ON A PROPERTY OF THE NUMBER 977731833235239280

WITOLD JARNICKI AND MACIEJ ŻENCZYKOWSKI

arXiv:0709.3361v2 [math.NT] 10 Oct 2007

Abstract. We solve a theoretical arithmetics problem stated by Wacław Sier-

piński. The problem has remained open for a couple of decades.

1. Introduction
Consider the following puzzle from number theory, presented almost 50 years
ago.
Problem 1 (W. Sierpiński). Find a composite number such that it remains com-
posite after altering any two digits in its decimal representation.
We found the problem in [Mat 1977]. It was stated as a puzzle for the readers.
It turned out (see [Mat 1978]) that no one had solved it.
A ternary version of the problem was stated in [Mat 1978] with a solution given
in [Mat 1980]. The basic idea was that altering any two digits in 40 = 11113 keeps it
even and different from 2. Observe that such an assumption (changing exactly two
digits) makes the problem easy to solve (see Table 1). Therefore, it is reasonable
to assume the following statement.
Problem 2 (W. Sierpiński). Find a composite number such that it remains com-
posite after altering at most two digits in its base b representation.

allowed not allowed

base
solution decimal solution decimal
2 1010100 84 1001 9
3 1111 40 11 4
4 20130000 34560 12321230 28268
>4 4 4 4 4
Table 1. Minimal solutions to Problem 1, depending on whether
one is allowed to change the most significant digit to zero or not.

1991 Mathematics Subject Classification. 11Y11, 65Y05.

Key words and phrases. prime number, composite number, sieve of Atkin, numerical methods.
Part of the computation was done on High Performance Computers of The Academic Computer
Centre CYFRONET AGH: SGI Altix 3700 (baribal, grant number MNiSW/SGI3700/UJ/097/2007),
IBM BladeCenter HS21 (mars, grant number MNiSW/IBM BC HS21/UJ/097/2007), and PC Rack-
Saver cluster (zeus, grant number MNiSW/PC RS/UJ/097/2007).
The authors obtained a significant increase of computing speed by using their program to test
newly installed Google machines.
1
2 WITOLD JARNICKI AND MACIEJ ŻENCZYKOWSKI

Another question one may ask is whether it is permitted to change the most
significant digit to zero. However, disallowing that gives little help in finding the
number (see Tables 2 and 3), so it is safer to assume it may be done.
The problem was investigated from the theoretical side in [Sch 1992]. There
the author shows, that there are infinitely many solutions to Problem 2 (for any
b), provided that Erdős’s “favorite” conjecture on covering systems of congruences
(see [Erd 1952]) is true. However, since Erdős’s conjecture is open, so remained the
problem.

2. Main results
We present the solution to Problem 2 (bases 2–10) in Tables 2 and 3.

base solution decimal

2 1010100 84
3 2200100 1953
4 20130000 34560
5 3243003420 7000485
6 55111253530 354748446
7 5411665056000 77478704205
8 33254100107730 1878528135128
9 210324811482600 48398467146642
10 977731833235239280 977731833235239280
Table 2. Solutions to Problem 2 found for bases 2–10 (changing
the most significant digit to zero is allowed). They are known to
be minimal for bases 2–9.

base solution decimal

2 1010100 84
3 2200100 1953
4 12321230 28268
5 324322330 1401590
6 43040303150 273241578
7 5411665056000 77478704205
8 33254100107730 1878528135128
9 210324811482600 48398467146642
10 977731833235239280 977731833235239280
Table 3. Solutions to Problem 2 found for bases 2–10 (changing
the most significant digit to zero is disallowed). They are known
to be minimal for bases 2–9.
 ON A PROPERTY OF THE NUMBER 977731833235239280 3

3. Motivation and methods

Since there are many ways to modify a number by altering two of its digits, it
initially seems impossible to find a solution to the problem. Therefore, a natural
way to approach the puzzle is to consider it in bases smaller than 10 — in such
situation the number of ways of changing a number is much smaller.
We managed to solve the problem for bases between 2 and 9 using a small “grid”
of computers at our university. The computation lasted several weeks. Looking at
the results gives grounds to suppose that solving the problem for base k + 1 requires
about 100 times the computing power needed to solve it for base k. This estimate
inspired us to use grid computing to solve the decimal case.
To present the main idea behind the solution method, consider the following
lemma.
Lemma 3. Let n contain at least 5 nonzero digits in base b. Assume that n is a
solution to Problem 2. Define n
e := b⌊n/b⌋. Then n
e is a solution to Problem 2.
Proof. The number n e is composite, because b|e n and 1 < b < n e. For the same
reason it will remain composite whenever we decide to leave the least significant
digit intact. To complete the proof, observe that altering two digits of ne, one of
which is the least significant one, leads us to a number that is obtained by exactly
the same alteration of n. 
Observe that, for n having 1 nonzero digit, either its least significant digit is zero
(then n e = n and Lemma 3 holds trivially), or nonzero (then n can be transformed
into 1, which is not composite, and Lemma 3 holds trivially).
Additionally, for n having 4 nonzero digits, the only situation when Lemma 3
fails is when the second least significant digit of n is 1.
This means that, when looking for a solution to Problem 2, it is sufficient to
consider a relatively small number of cases where n has 2, 3, or 4 nonzero digits (in
the latter case with second least significant digit being 1) and the numbers divisible
by b. The former can be done directly. The latter is accomplished by the following
algorithm.
Algorithm 4. Given integers b ≥ 2, e ≥ 0, u ≥ 0, to find all solutions n ∈
bZ ∩ [ube+1 , (u + 1)be+1 ) to Problem 2:
1. For each integer k ∈ [0, be+1 ):
2. If ube+1 + k is prime: set ak ←− 1; otherwise set ak ←− 0.
3. For each integer k ∈ [0, be ):
4. If any of abk , abk+1 , . . . , abk+b−1 is 1: set bk ←− 1; otherwise set bk ←− 0.
5. For each integer k ∈ [0, be ):
6. Set ck ←− 1.
7. For each integer ℓ ∈ [0, be ) with k differing from ℓ at at most one digit:
8. If bℓ = 1: set ck ←− 0.
9. For each integer k ∈ [0, be ) with ck = 1:
10. Check directly if b(ube + k) is a solution and output if it is.
The steps 1–2 require finding all prime numbers in an interval. We have used
the following algorithms, all with running time close to O(be ):
(a) a straightforward implementation of the sieve of Eratosthenes,
(b) the implementation of the sieve of Atkin from [Ber 2007],
4 WITOLD JARNICKI AND MACIEJ ŻENCZYKOWSKI

(c) our own implementations of the sieve of Atkin with W = 12, W = 60, and
W = 420 (see [Atk-Ber 2004] for details).
The author of [Ber 2007] claims that the program works for primes up to 1015 . The
code is pretty complicated, so we could not figure out whether it works past that
boundary. That is why we decided to create our own implementation. Surprisingly,
the version W = 12 worked best for really large numbers.
The steps 7–8 can be done in O(eb) time. Consequently, we implemented the
whole loop 6–8 in O(ebe ) time.
In practice, the step 10 is involved only for a couple values of k, so its influence
on running time is negligible.

If we are interested in finding the smallest solution, it is enough to call Algorithm

4 for fixed b, e and sequential units u. However, due to limited computational
resources, it is more important to find any solution rather than to prove that it is
minimal. Therefore, we first scanned the units, for which the probability of finding
a solution is high.
The method we used is neither strict nor formal, but worked in practice. For
the sake of the estimation, we assumed that primality of numbers is a result of
a sequence of independent random experiments. The probablity of a number in
unit u being prime is (π((u + 1)be+1 ) − π(ube+1 ))b−(e+1) . Using the approximation
π(n) ≈ n/ ln(n) we obtained the approximate value of the probability p(u) of a
number from block u being a solution to Problem 2. Then, we considered the
blocks in order of decreasing p(u). The computation done is summed up in Table
4.
b e units scanned solution unit
2 6 0 0
3 6 0 0
4 7 0 0
5 9 0 0
6 10 0 0
7 10 0–39 39
8 10 0–218 218
9 10 0–1542 1542
10 9 0–97773183 97773183
Table 4. Summary of units considered

References
[Atk-Ber 2004] A.O.L. Atkin, D.J. Bernstein, Prime sieves using binary quadratic forms,
Math. Comp. 73 (2004), 1023–1030.
[Ber 2007] D.J. Bernstein, http://cr.yp.to/primegen.html (seen on June 19th, 2007).
[Erd 1952] P. Erdős, On a problem concerning congruence systems (Hungarian), Mat. Lapok
3 (1952), 122–128.
[Mat 1977] Problem corner (Polish), Matematyka 1(147) (1977), p. 55.
[Mat 1978] Problem corner (Polish), Matematyka 6(158) (1978), p. 362.
[Mat 1980] Problem corner (Polish), Matematyka 6(170) (1980), p. 369.
[Sch 1992] A. Schinzel, On a problem by Wacław Sierpiński (Polish), Gradient 1 (1992), No
2, 6–9.
 ON A PROPERTY OF THE NUMBER 977731833235239280 5

Institute of Mathematics, Jagiellonian University, Reymonta 4, 30–150 Kraków,

Poland
E-mail address: [email protected]
Current address: Google R&E, Kraków, Krupnicza 16, 31–123 Kraków
E-mail address: [email protected]

Theoretical Computer Science Department, Jagiellonian University, Gronostajowa

3, 30–387 Kraków, Poland
E-mail address: [email protected]
Current address: Google, 1600 Amphitheatre Parkway, Mountain View, CA 94043
E-mail address: [email protected]