LXXIII.3 (1995)
Divisor problems of 4 and 3 dimensions
by
Hong-Quan Liu (Harbin)
Introduction. For a positive integer n, let the divisor functions d(4, 5, 6, 7; n), d(1, 1, 2, 2; n) and d(1, 1, 2; n) be defined as in [3], [4]. In this paper we will sharpen our former arguments by proving the following new results regarding the errors of distribution of these divisor functions. We have (ε and x are as usual):
Theorem 1.
X
n≤x
d(4, 5, 6, 7; n) = main terms + O(x
87/869+ε).
Theorem 2.
X
n≤x
d(1, 1, 2, 2; n) = main terms + O(x
7/19+ε).
Theorem 3.
X
n≤x
d(1, 1, 2; n) = main terms + O(x
29/80+ε).
Let Q
4(x) be the number of 4-full numbers not exceeding x, let τ (G) be the number of direct factors of a finite Abelian group G, and t(G) be the number of unitary factors of G, and T (x) = P
τ (G), T
∗(x) = P t(G), where the summations are over all G of order not exceeding x. Then, as in [3], [4], we have
Corollary 1. Q
4(x) = main terms + O(x
87/869+2ε).
Corollary 2. T (x) = main terms + O(x
7/19+2ε).
Corollary 3. T
∗(x) = main terms + O(x
29/80+2ε).
Note that 87/869 = 0.1001150 . . . , which improves the corresponding exponent 6/59 = 0.10169 . . . established in Theorem 2 of [3], and 7/19 = 0.3684 . . . , 29/80 = 0.3625 improve respectively the exponents 0.4 and 77/208 = 0.3701 . . . given by Theorems 2 and 1 of [4].
[249]
In demonstrating these theorems, Theorem 3 of [1] will again play an important role. We will also need to combine other tools existing in papers [2] to [5] of the author. Needless to say, many tedious and elementary cal- culations will emerge in our treatment, which is inherent in such divisor problems. We will do our best to avoid redundancy.
1. Proof of Theorem 1. We recall a useful lemma (Theorem 3 of [1]).
Lemma 1.1. Let H ≥ 1, X ≥ 1, Y ≥ 1000; let α, β and γ be real numbers with αγ(γ − 1)(β − 1) 6= 0, and let A > C(α, β, γ) > 0 and f (h, x, y) = Ah
αx
βy
γ. Define
S(H, X, Y ) = X
(h,x,y)∈D
C
1(h, x)C
2(y)e(f (h, x, y)),
where D is a region contained in the rectangle {(h, x, y) | h ∼ H, x ∼ X, y ∼ Y } such that for any fixed pair (h
0, x
0), the intersection D∩{(h
0, x
0, y) | y ∼ Y } has at most O(1) segments. Also, suppose that |C
1(h, x)| ≤ 1, |C
2(y)|
≤ 1 and F = AH
αX
βY
γY . Then, for L = ln((A + 1)HXY + 2) and M = max(1, F Y
−2),
L
−3S(H, X, Y )
22√
(HX)
19Y
13F
3+ HXY
5/8(1 + Y
7F
−4)
1/16+
32√
(HX)
29Y
28F
−2M
5+ √
4(HX)
3Y
4M .
We stress that the condition F Y is needed in the proof of this lemma.
We adopt the notations introduced in [3]. In particular, from (7) of [3], we have the following estimate:
Φ(H; N ) H
−1(N
32H
−1G
−1)
1/2X
h∼H
X
5
g
1(n
1)g
2(n
2)g
3(u)e(g) (1.1)
+ N
1(HG)
1/2ln x + x
13/132.
From (23) to (31) of [3], and the estimates on p. 175 there (η = ε/8), x
−ηS(a, b, c, d; N )
33√
x
3N
112N
22+
888√
x
73N
1377N
2162(1.2)
+
76√
x
6N
134N
219+
289√
x
24N
1121N
251+ x
0.1. We need two more estimates for S(a, b, c, d; N ). First we employ Lemma 1.1 to the triple summation over n
1, n
2and u in (1.1), with the choice (h, x, y)
= (n
1, n
2, u). Note that U ∼ = HG/N
3; this yields x
−ηΦ(H; N )
22√
(HG)
5N
119N
219N
39+ √
8HG(N
1N
2)
8N
33+
16√
(HG)
5(N
1N
2)
16N
3−1+
32√
(HG)
10(N
1N
2)
29N
34+
32√
(HG)
5(N
1N
2)
29N
314+ √
4(HG)
2N
13N
23+ √
4HG(N
1N
2)
3N
32+ x
0.1.
We put the above estimate in (1) of [3] and choose the parameter K opti- mally via a well-known lemma (cf. Lemma 3 of [3]) to get
x
−2ηS(a, b, c, d; N )
27√
G
5(N
1N
2)
24N
314+ √
9G(N
1N
2)
9N
34(1.3)
+
21√
G
5(N
1N
2)
21N
34+
42√
G
10(N
1N
2)
39N
314+
37√
G
5(N
1N
2)
34N
319+ √
5GN
14N
24N
33+ √
6G
2N
15N
25N
32+ x
0.1108
√
x
5N
161N
266N
331+
36√
xN
129N
230N
311(1.4)
+
84√
x
5N
143N
248N
33+
148√
x
5N
1101N
2106N
351+
12√
xN
13N
24N
3−1+
20√
xN
19N
210N
37+ x
0.1. To pass from (1.3) to (1.4) we have invoked (18) of [3]. By (21) of [3], (1.5) x
−ηS(a, b, c, d; N ) √
8xN
1−3N
2−2N
3−1+ x
0.1. From (1.4) and (1.5) we infer that
(1.6) x
−2ηS(a, b, c, d; N ) X
1≤i≤6
P
i+ x
0.1,
where
P
1= min(
108√
x
5N
161N
266N
331, √
8xN
1−3N
2−2N
3−1)
89√
x
9N
1−8N
2, (1.7)
P
2= min(
36√
xN
129N
230N
311, √
8xN
1−3N
2−2N
3−1)
31√
x
3N
1−1N
22, (1.8)
P
3= min(
84√
x
5N
143N
248N
33, √
8xN
1−3N
2−2N
3−1) ≤
54√
x
4N
117N
221, (1.9)
P
4= min(
148√
x
5N
1101N
2106N
351, √
8xN
1−3N
2−2N
3−1) (1.10)
≤
139√
x
14N
1−13N
2, P
5= min(
12√
xN
13N
23, √
8xN
1−3N
2−3) x
0.1, (1.11)
P
6= min(
20√
xN
19N
210N
37, √
8xN
1−3N
2−2N
3−1)
19√
x
2N
1−3N
2−1. (1.12)
Next, we again apply Lemma 1.1 to the triple summation over n
1, n
2and u in (1.1), but with the choice (h, x, y) = (n
1, u, n
2). This gives
x
−ηΦ(H; N )
22√
(HG)
11N
119N
213N
33+ (HG)
1/2N
1N
25/8+ (HG)
1/4N
1N
217/16+
32√
(HG)
11N
129N
228N
33+
32√
(HG)
16N
129N
218N
33+ √
4HGN
13N
24N
3+ √
4(HG)
2N
13N
22N
3+ x
0.1.
We put the above estimate in (1) of [3] and choose K optimally to get
x
−2ηS(a, b, c, d; N )
33√
G
11N
130N
224N
314+
12√
G
4N
112N
29N
34(1.13)
+
20√
G
4N
120N
221N
34+
43√
G
11N
140N
239N
314+
48√
G
16N
145N
234N
319+ √
5GN
14N
25N
32+ √
6G
2N
15N
24N
33+ x
0.1132
√
x
11N
143N
230N
3+
12√
xN
15N
22+
20√
xN
113N
214+
172√
x
11N
183N
290N
3+
48√
x
4N
117N
29+
20√
xN
19N
214N
33+
12√
xN
13N
22N
3+ x
0.1. From (1.5) and (1.13) we get
x
−2ηS(a, b, c, d; N ) x
0.1+ X
1≤i≤4
Q
i+
12√
xN
15N
22(1.14)
+
20√
xN
113N
214+
48√
x
4N
117N
29, where
Q
1= min(
132√
x
11N
143N
230N
3, √
8xN
1−3N
2−2N
3−1) (1.15)
≤
35√
x
3N
110N
27, Q
2= min(
172√
x
11N
183N
290N
3, √
8xN
1−3N
2−2N
3−1) (1.16)
≤
45√
x
3N
120N
222, Q
3= min(
20√
xN
19N
214N
33, √
8xN
1−3N
2−2N
3−1) ≤
11√ xN
22, (1.17)
Q
4= min(
12√
xN
13N
22N
3, √
8xN
1−3N
2−2N
3−1) ≤ x
0.1. (1.18)
From (31) of [3] we have
(1.19) x
−ηS(a, b, c, d; N )
28√
(x(N
1N
2)
−1)
3+ x
0.1. By (1.2) and (1.19) we have
(1.20) x
−ηS(a, b, c, d; N ) X
1≤i≤4
R
i+ x
0.1,
where
R
1= min(
33√
x
3N
112N
22,
28√
(x(N
1N
2)
−1)
3) ≤
31√ x
3N
16, (1.21)
R
2= min(
888√
x
73N
1377N
2162,
28√
(x(N
1N
2)
−1)
3) ≤
480√
x
47N
143, (1.22)
R
3= min(
76√
x
6N
134N
219,
28√
(x(N
1N
2)
−1)
3) ≤
152√
x
15N
19, (1.23)
R
4= min(
289√
x
24N
1121N
251,
28√
(x(N
1N
2)
−1)
3) ≤
153√
x
15N
114.
(1.24)
From (1.20) to (1.24), we find that the required estimate follows if N
1≤ x
15/869. We assume hereafter that N
1> x
15/869. From (1.19) and (1.6) to (1.12) we have
(1.25) x
−2ηS(a, b, c, d; N )
19√
x
2N
1−3N
2−1+ X
1≤i≤4
S
i+ x
0.1, where
S
1= min(
89√
x
9N
1−8N
2,
28√
(x(N
1N
2)
−1)
3) (1.26)
≤
295√
x
30N
1−27< x
87/869, S
2= min(
31√
x
3N
1−1N
22,
28√
(x(N
1N
2)
−1)
3) (1.27)
≤
149√
x
15N
1−9≤ x
0.1, S
3= min(
54√
x
4N
117N
221,
28√
(x(N
1N
2)
−1)
3) (1.28)
≤
250√
x
25N
1−4≤ x
0.1, S
4= min(
139√
x
14N
1−13N
2,
28√
(x(N
1N
2)
−1)
3) (1.29)
≤
445√
x
45N
1−42≤ x
0.1. From (1.25) to (1.29) we have
(1.30) x
−2ηS(a, b, c, d; N )
19√
x
2N
1−3N
2−1+ x
87/869. By (1.30) and (1.14) to (1.18) we have
(1.31) x
−2ηS(a, b, c, d; N ) X
1≤i≤6
T
i+ x
87/869, where
T
1= min(
19√
x
2N
1−3N
2−1,
35√
x
3N
110N
27) (1.32)
≤ (x
17N
1−11)
1/168≤ x
0.10007, T
2= min(
19√
x
2N
1−3N
2−1,
45√
x
3N
120N
222) (1.33)
≤ (x
47N
1−46)
1/463≤ x
0.1, T
3= min(
19√
x
2N
1−3N
2−1,
11√
xN
22) ≤ (x
5N
1−6)
1/49≤ x
0.1, (1.34)
T
4= min(
19√
x
2N
1−3N
2−1,
12√
xN
15N
22) ≤ (x
5N
1−1)
1/50≤ x
0.1, (1.35)
T
5= min(
19√
x
2N
1−3N
2−1,
20√
xN
113N
214) (1.36)
≤ (x
29N
1−29)
1/286≤ x
0.1, T
6= min(
19√
x
2N
1−3N
2−1,
48√
x
4N
117N
29) (1.37)
≤ (x
22N
1−10)
1/219≤ x
0.1.
By (1.30) to (1.37), we have completed the proof.
2. Proof of Theorem 2. Let (a, b, c, d) be any permutation of (1, 1, 2, 2).
It suffices to obtain
(2.1) S(a, b, c, d; N ) x
7/19+4η,
where η = ε/8, N = (N
1, N
2, N
3), N
1, N
2and N
3are positive integers with (2.2) N
1N
2N
3, N
1aN
2bN
3c+dx, N
1N
2N
3> x
1/3,
and the sum S(a, b, c, d; N ) is defined on p. 199 of [4]. We will retain many familiar notations used in both [3] and [4].
The case of (a, b, c, d) = (1, 1, 2, 2) can be dealt with immediately. In fact, from (2.2) we have N
1N
2x
1/3, thus (GN
1N
2N
3)
1/2(xN
1N
2)
1/4x
1/3, and the required estimate follows from Lemma 6 of [4].
For (a, b, c, d) = (1, 2, 1, 2), by (2.2) we have N
13N
33N
1N
22N
33x, thus again (GN
1N
2N
3)
1/2x
1/3, and the required estimate follows.
We now show
(2.3) S(2, 1, 1, 2; N ) x
4/11+ε.
To this end, we first proceed similarly to pp. 167–170 of [3]. This yields, similarly to (7) of [3],
Φ(H; N ) H
−1(N
32H
−1G
−1)
1/2X
h∼H
X
1
F (n
1)R(n
2)S(u)e(g) (2.4)
+ N
1(HG)
1/2ln x + x
13/36(for an explanation of the error term x
13/36, cf. p. 199 of [4]), where P means summation over n
1, n
2and u with
11 < n
1< n
2, N
v≤ n
v< 2N
v(v = 1, 2), U
1< u < U
2,
and G, U
1, U
2and the function g are defined on p. 169 of [3]. In particular, g = C
2(xn
−21n
−12h
2u)
1/3. Moreover, F (·), R(·), S(·) are suitable monomials with absolute values ∼ = 1. We can apply Lemma 1 of [3] one more time, to the variable n
2of (2.4). We have
Φ(H; N ) N
2N
3(H
2G)
−1(2.5)
× X
h∼H
X
n1∼N1
X
U1<u<U2
X
V1<v<V2
T (u)Q(v)e(g
1) + N
1(HG)
1/2ln x + x
13/36,
where V
i= V
i(h, n
1, u) (i = 1, 2), |T (u)| ≤ 1, |Q(v)| ≤ 1, and g
1= C
3(h
2xuvn
−21)
1/4. We can relax the condition U
1< u < U
2to u ∼ = U :=
HGN
3−1and the condition V
1< v < V
2to v ∼ = V := HGN
2−1consecutively
by means of Lemma 5 of [3] (note that we can assume that x is quadratic
irrational, cf. p. 168 of [3]); we thus deduce from (2.5) that
x
−ηΦ(H; N ) N
2N
3(H
2G)
−1(2.6)
× X
h∼H
X
n1∼N1
X
u∼=U
X
v∼=V
T
1(u)Q
1(v)e(g
1) + N
1(HG)
1/2+ x
13/36,
where |T
1(u)|, |Q
1(v)| ≤ 1. From (2.6) it is evident that x
−2ηΦ(H; N ) N
2N
3(H
2G)
−1(2.7)
× X
h∼H
X
n1∼N1
X
w∼=W
K(w)e(C
3(h
2xwn
−21)
1/4) + N
1(HG)
1/2+ x
13/36,
where W = U V = (HG)
2N
2−1N
3−1and |K(w)| ≤ 1.
If HG N
2N
3, we apply Lemma 1.1 to the triple exponential sum in (2.7), with (h, x, y) = (h, n
1, w), to get
x
−3ηΦ(H; N )
22√
H
4G
7N
119(N
2N
3)
9+ (HG)
1/4N
1(N
2N
3)
3/8(2.8)
+ (HG)
7/8N
1(N
2N
3)
−1/16+
32√
H
19G
22N
129(N
2N
3)
4+
32√
H
4G
4N
129(N
2N
3)
14+ √
4H
3G
4N
13+ √
4GN
13(N
2N
3)
2+ x
13/3622
√
G
3N
119(N
2N
3)
13+ N
1(N
2N
3)
5/8+
32√
G
3N
129(N
2N
3)
18+ √
4GN
13(N
2N
3)
2+ N
1(HG)
13/16+
32√
H
19G
22N
129(N
2N
3)
4+ √
4H
3G
4N
13+ x
13/36.
If N
1≥ x
1/22, we have (GN
1N
2N
3)
1/2= (xN
2N
3)
1/4(x
3N
1−2)
1/8x
4/11, and (2.3) follows from Lemma 6 of [4]. We now assume that N
1<
x
1/22. Then we easily see that the total contribution of the first four terms in (2.8) is x
0.34. In fact, since N
1N
2N
3x
1/2,
22
√
G
3N
119(N
2N
3)
1344
√
x
3N
132(N
2N
3)
2388
√
x
29N
118x
0.34, N
1(N
2N
3)
5/816
√
x
5N
16x
0.33,
32
√
G
3N
129(N
2N
3)
1864
√
x
3N
152(N
2N
3)
33128
√
x
39N
138x
0.32,
√
4GN
13(N
2N
3)
2√
8xN
14(N
2N
3)
316
√
x
5N
12x
0.32. Thus from (2.8) we get
x
−3ηΦ(H; N ) N
1(HG)
13/16+
32√
H
19G
22N
129(N
2N
3)
4(2.9)
+ √
4H
3G
4N
13+ x
13/36.
If HG N
2N
3, we go back to the original definition for Φ(H; N ), and we produce a new integral variable q from n
2and n
3such that q = n
2n
3. Since HG N
2N
3, Lemma 1.1 is applicable with (h, x, y) = (h, n
1, q), and we get
x
−ηΦ(H; N )
22√
G
3N
119(N
2N
3)
13+ N
1(N
2N
3)
5/8+ N
1(HG)
13/16+
32√
H
19G
22N
129(N
2N
3)
4+
32√
G
3N
129(N
2N
3)
18+ √
4H
3G
4N
13+ √
4GN
13(N
2N
3)
2+ x
13/36N
1(HG)
13/16+
32√
H
19G
22N
129(N
2N
3)
4+ √
4H
3G
4N
13+ x
13/36.
Thus we see that (2.9) always holds. We put the estimate (2.9) in (1) of [3]
and choose the parameter K optimally via Lemma 3 of [3] to get x
−4ηS(2, 1, 1, 2; N )
29√
G
13N
129(N
2N
3)
13+
51√
G
22N
148(N
2N
3)
23+ √
7G
4N
16(N
2N
3)
3+ x
13/3658
√
x
13N
132(N
2N
3)
13+
51√
x
11N
126(N
2N
3)
12+ √
7x
2N
12N
2N
3+ x
13/36116
√
x
39N
138+
51√
x
17N
114+
14√
x
5N
12+ x
13/36x
4/11, which proves (2.3).
We proceed to estimate S(2, 2, 1, 1; N ); the remaining two cases with (a, b, c, d) = (2, 1, 2, 1) and (1, 2, 2, 1) can be treated similarly. As in (7) of [3], we get
Φ(H; N ) H
−1(N
32H
−1G
−1)
1/2(2.10)
× X
h∼H
X
2
g
1(n
1)g
2(n
2)g
3(u)e
C
xhu n
21n
22 1/2+ N
1N
2ln x + (Hx(N
1N
2N
3)
−1)
1/2+ x
13/36, where P
2