Annales Mathematicae et Informaticae

41 (2013) pp. 219–234

Proceedings of the
15th International Conference on Fibonacci Numbers and Their Applications
Institute of Mathematics and Informatics, Eszterházy Károly College
Eger, Hungary, June 25–30, 2012

The On-Line Encyclopedia of Integer

Sequences

N. J. A. Sloane

The OEIS Foundation Inc., USA

[email protected]

Abstract
We all recognize 0, 1, 1, 2, 3, 5, 8, 13, . . . but what about 1, 2, 4, 6, 3, 9, 12, 8,
10, 5, 15, . . .? If you come across a number sequence and want to know if it has
been studied before, there is only one place to look, the On-Line Encyclopedia
of Integer Sequences (or OEIS). Now in its 49th year, the OEIS contains
over 220,000 sequences and 20,000 new entries are added each year. This
article will briefly describe the OEIS and its history. It will also discuss some
sequences generated by recurrences that are less familiar than Fibonacci’s,
due to Greg Back and Mihai Caragiu, Reed Kelly, Jonathan Ayres, Dion
Gijswijt, and Jan Ritsema van Eck.
Keywords: Fibonacci, sequences, recurrences.
MSC: Primary 11B

1. The Fibonacci numbers

The Fibonacci numbers have been in the On-Line Encyclopedia of Integer Se-
quences R
(or OEIS R
) right from the beginning. When I started collecting se-
quences as a graduate student in 1964, the Fibonacci numbers became sequence
A000045 (incidentally, 49 years later, sequences being added have A-numbers
around A2220001 ). Over 3000 sequences in the OEIS mention Fibonacci’s name in
their definition.
Some especially noteworthy variations on the Fibonacci numbers were recently
defined by Back and Caragiu [2] in the Fibonacci Quarterly. The simplest of their
1 As of February 2013. Throughout this article, six-digit numbers prefixed by A refer to entries

in the OEIS [15]. As in the OEIS, we adopt the convention that a(n) denotes the nth term of the
sequence being discussed.

219
220 N. J. A. Sloane

examples replaces the Fibonacci recurrence by

a(n) = gpf(a(n − 1) + a(n − 2)), (1.1)

where gpf stands for greatest prime factor (A006530). If we start with 1, 1 we get

1, 1, 2, 3, 5, 2, 7, 3, 5, 2, 7, . . . (1.2)

(A175723), and the cycle 3, 5, 2, 7 repeats for ever. Back and Caragiu show that no
matter what the initial values are, (1.1) always becomes periodic and that 3, 5, 2, 7
is the only nontrivial cycle. On the other hand, consider

a(n) = gpf(a(n − 1) + a(n − 2) + a(n − 3)). (1.3)

If we start with 1, 1, 1 we get

1, 1, 1, 3, 5, 3, 11, 19, 11, 41, 71, 41, 17, 43, 101, 23, . . . (1.4)

(A177904), which after 86 steps enters a cycle of length 212. Now it is only a
conjecture that (1.3) always becomes periodic, for any initial values.
Another interesting variant of the Fibonacci sequence2 was very recently intro-
duced into the OEIS by Reed Kelly [12]. Kelly’s recurrence is

a(n − 1) + a(n − 3)
a(n) = , (1.5)
gcd{a(n − 1), a(n − 3)}

with initial values 1, 1, 1:

1, 1, 1, 2, 3, 4, 3, 2, 3, 2, 2, 5, 7, 9, 14, 3, 4, 9, 4, 2, . . . (1.6)

(A214551). This sequence appears to grow exponentially (a(n) ≈ const.·e0.123...n ?),

but essentially nothing has been proved about it.
The OEIS is an endless source of lovely problems!

2. How the OEIS is used

However, the main use for the OEIS is as a reference work for identifying sequences
and telling you what is known about them. If you come across a sequence of
numbers, and you want to know if it has been studied before, there is only one
place to look, the OEIS [15] (http://oeis.org).
You enter the first few terms3 , and click “Submit”. If you are lucky, the OEIS
will return one or more sequences that match what you entered, and, for each one,
it will tell you such things as:
2 Or, more precisely, of another medieval sequence, the Narayana cows sequence, A000930.
3 When looking up a sequence, it is recommended that you omit the first term or two, since
different people may start a sequence in different ways.
The On-Line Encyclopedia of Integer Sequences 221

– The definition of the sequence

– The first 10, or 10,000, or sometimes 500,000 terms
– Comments explaining further properties of the sequence
– Formulas for generating the sequence
– Computer programs for producing the sequence
– References to books and articles where the sequence is mentioned
– Links to web pages on the Internet where the sequence has appeared
– The name of the person who submitted the sequence to the OEIS
– Examples illustrating some of the terms of the sequence (for example, se-
quence A000124, which gives the maximal number of pieces that can be
obtained when cutting a circular pancake with n cuts, is illustrated with
pictures showing the pieces obtained with 1, 2, 3, 4 and 5 cuts)
– The history of each entry in the OEIS as it has evolved over time
You can also view graphs or plots of the sequence, or listen to it when it is converted
to sounds.
If your sequence is not found, you will be encouraged to submit it. This will
establish your priority over the sequence, and will help the next person who comes
across it. Only sequences of general interest should be submitted. The sequence of
primes whose decimal expansion begins with 2012 is an example of a sequence that
would not be of general interest. Published sequences are almost always acceptable.
If your sequence was not in the OEIS, you should also try sending it to our
email server Superseeker (see http://oeis.org/ol.html), which will try hard to
find an explanation for your sequence. For example, Superseeker might suggest
a recurrence or generating function for your sequence, or tell you that it can be
obtained by applying one of over a hundred different transformations to one of
the over 200,000 sequences in the OEIS. Superseeker is a very powerful tool for
analyzing sequences.
Accuracy has always been one of the top priorities in the OEIS. Its standards are
those of a mathematics reference work. Ideally, every number, formula, computer
program, etc., should be absolutely correct. Formulas that are stated uncondi-
tionally should be capable of being proved, and otherwise should be labeled as
conjectures. Of course, as the database has grown, these goals have become harder
and harder to achieve. Many non-mathematicians have difficulty in understanding
the difference between a theorem and a conjecture. (“My formula fits the first 30
terms, so obviously it must be correct.”)
The OEIS has often been called one of the most useful mathematical sites on the
Internet. There is a web page (http://oeis.org/wiki/Works\_Citing\_OEIS)
that lists over 3000 articles and books that reference it.
222 N. J. A. Sloane

3. History of the OEIS

I started collecting sequences in 1964, entering them on punched cards (the original
motivation was to find an explanation for various sequences that had arisen in
my dissertation, the simplest of which was the sequence that became A000435).
Eventually two books were published ([16] in 1973, with 2372 entries, and [17],
written with Simon Plouffe, in 1995, with 5847 sequences).
In 1996, when the number of entries had risen to 10,000, I put the database
on the Internet, calling it the The On-Line Encyclopedia of Integer Sequences or
OEIS. By 2009, the database had grown to over 150,000 entries, and was becoming
too big for one person to manage, so I set up a foundation, The OEIS Foundation
Inc (http://oeisf.org), whose goals are to own the intellectual property of the
OEIS, to maintain it, and to raise funds to support it.
With major help from Russ Cox (of Google) and my colleague David Applegate
(at AT&T), I moved the OEIS off my home page at AT&T to a commercial hosting
service, and attempted to set it up as a “wiki.” However, this proved to be extremely
difficult, and it required a tremendous amount of work by Russ Cox before it started
working properly. It was not until November 11, 2010 that the OEIS was officially
launched in its new home at http://oeis.org. This would not have been possible
without the help that Russ Cox and David Applegate provided.
The fact that the OEIS is now a wiki means that I no longer have to process
all the updates myself. Once a user has registered4 , he or she can propose new
sequences or updates to existing sequences. All submissions are reviewed by a
panel of about 80 editors. Nearly two years after it was launched, the wiki system is
working well. Since November 2010 the database has grown from 180,000 sequences
to its current number of around 220,000. From 1996 to the present, the database
has grown at between 10,000 and 20,000 new sequences per year, with about an
equal number of entries that are updated.
More about the history of the OEIS can be found on the OEIS Foundation web
site, http://oeisf.org.5

4. The poster and the OEIS movie

To celebrate the creation of the the OEIS Foundation, David Applegate and I
made a poster that shows 25 especially interesting sequences (several of which will
be mentioned in this article). It can be downloaded (along with a key) from the
Foundation web site.
Also, Tony Noe made a movie that shows graphs of the first thousand terms of a
thousand sequences from the OEIS: it is quite spectacular. It runs for 8.5 minutes,
4 Allreaders are encouraged to register: go to http://oeis.org/wiki and click “Register.”
5 AsPresident, it would be remiss of me not to mention that the OEIS Foundation is a charitable
organization and donations are tax-deductible in the USA. The web site is free, and none of the
trustees receive a salary. To make a donation, please go to http://oeisf.org.
The On-Line Encyclopedia of Integer Sequences 223

and it too can be found on the Foundation web site. It is also on YouTube (search
for “OEIS movie”).

5. Puzzles
One of the goals of the OEIS has always been to help people get higher scores on
IQ tests, and the database includes many sequences that have appeared as puzzles.
The following are a few examples. If you can’t solve them, you know where to find
the answers!

• 61, 21, 82, 43, . . .

• 2, 4, 6, 30, 32, 34, 36, 40, 42, 44, 46, 50, 52, 54, 56, 60, 62, 64, 66, 2000, . . .
• 0, 0, 0, 0, 4, 9, 5, 1, 1, 0, 55, 55, . . .
• 5, 8, 12, 18, 24, 30, 36, 42, 52, 60, . . .
• 1, 2, 6, 21, 85, 430, 2586, 18109, 144880, . . .

The last one is a bit tricky, but it did appear on a quiz.

6. Two sequences that agree for a long time

People often ask if it is possible for two sequences to agree for many terms yet not
be the same. Here is an extreme example. The sequences
   
2n 2
and
log 2 21/n − 1
both begin
2, 5, 8, 11, 14, 17, 20, 23, 25, 28, 31, 34, 37, . . .
(A078608). In fact they agree for the first 777451915729367 terms! There are
infinitely many disagreements, the positions of which form sequence A129935:
777451915729368, 140894092055857794, 1526223088619171207, 3052446177238342414, . . .

7. Theorems resulting from the OEIS

Another question that is often asked is if there are any theorems that have resulted
from the OEIS. The answer is that there are many such examples. In the list of
papers that cite the OEIS (http://oeis.org/wiki/Works\_Citing\_OEIS) one
will find numerous acknowledgments that say things like “This result was discovered
with the help of the OEIS.”
224 N. J. A. Sloane

I will give three concrete examples of theorems that were discovered with the
help of the OEIS. The first concerns the remainder term in Gregory’s series for
π/2,
π X∞
(−1)k+1  1 1 1 
=2 = 2 1 − + − + ··· , (7.1)
2 2k + 1 3 5 7
k=1

which is famous for converging very slowly. In 1987, Joseph North observed that if
one truncates the series after 50,000 terms, the answer is of course wrong. There is
an error in the fifth decimal place. Surprisingly, he noticed that the next nine digits
are correct, then there is an error, then there are nine more correct digits, another
error, and so on. Here is the decimal expansion of the truncated sum followed by
the true value of π/2 (the sequences of digits form A013706 and A019669). The
digits that differ are in bold-face.

1.570796326794896619231321691639751442098584699687 . . . (truncated)
1.570786326794897619231321191639752052098583314687 . . . (true value)

The differences between the corresponding bold-faced terms are

1, −1, 5, −61, 1385, . . . .

Jonathan Borwein looked up this sequence in [16], and found that (apart from signs)
it appeared to be the Euler numbers, A000364. The end result of this investigation
was a new theorem.
Theorem 7.1 (Borwein, Borwein and Dilcher [4]; see also [3, pp. 28–29], [5]).
N/2 ∞
π X (−1)k+1 X Em
−2 ∼ 2m+1
, (7.2)
2 2k + 1 m=0
N
k=1

where the Em are the Euler numbers (A000364):

1, 1, 5, 61, 1385, 50521, 2702765, 199360981, 19391512145, . . . .

The second example is one that I was involved with personally. It began when
Eric W. Weisstein (at Wolfram Research, and creator of MathWorld) wrote to
me about a discovery he had made. He had been classifying real matrices of 0’s
and 1’s according to various properties, and he found that the numbers of such
matrices all of whose eigenvalues were positive were 1, 3, 25, 543, 29281 for matrices
of orders 1, 2, . . . , 5. He observed that these numbers coincided with the beginning
of sequence A003024 (whose definition on the surface seemed to have nothing to do
with eigenvalues), and he conjectured that the sequences should in fact be identical.
He was right, and this led to the following theorem.
Theorem 7.2 ([14]). The number of acyclic directed graphs with n labeled vertices
is equal to the number of n × n matrices of 0’s and 1’s all of whose eigenvalues are
real and positive.
The On-Line Encyclopedia of Integer Sequences 225

The third example is a result of Deutsch and Sagan [8]. It is well-known that
the famous Catalan numbers
 
1 2n
Cn :=
n+1 n

(A000108) are odd if and only if n = 2k − 1 for some k. Deutsch and Sagan proved
(among other things) an analogous result for the almost equally-famous Motzkin
numbers (A001006),
Xn  
n
Mn := Ck .
2k
k=1

Theorem 7.3 ([8]). Mn is even if and only if n ∈ 4S − 2 or 4S − 1, where

S := (1, 3, 4, 5, 7, 9, 11, 12, 13, 15, . . .)

lists the numbers whose binary expansion ends with an even number of 0’s (A003159).

8. Three unusual recurrences

The Fibonacci recurrence is very nice, but it is 800 years old. In the last section
of this article I will discuss some modern recurrences that I find very appealing.

8.1. The EKG sequence

140
120
100
80
a(n)
60
40
20
0

0 20 40 60 80 100
n

Figure 1: The first 100 terms of the EKG sequence, with successive
points joined by lines
226 N. J. A. Sloane

1600
1400
1200
1000
a(n)
800
600
400

800 850 900 950 1000

n

Figure 2: Terms 800 to 1000 of the EKG sequence

Jonathan Ayres contributed this to the OEIS in 2001 [1]. The first 18 terms are
1, 2, 4, 6, 3, 9, 12, 8, 10, 5, 15, 18, 14, 7, 21, 24, 16, 20, . . .
(A064413), and the defining recurrence is a(1) = 1, a(2) = 2, and, for n ≥ 3,
a(n) is the smallest natural number not yet in the sequence
which has a common factor > 1 with the previous term.

.
..
1500

..
. .
.
. ..
..
.. .
..
. .
. .
. ..
. .
. .. . ..........
.............. .
1000

..
. ............. .
.. .
. ..... ................ .
. ... ...
. ............ .
. .. . . ........ ..
. .. . . . ......
.................
a(n)

.. .
.....
.
.. ....... .
. .. .. .......... .
. .. .. ........... .
....... .
..
.
....
............. .
.. . .
. . . .... .
... . ............
........... . .. . .. .
.
500

.. .. ........ .
.. . .......... .. . . .
.....
. .. . . . .
. .
..
..
. .
............... . . . . . . ..
.
.... . . ..
. .......... . . . ..
. ... . ........ . .. . .
.. .......... . . .. . . .
.. . . .................. . . . .. .. . .
..
. . . ............ .. . ...
. ...
. . ........... . ... . .
............... .. ... .
..
.................. . . . ... . . .
.
.
.... ... .
.............
0

0 200 400 600 800 1000

n

Figure 3: Scatter-plot of the first 1000 terms of the EKG sequence.

They lie roughly on three almost-straight lines.
The On-Line Encyclopedia of Integer Sequences 227

Thus a(3) must have a common factor with 2, i.e. it must be even, and 4 is the
smallest candidate, so a(3) = 4. The next term must also be even, so a(4) = 6.
The smallest number not yet in the sequence which has a common factor with 6
is 3, so a(5) = 3. Similarly, a(6) = 9, a(7) = 12, a(8) = 8, a(9) = 10, a(11) = 5,
a(12) = 15, and so on. Jeffrey Lagarias, Eric Rains and I studied this sequence in
[13]. We called it the EKG sequence, since it looks like an electrocardiogram when
plotted (Figs. 1, 2).
It is not difficult to show that the primes appear in increasing order, and that
each odd prime p is either preceded by 2p and followed by 3p, or is preceded by
3p and followed by 2p (as we just saw, 3 was preceded by 6 and followed by 9, 5 is
preceded by 10 and followed by 15).6
By definition, no number can can be repeated. But does every number appear?
The answer is Yes.
Theorem 8.1. The EKG sequence is a permutation of the natural numbers.
Sketch of Proof. (i) If infinitely many multiples of some prime p occur in the se-
quence, then every multiple of p must occur. (For if not, let kp be the smallest
missing multiple of p. Every number below kp either appears or it doesn’t, but
once we get to a multiple of p beyond all those terms, the next term must be kp,
which is a contradiction.) (ii) If every multiple of a prime p appears, then every
number appears. (The proof is similar.) (iii) Every number appears. (For if there
are only finitely many different primes among the prime factors of all the terms,
then some prime must divide infinitely many terms, and the result follows from
(i) and (ii). On the other hand, if infinitely many different primes p appear, then
there are infinitely many terms 2p, as noted above, so 2 appears infinitely often,
and again the result follows from (i) and (ii).)
Although the initial terms of the sequence jump around, when we look at the
big picture we find that the points lie very close to three almost-straight lines (Fig.
3). This is somewhat similar to the behavior of the prime numbers, which are
initially erratic, but lie close to a smooth curve (since the nth prime is roughly
n log n) when we look at the big picture – see Don Zagier’s lecture on “The first 50
million prime numbers” [18].
In fact, we have a precise conjecture about the three lines on which the points
lie. We believe – but are unable to prove – that almost all a(n) satisfy the asymp-
totic formula a(n) ∼ n(1 + 1/(3 log n)) (the central line in Fig. 3), and that the
exceptional values a(n) = p and a(n) = 3p, for p a prime, produce the points on
the lower and upper lines. We were able to show that the sequence has essentially
linear growth (there are constants c1 and c2 such that c1 n < a(n) < c2 n for all n),
but the proof of even this relatively weak result was quite difficult. It would be
nice to have better bounds.

6 We conjectured that p was always preceded by 2p rather than 3p. This was later proved by

Hofman and Pilipczuk [11].

228 N. J. A. Sloane

8.2. Gijswijt’s sequence and the Curling Number Conjecture

1 1 2
1 1 2 2 2 3
1 1 2
1 1 2 2 2 3 2
1 1 2
1 1 2 2 2 3
1 1 2
1 1 2 2 2 3 2 2 2 3 2 2 2 3 3 2
1 1 2
1 1 2 2 2 3
1 1 2
1 1 2 2 2 3 2
1 1 2
1 1 2 2 2 3
1 1 2
1 1 2 2 2 3 2 2 2 3 2 2 2 3 3 2 2 2 3 2

Table 1: The first 98 terms of Gijswijt’s sequence (A090822)

We start by defining the curling number of a sequence. Let S be a finite nonempty

sequence of integers. By grouping consecutive terms, it is always possible to write
it as S = X Y Y · · · Y = X Y k , where X and Y are sequences of integers and Y is
nonempty. There may be several ways to do this: choose the one that maximizes
the value of k: this k is the curling number of S.
For example, if S = 0 1 2 2 1 2 2 1 2 2, we could write it as X Y 2 , where X =
0 1 2 2 1 2 2 1 and Y = 2, or as X Y 3 , where X = 0 and Y = 1 2 2. The latter
representation is to be preferred, since it has k = 3, and as k = 4 is impossible, the
curling number of this S is 3.
In 2004, Dion Gijswijt, then a graduate student at the University of Amsterdam
and also the puzzle editor for the Dutch magazine Pythagoras, contributed the
following sequence to the OEIS. Start with a(1) = 1, and, for n ≥ 2, use the
recurrence
a(n) = curling number of a(1), . . . , a(n − 1).

The beginning of the sequence is shown in Table 1 (it has been broken up into
sections to show where the curling number drops back to 1):
This sequence was analyzed by Gijswijt, Fokko van de Bult, John Linderman,
Allan Wilks and myself [6]. The first time a 4 appears is at a(220). We computed
several million terms without finding a 5, and for a while we wondered if perhaps
no term greater than 4 was ever going to appear. However, we were able to show
that a 5 does eventually appear, although the universe would grow cold before a
The On-Line Encyclopedia of Integer Sequences 229

direct search would find it. The first 5 appears at about term
23
1010 .
We also showed that the sequence is actually unbounded, and we conjecture that
the first time that a number m (= 5, 6, 7, . . .) appears is at about term number
·m−1
··
4
23
2 ,
a tower of height m − 1.
Our arguments could be considerably simplified if the Curling Number Conjec-
ture were known to be true. This states that:
If one starts with any initial sequence of integers, and extends
it by repeatedly calculating the curling number and appending
it to the sequence, the sequence will eventually reach 1.
The conjecture is still open. One way to tackle it is to consider starting se-
quences S0 that contain only 2’s and 3’s, and to see how far such a sequence will
extend (by repeatedly appending the curling number) before reaching a 1.
Let µ(n) denote the maximal length that can be achieved before a 1 appears,
for any starting sequence S0 consisting of n 2’s and 3’s. For n = 4, for example,
S0 = 2 3 2 3 produces the sequence
232322231 ...,
and no other starting string does better, so µ(4) = 8. The Curling Number Conjec-
ture would imply that µ(n) < ∞ for all n. Reference [6] gave µ(n) for 1 ≤ n ≤ 30,
and Benjamin Chaffin and I have determined µ(n) for all n ≤ 48 [7]. By making
certain plausible assumptions about S0 , we have also computed lower bounds on
µ(n) (which we conjecture to be the true values) for all n ≤ 80. The results are
shown in Table 2 and Figure 4. The values of µ(n) also form sequence A094004 in
[15].

As can be seen from Fig. 4, up to n = 80, it appears that µ(n) increases in a

piecewise linear manner. At the values n = 1, 2, 4, 6, 8, 9, 10, 11, 14, 19, 22, 48, 68, 76,
77 (A160766), assuming that the values in Table 2 are correct, there is a jump, but
at the other values of n, µ(n) is simply µ(n − 1) + 1. Table 3 gives the starting
sequences where µ(n) > µ(n − 1) + 1 for n ≤ 48.

For example, Table 2 shows that

µ(n) = n + 120 for 22 ≤ n ≤ 47. (8.1)
In this range one cannot do any better than taking the starting sequence for n = 22
and prefixing it by an irrelevant sequence of 47 − n 2’s and 3’s. However, at n = 48
a new record-holder appears, and it seems that
µ(n) = n + 131 for 48 ≤ n ≤ 67. (8.2)
230 N. J. A. Sloane

n 1 2 3 4 5 6 7 8 9 10 11 12
µ(n) 1 4 5 8 9 14 15 66 68 70 123 124
n 13 14 15 16 17 18 19 20 21 22 23 24
µ(n) 125 132 133 134 135 136 138 139 140 142 143 144
n 25 26 27 28 29 30 31 32 33 34 35 36
µ(n) 145 146 147 148 149 150 151 152 153 154 155 156
n 37 38 39 40 41 42 43 44 45 46 47 48
µ(n) 157 158 159 160 161 162 163 164 165 166 167 179
n 49 50 51 52 53 54 55 56 57 58 59 60
µ(n) 180 181 182 183 184 185 186 187 188 189 190 191
n 61 62 63 64 65 66 67 68 69 70 71 72
µ(n) 192 193 194 195 196 197 198 200 201 202 203 204
n 73 74 75 76 77 78 79 80 81 82 83 84
µ(n) 205 206 207 209 250 251 252 253 ? ? ? ?
Table 2: Lower bounds on µ(n), the record for a starting sequence
of n 2’s and 3’s. Entries for n ≤ 48 are known to be exact (and we
conjecture the other entries are exact).

Figure 4: Scatter-plot of lower bounds on µ(n), the record for a

starting sequence of n 2’s and 3’s. Entries for n ≤ 48 are known to
be exact (and we conjecture the other entries are exact).

We have not succeeded in finding any algebraic constructions for good starting
sequences. For more about the Curling Number Conjecture see [7].
The On-Line Encyclopedia of Integer Sequences 231

n Starting sequence
1 2
2 22
4 2323
6 222322
8 23222323
9 223222323
10 2323222322
11 22323222322
14 22323222322323
19 2232232322232232232
22 2322322323222323223223
48 223223232223222322322232232322232223223222322323
Table 3: Starting sequences of n 2’s and 3’s for which µ(n) >
µ(n − 1) + 1. This is complete for n ≤ 48 and is believed to be
complete for n ≤ 67.

8.3. Van Eck’s sequence

In 2010, Jan Ritsema van Eck [9] contributed a sequence to the OEIS that is defined
by yet another unusual recurrence. Again we start with a(1) = 0, and then for
n ≥ 2,

a(n) is the number of steps backwards before the previous

appearance of a(n − 1), or a(n) = 0 if a(n − 1) has never
appeared before.

Since a(1) = 0 has never appeared before, a(2) = 0. Now 0 has appeared one
step before, at a(1), so a(3) = 1. We have not seen a 1 before, so a(4) = 0. We
had an earlier 0 at a(2), so a(5) = 4 − 2 = 2. This is the first 2 we have seen, so
a(6) = 0. And so on. The first 36 terms are shown in Table 4.

0, 0, 1, 0, 2, 0, 2, 2, 1, 6, 0, 5,
0, 2, 6, 5, 4, 0, 5, 3, 0, 3, 2, 9,
0, 4, 9, 3, 6, 14, 0, 6, 3, 5, 15, 0,
5, 3, 5, 2, 17, 0, 6, 11, 0, 3, 8, 0, . . .
Table 4: The first 48 terms of Van Eck’s sequence (A181391)

Figure 5 shows a scatter-plot of the first 800 terms. The plot suggests that
after n terms, there are occasionally terms around n, or in other words that
lim sup a(n)/n ≈ 1. This is confirmed by looking at the first million terms, and the
data also strongly suggests that every number appears in the sequence. However,
at present these are merely conjectures.
232 N. J. A. Sloane

Van Eck was able to show that there are infinitely many 0’s in the sequence,
or, equivalently, that the sequence is unbounded.

Figure 5: Scatter-plot of the first 800 terms of Van Eck’s sequence

A181391

Theorem 8.2 (Van Eck, personal communication). The sequence contains in-
finitely many 0’s.
Proof. Suppose, seeking a contradiction, that there are only finite number of 0’s in
the sequence. Then after a certain point no new terms can appear, so the sequence
is bounded. Let M be the largest term. This means that any block of M successive
terms determines the sequence. But there are only M M different possible blocks.
So a block must repeat and the sequence is eventually periodic. Furthermore, the
period cannot contain a 0.
Suppose the period has length p, and starts at term r, with a(r) = x, . . . , a(r +
p − 1) = z, a(r + p) = x, . . . . There is another z after q ≤ p steps, which is
immediately followed by q. But this q implies that a(r − 1) = z. Therefore the
periodic part really began at step r − 1.
Repeating this argument shows that the periodic part starts at a(1). But a(1) =
0, and the periodic part cannot contain a 0. Contradiction.

It would be nice to know more about this fascinating sequence!

9. Conclusion
I will end with a few general remarks.
The On-Line Encyclopedia of Integer Sequences 233

• The OEIS needs more editors. If you are interested in helping, please write
to me or one of the other Editors-in-Chief. There are no formal duties,
everything is done on a volunteer basis, and you will get to see a lot of
interesting new problems.

• Everyone should register with the OEIS – see Sect. 3.

• If you write a paper that mentions a sequence in the OEIS, please do two
things. Add it to the list of papers that cite the OEIS – see Sec. 2, and add a
reference pointing to your paper to any entries in the OEIS that it mentions.
• The same thing if you come across a sequence in your work, in the library,
or on a web site: send it in to the OEIS if it is missing (it need not be your
own sequence – just mention the source) or add a reference to the source if it
is already in the OEIS. It is these cross-connections that make the database
so valuable.

References
[1] J. Ayres, Sequence A064413 in [15], Sept. 30, 2001.
[2] G. Back and M. Caragiu, The greatest prime factor and recurrent sequences, Fib.
Quart., 48 (2010), 358–362.
[3] J. M. Borwein, D. H. Bailey and R. Girgensohn, Experimentation in Mathematics,
A K Peters, Natick, MA, 2004.
[4] J. M. Borwein, P. B. Borwein and K. Dilcher, Pi, Euler numbers and asymptotic
expansions, Amer. Math. Monthly, 96 (1989), 681–687.
[5] J. M. Borwein and R. M. Corless, Review of “An Encyclopedia of Integer Sequences”
by N. J. A. Sloane and Simon Plouffe, SIAM Review, 38 (1996), 333–337.
[6] F. J. van der Bult, D. C. Gijswijt, J. P. Linderman, N. J. A. Sloane and A. R. Wilks,
A slow-growing sequence defined by an unusual recurrence, J. Integer Sequences, 10
(2007), Article 07.1.2.
[7] B. Chaffin, J. P. Linderman, N. J. A. Sloane and A. R. Wilks, On curling numbers
of integer sequences, Preprint, 2012; arXiv:1212.6102.
[8] E. Deutsch and B. E. Sagan, Congruences for Catalan and Motzkin numbers and
related sequences, J. Num. Theory, 117 (2006), 191–215.
[9] J. R. van Eck, Sequence A181391 in [15], Oct. 17, 2010.
[10] D. Gijswijt, Sequence A090822 in [15], Feb. 27, 2004.
[11] P. Hofman and M. Pilipczuk, A few new facts about the EKG sequence, J. Integer
Sequences, 11 (2008), Article 08.4.2.
[12] R. Kelly, Sequence A214551 in [15], July 20, 2012.
[13] J. C. Lagarias, E. M. Rains and N. J. A. Sloane, The EKG sequence, Experimental
Math., 11 (2002), 437–446.
234 N. J. A. Sloane

[14] B. D. McKay, F. E. Oggier, N. J. A. Sloane, G. F. Royle, I. M. Wanless and H. S. Wilf,

J. Integer Sequences, 7 (2004), Article 04.3.3.
[15] The OEIS Foundation Inc., The On-Line Encyclopedia of Integer Sequences, http:
//oeis.org, 2012.
[16] N. J. A. Sloane, A Handbook of Integer Sequences, Academic Press, NY, 1973.
[17] N. J. A. Sloane and S. Plouffe, The Encyclopedia of Integer Sequences, Academic
Press, 1995.
[18] D. Zagier, The first 50 million prime numbers, Math. Intelligencer, 0 (1977), 7–19.