [r_k,r_{k-1},...,r_0]$$ The function $f(d,n)$ which gives the number of occurances of a digit $d$ up to the number $n$ is given by $$ f(d,n) = \sum_{j=0}^k\left(\sum_{i=0}^{r_j} (10^j\delta_{i-1,d})+ r_jE(j) +\delta_{r_j,d}(n[j:]+1)\right)$$

where the notation $n[j:]$ is used to signify the number formed by the last $j$ digits (e.g. for $n=1234=[1,2,3,4]$, $n[1] = 4$, $n[3] = 234$) and $E(j) = j*10^{j-1}$

Typing this into mathematica (and checking ($f(1,n) = n$ gives the next result of $\mathbf{199981}$).

NOTE: This can be generalized even further to any base $B$ by replacing occurrences of $10^k$ with $B^k$


First observe that in the number 0-9 any digit appears only 1 time.

In the numbers 0-99 any digit appears 10 + 10 times (10 times as the ones digit and 10 times in the range x0-x9)

In the numbers 0-999 any digit appears 20*10+100 = 300 times (20 times in each range of 100+ plus 100 in the range x00-x99)

It is not difficult to prove that this pattern continues, for any number given as $10^k-1$ there are $k*10^{k-1}$ appearances of any digit in that range.

This motivates my definition of the function

$$E(j) = j*10^{j-1}$$

However, the number given is unlikely to be as nice as $10^k-1$ we have to learn a clever way to add up the parts of that range. For a number (given in the above notation) like $[4,0,0,0]$. We see that the range $0-999$ occurs 4 times ($0-999$,$1000-1999$,$2000-2999$,$3000-3999$. So we are left with $4*E(3) = 1200$ occurrences of the digit $1$ (this motivates the $r_jE(j)$ term). However, in the range 1000-1999 the digit one occurs an extra 1000 (or $10^j$) times, this extra term only happens if $r_j$ is greater then the digit we are summing over (e.g. if we are adding occurrences of the digit $9$ the answer would be 1200 and not 2200), the best way I could think of to express this conditional statement analytically was with the sum $\sum_{i=1}^{r_j}(10^j\delta_{i-1,d})$

Calculating a number like $[4,1,2,4]$ is not much more difficult, we simply iterate over the sequence $[4,0,0,0]$, $[3,0,0]$, $[2,0]$,$[4]$ following the formula in the previous paragraph. However, the method above will miscount because it ignores the extra occurrence of the digit $1$ in the range $4100-4124$. So at every step of the iteration we have to make sure there aren't any tails that we are forgetting to count, which motivates the $\delta_{r_j,d}(n[j:]+1)$ term.

share|cite|improve this answer
Hi ! I was working over this problem 156, and came up with a similar solution (wrote a program) - but i have been struggling to find out HOW to know when to stop.. How does one know that we've exhausted all possible solutions for [f(d,n) - n = 0] ?? My program almost ran out of memory trying to figure out a pattern !! – trinity Aug 1 '13 at 12:56

Ok this is simple to do with Mathematica.

For[i = 0; j = 0, i <= 200000,
 i++, j += (Plus @@ Cases[IntegerDigits[i], 1]); If[j == i, Print[i]]]

The result is


Therefore the number you search has to be 199981.

share|cite|improve this answer
While this is a correct way to approach this particular question, the actual problem in question (which is a project Euler problem) requires the sum total of all numbers that satisfy this condition for all digits 1-10. Not so easy (or fast) do with this brute force method (where do you set the upper search limit for instance?). – crasic Jun 24 '11 at 23:26
You can easily make it stop after it finds the first wanted number. I leave this modification to you. I agree that with some thoughts you could still fasten up the calculation a lot (factor 100 at least). – Listing Jun 25 '11 at 7:21
I dont see why 199984, 199985,...,199990 are not solutions as well as we have $f(199984)=f(199983)+1=199984$ and same for every number until 199991. – S4M Jun 25 '11 at 19:53
You are right I didn't mark properly that I left out the other solutions, now its correct. – Listing Jun 25 '11 at 20:19
fasten up the calculation I like it. How about slowen down the calculation? – GEdgar Jun 25 '11 at 20:23

I think I have the beginning of an idea to solve this. We can start by noticing that $f(10^{n+1}-1)= 1+10^n+10\times f(10^n-1)$ because when you list all the numbers from 1 to $10^{n+1}$ you get: 1... $10^n-1$, $10^n$,..., 1999...99, 20...00,...$10^{n+1}-1$. from $10^n$ to 1999...99 you can see $10^n+f(n)+1$ the digit '1', and from 20...00 to $10^{n+1}-1$ you get $9\times f(10^n-1)$ the digity '1'. So let's call $u_n=10^n-1$ and $v_n=f(u_n)$.

I can see that $u_9 >v_9$ and $u_{10} < v_{10}$, so the anwswer should be in beetween (or at least smaller than $10^{10}-1$.

share|cite|improve this answer
$f(10^{n+1})=(n+1)10^n$. – André Nicolas Jun 25 '11 at 5:20
hmm that's true. So $10^{9+1}$ is also a solution. This problem keeps fascinating me, I want to find ALL the solutions to $f(n)=n$ ! – S4M Jun 26 '11 at 8:53
How many solutions are there to $f(n) = n!$ ?? $$$$ :) – The Chaz 2.0 Jun 27 '11 at 17:25

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.