GENERALIZED SIERPIŃSKI NUMBERS BASE b

AMY BRUNNER† , CHRIS K. CALDWELL, DANIEL KRYWARUCZENKO† ,

AND CHRIS LOWNSDALE†

Abstract. Sierpiński proved that there are infinitely many odd integers k
such that k · 2n +1 is composite for all n ≥ 0. These k are now called Sierpiński
numbers. We define a Sierpiński number base b to be an integer k > 1 for which
gcd(k+1, b−1) = 1, k is not a rational power of b, and k · bn +1 is composite
for all n > 0. We discuss ways that these can arise, offer conjectured least
Sierpiński number in each of the bases 2 < b ≤ 100 (34 are proven), and show
that all bases b admit Sierpiński numbers. We also show that under certain cir-
r
cumstances there are base b Sierpiński numbers k for which k, k2 , k3 , ..., k2 −1
are each base b Sierpiński numbers.

1. Introduction and History

In 1958, R. M. Robinson [26] formed a table of primes of the form k · 2n +1 for
odd integers 1 ≤ k < 100 and 0 ≤ n ≤ 512. He found primes for all k values except
47. Some then wondered “Is there an odd k value such that k · 2n +1 is always
composite?” In 1960, W. Sierpiński [29] proved that there were indeed infinitely
many such odd integers k. He did this by finding a small set of primes S such that
for a suitable choice of k, every term of the sequence k · 2n +1 (n > 0) is divisible
by a prime in his “cover” S. The values k which make every term in the sequence
composite are now called Sierpiński numbers. Sierpiński however neither gave a
value of k nor sought the least such k.
In 1962, Selfridge [unpublished] showed that k = 78557 is also a Sierpiński num-
ber, and this is now believed to be the least Sierpiński number. For three decades
mathematicians have been testing all of the values of k less than 78557 to prove this
conjecture [2, 9, 20, 22]. All values except 10223, 21181, 22699, 24737, 55459, and
67607 have now been eliminated by finding a prime in the corresponding sequence
[18].
There are two standard methods of generalizing Sierpiński numbers. Several
have generalized this idea by altering the restrictions on k [10, 11, 14, 21]. For
example, one may seek Sierpiński numbers k for which all of k, k 2 , k 3 , . . . , k r are
also Sierpiński numbers for arbitrarily large integers r [14]. We will provide a similar
(but weaker) result as Theorem 7.1.
On the Internet several groups have generalized Sierpiński’s result to other bases
b [4, 30, 31]. (See, for example, the results in Table 1.) There was also a short
note by Bowen in 1964 [5] which we will mention in the next section. But at the
time we began our investigation, none of these presented a systematic study of the
generalization or even a careful study of the definition. In this paper we will fill this

Key words and phrases. Sierpiński number, covering set, generalized Fermat number.
†Undergraduate student. The beginning of this work was partially supported by a University
of Tennessee at Martin College of Engineering and Natural Sciences undergraduate research grant.
1
2 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

Table 1. Conjectured least Sierpiński numbers k base b

b N k {cover} k’s not yet eliminated ref

2 36 78557 {3, 5, 7, 13, 19, 37, 73} {10223, 21181, 22699, 24737, [18]
55459, 67607}
3 144 125050976086 {5, 7, 13, 17, {2949008, 4273396, 4660218, [4, 6]
19, 37, 41, 193, 767} 6363484, 8058998, 8182316, . . .}
4 12 66741 {5, 7, 13, 17, 241} {18534, 20446, 21181, 22699, [18, 31]
23451, 49474, 55459, 60849, 64494}
5 12 159986 {3, 7, 13, 31, 601} {6436, 7528, 8644, 10918, 24032, [4]
26798, 29914, 31712, 36412, . . .}
6 12 174308 {7, 13, 31, 37, 97} {10107, 13215, 14505, 26375, 31340, [4]
33706, 36772, 50252, 51255, . . .}
7 24 1112646039348 {5, 13, {66936, 95626, 242334, 270636, [4]
19, 43, 73, 181, 193, 1201} 303366, 357132, 468552, . . .}

gap by providing a definition and then extending the studied bases systematically
to include all of the bases up through 100.
We will prove that Sierpiński numbers exist for all bases b > 1, and offer conjec-
tured least Sierpiński numbers for the bases 2 < b ≤ 100. For 34 of these bases we
are able to prove that the conjectured values are indeed the least.

2. Generalizing Sierpiński numbers to base b

A Sierpiński number is an odd integer k such that k · 2n +1 is composite for all
n > 0. Before generalizing this definition of a Sierpiński number to other bases b,
there are a couple of things we must consider.
First, when generalizing a definition it is traditional to exclude any cases that
are too trivial. So we begin by requiring that the sequence k · bn + 1 (n = 0, 1, 2, . . .)
does have not a “one-cover.” That is, there is no single prime p which divides every
value of the sequence. For example, if k and b are odd, then 2 divides every term
(and in fact if 2 divides any one term of any sequence it divides them all).
Theorem 2.1 (1-covers). The prime p divides k ·bn +1 for all non-negative integers
n if and only if p divides gcd(k+1, b−1).
Proof. First, suppose p divides k · bn +1 for all n, then it does so for n = 0 and 1,
that is p divides k + 1 and k · b+1. Subtracting these we see p divides k(b − 1) so p
divides gcd(k+1, b−1). If instead p divides gcd(k+1, b−1), then k · bn +1 ≡ k+1 ≡ 0
(mod p). 
In 1964, Bowen [5] showed there where choices of k for which k·bn +1 is composite
for all n ≥ 0, but he did so by using 1-covers for all bases except those which are a
power of 2 plus one.
Second, some have suggested that the restriction “k odd” appears in the above
definition because any factor of 2 in k can be absorbed into the exponent n, but
consider the number 2m 2n +1 for some fixed positive integers m and n. If this
n
number is to be prime, then it must be a Fermat number Fn = 22 + 1 and so n+m
must be a power of two. It is widely suspected that there are only finitely many
Fermat primes, which would mean there would be infinitely many even Sierpiński
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 3

Table 2. k · bn + 1 is infinitely often a generalized Fermat ‡

b k b k
6 6, 36, 216, 1296, 7776, 46656 46 46, 2116
8 2, 4, 8, 16, 32 48 48
10 10, 100, 1000 52 52, 2704
12 12, 144 58 58, 3364
16 16, 256, 4096, 65536 60 60, 3600
18 18, 324 64 4, 16
22 22, 484 66 66, 4356, 287496, 18974736
24 24, 576, 13824 70 70, 4900
26 26 72 72
28 28, 784 78 78, 6084
30 30 80 80
32 2, 4, 8 82 82, 6724
36 36, 1296 88 88
40 40, 1600, 64000 96 96, 9216
42 42, 1764 100 100
2 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048,
4096, 8192, 16384, 32768, 65536
4 4, 16, 64, 256, 1024, 4096, 16384, 65536
‡ Just those smaller than conjectured least base b Sierpiński and
with gcd(k + 1, b − 1) = 1.

numbers that are a power of 2. If the only Fermat numbers are the five known,
then 216 2n + 1 would be composite for n > 0, and therefore 216 = 65536 would be
the least Sierpiński number, not Selfridge’s 78557.
Since the existence of infinitely many Fermat primes is undecidable at this point
in time, it seems best to define generalized Sierpiński numbers in such a way as
to exclude the Fermat numbers and, for bases other than powers of 2, to exclude
n
the generalized Fermat numbers Fn (b) = b2 + 1 [12]. At the end of this section
(Theorem 2.3) we will show that this is equivalent to adding the requirement that
p
k is not a rational power of b (k 6= b q for integers p ≥ 0 and q > 0), and hence that
k > 1. Those values so omitted are listed in Table 2.
Combining these considerations we generalize Sierpiński numbers as follows.
Definition 2.2. Let b > 1 be an integer. A Sierpiński number base b (or b-
Sierpiński) is an integer k > 1 for which gcd(k+1, b−1) = 1, k is not a rational
power of b, and k · bn +1 is composite for all n > 0.
Notice that this definition extends the definition of Sierpiński numbers in base 2
as well as the larger integer bases—yet still 78557 likely remains the least possible
2-Sierpiński number.
We end this section by showing we have properly characterized those pairs k and
b which may generate infinitely many generalized Fermat numbers. Recall that the
order of b modulo a relatively prime integer p, denoted ordp (b), is the least positive
integer m for which p divides bm − 1. So in particular ordp (b) divides φ(p) (Euler’s
φ function of p).
4 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

Lemma 2.1. Let e > 1, f > 0 and c 6= 0 be integers. Write e = 2n e0 where e0 is

odd. Then gcd(cf − 1, ce + 1) > 1 if and only if c is odd or 2n+1 divides f .
Proof. Let d = gcd(cf − 1, ce + 1). First note that 2 divides d if and only if c is odd,
n
so assume c is even. Note that since e0 is odd, c2 + 1 divides ce + 1. If 2n+1 divides
n
f , then c2 + 1 divides d. Conversely, if any odd prime p divides d, then ordp (c)
divides both 2e and f , but not e. This means 2n+1 divides ordp (c) and therefore
divides f. 

Theorem 2.3. Let b > 1 and k > 0 be integers for which gcd(k + 1, b − 1) = 1.
There is an integer c > 1 for which k · bn + 1 = Fr (c) for infinitely many integer
values of r and n, if and only if k is a rational power of b.
Proof. Let b > 1 and k > 0 be fixed integers for which gcd(k + 1, b − 1) = 1.
Suppose there is an integer c for which k ·bn +1 (n > 1) is the generalized Fermat
number Fr (c) for infinitely many pairs of integers r and n. Choose two such pairs
(r, n) and (s, m) with n < m. Then
r s
k · bn + 1 = c2 + 1 and k · bm + 1 = c2 + 1.
s r 2s −2r m2r −n2s
Thus bm−n = c2 −2 , and it follows b = c m−n , k = c m−n , and therefore k is a
rational power of b (and both are rational powers of c).
Conversely, suppose k is a rational power of b, say k = be/f for relatively prime
integers e and f with e ≥ 0 and f > 0. Then because b is an integer, b = cf
and k = ce for some integer c. Write f = 2t f 0 where f 0 is an odd integer. Now
gcd(cf − 1, ce + 1) = 1, so by Lemma 2.1 c is even and the power of 2 which divides
e is at least as great as the power of 2 which divides f. So we may write e = 2t e0
for some integer (not necessarily odd) e0 . Note that if r is any positive multiple
of ordf 0 (2), then e0 ≡ e0 2r (mod f 0 ), so we may solve the following for a positive
integer n = n(r):
e0 + f 0 n = e0 2r .
So it follows
e + f n = 2t (e0 + f 0 n) = e0 2r+t ,
and there are infinitely many choices of r and n for which
0 r+t 0
k · bn + 1 = ce+f n + 1 = ce 2 + 1 = Fr+t (ce ).


3. N -covers: Covers and theorems

The use of covers was introduced by Paul Erdös in 1950 [13].
Definition 3.1. A cover for the sequence k · bn +1 (n > 0) is a finite set of primes
S = {p1 , p2 , . . . pm } for which each element of the sequence is divisible by a prime
in S. We ask that covers be minimal in the sense that no subset of S will also cover
the sequence. S is called an N -cover if N is the least positive integer for which
each prime p in S divides k · bn +1 if and only if p divides k · bn+N +1. We will call
this integer N the period of the cover S. Finally, we will say that the base b
has an N -cover if there is an integer k for which k ·bn +1 has a non-trivial N -cover
(N > 1).
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 5

Erdös apparently believed that all Sierpiński numbers arise from covers [16,
Section F13]. That is probably not the case [14]. In section 5 we will show that
not all b-Sierpiński numbers arise from covers. In practice, though, most small
examples do come from covers.
There are two basic ways of constructing covers: Sierpiński’s approach of using
the Fermat numbers (generalized Fermat numbers in our case) and Selfridge’s use
of factors of bn − 1. We begin with the latter.
Theorem 3.2. Every element of an N -cover S of k · bn +1 divides bN −1.
Proof. Choose p ∈ S. This p must divide k · bn +1 for some n ≤ N . It then also
divides k · bn+N +1, so divides their difference k · bn (bN −1). Since p does not divide
k · bn , this completes the proof. 

For example, Selfridge’s Sierpiński 78557 arises from the cover {3, 5, 7, 13, 19, 37, 73}.
Each prime of this cover divides 236 −1, so to prove 78557 is a Sierpiński number base
2, it is sufficient to show that each of the first 36 terms in the sequence 78557 · 2n +1
(n > 0) are divisible by one of these seven primes.
The previous theorem also tells us that for every element p of an N -cover of
k · bn +1, ordp (b) divides N . It is easy to show N = lcmp∈S (ordp (b)).
Unless we say otherwise, in the rest of this article “N -cover” will mean non-
trivial N cover, that is N > 1. Note that if S is an N -cover for one k, b pair, then
it is also a cover for infinitely many other multipliers k and bases b.
Theorem 3.3. An N -cover S of k · bn + 1 is also a cover of K · B n + 1 for all
integers K ≡ k, B ≡ b (mod P ) where P is the product of the primes in the cover
S.
It follows by Dirichlet’s theorem that there are infinitely many prime multipliers,
and infinitely many prime bases covered by any given N -cover.
In what follows it will be helpful to recall the cyclotomic polynomials Φn (x).
These are defined by
Y Y
Φn (x) = (xd − 1)µ(n/d) and so xn − 1 = Φd (x).
d|n d|n

n
This makes Φn (x) the “primitive part” of x − 1 when factoring, and the φ(n) zeros
of Φn (x) are the primitive nth roots of unity. For integers n and b greater than one,
if a prime p divides Φn (b) but not n, then p ≡ 1 (mod n).
Theorem 3.2 can now be greatly sharpened for more specific values of N .
Theorem 3.4. Let p be a prime number. The sequence base b has a p-cover S if
and only if Φp (b) has at least p distinct prime divisors greater than p.
Proof. Suppose first S is a p-cover of k · bn +1. So there is an element of S which
divides each element of T = {k · b1 +1, k · b2 +1, . . . , k · bp +1}. If q is an element of
S, the ordq (b) must divide the period of S, which is p. This order can not be one,
or {q} would be a trivial cover, so ordq (b) = p and therefore p | q − 1. This means
q can only divide one element of T , hence there are at least p primes in S. Finally,
these primes do not divide b − 1, so they each divide (bp − 1)/(b − 1) = Φp (b).
On the other hand, if Φp (b) has the p distinct prime divisors: q1 , q2 , q3 , . . . , qp ,
each greater than p, then none divide b − 1 so we can use the Chinese Remainder
6 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

Theorem to show they form a p-cover by solving the system of linear equations

k · b1 + 1 ≡ 0 (mod q1 )
2
k·b +1 ≡ 0 (mod q2 )
..
.
k · bp + 1 ≡ 0 (mod qp )

for the multiplier k. 

For example, the base b has a 2-cover if and only if b + 1 has at least two distinct
odd prime divisors. Examples of this include b = 14, 20, 29, 32, 34, 38, 41, 44, 50,
54, 56, 59, 62, 64, 65, 68, 69, 74, 76, 77, 83, 84, 86, 89, 90, 92, 94, 98 . . . For all of
these listed bases except 68 and 86, we have proven the 2-cover generates the least
generalized Sierpiński number base b (see Table 5). Bowen [5] also used 2-covers to
address the bases b = 2s + 1 where s 6= 2m + 1 and s > 5.
Similarly, the base b has a 3-cover if and only if Φ3 (b) = b2 + b + 1 has at least
three distinct divisors greater than 3. The first such bases are b = 74, 81, 87, 100,
102, 107, 121, and 135. Of those bases b ≤ 100, only for 100 does the 3-cover yield
the least generalized Sierpiński number. Bases 74, 81, and 87 have 3-covers, but
these produce larger multipliers k than can be generated by other methods.
The minimal base for longer prime period covers grows quickly: (p, minimal base
b) = (2, 14), (3, 74), (5, 339), (7, 2601), (11, 32400), and (13, 212574).
The structure of composite period covers are more interesting. For example,
4-covers usually arise from an odd prime factor p for which the base b has order 2
(a divisor of Φ2 (b) = b + 1), and two primes q1 , q2 for which b has order 4 (divisors
of Φ4 (b) = b2 + 1). Then the terms of the sequence k · bn + 1 (n = 1, 2, . . .) are
divisible by the primes of the 4-cover in a pattern like

p, q , p, q2 , p, q1 , p, q2 , . . . .
| 1{z } | {z }

So one choice of k could be found by solving the following system.

k · b1 + 1 ≡ 0 (mod p)
k · b2 + 1 ≡ 0 (mod q1 )
k · b4 + 1 ≡ 0 (mod q2 )

For 29 of the bases in Table 5, 4-covers provide the least known b-Sierpiński num-
bers.
Most 6-covers involve four primes. Often one prime in the cover, say p, has
period 2 (ordp (b) = 2), and there are three more of orders 3 or 6, say q1 , q2 , q3 ,
dividing the terms of k · bn + 1 in a pattern similar to

p, q , p, q , p, q3 , p, q1 , p, q2 , p, q3 , . . . .
| 1 {z 2 } | {z }

Most bases have a 12-cover. One way one of these can arise is if each of Φ2 (b),
Φ3 (b), Φ4 (b), Φ6 (b) and Φ12 (b) have a primitive divisor, call them p2 , p3 , p4 , p6 and
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 7

p12 respectively. Then by solving the following system for k

k · b1 + 1 ≡ 0 (mod p2 )
2
k·b +1 ≡ 0 (mod p3 )
4
k·b +1 ≡ 0 (mod p4 )
k · b6 + 1 ≡ 0 (mod p6 )
10
k·b +1 ≡ 0 (mod p12 )

we have the divisibility pattern

p , p , p , p , p , p , p , p , p , p , p , p , ...
| 2 3 2 4 2 6 {z 2 3 2 12 2 4}
Many other such patterns are possible with these five primes, but this one is suffi-
cient to prove the following.
Theorem 3.5. Every base b > 2 which is not a Mersenne number has a 12-cover.
The proof follows immediately from the congruences above and Bang’s result [3]
that bN − 1 has a primitive divisor except when N = 2 and b is a Mersenne number
(2n − 1, n a positive integer); or N = 6 and b = 2.
We can also use Bang’s theorem on the Mersenne numbers by using a 144-cover:
choose a primitive divisor p of Φn (b) and then solve the system of congruences
k · be + 1 ≡ 0 (mod p) for each of the pairs (n, e) = (3, 1), (4, 2), (6, 3), (8, 5), (9, 8),
(12, 12), (16, 20), (18, 11), (24, 32), (36, 23), (48, 92), and (72, 41). Similar systems
are easily found for bases such as 120 and 180, but these require more primes.
With Dirichlet’s Theorem we now have the following.
Theorem 3.6. There are infinitely many prime generalized Sierpiński numbers for
every base b.
Finally, when searching for possible covers the following results can be very
useful.
Theorem 3.7. If S is an N -cover of k · bn +1, then p∈S ord1p (b) ≥ 1
P

Theorem 3.8. If there is a non-trivial cover for k · bn +1, then k+1 has an odd
prime divisor.
The first of these was used by Stanton [32] in his analysis of possible covers for the
b = 2 case.

4. A Simple Program and Known Results

Theorem 3.2 can be turned into a surprisingly effective program to find N -covers
by looping on k until one is found for which gcd(k · bn +1, bN −1) > 1 for each of
k = 1, 2, . . . N . If this is done with a fairly large round value of N, such as 5040,
then most small covers with relatively small k (say less than 108 ) will be easily
spotted.
Daniel Adler, at the time a student at University of Tennessee at Martin, was
enlisted to write a program Sierpiński in C++ using the multiprecision package
GMP1. When the program finds an N -cover, it outputs k and a vector of length N
1http://gmplib.org/
8 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

where the ith component is gcd(k · bi +1, bN −1) (1 ≤ i ≤ N ). From this it was a
simple hand calculation to find the actual covering set of primes.
This program was run on the 16 nodes of our Beowulf cluster for about 80-CPU
days to find the constants k and the associated covers in the first columns of Table 5
except for b = 3, 7 and 15. Some individual values (e.g., 71), required substantially
longer search times.
The program Sierpiński has several limitations. First, one must know some-
thing of N in advance because the program is set up to seek all covers with period
N dividing a specified constant. For Table 5 we usually sought periods dividing
7! = 5040. It is possible that we missed some covers for smaller k values.
Second, the program Sierpiński only seeks values of k belonging to covers.
Such k values are Sierpiński numbers base b, but there may be smaller b-Sierpiński
numbers that do not arise from covers. We will discuss this in the next section.
Finally, the program Sierpiński is too slow to find the least covers for bases
like 3. For those bases we may begin by factoring bN −1 for various small values of
N and construct covers as described in the previous section. For example, Brennen
[7] used this method to find 3574321403229074 (48-cover) for b = 3. (This improved
earlier results of Bowen [5] and Saouter [28].) With an improved algorithm Bosma
[6] reduced this to k = 125050976086 (144-cover).

5. Polynomial Factorization and Partial Factorization

Another way that generalized Sierpiński numbers can arise is through factoriza-
tion as polynomials. For example, when b = 27 and k = 8, each term factors as a
difference of cubes:
8 · 27n +1 = (2 · 3n +1)(4 · 32n −2 · 3n +1).
Similarly 8 is b3 -Sierpiński number for all positive multiples of 3. Such b-Sierpiński
numbers arise whenever b is a perfect cube.
Consider also the factorization 4x4 + 1 = (2x2 + 2x + 1)(2x2 − 2x + 1). Anytime
b is fourth power and the multiplier k is 4 times a fourth power, every term of
the sequence k · bn + 1 will all factor in this manner. Small examples for which
this factorization generates the least known generalized Sierpiński numbers base b
include (k, b) = (2500, 16), and (2500, 81).
The cases where the least Sierpiński numbers arise by polynomial factorization
are marked by ] in Table 5.
A final possibility is a “partial factorization,” where part of the sequence is
covered by a set of primes, and the remainder of the terms factor as above. For
example, the least known b-Sierpiński number for base b = 63, k = 3511808, comes
from the partial 3-cover {37, 109} (which divide 3511808 · 63n + 1 when n ≡ 1, 2
(mod 3)) and the factorable x3 + 1 (for n ≡ 0 (mod 3)). This was discussed for the
usual base 2 Sierpiński numbers by Izotov [19] (see also [14]).
Another example is k · 2070n + 1 whose least base b Sierpiński appears to be 324.
Here 324 · 2070n + 1 factors as 4x4 + 1 when n ≡ 0 (mod 4), and then values of
n 6≡ 0 (mod 4) are covered by {17, 19}. To prove 324 is the least Sierpiński base
2070, we must find a prime for each of the following values of k : 77, 96, 132, 153,
and 305. All others are known to generate primes.
The generalized Fermat numbers base b allow neither factorizations nor finite
covers, yet it seems very likely that there are bases b such that all Fn (b) (n ≥ 0)
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 9

are composite. These have been excluded by our definition, but we see no reason
that there could not be other examples of b-Sierpiński numbers that have neither
covers nor factorizations. (This same possibility is addressed for base 2 in [19, 14]).

6. Checking the Results

The conjectured minimal values in Table 5 were compared against the published
results [4, 31] and against the results of Robert Gerbicz’s program which finds
covers very quickly2.
To prove the multipliers k conjectured in Table 5 are the least b-Sierpiński num-
bers is “simple:” just find a prime of the form K · bn + 1 (n > 0) for each potential
K < k. Though conceptually trivial, the amount of effort this can take may be
truly massive! This is shown by the original case b = 2, still unsettled after 45
years, and is still one of the larger distributive computing projects: Seventeen or
Bust [18]. The largest prime that they have had to find so far to eliminate a k
value was 19249 · 213018586 + 1 with 3, 918, 990 digits. They estimate they may need
to search to an exponent of n = 3, 400, 000, 000, 000 just to get a 50% chance of
finishing of the remaining cases [17].
To eliminate these small K values, we began with a Maple program. We then
used OpenPFGW [25] for anything larger than a dozen digits. This was done in two
passes: the first to trial factor by small primes and perform a probable primality
test (this took about five CPU years). Second we reran OpenPFGW on our list of
probable primes to provide classical n ± 1 primality proofs [8].
We compared against all published sources that we could find, especially [4, 31].
For many of the smaller bases, Barnes [4] has results from more extensive searches
than ours—so we include those results in Table 5 also. When comparing tables
it is necessary to be sensitive of the variety of different definitions of generalized
Sierpiński numbers being used.

7. Can k, k 2 , k 3 , . . . all be b-Sierpiński numbers?

Sierpiński’s original construction [29] was based on the factorization of Fermat
n n+1
numbers. Because 22 ≡ −1 (mod Fn ), we know ordp (2) = 22 for any divisor
d > 1 of Fn (this also means the Fermat numbers are pairwise relatively prime). So
taking advantage of the fact that F5 = 641 · 6700417 = p · q, we have the following
implications.
n ≡ 20 (mod 21 ), k ≡ 1 (mod F0 ) =⇒ k · 2n + 1 ≡ 0 (mod F0 )
n ≡ 21 (mod 22 ), k ≡ 1 (mod F1 ) =⇒ k · 2n + 1 ≡ 0 (mod F1 )
n ≡ 22 (mod 23 ), k ≡ 1 (mod F2 ) =⇒ k · 2n + 1 ≡ 0 (mod F2 )
n ≡ 23 (mod 24 ), k ≡ 1 (mod F3 ) =⇒ k · 2n + 1 ≡ 0 (mod F3 )
n ≡ 24 (mod 25 ), k ≡ 1 (mod F4 ) =⇒ k · 2n + 1 ≡ 0 (mod F4 )
n ≡ 25 (mod 26 ), k ≡ 1 (mod p) =⇒ k · 2n + 1 ≡ 0 (mod p)
n≡0 (mod 26 ), k ≡ −1 (mod q) =⇒ k · 2n + 1 ≡ 0 (mod q)
Using the Chinese Remainder Theorem to solve for k yields a cover of k · 2n + 1
{F0 , F1 , F2 , F3 , F4 , p, q} with period 64. In this construction it does not matter that
the first five terms were prime, it would still work if they were composite. Rather
than stop with F5 (as Sierpiński did), we could then stop with any Fn for which at
least one proper factor p is known and let q = Fpn .

2http://robert.gerbicz.googlepages.com/coveringsets
10 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

This cover has another interesting property: not only are the k so constructed
Sierpiński numbers, but so are k t for any odd integer t > 1. It turns out that by using
multiple different composite terms we may use essentially the same construction to
find k for which k t is also prime for t divisible by low powers of 2. This was done
by Filaseta et al. [14] for the regular Sierpiński numbers, and virtually the same
argument works here.
Theorem 7.1. Let b > 1 be an integer for which b + 1 is not a power of 2. If there
m
are at least r generalized Fermat numbers Fm (b) = b2 + 1 which are each divisible
by at least two distinct odd primes, then there are infinitely many integers k such
that k t is a b-Sierpiński number for all positive integers t not divisible by 2r .
0 0 0
Proof. Define the integer Fm (b) by Fm (b) = 2rm Fm (b) with rm ≥ 0 and Fm (b) odd.
0
In what follows we need the fact that Fm (b) has at least one prime factor. This
0 0
is the case unless Fm (b) = 1. If Fm (b) = 1, then b is odd. It follows that m = 0;
2m 0
otherwise b + 1 ≡ 2 (mod 8). So Fm (b) = 1 implies b + 1 = 2r0 . For the rest of
0
the proof we assume b + 1 is not a power of 2 (hence Fm (b) > 1).
Let m0 < m1 < · · · < mr−1 be non-negative integers for which the generalized
0
Fermat numbers Fmj (b), hence Fm j
(b), each have at least two distinct odd prime
0
factors, say pj and qj . Note these primes are all different as b, b − 1, and Fm (b)
(m ≥ 0) are pairwise relatively prime.
By the Chinese Remainder Theorem there are infinitely many solutions to the
following set of congruences.


 0 (mod b − 1)
 1 (mod b)


0
k≡ 1 (mod Fm (b)) for 0 ≤ m < mr−1 and m 6∈ {m0 , . . . , mr−1 }


 1 (mod p j ) for 0≤j ≤r−1
 2mj −j

b (mod qj ) for 0 ≤ j ≤ r − 1.
We further restrict k to those solutions which are greater than each of the moduli
above. The first of these modular restrictions guarantees gcd(k + 1, b − 1) = 1, and
the second guarantees that gcd(k, b) = 1, so k is not a rational power of b.
Given any positive integer t not divisible by 2r , say t = 2w t0 where t0 is odd and
0 ≤ w < r, we must show k t bn + 1 is composite for each positive integer n. Fix a
positive integer n and let n = 2i n0 where n0 is odd. We may complete this proof
by showing k t bn + 1 is divisible by
 0
 Fm (b) if i < mw and m 6∈ {m0 , . . . , mw }
d= pj if i = mj for some j with 0 ≤ j ≤ w
qw if i > mw .


Since d < k < k t bn + 1, this will show the latter term is composite.
i i 0
If i ≤ mw , then d divides b2 + 1 which divides b2 n + 1 = bn + 1. Because k ≡ 1
(mod d), it follows d divides k t bn + 1.
If instead i > mw , then
mw −w w 0 mw 0 0
k t ≡ (b2 )2 t
≡ (b2 )t ≡ (−1)t ≡ −1 (mod qw ).
2mw
Now d = qw divides b + 1 which divides
i i−1
b2 − 1 = (b − 1)(b + 1)(b2 + 1)(b4 + 1) · · · (b2 + 1).
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 11

Table 3. k such that k t is a 5-Sierpiński when 2r - t

k r
23140626796 1
3352282631064632411056 2
38454071854799507248067375352496 3
295612797233398523232282186442005794587542575896 4
1202250010386171287615458085\ 5
386724017477152933279927552922222324231610279296
4833\ 6
96281140918612511630787705875212985273405983905\
512852696056665671273849671134513427529509057456
18081740848967\ 7
53044039134711401288516658002520824319923798573\
210660688220428187289811356995735827761349820556

i
n0
Thus d divides b2 − 1 and it follows
k t · bn + 1 ≡ −(bn − 1) ≡ 0 (mod d).
This completes the proof of the theorem. 
0
For example, when b = 5, Fm (b) is prime for m = 0, 1, and 2. It is composite
(with distinct prime divisors) for m0 = 3, m1 = 4, . . . , m10 = 13. If we let qj be
0
the smallest prime factor of Fm j
(b) and pj be the second smallest, then we get
the b-Sierpiński numbers in Table 3. Of course, as noted in the discussion before
the proof, rather than use prime factors, we may use any two relatively prime
(non-trivial) proper divisors. So it is sufficient to know any odd prime divisor and,
after checking that the given generalized Fermat is not a power of that prime,
use the cofactor as the second “prime.” Table 4 shows that there are 244 known
composite generalized Fermat numbers Fn (5), so there are 5-Sierpiński k for which
244
k, k 2 , k 3 , ..., k 2 −1 are all Sierpiński numbers (from [23]).

Table 4. Number of generalized Fermat numbers known to be composite

form number form number

2m 2m
2 +1 235 6 +1 220
m m
(32 + 1)/2 256 102 + 1 230
m m
(52 + 1)/2 244 122 + 1 223

8. Conclusions
Of the many possible generalizations of the Sierpiński numbers, we have discussed
what seemed the most natural to us. It would be interesting, but difficult, to study
the generalized Fermat cases that we excluded in our definition. It seems likely that
bases b can be found so that the least Sierpiński number is arbitrarily large. One
can also ask the reverse question: given a value k, can we find a base b for which
12 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

k is a base b Sierpiński? A partial answer has been provided by one of the authors
[24].
Note that every cover of a sequence of the form k · bn + 1 (n > 0) is also a cover
of a sequence k 0 · bn − 1 (n > 0), and vice versa. Positive odd integers k for which
k · bn − 1 are composite for all n > 0 are called Riesel numbers after an article by
Riesel [27] in 1956 (so the Riesel numbers predate the Sierpiński numbers). Thus
another generalization to study would be the generalized Riesel numbers (k which
make k · bn − 1 composite for all n > 0 with suitable restrictions on k and b); as
well as the numbers that are both b-Riesels and b-Sierpińskis. Part of this work is
being done informally by Barnes and others [4], as are restrictive cases like seeking
the smallest b-Sierpiński numbers which are prime.
A. de Polignac conjectured (and quickly retracted) the guess that every positive
odd number can be written in the form 2n + p for a prime p and integer n > 0.
He did this even though Euler had previously shown that this was not the case for
127 or 929 [1]. Again every cover of k · bn + 1 (n > 0) is also a cover of a sequence
bn + k (n > 0), and vice versa.

References
1. L. Babai, C. Pomerance & P. Vértesi, The mathematics of Paul Erdös, Notices of the AMS,
45:1 (January 1998), 19–31.
2. R. Baillie, G. V. Cormack & H. C. Williams, The problem of Sierpiński concerning k·2n +1
Math. Comput., 37 (1981) 229–231; corrigendum, 39 (1982) 308.
3. A. S. Bang, Taltheoretiske Undersøgelser, Tidsskrift for Mat., 5(4) (1886), 70–80, 130–137.
4. G. Barnes, Sierpiński conjecture reservations, May 2008, http://gbarnes017.googlepages.
com/Sierp-conjecture-reserves.htm.
5. R. Bowen, The sequence kan + 1 composite for all n, Math. Monthly, 71:2 (1964), 175–176.
6. W. Bosma, Some computational experiments in number theory. In Discovering Mathematics
with Magma, volume 19 of Algorithms Comput. Math., pp. 1–30. Springer-Verlag, Berlin,
2006.
7. J. Brennen, PrimeForm e-mail discussion list, May 16, 2002, http://tech.groups.yahoo.com/
group/primenumbers/message/7147.
8. J. Brillhart, D.H. Lehmer & J.L. Selfridge, New primality criteria and factorizations of 2m ±1,
Math. Comp., 29 (1975) 620–647.
9. D. A. Buell & J. Young, Some large primes and the Sierpiński problem, SRC Technical Report
88-004, Supercomputing Research Center, Lanham, MD, May 1988.
10. Y. G. Chen, On integers of the forms kr −2n and kr 2n +1, J. Number Theory, 98:2 (2003),
310–319.
11. Y. G. Chen, On integers of the forms k ± 2n and k2n ± 1, J. Number Theory, 125:1 (2007),
14–25, MR 2333115.
12. H. Dubner & W. Keller, Factors of generalized Fermat numbers, Math. Comput., 64 (1995)
397-405, MR 1270618.
13. P. Erdös, On integers of the form 2k + p and some related problems, Summa Brasil. Math.,
2 (1950) 113-123, MR 0044558.
14. M. Filaseta, C. Finch & M. Kozek, On powers associated with Sierpiński numbers, Riesel
numbers and Polignac’s conjecture, J. Number Theory, 128:7 (2008)1916–1940, MR 2423742.
15. Y. Gallot, Proth.exe, primes.utm.edu/programs/gallot/, July 2005.
16. R. K. Guy, Unsolved Problems in Number Theory (3rd ed.), Problem Books in Mathematics,
Springer-Verlag, New York, 2004.
17. L. Helm, P. Moore, P. Samidoost & G. Woltman, Resolution of the mixed Sierpiński problem,
INTEGERS: Elec. J. Comb. Num. Th., to appear.
18. L. Helm & D. Norris, Seventeen or Bust–a distributed attack on the Sierpiński problem,
http://www.seventeenorbust.com/.
19. A. S. Izotov, A note on Sierpiński numbers, Fibonacci Quart., 33 (1995) 206–207; MR
96f:11020.
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 13

20. G. Jaeschke, On the smallest k such that all k · 2N + 1 are composite, Math. Comput., 40
(1983) 381–384; MR 84k:10006; corrigendum, 45 (1985) 637; MR 87b:11009.
21. L. Jones, Variations on a theme of Sierpiński, J. Integer Seq., 10 (2007).
22. W. Keller, Factors of Fermat numbers and large primes of the form k·2n +1, Math. Comput.,
41 (1983) 661–673; MR 85b:11119; II (incomplete draft, 92-02-19).
23. W. Keller, Factors of generalized Fermat numbers found after Björn & Riesel, http://www1.
uni-hamburg.de/RRZ/W.Keller/GFNfacs.html, Oct. 2008.
24. D. Krywaruczenko, A reverse Sierpiński number problem, Rose-Hulman Undergrad. Math. J.
(electronic) to appear.
25. C. Nash & J. Fougeron, OpenPFGW (Open source software), http://tech.groups.yahoo.
com/group/primeform/.
26. R. M. Robinson, A report on primes of the form k·2n +1 and on factors of Fermat numbers,
Proc. Amer. Math. Soc., 9 (1958) 673–681; MR 20 #3097.
27. H. Riesel, Några stora primtal (Swedish: Some large primes), Elementa, 39 (1956) 258–260.
28. Y. Saouter, A Fermat-like sequence and primes of the form 2h · 3n + 1, Research Report 2728,
Nov. 1995, citeseer.ist.psu.edu/saouter95fermatlike.html.
29. W. Sierpiński, Sur un problème concernant les nombres k·2n +1, Elem. Math., 15 (1960)
73–74; MR 22 #7983; corrigendum, 17 (1962) 85.
30. N. Sloane, The on-line encyclopedia of integer sequences, Conjectured smallest Sierpiński
numbers, www.research.att.com/~njas/sequences/A123159.
31. R. Smith, Sierpiński and Riesel bases 6 to 18, Conjectured smallest Sierpiński numbers, www.
mersenneforum.org/showthread.php?t=6895, August 2007.
32. R. G. Stanton, Further results on covering integers of the form 1 + k · 2N by primes, Com-
binatorial mathematics, VIII (Geelong, 1980), Springer Lecture Notes in Math., 884 (1981)
107–114; MR 84j:10009.
14 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

Table 5: Conjectured Least Sierpiński Numbers k base b

b N k {cover} k’s not yet eliminated ref

8 4 47 {3, 5, 13} proven
9 6 2344 {5, 7, 13, 73} {2036}
10 6 9175 {7, 11, 13, 73} {7666} [4]
11 6 1490 {3, 7, 19, 37} proven [4]
12 4 521 {5, 13, 29} {404}
13 4 132 {5, 7, 17} proven
14 2 4 {3, 5} proven
15 24 91218919470156 {13, 17, {114258, 148458, 215432, 405556,
113, 211, 241, 1489, 424074, . . .}
3877}
16 ] 2500 proven
17 4 278 {3, 5, 29} {244} [4]
18 4 398 {5, 13, 19} {122}
19 12 765174 {5, 7, 13, 127, 769} {634, 1446, 2526, 2716, 3714, 4506, . . .}
20 2 8 {3, 7} proven
21 4 1002 {11, 13, 17} proven
22 4 6694 {5, 23, 97} {1611, 1908, 4233, 5128} [4]
23 4 182 {3, 5, 53} {8, 68}
24 12 30651 {5, 7, 13, 73, 79} {319, 621, 656, 821, 1099, 1851, 1864, 2164,
2351, 2586, 3031, 3051, 3404, 3526, . . .}
25 6 262638 {7, 13, 31, 601} {222, 5550, 6082, 6436, 7528, 8644, 10218,
10918, 12864, 12988, 13026, 13548, . . .}
26 6 221 {3, 7, 19, 37} {32, 65, 155}
27 ] 8 proven
28 4 4554 {5, 29, 157} {871, 2377, 3394, 4233, 4552}
29 2 4 {3, 5} proven
30 6 867 {7, 13, 19, 31} {278, 588}
31 12 6360528 {7, 13, 19, 37, {10366, 13240, 69120, 70612, 76848,
331} 99450, 101980, 122806, 124812, . . .}
32 2 10 {3, 11} proven
33 4 1854 {5, 17, 109} {766, 1678, 1818}
34 2 6 {5, 7} proven
35 6 214018 {3, 13, 97, 397} {46, 1610, 2006, 2272, 2588, 3046, 3700,
3812, 5518, 8632, 8800, 9542, 10222, . . .}
36 6 1886 {13, 31, 37, 43} proven
37 4 2604 {5, 19, 137} {94, 1272, 1866, 2224}
38 2 14 {3, 13} proven
39 6 166134 {5, 7, 223, 1483} {2264, 2414, 2434, 3254, 3986, 4226, . . .}
40 6 826477 {7, 41, 223, 547} {4468, 7092, 9964, 11112, 18285, . . .}
41 2 8 {3, 7} proven
42 4 13372 {5, 43, 353} {116, 988, 1117, 1421, 2794, 2903,
3046, 3226, 3897, 4127, 4297, 4643, . . .}
43 4 2256 {5, 11, 37} {166, 648}
‡ partial factorization, ] factorization
 GENERALIZED SIERPIŃSKI NUMBERS BASE b 15

Table 5: Conjectured Least Sierpiński Numbers k base b – Continued

b N k {cover} k’s not yet eliminated ref
44 2 4 {3, 5} proven
45 6 53474 {7, 19, 23, 109} {474, 1908, 2444, 3106, 4530, 4990,
6510, 6586, 6624, 7108, 8026, 9774, . . .}
46 6 14992 {7, 19, 47, 103} {892, 976, 1132, 1798, 3261, 3477,
3961, 4842, 5395, 6015, 6391, 6816, . . .}
47 4 8 {3, 5, 13} proven
48 6 1219 {7, 13, 61, 181} {29, 36, 62, 153, 422, 1174}
49 12 2944 {5, 19, 73, 181, 193} {1134, 1414, 1456, 2694, 2746}
50 2 16 {3, 17} proven
51 6 5183582 {7, 13, 379, 2551} {5498, 6280, 6696, 7682, 8126, 8412, . . .}
52 4 28674 {5, 53, 541} {1483, 1591, 2386, 3181, 3232, 3418, 5619,
5776, 5988, 6147, 6891, 7147, 8638, . . .}
53 4 1966 {3, 5, 281} {1816, 1838, 1862, 1892}
54 2 21 {5, 11} proven
55 4‡ 2500 {7, 17} {1980, 2274}
56 2 20 {3, 19} proven
57 4 1188 {5, 13, 29} {378}
58 4 43071 {5, 59, 673} {222, 787, 886, 1102, 1923, 2182, 2656, 2713,
3246, 3511, 3541, 4021, 5274, 6046, . . .}
59 2 4 {3, 5} proven
60 4 16957 {13, 61, 277} {853, 1646, 2075, 2497, 4025, 4406, 4441,
5064, 5767, 6772, 7262, 7931, 8923, . . .}
61 6 15168 {7, 13, 31, 97} {1570, 1642, 3390, 3442, 3936, 6852, 7348,
8710, 8772, 8902, 9208, 9268, 9952, . . .}
62 2 8 {3, 7} proven
63 3‡ 3511808 {37, 109} {3092, 3230, 4106, 7622, . . .}
64 2 51 {5, 13} proven
65 2 10 {3, 11} proven
66 24 21314443 {7, 17, 37, 67, {470, 2076, 4153, 5442, 6835, 13201,
73, 4357} 17035, . . .}
67 4 18342 {5, 17, 449} {154, 460, 1494, 2196, 2362, 2806, 2872,
2874, 3384, 4062, 4618, 4996, 5668, . . .}
68 2 22 {3, 23} {12, 17}
69 2 6 {5, 7} proven
70 4 11077 {13, 29, 71} {3762, 4119, 5608, 9231, 10438}
71 18 5917678826 {3, 19, 37, 73, {172, 502, 508, 1942, 2782, 3776, 4490, 5002,
1657, 5113} 5078, 5266, 5330, 5632, 5950, 6338, . . .}
72 4 731 {5, 61, 73} {493, 647}
73 4 1444 {5, 13, 37} {778, 1344}
74 2 4 {3, 5} proven
75 6 4086 {7, 13, 19, 61} {2336, 2564, 3782}
76 2 43 {7, 11} proven
77 2 14 {3, 13} proven
78 4 186123 {5, 79, 1217} {2371, 4820, 4897, 5294, 5531, 6353, . . .}
‡ partial factorization, ] factorization
16 BRUNNER, CALDWELL, KRYWARUCZENKO AND LOWNSDALE

Table 5: Conjectured Least Sierpiński Numbers k base b – Continued

b N k {cover} k’s not yet eliminated ref
79 6 2212516 {5, 7, 43, 6163} {24, 594, 724, 1086, 1654, 1774, 1896, . . .}
80 12 1039 {3, 7, 13, 43, 173} {86, 92, 166, 188, 295, 326, 370, 433, 472, 556
623, 628, 692, 770, 778, 787, 818, 857, 968}
81 ] 2500 {558, 1650, 2036, 2182, 2350, 2378}
82 12 19587 {5, 7, 13, 37, 83} {1251, 1327, 1570, 1716, 1798, 1908, 2251,
2352, 2461, 2491, 2731, 2989, 3342, . . .}
83 2 8 {3, 7} proven
84 2 16 {5, 17} proven
85 6 346334170 {37, 43, 193, {7612, 11740, 27168, 31776, 32550, 34014,
2437} 35088, 36508, 43474, 48204, 50352, . . .}
86 2 28 {3, 29} {8}
87 6 274 {7, 11, 19, 31} {32}
88 12 4093 {5, 7, 31, 37, 89} {192, 244, 958, 978, 1452, 1585, 1678, 1779,
2007, 2617, 2838, 3396, . . .}
89 2 4 {3, 5} proven
90 2 27 {7, 13} proven
91 4 89586 {23, 41, 101} {252, 1678, 2008, 6970, 8902, 11706, 12306,
14236, 22932, 23520, 26472, 29488, . . .}
92 2 32 {3, 31} proven
93 4 24394 {5, 47, 173} {62, 306, 706, 866, 894, 902, 1652, 2208,
2678, 3218, 3244, 3384, 3750, 3996, . . .}
94 2 39 {5, 19} proven
95 6 41354 {3, 7, 13, 229} {244, 376, 692, 790, 848, 908, 926, 1004,
1012, 1024, 1096, 1312, 1396, 1662, . . .}
96 4 353081 {13, 97, 709} {1262, 2952, 3028, 4461, . . .}
97 4 15996 {5, 7, 941} {120, 202, 538, 666, 736, 762, 1042, 1044,
1098, 1114, 1156, 1252, 1308, 1518, . . .}
98 2 10 {3, 11} proven
99 4 684 {5, 13, 29} {284}
100 3 2469 {7, 13, 37} {64, 433, 684, 922, 2145}
‡ partial factorization, ] factorization

University of Tennessee at Martin

E-mail address: caldwell@utm.edu