View
1
Download
0
Category
Preview:
Citation preview
Multiple Fourier series and lattice point problems
Shigehiko Kuratsubo and Eiichi Nakai
Abstract
For the multiple Fourier series of the periodization of some radial functionson Rd, we investigate the behavior of the spherical partial sum. We show theGibbs-Wilbraham phenomenon, the Pinsky phenomenon and the third phe-nomenon for the multiple Fourier series, involving the convergence propertiesof them. The third phenomenon is closely related to the lattice point prob-lems, which is a classical theme of the analytic number theory. We also provethat, for the case of two or three dimension, the convergence problem on theFourier series is equivalent to the lattice point problems in a sense. In partic-ular, the convergence problem at the origin in two dimension is equivalent toHardy’s conjecture on Gauss’s circle problem.
1 Introduction
It is well known as the Gibbs-Wilbraham phenomenon that, for the Fourier series of
piecewise continuous functions, in the neighborhood of each jump, the partial sums
overshoot the jump by approx 9% of the jump. This phenomenon can be seen not
only in one dimension but also in higher dimensions (see for example [4, 24, 42]).
In one dimension, it is also well known as the localization property that, if
the function is zero on an interval, then the Fourier series converges to zero there.
However, in higher dimensions this property is no longer valid. In 1993, Pinsky,
Stanton and Trapa [36] showed that, for the Fourier series of the indicator function
of a d-dimensional ball with d ≥ 3, the spherical partial sum diverges at the centerof the ball. That is, the Fourier series diverges at a smooth point - even a point of
2010 Mathematics Subject Classification. Primary 42B05Key words and phrases. multiple Fourier series, lattice point problem, Fourier transform, Hardy’s
identity, Gibbs-Wilbraham phenomenon, Pinsky phenomenon, spherical partial sum, radial func-tion.This work was supported by JSPS KAKENHI Grant Numbers JP22540166, JP24540159,
JP17K18731.
1
arX
iv:2
011.
0617
8v1
[m
ath.
FA]
12
Nov
202
0
2 S. Kuratsubo and E. Nakai
local constancy - of the function, resulting from global rather than local properties.
This phenomenon is called the Pinsky phenomenon.
In 2010, the third phenomenon was discovered in [18, 22]. Namely, for the Fourier
series of the indicator function of a d-dimensional ball with d ≥ 5, the sphericalpartial sum diverges at all rational points, while it converges almost everywhere.
This third phenomenon was proved by using results on the lattice point problems.
The study of lattice point problems is a classical theme of analytic number
theory which is concerned with the number of integer points. It has a long history
and deep accumulations since G. Voronöı, G. H. Hardy, E. Landau, J. G. Van der
Corput, V. Jarńık and A. Walfisz, see [5, 15, 16]. For example, in R2 = {(x1, x2) :x1 and x2 are real numbers}, by A(s) we denote the number of lattice points, whereare points with integral co-ordinates, inside the circle
x21 + x22 = s,
see Figure 1. Then A(s) is the same as the area of the polygon in Figure 2, since the
polygon is the union of the unit squares whose centers are the lattice points inside
the circle. Let P (s) = A(s)− πs, where πs is the area of this circle. Gauss showedthat
P (s) = O(s1/2) as s→∞.
In the above O is Landau’s symbol, that is, f(s) = O(g(s)) as s→∞ means that
Figure 1: Gauss’s circle problem Figure 2: The union of the unit squares.
lim sups→∞ |f(s)|/g(s) < ∞ for the positive valued function g. Similarly, f(s) =
Multiple Fourier series and lattice point problems 3
o(g(s)) as s→∞ means that lims→∞ f(s)/g(s) = 0. In 1915 Hardy [8] proved that
P (s) 6= O(sθ) if θ ≤ 1/4.
In fact P (s) 6= o(s1/4 log1/4 s). The best bound on θ for P (s) = O(sθ) is a very fa-mous open problem known as Gauss’s circle problem. While Hardy’s conjecture([9])
is P (s) = O(s1/4+ε) for any ε > 0, the most sharp result up to now is P (s) =
O(s131/416(log s)18637/8320) by M. Huxley [12] in 2003, where 131/416 = 0.3149 . . . .
(Recently, Bourgain and Watt [1] gave θ = 517/1648 = 0.31371 . . . in arXiv, 2017.)
Note that A(s) is a special case of
∑m21+···+m2d≤s
exp
(2πi
d∑k=1
mkxk
), (1.1)
where m1, · · · ,md are integers and x1, · · · , xd are real numbers, that is, A(s) is thecase (x1, · · · , xd) = (0, · · · , 0) of (1.1) and d = 2. The sum (1.1) is related to theFourier series. Especially the research on the sum (1.1) by Czechoslovakian mathe-
matician B. Novák (1938–2003) is very important for the study of the convergence
problem of multiple Fourier series.
Recently, Taylor [43, 44] found that the Pinsky phenomenon arises even in two-
dimension. He treated the radial function
U(x) =
{1/√a2 − |x|2, |x| < a,
0, |x| ≥ a,x ∈ R2, a > 0,
Then U(x) is the fundamental solution to the wave equation on R × T2, evaluatedat t = a. Our aim in this paper which is motivated by Taylor [43, 44] is to study the
Gibbs-Wilbraham phenomenon, the Pinsky phenomenon and the third phenomenon
on the Fourier series of
Uβ,a(x) =
{(a2 − |x|2)β, |x| < a0, |x| ≥ a,
x ∈ Rd, a > 0, (1.2)
for β > −1, a > 0 and dimension d, involving the convergence properties of them.If β = 0, then Uβ,a(x) is the same as the indicator function of the ball centered at
the origin and of radius a.
4 S. Kuratsubo and E. Nakai
By Rd, Zd and Td = Rd/Zd we denote the d-dimensional Euclidean space, integerlattice and torus, respectively. In this paper, however, we always identify Td with(−1/2, 1/2]d, that is, x ∈ Td means x ∈ (−1/2, 1/2]d and Td ⊂ Rd. Let Q be the setof all rational numbers, and let Qd = {(x1, · · · , xd) : x1, · · · , xd ∈ Q}.
For an integrable function F (x) on Rd, its Fourier transform F̂ (ξ) and its Fourierspherical partial integral σλ(F )(x) of order λ ≥ 0 are defined by
F̂ (ξ) =
∫RdF (x)e−2πiξx dx, ξ = (ξ1, · · · , ξd) ∈ Rd, (1.3)
σλ(F )(x) =
∫|ξ|
Multiple Fourier series and lattice point problems 5
If β = 0, then u0,a is the periodization of the indicator function of the ball centered
at the origin and of radius a. See Figure 3 and also Figure 4. Let
Figure 3: uβ,a(x) (d = 2, β = 0, a = 3/4).
u0,a(x) =∑m∈Zd
U0,a(x+m), x ∈ Td, U0,a(x) =
1, |x| < 1,1/2, |x| = 1,0, |x| > 1,
x ∈ Rd.
If d = 2, then
Sλ(u0,a)(x) = πa2 + a
∑0
6 S. Kuratsubo and E. Nakai
We first show an identity (Theorem 2.1) for the periodization of any integrable
radial function with compact support. Then it turns out that the difference between
the Fourier partial sum and the Fourier partial integral is closely related to lattice
point problems. Therefore, to study the convergence problem on the Fourier series
of uβ,a(x) we must also investigate the behavior of σλ(Uβ,a)(x) as λ→∞ and latticepoint problems.
In the case β = 0, the convergence problem on the Fourier series of the function
u0,a has been studied in detail by [18, 19, 20, 21, 22, 23, 24, 36, 34, 35]. Some of
results in these papers were proved by using Novák’s results in [28, 29, 30, 32]. His
results were very useful and sufficient for the affirmative results of the convergence
problem on the Fourier series of the function u0,a.
On the other hand, in the case −1 < β < 0, Novák’s results on the sum (1.1) arenot sufficient to study the convergence problem on the Fourier series of uβ,a. Our
results on the convergence of the Fourier series of uβ,a are obtained by using the
best estimates up to now on lattice point problems. Therefore, if the lattice point
problems will be improved in the future, then our results can be also improved. Ac-
tually, the convergence problem on the Fourier series and the lattice point problems
are equivalent in a sense as we will show in Section 7. In particular, the conver-
gence problem at the origin in two dimension is equivalent to Hardy’s conjecture on
Gauss’s circle problem, see Remark 7.1.
To state our main results, for a > 0, let
Ea = {x ∈ Td : x 6= 0 and |x−m| 6= a for all m ∈ Zd}, (1.11)
Ga = {x ∈ Td : x 6= 0 and |x−m| = a for some m ∈ Zd}. (1.12)
Then Td = {0} ∪Ga ∪ Ea. See Figures 4. If 0 < a < 1/2, then
Ea = {x ∈ Td : x 6= 0 and |x| 6= a},
Ga = {x ∈ Td : |x| = a},
because x ∈ Td means x ∈ (−1/2, 1/2]d and then {x ∈ Td : |x −m| = a} is emptyfor m 6= 0. For a > 0, let also
rd(a : x) =∑
m∈Zd, |x−m|=a
1, x ∈ Td. (1.13)
Multiple Fourier series and lattice point problems 7
Then rd(a : x) = 0 for x ∈ Ea. If 0 < a < 1/2, then rd(a : 0) = 0 and rd(a : x) = 1for x ∈ Ga.
x1
x2
O
T2
R2
Ea
Ga
− 1 1
−1
1
x1
x2
O
T2
R2
−1 1
−1
1
Figure 4: Ea and Ga for a = 1/4 and a = 3/4
In the following, Theorem 1.1 and Corollary 1.2 deal with the behavior of Sλ(uβ,a)
at x = 0 including the Pinsky phenomenon, Theorem 1.3 deals with the point-
wise behaviors including the third phenomenon, Theorem 1.4 deals with the Gibbs-
Wilbraham phenomenon near Ga. Theorem 1.5 deals with the almost everywhere
convergence.
Our first result is on the behavior of Sλ(uβ,a)(0) which includes the Pinsky phe-
nomenon. Let Γ(s) be the Gamma function. For β > −1, a > 0 and the dimensiond, let
P[d]β,a =
Γ(β + 1)
Γ(d/2)a(d−3)/2+βπ(d−4)/2−β (1.14)
and
Lβ,a =Γ(β + 1)
2
(aπ
)βsin βπ2βπ2
, (1.15)where
(sin βπ
2
)/βπ
2is regarded as 1 if β = 0, that is, L0,a = 1/2.
Theorem 1.1. Let β > −1 and a > 0.
8 S. Kuratsubo and E. Nakai
(i) If d ≥ 1 and β > (d− 3)/2, then
Sλ(uβ,a)(0) = uβ,a(0) + rd(a : 0) (Lβ,a + o(1))λ−β +O(λ
d−32−β)
as λ→∞. (1.16)
(ii) If d ≥ 2 and −1 < β ≤ (d−3)/2, then Sλ(uβ,a) reveals the Pinsky phenomenon.More precisely,
Sλ(uβ,a)(0) = uβ,a(0) +(σλ(Uβ,a)(0)− Uβ,a(0)
)+ rd(a : 0)(Lβ,a + o(1))λ
−β + o(λd−32−β) as λ→∞,
where
σλ(Uβ,a)(0)−Uβ,a(0) = −P [d]β,a cos(
2πaλ− d− 1 + 2β4
π
)λd−32−β +O(λ
d−52−β).
If 0 < a < 1/2, then rd(a : 0) = 0. Hence we have the following corollary.
Corollary 1.2. Let d ≥ 1 and 0 < a < 1/2.
(i) If β > d−32
, then
limλ→∞
Sλ(uβ,a)(0) = uβ,a(0).
(ii) If −1 < β ≤ d−32
, then
lim infλ→∞
Sλ(uβ,a)(0)− uβ,a(0)λd−32−β
= −P [d]β,a,
lim supλ→∞
Sλ(uβ,a)(0)− uβ,a(0)λd−32−β
= P[d]β,a.
For example, even if d = 2, we can see the Pinsky phenomena on the graphs of
Sλ(uβ,a) with β = −1/2 and a = 1/4 for λ ∈ N, as Taylor pointed out in [43, 44],see Figure 6. In this case P
[d]β,a = 4.
If d = 3, β = 0 and a = 1/4, then P[d]β,a = 2/π, see Figure 7. If d = 4, β = 1/2
and a = 1/4, then P[d]β,a = 1/8, see Figure 8.
Next, we state the pointwise behaviors of Sλ(uβ,a) on Td \{0} which includes thethird phenomenon. We define the number c(d) for the dimension d as the following:
c(d) = d− 2dd+ 1
− d+ 12
=d− 5
2+
2
d+ 1=d(d− 4)− 1
2(d+ 1), (1.17)
Multiple Fourier series and lattice point problems 9
Figure 5: uβ,a(x1, x2) (d = 2, β = −1/2, a = 1/4).
Figure 6: Sλ(uβ,a)(x1, x2) (d = 2, β = −1/2, a = 1/4) for λ = 9, 10, 11, 12.
10 S. Kuratsubo and E. Nakai
-0.4 -0.2 0.2 0.4
0.5
1.0
1.5
-0.4 -0.2 0.2 0.4
0.5
1.0
1.5
-0.4 -0.2 0.2 0.4
0.5
1.0
1.5
-0.4 -0.2 0.2 0.4
0.5
1.0
1.5
Figure 7: Sλ(uβ,a)(x, 0, 0) (d = 3, β = 0, a = 1/4) for λ = 46, 47, 48, 49.
-0.4 -0.2 0.2 0.4-0.1
0.1
0.2
0.3
0.4
-0.4 -0.2 0.2 0.4-0.1
0.1
0.2
0.3
0.4
-0.4 -0.2 0.2 0.4-0.1
0.1
0.2
0.3
0.4
-0.4 -0.2 0.2 0.4-0.1
0.1
0.2
0.3
0.4
Figure 8: Sλ(uβ,a)(x, 0, 0, 0) (d = 4, β = 1/2, a = 1/4) for λ = 43, 44, 45, 46.
Multiple Fourier series and lattice point problems 11
that is,
c(1) = −1, c(2) = −5/6, c(3) = −1/2, c(4) = −1/10.
Theorem 1.3. Let β > −1 and a > 0.
(i) If 1 ≤ d ≤ 4, then,
(a) for β > −1,
limλ→∞
Sλ(uβ,a)(x)− uβ,a(x)λ−β
= rd(a : x)Lβ,a for all x ∈ Ga.
(b) for β > c(d),
limλ→∞
Sλ(uβ,a)(x) = uβ,a(x) uniformly on any compact set in Ea.
(ii) If d ≥ 5, then
limλ→∞
1
λd−52
∣∣∣∣Sλ(uβ,a)(x)− uβ,a(x)λ−β − rd(a : x)Lβ,a∣∣∣∣ = 0for all x ∈ (Ea ∪Ga) \Qd,
and
0 < lim supλ→∞
1
λd−52
∣∣∣∣Sλ(uβ,a)(x)− uβ,a(x)λ−β − rd(a : x)Lβ,a∣∣∣∣
12 S. Kuratsubo and E. Nakai
-0.4 -0.2 0.2 0.4
-20
-10
10
20
Figure 9: uβ,a(x, 0, 0, 0, 0) (d = 5, β = −1/2, a = 1/4).
-0.4 -0.2 0.2 0.4
-20
-10
10
20
-0.4 -0.2 0.2 0.4
-20
-10
10
20
-0.4 -0.2 0.2 0.4
-20
-10
10
20
-0.4 -0.2 0.2 0.4
-20
-10
10
20
Figure 10: Sλ(uβ,a)(x, 0, 0, 0, 0) (d = 5, β = −1/2, a = 1/4) for λ = 100, . . . , 103.
0.25 0.30 0.35 0.40 0.45 0.50
-4
-2
2
4
0.25 0.30 0.35 0.40 0.45 0.50
-4
-2
2
4
0.25 0.30 0.35 0.40 0.45 0.50
-4
-2
2
4
0.25 0.30 0.35 0.40 0.45 0.50
-4
-2
2
4
Figure 11: Sλ(uβ,a)(x, 0, 0, 0, 0) (d = 5, β = −1/2, a = 1/4) for λ = 100, . . . , 103:Expansion of the part 0.2 ≤ x ≤ 0.5 in Figure 10.
Multiple Fourier series and lattice point problems 13
Next, we state the Gibbs-Wilbraham phenomenon near Ga for 1 ≤ d ≤ 4.
Theorem 1.4. Let 1 ≤ d ≤ 4, c(d) < β ≤ 0 and 0 < a < 1/2. Then, on anyneighborhood of the set Ga, Sλ(uβ,a) reveals a phenomenon like the Gibbs-Wilbraham
phenomenon. More precisely, the following holds: For each x0 ∈ Ga, let {x±λ } ⊂ Td
be the sequences which satisfy limλ→∞
x±λ = x0 and |x±λ | = a∓ (2± β)/(4λ). Then
limλ→∞
Sλ(uβ,a)(x±λ )− uβ,a(x
±λ )
λ−β= G±β,a,
where
G±β,a = ∓Γ(β + 1)aβ(2± β)β
π2β
∫ ∞π
sin s
(s± β2π)β+1
ds. (1.18)
If d = 3, β = 0 and a = 1/4, then we can see both the Gibbs-Wilbraham
phenomena and the Pinsky phenomena on the graphs in Figure 7. If d = 4, β =
1/2 and a = 1/4, then we can see the Pinsky phenomena but not see the Gibbs-
Wilbraham phenomena on the graphs in Figure 8.
Remark 1.2. (i) The constant G+β,a is positive and G−β,a is negative. Especially
G±0,a = ∓1
π
∫ ∞π
sin s
sds = ∓1
2± 1π
∫ π0
sin s
sds = ±0.08949 · · · .
(ii) In the case 2 ≤ d ≤ 4 and −1 < β ≤ c(d), Theorem 1.4 is also an open problemby the same reason as Remark 1.1.
Finally, we state the almost everywhere convergence of Sλ(uβ,a) for d ≥ 4.
Theorem 1.5. Let d ≥ 4, β > −1/2 and a > 0. Then
limλ→∞
Sλ(uβ,a)(x) = uβ,a(x), a.e. x ∈ Td.
Remark 1.3. (i) From Theorems 1.3 and 1.5 we get Theorem in [18] as a corollary,
which is the case β = 0.
(ii) In the case d ≥ 4 and −1 < β ≤ −1/2, the almost everywhere convergence ofSλ(uβ,a) is an open problem.
We first prove in Section 2 a fundamental identity for the periodization f of any
integrable radial function F on Rd with compact support. This identity gives the
14 S. Kuratsubo and E. Nakai
relation among Sλ(f)(x), σλ(F )(x) and the term related to lattice point problems.
In Section 3 we collect the results with respect to Bessel functions. In Section 4
we investigate the behavior of σλ(Uβ,a)(x). We show the uniform convergence of
σλ(Uβ,a)(x) on any compact subset of Ea, the Pinsky phenomenon at the origin,
and, the Gibbs-Wilbraham phenomenon near the spherical surface. In Section 5
we prove lemmas related to the lattice point problems. Then, in Section 6, using
the results in Sections 2–5, we prove the main results, in which it turns out that
the third phenomenon is closely related to the lattice point problems. Finally, in
Section 7 we give the relation between the convergence of the spherical partial sum
and lattice point problems.
2 Fundamental identity
Let Jν be the Bessel function of order ν. If ν > −1, then
Jν(s)
sν→ 1
2νΓ(ν + 1)as s→ 0.
For this reason, in this paper we always regard
Jν(s)
sν=
1
2νΓ(ν + 1)for s = 0. (2.1)
For ν > −1, let
Λν(t : s) =Jν(2πt
√s)
sν2
, t ≥ 0, s > 0. (2.2)
By (2.1) we regard
Λν(t : 0) =(πt)ν
Γ(ν + 1), t ≥ 0.
For any α > −1 we define Dα(s : x), Dα(s : x) and ∆α(s : x) as the following:
Dα(s : x) =
1
Γ(α + 1)
∑|m|2 0,
0, s = 0,
x ∈ Rd,
Dα(s : x) =
1
Γ(α + 1)
∫|ξ|2 0,
0, s = 0,
x ∈ Rd,
Multiple Fourier series and lattice point problems 15
and
∆α(s : x) = Dα(s : x)−Dα(s : x), s ≥ 0, x ∈ Rd. (2.3)
We consider a function φ on [0,∞) and the periodization fφ of the functionFφ(x) = φ(|x|), x ∈ Rd, that is,
fφ(x) =∑m∈Zd
Fφ(x+m) =∑m∈Zd
φ(|x+m|), x ∈ Td. (2.4)
If the radial function Fφ is integrable on Rd, then the Fourier transform F̂φ is alsoradial and has the form
F̂φ(ξ) =
∫Rdφ(|x|)e−2πixξ dx = 2π
∫ ∞0
φ(t)J d
2−1(2πt|ξ|)
(t|ξ|) d2−1td−1 dt, (2.5)
that is,
F̂φ(ξ) = 2π
∫ ∞0
φ(t)Λ d2−1(t : |ξ|
2) td2 dt. (2.6)
For the equation (2.5), see [40, Theorem 3.3, page 155] if d ≥ 2. If d = 1, then (2.5)is also valid by the equation J− 1
2(s) =
√(2/πs) cos s. Let
Aφ(s) = 2π
∫ ∞0
φ(t)Λ d2−1(t : s) t
d2 dt, s ≥ 0. (2.7)
Then
F̂φ(ξ) = Aφ(|ξ|2). (2.8)
Moreover, if the support of φ is compact, then Aφ is infinitely differentiable and
A(j)φ (s) = 2π(−π)
j
∫ ∞0
φ(t)Λ d2−1+j(t : s) t
d2+j dt, s ≥ 0, j = 1, 2, · · · ,
since∂
∂sΛµ(t : s) = (−πt)Λµ+1(t : s), (2.9)
by the Bessel recurrence formula (see (2.15) below).
Now, we define d] as the smallest integer that is greater than (d− 1)/2, that is,
d] =
[d− 1
2
]+ 1, (2.10)
where [t] is the integer part of t ≥ 0. In this section we prove the following:
16 S. Kuratsubo and E. Nakai
Theorem 2.1. Let φ be a function on [0,∞) with compact support. Suppose thatFφ is integrable on Rd. Let fφ be the periodization of the function Fφ. Then
Sλ(fφ)(x) = σλ(Fφ)(x) +
d]∑j=0
(−1)j∆j(λ2 : x)A(j)φ (λ2)
+ (−1)d]+1∑
m∈Zd\{0}
∫ λ20
Dd](s : x−m)A(d]+1)
φ (s) ds, (2.11)
for all x ∈ Td.
As φ in Theorem 2.1, take
φβ,a(t) =
{(a2 − t2)β, 0 ≤ t < a,0, t ≥ a,
(2.12)
with β > −1 and a > 0. Then Uβ,a(x) = φβ,a(|x|), x ∈ Rd, and the function uβ,ais the periodization of Uβ,a(x) defined by (1.9). In this case we denote Aφ by Aβ,a.
Then Aβ,a can be calculated explicitly by using (2.5), (2.8) and (2.19) below. Its
derivatives A(j)β,a are also calculated by (2.9). That is,
Aβ,a(s) =Γ(β + 1)
πβad2+βΛ d
2+β(a : s),
A(j)β,a(s) = (−1)
jΓ(β + 1)
πβ−jad2+β+jΛ d
2+β+j(a : s).
(2.13)
Then we have the following corollary:
Corollary 2.2. Let β > −1 and a > 0. Then
Sλ(uβ,a)(x) = σλ(Uβ,a)(x) +
d]∑j=0
(−1)j∆j(λ2 : x)A(j)β,a(λ2)
+ (−1)d]+1∑
m∈Zd\{0}
∫ λ20
Dd](s : x−m)A(d]+1)
β,a (s) ds, (2.14)
for all x ∈ Td.
To prove Theorem 2.1, first we state the properties of the Bessel functions, Dα
and Dα. The following are the recurrence formulas for Bessel functions.d
ds(sνJν(s)) = s
νJν−1(s),d
ds
(Jν(s)
sν
)= −Jν+1(s)
sν. (2.15)
Multiple Fourier series and lattice point problems 17
The Bessel functions have also the following asymptotic behavior.
Jν(s) =sν
2νΓ(ν + 1)+O(sν+1) as s→ 0, (2.16)
Jν(s) =
√2
πscos
(s− 2ν + 1
4π
)+O(s−3/2) as s→∞. (2.17)
If α > −1, then
∫ t0
Dα(s : x) ds = Dα+1(t : x), t ≥ 0, x ∈ Rd, (2.18)
and
Dα(s : x) =1
Γ(α + 1)
∫|ξ|2 (d− 1)/2, for example, α = d], then
Dα(s : x) =∑m∈Zd
Dα(s : x−m) =sd2+α
πα
∑m∈Zd
J d2+α(2π
√s|x−m|)
(√s|x−m|) d2+α
, (2.20)
where the sum in (2.20) converges absolutely. In the above the equation (2.18)
follows from the definition. For the equalities (2.19) and (2.20), see (4.2) and (4.3)
in [24], respectively. By (2.15) and (2.19) we have
∂
∂sDα(s : x) = Dα−1(s : x), α > 0,
∂
∂sD0(s : x) = π
(√s
|x|
) d2−1
J d2−1(2π|x|
√s), α = 0.
(2.21)
Proof of Theorem 2.1. Combining (1.8) with (2.8) we have
f̂φ(m) = F̂φ(m) = Aφ(|m|2), m ∈ Zd.
18 S. Kuratsubo and E. Nakai
Then
Sλ(fφ)(x)−D0(λ2 : x)Aφ(λ2)
=∑|m|
Multiple Fourier series and lattice point problems 19
Using (2.21) and integration by parts again, we have∫ λ20
Dd](s : x)A(d]+1)
φ (s) ds
= Dd](λ2 : x)A(d])
φ (λ2)−
∫ λ20
Dd]−1(s : x)A(d])
φ (s) ds
= · · · · · ·
=
d]∑j=0
(−1)jDd]−j(λ2 : x)A(d]−j)φ (λ
2)
+ (−1)d]+1∫ λ20
π
(√s
|x|
) d2−1
J d2−1(2π|x|
√s)Aφ(s) ds
=
d]∑j=0
(−1)d]−jDj(λ2 : x)A(j)φ (λ2)
+ (−1)d]+1(2π)∫ λ0
Aφ(u2)J d
2−1(2π|x|u)
(|x|u) d2−1ud−1 du.
Here we note that, by (2.8) and (2.5),
σλ(Fφ)(x) =
∫|ξ|
20 S. Kuratsubo and E. Nakai
Bessel functions:
Jν(s) =
√2
πs
(cos
(s− 2ν + 1
4π
)− 4ν
2 − 18s
sin
(s− 2ν + 1
4π
))+O(s−5/2)
as s→∞. (3.1)
Lemma 3.1 (Gradshteyn and Ryzhik [6], page 676, 6.561, 14). Let ν + 1 > −µ >−1/2. Then ∫ ∞
0
xµJν(x) dx = 2µΓ((1 + ν + µ)/2)
Γ((1 + ν − µ)/2).
Take
ν =1
2+ β and µ = −1
2− β
in Lemma 3.1, and we have the following corollary.
Corollary 3.2. Let β > −1. Then∫ ∞0
J 12+β(s)
s12+β
ds =1
2βΓ(1 + β)
√π
2.
Lemma 3.3. Let β > −1. Then∫ ∞u
J 12+β(s)
s12+β
ds =
√2
πu−β−1 cos
(u− β
2π
)+O(u−β−2) as u→∞.
Proof. Use (3.1) and two equations∫ ∞u
cos(s− θ)s1+β
ds = u−β−1 cos(u− θ + π
2
)+O(u−2−β),∫ ∞
u
sin(s− θ)s2+β
ds = O(u−2−β),
and we have the conclusion.
Lemma 3.4 (Gradshteyn and Ryzhik [6], page 683, 6.575, 1). Let ν+1 > µ > −1,A > 0 and B > 0. Then
∫ ∞0
Jν+1(As)Jµ(Bs)sµ−ν ds =
0 (A < B)(A2 −B2)ν−µBµ2ν−µAν+1Γ(ν − µ+ 1)
(A > B).
Multiple Fourier series and lattice point problems 21
Take
ν + 1 = µ+ β + 1, A = 1 and B =t
a,
in Lemma 3.4, and we have the following corollary.
Corollary 3.5. Let µ > −1, a > 0, β > −1 and t > 0. Then
2βΓ(β + 1)a2β∫ ∞0
Jµ(tas)Jµ+β+1(s)(ta
)µsβ
ds =
{0 (t > a)
(a2 − t2)β (0 < t < a),
Lemma 3.6. Let ν > −1, µ > −1, A > 0 and B > 0. Then
Jν(Au)Jµ(Bu) =1
π√ABu
(cos
((A−B)u− ν − µ
2π
)+ cos
((A+B)u− ν + µ+ 1
2π
))+
1√AB
(1
A+
1
B
)O(u−2) as u→∞.
Proof. Use (2.17) and 2 cos θ cosφ = cos(θ + φ) + cos(θ − φ), and we have theconclusion.
Let
sign(r) =
1, r > 0,
0, r = 0,
−1, r < 0.
Lemma 3.7. For β > −1, let
ψβ(λ, r) = |r|β∫ ∞2π|r|λ
1
sβ+1cos
(s− sign(r)β + 1
2π
)ds, λ > 0, r ∈ R\{0}. (3.2)
Then ψβ has the following properties:
(i) For each r ∈ R \ {0},
ψβ(λ, r) = |r|−1O(λ−β−1) as λ→∞. (3.3)
(ii) If β > 0, then
ψβ(λ, r) = O(λ−β) as λ→∞ (3.4)
uniformly with respect to r ∈ R \ {0}.
22 S. Kuratsubo and E. Nakai
(iii) If −1 < β ≤ 0, then, for each λ > 0, −ψβ(λ, r) has values(2 + β
4λ
)β ∫ ∞π
− sin s(s+ β
2π)β+1
ds at r =2 + β
4λ,
and (2− β
4λ
)β ∫ ∞π
sin s
(s− β2π)β+1
ds at r = −2− β4λ
.
Proof. (i) Using the equation∫ ∞v
cos(s− θ)sβ+1
ds = O(v−β−1), (3.5)
we have ψβ(λ, r) = |r|βO((|r|λ)−β−1) = |r|−1O(λ−β−1).(ii) Changing of variables, we have
ψβ(λ, r) =1
(2πλ)β
∫ ∞1
cos(2π|r|λu− sign(r)β+12π)
uβ+1du,
which shows ψβ(λ, r) = O(λ−β) uniformly with respect to r, since β > 0.
(iii) If r > 0, then
−ψβ(λ, r) = rβ∫ ∞2πrλ
− sin(s− β
2π)
sβ+1ds = rβ
∫ ∞2πrλ−β
2π
− sin s(s+ β
2π)β+1 ds,
and, if r < 0, then
−ψβ(λ, r) = (−r)β∫ ∞2π(−r)λ
sin(s+ β
2π)
sβ+1ds = (−r)β
∫ ∞2π(−r)λ+β
2π
sin s(s− β
2π)β+1 ds.
Then we have the conclusion.
Lemma 3.8. Let µ > −1, β > −1 and a > 0. If 0 < t < a or t > a, then∫ ∞2πaλ
Jµ(tau)Jµ+β+1(u)(ta
)µuβ
du
=1
πaβ
(at
)µ+ 12ψβ(λ, a− t) +
1
aβ+1
(at
)µ+ 12(
1 +a
t
)O(λ−β−1) as λ→∞,
where the term O(λ−β−1) is uniform with respect to t.
Multiple Fourier series and lattice point problems 23
Proof. By Lemma 3.6 we have
Jµ(t
au)Jµ+β+1(u)
=1
πu
√a
t
(cos
(|a− t|a
u− sign(a− t)1 + β2
π
)+ cos
(a+ t
au− 2µ+ β + 1
2π
))+
√a
t
(1 +
a
t
)O(u−2).
Let
I1 =1
π
(at
)µ+ 12
∫ ∞2πaλ
cos(|a−t|au− sign(a− t)β+1
2π)
uβ+1du,
I2 =1
π
(at
)µ+ 12
∫ ∞2πaλ
cos(a+tau− 2µ+β+1
2π)
uβ+1du.
Then ∫ ∞2πaλ
Jµ(tau)Jµ+β+1(u)(ta
)µuβ
du = I1 + I2 +(at
)µ+ 12(
1 +a
t
)a−β−1O(λ−β−1).
Changing of variables and using (3.5), we have
I1 =1
π
(at
)µ+ 12
(|a− t|a
)β ∫ ∞2π|a−t|λ
cos(s− sign(a− t)β+1
2π)
sβ+1ds
=1
πaβ
(at
)µ+ 12ψβ(λ, a− t)
and
I2 =1
π
(at
)µ+ 12
(a+ t
a
)β ∫ ∞2π(a+t)λ
cos(s− 2µ+β+1
2π)
sβ+1ds
=(at
)µ+ 12
(a+ t
a
)β(a+ t)−β−1O(λ−β−1)
=1
aβ+1
(at
)µ+ 12
(a
a+ t
)O(λ−β−1).
Then we have the conclusion.
Lemma 3.9 (Gradshteyn and Ryzhik [6], page 683, 6.574, 2). Let ν+µ+1 > κ > 0
and b > 0. Then∫ ∞0
Jν(bs)Jµ(bs)
sκds =
bκ−1Γ(κ)Γ(µ+ν−κ+12
)
2κΓ(µ−ν+κ+12
)Γ(−µ+ν+κ+12
)Γ(µ+ν+κ+12
).
24 S. Kuratsubo and E. Nakai
Especially, for µ > −1 and β > 0,∫ ∞0
Jµ+β+1(s)Jµ(s)
sβds = 0.
Lemma 3.10 (Gradshteyn and Ryzhik [6], page 660, 6.512, 3). Let ν > 0. Then∫ ∞0
Jν(s)Jν−1(s) ds =1
2.
Lemma 3.11. Let µ > −1 and β ≥ 0. Then
∫ ∞u
Jµ+β+1(s)Jµ(s)
sβds =
−1
πβsin
(βπ
2
)u−β +O(u−1−β), if β > 0,
O(u−1), if β = 0,
as u→∞.
Proof. By Lemma 3.6 we have
Jµ+β+1(s)Jµ(s) =1
πs
(cos
(β + 1
2π
)+ cos
(2s− 2µ+ 2 + β
2π
))+O(s−2). (3.6)
Using
1
π
∫ ∞u
1
sβ+1cos
(β + 1
2π
)ds =
−1
πβsin
(βπ
2
)u−β, β > 0,
0, β = 0,
and (3.5), we have the conclusion.
Corollary 3.12. Let µ > −1 and β > −1. Then
2βΓ(β + 1)a2β∫ 2πaλ0
Jµ+β+1(s)Jµ(s)
sβds
= Lβ,aλ−β +
{O(λ−β−1), β ≥ 0,O(1), −1 < β < 0,
as λ→∞. (3.7)
In the above Lβ,a is defined by (1.15), that is,
Lβ,a =Γ(β + 1)
2
(aπ
)βsin βπ2βπ2
,where
(sin βπ
2
)/βπ
2is regarded as 1 if β = 0.
Multiple Fourier series and lattice point problems 25
Proof. Case 1. Let β ≥ 0. By Lemmas 3.9 and 3.10 we have
∫ ∞0
Jµ+β+1(s)Jµ(s)
sβds =
{0, if β > 0,12, if β = 0.
Then ∫ 2πaλ0
Jµ+β+1(s)Jµ(s)
sβds = −
∫ ∞2πaλ
Jµ+β+1(s)Jµ(s)
sβds+
{0, if β > 0,12, if β = 0.
By Lemma 3.11 we have (3.7).
Case 2. Let −1 < β < 0. By (3.6) we have∫ 2πaλ1
Jµ+β+1(s)Jµ(s)
sβds
=1
π
∫ 2πaλ1
1
sβ+1
{cos
(β + 1
2π
)+ cos
(2s− 2ν + 2 + β
2π
)}ds+O(λ−β−1)
=1
π
(− sin
(βπ
2
))(2πaλ)−β
−β+O(1).
Since ∫ 10
Jµ+β+1(s)Jµ(s)
sβds
is a constant, we have the conclusion.
The following lemma is an extension of [24, Lemma 3.3] which is for the case
β = 0.
Lemma 3.13. Let ν > 0, λ > 0, ω ≥ 0 and β > −1. Then∫ λ0
Jν+β+1(s)Jν(ωs)
ωνsβds =
∫ λ0
Jν+β(s)Jν−1(ωs)
ων−1sβds− Jν+β(λ)Jν(ωλ)
ωνλβ. (3.8)
In the above, if ω = 0, then by (2.1) we regard (3.8) as∫ λ0
Jν+β+1(s)
sβsν
2νΓ(ν + 1)ds =
∫ λ0
Jν+β(s)
sβsν−1
2ν−1Γ(ν)ds− Jν+β(λ)
λβλν
2νΓ(ν + 1).
26 S. Kuratsubo and E. Nakai
Proof. Case 1. Let ω > 0. Then, using (2.15) and integration by parts, we have∫ λ0
Jν+β(s)Jν−1(ωs)
sβων+1 ds
=
∫ λ0
Jν+β(s)
sν+β((ωs)νJν−1(ωs)
)ω ds
=
∫ λ0
Jν+β(s)
sν+β∂
∂s
((ωs)νJν(ωs)
)ds
=
[Jν+β(s)
sν+β((ωs)νJν(ωs)
)]λ0
+
∫ λ0
Jν+β+1(s)
sν+β((ωs)νJν(ωs)
)ds
=Jν+β(λ)Jν(ωλ)
λβων +
∫ λ0
Jν+β+1(s)Jν(ωs)
sβων ds.
Case 2. Let ω = 0. Then,∫ λ0
Jν+β(s)
sβsν−1
2ν−1Γ(ν)ds
=
∫ λ0
Jν+β(s)
sβ+νs2ν−1
2ν−1Γ(ν)ds
=
[Jν+β(s)
sβ+νs2ν
2ν2ν−1Γ(ν)
]λ0
+
∫ λ0
Jν+β+1(s)
sβ+νs2ν
2ν2ν−1Γ(ν)ds
=Jν+β(λ)
λβλν
2νΓ(ν + 1)+
∫ λ0
Jν+β+1(s)
sβsν
2νΓ(ν + 1)ds.
Therefore we have the conclusion.
Using Lemma 3.13 several times, we have the following.
Corollary 3.14. Let d ≥ 3, β > −1, a > 0 and t ≥ 0, and let d] be as in (2.10).
(i) If t > 0, then
∫ 2πaλ0
J d2+β(s)J d
2−1(
tas)
( ta)d2−1sβ
ds
=
∫ 2πaλ0
J d2+β−d]+1(s)J d2−d]
( tas)
( ta)d2−1−d]+1sβ
ds−d]−1∑`=1
J d2+β−`(2πaλ)J d
2−`(2πtλ)
( ta)d2−`(2πaλ)β
,
Multiple Fourier series and lattice point problems 27
where the first term of the right hand side in the above equation is equal to
∫ 2πaλ0
Jβ+ 12(s)J− 1
2( tas)
( ta)−
12 sβ
ds, if d is odd,
∫ 2πaλ0
Jβ+1(s)J0(tas)
sβds, if d is even.
(ii) If t = 0, then
∫ 2πaλ0
J d2+β(s)s
d2−1
2d2−1Γ(d
2)sβ
ds
=
∫ 2πaλ0
J d2+β−d]+1(s)s
d2−d]
2d2−d]Γ(d
2− d] + 1)sβ
ds−d]−1∑`=1
(πaλ)d2−`J d
2+β−`(2πaλ)
Γ(d2− `+ 1)(2πaλ)β
,
where the first term of the right hand side in the above equation is equal to
√2
π
∫ 2πaλ0
Jβ+ 12(s)
sβ+12
ds, if d is odd,
∫ 2πaλ0
Jβ+1(s)
sβds, if d is even.
4 Fourier inversion for the function Uβ,a(x)
Recall that Uβ,a(x) = φβ,a(|x|) with
φβ,a(t) =
{(a2 − t2)β, 0 ≤ t < a,0, t ≥ a.
Then φβ,a and Uβ,a have the following properties:
(i) φβ,a is continuous and bounded variation if and only if β > 0;
(ii) φ0,a is discontinuous and bounded variation and U0,a is the indicator function
of the ball in Rd centered at the origin with the radius a;
(iii) φβ,a is not bounded variation if and only if β < 0;
(iv) Uβ,a is integrable on Rd if and only if β > −1 for all dimensions d.
28 S. Kuratsubo and E. Nakai
In this section, for β > −1, we investigate the behavior of the Fourier sphericalpartial integral σλ(Uβ,a)(x) as λ→∞.
For a > 0, let
Ẽa = {x ∈ Rd : x 6= 0 and |x| 6= a}, (4.1)
G̃a = {x ∈ Rd : |x| = a}. (4.2)
Then Rd = {0} ∪ Ẽa ∪ G̃a.
Theorem 4.1. Let d ≥ 1, a > 0 and β > −1. Then
σλ(Uβ,a)(x) = 2βΓ(β + 1)a2β
∫ 2πaλ0
J d2−1(
|x|as)J d
2+β(s)(
|x|a
) d2−1sβ
ds, (4.3)
for all x ∈ Rd and λ > 0. Moreover, σλ(Uβ,a) has the following properties:
(i) At x = 0,
σλ(Uβ,a)(0) = Uβ,a(0)− P [d]β,a cos(
2πaλ− d− 1 + 2β4
π
)λd−32−β
+O(λd−52−β) as λ→∞, (4.4)
where P[d]β,a is the constant defined by (1.14). Consequently,
(a) if β > (d− 3)/2, then
σλ(Uβ,a)(0) = Uβ,a(0) +O(λd−32−β) as λ→∞,
(b) if −1 < β ≤ (d− 3)/2, then σλ(Uβ,a)(x) reveals the Pinsky phenomenon,that is,
lim infλ→∞
σλ(Uβ,a)(0)− Uβ,a(0)λ(d−3)/2−β
= −P [d]β,a,
lim supλ→∞
σλ(Uβ,a)(0)− Uβ,a(0)λ(d−3)/2−β
= P[d]β,a.
(ii) For x ∈ G̃a,
limλ→∞
σλ(Uβ,a)(x)
λ−β= Lβ,a,
Multiple Fourier series and lattice point problems 29
where Lβ,a is the constant defined by (1.15), and consequently,
limλ→∞
σλ(Uβ,a)(x) =
0, β > 0,1
2, β = 0,
∞, −1 < β < 0.
(iii) For x ∈ Ẽa,
σλ(Uβ,a)(x) = Uβ,a(x) +O(λ−β−1) as λ→∞,
where the last term O(λ−β−1) is uniform on any compact subset of Ẽa.
(iv) If β > 0, then
limλ→∞
σλ(Uβ,a)(x) = Uβ,a(x) for x ∈ Ẽa ∪ G̃a
where the convergence is uniform on any compact subset of Ẽa ∪ G̃a andσλ(Uβ,a) does not reveal the Gibbs-Wilbraham phenomenon.
(v) If −1 < β ≤ 0, then
limλ→∞
σλ(Uβ,a)(x) = Uβ,a(x) for x ∈ Ẽa,
where the convergence is uniform on any compact subset of Ẽa, and σλ(Uβ,a)
reveals a phenomenon like the Gibbs-Wilbraham phenomenon near G̃a. More
precisely, the following holds: For each x0 ∈ G̃a, let {x±λ } be the sequenceswhich satisfy lim
λ→∞x±λ = x0 and |x
±λ | = a∓ (2± β)/(4λ). Then
limλ→∞
σλ(Uβ,a)(x+λ )− Uβ,a(x
+λ )
λ−β= G+β,a,
and
limλ→∞
σλ(Uβ,a)(x−λ )− Uβ,a(x
−λ )
λ−β= G−β,a,
where G±β,a are the constants defined by (1.18).
Remark 4.1. The function Uβ,a is piecewise smooth in the sense of Pinsky [34] if
and only if β is a nonnegative integer. In this case the result in Theorem 4.1 (i) is
contained in [34, Theorem 1a].
30 S. Kuratsubo and E. Nakai
In the following we first prove (4.3) in Subsection 4.1. Then, using (4.3), we
prove (i)–(v) of Theorem 4.1 in Subsections 4.2–4.6, respectively. Let
U[d]β,a,λ(t) = 2
βΓ(β + 1)a2β∫ 2πaλ0
J d2−1(
tas)J d
2+β(s)(
ta
) d2−1sβ
ds. (4.5)
Then (4.3) means that
σλ(Uβ,a)(x) = U[d]β,a,λ(|x|). (4.6)
4.1 Proof of (4.3)
By (2.19) the Fourier transform of Uβ,a(x) is expressed by the following:
Ûβ,a(ξ) = Γ(β + 1)Dβ(a2 : ξ) = Γ(β + 1)ad+2β
πβ
J d2+β(2πa|ξ|)
(a|ξ|) d2+β. (4.7)
Since Ûβ,a(ξ) is a radial function, using (2.5), we have
σλ(Uβ,a)(x) =
∫|ξ| −1, a > 0 and λ > 0. Then
U[1]β,a,λ(0) = a
2β − P [1]β,a cos(
2πaλ− β2π
)λ−β−1 +O(λ−β−2) as λ→∞.
U[2]β,a,λ(0) = a
2β − P [2]β,a cos(
2πaλ− 2β + 14
π
)λ−β−
12 +O(λ−β−
32 ) as λ→∞.
Multiple Fourier series and lattice point problems 31
Lemma 4.3. Let d ≥ 3, β > −1, a > 0 and λ > 0, and let
P [d]β,a,λ(t) = −Γ(β + 1)aβ
(πλ)β
d]−1∑`=1
J d2−`+β(2πaλ)J d
2−`(2πtλ)(
ta
) d2−`
, t ≥ 0, (4.9)
where d] is defined by (2.10). Then
U[d]β,a,λ(t) =
{U
[1]β,a,λ(t) + P
[d]β,a,λ(t), if d is odd,
U[2]β,a,λ(t) + P
[d]β,a,λ(t), if d is even,
t ≥ 0. (4.10)
Moreover, if t 6= 0, then P [d]β,a,λ(t) = O(λ−β−1) as λ→∞, If t = 0, then
P [d]β,a,λ(0) = −P[d]β,a cos
(2πaλ− d− 1 + 2β
4π
)λd−32−β +O(λ
d−52−β) as λ→∞.
(4.11)
Remark 4.2. (i) In Lemma 4.2, if β = −1/2, then
U[2]−1/2,a,λ(0) =
1− cos (2πaλ)a
+O(λ−1) as λ→∞. (4.12)
since P[2]−1/2,a = a
−1. This shows the amplitude of Pinsky phenomenon which was
found by Taylor [44].
(ii) Recently, Grafakos and Teschl [7] showed some related results with Lemma 4.3.
Proof of Lemma 4.2. Let d = 1. By (4.5) with (2.1) and Corollary 3.2 we have
U[1]β,a,λ(0) = 2
βΓ(β + 1)a2β√
2
π
∫ 2πaλ0
J 12+β(s)
s12+β
ds
and
a2β = 2βΓ(β + 1)a2β√
2
π
∫ ∞0
J 12+β(s)
s12+β
ds,
respectively. Then by Lemma 3.3 we have
U[1]β,a,λ(0)− a
2β = −2βΓ(β + 1)a2β√
2
π
∫ ∞2πaλ
J 12+β(s)
s12+β
ds
= −P [1]β,a cos(
2πaλ− β2π
)λ−β−1 +O(λ−β−2).
32 S. Kuratsubo and E. Nakai
Let d = 2. By (4.5) with (2.1) we have
U[2]β,a,λ(0) = 2
βΓ(β + 1)a2β∫ 2πaλ0
J1+β(s)
sβds
= 2βΓ(β + 1)a2β[−Jβ(s)
sβ
]2πaλ0
= a2β − a2βΓ(β + 1)Jβ(2πaλ)(πaλ)β
= a2β − P [2]β,a cos(
2πaλ− 2β + 14
π
)λ−β−1/2 +O(λ−β−3/2).
Then the proof is complete.
Proof of Lemma 4.3. The equation (4.10) follows from Collorary 3.14 and the defi-
nition of U[d]β,a,λ(t) (see (4.5)). If t 6= 0, then from (2.17) it follows that
J d2−`+β(2πaλ)J d
2−`(2πtλ) = O(λ
−1), ` = 1, . . . , d] − 1.
Then we have P [d]β,a,λ(t) = O(λ−β−1). If t = 0, then by (2.1) we regard
J d2−`+β(2πaλ)J d
2−`(2πtλ)(
ta
) d2−`
=(πaλ)
d2−`
Γ(d2− `+ 1)
J d2−`+β(2πaλ), ` = 1, . . . , d] − 1.
Then, using (2.17), we have
P [d]β,a,λ(0)
= −Γ(β + 1)aβ
(πλ)β
d]−1∑`=1
(πaλ)d2−`
Γ(d2− `+ 1)
J d2−`+β(2πaλ)
= −Γ(β + 1)aβ
(πλ)β(πaλ)
d2−1
Γ(d2)
J d2−1+β(2πaλ) +O(λ
d−52−β)
=
(−Γ(β + 1)
Γ(d2)
ad−32
+βπd−42−β
)λd−32−β cos
(2πaλ− d− 1 + 2β
4π
)+O(λ
d−52−β).
This is the conclusion.
Multiple Fourier series and lattice point problems 33
4.3 Proof of Theorem 4.1 (ii)
For x ∈ G̃a, by (4.5) and (4.6) we have
σλ(Uβ,a)(x) = U[d]β,a,λ(a) = 2
βΓ(β + 1)a2β∫ 2πaλ0
J d2+β(s)J d
2−1(s)
sβds.
Then, using Corollary 3.12 with µ = d2− 1, we have
U[d]β,a,λ(a) = Lβ,aλ
−β +
{O(λ−β−1), β ≥ 0,O(1), −1 < β < 0,
as λ→∞, (4.13)
which shows the conclusion.
4.4 Proof of Theorem 4.1 (iii)
Let x ∈ Ẽa and |x| = t. Then 0 < t < a or t > a. In this case, by (4.5), (4.6) andCorollary 3.5, we have
σλ(Uβ,a)(x)− Uβ,a(x) = U [d]β,a,λ(t)− φβ,a(t)
= −2βΓ(β + 1)a2β∫ ∞2πaλ
J d2−1(
tas)J d
2+β(s)(
ta
) d2−1sβ
ds.
Using Lemma 3.8 with µ = d2− 1, we have
U[d]β,a,λ(t)− φβ,a(t)
= −2βΓ(β + 1)aβ
π
(at
) d−12ψβ(λ, a− t) +
(at
) d−12(
1 +a
t
)O(λ−β−1). (4.14)
Since ψβ(λ, a− t) = |a− t|−1O(λ−β−1) by Lemma 3.8 (i),
U[d]β,a,λ(t)− φβ,a(t) = O(λ
−β−1)
uniformly on any closed interval in (0, a) ∪ (a,∞). This shows the conclusion.
4.5 Proof of Theorem 4.1 (iv)
Let β > 0. By (4.13) we have
U[d]β,a,λ(a)− φβ,a(a) = U
[d]β,a,λ(a) = Lβ,aλ
−β +O(λ−β−1).
34 S. Kuratsubo and E. Nakai
If 0 < t < a or t > a, then by (4.14) and Lemma 3.7 (ii) we have
U[d]β,a,λ(t)− φβ,a(t) =
(at
) d−12O(λ−β) +
(at
) d−12(
1 +a
t
)O(λ−β−1),
where the terms O(λ−β) and O(λ−β−1) are uniform with respect to t. Hence, U[d]β,a,λ(t)
converges to φβ,a(t) uniformly on any closed interval in (0,∞). That is, σλ(Uβ,a)(x)converges to Uβ,a(x) uniformly on any compact subset of Ẽa ∪ G̃a = Rd \ {0},and consequently, it does not reveal any phenomenon like the Gibbs-Wilbraham
phenomenon.
4.6 Proof of Theorem 4.1 (v)
Let −1 < β ≤ 0. The first assertion is already shown by (iii). Next we show theGibbs-Wilbraham phenomenon. Here, we recall that
G±β,a = ∓Γ(β + 1)aβ(2± β)β
π2β
∫ ∞π
sin s
(s± β2π)β+1
ds.
By (4.14) we have
σλ(Uβ,a)(x+λ )− Uβ,a(x
+λ ) = U
[d]β,a,λ(|x
+λ |)− φβ,a(|x
+λ |)
= −2βΓ(β + 1)aβ
π
(a
|x+λ |
) d−12
ψβ(λ, a− |x+λ |) +(
a
|x+λ |
) d−12(
1 +a
|x+λ |
)O(λ−β−1).
Since a− |x+λ | =2+β4λ
, using Lemma 3.7 (iii), we have
−ψβ(λ, a− |x+λ |) =(
2 + β
4λ
)β ∫ ∞π
− sin s(s+ β
2π)β+1
ds.
Observing (a
|x+λ |
) d−12
=
(1− 2 + β
4λa
)− d−12
= 1 +O(λ−1),
we conclude that
σλ(Uβ,a)(x+λ )− Uβ,a(x
+λ )
λ−β
= −2βΓ(β + 1)aβ
π
(2 + β
4
)β ∫ ∞π
sin s
(s+ β2π)β+1
ds(1 +O(λ−1)
)+O(λ−1)
→ G+β,a as λ→∞.
Multiple Fourier series and lattice point problems 35
In a similar way we also have
σλ(Uβ,a)(x−λ )− Uβ,a(x
−λ )
λ−β→ G−β,a as λ→∞.
5 Results related to lattice point problems
The terms ∆j(s : x), j = 0, 1, · · · , are closely related to lattice point problems whichhave been studied by Landau, Jarńık, Szegö, Novák and others, see [5, 13, 15, 16,
26, 27, 46]. Recall that
∆α(s : x) = Dα(s : x)−Dα(s : x), α > −1, s ≥ 0, x ∈ Rd,
where
Dα(s : x) =
1
Γ(α + 1)
∑|m|2 0,
0, s = 0,
x ∈ Rd,
Dα(s : x) =
1
Γ(α + 1)
∫|ξ|2 0,
0, s = 0,
x ∈ Rd.
In this section we consider the behavior of ∆α(s : x) as s→∞. In Stein [39], theestimation of ∆ d−1
2(s : x) was treated. We shall prove the following four lemmas: In
the following f(s) = Ω(g(s)) means f(s) 6= o(g(s)).
Lemma 5.1. Let d ≥ 1. Then, as s→∞,
∆α(s : x) =
O(s
d2− dd+1 ), if α = 0,
O(sd2− dd+1
+ αd+1
+ε) for every ε > 0, if 0 < α ≤ d−12,
O(sd−14
+α2 ), if α > d−1
2,
(5.1)
uniformly with respected to x ∈ Td.
Lemma 5.2. Let d ≥ 5. If 0 ≤ α < (d− 4)/2, then, as s→∞,
∆α(s : x) =
{O(s
d2−1), Ω(s
d2−1) for x ∈ Td ∩Qd,
o(sd2−1) for x ∈ Td \Qd.
(5.2)
If (d− 4)/2 ≤ α ≤ (d− 1)/2, then, for every ε > 0, as s→∞,
∆α(s : x) =
{O(s
d−1+α3
+ε) for x ∈ Td ∩Qd,o(s
d−1+α3
+ε) for x ∈ Td \Qd.(5.3)
36 S. Kuratsubo and E. Nakai
The following lemma gives more precise information on the estimate (5.2) for
x ∈ Td ∩Qd.
Lemma 5.3. Let d ≥ 5, α = 0 and β > −1. For k ∈ N = {1, 2, · · · }, let
`k = `k(d, β) =1
2a
(k +
d+ 2β + 1
4− 1
4
), (5.4)
mk = mk(d, β) =1
2a
(k +
d+ 2β + 1
4+
1
4
). (5.5)
Then, for large k ∈ N, there exists sk ∈ [`2k,m2k] such that
|∆0(sk : x)| ≥ C(x)sd2−1
k , x ∈ Td ∩Qd, (5.6)
where C(x) is a positive constant dependent on x, but independent of sk for large k.
Lemma 5.4. Let d ≥ 4 and 0 ≤ α ≤ (d− 1)/2. Then, for every ε > 0, as s→∞,
∆α(s : x) = O(sd4+ d−2
2(d−1)α+ε) for a.e.x ∈ Td. (5.7)
To prove above four lemmas we state known results (Theorems 5.5–5.9): See also
Jarńık [14] for Theorem 5.5. For α ≥ 0, let
Pα(s : x) = Dα(s : x)−πd2 s
d2+α
Γ(d2
+ α + 1)δ(x), x ∈ Rn, s ≥ 0, (5.8)
where δ(x) is the indicator function of Zd.
Theorem 5.5 (Landau [25, 27]). Let d ≥ 2. Then, for x ∈ Rd,
Pα(s : x) =
O(s
d2+α− d
d+1−2α ), if 0 ≤ α < d−12,
O(sd−12 log s), if α = d−1
2,
O(sd−14
+α2 ), if α > d−1
2.
(5.9)
Theorem 5.6 (Novák [32]). Let d ≥ 5, and let 0 ≤ α < (d− 4)/2. Then
Pα(s : x) =
O(s
d2−1), Ω(s
d2−1) for x ∈ Qd,
o(sd2−1) for x /∈ Qd,
O(sd4+α
2 logτ s) for a.e.x ∈ Rd,(5.10)
where τ = 3d if α = 0 and τ = 3d− 1 if α > 0.
Multiple Fourier series and lattice point problems 37
Theorem 5.7 (Novák [30]). Let d ≥ 3. Then, for all x ∈ Qd, there exists a positiveconstant Kd(x) such that
∫ s0
|P0(t : x)|2 dt =
Kd(x)s
2 log s+O(s2 log1/2 s), if d = 3,
Kd(x)s3 +O(s5/2 log s), if d = 4,
Kd(x)s4 +O(s3 log2 s), if d = 5,
Kd(x)sd−1 +O(sd−2), if d ≥ 6.
(5.11)
Remark 5.1. In Theorem 5.7 the positive constant Kd(x) is given explicitly for
each x ∈ Qd, see [22]. For example, if d ≥ 4 and x = 0, then
Kd(0) =πd(2d + 8)ζ(d− 2)
12(d− 1)(2d − 1)ζ(d)Γ2(d/2),
where ζ is the Riemann’s zeta function, see [46].
Theorem 5.8 (Kuratsubo [17]). Let d ≥ 2 . Then, for every τ > 3/2,
P0(s : x) = O(sd4 logτ s) for a.e.x.
Remark 5.2. Theorems 5.5–5.8 valid for ∆α(s : x) instead of Pα(s : x), if x ∈ Td.Actually,
∆α(s : x)− Pα(s : x) =
{0, if x = 0,
−Dα(s : x) = O(sd−14
+α2 ), if x ∈ Td \ {0},
since (see (2.19) and (2.17))
Dα(s, x) =
πd2 s
d2+α
Γ(d2
+ α + 1), x = 0,
sd2+α
πα
J d2+α(2π
√s|x|)
(√s|x|) d2+α
= O(sd−14
+α2 ), x 6= 0.
Therefore,
|∆α(s : x)| = |Pα(s : x)|+O(sd−14
+α2 ), if x ∈ Td.
Remark 5.3. Let d = 1. Then
∆α(s : x) =
{O(1), if α = 0,
O(sα2 ), if α > 0,
(5.12)
38 S. Kuratsubo and E. Nakai
uniformly with respect to x ∈ T. Actually, for all s > 0, choosing N ∈ N such thatN <
√s ≤ N + 1, we have by an elementary calculation
∆0(s : x) =sin(2π(N + 1
2)x)
sin πx− sin(2π
√sx)
πx= O(1).
For α > 0, from (2.19) and (2.20) it follows that
∆α(s : x) =s
12+α
πα
∑m∈Z, m6=0
J 12+α(2π
√s|x−m|)
(√s|x−m|) 12+α
.
By (2.17) we have
|J 12+α(2π
√s|x−m|)|
(√s|x−m|) 12+α
≤ C(√s|x−m|)1+α
,
for some positive constant C. Since |x −m| ≥ 1/2 for x ∈ T and m 6= 0, the sumconverges absolutely and ∆α(s : x) = O(s
α2 ).
The following is the Riesz’ convexity theorem. (see [40, page 285] and [3, page
13]).
Theorem 5.9. Let 0 ≤ α0 < α1 < ∞. For x ∈ Rd, let V0(s : x) and V1(s : x) betwo positive nondecreasing functions with respect to s > 0. Assume that
|∆αi(s : x)| ≤ Vi(s : x), i = 0, 1.
Then, for 0 ≤ θ ≤ 1,
|∆(1−θ)α0+θα1(s : x)| ≤ CV0(s : x)1−θV1(s : x)θ,
where C is a positive constant dependent on α1, α0 and θ, and independent of s and
x.
Proof of Lemma 5.1. The case d = 1 has been already proven in Remark 5.3. Then
we consider the case d ≥ 2. In general, for a function f : [0,∞)→ R, the differenceof f with h > 0 is defined by
δhf(s) = δ1hf(s) = f(s+ h)− f(s),
Multiple Fourier series and lattice point problems 39
and
δk+1h f(s) = δkhf(s+ h)− δkhf(s), k = 1, 2, · · · .
Then
δkhf(s) =
∫ s+hs
ds1 · · ·∫ sk−1+hsk−1
f (k)(sk) dsk.
Let k = d] as in (2.10) and use this relation for f(s) = ∆k(s : x). Then, from
(2.18) and (2.21) we see that
δkh∆k(s : x)− hk∆0(s : x)
=
∫ s+hs
ds1 · · ·∫ sk−1+hsk−1
(∆0(sk : x)−∆0(s : x)) dsk
=
∫ s+hs
ds1 · · ·∫ sk−1+hsk−1
((D0(sk : x)−D0(s : x))− (D0(sk : x)−D0(s : x))
)dsk
=
∫ s+hs
ds1 · · ·∫ sk−1+hsk−1
∑s≤|m|2
40 S. Kuratsubo and E. Nakai
where vd is the volume of the d-dimensional unit ball. Hence∣∣δkh∆k(s : x)− hk∆0(s : x)∣∣≤∫ s+hs
ds1 · · ·∫ sk−1+hsk−1
∣∣∣∣∣∣∑
s≤|m|2
Multiple Fourier series and lattice point problems 41
Therefore, combining (5.13) and (5.14), we have
|∆0(s : x)| ≤ h−k(∣∣δkh∆k(s : x)− hk∆0(s : x)∣∣+ ∣∣δkh∆k(s : x)∣∣)
≤ C(sd2− dd+1 + hs
d2−1)
+ C( sh
) d−12, if kh ≤ s.
Then, setting h as hsd2−1 = (s/h)
d−12 , that is, h = s
1d+1 , we have that
|∆0(s : x)| ≤ Csd2− dd+1 , if s ≥ k
d+1d ,
where C is a positive constant independent of x ∈ Td and s ≥ k d+1d .If α > (d− 1)/2, then we have by (2.19) and (2.20) that
∆α(s : x) =sd2+α
πα
∑m∈Zd,m 6=0
J d2+α(2π
√s|x−m|)
(√s|x−m|) d2+α
.
Moreover, since |x−m| ≥ 1/2 for x ∈ Td and m 6= 0, the sum converges absolutelyand
|∆α(s : x)| ≤ Csd−14
+α2 ,
where C is a positive constant independent of x ∈ Td and s > 0. Therefore, we have
∆α(s : x) =
{O(s
d2− dd+1 ), if α = 0,
O(sd−14
+α2 ), if α > d−1
2,
uniformly with respect to x ∈ Td.Applying Theorem 5.9 as
α0 = 0, α1 =d− 1
2and α = θα1,
we have
∆α(s : x) = O(sd2− dd+1
+ αd+1
+ε) for every ε > 0, if 0 < α ≤ d− 12
,
since
(1− θ)(d
2− dd+ 1
)+ θ
(d− 1
4+α12
)=d
2− dd+ 1
+α
d+ 1.
Then the proof is complete.
42 S. Kuratsubo and E. Nakai
Remark 5.4. We have the following comparison between Lemma 5.1 and Landau’s
estimate (5.9):
d
2− dd+ 1
+α
d+ 1<d
2+ α− d
d+ 1− 2α, if 0 < α <
d− 12
.
Proof of Lemma 5.2. If 0 ≤ α < (d−4)/2, then, by Theorem 5.6 and Remark 5.2 wehave (5.2) immediately. Next we show (5.3). By Theorems 5.5, 5.6 and Remark 5.2
we have
∆α(s : x) =
{O(s
d2−1) for x ∈ Td ∩Qd,
o(sd2−1) for x ∈ Td \Qd,
if α <d− 4
2,
and
∆α(s : x) = O(sd−12 log s) for x ∈ Td, if α = d− 1
2.
Applying Theorem 5.9 as
α0 =d− 4
2, α1 =
d− 12
and α = (1− θ)α0 + θα1,
we have
∆α(s : x) = O(sd−1+α
3+ε) for every ε > 0, if
d− 42≤ α ≤ d− 1
2,
since
(1− θ)(d
2− 1)
+ θ
(d− 1
2
)=d− 1 + α
3.
Proof of Lemma 5.3. By Theorem 5.7 and Remark 5.2 we have∫ s0
|∆0(t : x)|2 dt = Kd(x)sd−1 +O(sd−2 logτ s),
where τ = 2 if d = 5, τ = 0 if d ≥ 6. Using
m2k − `2k =1
(2a)2
(k +
d+ 2β + 1
4
),
m2(d−1)k − `
2(d−1)k =
d− 1(2a)2(d−1)
k2d−3 +O(k2d−4),
we have
1
m2k − `2k
∫ m2k`2k
|∆0(t : x)|2 dt =1
m2k − `2k
(∫ m2k0
|∆0(t : x)|2 dt−∫ `2k0
|∆0(t : x)|2 dt
)
= Kd(x)m
2(d−1)k − `
2(d−1)k
m2k − `2k+O(k2(d−2)−1 logτ k)
=d− 1
(2a)2(d−2)Kd(x) k
2d−4 +O(k2(d−2)−1 logτ k).
Multiple Fourier series and lattice point problems 43
Hence we can take K̃d(x) such that
1
m2k − `2k
∫ m2k`2k
|∆0(t : x)|2 dt ≥ K̃d(x)2m2d−4k for large k.
Therefore, for large k, there exists sk ∈ [`2k,m2k] such that
|∆0(sk : x)| ≥ K̃d(x)md−2k ≥ K̃d(x)sd2−1
k .
Proof of Lemma 5.4. By Theorems 5.5, 5.8 and Remark 5.2 we have
∆α(s, x) =
{O(s
d4 logτ s), if α = 0,
O(sd−12 log s), if α = d−1
2,
for a.e.x ∈ Td.
Applying Theorem 5.9 as
α0 = 0, α1 =d− 1
2and α = θα1,
we have
∆α(s, x) = O(sd4+ d−2
2(d−1)α+ε) for every ε > 0, if 0 < α <d− 1
2,
since
(1− θ)d4
+ θd− 1
2=d
4+
d− 22(d− 1)
α.
Remark 5.5. We have the following comparison between Lemma 5.1 and Lemma 5.4:
d
4+
d− 22(d− 1)
α <d
2− dd+ 1
+α
d+ 1, if d ≥ 4 and 0 ≤ α < d− 1
2.
6 Proof of the main theorems
In this section we prove the pointwise convergence of the Fourier series of the func-
tion uβ,a(x) described in the main theorems (Theorems 1.1–1.5). First we state a
generalized Hardy’s identity (Theorem 6.1) and three lemmas (Lemmas 6.4–6.6).
Next, using the generalized Hardy’s identity and three lemmas, we will prove the
main theorems in Subsection 6.2. The proofs of Theorem 6.1 and Lemmas 6.4–6.6
are in Subsections 6.3, 6.4, 6.5 and 6.6, respectively.
44 S. Kuratsubo and E. Nakai
6.1 Generalized Hardy’s identity and three lemmas
Recall that uβ,a(x) is the periodization of Uβ,a(x) = φβ,a(|x|) with
φβ,a(t) =
{(a2 − t2)β, 0 ≤ t < a,0, t ≥ a.
That is,
uβ,a(x) =∑m∈Zd
Uβ,a(x+m) =∑
|x+m| 1/2 and then
supx∈Td
∑m∈Zd\{0}
1
|x−m| d+12 +d]
d. By Lemma 3.7 (i) we have that
ψβ(λ, |a− |x−m||) = |a− |x−m||−1O(λ−β−1) as λ→∞.
Letting
Ma(x) = maxm∈Zd\{0}, |x−m|6=a
|a− |x−m||−1,
we have ∑m∈Zd\{0}, |x−m|6=a
(a
|x−m|
) d+12
+d] ∣∣ψβ(λ, |a− |x−m||)∣∣≤ C
∑m∈Zd\{0}
1
|x−m| d+12 +d]
Ma(x)λ−β−1
Multiple Fourier series and lattice point problems 45
Hence,
Gβ,a(λ, x) = Ma(x)O(λ−β−1) as λ→∞. (6.6)
Next, recall that
rd(a : x) =∑
m∈Zd, |x−m|=a
1, x ∈ Td,
and let
r̃d(a : x) =∑
m∈Zd\{0}, |x−m|=a
1, x ∈ Td. (6.7)
Then
r̃d(a : 0) = rd(a : 0) and r̃d(a : x) = rd(a : x) = 0 for x ∈ Ea. (6.8)
Theorem 6.1 (Generalized Hardy’s identity). Let d ≥ 1, β > −1 and a > 0. Then
Sλ(uβ,a)(x) = uβ,a(x) +(σλ(Uβ,a)(x)− Uβ,a(x)
)+
(Gβ,a(λ : x) + r̃d(a : x)
(Lβ,a + o(1)
)λ−β
)+Kβ,a(λ2 : x) +O(λ−β−1) as λ→∞ (6.9)
for all x ∈ Td.
If 0 < a < 1/2, then r̃d(a : x) = 0 and uβ,a(x) = Uβ,a(x) in x ∈ Td. Combiningthese and (6.6), we have the following corollary of Theorem 6.1.
Corollary 6.2. Let d ≥ 1, β > −1 and 0 < a < 1/2. Then
Sλ(uβ,a)(x) = σλ(Uβ,a)(x) +Kβ,a(λ2 : x) +O(λ−β−1) as λ→∞ (6.10)
for all x ∈ Td.
If d = 2, β = 0 and a > 0, then U0,a(x) is the indicator function of the ball
{x ∈ R2 : |x| < a}. In this case, from Theorem 4.1 (i) (a), (ii) and (iii) it followsthat
limλ→∞
(σλ(U0,a)(x)− U0,a(x)
)=
L0,a =1
2, |x| = a,
0, |x| 6= a,x ∈ R2.
Then we conclude that
limλ→∞
(σλ(U0,a)(x)− U0,a(x)
)+ r̃2(a : x)L0,a =
1
2r2(a : x) =
1
2
∑|x+m|=a
1, x ∈ T2.
46 S. Kuratsubo and E. Nakai
By (6.6) and Lemma 6.4 bellow we also have
G0,a(λ : x) +K0,a(λ2 : x)→ 0 as λ→∞, x ∈ T2.
Since u0,a(x) =∑|x+m| 0.
Then
limλ→∞
Sλ(u0,a)(x) =∑
|x+m| 0. By (1.8) and (4.7) together with (2.1)
we have û0,a(0) = πa2 and û0,a(m) = aJ1(2πa|m|)/|m| for m 6= 0. Then
Sλ(u0,a)(x) = πa2 + a
∑0
Multiple Fourier series and lattice point problems 47
and Sλ(u− 12,a)(x) behaves about as nicely on the torus as does σλ(U− 1
2,a)(x) on the
Euclidean space. Now, from (6.12) we know that the set of these discrete values of
λ is
{λ > 0 : sin(2πaλ) = 0}.
For example, if a = 3/(4π), then λ = (2π/3)n for n = 1, 2, . . . , which are correspon-
dent with Figure 7A (n = 6) and Figure 7F (n = 7) in [43]. For the coefficient of
sin(2πaλ), we can calculate by Lemma 5.1 as
∆0(λ2 : x)
λ= O(λ−
13 ).
In the rest of this subsection we state three lemmas. Recall that
c(d) = d− 2dd+ 1
− d+ 12
=d− 5
2+
2
d+ 1=d(d− 4)− 1
2(d+ 1)=d− 3
2− d− 1d+ 1
.
Lemma 6.4. Let d ≥ 1, β > −1 and a > 0. Then
Kβ,a(λ2 : x) = O(λc(d)−β) uniformly on Td.
Consequently, if 1 ≤ d ≤ 4 and β > c(d), then
limλ→∞Kβ,a(λ2 : x) = 0 uniformly on Td,
and, if d ≥ 2, then
limλ→∞
|Kβ,a(λ2 : x)|λd−32−β
= 0 uniformly on Td.
Lemma 6.5. Let d ≥ 5, β > −1 and a > 0. Then
limλ→∞
|Kβ,a(λ2 : x)|λd−52−β
= 0, if x ∈ Td \Qd,
0 < lim supλ→∞
|Kβ,a(λ2 : x)|λd−52−β
−1/2 and a > 0. Then
limλ→∞Kβ,a(λ2 : x) = 0, a.e.x ∈ Td.
6.2 Proof of Theorems 1.1–1.5
In this section, using Theorem 6.1 and Lemmas 6.4–6.6, we prove the main theorems.
48 S. Kuratsubo and E. Nakai
6.2.1 Proof of Theorem 1.1
For all β > −1, from (6.6) and Lemma 6.4 it follows that
Gβ,a(λ : 0) = O(λ−β−1) and Kβ,a(λ2 : 0) = O(λ
d−32−β),
respectively. Since r̃d(a, 0) = rd(a, 0) as in (6.8), by Theorem 6.1 we have
Sλ(uβ,a)(0) = uβ,a(0) +(σλ(Uβ,a)(0)− Uβ,a(0)
)+ rd(a : 0) (Lβ,a + o(1))λ
−β +O(λd−32−β). (6.13)
If β > (d− 3)/2, then, using Theorem 4.1 (i) (a), we have
σλ(Uβ,a)(0)− Uβ,a(0) = O(λd−32−β).
Combining this with (6.13), we have
Sλ(uβ,a)(0) = uβ,a(0) + rd(a : 0) (Lβ,a + o(1))λ−β +O(λ
d−32−β),
which shows (i).
If d ≥ 2 and −1 < β ≤ (d− 3)/2, then, using (4.4) in Theorem 4.1 (i), we have
σλ(Uβ,a)(0)− Uβ,a(0) = −P [d]β,a cos(
2πaλ− d− 1 + 2β4
π
)λd−32−β +O(λ
d−52−β),
which shows (ii).
6.2.2 Proof of Theorem 1.3
By Theorem 4.1 (ii) and (iii) we have
limλ→∞
σλ(Uβ,a)(x)− Uβ,a(x)λ−β
=
{Lβ,a, x ∈ G̃a,0, x ∈ Ẽa,
which shows
limλ→∞
σλ(Uβ,a)(x)− Uβ,a(x)λ−β
+ r̃d(a : x)Lβ,a = rd(a : x)Lβ,a, x ∈ Ea ∪Ga. (6.14)
Proof of (i) (a) Let 1 ≤ d ≤ 4. By (6.6) and Lemma 6.4 we have
Gβ,a(λ : x)
λ−β= O(λ−1),
Kβ,a(λ2 : x)λ−β
= O(λd−52
+ 2d+1 ) = o(1).
Multiple Fourier series and lattice point problems 49
By Theorem 6.1 and (6.14) we have
Sλ(uβ,a)(x)− uβ,a(x)λ−β
=σλ(Uβ,a)(x)− Uβ,a(x)
λ−β+Gβ,a(λ : x)
λ−β+ r̃d(a : x)
(Lβ,a + o(1)
)+Kβ,a(λ2 : x)
λ−β+O(λ−1)
→ rd(a : x)Lβ,a as λ→∞,
which shows the conclusion.
Proof of (i) (b) Let β > c(d) = d−52
+ 2d+1
. Then by Lemma 6.4 we have
Kβ,a(λ2 : x) = O(λd−52
+ 2d+1−β) = o(1) uniformly on Td.
If x ∈ Ea, then r̃d(a : x) = 0 as (6.8). By Theorem 4.1 (iii) and (6.6) we have
limλ→∞
(σλ(Uβ,a)(x)− Uβ,a(x)) = 0, limλ→∞
Gβ,a(λ : x) = 0
uniformly on any compact set in Ea. Hence, by Theorem 6.1 we have
Sλ(uβ,a)(x)− uβ,a(x)
= (σλ(Uβ,a)(x)− Uβ,a(x)) +Gβ,a(λ : x) +Kβ,a(λ2 : x) +O(λ−β−1)
→ 0 as λ→∞
uniformly on any compact set in Ea, which shows the conclusion.
Proof of (ii) Let d ≥ 5 and x ∈ Ea ∪Ga. By Theorem 6.1 we have
1
λd−52
(Sλ(uβ,a)(x)− uβ,a(x)
λ−β− rd(a : x)Lβ,a
)=
1
λd−52
(σλ(Uβ,a)(x)− Uβ,a(x)
λ−β+ r̃d(a : x)
(Lβ,a + o(1)
)− rd(a : x)Lβ,a
)+Gβ,a(λ : x)
λd−52−β
+Kβ,a(λ2 : x)λd−52−β
+O(λ−d−52−1).
By (6.14), (6.6) and Lemma 6.5 we have
limλ→∞
1
λd−52
(Sλ(uβ,a)(x)− uβ,a(x)
λ−β− rd(a : x)Lβ,a
)= lim
λ→∞
Kβ,a(λ2 : x)λd−52−β
= 0
50 S. Kuratsubo and E. Nakai
for all x ∈ (Ea ∪Ga) \Qd, and
0 < lim supλ→∞
1
λd−52
∣∣∣∣Sλ(uβ,a)(x)− uβ,a(x)λ−β − rd(a : x)Lβ,a∣∣∣∣
= lim supλ→∞
|Kβ,a(λ2 : x)|λd−52−β
0.
Hence, by (6.6) we haveGβ,a(λ : x
±λ )
λ−β= O(λ−1).
Since x±λ ∈ Ea, r̃d(a : x±λ ) = 0 as in (6.8). By Theorem 6.1 and Theorem 4.1 (v) we
have
Sλ(uβ,a)(x±λ )− uβ,a(x
±λ )
λ−β
=
(σλ(Uβ,a)(x
±λ )− Uβ,a(x
±λ )
λ−β
)+Gβ,a(λ : x
±λ )
λ−β+Kβ,a(λ2 : x±λ )
λ−β+O(λ−1)
→ G±β,a as λ→∞,
which is the conclusion.
6.2.4 Proofs of Theorem 1.5
Let d ≥ 4, β > −1/2 and a > 0. Then by lemma 6.6 we have
limλ→∞Kβ,a(λ2 : x) = 0, a.e.x ∈ Td.
Let x ∈ Ea. Then r̃d(a : x) = 0 as in (6.8). By Theorem 4.1 (iii) and (6.6) we have
σλ(Uβ,a)(x)− Uβ,a(x) = O(λ−β−1) and Gβ,a(λ : x) = O(λ−β−1).
Multiple Fourier series and lattice point problems 51
Then by Theorem 6.1 we have, for a.e.x ∈ Ea,
Sλ(uβ,a)(x)− uβ,a(x)
=(σλ(Uβ,a)(x)− Uβ,a(x)
)+Gβ,a(λ : x) +Kβ,a(λ2 : x) +O(λ−β−1)
→ 0 as λ→∞,
which shows the conclusion, since the measure of Td \ Ea is zero.
6.3 Proof of Theorem 6.1
By Corollary 2.2 and (6.2) we have that, for all x ∈ Td,
Sλ(uβ,a)(x) = σλ(Uβ,a)(x) +Kβ,a(λ2 : x)
+ (−1)d]+1∑
m∈Zd\{0}
∫ λ20
Dd](s : x−m)A(d]+1)
β,a (s) ds.
Using (2.19) and (2.13) we have
Dd](s : x−m)A(d]+1)
β,a (s)
=sd2+d]
πd]
J d2+d]
(2π√s|x−m|)
(√s|x−m|) d2+d]
(−1)d]+1 Γ(β + 1)πβ−(d]+1)
ad2+β+d]+1
J d2+β+d]+1
(2πa√s)
sd2+β+d]+1
= (−1)d]+12βΓ(β + 1)a2βJ d
2+d]
(2π√s|x−m|)J d
2+β+d]+1
(2πa√s)
( |x−m|a
)d2+d](2πa
√s)β
πa√s.
Then, letting u = 2πa√s, we have
(−1)d]+1∫ λ20
Dd](s : x−m)A(d]+1)
β,a (s) ds
= 2βΓ(β + 1)a2β∫ 2πaλ0
J d2+d]
( |x−m|au) J d
2+β+d]+1
(u)
( |x−m|a
)d2+d] uβ
du. (6.15)
If |x−m| 6= a, then Corollary 3.5 shows that
2βΓ(β + 1)a2β∫ ∞0
J d2+d]
( |x−m|au) J d
2+β+d]+1
(u)
( |x−m|a
)d2+d] uβ
du
= Uβ,a(x−m) =
{0, if a < |x−m|,(a2 − |x−m|2)β, if 0 < |x−m| < a.
(6.16)
52 S. Kuratsubo and E. Nakai
If m 6= 0 and |x−m| 6= a, then Lemma 3.8 shows that
2βΓ(β + 1)a2β∫ ∞2πaλ
J d2+d]
( |x−m|au) J d
2+β+d]+1
(u)
( |x−m|a
)d2+d] uβ
du
=2βΓ(β + 1)aβ
π
(a
|x−m|
) d+12
+d]
ψβ(λ, a− |x−m|) +1
|x−m| d+12 +d]O(λ−β−1),
(6.17)
since |x − m| ≥ 1/2 for m 6= 0 and x ∈ Td. Combining (6.15), (6.16) and (6.17),and observing (6.3)–(6.5), we have
(−1)d]+1∑
m∈Zd\{0}, |x−m|6=a
∫ λ20
Dd](s : x−m)A(d]+1)
β,a (s) ds
=
∑m∈Zd\{0}, |x−m|6=a
Uβ,a(x−m)
−Gβ,a(λ : x) +O(λ−β−1).= uβ,a(x)− Uβ,a(x)−Gβ,a(λ : x) +O(λ−β−1).
On the other hand, if m 6= 0 and |x−m| = a, then, using (6.15) and Corollary 3.12,we have
(−1)d]+1∫ λ20
Dd](s : x−m)A(d]+1)
β,a (s) ds
= 2βΓ(β + 1)a2β∫ 2πaλ0
J d2+d]
(u) J d2+β+d]+1
(u)
uβdu
= Lβ,aλ−β +
{O(λ−β−1), β ≥ 0,O(1), −1 < β < 0
= (Lβ,a + o(1))λ−β.
Therefore, we have the conclusion.
6.4 Proof of Lemma 6.4
To estimate
Kβ,a(λ2 : x) =d]∑j=0
(−1)j∆j(λ2 : x)A(j)β,a(λ2),
Multiple Fourier series and lattice point problems 53
we combine the estimates of ∆j(λ2 : x) and A
(j)β,a(λ
2). Firstly, by (2.17) we see that
A(j)β,a(s) = (−1)
jΓ(β + 1)
πβ−jad2+β+j
J d2+β+j(2πa
√s)
s12( d2+β+j)
.
= (−1)jΓ(β + 1)πβ−j+1
ad2+β+j− 1
2
s12( d2+β+j+ 1
2)
cos
(2πa√s− d+ 2β + 2j + 1
4π
)+O(s−
12( d2+β+j+ 3
2)) as s→∞. (6.18)
That is, for some positive constant C,
|A(j)β,a(λ2)| ≤ Cλ−(
d2+β+j+ 1
2), λ ≥ 1, (6.19)
For the terms ∆j(s : x), we use Lemma 5.1, that is,
∆j(λ2 : x) =
O(λd−
2dd+1 ), if j = 0,
O(λd−2dd+1
+ 2jd+1
+ε) for every ε > 0, if 0 < j ≤ d−12,
O(λd−12
+j), if j > d−12.
If j = 0, then
|∆0(λ2 : x)Aβ,a(λ2)| ≤ Cλd−2dd+1λ−(
d2+β+ 1
2) = Cλ
d−52
+ 2(d+1)
−β. (6.20)
If 0 < j < d], then 0 < j ≤ (d− 1)/2 and then, for any small ε > 0,
|∆j(λ2 : x)A(j)β,a(λ2)| ≤ Cλd−
2dd+1
+ 2jd+1
+ελ−(d2+β+j+ 1
2) = Cλ
d−52
+2−j(d−1)
(d+1)−β+ε. (6.21)
If j = d], then j > (d− 1)/2 and then
|∆d](λ2 : x)A(d])
β,a (λ2)| ≤ Cλ
d−12
+d]λ−(d2+β+d]+
12) = Cλ−1−β. (6.22)
Comparing (6.20), (6.21) and (6.22), we have the conclusion.
6.5 Proof of Lemma 6.5
By Lemma 5.2 we have that, as λ→∞,
∆j(λ2 : x) =
O(λd−2) for x ∈ Td ∩Qd and if 0 ≤ j < (d− 4)/2,o(λd−2) for x ∈ Td \Qd and if 0 ≤ j < (d− 4)/2,O(λ
2(d−1+j)3
+ε) for x ∈ Td and if (d− 4)/2 ≤ j ≤ (d− 1)/2.
54 S. Kuratsubo and E. Nakai
Combining this and (6.19), we have the following estimates: If 0 ≤ j < (d − 4)/2and x ∈ Td \ Qd, then, for some decreasing function ϕ(λ) which satisfies ϕ(λ) → 0as λ→∞,
|∆j(λ2 : x)A(j)β,a(λ2)| ≤ Cλd−2ϕ(λ)λ−(
d2+β+j+ 1
2) = Cλ
d−52−β−jϕ(λ).
If 0 ≤ j < (d− 4)/2 and x ∈ Td ∩Qd, then
|∆j(λ2 : x)A(j)β,a(λ2)| ≤ Cλd−2λ−(
d2+β+j+ 1
2) = Cλ
d−52−β−j.
If (d− 4)/2 ≤ j ≤ (d− 1)/2, then, for small ε > 0,
|∆j(λ2 : x)A(j)β,a(λ2)| ≤ Cλ
2(d−1+j)3
+ελ−(d2+β+j+ 1
2) = Cλ
d−76−β− j
3+ε ≤ Cλ−
12−β+ε.
If j = d], then we have (6.22). Comparing these estimates, we have
|Kβ,a(λ2 : x)| ≤
{Cλ
d−52−βϕ(λ) for x ∈ Td \Qd,
Cλd−52−β for x ∈ Td ∩Qd,
which shows
limλ→∞
|Kβ,a(λ2 : x)|λd−52−β
= 0, if x ∈ Td \Qd,
lim supλ→∞
|Kβ,a(λ2 : x)|λd−52−β
Multiple Fourier series and lattice point problems 55
If λ ∈ [`k,mk], then
kπ − 14π ≤ 2πaλ− d+ 2β + 1
4π ≤ kπ + 1
4π,
that is, ∣∣∣∣cos(2πaλ− d+ 2β + 14 π)∣∣∣∣ ≥ 1√2 .
Therefore, if x ∈ Td ∩Qd, then, for λk ∈ [`k,mk],
|∆0(λ2k : x)Aβ,a(λ2k)| ≥ Cβ,aC(x)λd−2k λ−( d
2+β+ 1
2)
k = Cβ,aC(x)λd−52−β
k ,
that is,
0 < lim supλ→∞
|∆0(λ2 : x)Aβ,a(λ2)|λd−52−β
.
The proof is complete.
6.6 Proof of Lemma 6.6
If j = d], then we have (6.22). If 0 ≤ j ≤ (d − 1)/2, then, using Lemma 5.4 and(6.19), we have, for small ε > 0,
|∆j(λ2 : x)A(j)β,a(λ2)| ≤ Cλ
d2+ d−2d−1 j+ελ−(
d2+β+j+ 1
2) = Cλ−
12−β− j
d−1+ε ≤ Cλ−12−β+ε,
λ ≥ 1, a.e.x ∈ Td.
This shows the conclusion.
7 A relation between multiple Fourier series and
lattice point problems
Let d ≥ 1, β > −1 and 0 < a < 1/2. Then Corollary 6.2 has shown that
Sλ(uβ,a)(x) = σλ(Uβ,a)(x) +Kβ,a(λ2 : x) +O(λ−β−1) as λ→∞
for all x ∈ Td. The behavior of σλ(Uβ,a)(x) has been clarified by Theorem 4.1.Therefore, to clarify the behavior of Sλ(uβ,a)(x) we need to investigate the term
Kβ,a(λ2 : x) =d]∑j=0
(−1)j∆j(λ2 : x)A(j)β,a(λ2).
56 S. Kuratsubo and E. Nakai
Since the estimates for the terms A(j)β,a(s) are gotten as (6.19), the convergence prob-
lem depends only on the terms ∆j(s : x), which are connected with the lattice point
problem. Especially, the estimate for ∆0(s : x) is very important and difficult.
First we state the following theorem, which will be proven later:
Theorem 7.1. Let d = 2 or 3, 0 < a < 1/2 and x0 ∈ Td. Then,
Sλ(uβ,a)(x0)− σλ(Uβ,a)(x0) = o(1) as λ→∞ for all β > −1,
if and only if
∆0(s : x0) = O(sd−14
+ε) as s→∞ for all ε > 0. (7.1)
For d = 1 the convergence property of Sλ(uβ,a)(x) and σλ(Uβ,a)(x) are well known
and the estimate (7.1) holds. Actually, we have ∆0(s : x) = O(1) for all x ∈ T, seeRemark 5.3. However, for d = 2 and d = 3, the estimate (7.1) is an open problem,
see Remarks 7.1 and 7.2, respectively. For d ≥ 4, see Remarks 7.3 and 7.4.
Remark 7.1. If d = 2, then the following fact is well known (see [30, Hauptsatz 3]
and Remark 5.2):
C2(x) t14 <
(1
t
∫ t0
|∆0(s : x)|2 ds) 1
2
< D2(x) t14 for all x ∈ T2, (7.2)
where C2(x) and D2(x) are positive constants depending on x ∈ T2. Therefore it isnatural to conjecture that ∆0(s : x) = O(s
14+ε) for all x ∈ T2. However,
“∆0(s : 0) = O(s14+ε) for all ε > 0”
is an open problem as famous Hardy’s conjecture on Gauss’s circle problem, since
∆0(s : 0) = D0(s : 0)−D0(s : 0) =∑|m|2
Multiple Fourier series and lattice point problems 57
Remark 7.2. If d = 3 and for all x ∈ T3 then (see [30, Hauptsatz 3] and Remark 5.2)
C3(x) t12 <
(1
t
∫ t0
|∆0(s : x)|2 ds) 1
2
< D3(x) t12 log
12 t for all x ∈ T3, (7.3)
where C3(x) and D3(x) are positive constants depending on x ∈ T3. Therefore it isnatural to conjecture that ∆0(s : x) = O(s
12+ε) for all x ∈ T3. However, this problem
is also very hard. Up to now the best result on this problem is ∆0(s : 0) = O(s21/32+ε)
by D. R. Heath-Brown [11] in 1999.
For d ≥ 4 we have the following theorem. The proof will be given later:
Theorem 7.2. Let d ≥ 4, β > −1 and 0 < a < 1/2. For x0 ∈ Td, if (7.1) holds,then
Sλ(uβ,a)(x0)− σλ(Uβ,a)(x0) = o(1) as λ→∞.
Remark 7.3. If d ≥ 4 and x ∈ Td∩Qd, then (see [30, Hauptsatz 1] and Remark 5.2)(1
t
∫ t0
|∆0(s : x)|2 ds) 1
2
= Md(x) td2−1 +O(t
d2− 3
2+ε),
where Md(x) is a positive constant depending on x. Therefore, the estimate (7.1)
fails at x ∈ Td ∩Qd for d ≥ 4.
By Propositions 7.1 and 7.2 we have the following corollary immediately:
Corollary 7.3. Let d ≥ 2, β > −1 and 0 < a < 1/2. Assume that, for all ε > 0,
∆0(s : x) = O(sd−14
+ε) as s→∞ for a.e. x ∈ Td. (7.4)
Then
Sλ(uβ,a)(x)− σλ(Uβ,a)(x) = o(1) as λ→∞ for a.e. x ∈ Td.
Consequently,
limλ→∞
Sλ(uβ,a)(x) = uβ,a(x) for a.e. x ∈ Td.
Remark 7.4. If d = 2 or d = 3, then (7.4) is conjectured naturally by (7.2) and
(7.3). If d ≥ 4, then the following is known (see [31, Theorem 5] and Remark 5.2)(1
t
∫ t0
|∆0(s : x)|2 ds)1/2
= O(td−14 ) for a.e. x.
Therefore (7.4) is conjectured naturally for d ≥ 4 also.
58 S. Kuratsubo and E. Nakai
Proof of Theorem 7.1. First observe that Theorem 6.1 implies that
Sλ(uβ,a)(x0) = σλ(Uβ,a)(x0) + o(1) as λ→∞,
if and only if
Kβ,a(λ2 : x0) =d]∑j=0
(−1)j∆j(λ2 : x0)A(j)β,a(λ2) = o(1). (7.5)
If j = d], then by (6.22) we have
∆d](λ2 : x0)A
(d])
β,a (λ2) = O(λ−1−β).
Note that d] = 1 if d = 2 and d] = 2 if d = 3. For the case d = 3 and j = 1, by
(6.21), we have, for small ε0 > 0,
∆1(λ2 : x0)A
(1)β,a(λ
2) = O(λ−1−β+ε0).
Hence
Kβ,a(λ2 : x0) =
{∆0(λ
2 : x0)Aβ,a(λ2) +O(λ−1−β), d = 2,
∆0(λ2 : x0)Aβ,a(λ
2) +O(λ−1−β+ε0), d = 3.
Since we can take ε0 > 0 as −1− β + ε0 < 0 in the case d = 3, we see that (7.5) isequivalent to
∆0(λ2 : x0)Aβ,a(λ
2) = o(1). (7.6)
(i) Assume that the estimate (7.1) holds. Then, for all β > −1, we can takeε > 0 as −1− β + ε < 0 and
|∆0(λ2 : x0)Aβ,a(λ2)| ≤ Cλd−12
+ελ−(d2+β+ 1
2) = Cλ−1−β+ε,
where we use (7.1) and (6.19). This shows (7.6).
(ii) Conversely, assume that the estimate (7.6) holds for all β > −1. Then by(6.18) we have that
∆0(λ2 : x)
λd2+β+ 1
2
cos
(2πaλ− d+ 2β + 1
4π
)= o(1)
for all β > −1. Now, for all ε > 0, take β(1) and β(2) such that −1 < β(1) <−1 + � = β(2) < β(1) + 1. Then
∆0(λ2 : x)
λd2+β(i)+ 1
2
cos
(2πaλ− d+ 2β(i) + 1
4π
)= o(1) (i = 1, 2). (7.7)
Multiple Fourier series and lattice point problems 59
Let θ0 = (β(2)− β(1))π/4. Then 0 < θ0 < π/4 and
minθ∈R
(max {| cos θ|, | cos(θ − 2θ0)|}) = min−π≤θ≤π
(max {| sin θ|, | sin(θ − 2θ0)|}) ≥ sin θ0.
Therefore,
minλ>0
(max
{∣∣∣∣cos(2πaλ− d+ 2β(i) + 14 π)∣∣∣∣ : i = 1, 2}) ≥ sin θ0.
Combining this and (7.7), and observing that β(1) < β(2), we conclude that
∆0(λ2 : x0)
λd2+β(2)+ 1
2
=∆0(λ
2 : x0)
λd−12
+�= o(1).
This shows the conclusion.
Proof of Theorem 7.2. From Theorem 6.1 it is enough to prove that
Kβ,a(λ2 : x0) =d]∑j=0
(−1)j∆j(λ2 : x0)A(j)β,a(λ2) = o(1).
By the assumption and Lemma 5.1, for all ε > 0,
∆α(s : x0) =
{O(s
d−14
+ε), if α = 0,
O(sd−12
+ε), if α = d−12.
Applying Theorem 5.9 as
α0 = 0, α1 =d− 1
2and α = θα1,
we have
∆α(s : x0) = O(sd−14
+α2+ε), if 0 ≤ α ≤ d− 1
2,
since
(1− θ)d− 14
+ θd− 1
2=d− 1
4+α
2.
Take ε > 0 as −1− β + ε < 0. Then we have by (6.19) and (6.22)
|Kβ,a(λ2 : x0)| ≤d]−1∑j=0
|∆j(λ2 : x0)A(j)β,a(λ2)|+O(λ−1−β)
≤ Cd]−1∑j=0
λd−12
+j+ελ−(d2+β+j+ 1
2) +O(λ−1−β)
= O(λ−1−β+ε)
= o(1).
The proof is complete.
60 S. Kuratsubo and E. Nakai
References
[1] J. Bourgain and N. Watt, Mean square of zeta function, circle problem and
divisor problem revisited, arXiv:1709.04340v1.
[2] L. Brandolini and L. Colzani, Localization and convergence of eigenfunction
expansions, J. Fourier Anal. Appl. 5 (1999), no. 5, 431–447.
[3] K. Chandrasekharan and S. Minakushisundaram, Typical means, Oxford Univ.
Press, 1952.
[4] L. Colzani and M. Vignati The Gibbs Phenomenon for Multiple Fourier Inte-
grals, J. Approximation theory. 80 (1995), 119-131.
[5] F. Fricker, Einführung in die Gitterpunktlehre, Birhäuser, 1982
[6] I. S. Gradshteyn and I. M. Ryzhik, Tables of integrals, series, and products
(Seventh edition), Academic Press, 2007.
[7] L. Grafakos and G. Teschl, On Fourier Transforms of Radial Functions and
Distributions, J. Fourier Anal. Appl. 19 (2013), 167–179.
[8] G. H. Hardy, On the expression of a number as the sum of two squares, Quart.
J. Math., 46 (1915), 263–283.
[9] G. H. Hardy, The average order of the arithmetical functions P (x) and ∆(x),
Proc. London Math. Soc. 15 (1917), 192–213.
[10] G. H. Hardy and E. Landau, The lattice points of a circle, Proc. Roy. Soc.
London Ser. A, 105 (1924), 244–258.
[11] D. R. Heath-Brown, Lattice points in the sphere, Number theory in progress,
Vol. 2 (Zakopane-Kościelisko, 1997), 883-892, de Gruyter, Berlin, 1999.
[12] M. N. Huxley, Exponential sums and lattice points. III. Proc. London Math.
Soc., (3) 87 (2003), no. 3, 591–609.
http://arxiv.org/abs/1709.04340
Multiple Fourier series and lattice point problems 61
[13] A. Ivic, E. Krätzel, M. Kühleitner, and W. G. Nowak, Lattice points in large
regions and related arithmetic functions: recent developments in a very classic
topic, Elemenare und analytische Zahlentheorie, Franz Steiner Verlag Stuttgart,
2006, 89-128.
[14] V. Jarńık, Bemerkungen zu Landauschen Methoden in der Gitterpunktlehre,
Number Theory and Analysis, (A collection of papers in honor of Edmund
Landau (1877-1938), Edited by Turan, Plenum, New York, 1969, 139–159.
[15] E. Krätzel, Lattice points, Kluwer Academic Publication, 1988.
[16] E. Krätzel, Analytische Funktion in der Zahlentheorie, B. G. Teubner Stuttgart·Leibzig· Weisbaden, 2000.
[17] S. Kuratsubo, On a theorem of B. Novák on lattice-point problem, Science
Reports Hirosaki Univ., 29(1982), 111-113.
[18] S. Kuratsubo, On pointwise convergence of Fourier series of indicator function
of N dimensional ball, Sci. Report Hirosaki Univ., 43 (1996), 199–208.
[19] S. Kuratsubo, On pointwise convergence of Fourier series of radial function in
several variables, Proc. Amer. Math. Soc., 127 (1999), 2987–2994.
[20] S. Kuratsubo, Analysis of the Fourier inversion formula of the indicator func-
tion of a d dimensional ball and an extension of the Voronöı-Hardy’s identity,
In: Proceedings of the 4th International Conference on Analytic Number The-
ory and Spatial Tesselations, Drahomanov National Pedagogical Univ. (Kyiv
Ukraine), 2008, 22–28.
[21] S. Kuratsubo, On an extension of the Voronöı-Hardy identity and multiple
Fourier series, RIMS Kokyuroku Bessatsu, B14, 2009, 35–51.
[22] S. Kuratsubo, On pointwise convergence of Fourier series of the indicator func-
tion of d dimensional ball, J. Fourier Anal. Appl., 16 (2010), 52–59.
[23] S. Kuratsubo, E. Nakai and K. Ootsubo, On the Pinsky phenomenon of Fourier
series of the indicator function in several variables, Mem. Osaka-Kyoiku Univ.
Ser. III Nat. Sci. Appl. Sci., 55 (2006), 1–20.
62 S. Kuratsubo and E. Nakai
[24] S. Kuratsubo, E. Nakai and K. Ootsubo, Generalized Hardy identity and rela-
tions to Gibbs-Wilbraham and Pinsky phnomena, J. Funct. Anal., 259 (2010),
315–342.
[25] E. Landau, Zur analytischen Zahlentheorie der definiten quadratischen Formen
Über die Gitterpunkte in einem mehrdimensionalen Ellipsoid), Sitzungsber. der
Königlich Rreissischen Akad. Wiss., 31 (1915), 458-476.
[26] E. Landau, Vorlesungen Über Zahlentheorie II, Hirzel, Leibzig, 1927; Chelsea,
New York, 1969.
[27] E. Landau, Ausgewählte Abhandlungen zur Gitterpunktlehre, Herausgegeben
von Arnold Walfisz, VEB Deutscher Verlag der Wissenschaften, Berlin 1962.
[28] B. Novák, Über Gitterpunkte mit Gewichten in mehrdimensionalen Ellipsoiden,
Czechoslovak Math, J., 17 (1967), 609–623.
[29] B. Novák, Verallgemeinerung eines Petersonschen Satzes und Gitterpunkte mit
Gewichten, Acta Arith., 13 (1967/1968), 423–454.
[30] B. Novák, Mittelwertsätze der Gitterpunktlehre, Chechoslovak Math. J., 19
(1969), 154–180.
[31] B. Novák, Mean value theorems in the theory of lattice points with weight II,
Comm. Math. Univ. Carolinae, 11(1970), 53–81.
[32] B. Novák, Über Gitterpunkte in mehrdimensionalen Ellipsoiden, Czechoslovak
Math, J., 22 (1972), 495–507.
[33] K. Ootsubo, S. Fujima, S. Kuratsubo and E. Nakai, Kuratsubo phenomenon of
the Fourier series of some radial functions in 4 dimension, in preparation.
[34] M. Pinsky, Pointwise Fourier inversion and related eigenfunction expasions,
Comm. Pure Appl. Math., 47 (1994), 653–681.
[35] M. Pinsky, Introduction to Fourier Analysis and Wavelets, Brooks/Cole, CA,
2002.
Multiple Fourier series and lattice point problems 63
[36] M. Pinsky, N. Stanton and P. Trapa, Fourier series of radial function in several
variables, J. Funct. Anal. 116 (1993), 111–132.
[37] M. Pinsky and M. Taylor, Pointwise Fourier inversion: a Wave Equation Ap-
proach, J. Fourier Anal. Appl., 3 (1997), 647–703.
[38] E. M. Stein, Localization and summability of multiple Fourier series, Acta
Math., 100 (1958), 93–147.
[39] E. M. Stein, On certain exponential sums arising in multiple Fourier series,
Annals of Math., 73(1961), 87-109.
[40] E. M. Stein and G. Weiss, Introduction to Fourier analysis on Euclidean spaces,
Princeton Univ. Press, 1971.
[41] M. Taylor, Pointwise Fourier inversion on tori and other compact manifolds, J.
Fourier Anal. Appl., 5 (1999), 449–463.
[42] M. Taylor, The Gibbs phenomenon, the Pinsky phenomenon, and variants
for eigenfunction expansions, Comm. Partial Differential Equations, 27 (2002),
565–605.
[43] M. Taylor, Double Fourier series with simple singularities–A graphical case
study, preprint. http://www.unc.edu/math/Faculty/met/fourier.html
[44] M. Taylor, Serendipitous Fourier inversion, preprint.
http://www.unc.edu/math/Faculty/met/SERENE.pdf
[45] G. Voronöı, Sur le développement, à l’aide des fonctions cylindriques, des
sommes doubles∑f(pm2 + 2qmn+ rn2), où pm2 + 2qmn+ rn2 est une forme
positive à coefficients entiers, Verhandlungen des dritten Int. Math. -Kongresses
in Heidelberg, 241–245, 1905.
[46] A. Walfisz, Gitterpunkte in mehrdimensionalen Kugeln, Warszawa, 1957.
[47] A. Zygmund, Trigonometric Series. 2nd ed., Cambridge, UK: Cambridge Uni-
versity Press, 1959.
http://www.unc.edu/math/Faculty/met/fourier.htmlhttp://www.unc.edu/math/Faculty/met/SERENE.pdf
64 S. Kuratsubo and E. Nakai
Shigehiko KuratsuboDepartment of Mathematical SciencesHirosaki UniversityHirosaki 036-8561, JapanE-mail: kuratubo@hirosaki-u.ac.jp
Eiichi NakaiDepartment of Math
Recommended