On Generalized Choral Sequences

Joel Reyes Noche

Department of Mathematics and Natural Sciences
Ateneo de Naga University
Naga City, Camarines Sur
email: jrnoche@adnu.edu.ph

Abstract

Generalized choral sequences are infinite binary words (cn )n≥0 defined by c3i+r0 =
0, c3i+r1 = 1, and c3i+rc = ci (where the r’s are distinct fixed elements of {0, 1, 2})
for all non-negative integers i. We present some of their properties. In particular, we
show how each is generated by a deterministic finite automaton with output and how
each is a fixed point of a uniform morphism. We also look at their subword complexity
and Lyndon factorizations.

1 Introduction
We use definitions and notation from Allouche and Shallit [1]. Denote by N the set of
non-negative integers and by Z+ the set of positive integers. A word is a concatenation (a
sequence) of letters chosen from an alphabet (a non-empty set of letters). A finite word has
a finite length (the number of letters it contains) and is denoted by a lowercase italic letter.
The length of a finite word w is denoted by |w|. The empty word, denoted by , has a length
of zero. A one-sided right-infinite word (which we will simply call an infinite word) is a map
from N to an alphabet and is denoted by a lowercase boldface letter. We call a word over
the alphabet {0,1} a binary word.
Concatenation of words is denoted by the juxtaposition of their symbols. For example,
if w = 0 and x = 01, then wx = 001, xw = 010, w1 = x, and wwxwx = w(wx)2 = 03 102 1.
We define w1 = w and w0 =  for any finite word w. If x is a finite non-empty word, then
xω is the infinite word xxx · · ·.
A word y is a subword of w if there exist x and z such that w = xyz. If x = , then y is
a prefix of w. If z = , then y is a suffix of w. If x =  and z 6= , then y is a proper prefix
of w. If x 6=  and z = , then y is a proper suffix of w.
We extend these definitions to infinite words. An infinite word w can be written as an
infinite sequence of finite subwords (wn )n≥0 = w0 w1 w2 · · ·. A word y is a subword of w if
there exist x and z such that w = xyz. If x = , then y is a prefix (and a proper prefix) of
w. A word z is a subword (and a suffix) of w if there exists a y such that w = yz. If y 6= ,
then z is a proper suffix of w.
The set of all finite words made up of letters chosen from an alphabet Σ is denoted by
Σ∗ . Note that  ∈ Σ∗ . If a ∈ Σ and w ∈ Σ∗ , then |w|a denotes the number of occurrences
of the letter a in the word w. The frequency of a letter a in an infinite word w = (wn )n≥0

1
(where the w’s are letters), denoted by Freqw (a), is limn→∞ n1 |w0 w1 · · · wn−1 |a , if this limit
exists.
Let Σ and ∆ be alphabets. A morphism is a map µ from Σ∗ to ∆∗ that obeys the
identity µ(wx) = µ(w)µ(x) for all words w, x ∈ Σ∗ . If Σ = ∆, then the application of a
morphism can be iterated. For example, if µ is the morphism mapping 0 to 1 and 1 to 10,
then µ(0) = 1, µ2 (0) = µ(µ(0)) = µ(1) = 10, and so on. We define µ1 (w) = µ(w) and
µ0 (w) = w for any word w.
A morphism µ : Σ∗ → ∆∗ is k-uniform if there is a constant k such that |µ(a)| = k for
all a ∈ Σ. A coding is a 1-uniform morphism.
A fixed point of a morphism µ : Σ∗ → Σ∗ is a finite word w (or infinite word w) such
that µ(w) = w (or µ(w) = w). If there exists a letter a ∈ Σ such that µ(a) = ax and x is a
word composed of letters xi ∈ Σ such that µm (xi ) 6=  for any m ∈ Z+ , then the morphism
µ is prolongable on the letter a. If so, then limm→∞ µm (a) (denoted by µω (a)) is the fixed
point of µ iterated on a, where the length of the iterates from the letter a tends to infinity.
Let w denote the complement of the finite binary word w (and w denote the complement
of the infinite binary word w) where the overbar represents the morphism mapping 0 → 1
and 1 → 0. For example, if w = 001, then w = 110.
Given a finite word w = a0 a1 · · · an , where the a’s are letters, its reversal, denoted by
wR , is an · · · a1 a0 .
Definition 1. A generalized choral sequence is an infinite binary word c(r0 , r1 , rc , z) =
(cn )n≥0 defined by c3i+r0 = 0, c3i+r1 = 1, c3i+rc = ci , and c0 = z (where the r’s are distinct
fixed elements of {0, 1, 2} and z = 0 if r0 = 0, z = 1 if r1 = 0, and z could either be 0 or 1
if rc = 0) for all i ∈ N.
There are eight distinct choral sequences: (Spaces have been inserted to improve read-
ability.)
c(0, 2, 1, 0) = 001 001 011 001 001 011 001 011 011 · · ·
c(1, 2, 0, 0) = 001 001 101 001 001 101 101 001 101 · · ·
c(0, 1, 2, 0) = 010 011 010 010 011 011 010 011 010 · · ·
c(2, 1, 0, 0) = 010 110 010 110 110 010 010 110 010 · · ·
c(1, 2, 0, 1) = 101 001 101 001 001 101 101 001 101 · · ·
c(1, 0, 2, 1) = 101 100 101 101 100 100 101 100 101 · · ·
c(2, 1, 0, 1) = 110 110 010 110 110 010 010 110 010 · · ·
c(2, 0, 1, 1) = 110 110 100 110 110 100 110 100 100 · · ·
Sequence c(0, 2, 1, 0) is Stewart’s choral sequence. (Stewart [11] presented the sequence
(cn )n≥1 and not (cn )n≥0 .) Sequence c(0, 1, 2, 0) is from Berstel and Karhumäki [2].
Definition 2. A generalized choral sequence c(r0 , r1 , rc , z) is called a type-012 sequence
if (r0 , r1 , rc ) is a circular permutation of (0, 1, 2). Otherwise (if (r0 , r1 , rc ) is a circular
permutation of (2, 1, 0)), it is called a type-210 sequence.

2 Some Properties
A previous work [8] presented a characteristic function for generalized choral sequences as
well as proofs of the following two theorems.
Theorem 1. A generalized choral sequence is cube-free, that is, it does not contain any
subword of the form xxx, where x is a non-empty finite subword.

2
Theorem 2. Given two generalized choral sequences, if they are both type-012 (or if they
are both type-210) then any finite subword of one is a subword of the other. Otherwise, any
finite subword of one is the complement of a subword of the other.

The following theorem is related to the previous theorem and their proofs are similar.

Theorem 3. Given a type-012 sequence and a type-210 sequence, any finite subword of one
is the reversal of a subword of the other.

Proof. Let v = (vn )n≥0 be the given type-012 sequence and w = (wn )n≥0 be the given
type-210 sequence. We will show that any finite subword of v is the reversal of a subword
of w. (The proof that any finite subword of w is the reversal of a subword of v is similar.)
Because any generalized choral sequence has all the subwords 001, 010, 011, 100, 101,
and 110 [8], any length-3 subword of v is the reversal of a subword of w. That is, for a
given j ∈ N, there exists a k ∈ N such that vj vj+1 vj+2 = wk+2 wk+1 wk .
By Definition 2, there exists a subword v3j+rc v3j+rc +1 v3j+rc +2 v3j+rc +3 v3j+rc +4 v3j+rc +5
v3j+rc +6 = vj 01vj+1 01vj+2 and a subword w3k+rc w3k+rc +1 w3k+rc +2 w3k+rc +3 w3k+rc +4
w3k+rc +5 w3k+rc +6 = wk 10wk+1 10wk+2 such that vj 01vj+1 01vj+2 = wk+2 01wk+1 01wk .
Any length-5 subword of v is a subword of a length-7 subword of the form v3j+rc v3j+rc +1
v3j+rc +2 v3j+rc +3 v3j+rc +4 v3j+rc +5 v3j+rc +6 . Thus, any length-5 subword of v is the reversal
of a subword of w. (This is also true for subwords of length less than 5.)
Extending this reasoning to arbitrarily long finite subwords of similar form yields the
result that any finite subword of v is the reversal of a subword of w.
1
Remark 1. For any generalized choral sequence c, Freqc (0) = 2 and Freqc (1) = 12 .

We first show that Freqc (0) = 12 . Let c = (cn )n≥0 . Consider the subword c0 c1 · · · cn−1
with length n. As n → ∞, around 13 of the letters of c0 c1 · · · cn−1 will be due to the subwords
c3i+r0 , around 13 will be due to c3i+r1 , and around 31 will be due to c3i+rc , for i ∈ N such
that 3i + r < n.
Since c3i+r0 = 0, the number of occurrences of 0 due to the subwords c3i+r0 approaches
n
3 as n → ∞. Since c3i+r1 = 1, the number of occurrences of 0 due to the subwords c3i+r0
and c3i+r1 still approaches n3 as n → ∞.
Now consider the subwords c3i+rc . Since c3i+rc = ci , one third of the subwords c3i+rc
will be 0, and one third will be 1. Of the remaining subwords, one third will be 0, one third
will be 1 and so on.
Thus, the number of

occurrences of 0 in the subword c0 c1 · · · cn−1 as n → ∞ approaches

n 1 1
3 1 + 1 + (1 + · · ·) = n3 + 3n2 + 3n3 +· · · and Freqc (0) = limn→∞ n1 n3 + 3n2 + 3n3 + · · ·
P∞ 3 1 3
P∞ 1
i−1
= i=1 3i = −1 + i=1 3 = −1 + 1−1 1 = 12 .
3
Using similar reasoning, we can show that Freqc (1) = 21 .

Proposition 1. c(1, 2, 0, z) = zc(0, 1, 2, 0) and c(2, 1, 0, z) = zc(1, 0, 2, 1)

Proof. We first show that c(1, 2, 0, z) = zc(0, 1, 2, 0). Let (an )n≥0 = c(1, 2, 0, z), (bn )n≥0 =
c(0, 1, 2, 0), and (cn )n≥0 = zc(0, 1, 2, 0). Thus, a3i+1 = 0, a3i+2 = 1, a3i = ai , b3i = 0,
b3i+1 = 1, and b3i+2 = bi for all i ∈ N. Also, ci+1 = bi for all i ∈ N so c3i+1 = b3i = 0,
c3i+2 = b3i+1 = 1, and c3i+3 = b3i+2 = bi = ci+1 for all i ∈ N. Since c3(i+1) = ci+1 for i ∈ N
and c3·0 = c0 , c3i = ci for all i ∈ N. Also, a3i+1 = 0 = c3i+1 and a3i+2 = 1 = c3i+2 for all
i ∈ N.

3
 We will prove that (an )n≥0 = (cn )n≥0 using induction. Assume an = cn for all n ≤ k,
where n, k ∈ N and k is fixed. There are three possibilities: k ≡ 0 (mod 3), k ≡ 1 (mod
3), and k ≡ 2 (mod 3). If k ≡ 0 (mod 3), then k + 1 = 3i + 1 for some i ∈ N and
ak+1 = a3i+1 = c3i+1 = ck+1 . If k ≡ 1 (mod 3), then k + 1 = 3i + 2 for some i ∈ N and
ak+1 = a3i+2 = c3i+2 = ck+1 . If k ≡ 2 (mod 3), then k + 1 = 3i for some i ∈ Z+ and
ak+1 = a3i = ai = ci = c3i = ck+1 (where ai = ci because i ≥ 1 and i ≤ 3i − 1 = k). Thus,
if an = cn for all n ≤ k, then an = cn for all n ≤ k + 1. Since a0 = c0 = z (that is, an = cn
for all n ≤ 0), it follows that an = cn for all n ∈ N.
The proof that c(2, 1, 0, z) = zc(1, 0, 2, 1) is similar.
Proposition 2. The complement of a generalized choral sequence is also a generalized choral
sequence.
Proof. Let c(r0 , r1 , rc , z) = (an )n≥0 . Thus, a3i+r0 = 0, a3i+r1 = 1, and a3i+rc = ai for all
i ∈ N. Also, a0 = z where z = 0 if r0 = 0, z = 1 if r1 = 0, and z = 0 or z = 1 if rc = 0.
Let (bn )n≥0 = (an )n≥0 so that bi = ai for all i ∈ N. Thus, b3i+r0 = a3i+r0 = 1, b3i+r1 =
a3i+r1 = 0, and b3i+rc = a3i+rc = ai = bi for all i ∈ N. Also, b0 = a0 = z where z = 1 if
r0 = 0, z = 0 if r1 = 0, and z = 1 or z = 0 if rc = 0. Thus, c(r0 , r1 , rc , z) = c(r00 , r10 , rc0 , z 0 )
where r00 = r1 , r10 = r0 , rc0 = rc , and z 0 = 0 if r00 = 0, z 0 = 1 if r10 = 0 and z 0 = 0 or z 0 = 1 if
rc0 = 0.
We follow Cassaigne and Karhumäki [3]. Let Σ be an alphabet and ? be a letter not
in Σ. For a word w = xy with x ∈ Σ and y ∈ (Σ ∪ {?})∗ , let T0 (w) = wω and let Ti (w)
for i ∈ Z+ be the word obtained from Ti−1 (w) by replacing the first occurrence of ? in
Ti−1 (w) by the i-th letter of Ti−1 (w). The Toeplitz word determined by the pattern w is
T (w) = limi→∞ Ti (w), an infinite word over Σ.
Using this definition of Toeplitz words and Definition 1, we see that some generalized
choral sequences are Toeplitz words over {0,1} [3, Example 4].
Remark 2. The sequences c(0, 2, 1, 0), c(0, 1, 2, 0), c(1, 0, 2, 1), and c(2, 0, 1, 1) are the
Toeplitz words T (0?1), T (01?), T (10?), and T (1?0), respectively.
The other generalized choral sequences are not Toeplitz words (unless we relax the con-
dition that w start with a letter from Σ [3, Example 2]).

2.1 Automatic Sequence

We follow Allouche and Shallit [1] again. A deterministic finite automaton with output
(DFAO) is a model of computation defined by a finite set of states Q, a finite input alphabet
Σ, a transition function δ : Q×Σ → Q (which we extend to δ : Q×Σ∗ → Q so that δ(q, ) = q
and δ(q, xa) = δ(δ(q, x), a) for all q ∈ Q, x ∈ Σ∗ , and a ∈ Σ [1, p. 129]), an initial state
q0 ∈ Q, a finite output alphabet ∆, and an output function τ : Q → ∆. A DFAO has a word
w ∈ Σ∗ as input and it moves from state to state according to δ while reading the letters
of w (in order from left to right). When the end of w is reached, the automaton halts in a
state q and outputs the letter τ (q).
A DFAO with representations of base-k numbers as input is called a k-DFAO. A finite-
state function f : Σ∗ → ∆ is one that can be computed by a DFAO such that f (w) =
τ (δ(q0 , w)). Informally, a word w = (wn )n≥0 is k-automatic if wn is a finite-state function
of the base-k digits of n (starting with the most significant digit). (Note that the input
can have an arbitrary finite number of leading zeros.) A k-automatic infinite word has an
associated k-DFAO.

4
Proposition 3. A generalized choral sequence is 3-automatic.

Proof. The 3-DFAO of a generalized choral sequence c(r0 , r1 , rc , z) is shown below as a table
and as a transition diagram, where Q = {qr0 ,0 , qr1 ,1 , qrc ,0 , qrc ,1 }, Σ = {0, 1, 2}, ∆ = {0, 1},
and the initial state is q0,z . ...... ......
....
...... .......... ....... ...........
... ...
... ... ... ...
... ..... .
... r
..... 0 ... ...........................
...
r 1 ........... ...
...
...
.
. . .
.... r 1 ......
.... .......... . .
......... ..
.............................
. .......
........................................................................................... .... ........................
. ... .. .... ....
... ... ...
... .. .... ... ..
..
.... q /0 .......................................................................
. q /1 .
..
... r0 ,0 ... ..
... r1 ,1 ..
... .. r . ....
.... ...
. 0 ....
. .. .
......... ..... ..... ..
q δ(q, r0 ) δ(q, r1 ) δ(q, rc ) τ (q) . ... ......... ............... .............. ..........................................
..... ...................... .... ... . .. ..
....... .. .... .... ........ ..
.. .
. ... .... .. . ..
....
qr0 ,0 qr0 ,0 qr1 ,1 qrc ,0 0 ..
..
..
..
..
..
....
.... ......
.... ..
..
..
..
..
..
.... .... ..
r .. r .. ...... r ... ..
r
qr1 ,1 qr0 ,0 qr1 ,1 qrc ,1 1 0 ....
..
... c
.. ..
...
... .......
.. 1
.
.
.
.
...
.. c
... .. ..... .... .
. .
.
qrc ,0 qr0 ,0 qr1 ,1 qrc ,0 0 ..
..
........
... ....
. ..... ....
.....
....
.
.
.
. .
.
.
........
.. ............ .... .... .. .
qrc ,1 qr0 ,0 qr1 ,1 qrc ,1 1 .............. ....................... r r . ....................................
. ..... .... 1 0 ................. ....
...
...
..... ... ..... ...
. .. . ..
... ... ..
..
...
q r c ,0 /0 .
.
.
.
..
..
... rq c ,1 /1 ...
..
... ... ... . .
. ...
.. .. . ............
............. ......... .. .....
..... ....................
......... . ...... .........
...............................
..... ... ............................ ....
.. r .......... ...
r .
...
... c ... ...
...
c .....
..... . . ..... ..
.................... ....................

Let w = (wn )n≥0 be the word generated by this automaton. Each wn is a finite-state
function of s = s0 s1 · · · sk−1 , a base-3 representation of n. (Because leading zeros are
allowed, this representation is not unique.) We will show that w = c(r0 , r1 , rc , z).
Let i be the number of leading zeros of s. Thus, 0 ≤ i ≤ k and s = 0i si · · · sk−1 . The
automaton starts at state q0,z . Note that δ(q0,z , 0i ) = q0,z , that is, leading zeros cause the
automaton to remain in the initial state. If i = k, then n = 0 and wn = w0 = τ (δ(q0,z , s)) =
τ (δ(q0,z , 0k )) = τ (q0,z ) = z.
Note that δ(q, r0 ) = qr0 ,0 for any q ∈ Q. Thus, if sk−1 = r0 , then wn = τ (δ(q0,z , s)) =
τ (δ(δ(q0,z , s0 s1 · · · sk−2 ), sk−1 )) = τ (δ(δ(q0,z , s0 s1 · · · sk−2 ), r0 )) = τ (qr0 ,0 ) = 0. If the last
ternary digit of n is r0 , then n ≡ r0 (mod 3), that is, n = 3i + r0 for a unique i ∈ N (since
n ≥ 0 and r0 ≥ 0). Thus, w3i+r0 = 0 for all i ∈ N.
Also, since δ(q, r1 ) = qr1 ,1 for any q ∈ Q, we see using the same reasoning that w3i+r1 = 1
for all i ∈ N.
If the last ternary digit of n is rc , that is, if sk−1 = rc , then n = 3i + rc for some
i ∈ N. Since s0 s1 · · · sk−2 sk−1 = s0 s1 · · · sk−2 rc is the base-3 representation of 3i + rc ,
then s0 s1 · · · sk−2 is the base-3 representation of i. Let q 0 = δ(q0,z , s0 s1 · · · sk−2 ) so that
wi = τ (δ(q0,z , s0 s1 · · · sk−2 )) = τ (q 0 ) and w3i+rc = τ (δ(q0,z , s0 s1 · · · sk−2 rc )) = τ (δ(δ(q0,z ,
s0 s1 · · · sk−2 ), rc )) = τ (δ(q 0 , rc )). Note that τ (q 0 ) = τ (δ(q 0 , rc )). Thus, w3i+rc = wi for all
i ∈ N.
By Definition 1, w = c(r0 , r1 , rc , z).

2.2 Fixed Point of a Uniform Morphism

Since a generalized choral sequence is 3-automatic, then (by Cobham’s theorem [1, p. 175])
it is the image, under a coding τ , of a fixed point of a 3-uniform morphism µ. That is,
c = τ (w) where w = µ(w). The following theorem is a generalization of one by Noche [7]
stating the well-known fact that c(0, 2, 1, 0) is the fixed point of the morphism 0 → 001 and
1 → 011 iterated on 0.

5
Proposition 4. A generalized choral sequence c(r0 , r1 , rc , z) = (cn )n≥0 is the fixed point of
the morphism µ iterated on z mapping 0 → a0 a1 a2 and 1 → b0 b1 b2 , where ar0 = br0 = 0,
ar1 = br1 = 1, arc = 0, and brc = 1.

Proof. Recall that z = 0 if r0 = 0, z = 1 if r1 = 0, and z could either be 0 or 1 if rc = 0. If

z = 0, then either r0 = 0 or rc = 0. Either way, a0 = 0 and 0 maps to 0a1 a2 . If z = 1, then
either r1 = 0 or rc = 0. Either way, b0 = 1 and 1 maps to 1b1 b2 . Since a1 , a2 , b1 , b2 ∈ {0, 1},
the morphism µ is prolongable on the letter z. Thus, µ iterated on z has a unique fixed
point µω (z) which we will call w = (wn )n≥0 , where the w’s are letters. Note that w0 = z.
The morphism µ is 3-uniform (because µ(0) = a0 a1 a2 and µ(1) = b0 b1 b2 ). Thus (by a
lemma in [1, p. 174]), µ(wi ) = w3i+0 w3i+1 w3i+2 for all i ∈ N. If r0 = 0, then µ(0) = 0a1 a2
and µ(1) = 0b1 b2 . If r0 = 1, then µ(0) = a0 0a2 and µ(1) = b0 0b2 . If r0 = 2, then
µ(0) = a0 a1 0 and µ(1) = b0 b1 0. In any case, w3i+r0 = 0 for all i ∈ N. Similarly, it can be
seen that w3i+r1 = 1 for all i ∈ N. If rc = 0, then µ(0) = 0a1 a2 and µ(1) = 1b1 b2 . If rc = 1,
then µ(0) = a0 0a2 and µ(1) = b0 1b2 . If rc = 2, then µ(0) = a0 a1 0 and µ(1) = b0 b1 1. In
any case, w3i+rc = wi for all i ∈ N. By Definition 1, c = w = µω (z).

A morphism µ : Σ∗ → Σ∗ is said to be primitive if there exists an integer n ≥ 1 such that

for all a, b ∈ Σ, a occurs in µn (b). The morphism µ in Proposition 4 is primitive because 0
occurs in µn (1) and 1 occurs in µn (0) for any n ∈ Z+ .
An infinite word w is uniformly recurrent if, for every finite subword y of w, there exists
an integer k such that every subword of length k of w contains y.
If µ is a primitive morphism prolongable on z, then µω (z) is uniformly recurrent [1,
Theorem 10.9.5].

Remark 3. A generalized choral sequence is uniformly recurrent.

2.3 Subword Complexity

Let Subw (n) denote the set of all subwords of length n of an infinite word w and pw (n)
denote the subword complexity function of w, the function counting the number of distinct
length-n subwords of w.
For a generalized choral sequence c, Subc (1) = {0, 1}, Subc (2) = {00, 01, 10, 11}, and
Subc (3) = {001, 010, 011, 100, 101, 110}. Thus, pc (1) = 2, pc (2) = 4, and pc (3) = 6. If c
is a type-012 sequence, then Subc (4) = {0010, 0011, 0100, 0110, 1001, 1010, 1011, 1101}; if
c is a type-210 sequence, then Subc (4) = {0010, 0100, 0101, 0110, 1001, 1011, 1100, 1101}.
In either case, pc (4) = 8.

Theorem 4. For a generalized choral sequence c, pc (n) = 2n for all n ∈ Z+ .

Proof. Let Sn = {cj cj+1 · · · cj+n−1 : j ∈ N} be the set of all length-n subwords of a gener-
alized choral sequence c(r0 , r1 , rc , z) = (cn )n≥0 . The set Sn can be partitioned into three
subsets: Sn,0 contains only all the subwords with initial index j ≡ r0 (mod 3), Sn,1 contains
those with j ≡ r1 (mod 3), and Sn,c contains those with j ≡ rc (mod 3). Let |S| denote the
number of elements in a set S. Thus, pc (n) = |Sn | = |Sn,0 | + |Sn,1 | + |Sn,c |.
The proof is by induction: given pc (i) = 2i for all i ≤ k where i, k ∈ Z+ and k is fixed,
we will show that pc (i) = 2i for all i ≤ k + 1. It is easy to see that pc (i) = 2i for all i ≤ 3,
so we only need to look at k ≥ 3.
If c is, say, a type-012 sequence then length-k subwords of c can be visualized as shown
below, where the bottom overline shows a subword from Sk,0 , the middle overline shows

6
a subword from Sk,1 , and the top overline shows a subword from Sk,c . (If c is a type-210
sequence, then the subwords would be of the form · · · 10ci 10 · · ·; the end result is the same.)

If k ≡ 0 (mod 3): ··· 0 1 ci 0 1 ··· ci+m−1 0 1 ci+m 0 1 ···

If k ≡ 1 (mod 3): ··· 0 1 ci 0 1 ··· ci+m−1 0 1 ci+m 0 1 ···
If k ≡ 2 (mod 3): ··· 0 1 ci 0 1 ··· ci+m−1 0 1 ci+m 0 1 ···
Note that 0 < 3m ≤ k since k ≥ 3. Thus, m < k and m + 1 < k.
We are given |Sk,0 | + |Sk,1 | + |Sk,c | = 2k. Consider what happens when k ≡ 0 (mod 3)
and we now look at subwords of length k + 1:

··· 0 1 ci 0 1 ··· ci+m−1 0 1 ci+m 0 1 ···

All the subwords in Sk+1,0 are just the subwords in Sk,0 with the letter 0 concatenated
at the end; the number of subwords remains the same. That is, |Sk+1,0 | = |Sk,0 |. All
the subwords in Sk+1,1 are just those in Sk,1 with a 1 concatenated at the end. Thus,
|Sk+1,1 | = |Sk,1 |.
Note that |Sk,c | is equal to the number of distinct subwords ci ci+1 · · · ci+m−1 , that is,
|Sk,c | = |Sm,c |. Since m < k, |Sm,c | = 2m. Also, |Sk+1,c | is equal to the number of distinct
subwords ci ci+1 · · · ci+m , that is, |Sk+1,c | = |Sm+1,c | = 2(m + 1) = 2m + 2 = |Sm,c | + 2 =
|Sk,c | + 2. (Since m + 1 < k, |Sm+1,c | = 2(m + 1).) Thus, |Sk+1 | = |Sk+1,0 | + |Sk+1,1 | +
|Sk+1,c | = |Sk,0 |+|Sk,1 |+(|Sk,c |+2) = |Sk |+2 = 2k+2 = 2(k+1), that is, pc (k+1) = 2(k+1).
Thus, pc (i) = 2i for all i ≤ k + 1.
Using similar reasoning, we find that when k ≡ 1 (mod 3), |Sk+1,0 | + |Sk+1,1 | + |Sk+1,c | =
|Sk,0 | + (|Sk,1 | + 2) + |Sk,c |; when k ≡ 2 (mod 3), |Sk+1,0 | + |Sk+1,1 | + |Sk+1,c | = (|Sk,0 | +
2) + |Sk,1 | + |Sk,c |. In any case, the end result is the same.
It is easy to see that pc (i) = 2i for all i ≤ 3, i ∈ Z+ . Thus, pc (n) = 2n for all n ∈ Z+ .

2.4 Lyndon Factorization

We now follow Richomme [9]. Words may be ordered lexicographically. Let the alphabet
{0,1} be ordered such that 0 < 1. We say that v ≤ w (or w ≥ v) if and only if either v is
a prefix of w or there exist words x, y, z and letters a, b such that v = xay, w = xbz, and
a < b. We say that v < w (or w > v) if v ≤ w and v 6= w.
Remark 4. [4, p. 82] Let a, b, c, and d be finite words over an ordered alphabet. If a < b
and |a| ≥ |b|, then ac < bd.
We extend lexicographic order to the set of finite or infinite words [6]. We say that v < w
if and only if either v is a prefix of w or there exist words x, y, z and letters a, b such that
v = xay, w = xbz, and a < b. We say that v < w if and only if there exist words x, y, z
and letters a, b such that v = xay, w = xbz, and a < b. We say that v < w if and only if
there exist words x, y, z and letters a, b such that v = xay, w = xbz, and a < b.
Chen, Fox, and Lyndon [4] introduced what they called standard sequences but which are
now called Lyndon words. A Lyndon word is a word that is less than any of its non-empty
proper suffixes. Lyndon words were originally defined as finite words [4] but the definition
was eventually extended to include infinite words [10, Proposition 2.2]. For example, letters
are Lyndon words; 01011 is a finite Lyndon word while 01101 is not; and 01ω = 0111 · · · is
ω
an infinite Lyndon word while (01) = 010101 · · · is not.

7
Lemma 1. [10] An infinite word is an infinite Lyndon word if and only if it has an infinite
number of prefixes which are Lyndon words.
A morphism µ over an ordered alphabet Σ is order-preserving if for all u, v ∈ Σ∗ , u ≤ v
implies µ(u) ≤ µ(v) [9]. For an order-preserving morphism µ, if u < v then µ(u) < µ(v) [9,
Lemma 3.2].
Remark 5. For an order-preserving morphism µ, if u < v then µn (u) < µn (v) for n ∈ Z+ .
A morphism is a Lyndon morphism [9] if it preserves (finite) Lyndon words. A morphism
µ on {0,1} such that 0 < 1 is a Lyndon morphism if and only if µ(0) and µ(1) are Lyndon
words and µ(0) < µ(1) [9, Proposition 4.7].
Remark 6. The morphism µ mapping 0 → 001 and 1 → 011 is a Lyndon morphism.
Lemma 2. For the morphism µ mapping 0 → 001 and 1 → 101, µn (0i 1j ) for n, j ∈ N,
i ∈ Z+ is a Lyndon word.
Proof. We first show that µn (0) for n ∈ N is a Lyndon word (the case where i = 1 and
j = 0).
The words µ(0) = 001 and µ(1) = 101 differ only in the first letter. The words µ2 (0) =
µ(0)µ(0)µ(1) and µ2 (1) = µ(1)µ(0)µ(1) also differ only in the first letter because µ(0) and
µ(1) differ only in the first letter. Continuing this reasoning, it can be seen that µn (0) and
µn (1) for n ∈ Z+ differ only in the first letter and that if µn (0) = 0x then µn (1) = 1x.
Assume that µk (0) is a Lyndon word for some k ∈ Z+ . Now, µk+1 (0) = µk (0)µk (0)µk (1)
= 0x0x1x for some x = a1 a2 · · · an where the a’s are letters. Since µk (0) is a Lyndon word,
it is less than any of its non-empty proper suffixes. Thus, 0x < a1 a2 · · · an , 0x < a2 · · · an ,
and so on up to 0x < an .
Using Remark 4 and starting from 0x < x, we get 0x0x1x < x0x1x. From 0x < a2 · · · an ,
we get 0x0x1x < a2 · · · an 0x1x. Similarly, we go on up to 0x0x1x < an 0x1x.
Clearly, 0x0x1x < 0x1x. Using Remark 4 and starting from 0x < x, we get 0x0x1x <
x1x. Similarly, 0x0x1x < a2 · · · an 1x and so on up to 0x0x1x < an 1x.
Clearly, 0x0x1x < 1x. Using Remark 4 and starting from 0x < x, we get 0x0x1x < x.
Similarly, 0x0x1x < a2 · · · an and so on up to 0x0x1x < an .
Because 0x0x1x is less than any of its non-empty proper suffixes, µk+1 (0) is a Lyndon
word. Now, µ0 (0) = 0 is a Lyndon word. Also, µk (0) is a Lyndon word for k = 1. Thus,
µn (0) is a Lyndon word for any n ∈ N.
The proof that µn (0i 1j ) for n, i, j ∈ Z+ is a Lyndon word is similar, but now we use
µ (0 1 ) = (µk (0))i (µk (1))j = (0x)i (1x)j .
k i j

Lemma 3. For the morphism µ mapping 0 → 001 and 1 → 011, µn (1) > µn+1 (1) for
n ∈ N.
Proof. By Remark 6, µ is a Lyndon morphism. Thus, it is order-preserving [9, Proposition
4.2]. From Remark 5, since 011 < 1, then µn (011) = µn+1 (1) < µn (1) for n ∈ N.
Lemma 4. For the morphism µ mapping 0 → 001 and 1 → 101, µn (01m ) > µn+1 (01m )
for n, m ∈ Z+ .
Proof. The morphism µ is order-preserving because µ(01) < µ(1) [9, Lemma 3.13]. Fix
m ∈ Z+ . From Remark 5, since 001(101)m < 01m , then µn (001(101)m ) = µn+1 (01m ) <
µn (01m ) for n ∈ Z+ .

8
 To factorize a word is to express it as a sequence of subwords. Theorems 5 and 6 describe
Lyndon factorization.

Theorem 5. [4][5, Theorem 5.1.5] Any non-empty finite word w may be uniquely factorized
as a non-increasing finite sequence of finite Lyndon words (`k )0≤k≤n . That is, w = `0 `1 · · · `n
where `0 ≥ `1 ≥ · · · ≥ `n .

For example, the Lyndon factorization of 01011 is (01011) while that of 01101 is
(011)(01).

Theorem 6. [10, Theorem 2.3] Any infinite word may w be uniquely factorized as either
a non-increasing infinite sequence of finite Lyndon words (`k )k≥0 or a non-increasing finite
sequence of finite Lyndon words (`k )0≤k≤n followed by an infinite Lyndon word x ≤ `n .
That is, either w = `0 `1 `2 · · · where `0 ≥ `1 ≥ `2 · · · or w = `0 `1 · · · `n x where `0 ≥ `1 ≥
· · · ≥ `n ≥ x.

For example, the Lyndon factorization of 01ω = 0111 · · · is (0111 · · ·) while that of
ω
(01) = 010101 · · · is (01)(01)(01)· · ·.
We will show that the Lyndon factorizations of the generalized choral sequences are as
follows:
c(0, 2, 1, 0) = (001001011001001011001011011 · · ·)
c(1, 2, 0, 0) = (001001101001001101101001101 · · ·)
c(0, 1, 2, 0) = (01)(001101)(001001101101001101) · · ·
c(2, 1, 0, 0) = (01011)(001011011)(001001011001011011001011011) · · ·
c(1, 2, 0, 1) = (1)(01)(001101)(001001101101001101) · · ·
c(1, 0, 2, 1) = (1)(011)(001011011)(001001011001011011001011011) · · ·
c(2, 1, 0, 1) = (1)(1)(011)(001011011)(001001011001011011001011011) · · ·
c(2, 0, 1, 1) = (1)(1)(011)(01)(001101101)(001101) · · ·

Proposition 5. The sequence c(0, 2, 1, 0) is an infinite Lyndon word.

Proof. By Proposition 4, c(0, 2, 1, 0) = µω (0) where µ maps 0 → 001 and 1 → 011. Note
that µk (0) = c0 c1 · · · c3k −1 for k ∈ N. From Remark 6, µ is a Lyndon morphism. Thus,
since 0 is a Lyndon word, then µk (0) is a Lyndon word for k ∈ N. The proper prefixes
c0 c1 · · · c3k −1 for k ∈ N are all Lyndon words and by Lemma 1, c(0, 2, 1, 0) is an infinite
Lyndon word.

Proposition 6. The sequence c(1, 2, 0, 0) is an infinite Lyndon word.

Proof. By Proposition 4, c(1, 2, 0, 0) = µω (0) where µ maps 0 → 001 and 1 → 101. Note
that µk (0) = c0 c1 · · · c3k −1 for k ∈ N. From Lemma 2, µk (0) is a Lyndon word for k ∈ N. The
proper prefixes c0 c1 · · · c3k −1 for k ∈ N are all Lyndon words and by Lemma 1, c(1, 2, 0, 0)
is an infinite Lyndon word.
Pb
We define n=a f (n) to be 0 if b < a.

Proposition 7. The sequence c(0, 1, 2, 0) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = 01 and `k = µk (`0 ) for k ∈ Z+ , where µ is the morphism
mapping 0 → 001 and 1 → 101.

9
Proof. Let c(0, 1, 2, 0) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}.
From Definition 1, c3i = 0, c3i+1 = 1, and c3i+2 = ci for i ∈ N.
From `k = µk (01) for k ∈ Z+ and given that µ is 3-uniform, we get |`k | = 2 · 3k . Thus,
Pk
`k = wm wm+1 · · · wm+2·3k −1 where m = j=1 2 · 3j−1 . From Definition 1 and Proposition
Pk−1
4, wm+3p = wm0 +p , wm+3p+1 = 0, and wm+3p+2 = 1 where m0 = j=1 2 · 3j−1 and p goes
from 0 to 2 · 3k−1 − 1. Note that as k goes from 1 onwards, the indices m + 3p, m + 3p + 1,
and m + 3p + 2 cover all the integers from 2 onwards.
Thus, the wn ’s are characterized by w0 w1 w2 = 010, wm+3p = wm0 +p , wm+3p+1 = 0, and
Pk Pk−1
wm+3p+2 = 1 where m = j=1 2 · 3j−1 and m0 = j=1 2 · 3j−1 , for all p ∈ N such that
p ≤ 2 · 3k−1 − 1, and for all k ∈ Z+ .
We show that (cn )n≥0 = (wn )n≥0 . Clearly, c0 c1 = w0 w1 , so we now consider n ≥ 2.
Pk
Since m = 2 + j=2 2 · 3j−1 , if follows that m ≡ 2 (mod 3) for k ≥ 1 (that is, for n ≥ 2). If
n ≡ 0 (mod 3), then cn = c3i for some i ≥ 0 and wn = wm+3p+1 for some m and p for k ≥ 1.
If n ≡ 1 (mod 3), then cn = c3i+1 and wn = wm+3p+2 . If n ≡ 2 (mod 3), then cn = c3i+2
and wn = wm+3p .
Now, c3i = 0 = wm+3p+1 and c3i+1 = 1 = wm+3p+2 , so all we need to show is that
c3i+2 = wm+3p . Consider the letter cn = c3i+2 where i ≥ 0 and n ≥ 2. It is part of a subword
`k for some k ≥ 1, that is, since `k = wm wm+1 · · · wm+2·3k −1 , then m ≤ 3i+2 ≤ m+2·3k −1.
But 3i+2 ≡ 2 (mod 3), so m ≤ 3i+2 ≤ m+2·3k . Thus, (m−2)/3 ≤ i ≤ (m+2·3k −2)/3. But
Pk Pk Pk−1
(m − 2)/3 = ( j=2 2 · 3j−1 )/3 = j=2 2 · 3j−2 = j=1 2 · 3j−1 = m0 and (m + 2 · 3k − 2)/3 =
m0 + 2 · 3k−1 . Thus, m0 ≤ i ≤ m0 + 2 · 3k−1 , and ci is part of the subword `k−1 , that is,
ci = wm0 +p . Finally, c3i+2 = ci = wm0 +p = wm+3p .
From Lemma 2, `k = µk (01) is a Lyndon word for k ∈ N. Clearly, `0 = 01 > `1 =
001101. From Lemma 4, `1 > `2 > `3 > · · ·.

The proof of Proposition 7 is relatively detailed. For brevity, we omit some details in
the remaining proofs.

Proposition 8. The sequence c(2, 1, 0, 0) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = 01011 and `k = µk (011) for k ∈ Z+ , where µ is the
morphism mapping 0 → 001 and 1 → 011.

Proof. Let c(2, 1, 0, 0) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}. Now,
c3i = ci , c3i+1 = 1, c3i+2 = 0, and c0 = 0 for i ∈ N. The wn ’s are characterized by
w0 w1 w2 w3 w4 = 01011, wm+3p = 0, wm+3p+1 = wm0 +p , and wm+3p+2 = 1 where m =
Pk Pk−1
2 + j=1 3j and m0 = 2 + j=1 3j , for all p ∈ N such that p ≤ 3k − 1, and for all k ∈ Z+ .
Clearly, c0 c1 c2 c3 c4 = w0 w1 w2 w3 w4 . Because m ≡ 2 (mod 3), it follows that c3i = wm+3p+1 ,
c3i+1 = wm+3p+2 , and c3i+2 = wm+3p .
Subword `0 is a Lyndon word and so is each `k for k ∈ Z+ because µ is a Lyndon
morphism (Remark 6) and 011 is a Lyndon word. Clearly, `0 = 01011 > `1 = 001011011.
Note that `k = µk+1 (1). From Lemma 3, `1 > `2 > `3 > · · ·.

Proposition 9. The sequence c(1, 2, 0, 1) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = 1, `1 = 01, and `k = µk−1 (`1 ) for k ≥ 2, where µ is the
morphism mapping 0 → 001 and 1 → 101.

Proof. Let c(1, 2, 0, 1) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}. Now,
c3i = ci , c3i+1 = 0, c3i+2 = 1, and c0 = 1 for i ∈ N. The wn ’s are characterized by

10
w0 w1 w2 w3 = 1010, wm+3p = wm0 +p , wm+3p+1 = 0, and wm+3p+2 = 1 where m = 1 +
Pk Pk−1
j=2 2 · 3
j−2
and m0 = 1 + j=2 2 · 3j−2 , for all p ∈ N such that p ≤ 2 · 3k−2 − 1, and for
Pk
all k ≥ 2. Clearly, c0 c1 c2 = w0 w1 w2 . Because m = 3 + j=3 2 · 3j−2 and m ≡ 0 (mod 3),
it follows that c3i = wm+3p , c3i+1 = wm+3p+1 , and c3i+2 = wm+3p+2 .
Subword `0 is a Lyndon word and, by Lemma 2, so is each `k+1 = µk (01) for k ∈ N.
Clearly, `0 = 1 > `1 = 01 > `2 = 001101. From Lemma 4, `2 > `3 > `4 > · · ·.

Proposition 10. The sequence c(1, 0, 2, 1) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = 1 and `k = µk (`0 ) for k ∈ Z+ , where µ is the morphism
mapping 0 → 001 and 1 → 011.

Proof. Let c(1, 0, 2, 1) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}. Now, c3i =
1, c3i+1 = 0, and c3i+2 = ci for i ∈ N. The wn ’s are characterized by w0 = 1, wm+3p = 0,
Pk Pk−1 j−1
wm+3p+1 = wm0 +p , and wm+3p+2 = 1 where m = j=1 3
j−1
and m0 = j=1 3 , for
k−1 +
all p ∈ PN such that p ≤ 3 − 1, and for all k ∈ Z . Clearly, c0 = w0 . Because
k
m = 1 + j=2 3j−1 and m ≡ 1 (mod 3), it follows that c3i = wm+3p+2 , c3i+1 = wm+3p , and
c3i+2 = wm+3p+1 .
Subword `0 is a Lyndon word and so is each `k for k ∈ Z+ because µ is a Lyndon
morphism (Remark 6). From Lemma 3, `0 > `1 > `2 > · · ·.

Proposition 11. The sequence c(2, 1, 0, 1) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = `1 = 1 and `k = µk−1 (`1 ) for k ≥ 2, where µ is the
morphism mapping 0 → 001 and 1 → 011.

Proof. Let c(2, 1, 0, 1) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}. Now,
c3i = ci , c3i+1 = 1, c3i+2 = 0, and c0 = 1 for i ∈ N. The wn ’s are characterized by
Pk
w0 w1 = 11, wm+3p = 0, wm+3p+1 = wm0 +p , and wm+3p+2 = 1 where m = 1 + j=2 3j−2
Pk−1
and m0 = 1 + j=2 3j−2 , for all p ∈ N such that p ≤ 3k−2 − 1, and for all k ≥ 2. Clearly,
Pk
c0 c1 = w0 w1 . Because m = 2+ j=3 3j−2 and m ≡ 2 (mod 3), it follows that c3i = wm+3p+1 ,
c3i+1 = wm+3p+2 , and c3i+2 = wm+3p .
Subwords `0 and `1 are Lyndon words and so is each `k for k ≥ 2 because µ is a Lyndon
morphism (Remark 6). Note that `0 = `1 = 1. From Lemma 3, `1 > `2 > `3 > · · ·.

Proposition 12. The sequence c(2, 0, 1, 1) is an infinite non-increasing sequence of finite

Lyndon words (`k )k≥0 with `0 = `1 = 1, `2 = 011, `3 = 01, and `k = µ(`k−2 ) for k ≥ 4,
where µ is the morphism mapping 0 → 001 and 1 → 101.

Proof. Let c(2, 0, 1, 1) = (cn )n≥0 and (`k )k≥0 = (wn )n≥0 where cn , wn ∈ {0, 1}. Now,
c3i = 1, c3i+1 = ci , and c3i+2 = 0 for i ∈ N.
Note that for k = 2j + 2 and j ∈ Z+ , k is an even number greater than or equal to 4.
For k = 2j + 3 and j ∈ Z+ , k is an odd number greater than or equal to 5.
From `2j+2 = µj (011) for j ∈ Z+ , we get |`2j+2 | = 3j+1 . Thus, `2j+2 = wq wq+1 · · ·
Pj
wq+3j+1 −1 where q = 2 + i=1 (3i + 2 · 3i−1 ). Also, wq+3p = wq0 +p , wq+3p+1 = 0, and
Pj−1
wq+3p+2 = 1 where q 0 = 2 + i=1 (3i + 2 · 3i−1 ) and p goes from 0 to 3j − 1.
From `2j+3 = µj (01) for j ∈ Z+ , we get |`2j+3 | = 2 · 3j . Thus, `2j+3 = wr wr+1 · · ·
Pj
wr+2·3j −1 where r = 5 + i=1 (3i+1 + 2 · 3i−1 ). Also, wr+3p = wr0 +p , wr+3p+1 = 0, and
Pj−1
wr+3p+2 = 1 where r0 = 5 + i=1 (3i+1 + 2 · 3i−1 ) and p goes from 0 to 2 · 3j−1 − 1.

11
 Pj
Now, c0 c1 c2 c3 c4 c5 c6 = w0 w1 w2 w3 w4 w5 w6 . Note that q = 7 + i=2 (3i + 2 · 3i−1 ) and
Pj
r = 16 + i=2 (3i+1 + 2 · 3i−1 ) so that q ≡ r ≡ 1 (mod 3). Thus, it follows that c3i =
wq+3p+2 = wr+3p+2 , c3i+1 = wq+3p = wr+3p , and c3i+2 = wq+3p+1 = wr+3p+1 .
Subwords `0 and `1 are Lyndon words and, by Lemma 2, so is each `2j+2 = µj (011) and
each `2j+3 = µj (01) for j ∈ N. Clearly, `0 = `1 = 1 > `2 = 011. Note that µj (01)µj (1) =
µj (011) > µj (01) for j ∈ N. Thus, `2j+2 > `2j+3 for j ∈ N, that is, `2 > `3 , `4 > `5 ,
and so on. The morphism µ is order-preserving because µ(01) < µ(1) [9, Lemma 3.13].
From Remark 5, since 01 > 001101101, then µj (01) > µj (001101101) = µj+1 (011) for
j ∈ N. Thus, `2j+3 > `2j+4 for j ∈ N, that is, `3 > `4 , `5 > `6 , and so on. We then get
`2 > `3 > `4 > `5 > `6 > · · ·.

Acknowledgment
I thank an anonymous colleague for pointing out the connection to Toeplitz words.

References
[1] Jean-Paul Allouche and Jeffrey Shallit. Automatic Sequences: Theory, Applications,
Generalizations. Cambridge University Press, 2003.
[2] J. Berstel and J. Karhumäki. Combinatorics on words—a tutorial. Bulletin of the
European Association of Theoretical Computer Science, (79):178–228, 2003.
[3] Julien Cassaigne and Juhani Karhumäki. Toeplitz words, generalized periodicity and
periodically iterated morphisms. European Journal of Combinatorics, 18:497–510, 1997.

[4] K. T. Chen, R. H. Fox, and R. C. Lyndon. Free differential calculus, IV. The quotient
groups of the lower central series. Annals of Mathematics, 68(1):81–95, 1958.
[5] M. Lothaire. Combinatorics on Words. Cambridge Mathematical Library. Cambridge
University Press, 2nd edition, 1997.

[6] Guy Melançon. Lyndon factorization of infinite words. In STACS 96, volume 1046/1996
of Lecture Notes in Computer Science, pages 147–154. Springer Berlin/Heidelberg,
1996.
[7] Joel Reyes Noche. On Stewart’s choral sequence Gibón, 8(1):1–5, 2008.

[8] Joel Reyes Noche. Generalized choral sequences. Matimyás Matematika, 31(1–3):25–28,
2008.
[9] G. Richomme. Lyndon morphisms. Bulletin of the Belgian Mathematical Society,
10:761–785, 2003.
[10] Rani Siromoney, Lisa Mathew, V. R. Dare, and K. G. Subramanian. Infinite Lyndon
words. Information Processing Letters, 50(2):101–104, 1994.
[11] Ian Stewart. Mathematical recreations: The never-ending chess game. Scientific Amer-
ican, 273(4):158, 160, October 1995.

12