SOME PROBLEMS IN ADDITIVE NUMBER THEORY

A dissertation submitted to
Kent State University in partial
fulfillment of the requirements for the
degree of Doctor of Philosophy

by

John W. Hoffman

August 2015
Dissertation written by

John W. Hoffman

B.S., Youngstown State University, 2009

M.A., Kent State University, 2011

Ph.D., Kent State University, 2015

Approved by

Dr. G. Yu, Chair, Doctoral Dissertation Committee

Dr. U. Vorhauer, Member, Doctoral Dissertation Committee

Dr. M. Davidson, Member, Doctoral Dissertation Committee

Dr. S. Gagola, Member, Doctoral Dissertation Committee

Dr. J. Maletic, Member, Outside Discipline

Dr. J. Khan, Member, Graduate Representative

Accepted by

Dr. A. Tonge, Chair, Department of

Mathematical Sciences

Dr. J. Blank, Dean, College of Arts and Sciences

ii
TABLE OF CONTENTS

ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

NOTATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Waring’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Goldbach’s Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Waring-Goldbach Problems . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Mixed Power Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.5 Thin Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 The Hardy-Littlewood Method . . . . . . . . . . . . . . . . . . . . . . 15

2.2 Weyl’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.3 Exponential Sums over Prime Variables . . . . . . . . . . . . . . . . . 20

2.4 Hua and van der Corput Inequalities . . . . . . . . . . . . . . . . . . . 23

2.5 Exponent Pair Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.6 Double Large Sieve Method . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Representations of n = p21 + p22 + p23 + p2γ + qγ2 . . . . . . . . . . . . . . . . . . 32

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2 The Minor Arcs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.3 The Major Arcs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.4 Estimating |fγ − g| . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

iii
3.5 Estimating Type I and Type II Sums . . . . . . . . . . . . . . . . . . 41

4 Representations of n = [mc ] + pk . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4.1 The Treatment of S1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.2 The Treatment of S2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.2.1 The Error Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.2.2 The “Main” Error Term . . . . . . . . . . . . . . . . . . . . . . . 50

5 Representations of N = m2 + p3 + q 5 . . . . . . . . . . . . . . . . . . . . . . . 55

5.1 Exponential Sums over Squarefree Numbers . . . . . . . . . . . . . . 56

5.2 Proof of Theorem 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.3 Proof of Theorem 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.4 Proof of Theorem 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5.5 Singular Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

iv
ACKNOWLEDGEMENTS

In my nine year educational journey, I have had the help of many people along the way.

This dissertation would not be possible without their support.

First, I would like to thank my committee members for their suggestions and corrections

to my work. In particular, the number theory seminar group have been wonderful teachers,

mentors and friends during my time at Kent. I want to thank Dr. Yu for his patience and

wisdom throughout my research process. Every conversation with him always ends with

me feeling confident and capable. He has been a great mentor, and he helped me unlock

my potential as a number theorist.

Also, I want to thank my friends at YSU for their encouragement and support through-

out the years. My experience in Dr. Faires’ calculus class is the reason I decided to pursue

mathematics. Similarly, my love of number theory began in Dr. Fabrykowski’s number

theory class.

Finally, I want to thank my family and friends for their love and support. I want to

thank my parents for instilling a high value in education since grade school, and my big

sister (the original Dr. Hoffman) for being the best role model in the world. Lastly, I want

to thank Lindsay for being my even prime and supporting me throughout my education.

v
NOTATION

Standard notation is used whenever possible. Common notation used throughout the

text is explained here, while more specific notation will be given in the body of the text.

k, l, m, n, M, N Natural numbers.

p A prime number.

α, x, y Real variables.
P∞ −s
ζ(s) Riemann’s zeta function defined by ζ(s) = n=1 n when

Re(s) > 1 and the analytic continuation to the whole com-

plex plane.

exp z = ez .

e(x) = e2πix .

log x =ln x.

||t|| =minn∈Z |t − n|.

[x] The greatest integer not exceeding x.

{x} = x − [x].

ψ(x) = x − [x] − 12 .

Λ(n) The von Mangoldt function defined by Λ(n) = log p if n =

pm and zero otherwise.

µ(n) The Möbius function defined by µ(n) = (−1)k if n =

p1 · · · pk (the pj being distinct), µ(1) = 1, and zero oth-

erwise.
P
d|n A sum taken over all positive divisors of n.

π(x) The number of primes not exceeding x.

vi
f (x) = O(g(x)) Means |f (x)| ≤ Cg(x) for x ≥ x0 and some absolute con-

stant C > 0.

f (x) g(x) Means f (x) = O(g(x)).

f (x) g(x) Means g(x) = O(f (x)).

f (x) g(x) Means g(x) f (x) g(x)

m∼M Means M < m ≤ 2M .

ε, δ An arbitrarily small constant, not necessarily the same at

each occurence.

vii
CHAPTER 1

Introduction

In this chapter, we introduce the historical background of the techniques and problems

investigated in the later chapters. We will begin with the classic Waring and Goldbach

problems and discuss the results on these problems. In addition, we will explore variations

on these classics, including Waring-Goldbach and mixed power problems. Lastly, we will

introduce the history of Piatetski-Shapiro primes and the results on their existence. This

chapter also includes the theorems that will be proved in the later chapters.

1.1 Waring’s Problem

Additive problems have a long history in mathematics. The classic example is regarding

representing integers as sums of squares. This problem was studied throughout history by

many different people. The well known four square theorem was first stated by Bachet

in 1621 (the result was most likely known by Diophantus), and its proof was claimed by

Fermat. Fermat, however, did not provide the details of the proof. It wasn’t until 1770

when Lagrange provided a proof of the theorem by building on the earlier work of Euler.

This problem can be generalized to any k-th powers. In 1770 Edward Waring stated

(without proof) in his book Meditationes Algebraicae that every integer is a sum of 9 cubes,

19 fourth powers, and so on. That is, there exists an s such that every natural number

n can be represented as a sum of s k-th powers. Let g(k) represent the minimum such s.

Waring claimed that g(3) = 9, g(4) = 19 and so on. Of course Lagrange proved g(2) = 4.

In 1859 Liouville gave the next Waring-type result by showing g(4) ≤ 53. The existence of

g(k) for all k was done by Hilbert in 1909 [40]. Hilbert’s proof is complicated and provides a

1
poor bound for g(k). With the establishment of the existence of g(k), the natural problem

is to determine the size of g(k). Hardy and Littlewood established an analytic method

which greatly improved Hilbert’s bound for g(k). Their method, also known as the circle

method, is the basis for many problems in additive number theory. Through refinements of

the Hardy-Littlewood method, the value, g(k) is almost completely settled. In particular,

the integer n = 2k [(3/2)k ] − 1 is less than 3k and thus can only be represented by k-th

powers of 1 and 2. The best representation includes [(3/2)k ] − 1 k-th powers of 2 and 2k − 1

powers of 1. Thus g(k) ≥ [(3/2)k ] + 2k − 2. It is likely that this always holds with equality.

In fact Mahler (1957) [65] showed this holds for all but finitely many k. No exceptions

are currently known, and unfortunately there is no known bound for which there are no

exceptions.

In 1942 Linnik [56] proved only finitely many integers require 8 or more cubes. That

is if n is sufficiently large, it can be represented as a sum of 7 cubes. Since g(k) is usually

determined by the requirements of a few specific integers (for instance 23 requires 9 cubes), it

is more interesting to study the case for sufficiently large integers. Let G(k) be the minimum

s such that all sufficiently large integers are representable by s k-th powers. Linnik’s result

can be restated as G(3) ≤ 7. Clearly G(k) ≤ g(k). Since G(k) is much smaller than g(k)

when k is large, its evaluation is more difficult. Currently, the only known values of G(k)

are G(2) = 4 and G(4) = 16, the latter result due to Davenport in 1939 [25].

1.2 Goldbach’s Problem

Another historically famous additive problem is Goldbach’s conjecture. In 1742 Chris-

tian Goldbach conjectured in a letter to Euler that every even integer is a sum of 2 primes

and every integer is a sum of 3 primes. Goldbach included 1 as a prime number and thus

Goldbach’s conjectures have become the statements every even integer greater than 2 is the

sum of two primes and every odd integer greater than 5 is the sum of 3 primes. Hardy and

2
Littlewood [35] showed all sufficiently large odd integers are representable as a sum of three

primes under the assumption of the generalized Riemann hypothesis. In 1937 Vinogradov

removed the dependence of the generalized Riemann hypothesis and gave an unconditional

proof of Hardy and Littlewood’s result. In particular Vinogradov [89] proved the following

result:

For a positive integer odd n, let R(n) denote the number of representations of n as the

sum of three primes. Then

n2
R(n) = S0 (n) + O(n2 (log n)−4 )
2(log n)3

where
Y 1
Y
1

S0 (n) = 1− 1+ 1.
(p − 1)2 (p − 1)3
p|n p-n

In particular, every sufficiently large odd integer is a sum of three primes.

It should be remarked that the removal of the dependence on GRH comes at the cost

of a large implied constant. Hardy and Littlewood’s proof (GRH dependent) establishes

the ternary Goldbach conjecture for all n ≥ 1050 , where as Vinogradov’s method requires

n ≥ 106800000 . The bound, 1050 , is beyond checking the remaining cases by computer.

However, recently Deshouillers et al. [46] proved that if GRH is true then the ternary

Goldbach conjecture holds for all odd n ≥ 7. Most recently, Harold Helfgott claims to have

settled the problem unconditionally [39] [38].

Work on the binary Goldbach problem has been less conclusive. Following Vinogradov’s

method, Chudakov [22], van der Corput [82], and Estermann [29] each showed that almost

all even integers n ≤ x are sums of two primes. More precisely, let E(x) denote the number

of integers n ≤ x such that n is not the sum of two primes. They showed that for any A > 0

we have

E(x) = O(x(log x)−A ).

3
Some improvements have been made on this result. Montgomery and Vaughan [67] proved

the existence of an absolute constant δ > 0 such that

E(x) = O(x1−δ ).

Chen and Pan [20] computed a numerical value for the constant. They showed that the

result of Montgomery and Vaughan gives the constant δ = 0.01. This result has been

improved by a number of authors and in 2004 Pintz established the bound with δ = 1/3

[69].

Although the binary Goldbach conjecture is out of reach, there have been some results

just short of the conjecture. We denote Pr as a number having at most r prime factors

counting multiplicity. Such a number is called an almost prime of order r. In 1973, Chen

proved a result that is just short of the binary Goldbach conjecture. His result is stated

below. [19]

For an even integer n, let r(n) denote the number of representations of n in the form

n = p + P2 , where p is a prime and P2 is an almost prime of order 2. There exists an

absolute constant n0 such that if n ≥ n0 , then

Y 1
Y
p−1

n
r(n) > 0.67 1− .
(p − 1)2 p − 2 (log n)2
p>2 p>2
p|n

In particular, every sufficiently large even integer n can be represented in the form n = p+P2 .

1.3 Waring-Goldbach Problems

In the last century several variations to Waring’s problem have been considered. Vino-

gradov’s work on Goldbach’s conjecture provides a model to adapt Hardy and Littlewood’s

method to problems involving sums of primes. We will begin by describing the necessary

congruence conditions for Waring’s problem over primes. Suppose k is a positive integer

and p is a prime, and denote θ = θ(k, p) to be the power of p that exactly divides k i.e. pθ |k

4
but pθ+1 - k. Now, define γ = γ(k, p) by



θ + 2,
 when p = 2 and θ > 0,
γ(k, p) =

θ + 1, otherwise.



Lastly, define
Y
K(k) = pγ .
(p−1)|k

The Diophantine equation in consideration is everywhere locally solvable if it is solvable

modulo K(k). Now, we let H(k) denote the minimal s such that every sufficiently large n

congruent to s modulo K(k) can be represented as a sum of s kth powers of primes. Fol-

lowing the earlier work of Vinogradov [89] [90], Hua studied a variety of problems involving

primes in his book. Indeed, in [44] he proved

H(k) ≤ 2k + 1

and when k is large

1
H(k) ≤ 4k(log k + log log k + O(1)).
2

The approach to the Waring-Goldbach problem is to proceed with the Hardy-Littlewood

method as in the classic Waring’s problem and removing the primality condition on some

estimates to obtain an upper bound. There have been some improvements to H(k) for small

values of k. In particular Kawada and Wooley proved H(4) ≤ 14, and H(5) ≤ 21 [50]. Also,

Thanigasalam proved the estimates H(6) ≤ 33, H(7) ≤ 47, H(8) ≤ 63, H(9) ≤ 83, and

H(10) ≤ 107 [81].

1.4 Mixed Power Problems

The Hardy-Littlewood method can be adapted to attack many different types of additive

problems. In addition to Waring-Golbach problems, there has been some attention given to

mixed power problems. Hardy and Littlewood formulated several conjectures on asymptotic

5
formulas concerning different representations of integers. In particular, they conjectured an

asymptotic formula for the number of representations of a sufficiently large integer n in the

form

n = p + x2 + y 2

where p is a prime and x and y are integers. Their conjecture was confirmed by Hooley [41]

under the conditions of GRH and later unconditionally by Linnik [58].

Another famous conjecture of Hardy and Littlewood is concerning an asymptotic formula

for the number of representations of a sufficiently large integer, n, as a sum of a prime and

a square. For non-square n, let R(n) be the number of representations of n as n = m2 + p.

The conjectured asymptotic formula is given by

√ Y
( np )

n
R(n) = 1− (1.1)
log n p−1
p>2

where ( np ) represents the Legendre symbol. This result appears to be out of reach with

current methods, but a number of authors have considered this problem.

In 1936 Davenport and Heilbronn [26] gave a proof that the asymptotic formula (1.1)

holds for all but O(N (log N )−A ) integers n ≤ N and some positive constant A. Meich

[66] provided the first improvement to Davenport and Heilbronn. In particular, let Ek (N )

denote the number of integers n ≤ N not representable in the form n = p + xk where p is

a prime and x an integer. Meich proved E2 (N ) N (log N )−A for any fixed A > 0. The

next improvement came from Polyakov [71]. He showed that

√
( np ) log1/2 n

n Y
R(n) = 1− 1 + O exp (− 3 )
log n p−1 log log n
p>2
√
for all but O(N exp (−c log N )) integers n ≤ N for some constant c. Polyakov’s improve-

ment was followed by a result of Vinogradov. Vinogradov [88] showed that the number

of exceptions is at most O(N θ ) for some constant θ < 1. Brünner, Perelli, and Pintz [18]

6
independently proved the same result as Vinogradov, but with a different approach. Vino-

gradov used the original approach of Hardy and Littlewood. where Brünner, Perelli, and

Pintz used a method developed by Montgomery and Vaughan.

In addition to the study of the classic conjecture, a number of authors have considered

the more challenging generalization of representing a positive integer n in the form n =

mk + p. In the study of this generalization, Davenport and Heilbronn’s 1936 paper included

a result on the general problem. They proved that almost all positive integers n can be

represented in the form n = mk + p. The most recent result concerning this problem is due

to A. Zaccagnini in 1992. Zaccagnini [92] proved that the number of exceptions is bounded

by Ek (X) X 1−δ for some δ = δ(k) > 0.

One final variation of this conjecture has been studied throughout the years. This

variation concerns representing an integer n in the form n = p + q k where p and q are

primes. This problem was first considered by W. Schwarz in 1961. Schwarz built on the

earlier work of Prachar to establish a number of results for various prime diophantine

equations. Plaksin [70] improved this result in 1990. Plaksin showed that the number

of even integers, n ≤ N , not representable in the form n = p + q k is N γ for some

1
γ < 1 and γ < 1 − 137k3 log k
for large k. Liu and Chiu [21] considered variations of the

Hardy-Littlewood conjecture which included coefficients. They considered representations

of integers in the forms n = a1 p + a2 mk and n = a1 p + a2 q k , where p and q are primes

and m is an integer. If we let E1 (N ), and E2 (N ) denote the number of exceptions n ≤ N

of these problems respectively, Liu and Chiu proved E1 (N ) N θ1 and E2 (N ) N θ2 for

some θ1 , θ2 < 1.

Although this conjecture is presently unobtainable, we prove the following result inspired

by this conjecture.

7
Theorem 1. Every sufficiently large n is representable in the form

n = [mc ] + pk

1
where m is an integer and p is a prime provided c ≤ 1−θ . Here,
1

+ µ1 − λ1 ) 1 − λ2 − (k − 1)µ2 − µ1 (1 − λ2 )
k (1
θ=
2(µ2 + 1)(1 + µ1 − λ1 ) + 2(1 − λ2 )(1 + µ1 − λ1 ) + (1 − λ2 )

493
where (µ1 , λ1 ) and (µ2 , λ2 ) are exponent pairs. In particular, if k = 2, c ≤ 489 .

We remark that the result for k = 2 is not optimal. Here we obtain a representation

of all sufficiently large integers, but in this case [mc ] is far from a square. We also have a

generalization in that we include a power of p as well.

In addition to these conjectures, many other variations of mixed power problems have

been considered. A special case concerns the representation of n as a sum of s consecutive

powers starting with a square. This question was answered by Roth [79] in 1951 for suffi-

ciently large n with s = 50. This result was later improved by Thanigasalam [81], Vaughan

[83], Brüdern [13], and Ford [30]. The current best result belongs to Ford with s = 15.

In general, we consider representations as non-consecutive mixed powers. That is suppose

that k1 , k2 , . . . ks are natural numbers with 2 ≤ k1 ≤ k2 ≤ · · · ≤ ks . Then we consider

representations of integers n as

n = xk11 + xk22 + · · · + xks s

where each xi is a positive integer. Heuristic applications of the Hardy-Littlewood method

suggest that all sufficiently large n have such a representation provided that

k1−1 + k2−1 + · · · + ks−1 > 2, (1.2)

and a representation exists for almost all n provided

k1−1 + k2−1 + · · · + ks−1 > 1. (1.3)

8
Table 1.1: Representations of almost all integers
Davenport and Heilbronn [27] x21 + x22 + y k
Davenport and Heilbronn[28] x2 + y13 + y23
Roth [78] x2 + y 3 + z 4
Vaughan[84] x2 + y 3 + z 5
Hooley [43] x2 + y 3 + z 6 + w k
Davenport [24] x31 + x32 + x33 + x34
Brüdern [10] x31 + x32 + x33 + y 4
Brüdern [10] and Lu [64] x31 + x2 + x33 + y 5
Brüdern and Wooley [17] x31 + x2 + x33 + y 6
Kawada and Wooley [49] x3 + y14 + y24 + y34 + y44
Vaughan [87] x41 + x42 + x43 + x44 + x45 + x46
Kawada and Wooley [49] x41 + x42 + x43 + x44 + x45 + y k (k odd)

Of course, the condition (1.3) is necessary for such a representation of all integers. The study

of such problems has stimulated progress on the methods used in the classical Waring’s

problem.

Problems involving smaller powers tend to be of greater interest than those of higher

powers, and a great deal is known about these cases. In particular, the problem of repre-

senting a natural number n in the form

n = x2 + y 3 + z 5

for x, y and z in the naturals has been of particular interest. R.C. Vaughan [84] first showed

that almost all integers have such a form. Later Brüdern improved Vaughan’s result by

showing there are at most O(N 1−1/42+ε ) exceptions up to N [12]. He later improved this

result to at most O(N 1−1/30+ε ) exceptions [14]. This is the best result that can be expected

with the current tools available. Many other variations have been studied as well. Here

we summarize the results concerning mixed power problems. In particular, for all ternary

problems subject to (1.3), an “almost all” type of result has been proved.

As with Waring’s problem, it is natural to consider mixed power problems with prime

variables. That is, we also consider representations of integers as a sum of mixed powers of

9
Table 1.2: Representations of all sufficiently large integers
Gauss [33] x21 + x22 + x23
Hooley [42] x21 + x22 + y13 + y23 + y33
Hooley [43] x21 + x22 + assorted powers
Vaughan [86] x21 + y13 + y23 + y33 + y43 + y53
Brüdern and Wooley [15] x21 + y13 + y23 + y33 + y43 + y53 + z 4
Ford [31] x21 + x32 + x43 + · · · + x15
14
Linnik [57] x31 + x32 + x33 + x34 + x35 + x36 + x37
Brüdern [11] x31 + x32 + x33 + x34 + x35 + x36 = Y14 + y24
Brüdern [11] x31 + x32 + x33 + x34 + x35 + y14 + y24 + y34
Kawada and Wooley [49] x31 + x32 + x33 + y14 + y24 + y34 + y44 + y54 + y64
Brüdern and Wooley [16] x31 + x32 + y14 + y24 + y34 + y44 + y54 + y64 + y74
Kawada and Wooley [49] x31 + y14 + y24 + y34 + y44 + y54 + y64 + y74 + y84 + y94
Vaughan [87] x41 + x42 + · · · + x412
Kawada and Wooley [49] x41 + x42 + · · · + x410 + y k (k odd)

primes. In particular, the problem of representing an integer n in the form

n = p1 2 + p2 3 + p3 5 + p4 k (1.4)

for prime numbers pj , j = 1, 2, 3, 4. We should mention that, when the pj ’s are replaced by

natural numbers, the equation (1.4) has received a lot of attentions (c.f. [80], [81], [23] for

example). Let E(k, N ) be the number of even integers n ≤ N that can not be represented

by (1.4). In 1953, Prachar [72] showed that E(4, N ) = O(N (log N )−30/47+ε ). This was later

improved by a number of authors (c.f. [5], [6], [7], [73], [74]). The current best result is

47
E(4, N ) N 1− 1680 +ε , (1.5)

which belongs to Bauer [8]. For general k, Lu and Shan proved in [63] that E(k, N ) =

O(N (log N )−c ) for some c > 0. This was recently improved to

1
1− +ε
E(k, N ) N 3k×2k−2 (1.6)

by Liu [62]. Here we give improvements to (1.5) and (1.6).

Theorem 2. We have
167
E(4, N ) N 1− 5040 +ε , (1.7)

10
and, for k ≥ 5,
47
E(k, N ) N 1− 420·2s +ε , (1.8)

where s = [ k+1
2 ].

The improvements (1.7) and (1.8) are obtained by improving the minor arc estimates in [8]

and [62], respectively.

Additionally, we consider the ternary problem of representing an integer n in the form

n = m2 + p3 + q 5 (1.9)

where m is squarefree and p and q are primes. We will establish the following theorem:

Theorem 3. Let E(n) be the number of positive integers not exceeding n which cannot be

written in the form (1.9). Then we have

1
E(n) n1− 45 +ε ,

where ε is any fixed positive number.

1 1
We remark that the number 45 in the exponent can be improved to 30 if, in (1.9), m

is relaxed to ordinary integers. Concerning representing integers in the form of (1.9), the
1
most challenging problems are probably to improve Brüdern’s result to O(N 1− 30 −δ ) for any

δ > 0, or show that almost all odd integers are representable by (1.9) with m a prime as

well.

1.5 Thin Sets

Another application of exponential sums is the study of primes in thin sets. That is a

set of primes, S, is said to be thin if

X
1 = o(π(x)).
p≤x
p∈S

11
In particular, the set Pc = {p : p prime and p = [nc ] for some integer n} is a well known

thin set of primes known as Piatetski-Shapiro primes. Let πc (x) denote the number of

primes, p ≤ x, in this form. It is a straightforward result from the prime number theorem

to show that

x x
πc (x) = +O (1.10)
c log x c log2 x

holds for 0 < c ≤ 1. This result is expected to hold for c in the range 1 < c < 2 as well.

The problem of representing primes by linear polynomials is settled by Dirichlet’s theorem

on primes in arithmetic progressions, but it is not known if there is a quadratic polynomial

which takes on infinitely many primes. Representing primes in the form [nc ] for 1 < c < 2

can be regarded as representing primes as a polynomial of degree c. Piatetski-Shapiro

proved that this result holds when 0 < c < 12/11 = 1.0909... [68], and such primes are

called Piatetski-Shapiro primes in his honor. A number of authors have made improvements

to the range of c for which this asymptotic formula holds. Kolesnik [51] was the first to

provide a improvement. He proved that (1.10) holds for 0 < c < 10/9 = 1.111.... Graham

(unpublished) and Leitmann [55] improved this result to 0 < c < 69/62 = 1.1129... using

the method of exponent pairs. Heath-Brown [37] gave the improvement 0 < c < 755/662 =

1.1404... using a ‘discrete’ Weyl shift and exponent pairs along with his decomposition of

the von Mangoldt function. Kolesnik [52] provided another improvement on Heath-Brown’s

result. He gave the range 0 < c < 39/34 = 1.1470... using multiple exponential sums. Liu

and Rivat [59] used the double large sieve technique of Iwaniec and Fouvry to obtain the

range 0 < c < 15/13 = 1.1538.... The current best result is due to Rivat and Sargos [76].

They gave the range 0 < c < 2817/2426 = 1.16117....

In addition to the studying the range of c for which (1.10) holds, some authors have

considered the range for which

x
πc (x) . (1.11)
c log x

12
J. Rivat [75] was the first to consider this problem. In his doctoral thesis, he proved that

there exists an absolute constant ρ0 > 0 such that

x
πc (x) ≥ ρ0
c log x

for each fixed c such that 1 < c < 7/6 = 1.1666.... Subsequent improvements came inde-

pendently from Baker, Harmon, and Rivat [3] and Jia [48] using the double large sieve of

Iwaniec and Fouvry. Baker et al proved (1.11) holds for 1 < c < 20/17 = 1.1764.... This

was improved by Jia [47] to the range 1 < c < 13/11 = 1.1818.... The next improvement

was due to Kumchev [53] who used a refined version of the double large sieve of Iwaniec

and Fouvry to obtain the range 1 < c < 45/38 = 1.1842.... The current best result is due

to Rivat and Wu who proved that (1.11) holds when 1 < c < 243/205 [77].

The introduction of Piatetski-Shapiro primes (and other thin sets of primes) has led

to hybrid variations in Waring-Goldbach type problems. Balog and Friedlander proved a

variation of Vinogradov’s three prime theorem involving Piatetski-Shapiro primes. They

showed that every sufficiently large odd integer can be expressed as a sum of three primes

of the form [nc ] for 1 ≤ c < 21/20 [4]. Their result was later improved by Kumchev to

include c in the range 1 < c < 53/50. Balog and Friedlander’s proof doesn’t directly use

the Hardy-Littlewood method. The proof of this result comes from reducing the problem

to the standard three prime problem and showing the error term is admissible. Bounding

the error term is done by directly estimating exponential sums.

The construction of a thin set of primes has led to many other variations in mixed power

and Waring-Goldbach problems. In view of Balog and Friedlander’s work on the hybrid

Vinogradov problem, there has been some attention given to using Piatetski-Shapiro primes

in other Waring-Goldbach problems. In particular, W. G. Zhai proved a variation of the

Waring-Goldbach problem for k = 2 [93]. He gave an asymptotic formula for the number

of representations of a large integer n as a sum of five Piatetski-Shapiro prime squares

13
provided 1 < c < 44/43. His proof was carried out similarly to Balog and Friedlander’s

with a slight modification to account for the squares. We also consider a variation of this

problem. We examine the representation of integers n as a sum of three prime squares and

two Piatetski-Shapiro prime squares. We prove the following theorem:

Theorem 4. Let R2,γ (n) denote the number of solutions of the equation

n = p21 + p22 + p23 + p2γ + qγ2

where p1 , p2 , p3 are primes and pγ , qγ are Piatetski-Shapiro primes of type c = 1/γ. Then

Γ5 (3/2) n3/2
R2,γ (n) ∼ S(n)
Γ(5/2) (log n)5

provided 7/8 < γ ≤ 1. Here

q
∞ X
S 5 (q, a) −an
X
S(n) = e ,
ϕ5 (q) q
q=1 a=1
(a,q)=1

where
q
X am2
S(q, a) = e .
q
m=1
(m,q)=1

Moreover, S(n) is absolutely convergent and S(n) ≥ δ > 0 for all n ≡ 5 (mod 24)

The proof of this theorem uses a combination of the Hardy-Littlewood method and

exponential sum estimation. We use a pruning technique to shrink the size of the major

arcs in order to obtain an acceptable error for replacing two of the primes with Piatetski-

Shapiro primes.

14
CHAPTER 2

Preliminaries

In this chapter, we provide some of the machinery that will be used to prove the results

in later chapters. In particular, we will introduce the basic idea of the Hardy-Littlewood

method, and include the proofs of the classic Weyl and Hua inequalities. Later in this

chapter, we discuss some more recent contributions to the study of exponential sums such

as the exponent pair and double large sieve methods.

2.1 The Hardy-Littlewood Method

In the study of additive problems, the Hardy-Littlewood method is a powerful method

of attack. Here we outline its basic ideas. Hardy and Littlewood developed this method

to determine values of G(k), but their method is easily adapted to more general additive

problems. Suppose that for each j = 1, 2, . . . s, Aj is a set of non-negative integers. For a

fixed large integer n, let

X
fj (α) = fj (α, n) = e(aj α), j = 1, 2, . . . s
T
aj ∈Aj [0,n]

and
s
Y
F (α) = fj (α).
j=1

It is clear that

X X X
F (α) = ··· e((a1 + · · · + as )α) = R(m, n)e(mα)
T T
a1 ∈A1 [0,n] as ∈As [0,n] m≤sn

where R(m, n) is the number of representations of m as m = a1 + a2 + · · · + as with

T
aj ∈ Aj [0, n]. In particular, we define R(n) = R(n, n), the number of representations of

15
n as n = a1 + · · · + as . Now from the orthogonality relation,

Z 1  1 if h = 0,

e(αh)dα =
0  0 if h ∈ Z\{0}


we have that
Z 1
R(n) = F (α)e(−nα)dα.
0

Now if Aj is a “well-distributed” subset of non-negative integers, then |fj (α)| is expected

to be large when α is close to a rational number with relatively small denominator and

small otherwise. In view of this observation, it is natural to dissect the unit interval into

two parts; the major arcs and minor arcs denoted M and m respectively. The major arcs

are made up of subintervals centered at reduced rational numbers a/q with q small, and

the minor arcs are the complement of the major arcs in the unit interval. In most cases,

we expect the integral over the major arcs to give us the main term and the integral over

the minor arcs gives a negligible contribution. Various analytic tools are used to estimate

these integrals.

2.2 Weyl’s Inequality

Bounds for exponential sums are extremely useful. We will develop some of the tools

necessary for estimating exponential sums. In Waring’s problem (and many other additive

problems), we are interested in representing a natural number n as a sum of k-th powers.

Here f (α) has the form

X
f (α) = e(αnk ),
1≤n≤N

so we will focus our attention on sums of the form

X
S= e(g(n))
1≤n≤N

16
where g(n) = αnk + · · · is a polynomial of degree k. The case for k = 1 is a sum over a

linear polynomial which is a geometric series. We have

X |1 − e(αN ) | sin παN |
e(αn) ≤ e(α)
≤ .

1≤n≤N 1 − e(α) | sin πα|

Since | sin πα| ≥ 2||α||, we have for α 6∈ Z

X
1
e(αn) ≤ min N,
. (2.1)

n≤N 2||α||

We now give an estimate for a sum involving a polynomial of any degree k ≥ 2. We begin

with
2

X X X

e(g(n)) =
e(g(m) − g(n))
n≤N m≤N n≤N
−n
X NX
= e(g(h + n) − g(n)).
n≤N h=1−n

Now if g(n) is a polynomial of degree k, then g(h + n) − g(n) is a polynomial of degree

k − 1 (for h 6= 0). By repeating this differencing process we arrive at a sum involving a

polynomial of degree one for which we have a non-trivial bound. Thus by induction, we

obtain

Proposition 1. If g(x) = αxk + · · · is a polynomial of degree k with k ≥ 1, then

X X X 21−k
| e(g(n))| ≤ 2N N −k ··· min (N, ||αk!h1 . . . hk−1 ||−1 )
n≤N −N <h1 ,...,hk−1 <N

Proof: For k = 1 we interpret this result as that given by (2.1). Now suppose this result

holds for some k ≥ 1. Let g(x) = αxk+1 + · · · be a polynomial of degree k + 1. Then we

have g(x + h) − g(x) = α(k + 1)hxk + · · · is a polynomial of degree k. Thus,

X X X
e(g(n))|2 ≤ k

|
e(α(k + 1)hn + · · · )

n≤N |h|<N n≤N
0<h+n≤N

17
X X X 21−k
≤ 2N N −k ··· min (N, ||α(k + 1)!h1 · · · hk−1 h||−1 )
|h|<N −N <h1 ,...,hk−1 <N

by the induction hypothesis. From an application of Hölder’s inequality, we have

X X X 21−k
| e(g(n))| ≤ 4N N −k−1
2 2
··· min (N, ||α(k + 1)!h1 · · · hk−1 hk || −1
) .
n≤N −N <h1 ,...,hk <N

This gives the result for k + 1.

To finish Weyl’s inequality, we require the following result.

Proposition 2. Suppose that X, Y, α are real numbers with X ≥ 1, Y ≥ 1 and that |α −

a/q| ≤ q −2 with (a, q) = 1. Then

X
−1 −1 1 1 q
min (XY x , ||αx|| ) XY + + log (2Xq).
q Y XY
x≤X

Proof: Let
X
S= min (XY x−1 , ||αx||−1 ).
x≤X

We have
q
X X XY
S≤ min , ||α(qj + r)||−1 .
qj + r
0≤j≤X/q r=1

For each j, let yj = [αjq 2 ] and write θ = q 2 (α − a/q). Then

yj + ar {αjq 2 }
α(qj + r) = + + θrq −2 .
q q

When j = 0 and r ≤ q/2,

1 1
||α(qj + r)|| ≥ ||ar/q|| − ≥ ||ar/q||.
2q 2

Otherwise for each j, there are at most O(1) values of r for which ||α(qj + r)|| ≥ 12 ||(yj +

18
ar)/q|| fails to hold. Moreover qj + r q(j + 1). Thus,
q
X
−1
X XY X
−1
S ||ar/q|| + + ||(yj + ar)/q||
q(j + 1)
1≤r≤q/2 0≤j≤X/q r=1
q|yj +ar
X 1 X q
XY q −1 + (Xq −1 + 1)
j+1 h
0≤j≤X/q 1≤h≤q/2

1 1 q
XY + + log (2Xq)
q Y XY

which is the desired result.

Estimating exponential sums is critical in applications of the Hardy-Littlewood method.

Here, we combine the previous propositions into an estimate for exponential sums.

Lemma 1 (Weyl’s Inequality). Suppose that (a, q) = 1, |α − a/q| ≤ q −2 , g(x) = αxk +

α1 xk−1 + · · · + αk , and
X
Sg (N ) = e(g(n)).
n≤N

Then
1−k
Sg (N ) N 1+ε (q −1 + N −1 + qN −k )2 .

Proof: From Proposition 1 we have that

X X X 21−k
| e(g(n))| ≤ 2N N −k ··· min (N, ||αk!h1 . . . hk−1 ||−1 ) .
n≤N −N <h1 ,...,hk−1 <N

When h1 · · · hk−1 = 0, each term is bounded by N , so we have

X X X 21−k
| e(g(n))| N N −k (N k−1 + ··· −1
min (N, ||αk!h1 · · · hk−1 || )) .
n≤N 1≤|h1 |,...,|hk−1 |<N

19
Now let h = k!h1 · · · hk−1 , then we have
k!N k−1 21−k
X X
−k k−1 ε −1
| e(g(n))| N N (N +N min (N, ||αh|| ))
n≤N h=1
k!N k−1 21−k
X
−k k−1 ε k −1 −1
N N (N +N min (N h , ||αh|| )) .
h=1

Now from Proposition 2 this is bounded by

21−k
−k k−1 1 ε1 q k
N N (N + N (N ( + + k )))
q N N
21−k
N 1+ε N −1 + q −1 + qN −k

which completes the proof.

2.3 Exponential Sums over Prime Variables

We remark that although we allowed a general polynomial in Weyl’s inequality, the

result only relies on the leading term. In addition to the standard Weyl inequality, it is

necessary to consider an estimate for sums over a prime variable. Such sums are necessary

in additive problems involving primes. Vinogradov introduced these sums in the late 1930s

and provided an unconditional estimate in the case k = 1. This estimate is the essential

innovation in his celebrated proof of the ternary Goldbach problem. Let

X
g(α) = e(αpk )
p∼P

where the sum is over primes. Vinogradov’s result essentially states that for k = 1 if a and

q are integers satisfying q ≥ 1, (a, q) = 1, and |α − a/q| ≤ q −2 then

g(α) q ε P (q −1 + P −2/5 + qP −1 )1/2 .

20
In addition, Vinogradov gave estimates for the case k ≥ 2 and used them to provide the first

unconditional results on the Waring-Goldbach problem. For k ≥ 2 the sharpest estimates

using Vinogradov’s approach were provided by Harman [36]. He showed

1−k
g(α) P 1+ε (q −1 + P −1/2 + qP −k )4 .

A number of authors have worked to improve the estimates over the years. Here we provide

the current best result due to A. Kumchev [54].

Lemma 2. Let k ∈ N and α ∈ R and suppose that there exists a ∈ Z and q ∈ N satisfying

a Q
1 ≤ q ≤ Q, (a, q) = 1, |α − | <
q qP k

with Q ≤ P . Then for any fixed ε > 0,

q ε P (log P )c
g(α) Q1/2 P 11/20+ε + ,
q 1/2 (1 + P k |α − a/q|)1/2

where c > 0 is an absolute constant and where the constant implied in the depends on at

most k and ε.

When dealing with sums over prime variables, it is often convenient to weight the sums

by Λ(n) where Λ is the von Mangoldt function. Thus, we are required to estimate a sum

in the form
X
Λ(n)f (n)

for an appropriate function f (n). To bound a sum in this form, we use a decomposition

of Λ(n) known as Vaughan’s identity. We begin with an identity of ζ(s). We have the

following lemma.

Lemma 3. Let µ(n) be the Möbius function and define

X µ(n) X Λ(n)
MX (s) = M (s) = and N (s) = .
ns ns
n≤X n≤X

21
Then we have

ζ0 ζ0

(s) = ζ 0 (s)M (s) + ζ(s)M (s)N (s) + (s) + N (s) (1 − ζ(s)M (s)) − N (s) (2.2)
ζ ζ

From this identity, we select the coefficients of n−s on each side of (2.2) to obtain

Vaughan’s identity. For any arithmetic function f , we have

X
Λ(n)f (n) = S1 − S2 − S3 ,
X<n≤x

where

X X
S1 = µ(m) (log n)f (mn),
m≤X X<mn≤x
X X
S2 = µ(m)Λ(n) f (mnr),
m,n≤X X<mnr≤x
X
S3 = Λ(m)c(n)f (mn)
m,n>X
X<mn≤x

with
X
c(n) = µ(d).
d|n
d≤X

The sums S1 and S2 are estimated by considering a sum in the form

X X
SI = am f (mn), (2.3)
m n

The sum S3 can be treated in the form

XX
SII = am bn f (mn). (2.4)
m n

These sums are known as type I and type II sums respectively. Various methods are applied

to bound these sums with the quality of estimate dependent upon the function f , as well

as the ranges of m and n.

22
2.4 Hua and van der Corput Inequalities

Estimating exponential sums directly is essential, but it is also useful to estimate in-

tegrals of exponential sums. Here we give a result on such estimates. In this section, we

define
X
f (α) = e(αmk ).
1≤m≤N

Lemma 4 (Hua’s Inequality). Suppose that 1 ≤ j ≤ k. Then

Z 1
j j −j+ε
|f (α)|2 dα N 2 .
0

Proof: We proceed by induction on j. The case j = 1 is immediate from Parseval’s

identity. Now suppose the lemma holds for 1 ≤ j ≤ k − 1. By a repeated differencing

process described above, we have

j −j−1
X XX
|f (α)|2j ≤ (2N )2 ··· e(αh1 · · · hj pj (x; h1 , . . . , hj ))
h1 hj x∈Ij
|hi |≤N

where pj (x; h1 , . . . , hj ) is a polynomial in x of degree k − j with integer coefficients. Thus,

X
|f (α)|2j (2N )2j−j−1 ch e(αh) (2.5)
h

where ch is the number of solutions to the equation

h1 · · · hj pj (x; h1 , . . . , hj ) = h

with |hi | < N and x ∈ Ij . We have c0 N j and ch N ε (h 6= 0).

Now, considering
j j−1 j−1
|f (α)|2 = f (α)2 f (−α)2

we also have that

j
X
|f (α)|2 = bh e(−αh) (2.6)
h

23
where bh is the number of solutions to the equation

xk1 + · · · + xk2j−1 − y1k − · · · − y2kj−1 = h

with xi , yi ≤ N . Thus
X
bh = f (0)2j = N 2j
h

and, by the inductive hypothesis,

Z 1
b0 = |f (α)|2j dα N 2j−j+ε .
0

Now by (2.5), (2.6), and Parseval’s identity, we have

Z 1 X
|f (α)|2j+1 dα (2N )2j−j−1 ch bh .
0 h

Moreover,

X X
ch bh c0 b0 + N ε bh
h h6=0

N j N 2j−j+ε + N ε N 2j

which proves the lemma.

Lemma 5 (Weyl-van der Corput Inequality). Suppose ξ(n) is a complex valued function

such that ξ(n) = 0 if n 6∈ I = (a, b]. If Q is a positive integer then

2 X X
≤ |I| + Q |q|
X
ξ(n) 1 − ξ(n)ξ(n + q).
Q Q
n∈I q≤Q n,n+q∈I

Proof: We have

X Q X
X Q
XX
Q ξ(n) = ξ(n + q) = ξ(n + q).
n∈I q=1 n n q=1

24
Note that the inner sum is empty unless a − Q < n ≤ b − 1. By Cauchy’s inequality
2 Q 2
X
2
XX
Q ξ(n) ≤ (|I| + Q)

ξ(n + q)

n n k=1
Q Q
XXX
= (|I| + Q) ξ(n + k)ξ(n + l).
k=1 l=1 n

Now we collect the terms with l − k = q to obtain the result.

In applications of the Weyl-van der Corput inequality we take


 e(f (n)) if n ∈ I,

ξ(n) =
 0

otherwise

It is often necessary to apply a pruning technique to the major arcs. This is essentially

a result (in slightly different forms) that has been used in several previous papers (e.g.,

Lemma 2.6 of [91]). First, we give a definition of the major arcs, M(K),
K q
[ [ a K a K
M(K) = − , + .
q qN q qN
q=1 a=1
(a,q)=1

Here we provide the pruning lemma.

Lemma 6. For a large n and 1 < K n suppose G(α) is a function satisfying

G(α) q −(1+θ1 ) (1 + n|α − a/q|)−(1+θ2 ) , α ∈ M(2K)

where θ1 , θ2 > 0 are certain constants. The function

X
Φ(α) = η(h)e(αh)
|h|≤H

satisfies Φ(α) ≥ 0, η(h) ≥ 0 for all α, h and log(H) log(n). Then

Z X
−1+ −θ2
1−θ1
−θ1 −θ2
G(α)Φ(α) dα n η(0)(K +K )+ η(h) (K +K )
M(2K)\M(K) |h|≤H
h6=0

25
Proof: We have
q Z K/qn
q −(1+θ1 )
Z XX X
G(α)Φ(α) dα η(h)e(ah/q)e(hβ)dβ
M(2K)\M(K) −K/qn (1 + n|β|)1+θ2
q≤K a=1 |h|≤H
q Z K/qn
X X X
−(1+θ1 ) 1
η(h) e(ah/q)q e(hβ)dβ.
−K/qn (1 + n|β|)1+θ2
|h|≤H q≤K a=1

Here we bound the integral trivially to obtain

q
X XX K
η(h) e(ah/q)q −(1+θ1 ) (1 + K/q)−(1+θ2 ) .
qn
|h|≤H q≤K a=1

Noting that the sum over a is equal to q if q divides h and 0 otherwise, we have

X X K
η(h) q −(1+θ1 ) (1 + K/q)−(1+θ2 ) .
qn
|h|≤H q≤K
q|h

Thus, we have
X X
n−1 ( η(h) (q −(1+θ1 ) K + q θ2 −θ1 K −θ2 ).
|h|≤H q≤K
q|h

Hence
X
−1+ε 1−θ1 −θ1
n η(0)K + η(h)K .
|h|≤H
h6=0

Here, we also give a lemma for the special case with G(α) 1. We omit the proof as it

can be carried out similarly.

Lemma 7. Suppose the function

X
Φ(α) = η(h)e(αh)
|h|≤H

satisfies Φ(α) ≥ 0, η(h) ≥ 0 for all α, h and H N γ for some fixed γ > 0. Then, for any
1
Y N 2 L−c , we have
Z X
−1+
Φ(α) dα N Y η(0)Y + η(h) .
M(Y ) |h|≤H
h6=0

26
2.5 Exponent Pair Method

Weyl’s inequality is a useful bound for many applications of exponential sums, but some

problems require more sophisticated estimation. Additionally we often require an estimate

for an exponential sum where the exponent is not a polynomial. Here, we appeal to van

der Corput’s method of exponent pairs. Suppose F is differentiable on [N, 2N ]. We expect

an estimate of the form

X
S= e(F (n)) (A/N )k N l
N ≤n≤2N

where AN −1 |F 0 (n)| AN −1 . We call such a pair (k, l) an exponent pair. Clearly if (k, l)

and (µ, λ) are exponent pairs, then so is any linear combination (tk + (1 − t)µ, tl + (1 − t)λ)

for t ∈ [0, 1] because

S = S t S 1−t (A/N )tk+(1−t)µ N tl+(1−t)λ .

Trivially, we have that (0, 1) is an exponent pair. The power of the theory of exponent pairs

is the ability to obtain new pairs from known pairs. Thus, we can estimate an exponential

sum with this technique and choose the pair that yields the best bound. There are two

processes to determine new pairs from old pairs. They are known as the A process and B

process.

k k+l+1
ˆ A process: If (k, l) is an exponent pair, so is , .
2(k + 1) 2(k + 1)

ˆ B process: If (k, l) is an exponent pair, so is (l − 1/2, k + 1/2).

Here we include a list of commonly used exponent pairs. These are: ( 12 , 21 ) = B(0, 1);

( 61 , 23 ) = A( 12 , 12 ); ( 72 , 47 ) = BA( 61 , 23 ); ( 11 16 2 1 2 13 22 2 2 4 13 16
30 , 30 ) = BA ( 6 , 3 ); ( 40 , 40 ) = BA ( 7 , 7 ); ( 31 , 31 ) =

BAB( 11 16
30 , 30 ). For a complete treatment of exponent pair theory, we refer the interested

reader to chapter 2 of [45].

Exponent pairs are useful for a variety of exponential sum estimates. In particular,

we frequently use exponent pairs to bound the error term of the Fourier expansion of

27
ψ(t) = t − [t] − 1/2. Here we provide the expansion for ψ(t) that will be used in later

chapters. For t 6∈ Z and any J ≥ 1,

X 1 1
ψ(t) = − e(ht) + O min 1, , (2.7)
2πih J||t||
0<|h|≤J

where
∞
1 X
min 1, = bh e(ht),
J||t||
h=−∞
with

log 2J J
bh min , 2 .
J h
In addition to the exponent pair technique, van der Corput’s method can be used to

obtain other useful estimates for exponential sums. We include one such estimate which we

will use in the later chapters.

Lemma 8. If ∆ > 0 and F 00 (t) ∆ or (−F 00 (t) ∆) for N < t ≤ 2N then

X
e(F (n)) N ∆1/2 + ∆−1/2 .
n∼N

2.6 Double Large Sieve Method

We turn our attention to another method of exponential sum estimation known as the

double large sieve. The basis of this method was introduced in a work of Fouvry and Iwaniec

on exponential sums with monomials [32]. In their paper, they consider sums of the type
X X
··· e(f (m1 , . . . , mj )),
m1 mj
α
where the mi range over integers from an interval and f (m1 , . . . , mj ) = xmα1 1 · · · mj j is a

monomial. In the treatment of such sums, they regard them as a special case of bilinear

forms
XX
Bφψ (X , Y ) = φr ψs e(xr ys ),
r s

where X = (xr ), Y = (ys ) are finite sequences of real numbers with |xr | ≤ X, |ys | ≤ Y and

φr , ψs ∈ C. Their method is based off the following inequality of [9].

28
Lemma 9.

|Bφψ (X , Y )|2 ≤ 20(1 + XY )Bφ (X , Y )Bψ (Y , X)

with
X
Bφ (X , Y ) = |φr1 φr2 |
|xr1 −xr2 |≤Y −1

and Bψ (Y , X) defined similarly.

In the application of this lemma, we must determine the length of the summations

involved in the definition of Bφ (X , Y ). In our applications, we typically have

mα nβ
X = (xm ) = and Y = (yn ) =
M N

where m ∼ M and n ∼ N . Thus, it is natural to consider the spacing of the monomial term

mα nβ . We provide Fouvry and Iwaniec’s result in the following lemma.

Lemma 10. Let αβ 6= 0, ∆ > 0, M ≥ 1 and N ≥ 1. Let A (M, N ; ∆) be the number of

quadruples (m1 , m2 , n1 , n2 ) such that

m1 α
β
n1

m2 − < ∆,
n2

with m1 , m2 ∼ M and n1 , n2 ∼ N . We have

A (M, N ; ∆) M N log 2M N + ∆M 2 N 2 .

This lemma provides a method to estimate a common form of exponential sum. Oc-

casionally, more sophisticated spacing estimates are needed. In particular, if we apply a

Weyl shift, we introduce a term of the form tα (m, q) = (m + q)α − (m − q)α . Note that for

m ∼ M and q ∼ Q, |tα (m, q)| ∼ M α−1 Q = T . Thus, we require an estimate for the number

of quadruples (m1 , m2 , q1 , q2 ) such that

|tα (m1 , q1 ) − tα (m2 , q2 )| < ∆T. (2.8)

Let B(M, Q, ∆) denote the number of quadruples (m1 , m2 , q1 , q2 ) subject to (2.8). An upper

bound for B(M, Q, ∆) is provided in the following lemma due to Fouvry and Iwaniec.

29
Lemma 11. If Q ≤ M 2/3 , we have

B(M, Q, ∆) (M Q + ∆M 2 Q2 + M −2 Q6 )(log 2M )2

where the constant implied in depends on α only.

In their paper, Fouvry and Iwaniec applied Lemma 9, Lemma 10, and Lemma 11 in

combination with other analytic methods (such as Weyl-van der Corput inequality) to

obtain a number of bounds for exponential sums with monomials. Here we give one of their

results as well as the proof for completeness.

Lemma 12. Let α, α1 , α2 be real constants such that α 6= 1 and αα1 α2 6= 0. Let M, M1 , M2 , x ≥

1 and ϕm , ψm1 m2 be complex numbers with |ϕm |, |ψm1 m2 | ≤ 1. We then have,

mα mα1 1 mα2 2
X X X
Sϕψ (M, M1 , M2 ) = ϕm ψm1 m2 e x α α1 α2
M M1 M2
m∼M m1 ∼M1 m2 ∼M2

x1/4 M 1/2 (M1 M2 )3/4 + M 7/10 M1 M2

3/4 −1/4 11/10
M (M1 M2 ) + x M M1 M2 (log 2M M1 M2 )2 .

Proof: By the Weyl-van der Corput inequality we have

|Sϕψ |2 Q−1 M M1 M2 (M M1 M2 + |S(Q0 )| log 2Q),

for any Q ≤ 31 M and some Q0 ≤ Q, where

X t(m, q)mα1 1 mα2 2

XXX
q
S(Q0 ) = 1− ϕm+q ϕm−q e x
Q m m1 m2
M α M1α1 M2α2
q∼Q0

and

t(m, q) = (m + q)α − (m − q)α ∼ Q0 M α−1 .

Thus, by lemma 9 we have

S(Q0 ) (A BxQ0 M −1 )1/2 , (2.9)

30
where A is the number of quadruples (m1 , m2 , m̃1 m̃2 ) such that

m̃1 α1 m̃2 α2

−1

m1 − (xQ0 ) M,
m2

and B is the number of quadruples (m, m̃, q, q̃) such that

|t(m, q) − t(m̃, q̃)| x−1 M α .

Now from lemmas 10 and 11 we have

A M1 M2 log 2M1 M2 + (xQ0 )−1 M M12 M22 , (2.10)

and

B Q0 M (1 + x−1 M 2 )(log 2M )4 , (2.11)

provided 3Q0 M 3/5 . Now combining (2.9), (2.10), and (2.11), we have

S(Q0 ) Q0 (xM1 M2 )1/2 (1 + x−1 Q−1

0 M M 1 M2 )
1/2
(1 + x−1 M 2 )1/2 (log 2M M1 M2 )5/2 .

Notice that the worst value for Q0 is Q so we set Q = 13 M 3/5 to conclude the proof.

This lemma is one of many results from Fouvry and Iwaniec paper. For a complete list of

their results, we refer the reader to their work, [32].

31
CHAPTER 3

Representations of n = p21 + p22 + p23 + p2γ + qγ2

3.1 Introduction

In this chapter, suppose N ≡ 5 (mod 24) is a large integer. We consider the represen-

tations of a large integer, N , in the form N = p21 + p22 + p23 + p2γ + qγ2 , where p1 , p2 , p3 are

ordinary prime numbers and pγ , qγ are Piatetski-Shapiro primes of type γ. The first part of

the chapter is dedicated to using the Hardy-Littlewood method to reduce the problem to an

estimation on “small” major arcs. From there, the problem reduces to an exponential sum

estimate over a prime variable. The remainder of the chapter is dedicated to that estimate.

We begin by noting the number of representations is given by

Z 1 X 3 X 2
Rγ∗ (N ) = e(αp ) 2
e(αp2γ ) e(−N α)dα. (3.1)
0
p≤N 1/2 pγ≤N 1/2
pγ =[nc ]

For convenience we will consider a weighted number of representations instead of Rγ∗ (N ).

We define

X
g(α) = Λ(m)e(αm2 ),
m≤N 1/2
1 X
m1−γ Λ(m)e(αm2 ) [(m + 1)γ ] − [mγ ] ,

fγ (α) =
γ
m≤N 1/2

and let
Z 1
Rγ (N ) := g 3 (α)fγ2 (α)e(−N α)dα. (3.2)
0

In this section we will prove the following theorem.

32
Theorem 5. For Rγ (N ) defined above and γ > 7/8, we have

Γ5 (3/2)
S(N )N 3/2 1 + o(1) ,

Rγ (N ) =
Γ(5/2)

where S(N ) is the singular series, defined by

q
∞ X
S 5 (q, a) −aN
X
S(N ) = e .
ϕ5 (q) q
q=1 a=1
(a,q)=1

Moreover, S(N ) is absolutely convergent and S(N ) ≥ c > 0 for all N ≡ 5 (mod 24).

From this theorem, we have the following corollary.

Corollary 1. Every sufficiently large integer N can be represented in the form N = p21 +

p22 + p23 + p2γ + qγ2 provided γ > 7/8.

Note that the asymptotic formula

1
Γ5 (3/2) N 2 +γ
Rγ∗ (N ) = S(N ) (1 + o(1))
Γ(5/2) (log N )5

follows from the result of Theorem 5. The goal is to use the existing work on the Waring-

Goldbach problem of representing an integer as a sum of five prime squares. Recall that if

R2 (N ) is the weighted number of representations of N as a sum of five prime squares, we

have
Γ5 (3/2)
R2 (N ) ∼ S(N )N 3/2 (log N )−5 , (3.3)
Γ(5/2)

where the singular series is given in the above theorem. In this framework, we wish to show

that by replacing two of the primes with Piatetski-Shapiro primes, we obtain an analogous

result provided γ is not too small. That is, for Rγ (N ) given by (3.2), we wish to show

R2 (N ) − Rγ (N ) N 3/2−ε .

However, the presence of e(αm2 ) term prevents a suitable estimate for |g(α) − fγ (α)| for

large major arcs. To overcome this obstacle, we prune the major arcs to be as small as

33
possible. This works best when there are two or fewer Piatetski-Shapiro prime squares

among the five terms. We could replace more of the terms by Piatetski-Shapiro prime

squares, but the range for γ will be much more restrictive. For α in small major arcs, the

term e(αm2 ) can be well handled.

We proceed by using the Hardy-Littlewood method to split Rγ (N ) into two integrals

respectively over major and minor arcs. Suppose N is sufficiently large, and define

P = N 1/4 and Q = N ε . (3.4)

We define the major arcs by M = M(P ), where

K q
[ [ a K a K
M(K) = − , +
q qN q qN
q=1 a=1
(a,q)=1

and the minor arcs

m = U \M,

where U = [P/N, 1 + P/N ]. Thus, we rewrite Rγ (N ) as

Rγ (N ) = RM (N ) + Rm (N ),

where RM (N ) and Rm (N ) represent integrals over the major and minor arcs respectively.

3.2 The Minor Arcs

We bound the contribution from the minor arcs by

Z 1 1/2 Z 1 1/2
4 4
Rm (N ) max |g(α)| |g(α)| dα |fγ (α)| dα . (3.5)
α∈m 0 0

For the two integrals on the right side of (3.5), we will bound them by Hua’s inequality and

the next lemma respectively.

Lemma 13.
Z 1
|fγ (α)|4 dα N 2−γ+ε . (3.6)
0

34
Proof: Note that this integral is bounded by the number of integer solutions to

m21 − m22 = m23 − m24

with each mi weighted by a factor mi1−γ and mi ≤ N 1/2 for i = 1, 2, 3, 4. We consider the

solutions in two cases.

Case 1: m1 = m2 .

Note that this implies that m3 = m4 . Now, we have N 1/2 choices for m1 and N 1/2 choices

for m3 both weighted by m1−γ

i ≤ N (1−γ)/2 for i = 1, 3. Thus, we have at most

(N 1/2 N (1−γ)/2 )2 = N 2−γ

weighted solutions.

Case 2: m1 6= m2 . In this case, we have N 1/2 choices for m1 and m2 weighted by m1−γ
i ≤

N (1−γ)/2 for i = 1, 2. Thus we have at most

(N 1/2 N (1−γ)/2 )2 = N 2−γ

choices for m1 and m2 . In addition we have at most N ε choices for m3 and m4 . Thus the

number of solutions is at most O(N 2−γ+ε ), which completes the proof.

Also, we use the estimate on g(α) of Ghosh. We state the result in the following lemma.

Lemma 14. (Ghosh [34]) Let α ∈ R and suppose there exists a ∈ Z and q ∈ N such that

(a, q) = 1 and

|α − a/q| < q −2 .

Then for any positive ε, we have

g(α) N 1/2+ε (q −1 + N −1/4 + qN −1 )1/4 (3.7)

where the constant implied in the “” depends on at most ε.

35
Applying Hua’s inequality to the integral over g(α) and combining (3.6), and (3.7) into

(3.5), we have
2−γ
Rm (N ) N 1/2+ε (P −1 + N −1/4 )1/4 · N 2 N 1/2 .

Thus, we have

Rm (N ) N 2−γ/2+ε (P −1/4 + N −1/16 ),

which gives, by (3.4)

Rm (N ) N 2−1/16−γ/2+ε .

We clearly have Rm (N ) N 3/2−ε provided γ > 7/8 (and we can take any ε > 0 satisfying

ε < 41 (γ − 7/8)).

3.3 The Major Arcs

Now, our goal is to prune the major arcs to M(Q). We use the pruning technique

described in the preliminaries section. First, we require a major arc estimate for g(α). We

apply the bound (3.9) in lemma 15. First, we have

Z Z
|g(α)|3 |fγ (α)|2 dα log N max |g(α)|3 |fγ (α)|2 dα. (3.8)
M(P )\M(Q) QKP M(2K)\M(K)

Here we apply a theorem of Kumchev to the function g(α). We include the theorem here

as a lemma.

Lemma 15. (Kumchev [?]) Let k ∈ N and α ∈ R, and suppose that there exists a ∈ Z and

q ∈ N satisfying

1 ≤ q ≤ Q, (a, q) = 1, |qα − a| ≤ QN −1

with Q ≤ N 1/2 . Then, for any fixed ε > 0,

q ε N 1/2 (log N )c
g(α) Q1/2 N 11/40+ε + , (3.9)
(q + N |qα − a|)1/2

where c > 0 is an absolute constant and where the constant implied by the “” depends on

ε only.

36
Applying lemma 15 we see that (3.8) is bounded by

log N max (I1 + I2 ),

QKP

where
Z 1
I1 = K 1/2 N 11/40+ε |g(α)|2 |fγ (α)|2 dα, (3.10)
0

and
|fγ (α)|2
Z
3/2+ε
I2 = N dα. (3.11)
M(2K)\M(K) q 3/2 (1 + N |α − a/q|)3/2

From the estimate (3.6), Hua’s inequality, and P = N 1/4 , we have

1 11 1 2−γ 19 γ 3
I1 N 8 + 40 +ε · N 2 + 2
+ε
N 10 − 2 +ε N 2 −ε (3.12)

provided γ > 4/5 (which is less restrictive than γ > 7/8). Now, from lemma 6, we have
Z 1
3/2 −1+ε 1/2 2 −1/2 2
I2 N ·N K |fγ (α)| dα + K |fγ (0)| .
0

Note

X
Λ(m)m1−γ [(m + 1)γ ] − [mγ ]

|fγ (0)| =
m≤N 1/2
1−γ X
[(m + 1)γ ] − [mγ ] N 1/2 log N,

N 2 log N
m≤N 1/2

and
Z 1 X
|fγ (α)|2 dα m2−2γ Λ2 (m)([(m + 1)γ ] − [mγ ])2
0
m≤N 1/2
X
N 1−γ log2 N [(m + 1)γ ] − [mγ ]
m≤N 1/2

N 1−γ/2 log2 N.

37
We thus have,

I2 N 1/2+ε P 1/2 N 1−γ/2 Q−1/2 N N 3/2 N 1/8−γ/2+ε + N ε Q−1/2 .

Hence,

I2 N 3/2−ε (3.13)

provided Q N ε for some ε > 0 (not necessarily the same as the above ε). Therefore, we

have
Z
Rγ (N ) = g 3 (α)fγ2 (α)e(−N α)dα + O(N 3/2−ε ).
M(Q)

Hence, we have
Z
g 5 (α) − g 3 (α) fγ2 (α)dα + O(N 3/2−ε )

R2 (N ) − Rγ (N ) = (3.14)
M(Q)

by noting that
Z Z 1
|g(α)|5 dα max |g(α)| |g(α)|4 dα
U \M(Q) α∈U \M(Q) 0
0
N 3/2+ε Q−1/4 N 3/2−ε .

Thus, it remains to show the integral in (3.14) is O(N 3/2− ). Note

Z Z
g 5 (α) − g 3 (α) fγ2 (α)dα = g 3 (α)(g(α) + fγ (α))(g(α) − fγ (α))dα

M(Q) M(Q)

max |fγ − g| · I3 ,
α∈M(Q)

where
Z
I3 = |g 3 (α)(g(α) + fγ (α))|dα. (3.15)
M(Q)

Trivially, we have,

Q2 2+ε
|I3 | |M(Q)||g(0)|3 |fγ (0)| N = N 1+ε .
N
38
Thus, we have
Z
g 5 (α) − g 3 (α)fγ2 (α) dα max |fγ − g| · N 1+ε .

M(Q) α∈M(Q)

It is then clear that the theorem follows from the next lemma.

Lemma 16.

max |fγ − g| N 1/2−ε . (3.16)

α∈M(Q)

The remainder of the chapter is dedicated to proving this lemma.

3.4 Estimating |fγ − g|

From the definitions of fγ and g, we have

X 1 1−γ
|fγ (α) − g(α)| = Λ(m)e(αm2 ) m [(m + 1)γ ] − [mγ ] − 1
1/2
γ
m≤N
X 1
= Λ(m)e(αm2 ) m1−γ (ψ(mγ ) − ψ((m + 1)γ ))
γ
m≤N 1/2

X 1 1−γ
+ e(αm2 )Λ(m) m (m + 1)γ − mγ − 1 .
1/2
γ
m≤N

Note that the last sum is bounded by

X 1 1−γ γ γ

Λ(m) m (m + 1) − m − 1
γ
m≤N 1/2

X 1 1−γ γ γ−1 γ−2 γ

Λ(m)
m m + γm + O(m ) − m − 1
1/2
γ
m≤N
X
Λ(m)m−1 log N.
m≤N 1/2

Thus, we have


2 1 1−γ
X γ γ

|fγ − g|
Λ(m)e(αm ) m (ψ(m ) − ψ((m + 1) )) + O(log N ). (3.17)
m≤N 1/2 γ

39
1
−ε
We also note that the terms with m ≤ N 2(2−γ) contribute at most O(N 1/2−ε ). It is then
1
sufficient to show that, for N 2(2−γ) ≤ x ≤ N 1/2 ,

X
Λ(m)e(αm2 )m1−γ (ψ(mγ ) − ψ((m + 1)γ )) x1−ε .
m∼x

To estimate this sum, we replace ψ(t) by (2.7). First, we estimate the error term of the

expansion. We have
∞

X X x2−γ log J X X
Λ(m)e(αm2 )m1−γ bh e(hmγ ) + x1−γ log x |bh | e(hmγ ) .

m∼x
J m∼x
h=−∞ h6=0
(3.18)

Applying lemma 8 with ∆ = |h|xγ−2 to the last sum in (3.18), we have

X x2−γ log J
x1−γ log x |bh |(|h|1/2 xγ/2 + |h|−1/2 x1−γ/2 ) +
J
h6=0
 
X log J X J 2−γ log J
x1−γ log x  +  (|h|1/2 xγ/2 + |h|−1/2 x1−γ/2 ) + x
J h2 J
|h|≤J |h|>J

x2−γ log J
(x1−γ/2 J 1/2 + J −1/2 x2−3γ/2 ) log x +
J
1
x1−ε N 2 −ε

1−γ
+ε
if taking J = x1−γ+ε N 2 .

Thus, it remains to estimate the “main” error term of these expansions. That is, we

require an estimate for

X 1 X
1−γ 2 γ γ
Λ(m)m e(αm ) e(h(m + 1) ) − e(hm . (3.19)
h m∼x
h∼H

40
Note that
Z 1
γ γ
e(h(m + 1) ) − e(hm ) = 2πihγ(m + u)γ−1 e(h(m + u)γ )du
0
Z 1
1
= 2πihγ mγ−1 (1 + O( ))e(h(m + u)γ )du
0 m
Z 1
= 2πiγhmγ−1 e(h(m + u)γ )du + O(hmγ−2 ),
0

and the error term O(hmγ−2 ) contributes at most

X 1 X
m1−γ Λ(m)hmγ−2 H log x x1−γ+ε x1−ε ,
h m∼x
h∼H

which is admissible. By partial summation, it remains to show that, for any u ∈ [0, 1] and

H ≤ J = x1−γ+ε ,
X X
Λ(m)e(αm2 + h(m + u)γ ) x1−ε . (3.20)
h∼H m∼x

3.5 Estimating Type I and Type II Sums

In this section we show that (3.20) holds for γ > 7/8. Following the use of Vaughan’s

identity, we require estimates for sums of the form

X X X
TypeIsum : a(m)e(αm2 n2 + h(mn + u)γ )
h∼H m∼X n∼Y

and
X X X
TypeIIsum : a(m)b(n)e(αm2 n2 + h(mn + u)γ ),
h∼H m∼X n∼Y
1
known as type I and type II sums respectively. Here XY ∼ x with N 2(2−γ) x N 1/2 .

• Type II Sum Estimate

a Q
We rewrite α = q + β with 1 ≤ a ≤ q ≤ Q, (a, q) = 1 and |β| ≤ N. Thus, the type II sums

are in the form

am2 r2
X X X X
2 2 γ
a(m)b(ql + r)e e βm (ql + r) + h(m(ql + r) + u) . (3.21)
q
h∼H m∼X r≤q l∼Y /q

41
Note that |e(am2 r2 /q)| ≤ 1 and that this term does not depend on l. Thus, we absorb

it in the coefficient a(m) to get a1 (m), say (we will let q, a, r be fixed in our treatment

of the sums). From here, we look to reduce the problem to the traditional problem of

finding Piatetski-Shapiro primes. In this regard, we will use some known estimates for the

Piatetski-Shapiro prime problem to bound our sums. Note that (3.21) is bounded by

q max |SII |,
r≤q

where
X X X
a1 (m)b1 (n)e βm2 n2 + h(mn + u)γ ,

SII =
h∼H m∼X n∼Y

where a1 (m), b1 (m) xε . Here, after taking out the factor e(am2 r2 /q), we have recovered

n = ql + r but with b(n) = 0 if n 6≡ r (mod q). Next we reduce this SII to the form

X X X
â(m)b̂(n, h)e(hmγ nγ ), (3.22)
h∼H m∼X n∼Y

where

|â(m)|, |b̂(n, h)| xε .

in order to apply some results on the Piatetski-Shaprio prime number theorem. To begin,

we need to break apart the term h(m(ql + r) + u)γ . We have,

h(mn + u)γ = hmγ nγ + γuhmγ−1 nγ−1 + O(Hxγ−2 ). (3.23)

Substituting this into SII gives

X X X
SII = a1 (m)b1 (n)e(βm2 n2 )e(γuhmγ−1 nγ−1 )e(hmγ nγ ) + O(xε ).
h∼H m∼X n∼Y

Here, our goal is to separate m and n in the terms e(βm2 n2 ) and e(hmγ−1 nγ−1 ). This

way we can group the appropriate terms into new coefficients, say â and b̂, and apply some

results from the double large sieve. To accomplish this, let ν > 0 be absolute and sufficiently

42
small, and partition the sum over m into xν intervals. We write
xν
[ X(s − 1) Xs
(X, 2X] = I(X, s) where I(X, s) = X + , X + ν = (Ms , Ms+1 ].
xν x
s=1
(3.24)

For m ∈ I(X, s) consider

βm2 n2 − βMs2 n2 = βn2 (m2 − Ms2 )

Q 2 X x2−ν
Y X · 4ε 1−ε N −ε .
N x N

Similarly,

γuhnγ−1 mγ−1 − γuhnγ−1 Msγ−1 = γuhnγ−1 (mγ−1 − Msγ−1 )

X γ−1
HY γ−1 N −ε .
xν

Thus for m ∈ I(X, s), we rewrite

e(βm2 n2 ) = e(βMs2 n2 )e(βn2 (m2 − Ms2 ))

and

e(γuhmγ−1 nγ−1 ) = e(γuhMsγ−1 nγ−1 )e(γuhnγ−1 (mγ−1 − Msγ−1 )).

Note that the terms e(βMs2 n2 ) and e(γuhMsγ−1 nγ−1 ) are only dependent on n and h. Thus,

we can absorb these terms into a new constant say b2 (n, h) in the treatment of SII . Note

that |b2 (n, h)| ≤ 1 since the absolute value of these exponential terms is always less or equal

to 1. Now we must separate the remaining terms. For given δ, η > 0 we can take Kδ = [δ/ε]

and Tη = [η/ε], and then

e(βn2 (m2 − Ms2 )) = S1 (Kδ ) + O(x−δ ), (3.25)

43
where
Kδ k
X 2πiβn2 (m2 − Ms2 )
S1 (Kδ ) = ,
k!
k=0

and

e(γuhnγ−1 (mγ−1 − Msγ−1 )) = S2 (Tη ) + O(x−η ), (3.26)

where
Tη t
X 2πiγuhnγ−1 (mγ−1 − Msγ−1 )
S2 (Tη ) = .
t!
t=0

Substituting (3.25) and (3.26) into SII gives

X XX X X
SII = a1 (m)b2 (n, h)S1 (Kδ )S2 (Tη )e(hmγ nγ )
h∼H s≤xν r≤q l∼Y /q m∈I(X,s)

+ O(Hx1+ν−δ−η ),

which we rewrite as

XXXXXX
= ât,k (m)b̂t,k (n, h)e(hmγ nγ ) + O(Hx1+ν−δ−η ). (3.27)
t k h s n m

Note that the error term is admissible provided δ + η > 1 − γ + 5ε. Thus, we choose

Kδ = Tη = 1ε ε 1 such that this condition is met. Here we have absorbed the terms

from the power expansions, S(Kδ ) and S(Tη ), into new constants ât,k (m) and b̂t,k (n, h).

Note that |ât,k (m)| xε and |b̂t,k (n, h)| xε because |β k n2k (m2 − Ms2 )k | (N −ε )k and

|(γuh)t nt(γ−1) (mγ−1 − Msγ−1 )t | (N −ε )t . Thus, we estimate

Tη
Kδ X
X XX X X
ât,k (m)b̂t,k (n, h)e(hmγ nγ ).
h∼H s≤xν k=0 t=0 n∼Y m∈I(X,s)

We apply a result of Fouvry and Iwaniec [32] to bound SII . For convenience, we state the

lemma next.

Lemma 17. (Fouvry and Iwaniec) [32] Let α, α1 , α2 be real constants such that α 6=

1, αα1 α2 6= 0. Let M, M1 , M2 , x ≥ 1 and φm , ψm1 m2 be complex numbers with |φm |, |ψm1 m2 | ≤

44
1. We then have

mα mα1 mα2
X X X
Sφψ = φm ψm1 m2 e x α 1α1 2α2
M M1 M2
m∼M m1 ∼M1 m2 ∼M2

x1/4 M 1/2 (M1 M2 )3/4 + M 7/10 M1 M2 + M (M1 M2 )3/4 + x−1/4 M 11/10 M1 M2 log2 M M1 M2 .

We apply the above lemma to SII with x = Hxγ , M = X, M1 = H, M2 = Y . We have,

SII (H 1/4 xγ/4 X 1/2 (HY )3/4 +X 7/10 HY +X(HY )3/4 +H −1/4 x−γ/4 X 11/10 HY )xε . (3.28)

Simplifying this bound and using XY = x and H = x1−γ+ε , we obtain

SII (HY 1/4 x1/2+γ/4 + HY 3/10 x7/10 + H 3/4 X 1/4 x3/4 + H 3/4 X 1/10 x1−γ/4 )xε (3.29)

Y 1/4 x3/2−3γ/4+ε + Y 3/10 x17/10−γ+ε + X 1/4 x3/2−3γ/4+ε + X 1/10 x7/4−γ+ε (3.30)

(3.31)

Thus we have |SII | x1−ε provided X and Y satisfy

10 7
X x3γ−2−ε and Y x 3 γ− 3 −ε . (3.32)

From the symmetry of m and n, we are left to treat the case when either X or Y is

x3(1−γ)+ε . The result follows from an estimate of Heath-Brown [37] for the type II sum.

This result is stated as lemma 4 in [37].

Lemma 18. Let η > 0 be constant, and suppose 1/2 + η ≤ γ ≤ 1. Let

X∞
X X
γ γ

L = sup d4 (m) g(n) εh e(hm n ) ,
m=1 u<n≤v
0<h≤H

N <mn≤N1

where the supremum is taken over all arithmetic functions g(n) for which |g(n)| ≤ d3 (n).

Then

L N 1−η/γ

when

u = N 1−γ+η , v = N 5γ−4−η .

45
Thus, applying this lemma to our type II sum, we have that |SII | x1−ε , provided

x1−γ+ε Y x5γ−4−ε (3.33)

Note that 5γ − 4 > 3(1 − γ) if γ > 7/8. Thus we have shown that SII x1−ε provided

X x1−γ+ε and Y x1−γ+ε .

• Type I estimate

Now we give an upper bound for

X X X
SI = a(m)e(αm2 n2 + hmγ nγ )
h∼H m∼X n∼Y
X X X X
= a(m)e(αm2 (ql + r)2 + hmγ (ql + r)γ ).
h∼H m∼X r≤q l∼Y /q

Q
Now, let α = a/q + β , |β| ≤ N, and q ≤ Q. We have

am2 r2
X X X X
SI = a(m)e(βm2 (ql + r)2 )e e(hmγ (ql + r)γ ).
q
h∼H m∼X r≤q l∼Y /q

We absorb the e(am2 r2 /q) term into a(m) to get a new constant â(m). Thus, we have

X X X X
SI = â(m)e(βm2 (ql + r)2 + hmγ (ql + r)γ ).
h∼H m∼X r≤q l∼Y /q

We let f (l) = βm2 (ql + r)2 + hmγ (ql + r)γ . Then, f 0 (l) = β2m2 q(ql + r) + γqhmγ (ql + r)γ−1 .

Now, |β| ≤ Q/N and thus, βm2 q(ql + r) ≤ Q2 /Y . Hence we have

|f 0 (l)| hqX γ Y γ−1 Hqxγ Y −1 > 1.

Here we apply the exponent pair (2/7, 4/7) to the sum over l to obtain the estimate

X
e(hmγ (ql + r)γ ) H 2/7 q −2/7 x2γ/7 Y 2/7 .
l∼Y /q

46
Thus, we have that

SI H 9/7 q 5/7 x2γ/7 XY 2/7 . (3.34)

Since H x1−γ+ε and q Q xε (3.34) gives

9 2γ
+ 27
SI x 7 (1−γ)+ε+ 7 X 5/7 x11/7−γ+ε X 5/7 (3.35)

which is x1−ε provided

7γ−4
+ε
Xx 5 . (3.36)

• The Vaughan Identity

Here, we combine the estimates for the type I and type II sums in the Vaughan identity.

We begin with the required ranges for our type I and type II estimations. We have,

X ≤ uv for type I sum

v ≤ Y ≤ x/u for type II sum.

We choose u = v = x1−γ+ε . Recall that SII x1−ε provided Y x1−γ+ε = v. Finally, we

7γ−4
+ε
have that SI x1−ε provided X x 5 , and note that

7γ − 4
2(1 − γ) <
5

when γ > 7/8. Thus, (3.20) holds provided γ > 7/8.

47
CHAPTER 4

Representations of n = [mc ] + pk

In this chapter we consider the number of representations of all sufficiently large integers,

N , in the form N = [mc ] + pk for a fixed k. The goal is to find the largest value of c > 1.

We begin by determining a lower bound for the main term in the asymptotic formula of

the number of such representations. From there, we estimate the exponent pair technique

to bound the error term. We have a representation of N = [mc ] + pk if and only if

(N − pk )γ ≤ m < (N − pk + 1)γ ,

where γ = 1/c. Let

X
S= Λ(n) [−(N − nk )γ ] − [−(N + 1 − nk )γ ] . (4.1)
n≤ 12 N 1/k

1 1/k
Then S essentially counts the (weighted) number of representations with p ≤ 2N .

Clearly, we have S = S1 + S2 where

X
S1 = Λ(n) (N + 1 − nk )γ − (N − nk )γ (4.2)
n≤ 21 N 1/k

and
X
S2 = Λ(n) ψ(−(N + 1 − nk )γ ) − ψ(−(N − nk )γ ) . (4.3)
n≤ 21 N 1/k

48
4.1 The Treatment of S1

We have,

X
k γ k γ
X
k γ 1 γ
Λ(n) (N + 1 − n ) − (N − n ) = Λ(n)(N − n ) (1 + ) −1
N − nk
n≤ 12 N 1/k n≤ 12 N 1/k

X
k γ γ −2
= Λ(n)(N − n ) + O(N )
N − nk
n≤ 12 N 1/k
X 1
=γ Λ(n)(N − nk )γ−1 + O(N γ−2+ k +ε )
n≤ 12 N 1/k
X 1
≥ γN γ−1 Λ(n) + O(N γ−2+ k +ε
n≤ 12 N 1/k

1
N γ+ k −1 .

Remark: For k = 1, it is fairly straightforward to obtain an asymptotic formula for the

number of weighted representations by extending the range of n in (4.1) to N . For k ≥ 2,

things become more difficult because we have very different situations according to how

close N is to a k-th power. Here, we restrict the range of n to avoid this unnecessary

complication as our main purpose is to show the existence of such representations for c > 1.

4.2 The Treatment of S2

We look to estimate
X
Λ(n)ψ −(N − nk )γ . (4.4)
n≤ 12 N 1/k

In this regard, we apply the Fourier expansion of ψ(t). We replace ψ(−(N − nk )γ )) in (4.4)

with the exponential sum given in (2.7). We have

∞
!
X X 1 k γ
X X
k γ
Λ(n) e(h(N − n ) ) + O Λ(n) bh e(h(N − n ) ) . (4.5)
2πih n
n≤ 21 N 1/k 0<|h|≤H h=−∞

49
4.2.1 The Error Term

We begin by analyzing the error term in (4.5). We have

X ∞
X ∞
X

k γ
X k γ

Λ(n) bh e(h(N − n ) ) log N |bh | e(h(N − n ) ).
n≤ 12 N 1/k h=−∞ h=−∞ n

To estimate this term, we break the sum over n into O(log N ) sums of the form x ≤ n < 2x

for each x ≤ N 1/k . This gives,

∞
X

2
X k γ

log N |bh |
e(h(N − n ) ).
h=−∞ n∼x

From here, we apply the exponent pair (µ, λ) to the sum over n. Thus, we have that
X
e(h(N − nk )γ ) (hN γ−1 xk−1 )µ xλ = (hN γ−1 )µ xµ(k−1)+λ .
n∼x

Now, the sum over h is bounded by

∞
X X log 2H X H
|bh |hµ hµ + hµ
H h2
h=−∞ |h|≤H |h|>H

1 µ+1+ε X
H +H hµ−2
H
|h|>H

H µ+ε .

Thus, the error term is bounded by

(HN γ−1+ε )µ xµ(k−1)+λ

for some exponent pair (µ, λ). Here we choose H xk(1−γ)+ε , which gives the bound

xµ(k−1)+λ+ε . (4.6)

4.2.2 The “Main” Error Term

Here we consider the sum

X X
Λ(n)e(h(N − nk )γ ) (4.7)



h∼H n∼x

50
We use a combinatorial identity known as Vaughan’s identity to split (4.7) into the so called

type I and type II sums. We begin with an estimate for the type I sum.

• Type 1 Sum

We begin with,
XX
ar e(h(N − nk )γ ) (4.8)
r∼R t∼T

where RT = x. We estimate (4.8) by

X X X X
ar e(h(N − (rt)k )γ ) ≤ |ar | e(h(N − (rt)k )γ ) .



r∼R t∼T r∼R t∼T

Here we apply an exponent pair to the sum over t. We apply the generic pair (µ, λ), and

we have, µ
HN γ−1 xk
X
k γ
e(h(N − (rt) ) ) T λ.

T


t∼T

Thus (4.8) is bounded by

R(HN γ−1 xk )µ T λ−µ xkµ+ε RT λ−µ (4.9)

from the choice of H.

• Type 2 Sum

Now we look to bound the type II sum. That is, we must estimate a sum of the form

XX
ar bt e(h(N − (rt)k )γ ). (4.10)
r∼R t∼T

Here we take the square of (4.10) and apply the Weyl-van der Corput inequality. For Q ≤ T ,

we have
2
x2 x X X X
X X
k γ k k γ k γ

ar bt e(h(N − (rt) ) ) + bt+q bt e h (N −r (t+q) ) −(N −(rt) ) .

Q Q


r∼R t∼T q≤Q t,t+q∼T r∼R
(4.11)

Let
Z q
k k γ k γ
f (r, t) = h (N − r (t + q) ) − (N − (rt) ) = −hγk (N − rk (t + u)k )γ−1 (t + u)k−1 du.
0

51
Then,

f (r, t) HqN γ−1 xk T −1 ,

and
∂
f (r, t) HqN γ−1 xk T −1 R−1 .
∂r

Again, we apply an exponent pair (probably different from the type I pair) (µ, λ) to the

sum over r and we have that this sum is bounded by

(HQN γ−1 xk T −1 )µ Rλ−µ . (4.12)

Combining the estimate (4.12) with (4.11), we have that

x2
|SII |2 + xT (HQN γ−1 xk T −1 )µ Rλ−µ .
Q

Now from H xk(1−γ)+ε , we have that

1−µ 1+kµ λ−µ

|SII | xQ−1/2 + T 2 x 2
+ε
Qµ/2 R 2 . (4.13)

• Restrictions on R and T

For the type I sum, we have

|SI | RT λ−µ (HN γ−1 xk )µ .

We choose H = N 1−γ+ε . Thus,

|SI | RT λ−µ xkµ .

We require this estimate to be bounded by N γ−1+1/k−ε with (x = N 1/k ). Thus, for the

type I sum, we have

1
RT λ−µ xkµ N γ−1+ k −ε and T ≥ Q = N 2(1−γ)+ε .

This gives the restriction

γ−1−µ 1 γ−1−µ
− λ−µ−1
T N λ−µ−1 or R Nk . (4.14)

52
We require the same restriction on the type II sums. Here we let Q = N 2(1−γ)+ε in (4.13).

Thus, we require

1−µ 1+kµ λ−µ

T 2 N (1−γ)µ x 2 R 2 N γ−1+1/k−ε and T ≥ Q = N 2(1−γ)+ε .

This gives the restrictions:

kµ−µ+2k(1−γ)(µ+1)
RN k(1−λ) and T N 2(1−γ)+ε . (4.15)

• The Vaughan Identity

Using the Vaughan identity restricts the ranges of our sums on r and t. In particular, we

must find a u and v satisfying

x
Type I: R ≤ uv, T ∼
r
x x
Type II: u<R< , v<T < .
v u

Thus, from (4.15) we require

(k−1)µ+2k(1−γ)(µ+1)
uN k(1−λ) and v N 2(1−γ)+ε . (4.16)

From (4.14), we have

γ−1−µ x 1+µ−γ
T N λ−µ−1 ⇒ N 1+µ−λ .
uv

Thus, we have an upper bound for uv given by

1 1+µ−γ
− 1+µ−λ
uv N k . (4.17)

In both (4.16) and (4.17), (µ, λ) represent an exponent pair. However, the pair for the

upper bound and lower bound may be different. Thus, to ensure a representation of N as

N = [mc ] + pk , we require

(k − 1)µ2 + 2k(1 − γ)(µ2 + 1) 1 1 − γ + µ1

< + − 2(1 − γ),
k(1 − λ2 ) k λ1 − µ 1 − 1

53
where (µ1 , λ1 ) and (µ2 , λ2 ) are exponent pairs. This imposes the restriction

1

+ µ1 − λ1 ) 1 − λ2 − (k − 1)µ2 − µ1 (1 − λ2 )
k (1
1−γ < . (4.18)
2(µ2 + 1)(1 + µ1 − λ1 ) + 2(1 − λ2 )(1 + µ1 − λ1 ) + (1 − λ2 )

5 663 10 510
In the case k = 2, we let (µ1 , λ1 ) = ( 199 , 746 ) and (µ2 , λ2 ) = ( 189 , 623 ) which gives

4
1−γ < .
493

54
CHAPTER 5

Representations of N = m2 + p3 + q 5

In this chapter, we consider the representations of

n = m2 + p3 + q 5 , (5.1)

where m is squarefree and p, q are primes. We begin by providing some Weyl type esti-

mates for exponential sums over squarefree numbers. From there, we apply some pruning

techniques to reduce the size of the major arcs. The remainder of the chapter consists

of factoring the integral over the major arcs into the singular series and singular integral,

bounding the contribution from the minor arcs, and the treatment of the singular series. In

this chapter, we use the following notations. For α ∈ R, and a ∈ Z, q ∈ N, we define

X X
fk (α) = e(αmk ) gk (α) = Λ(m)e(αmk )
N N
10
<mk ≤N 10
<mk ≤N
1
Z Nk
X
2 2
h(α) = µ (m)e(αm ) vk (α) = e(αtk )dt
N 1
10
<m2 ≤N N )k
( 10
q k q
ahk

X ah X
Sk (q, a) = e Ck (χ, a) = χ(h)e
q q
h=1 h=1

We also let Ck (χ0 , a) = Ck (q, a). We should note that, by Euler-MacLaurin summation

formula, we easily have

fk (α) = vk (α) + O(1 + |α|N ). (5.2)

For two Dirichlet characters χ1 , χ2 (mod q), let

q
X µ(d) X
4 an
BD (n, q, χ1 , χ2 ) = S2 (q, ad )C3 (χ1 , a)C5 (χ2 , a)e − , (5.3)
d2 a=1 q
d≤D
(a,q)=1

55
and
X BD (n, q)
BD (n, q) := BD (n, q, χ0 , χ0 ), SD (n, P ) := , (5.4)
q(φ(q))2
q≤P

and S(n, P ) = S∞ (n, P ).

5.1 Exponential Sums over Squarefree Numbers

In this section, we study h(α). First, we note that

X X
h(α) := µ(d) e(αd4 s2 ). (5.5)
1 N
d≤N 4 <s2 ≤ N4
10d4 d

We let
X X
hD (α) := µ(d) e(αd4 s2 ),
N
d≤D <s2 ≤ N4
10d4 d

and

GK (α) := h2K (α) − hK (α).

We first give a minor arc estimate for hD (α). It should be remarked that [91] contains a

minor arc estimate (Lemma 3.2) which is not in ready-to-use form for our purpose. Following

the proof of Lemma 3.2 of [91] (which is a modification of Lemma 6 of [2]), one should be

able to get an estimate of the same strength as the following lemma, but our proof is much

easier and more straightforward.

1 1
a
Lemma 19. Suppose D N 6 , Q < N 2 , α = q + β where a ∈ Z, q ∈ N satisfy (a, q) = 1,
N Q
q≤ Q, and |β| < qN . Then we have

1 1 1 1 1
hD (α) N (N 2 q − 2 D 2 + q 2 D + Q 2 D).

Proof: It suffices to show

1 1 1 1 1
GK (α) N (N 2 q − 2 K 2 + q 2 K + Q 2 K). (5.6)

56
for K D. By Cauchy’s inequality, we have
2
X X
|GK (α)|2 K 4 2


e(αd s )
N
d∼K <s2 ≤ N4
10d4 d
X X X
= K e(αd4 ((y + x)2 − y 2 ))
√
d∼K |x|≤
q q
N N N
d2
<y+x,y≤
10d4 d4
√
X X N 1
K min , .
√ d2 ||2αd4 x||
d∼K |x|≤ N
d2

4QK 2
Note that |2βd4 x| < In the sum, we write q1 := (2d4 , q), and q = q0 q1 , then we get
√ .
q N
√
X X N
|GK (α)|2 K
d∼K
√ K2
|x|≤ N
4
(2d ,q)=q1 K 2 2 2
2ad4qq1 −1 x ≤ 4QK
√ +√ K
0 q N N

2ad4 q1 −1 x −1

X X
+K . (5.7)
d∼K
√
q 0

|x|≤ 2 N
(2d4 ,q)=q1 K
2ad4qq1 −1 x > 4QK
√
2
+√K2
0 q N N

The first sum in (5.7) is simply bounded by

√ √
q0 K 2 QK 2

N X N
1+ 1+ √ + √
K d∼K
q0 K 2 N q1 N
(2d4 ,q)=q1
√ √
QK 2 q0 K 2

N X Q N
+ √ + √ +
K d∼K
q q1 N N q0 K 2
(2d4 ,q)=q1
√ X 1 N X
N Qq −1 + (q + Q)K + q1
d∼K
q1 qK 3 d∼K
(2d4 ,q)=q1 (2d4 ,q)=q1
√
N ( N Qq −1 + qK 2 + QK 2 + N Kq −1 ),

where for the last sum we used trivial estimate

X X X
q1 q1 1 qK 4.
d∼K q1 |q m∼K 4
(2d4 ,q)=q1 q1 K 4 q1 |m

The second sum in (5.7) is bounded by

√ √
X N X
N
K 1+ 2
q0 log(2q0 ) N K q0 + 2
d∼K
q 0 K d∼K
K
(2d4 ,q)=q1 (2d4 ,q)=q1
√
N (qK 2 + N ).

57
√
From the above estimates and (5.7), and that N qK 2 + N Kq −1 , we have proved (5.6).

Next, we state some major arc approximations of hD (α) and h(α).

1 1
P Lc
Lemma 20. Suppose D N 6 , P N 2 − . α = a
q + β with |β| < qN and a ∈ Z, q ∈ N

satisfying (a, q) = 1, and q ≤ P . Then we have

1 1
hD (α) = VD (α) + O(Dq 2 + (1 + |β|N ) 2 ), (5.8)

where
µ(d)S2 (q, ad4 )
X
VD (α) = v2 (β)
qd2
d≤D

satisfies
1 1
VD (α) N 2 q − 2 + (1 + N |β|)−1 . (5.9)

Proof: (5.8) is essentially a special case of Lemma 2.2 (a) of [91]. (5.9) follows from the

estimates
√
N X µ(d)S(q, ad4 ) X q 21 (q, d4 ) 12 1
v2 (β) , and 2
2
q − 2 + .
1 + N |β| qd qd
d≤D d≤D

1
Lemma 21. For K N 4 , we have
Z 1 16 1
H(K) := |GK (α)g3 (α)g5 (α)|2 dα N 15 + (N − 30 + K −2 ). (5.10)
0

Proof: (5.10) is a direct consequence of the following estimates

16 1 1 1
H(K) N 15 + K − 2 (1 + N 20 K − 2 ), (5.11)

31 16 2 2 22
H(K) N 30 + + N 15 + K − 3 + N 5 + K 3 , (5.12)

and
31 16 2
H(K) N 30 + + N 15 + K −2 + N 3 + K 6 + K 14+ , (5.13)

58
1 1 1 1
(which respectively yield (5.10) in the ranges K N 12 , N 20 K N 12 , and K N 20 .)

By Hua’s inequality (one can also refer to [91], Lemma 2.8), we have
Z 1 5
Z 1
8
|g3 (α)| dα N 3
+
and |GK (α)|4 dα N 1+ K −2 ,
0 0

and
Z 1 4
Z 1 Z 1 4 9
|GK (α)|2 |g5 (α)|4 dα N 5 + + |GK (α)|2 dα |g5 (α)|4 dα N 5 + + N 10 K −1 .
0 0 0

Hence by Hölder’s inequality, we have

Z 1 1 Z 1 1 Z 1 1
4 2 4
8 2 4 4
H(K) |g3 (α)| dα |GK (α)| |g5 (α)| dα |GK (α)| dα
0 0 0
5 2 9
− 12 1
− 12
N N 12 (N + N 5 20 K )N K 4

16 1 1 1
N 15 + K − 2 (1 + N 20 K − 2 ),

which proves (5.11). Next we prove (5.13). By Cauchy’s inequality, we have

XZ 1 X
H(K) K | e(αd4 s2 )|2 |g3 (α)g5 (α)|2 dα KR, say,
0 N
d∼K <s2 ≤ N4
10d4 d

where R is the number of solutions of the Diophantine equation

x1 3 − x2 3 + y1 5 − y2 5 = d4 (z1 2 − z2 2 )

subject to the conditions

N
K < d ≤ 2K, and < xj 3 , yj 5 , d4 zj 2 ≤ N.
10

We note that

1
Z 1 X X
R N 2 K −1 |f3 (α)f5 (α)|2 dα + N 1
0 d∼K xj 3 ,yj 5 ∼N
x1 3 −x2 3 +y1 5 −y2 5 ≡0( mod d4 )
31
N 30 + X N 13 2 1
N5
2 X

+N +1 +1 1, (5.14)
K K4 K4
d∼K xj ,yj ( mod d4 )

59
where the last sum over xj , yj is also subject to x1 3 − x2 3 + y1 5 − y2 5 ≡ 0(modd4 ). Thus

(5.13) immediately follows if we can prove that, for every positive integer q,

X
T (q) := 1 q 3+ . (5.15)
xj ,yj ( mod q)
x1 3 −x2 3 +y1 5 −y2 5 ≡0( mod q)

First we have
q q q
3 2 X 5 2
1 X X hx hy
T (q) = e e
q q q
h=1 x=1 y=1
q0
1 X 4 X
= t |S3 (q0 , h0 )S5 (q0 , h0 )|2 .
q q=q t h0 =1
0
(h0 ,q0 )=1

1
From the estimate Sk (q0 , h0 ) q0 1− k (c.f., [85] Theorem 4.2), we have
q0
1 X 4 X 44 X 1
T (q) t q0 15 + q 3 q0 − 15 + q 3+ .
q q=q t h0 =1
0 q0 |q
(h0 ,q0 )=1

Hence we have proved (5.15).

To show (5.12), we follow the proof of (5.13) except that, in the last step of (5.14), we

replace the sum by

X N 31 2 X
+1 1.
K4 1
d∼K
xj ( mod d4 ),yj ∼N 5
x1 −x2 +y1 −y2 ≡0( mod d4 )
3 3 5 5

It is then clear that (5.12) follows from the following estimate

X X 2 19
1 N 5 + K 3 . (5.16)
d∼K 1
xj ( mod d4 ),yj ∼N 5
3 3
x1 −x2 +y1 −y2 ≡0( mod d4 )
5 5

60
Note that, for a positive integer q satisfying log q L, we have
q q 3 2 X 5 2
X 1 X X hx hy
1 = e e
q q q
1 h=1 x=1 1
xj ( mod q),yj ∼N 5 y∼N 5
x1 3 −x2 3 +y1 5 −y2 5 ≡0( mod q)
2 q q 3 2
N 5 X X hx
e
q q
h=1 x=1
2 q0
N 5 X
2
X
= t |S3 (q0 , h0 )|2
q q=q0 t h0 =1
(h0 ,q0 )=1
2
N 5 X 2 7 + 2 4
t q0 3 N 5 + q 3
q q=q t
0

2
where we have used the estimate S3 (q0 , h0 ) q0 3 + . Thus (5.16) follows from this by

letting q = d4 .

5.2 Proof of Theorem 3

From now on, we let

1 1 1
D = N 90 , P0 = N 2 − , P = N 19

Let c be any fixed positive number. For a given K ≥ 1, we define the major arcs by
q
KLc
[ [
M(K) = α ∈ U : |α − a/q| ≤
qN
q≤K a=1

where U = (P Lc /N, 1 + P0 Lc /N ]. We define M = M(P0 ), R = M(P ), and m = U\M.

For an n ∈ [N/2, N ], and a subset F of U, let

Z
RF (n) := h(α)g3 (α)g5 (α)e(−nα)dα.
F

For the above D, we let

Z
RF (D, n) := hD (α)g3 (α)g5 (α)e(−nα)dα,
F

and
Z
∗
R F (D, n) := VD (α)g3 (α)g5 (α)e(−nα)dα.
F

61
It is clear that

RU (n) = RM (D, n) + Rm (D, n) + Θ(D, n), (5.17)

where
Z X X
4 2
Θ(D, n) = µ(d) e(αd s ) g3 (α)g5 (α)e(−nα)dα.
U 1 N
D<d≤N 4 <s2 ≤ N4
10d4 d

We further split RM (D, n) into several parts as

RM (D, n) = R∗ R (D, n) + R∗ M\R (D, n) + ∆(D, n), (5.18)

where
Z
∆(D, n) = ∆(M, D, n) = (hD (α) − VD (α))g3 (α)g5 (α)e(−nα)dα.
M

In the next sections, we devote to proving the following results.

Theorem 6. Suppose B > 0 is any fixed constant. For any n ∈ [N/2, N ], we have

1
R∗ R (D, n) = SD (n, P )I(n) + O(N 30 L−2B ), (5.19)

1
where I(n) n 30 .

Theorem 7. Let Ψ(D, n) be any of Rm (D, n), Θ(D, n), R∗ M\R (D, n) and ∆(D, n). We

have
X 16 1
|Ψ(D, n)|2 N 15 − 45 +ε .
N/2<n≤N

Theorem 8. There are absolute constants δ, A > 0 such that

SD (n, P ) = (1 + o(N −δ ))S(n, P ),

1
and S(n, P ) L−A for all n ∈ [N/2, N ] with at most O(N 1− 40 ) exceptions.

1
Proof of Theorem 3: Theorems 6, 7 and 8 together yield that, for all but O(N 1− 45 +ε )

integers n ∈ [N/2, N ], we have

1
RU (n) N 30 L−A .

62
Note that RU (n) differs from Z(n), the number of representations of n as (5.1) weighted

by log p3 log p5 and subject to m2 , p3 3 , p5 5 ∈ [N/10, N ], by at most L2 times Z∗ (n), say, the

number of solutions to the following equations

n = m2 + p3 3s + p5 5t , s, t ∈ N and max{s, t} ≥ 2.

14
Clearly, Z∗ (n) 6= 0 for at most O(N 15 ) integers n ∈ [N/2, N ]. Thus we have shown that all
1
but O(N 1− 45 +ε ) integers n ∈ [N/2, N ] is expressed by (5.1). The theorem then follows by

a diadic interval argument.

5.3 Proof of Theorem 7

In this section, we shall prove Theorem 7. First of all, by (5.10) and Bessel’s inequality,

we have
X 16 1
|Θ(D, n)|2 L2 max H(K) N 15 − 45 +ε . (5.20)
1
N/2<n≤N D<K≤N 4

P0 Lc
If α ∈ m, then |α − a/q| ≤ qN for some integers a, q satisfying (q, a) = 1, and P0 < q ≤

N P0 −1 L−c . Thus, from (5.6), we have

1
max |GK (α)| N 4 + K.
α∈m

Hence
Z Z 1 12 Z 1 12
2 2 4 2 4
|hD (α)g3 (α)g5 (α)| dα L max |GK (α)| |g3 | dα |GK | |g5 | dα
KD
m α∈m 0 0

1 1 2 9 1 16 1
max N 4 +ε K · N 3 · (N 5 + N 20 K − 2 ) N 15 − 36 +ε .
KD

From this and Bessel’s inequality, we have proved

X 16 1
|Rm (D, n)|2 N 15 − 36 +ε . (5.21)
N/2<n≤N

63
The proof of Theorem 7 for Ψ(D, n) = ∆(D, n) is also straightforward. For convenience,

let V (K, α) := V2K (α) − VK (α). First, we have

X Z
2
|∆(D, n)| |(hD (α) − VD (α))g3 (α)g5 (α)|2 dα
N/2<n≤N M
Z
2
L max |(GK (α) − V (K, α))g3 g5 |2 dα
KD
ZM
L2 max |GK − V |(|GK | + |V |)|g3 g5 |2 dα. (5.22)
KD M

Then from Lemma 20, we have

Z Z 1 1 Z 1 1
1 2 2
2 +ε 2 4 4
|GK − V ||GK ||g3 g5 | dα N 4 K |GK | |g5 | |g3 |
M 0 0
1 2 9
− 21 1 16 1
N 4
+ε
K(N + N
5 20 K )N N 15 − 36 +ε ,
3

and
Z
1
Z 1 16 1
|GK (α) − V (K, α)||V (K, α)||g3 g5 | dα N 2 2
+ε
D |g3 g5 |2 dα N 15 − 45 +ε
M 0

which, along with (5.22), give the desired bound for the second moment of ∆(D, n). It is

now clear that, to prove Theorem 7, it is more than enough to prove the following lemma.

Lemma 22. For any K satisfying P < K ≤ 12 P0 , we have

Z
16 1
|VD (α)g3 (α)g5 (α)|2 dα N 15 − 30 +ε .
M(2K)\M(K)

We shall apply some pruning techniques to prove Lemma 22. We first show a result

that will be frequently referred to in our pruning procedure. This is essentially a result (in

slightly different forms) that has been used in several previous papers (e.g., Lemma 2.6 of

[91]). We give it an easier form (yet with a much shorter proof) here.

Lemma 23. Suppose the function

X
Φ(α) = η(h)e(αh)
|h|≤H

64
satisfies Φ(α) ≥ 0, η(h) ≥ 0 for all α, h and H N γ for some fixed γ > 0. Then, for any
1
Y N 2 L−c , we have
Z X
−1+
Φ(α) dα N Y η(0)Y + η(h) .
M(Y ) |h|≤H
h6=0

Proof. Since Φ(α) is non-negative, we have

Z XX q Z Y Lc /qN
Φ(α) dα Φ(a/q + β)dβ
M(Y ) q≤Y a=1 −Y Lc /qN
q Z Y Lc /qN
X X X ah
= η(h) e e(hβ)dβ.
q −Y Lc /qN
q≤Y |h|≤H a=1

Note that the sum over a is equal to q if q | h, and 0 otherwise, and the last integral is
c
trivially bounded by YqN
L
. Thus, we have
Z X X X
Φ(α) dα N −1 Y Lc 1 N −1+ Y η(0)Y +

η(h) η(h) .
M(Y ) |h|≤H q≤Y |h|≤H
q|h
h6=0

Proof of Lemma 22. We prove the lemma in accordance with the size of K. First of all,

we note that for s ∈ N, Lemma 23 gives

Z Z 1 2s

2s −1+ε 2s
|gk (α)| dα N K K |gk (α)| dα + N k . (5.23)
M(2K) 0

From (5.9), we also have

1 1
VD (α) N 2 +ε K − 2 (5.24)

if α ∈ M(2K)\M(K).

2
• N 5 K P0

By (5.23) and (5.24), we have

!1/2 Z 1/2
Z Z 1
2 1+ε −1
|VD g3 g5 | dα N K |g54 | dα 4
|g3 | dα
M(2K)\M(K) M(2K) 0
2 4 1 1
N 1+ε K −1 N −1+ε K KN 5 + N 5 2
·N3
31 37 1 16 1
N 30 +ε + N 30 +ε K − 2 N 15 − 30 +ε .

65
1 2
• N 5 K N 5.

By (5.23), (5.24), and Hua’s inequality, the integral is bounded by

!1 ! 1 Z 16
Z 3 Z 2 1
1+ε −1
N K |g32 | dα |g54 | dα |g3 | dα8
M(2K) M(2K) 0

1 2 1 2 4 1 5
N 1+ε K −1 N −1+ε K KN 3 + N 3 3
· N −1+ε K KN 5 + N 5 2
· N 18
16 1 1 1 16 1
N 15 +ε N − 9 K 6 + K − 6 N 15 − 30 +ε .

1
• P K N 5.

The Theorem 2 of Kumchev [54] implies that, if 1 ≤ q ≤ Q, (a, q) = 1, |qα − a| < QN −1

1
with Q ≤ N k , then
1
X
k 1 11
+ε N k +ε
e(αp ) Q N 2 20k + 1 . (5.25)
1 (q + N |qα − a|) 2
p∼N k

1
Thus, if α ∈ M(2K)\M(K) with K N 5 , then by partial summation we have

11 1 1 1
g3 (α) N 60 +ε K 2 + N 3 +ε K − 2 .

Thus by this and (5.23), (5.24), we get

N 1+ε
Z Z
11 2
2 +ε +ε −1
|g5 |2 dα

|VD g3 g5 | dα · N 30 K + N 3 K
M(2K)\M(K) K M(2K)
41 5 1 2
N 30 +ε + N 3 +ε K −2 N −1+ε K(KN 5 + N 5 )


16 23 16 1
N 15 +ε K −1 + N 30 +ε K N 15 − 30 +ε .

We thus have finished the proof of Lemma 22.

66
5.4 Proof of Theorem 6
a
If α = q + β with q ∈ N, a ∈ Z, and (a, q) = 1, then
q
ahk
X X X
gk (α) = e Λ(m)e(βmk ) + O Λ(m)
h=1
q 1 1
1
(h,q)=1 N k <m≤N k m≤N k
10 (m,q)>1
m≡h( mod q)

1 X X
= Ck (χ, a) Λ(m)χ(m)e(βmk ) + O(log N log q)
φ(q) k1
χ( mod q) N
1
10
<m≤N k

Ck (χ0 , a) 1 X
= fk (β) + Ck (χ, a)Wk (β, χ) + O(L2 ), (5.26)
φ(q) φ(q)
χ( mod q)

where
X
Wk (β, χ) = (Λ(m)χ(m) − δχ )e(mk β),
N
10
<mk ≤N

with δχ = 1 if χ = χ0 , and δχ = 0 if χ 6= χ0 .

We state two lemmas that will be used in the proof of Theorem 6.

N
Lemma 24. Suppose δ ∈ (0, 1) is a fixed constant, Q = HLc for some constant c > 0,
5
−
1≤R≤HN 12(2−δ)k for some constant > 0. Let
X X ∗
Jk = Jk (δ, R, Q) := r−δ max |Wk (β, χ)|.
|β|<(RQ)−1
r∼R χ( mod r)

Then for any fixed B > 0, we have

1
Jk N k L−B . (5.27)

Proof: This lemma is more general than Lemma 4.3 of [62], and it follows from the same

argument.

Lemma 25. Let χj (modrj ), j = 1, 2 be primitive characters, r0 = [r1 , r2 ], and χ0 is the

principal character modulo q. Then for any x N , we have

X |BD (n, q, χ0 χ1 , χ0 χ2 )| 1
r0 − 2 + LB , (5.28)
q≤x
q(φ(q))2
r0 |q

for some absolute constant B and any > 0.

67
Proof: Similar to Lemma 2.1 of [60], this lemma follows from the first estimate in (5.38)

(section 7) and the fact that BD (n, q, χ0 χ1 , χ0 χ2 ) is multiplicative.

a
For α = q + β ∈ R(q, a), (5.2) and (5.26) give

Ck (χ0 , a) 1 X
gk (α) = vk (β) + Ck (χ, q, a)Wk (β, χ) + O(P Lc ).
φ(q) φ(q)
χ( mod q)

Hence we have
Z
1
VD (α)g3 (α)g5 (α)e(−nα) = I0 + I3 + I5 + I35 + O N − 6 + P 3 ,

(5.29)
R

where the terms I0 , I3 , I5 , and I35 are given by

c
X BD (n, q) Z PqN
L

I0 = v2 (β)v3 (β)v5 (β)e(−βn)dβ,

q(φ(q))2 − P Lc
q≤P qN

Z P Lc
X X BD (n, q, χ, χ0 ) qN
I3 = v2 (β)W3 (β, χ)v5 (β)e(−βn)dβ,
q(φ(q))2 − PqN
L c
q≤P χ( mod q)
Z P Lc
X X BD (n, q, χ0 , χ) qN
I5 = v2 (β)v3 (β)W5 (β, χ)e(−βn)dβ,
q(φ(q))2 − PqN
L c
q≤P χ( mod q)
Z P Lc
X X BD (n, q, χ1 , χ2 ) qN
I35 = v2 (β)W3 (β, χ1 )W5 (β, χ2 )e(−βn)dβ,
q(φ(q))2 − PqN
L c
q≤P χ1 ,χ2 ( mod q)
(5.30)

1
and the error term O N − 6 + P 3 in (5.29) comes from the contribution of the other terms

involving the O(P Lc ) which is trivially

P2
Z
1 1 1
PL c
|V2 |(|g3 | + |g5 | + P Lc )dα P Lc N 2 + 3 · N − 6 P 3 Lc .
R N

We note that
1
Nk
vk (β) .
1 + |β|N

Thus,
Z P Lc Z P Lc 1
N N N2 1
|v2 (β)|dβ dβ N − 2 L, (5.31)
− PN
L c
− PN
L c 1 + |β|N

68
and for k = 3, 5, we have
Z P Lc P Lc 1 1
N 2+k
Z
N N 1 1
|v2 (β)vk (β)|dβ 2
dβ N k − 2 . (5.32)
c
− PN
L
− PN
L c (1 + |β|N )

Therefore, from Lemmas 25 and 24, and (5.32) we get

Z PL c
X X ∗ X |BD (n, q, χχ0 , χ0 )| rN
I3 max |W 3 (β, χχ 0 )| |v2 (β)v5 (β)|dβ
q(φ(q))2 Lc
|β|< PrN −PL
c
r≤P χ( mod r) q≤P rN
r|q

3 X X ∗ 1
N − 10 LB r − 2 +
L + max c |W3 (β, χ)|
r≤P χ( mod r) |β|< PrN
L

3 3 1 1
N − 10 LB+1 P 2 + + N 30 L−C N 30 L−C . (5.33)

I5 can be bounded in exactly the same way. Similarly, we have

X X ∗ X X ∗ X |BD (n, q, χ1 χ0 , χ2 χ0 )|
I35 ×
q(φ(q))2
r1 ≤P χ1 ( mod r1 ) r2 ≤P χ2 ( mod r2 ) q≤P
[r1 ,r2 ]|q
Z P Lc /N
× max |W3 (β, χ1 χ0 )||W5 (β, χ2 χ0 )| |v2 (β)|dβ
|β|<P Lc /N −P Lc /N
LB X X ∗ ∗ 1
[r1 , r2 ]− 2 + ×
X X
1
N 2
r1 ≤P χ1 (r1 ) r2 ≤P χ2 (r2 )

× L + max |W3 (β, χ1 )| L + max |W5 (β, χ2 )|
β β
B
XX
L Y ∗ −δ +
1 r k
L + max |Wk (β, χ)|
N 2 k=3,5 r≤P χ(r) β

LB Y 1 1
P 2 + N k L−C N 30 L−C ,

1 (5.34)
N 2
k=3,5
1 5
where δ3 = 12 and δ5 = 12 so that the condition in Lemma 24 is satisfied.

To get an asymptotic formula for I0 , we first note that

Z 1 Z 1 31
2 2 N 30 1
|v2 (β)v3 (β)v5 (β)|dβ 3
dβ N 30 L−2c .
PL c P L (1 + |β|N )
c
qN qN

We also note that

Z 1
2
I(n) : = v2 (β)v3 (β)v5 (β)e(−βn)dβ
− 21
1 X 1 1
= 1 2 4 N 30 . (5.35)
30 N m 12 m 23 m
35
10 <m1 ,m2 ,m3 ≤N
m1 +m2 +m3 =n

69
We thus have proved (5.19) from (5.29)-(5.35).

5.5 Singular Series

We prove Theorem 8 in this section. We first quote a result from [61].

Lemma 26. Let a, α ∈ Z, p be a prime, p - a and α ≥ 0. χ (mod pα ) is any character,

and χ0 is the principal character modulo pt . Then Ck (χχ0 , a) = 0 if t ≥ θp + max{θp , α},

where θp = 1 + ordp (k).

Proof. This is part of Lemma 3.1 of [61].

Lemma 26 shows that BD (n, q) 6= 0 only when, apart from a possible divisor 4, q is square-

free. Define

1 X ∗
Ad (q) = Ad (q; n) := S2 (q, ad4 )C3 (q, a)C5 (q, a)e(−an/q). (5.36)
q(ϕ(q))2
a( mod q)

It is clear that Ad (q) is a multiplicative function of q.

Lemma 27. For every integer d and prime p > 5, we have


6
if p 6≡ 1 (mod 30),


p




|Ad (p)| < 16
if p ≡ 1 (mod 30) and p - d,
 p


10

 √ if p ≡ 1 (mod 30) and p | d.

p

Proof: We prove the two cases separately.

Case 1: p 6≡ 1 (mod 30). First note that

p
1 X
Ad (p) = S2 (p, ad4 )C3 (p, a)C5 (p, a)e(−an/p) − 1
p(p − 1)2
a=1
p X
p−1 X
p−1
1 X
= 1 − 1.
(p − 1)2
x=1 y=1 z=1
d4 x2 +y 3 +z 5 ≡n (mod p)

70
For {k1 , k2 } = {3, 5}, if k1 - (p − 1) (no matter k1 = 3 or 5), then y k1 ≡ n − d4 x2 − z k2

(mod p) has exactly one solution for any x, z. Since


XXp p−1  p − 1 + (k2 − 1)θ√p

if p - d
1 = (5.37)
c(k2 )p if p | d

x=1 z=1 
d4 x2 +z k2 ≡n (mod p)

where the first equality, with |θ| < 1, is from Weil’s bound, and c(k2 ) ∈ {0, 1, k2 }. Then

from
p−1
p X
p(p − 1) 1 X
Ad (p) = 2
−1− 1 1
(p − 1) (p − 1)2
x=1 z=1
d4 x2 +z k2 ≡n (mod p)

we have the result in this case. Case 2: p ≡ 1 (mod 30). We first consider the case p - d

(then S2 (p, ad4 ) = S2 (p, a)). We write Sk for Sk (p, a). Note that
p−1
X S2 (S3 − 1)(S5 − 1)
Ad (p) = e(−an/p)
p(p − 1)2
a=1
p−1
1 X
= (S2 S3 S5 − S2 S3 − S2 S5 + S2 )e(−an/p).
p(p − 1)2
a=1
√
Trivially, by Weil’s bound |Sk | < (k − 1) p, we have

p−1
1 X
(|S2 S3 | + |S2 S5 | + |S2 |)
p(p − 1)2
a=1
√
1 √ 6p + p 7
≤ 2
(p − 1)(2p + 4p + p) ≤ <
p(p − 1) p(p − 1) p

Now, we are left with

p−1
1 X
S2 S3 S5 (p, a)e(−an/p).
p(p − 1)2
a=1

Let χ (mod p) have order 30. Then from

X p
X
Sk (p, a) = χ(a)τ (χ), where τ (χ) = χ(m)e(m/p),
χk =χ0 m=1
χ6=χ0

71
we have,

S2 (p, a) = χ15 (a)τ (χ15 ),

S3 (p, a) = χ10 (a)τ (χ10 ) + χ20 (a)τ (χ20 ),

S5 (p, a) = χ6 (a)τ (χ6 ) + χ12 (a)τ (χ12 ) + χ18 (a)τ (χ18 ) + χ24 (a)τ (χ24 ).

Thus,
p−1
1 X
S2 S3 S5 e(−an/p)
p(p − 1)2
a=1
 
2 X 4 p−1
1 15 
X
10i 6j 
X
= τ (χ ) τ (χ )τ (χ ) χ15+10i+6j (a)e(−an/p)
p(p − 1)2
i=1 j=1 a=1
 
2 4
τ (χ15 ) X X
= τ (χ10i )τ (χ6j ) χ15+10i+6j (−n)τ (χ15+10i+6j ).
p(p − 1)2
i=1 j=1

√
If 30 - j, then the Gauss sum τ (χj ) satisfies |τ (χj )| ≤ p. So we have,
p−1

1 X 8p2 9
S S S e(−an/p) ≤ <

2 2 3 5 2
p(p − 1) p(p − 1) p

a=1

Therefore, in the case p ≡ 1 (mod 30) and p - d, we have

16
|Ad (p)| < .
p

If p | d, then we have
p−1 p−1
X (S3 − 1)(S5 − 1) 1 X
Ad (p) = e(−an/p) = (S3 S5 − S3 − S5 + 1)e(−an/p).
(p − 1)2 (p − 1)2
a=1 a=1

Note
p−1
X p
X
(−S3 − S5 + 1)e(−an/p) = (−S3 − S5 + 1)e(−an/p) + 2p − 1
a=1 a=1
= cp − 1 − c3 p − c5 p

72
where c = 2 or 3 depending on whether p | n, and ck (k = 3, 5) is the number of solutions

of xk ≡ n(modp). So we have
p−1
X
(−S3 − S5 + 1)e(−an/p) ∈ [−6p − 1, 2p − 1],
a=1

which, along with

 
p−1 2 4 p−1
X XX 10i 6j
X

S3 S5 e(−an/p)=  τ (χ )τ (χ ) χ10i+6j (a)e(−an/p)
a=1 i=1 j=1 a=1
 
2 4
XX 3
10i 6j  10i+6j 10i+6j

= 
τ (χ )τ (χ ) χ (−n)τ (χ )< 8p 2 ,
i=1 j=1

yields the result for the case p | d.

From Lemma 27, (5.3) and (5.4), it is easily seen that

BD (n, q) q 2+ , and BD (n, q) = B∞ (n, q) + O(q 2 D−1+ ), (5.38)

which yield
1
SD (n, P ) = (1 + O(N − 90 + ))S(n, P ). (5.39)

We note that, by (5.3), (5.4), and a similar argument as Lemma 7 of [1], we have

6 X
S(n, P ) = B(q; n), (5.40)
π2
q≤P

where
Y 1 −1 X µ(d)

B(q; n) := 1− 2 Ad (q; n) (5.41)
p d2
p|q d|q

is clearly multiplicative and, by Lemmas 26 and 27,

B(pk ; n) = 0 if k > 1 for p > 2, or k > 2 for p = 2, (5.42)

and, for p > 5,‘ 

6
if p 6≡ 1 (mod 30)


p
|B(p; n)| < (5.43)
18
if p ≡ 1 (mod 30).


p

73
For p > 5, from the estimates in this section we get

1 −1

Ap (p; n)
B(p; n) = 1− 2 A1 (p; n) −
p p2
p−1
1 X 1 an
= S2 (p, a) − (S3 (p, a) − 1)(S 5 (p, a) − 1)e −
p(p − 1)2 p p
a=1
= : B1 (p; n) + B2 (p; n), (5.44)

where
p−1
1 X an c0
B1 (p; n) = 3
S2 S3 S5 e − , and |B2 (p; a)| < 3 , (5.45)
p p p2
a=1

for some absolute constant c0 > 0. We also have |B1 (p; n)| < 8p−1 from Lemma 8.1 of [85].

Thus, for every square-free q prime to 30, we can write

Y X
B(q; n) = (B1 (p; n) + B2 (p; n)) = B1 (r; n)B2 (s; n), say. (5.46)
p|q q=rs

Let D be the set of integers each of which has no prime factor exceeding N and, apart

from a possible factor 4, is square-free. Let C be the set of square-free integers consisting of

prime factors p ∈ [7, N ]. Then every integer q ∈ D can be uniquely written as q = rs, with

r | 60 and s ∈ C.

1
Lemma 28. Suppose δ ∈ 0, 14 is fixed. V = exp(L1+2δ ). Then, for all but O(N 1− 40 )

values of n ∈ [N/2, N ], we have

X
B(q; n) N −ε ,
P <q≤V
q∈D

where the -constant depends on ε only.

Proof: We first note that

X X X
B(q; n) = B(t; n) B(h; n).
P <q≤V t|60 P t−1 <h≤V t−1
q∈D h∈C

74
Hence, to prove the lemma, it suffices to show, for every t | 60

X
B(h; n) N −ε
P/t<h≤V /t
h∈C

1
for all but O(N 1− 40 −ε ) values of n ∈ [N/2, N ]. On the other hand, from (5.46), we have

X X
B(h; n) = B1 (r; n)B2 (s; n). (5.47)
P/t<h≤V /t P/t<rs≤V /t
h∈C rs∈C

ω(s) − 3
Since B1 (r; n) 8ω(r) r−1 and B2 (s; n) c0 s 2 , the terms in (5.47) with s > N ε

contribute at most

ε X (8 + c0 )ω(rs) ε X (16 + 2c0 )

ω(q) ε ε
N−2 N−2 N − 2 (log V )16+2c0 N − 3 .
rs q
P/t<rs≤V /t q≤V
rs∈C

Let r ≤ N be any fixed number, U ∈ [P N −ε , P ], V = exp(L1+δ ) for some fixed δ > 0. It

is now clear from the above argument that, to prove the lemma, it suffices to show

X X 39
S(U, V ) = S(U, V ; C, r, n) :=
B1 (q; n) N 40 . (5.48)
N/2<n≤N U <q≤V
q∈C,(q,r)=1

Now, exactly following the argument of Vaughan [85], pp. 141 − 143, we have

1 1 X l−1 l4 −1
S(U, V ) N U − 2 r 2 L4 + N L8 N− 4l (l(L + 2)) 16l

1<l≤ 2 log
L
V

1 7 1
N 1− 38 + + N 8 + N 1− 40 ,

which proves the lemma.

Similar to Vaughan [85], p144, we let γ = L−1 . Then

X X
|B(q; n)| ≤ (q/V )γ |B(q; n)|
q>V q∈D
q∈D
Y
−γ γ
= V (1 + B(2; n) + B(4; n)) 1 + p |B(p; n)|
2<p≤N

75
which, by (5.43), is bounded by

Y 18e

2δ
exp(−L ) 1+ exp(−Lδ ).
p
p≤N

From this, Lemma 28, and (5.40), we have

6 Y
S(n; P ) = (1 + B(2; n) + B(4; n)) (1 + B(p; n)) + O(exp(−Lδ )) (5.49)
π2
2<p≤P

1
for all but O(N 1− 40 ) values of n ∈ [N/2, N ]. From (5.43), we see that

Y Y 18
(1 + B(p; n)) 1− L−18 .
p
5<p≤P 17<p≤P

Thus, to finish the proof of Theorem 8, it suffices to show that

(1 + B(2; n) + B(4; n))(1 + B(3; n))(1 + B(5; n)) > 0. (5.50)

(−1)n+1
A direct calculation based on (5.41) and (5.36) gives B(2; n) = 3 , B(4; n) = 31 , − 16 , − 31 , 16
1
when n ≡ 1, 2, 3, 4(mod4), respectively, and B(3; n) = √
2 3
sin 2nπ
3 . In any case, we have

(1 + B(2; n) + B(4; n))(1 + B(3; n)) > 0.

From the proof of Lemma 27, we have

5 X
4 X
4
1 X
Ad (5; n) = 1−1
16
x=1 y=1 z=1
d4 x2 +y 3 +z 5 ≡n (mod 5)
5 X
4
20 1 X
= −1− 1
16 16
x=1 z=1
d4 x2 +z 5 ≡n (mod 5)
1 5 1
= − + #{x(mod5) : d4 x2 ≡ n(mod5)}.
4 16 16

1
This gives |A1 (5; n)| ≤ 16 , and |A5 (5; n)| ≤ 41 which together yield |B(5; n)| ≤ 25
24 |A1 (5; n)|+
1

52
|A5 (5; n)| < 1. So we have 1 + B(5; n) > 0 as well, and thus (5.50) is proved.

76
BIBLIOGRAPHY

[1] R. C. Baker and J. Brüdern. Sums of cubes of square-free numbers. Monatsh. Math.,
111:1–21, 1991.

[2] R.C. Baker and J. Brüdern. Sums of cubes of square-free numbers ii. Monatsh. Math.,
112:177–207, 1991.

[3] R.C. Baker, G. Harman, and J. Rivat. Primes of the form [nc ]. Journal of Number
Theory, 50:261–277, 1995.

[4] A. Balog and J. Friedlander. A hybrid theorem of Vinogradov and Piatetski-Shapiro.

Pacific. J. Math., 156:45–62, 1992.

[5] C. Bauer. On a problem of the Goldbach-Waring type. Acta Math. Sinica, New Series,
14:223–234, 1998.

[6] C. Bauer. An improvement on a theorem of the Goldbach-Waring type. Rocky Mount.

J. Math., 31:1151–1170, 2001.

[7] C. Bauer. A remark on a theorem of the Goldbach-Waring type. Studia Sci. Math.
Hungar., 41:309–324, 2004.

[8] C. Bauer. A Goldbach-Waring problem for unequal powers of primes. Rocky Mount.
J. Math., 38(4):1073–1090, 2008.

[9] E. Bombieri and H. Iwaniec. On the order of ζ( 12 + it). Ann. Scuola Norm Sup. Pisa,
13:449–472, 1986.

[10] J. Brüdern. Iterationsmethoden in der additiven Zahlentheorie. PhD thesis, Göttingen,

1988.

[11] J. Brüdern. On Waring’s problem for cubes and biquadrates. J. London Math. Soc.,
37:25–42, 1988.

[12] J. Brüdern. A problem in additive number theory. Math. Proc. Camb. Phil. Soc.,
103:27–33, 1988.

[13] J. Brüdern. Ternary additive problems of Waring’s type. Math. Scand., 68:37–45,
1991.

[14] J. Brüdern. Ternary additive problems of Waring’s type. Math. Scand., 68:37–45,
1991.

[15] J. Brüdern and T. Wooley. On Waring’s problem: a square, four cubes and a bi-
quadrate. Math. Proc. Cambridge Philos. Soc., 127:193–200, 1999.

77
[16] J. Brüdern and T. Wooley. On waring’s problem: two cubes and seven biquadrates.
Tsukuba J. Math., 24:387–417, 2000.

[17] J. Brüdern and T. Wooley. On Waring’s problem: three cubes and a sixth power.
Nagoya Math. J., 163:13–53, 2001.

[18] R. Brünner, A. Perelli, and J. Pintz. The exceptional set for the sum of a prime and
a square. Acta Math Hung., 53:347–365, 1989.

[19] J. R. Chen. On the representation of large even integer as the sum of a prime and the
product of at most two primes. Sci Sinica., 16:157–176, 1973.

[20] J. R. Chen and C. D. Pan. The exceptional set of Goldbach numbers. Sci. Sinica,
23:416–430, 1980.

[21] S-F Chiu and M-C Liu. On exceptional sets for numbers representable by binary sums.
Rocky Mountain Journal of Math., 26:959–986, 1996.

[22] N. G. Chudakov. On the density of the sets of even integers which are not representable
as a sum of two odd primes. Izv. Akad. Nauk SSSR Ser. Mat., 2:25–40, 1938.

[23] R. Cook. On sums of powers of integers. J. Number Theory, 4:516–528, 1979.

[24] H. Davenport. On Waring’s problem for cubes. Acta Math., 71:123–143, 1939.

[25] H. Davenport. On Waring’s problem for fourth powers. Ann. Math., 40:731–747, 1939.

[26] H. Davenport and H. Heilbronn. Note on a result in the additive theory of numbers.
Math Annalen, 109:142–151, 1936.

[27] H. Davenport and H. Heilbronn. Note on a result in the additive theory of numbers.
Proc. London Math. Soc., 43:142–151, 1937.

[28] H. Davenport and H. Heilbronn. On Waring’s problem: two cubes and one square.
Proc. London Math. Soc., 43:73–104, 1937.

[29] T. Estermann. On Goldbach’s problem: Proof that almost all even positive integers
are sums of two primes. Proc. London Math. Soc., 44:307–314, 1938.

[30] K. Ford. The representation of numbers as sums of unlike powers. J. London Math.
Soc., 51:14–26, 1995.

[31] K. Ford. The representation of numbers as sums of unlike powers. J. Amer. Math.
Soc., 9:919–940, 1996.

[32] E. Fouvry and H. Iwaniec. Exponential sums with monomials. Journal of Number
Theory, 33:311–333, 1989.

[33] C.F. Gauss. Disquisitiones arithmeticae. PhD thesis, Leipzig, 1801.

[34] A. Ghosh. The distribution of αp2 modulo 1. Proc. London Math. Soc., 42:252–269,
1981.

78
[35] G. H. Hardy. Goldbach’s theorem. Math. Tid., B:1–16, 1922.

[36] G. Harman. Trigonometric sums over primes. I, Mathematika, 28:249–254, 1981.

[37] D.R. Heath-Brown. The Pjateckii-Sapiro prime number theorem. J. Number Theory,
16:242–266, 1983.

[38] H. Helfgott. Major arcs for Goldbach’s problem. arxiv:1305.2897, 2013.

[39] H. Helfgott. Minor arcs for Goldbach’s problem. arXiv:1205.5252, 2013.

[40] D. Hilbert. Beweis für darstellvrkeit der ganzen zahlen durch eine feste anzahl nter
potenzen (waringsche problem). nachrichten von der Königlichen geselllchaft der wis-
senschaften zu Göttingen, mathematischphysikalische klasse aus den jaren. Math. An-
nalen, 67:281–300, 1909.

[41] C. Hooley. On the representation of a number as the sum of two squares and a prime.
Acta. Math., 97:189–210, 1957.

[42] C. Hooley. On Waring’s problem for two squares and three cubes. J. Reine Angew
Math., 328:161–207, 1981.

[43] C. Hooley. Recent progress in analytic number theory, Vol 1, chapter On a new ap-
proach to various problems of Waring’s type, pages 127–191. Academic Press, 1981.

[44] L. K. Hua. Additive theory of prime numbers. American Math. Soc., Providence,
Rhode Island, 1965.

[45] A. Ivic. The Riemann zeta function: Theory and applications. Dover Publications,
2003.

[46] J. M. Deshouillers, G. Effinger, H. te Riele, and D. Zinoviev. A complete Vinogradov

3-primes theorem under the Riemann hypothesis. Amer. Math. Soc., 3:99–104, 1997.

[47] C.-H Jia. On Piatetski-Shapiro prime number theorem ii. Science in China Ser. A,
35:913–922, 1993.

[48] C.-H. Jia. On Piatetski-Shapiro prime number theorem. Chinese Ann. Math, 15B:1:9–
22, 1994.

[49] K. Kawada and T. Wooley. Sums of fourth powers and related topics. J. Reine Angew.
Math., 512:173–223, 1999.

[50] K. Kawada and T.D. Wooley. On the Waring-Goldbach problem for fourth and fifth
powers. Proc. London Math. Soc., 83:1–50, 2001.

[51] G.A. Kolesnik. The distribution of primes in sequences of the form [nc ]. Mat. Zametki,
2:117–128, 1972.

[52] G.A. Kolesnik. Primes of the form [nc ]. Pacific J. Math, 118:437–447, 1985.

[53] A. Kumchev. On the distribution of prime numbers of the form [nc ]. Glasgow Math.
J., 41:85–102, 1999.

79
[54] A. Kumchev. On Weyl sums over primes and almost primes. Michigan Math. J.,
54:243–268, 2006.

[55] D. Leitmann. Abschätzung trigonometrischer summen. J. Reine Angew. Math.,

317:209–219, 1980.

[56] J. Linnik. On the representation of large numbers as a sum of seven cubes. Mat.
Sbornik, 12:218–224, 1942.

[57] Ju. V. Linnik. On the representation of large numbers as sums of seven cubes. Mat.
Sb., 12:218–224, 1943.

[58] V. Linnik. An asymptotic formula in the Hardy-Littlewood additive problem. Izv.

Akad. Nauk SSSR Ser Mat., 24:629–706, 1960.

[59] H.Q. Liu and J. Rivat. On the Piatetski-Shapiro prime number theorem. Bull. London
Math. Soc., 24:143–147, 1992.

[60] J. Liu and T. Zhan. Distribution of integers that are sums of three squares of primes.
Acta Arith., 98(3):207–228, 2001.

[61] M-C Liu and K-M Tsang. Small prime solutions of some additive equations. Monatsh.
Math., 111:147–169, 1991.

[62] Z. Liu. An improvement on Waring-Goldbach problem for unlike powers. Acta Math.
Hungar., 130:118–139, 2011.

[63] M. Lu and Z. Shan. A problem of Waring-Goldbach type (in chinese). J. China Univ.
Sci. Tech., pages 1–8, 1982.

[64] M.G. Lu. On Waring’s problem for cubes and fifth power. Sci. China Ser. A, 36:641–
662, 1993.

[65] K. Mahler. On the fractional parts of the powers of a rational number ii. Michigan
Math., 35:122–124, 1957.

[66] R. J. Meich. On the equation n = p + x2 . Trans. Amer. Math. Soc., 130:494–512, 1968.

[67] H. L. Montgomery and R. C. Vaughan. The exeptional set in Goldbach’s problem.

Acta. Arith., 27:353–370, 1975.

[68] I. Piatetski-Shapiro. On the distribution of prime number in the sequence of the form
[f(n)]. Mat. Sb., 33:559–566, 1953.

[69] J. Pintz. Explicit formulas and the exceptional set in Goldbach’s problem. lecture at
the CNTA VIII meeting in Toronto, 2004.

[70] V.A. Plaksin. On a question of Hua Loo-Keng. Math. Notes, 47:278–286, 1990.

[71] I.V. Polyakov. On the exceptional set for the sum of a prime and a perfect square.
Math USSR Izvestija, 19:1982, 611-641.

80
[72] K. Prachar. Über ein problem vom Waring-Goldbach’schen type. Monatsh Math,
57:66–74, 1953.

[73] X. Ren and K. Tsang. Waring-Goldbach problem for unlike powers. Acta Math. Sinica,
English Series(2), 23:265–280, 2007.

[74] X. Ren and K. Tsang. Waring-Goldbach problem for unlike powers (ii). Acta Math.
Sinica, Chinese Series, 50:175–182, 2007.

[75] J. Rivat. Autour d’un theorem de Piatetski-Shapiro. PhD thesis, Universite de Paris
Sud, 1992.

[76] J. Rivat and P. Sargos. Nombres premiers de la forme [nc ]. Canad. J. Math, 53:414–
433, 2001.

[77] J. Rivat and J. Wu. Prime numbers of the form [nc ]. Glasg. Math. J., 43:237–254,
2001.

[78] K. F. Roth. Proof taht almost all positive integers are sums of a square, a positive
cube and a fourth power. J. London Math. Soc., 24:4–13, 1949.

[79] K. F. Roth. A problem in additive number theory. Proc. London Math. Soc., 53:381–
395, 1951.

[80] K. F. Roth. A problem in additive number theory. Proc. London Math. Soc., 53:381–
395, 1951.

[81] K. Thanigasalam. On additive number theory. Acta Arith., 13:237–258, 1967.

[82] J. G. van der Corput. Sur l’hypothese de Goldbach. Proc. Akad. Wet. Amsterdam,
41:76–80, 1938.

[83] R. C. Vaughan. On the representation of numbers as sums of powers of natural num-

bers. Proc. London Math. Soc., 21:160–180, 1970.

[84] R. C. Vaughan. A ternary additive problem. Proc. Lond. Math. Soc., 41:516–532,
1980.

[85] R. C. Vaughan. The Hardy-Littlewood Method 2nd edition. Cambridge University

Press, 1997.

[86] R.C. Vaughan. On Waring’s problem: one square and five cubes. Quart. J. Math.
Oxford, 37:117–127, 1986.

[87] R.C. Vaughan. A new iterative method in Waring’s problem. Acta Math., 162:1–71,
1989.

[88] A.I. Vinogradov. On a binary problem of Hardy-Littlewood. Acta Arith., 46:33–56,

1985.

[89] I. M. Vinogradov. Representation of an odd number as a sum of three primes. C. R.

Acad. Sci. URSS, 15:6–7, 1937.

81
[90] I. M. Vinogradov. Some theorems concerning the theory of primes. Mat. Sbornik N.S.,
2:179–195, 1937.

[91] Y-L Lau, M-C Leung, and M-C Liu. chapter On ternary equations in square-free and
prime variables. World Sci. Publ.

[92] A. Zaccagnini. On the exceptional set for the sum of a prime and a kth power. Math-
ematika, 39:400–421, 1992.

[93] W. Zhai. On the Waring-Goldbach problem in thin sets of primes. Acta. Math. Sinica.,
41:595–608, 1998.

82