LXXVI.2 (1996)
A certain power series associated with a Beatty sequence
by
Takao Komatsu (North Ryde, N.S.W.)
0. Introduction. We consider the function (1) f (θ, φ; x, y) =
X
∞ k=1X
1≤m≤kθ+φ
x
ky
m. Putting y = 1 entails that
(2) f (θ, φ; x, 1) =
X
∞ k=1[kθ + φ]x
k.
The sequence {[kθ + φ]}
∞k=1, which appears in this power series, is called a Beatty sequence. In that context it is natural to consider the sequence of differences
(3) {[(k + 1)θ + φ] − [kθ + φ]}
∞k=1.
The function f (θ, 0; x, y) and the sequence {[(k + 1)θ] − [kθ]}
∞k=1in the homogeneous case have been treated independently by many authors (see e.g. [1], [7], [8] and [2], [10] respectively). The inhomogeneous case of (3) has also been treated by several authors (see e.g. [3]–[5]).
In 1992 Nishioka, Shiokawa and Tamura [9] described the sequence (3) in the inhomogeneous case by using the characteristic properties of (1), but their result (Theorem 3 of [9]) is incorrect. The arguments only hold when φ is an integer or when b
n= 1 for all positive integers n (for the definition of b
nsee the next section).
In this paper we base on the arguments corrected by the author [6] and describe the sequence (3) completely in the new form. Of course, Theorem 2 of [9] holds because φ = 0. Lemmas 2 and 3 of [9], which were used to prove Theorem 3 of [9], work and have meaning only in the original context. After
1991 Mathematics Subject Classification: Primary 11B83.
The author thanks Dr. Peter A. B. Pleasants for his useful advice.
[109]
correcting the arguments properly, both lemmas are no longer useful and we need different new arguments to obtain a correction to Theorem 3 of [9].
1. Preliminary remarks and notation. Throughout this paper θ > 0 is irrational and kθ + φ is never integral for any positive integer k. As usual, θ = [a
0, a
1, a
2, . . .] denotes the continued fraction expansion of θ, where
θ = a
0+ θ
0, a
0= [θ],
1/θ
n−1= a
n+ θ
n, a
n= [1/θ
n−1] (n = 1, 2, . . .).
The nth convergent p
n/q
n= [a
0, a
1, . . . , a
n] of θ is then given by the recur- rence relations
p
n= a
np
n−1+ p
n−2(n = 0, 1, . . .), p
−2= 0, p
−1= 1, q
n= a
nq
n−1+ q
n−2(n = 0, 1, . . .), q
−2= 1, q
−1= 0.
One now expands φ in terms of the sequence {θ
0, θ
1, . . .} by setting φ = b
0− φ
0, b
0= dφe,
φ
n−1/θ
n−1= b
n− φ
n, b
n= dφ
n−1/θ
n−1e (n = 1, 2, . . .).
Furthermore, the quantities s
nand t
nare defined by s
n=
X
n ν=0b
νp
ν−1(n = 0, 1, . . .), s
n= 0 (n < 0),
t
n= X
n ν=0b
νq
ν−1(n = 0, 1, . . .), t
n= 0 (n < 0).
We can assume 0 < θ, φ < 1 without loss of generality. As shown in Sections 1 and 2 of [6],
f (θ, φ; x, y) = X
∞ n=1(−1)
n−1x
tny
sn(1 − x
qny
pn)(1 − x
qn−1y
pn−1) , which yields
X
∞ k=0([(k + 1)θ + φ] − [kθ + φ])x
k= 1 x lim
n→∞
P
n∗(x), |x| < 1.
Here, P
n∗(x) is defined recursively by
P
n∗(x) = A
∗n(x)P
n−1∗(x) + x
bnqn−1P
n−2∗(x) (n ≥ 1) with P
−1∗(x) = 1, P
0∗(x) = 0, where
A
∗n(x) = 1 − x
qn− x
bnqn−1(1 − x
qn−2)
1 − x
qn−1(n ≥ 1).
Let P
n∗(x) = d
1x + d
2x
2+ d
3x
3+ . . . be the power series expansion.
Put P
n∗= d
1d
2d
3. . . , which is the string of coefficients of the power series beginning from that of x
1.
Define
Γ
n= {a
3− b
3, a
4− b
4, . . . , a
n− b
n} (n ≥ 3)
and write π
n= a
n− b
nif a
n> b
n, $
n= a
n− b
nif a
n≥ b
n—to account for the case when the entry 0 is permitted.
We consider the following situations:
Γ
n∈ O if Γ
n= $
3$
4. . . $
n,
Γ
n∈ A
k,l(or simply A) if Γ
nends in (−1)0
2k−1π
n−l$ |
n−l+1{z . . . $
n}
l
, Γ
n∈ B
k(or simply B) if Γ
nends in (−1)0
2k−1,
Γ
n∈ C
k(or simply C) if Γ
nends in (−1)0
2k−2(k ≥ 2), Γ
n∈ C
1if Γ
nends in π
n−l−1$
n−l. . . $
n−1| {z }
l
(−1), Γ
n∈ D
k(or simply D) if Γ
nends in (−1)0
2k−2(−1),
where k is a positive integer and l is a non-negative integer. (Note that Γ
3∈ O if a
3≥ b
3and Γ
3∈ C if a
3= b
3− 1.)
Let β
n= t
n− q
n−1− b
1+ 1 = (b
n− 1)q
n−1+ b
n−1q
n−2+ . . . + b
2q
1+ 1.
We define the words u, v and ∆
nas u = 0 . . . 0 | {z }
a1−1
1, v = 0 . . . 0 | {z }
b1−1
1 and ∆
n= 0 . . . 0 | {z }
βn−2
(−1)
n+1(−1)
n.
2. Main results. Our main result, which replaces the alleged Theorem 3 of [9], is
Theorem. Let θ be irrational with 0 < θ, φ < 1. Then either {[(k + 1)θ + φ] − [kθ + φ]}
∞k=0= lim
n→∞
P
n∗or
{[(k + 1)θ + φ] − [kθ + φ]}
∞k=1= lim
n→∞
0 . . . 01 | {z }
b1−1
w
n.
Here (w
n) is the sequence of words of respective lengths q
n, with letters 0 or 1, given inductively by
w
1= u, w
2= w
1b2−10w
1a2−b2+1, w
n= w
n−1cnw
n−2w
an−1n−cn, where
c
n=
b
n+ 1 if Γ
n−1∈ B and a
n> b
n,
0 if Γ
n−1∈ C,
1 if Γ
n−1∈ D,
min(a
n, b
n) otherwise.
R e m a r k. By Lemma 1 below, a
n≤ b
nif Γ
n−1∈ C, D. Other possible cases are limited to Γ
n−1∈ B and a
n= b
n, and Γ
n−1∈ O, A.
The Theorem is a direct consequence of the following Proposition, which describes P
n∗. From now on the underline means to add (−1) to the last one part in that word. For example, if W = 00101, then W = 00100. If W = 00100, then W = 0010(−1), W
2= 00100 0010(−1) and (W )
2= 0010(−1)0010(−1).
Proposition. For every n = 1, 2, . . . , we have P
n∗= vw
nw
00n. Here,
|w
n| = q
nfor every n, and w
1= u, w
2= u
b2−10u
a2−b2+1; w
100and w
002are empty; and w
nand w
n00(n ≥ 3) are determined as follows:
(1) If n = 3 and Γ
n−1∈ O or A (n ≥ 4), then
w
n= w
bn−1nw
n−2w
n−1an−bnand w
00n= empty if a
n≥ b
n, w
n= w
an−1nw
n−2and w
00n= ∆
n−1if a
n= b
n− 1.
(2) If Γ
n−1∈ B (n ≥ 5), then
w
n= w
n−1bn+1w
n−2w
an−1n−bn−1and w
n00= empty if a
n> b
n, w
n= w
n−1anw
n−2and w
n00= ∆
n−2k−1if a
n= b
n, (k = 1 if Γ
n−2∈ D).
(3) If Γ
n−1∈ C (n ≥ 4), then w
n= w
n−2w
an−1nand w
n00=
empty if a
n= b
n,
∆
n−1if a
n= b
n− 1.
(4) If Γ
n−1∈ D (n ≥ 5), then
w
n= w
n−1w
n−2w
n−1an−1and w
00n=
empty if a
n= b
n,
∆
n−1if a
n= b
n− 1.
We detail the initial cases n = 1, 2, 3 here. We notice that A
∗n(x) = 1 + x
qn−1+ . . . + x
(bn−1)qn−1+
x
bnqn−1+qn−2(1 + x
qn−1+ . . . + x
(an−bn−1)qn−1) if a
n> b
n,
0 if a
n= b
n,
−x
qnif a
n= b
n− 1.
Since P
1∗(x) = x
b1, we have P
1∗= v = vu. Since P
2∗(x) = x
b1A
∗2(x), we have
P
2∗=
vu
b2−10u
a2−b2= vu
b2−10u
a2−b2u if a
2> b
2,
vu
b2−1= vu
b2−10u if a
2= b
2,
vu
b2−1(−1) if a
2= b
2− 1.
Thus, w
2= u
b2−10u
a2−b2+1. Since P
3∗(x) = x
b1(A
∗3(x)A
∗2(x) + 1), we have
P
3∗=
vw
b23uw
a23−b3−1u
b2−10u
a2−b2if a
3> b
3, vw
b23= vw
b23u if a
3= b
3, vw
a2300 . . . 00 | {z }
a1+β2−2
(−1)1 if a
3= b
3− 1.
3. Lemmas. We need the following lemmas to complete the proof of the Proposition.
Lemma 1. (1) If Γ
n−1∈ C or D, then a
n≤ b
n. (2) If Γ
n−1∈ B, then a
n≥ b
n.
Fig. 1
P r o o f. We prove (1) and (2) together. Notice that as long as a
i≥ b
ifor i = 3, 4, . . . , always Γ
i∈ O. Suppose that a
3≥ b
3, . . . , a
n−2≥ b
n−2and a
n−1= b
n−1− 1 for some fixed n ≥ 4, which means Γ
n−1∈ C
1. From the definition we have
θ
n−1+ φ
n−1=
1 θ
n−2− a
n−1+
b
n−1− φ
n−2θ
n−2= 1 − φ
n−2θ
n−2+ 1 > 1 or
0 < 1
θ
n−1− φ
n−1θ
n−1< 1.
Therefore,
a
n=
1 θ
n−1≤ b
n=
φ
n−1θ
n−1. The case Γ
n−1∈ C
1is proved.
If a
n= b
n, that is, Γ
n∈ B
1, we get θ
n+ φ
n=
1
θ
n−1− a
n+
b
n− φ
n−1θ
n−1= 1
θ
n−1− φ
n−1θ
n−1< 1.
Therefore,
a
n+1=
1 θ
n≥ b
n+1=
φ
nθ
n.
The case Γ
n∈ B
1is proved. If a
n+1> b
n+1, Γ
n+1∈ A
1,0. If a
n+1= b
n+1, Γ
n+1∈ C
2.
If a
n< b
n, that is, Γ
n∈ D
1, similarly to the case Γ
n−1∈ C
1, we get θ
n+ φ
n> 1 and a
n+1≤ b
n+1. The case Γ
n∈ D
1is proved. If a
n+1= b
n+1, Γ
n+1∈ B
1. If a
n+1< b
n+1, Γ
n+1∈ D
1again.
Now, we consider each case for an arbitrary positive integer k (≥ 2). Let Γ
i−1∈ C
kfor some integer i (≥ 6). Since Γ
i−2∈ B
k−1,
1
θ
i−2− φ
i−2θ
i−2> 1.
Hence,
θ
i−1+ φ
i−1=
1
θ
i−2− a
i−1+
b
i−1− φ
i−2θ
i−2= 1 − φ
i−2θ
i−2> 1 or
0 < 1
θ
i−1− φ
i−1θ
i−1< 1.
Therefore, a
i≤ b
i. If a
i= b
i, Γ
i∈ B
k. If a
i< b
i, Γ
i∈ D
k. The general case Γ
i∈ B
kor Γ
i∈ D
kis treated similarly.
The situation in Lemma 1 is illustrated in Figure 1.
Lemma 2. (1) If Γ
n−1∈ O or A, then β
n−1≤ q
n−1. (2) If Γ
n−2∈ C
kand Γ
n−1∈ B
k, then β
n−2k−1≤ q
n−3. (3) If Γ
n−2∈ D
kand Γ
n−1∈ B
1, then β
n−3≤ q
n−2+ q
n−3. (4) If Γ
n−1∈ C
k, then β
n−2k≤ q
n−2.
(5) If Γ
n−1∈ D
k, then β
n−2≤ q
n−1+ q
n−2. P r o o f. If a
i≥ b
ifor any i = 3, 4, . . . , n, then
β
n= (b
n− 1)q
n−1+ b
n−1q
n−2+ . . . + b
3q
2+ b
2q
1+ 1
≤ (a
n− 1)q
n−1+ a
n−1q
n−2+ . . . + a
3q
2+ (a
2+ 1)q
1+ 1 = q
n.
The other cases will be proved inductively in the proof of the Proposition.
Lemma 3. (1) If Γ
n−1∈ O or A and a
n≥ b
n, then w
nw
n−1− w
n−1w
n= 0 . . . 0 | {z }
qn−1
∆
n.
(2) If Γ
n−1∈ O or A and a
n< b
n, then
−w
nw
n−1+ w
n−1w
n= 0 . . . 0 | {z }
qn
∆
n−1.
(3) If Γ
n−1∈ B
kand a
n> b
n, then
−w
nw
n−1+ w
n−1w
n= 00 . . . 00 | {z }
(bn+1)qn−1+qn−2
∆
n−2k−1.
(4) If Γ
n−1∈ B
kand a
n= b
n, then
−w
nw
n−1+ w
n−1w
n= 0 . . . 0 | {z }
qn
∆
n−2k−1.
(5) If Γ
n−1∈ C
kand a
n≤ b
n, then
w
nw
n−1− w
n−1w
n= 0 . . . 0 | {z }
qn−1
∆
n−2k.
(6) If Γ
n−1∈ D
kand a
n≤ b
n, then
w
nw
n−1− w
n−1w
n= 0 . . . 0 | {z }
qn−1
∆
n−2.
P r o o f. Here, we shall prove only the case when Γ
n−1∈ O and a
n≥ b
n. The others will be proved inductively in the proof of the Proposition. Both w
00n−2and w
00n−1are empty by induction. Set X = x
qn−1for brevity. If a
n> b
n, then
P
n∗(x) = (1 + X + . . . + X
bn−1+ X
bnx
qn−2(1 + X + . . . + X
an−bn−1))
× P
n−1∗(x) + X
bnP
n−2∗(x), which yields
P
n∗= vw
bn−1nw
n−2w
an−1n−bn.
If a
n= b
n, we have P
n∗(x) = (1+X +. . .+X
bn−1)P
n−1∗(x)+X
bnP
n−2∗(x), yielding P
n∗= vw
bn−1nw
n−2. Hence, we have w
n= w
bn−1nw
n−2w
an−1n−bn. There- fore, if n is odd, then
w
nw
n−1− w
n−1w
n= w
bn−1nw
n−2w
an−1n−bnw
n−1− w
n−1w
bn−1nw
n−2w
n−1an−bn= 0 . . . 0 | {z }
bnqn−1
(w
n−2w
n−1− w
n−1w
n−2)
= 0 . . . 0 | {z }
bnqn−1
(w
n−2w
n−2bn−1w
n−3w
an−2n−1−bn−1− w
bn−2n−1w
n−3w
an−2n−1−bn−1w
n−2)
= 00 . . . 00 | {z }
bnqn−1+bn−1qn−2
(w
n−2w
n−3− w
n−3w
n−2) = . . .
= 000 . . . 000 | {z }
bnqn−1+bn−1qn−2+...+b3q2
(w
1w
2− w
2w
1)
= 000 . . . 000 | {z }
bnqn−1+bn−1qn−2+...+b3q2
(uu
b2−10u
a2−b2+1− u
b2−10u
a2−b2+1u)
= 0000 . . . 0000 | {z }
bnqn−1+bn−1qn−2+...+b3q2+(b2−1)q1
(0 . . . 0 | {z }
q1−1
10 − 0 0 . . . 0 | {z }
q1−1
1)
= 00 . . . 00 | {z }
qn−1+βn−2
1(−1).
If n is even, then w
1and w
2above are interchanged, so we obtain 00 . . . 00
| {z }
qn−1+βn−2
(−1)1.
4. Proof of Proposition. We prove the Proposition together with Lemmas 2 and 3. We write [B
k−1C
kD
k] for brevity when Γ
n−3∈ B
k−1, Γ
n−2∈ C
kand Γ
n−1∈ D
k. From Lemma 1 all cases are classified into one of O, A, B, C, D and the number of patterns like [B
k−1C
kD
k] is limited.
We denote by S the sequence of the patterns of [Γ
n−3, Γ
n−2, Γ
n−1].
4.1. Case Γ
n−1∈ O. The only possible pattern is [OOO]. Then, both w
00n−2and w
00n−1are empty. As we have already seen in the proof of Lemma 3,
w
n= w
bn−1nw
n−2w
n−1an−bnand w
n00= empty if a
n≥ b
n.
If a
n= b
n− 1, by using Lemma 3(1) with Γ
n−3∈ O and β
n−1= (b
n−1− 1)q
n−2+ q
n−3+ β
n−2we have P
n∗(x) = (1 + X + . . . + X
bn−1− x
qn)P
n−1∗(x) + X
bnP
n−2∗(x), which yields
P
n∗= vw
bn−1nw
n−2− 0 . . . 0 | {z }
qn
vw
n−1= v w
bn−1n| {z }
firstqn
w
n−2− 0 . . . 0 | {z }
b1+qn
w
n−1= vw
an−1nw
n−200 . . . 00 | {z }
(bn−1−1)qn−2
(w
n−3w
n−2− w
n−2w
n−3)
= vw
an−1nw
n−2∆
n−1.
Therefore, w
n= w
an−1nw
n−2and w
00n= ∆
n−1.
Using the results here and Lemma 3(1) with Γ
n−2∈ O, we obtain Lemma 3(2), that is,
−w
nw
n−1+ w
n−1w
n= − w
an−1nw
n−2w
n−1+ w
an−1nw
n−1w
n−2= 0 . . . 0 | {z }
anqn−1
0 . . . 0
| {z }
qn−2
∆
n−1= 0 . . . 0 | {z }
qn
∆
n−1.
As long as a
i≥ b
ifor i = 3, 4, . . . , there is no other pattern. But once a
n< b
nfor some n, the pattern [OOC
1] follows [OOO] in the sequence S and the loop starts. The situation after this can be seen in Figure 2. “//”
stands for a
i≥ b
i, “/” for a
i> b
i, “−” for a
i= b
i, “\” for a
i< b
i. Once we encounter C
1(or B
1, D
1) again, the situation after that is the same as the situation after the first C
1(or B
1, D
1).
We shall indicate the loop in all patterns according to the class of Γ
n−1. Some initial cases are omitted, but it is easy to see that they are special cases of the general ones and they are included in them.
4.2. Case Γ
n−1∈ C. From Lemma 1 the possible patterns are [OOC
1], [B
kA
k,0C
1], [A
k,l−1A
k,lC
1], [C
kB
kC
k+1], [D
kB
1C
2].
• [OOC
1]. This follows [OOO] in the sequence S.
Since Γ
n−1= $
3. . . $
n−2(−1) (Γ
3= (−1) when n = 4), w
00n−2is empty and w
n−1= w
an−2n−1w
n−3and w
00n−1= ∆
n−2.
If a
n= b
n, we have P
n∗(x) = (1+X +. . .+X
an−1)P
n−1∗(x)+X
anP
n−2∗(x).
Since the string of coefficients of
(1 + X + . . . + X
bn−1) × x
b1X((−1)
n−1x
βn−2−1+ (−1)
n−2x
βn−2) is
0 . . . 0
| {z }
b1+βn−2
( 0 . . . 0 | {z }
qn−1−2
(−1)
n−1(−1)
n)
an, we obtain
P
n∗= vw
n−1anw
n−2+ 0 . . . 0 | {z }
b1+βn−2
( 0 . . . 0 | {z }
qn−1−2
(−1)
n−1(−1)
n)
an.
From Lemma 2(1) with Γ
n−2∈ O we get 0 < β
n−2≤ q
n−2. Therefore, w
n00is empty and the conclusion of Lemma 2(4) is satisfied.
If a
n= b
n− 1, we have P
n∗(x) = (1 + X + . . . + X
bn−1− x
qn)P
n−1∗(x) + X
bnP
n−2∗(x). Since the string of coefficients of
(1 + X + . . . + X
bn−1− x
qn) × x
b1X((−1)
n−1x
βn−2−1+ (−1)
n−2x
βn−2) is
0 . . . 0
| {z }
b1+βn−2
( 0 . . . 0 | {z }
qn−1−2
(−1)
n−1(−1)
n)
bn0 . . . 0 | {z }
qn−2−2
(−1)
n(−1)
n−1,
Fig. 2