ON PRIMES REPRESENTED BY QUARTIC POLYNOMIALS

ON AVERAGE
arXiv:1908.09439v1 [math.NT] 26 Aug 2019

KAM HUNG YAU

Abstract. We obtain an upper bound for the distribution of

primes in the form n4 + k up to x, averaged over k with small
square-full part. As a corollary, we show that for almost all k ,
there is an expected amount of primes in the form n4 + k up to x.

1. Introduction
Dirichlet proved that any linear polynomial with coprime coefficients
take prime values infinitely often. The case of showing a polynomial
of degree two or more representing prime values infinitely often is still
open, although there are stunning partial results due to Kuhn [18]
and later by Iwaniec [16]. Kuhn’s result [18] can be extended to show
that any irreducible polynomial an2 + bn + c with a > 0 and c is a
product of at most three primes for infinitely many n. Iwaniec [16]
improved Kuhn’s result [18] for the particular polynomial n2 + 1 by
showing n2 + 1 is a product of at most two primes for infinitely many
n. For polynomials of two variables, Friedlander & Iwaniec [7] proved a
spectacular theorem stating that the polynomial x2 + y 4 attain prime
values infinitely often. Later Heath-Brown [12] and Heath-Brown &
Li [13] showed the polynomials x3 + 2y 3 and a2 + p4 (p prime) takes
prime values infinitely often respectively.
A conjecture of Bateman-Horn [4] says that if f is an irreducible
polynomial in Z[x] satisfying gcd{f (n) : n ∈ Z} = 1 then
X Y np − 1

Λ (f (n)) ∼ 1− x,
n≤x p
p − 1

where np is the number of solutions to the congruence

f (n) ≡ 0 mod p, n ∈ Z/pZ.

Date: August 27, 2019.

2010 Mathematics Subject Classification. 11P55, 11L07, 11N13, 11N32.
Key words and phrases. Circle method, Primes in quartic progressions, Large
sieve for quartic characters.
1
2 KAM HUNG YAU

Here Λ is the von-Mongoldt function. See [2] for a discussion on this

topic. We note that in the special case of f (n) = nr + k , we can write
gcd(p−1,r)−1
X
np = χj (−k),
j=0

where χ is a multiplicative character of order gcd(p − 1, r). The rep-

resentation of np for this specific f is crucial for the following series of
results.
Baier & Zhao [1] showed that for fixed A, B > 0, and xr (log)−A ≤
y ≤ xr , we have
  2
X X Y np − 1 yx2
Λ(nr + k) − 1− x ≪ ,

p−1 (log x)B

 
k≤y n≤x p>2
µ2 (k)=1

for r = 2. Later, Foo & Zhao [6] proved the case of r = 3. We

remark that in the speacial case of r = 2, Baier & Zhao [3] widened
the region to x1/2+ε ≤ y ≤ x by instead using the dispersion method
of Linnik [19].
In this paper we are concerned with showing the case of r = 4. As in
all the previous approaches [1, 6], we use the circle method to estimate
X Z 1 X X
4
Λ(n + k) = Λ(m)e(αm) e(−α(n4 + k)) dα.
n≤x 0 m≤x4 +k n≤x

Next, we divide the interval [0, 1] into two sets, the major arc M and
the minor arc m, which corresponds to the main and error term respec-
tively. The computation for the integral over the major arc M and the
corresponding error terms are done in sections 4–7. The computation
for the integral over the minor arc m are done in section 8.

2. Results
We write f ≪ g or f = O(g) to mean there exist a C > 0 such that
f ≤ Cg . When M ≤ m < 2M we write m ∼ M .
For any integer n ≥ 1, we can write n = ℓ2 m where µ2 (m) = 1. We
denote the square-full part of n to be κ(n) = ℓ2 , and so n is square-free
if and only if κ(n) = 1.
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 3

Theorem 2.1. For any fixed A, B, ε > 0, we have for x4 (log x)−A ≤
y ≤ x4 , that
 2
X X  yx2
Λ(n4 + k) − S(k)x ≪ ,

(log x)B

 
k≤y n≤x
κ(k)≤y 1/2−ε

where the singular series is given by

Y np − 1

S(k) = 1− ,
p>2
p − 1
and np is the number of solutions to n4 + k ≡ 0 mod p in Z/pZ.
Theorem 2.1 implies the following result immediately.
Corollary 2.2. For any fixed A, B, C, ε > 0 and S(k) as defined in
Theorem 2.1, we have for x4 (log x)−A ≤ y ≤ x4 that
 
X
4 x
Λ(n + k) = S(k)x + O
n≤x
(log x)B

for all k up to y such that κ(k) ≤ y 1/2−ε with at most O y(log x)−C


exceptions.

3. Preliminaries
We recall a special case of the large sieve over quartic characters
from [9, Theorem 1.2].
Lemma 3.1. Let (am )m∈N be a sequence of complex numbers. Then
 2
X X⋆  X 
2
µ (m)am χ(m)


 
q∼Q χ mod q m∼M
χ4 =χ0 ,χ2 6=χ0
X
≪ (QM)ε (Q5/4 + Q2/3 M) µ2 (m)|am |2 ,
m∼M
where the ⋆ on the sum over χ restricts the sum to primitive characters.
Next, we recall a bound for linear multiplicative character sum inde-
pendently proved by Pólya, and Vinogradov, see [17, Theorem 12.5].
Lemma 3.2 (Pólya-Vinogradov). For any non-principle character χ
modulo q , we have
 
 X 
χ(n) ≪ q 1/2 log q.
 

 
M <n≤M +N
4 KAM HUNG YAU

We state a large sieve inequality for number fields from [14, Theorem
1].
Lemma 3.3. Let K be a number field and r denote an ideal in K .
Suppose u(r) is a complex-valued function defined on the set of ideals
in K . We have
 2
 
X N (f ) X∗  X X
u(r)χ(r) ≪ (z + Q2 ) |u(r)|2,


Φ(f) χ mod f 

N (f)≤Q N (r)≤z  N (r)≤z

where N (f) denotes the norm of the ideal f, Φ(f) is Euler’s totient
function generalized to the setting of number fields, the ∗ over the sum-
mation over χ indicates that χ is a primitive character of narrow ideal
class group modulo f and the implicit constant depends on K .
We recall the Duality principle in [11, Theorem 228].
Lemma 3.4 (Duality principle). For a finite square T = (tmn ) ma-
trix with entries in the complex numbers. The follow statements are
equivalent:
For any complex sequence (an ), we have
 2
X X  X
an mn  ≤ D
t |an |2 .


 
m n n

For any complex sequence (bn ), we have

 2
X X  X
bm tmn  ≤ D |bm |2 .


 
n m m

Recall the Perron formula from [5].

Lemma 3.5 (Perron). Suppose that y 6= 1 is a positive real number then
for c, T > 0 we have
Z c+iT s (
1 y 1 + O (y c min{1, T −1 | log y|−1 }) if y > 1,
ds =
2πi c−iT s O (y c min{1, T −1 | log y|−1}) otherwise.
Recall the Weyl bound from [17, Proposition 8.2].
Lemma 3.6 (Weyl). If f (x) = αxk + . . . + a0 is a polynomial with real
coefficients and k ≥ 1 then
    21−k
X   X 1 
e(f (n)) ≤ 2N N −k min N, .
  
 Qk−1 
αk! ℓ
  
n≤N −N <ℓ1 ,...,ℓk−1 <N i=1 i

Here kxk is the distance of x to the nearest integer .

 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 5

Recall a result from Mikawa [20].

Lemma 3.7. Let

 #
2
X Z 2N  X 
J(q, ∆) = Λ(n)χ(n) dt
 

χ mod q N 
t<n<t+q∆


where the # over the summation symbol means that if χ = χ0 , then

χ(n)Λ(n) is replaced by Λ(n) − 1. Let ε, A, B > 0 be given. If q ≤
(log N)B and N 1/5+ε < ∆ < N 1−ε , then we have

J(q, ∆) ≪ε,A,B (q∆)2 N(log N)−A .

Recall a result by Gallagher [8, Lemma 1].

Lemma 3.8. Let 2 < ∆ < N/2 and N < N ′ < 2N . For arbitrary
complex sequence (an )n∈N , we have
 2  2
Z  X  Z N 
X

−2
 
an e(βn) dβ ≪ ∆ a  dt.
 
n
 
 
|β|<∆−1 N <n<N ′  N −∆/2 max{t,N }<n<min{t+∆/2,N ′ } 

We recall a classical result from [14].

Lemma 3.9 (Bessel). Let φ1 , φ2 , . . . , φR be orthonormal members of an

inner product spaces V over C and let ξ ∈ V . Then
R
X
|(ξ, φr )|2 ≤ (ξ, ξ).
r=1

We recall a zero-free region for L-function from [17, Theorem 5.35].

Lemma 3.10. Let K\Q be a number field, ξ a Hecke Grossencharakter

modulo (m, Ω) where m is a non-zero integral ideal in K and Ω is
a set of real infinite places where ξ is ramified. Let the conductor
∆ = |dK |NK/Q m. There exists an absolute effective constant c′ > 0
such that the L-function L(ξ, s) of degree d = [K : Q] has at most a
simple real zero in the region
c′
σ >1− .
d log ∆(|t| + 3)
The exceptional zero can occur only for a real character and it is strictly
less than 1.
6 KAM HUNG YAU

4. The major arc

The major arc is written as follows
q
[ [
M= Ia,q ,
q≤Q1 a=1
(a,q)=1

where
 
a 1 a 1
Ia,q = − , + , Q1 = (log x)c1 , Q2 = x1−ε ,
q qQ2 q qQ2
for some suitable fixed c1 > 0. We remark that when x is sufficiently
large, Q2 > Q1 so that the intervals Ia,q with q ≤ Q1 are disjoint.
We write α ∈ M as
a 1
α = + β with |β| ≤ .
q qQ2
Let
X X
S1 (α) = Λ(m)e(αm) and S2 (α) = e(−αn4 ).
m≤x4 +k n≤x

We will first estimate S1 . For convenience we set

(4.1) z = x4 + k.
We have
X
S1 (α) = Λ(m)e(am/q)e(βm)
m≤z
X
Λ(m)e(am/q)e(βm) + O log2 z .


(4.2) =
m≤z
(m,q)=1

For (am, q) = 1, we recall the identity

1 X
(4.3) e(am/q) = χ(am)τ (χ),
ϕ(q) χ mod q

where
q
X
τ (χ) = χ(n)e(an/q)
n=1
is the well-known Gauss sum. Applying this identity, the main term
of (4.2) is transformed into
1 X X
τ (χ)χ(a) χ(m)Λ(m)e(βm).
ϕ(q) χ mod q m≤z
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 7

Separating the main term corresponding to the principal character χ0

and using a property of Ramanujan sum, we obtain
!
µ(q) X X
e(βm) + (Λ(m) − 1)e(βm)
ϕ(q) m≤z m≤z
1 X X
+ τ (χ)χ(a) χ(m)Λ(m)e(βm)
ϕ(q) m≤z
χ mod q
χ6=χ0
#
µ(q) X 1 X X
= e(βm) + τ (χ)χ(a) χ(m)Λ(m)e(βm)
ϕ(q) m≤z ϕ(q) χ mod q m≤z

= T1 (α) + E1 (α), say.

Here the # over the summation symbol means that if χ = χ0 , then
χ(n)Λ(n) is replaced by Λ(n) − 1. Hence
S1 (α) = T1 (α) + E1 (α) + O log2 z .

Next we work on S2 (α). Recalling (4.3), we get

X
S2 (α) = e(−an4 /q)e(−βn4 )
n≤x
1 X X
= χ(−a)τ (χ̄) χ4 (n)e(−βn4 )
ϕ(q) χ mod q n≤x
X 1 X X
∗
= χ(−ad )τ (χ) χ4 (n∗ )e(−βn4 ),
ϕ(q1∗ ) χ mod q∗ n≤x
d|q 1
(n,q)=1

where
q n d3 q∗
q∗ = , n∗ = , d∗ = , q1∗ = .
d d (d3 , q ∗ ) (d3 , q ∗ )
Therefore by separating the the main term corresponding to χ = χ0 ,
we obtain
X 1 X X
∗
S2 (α) = χ(−ad )τ (χ) e(−βn4 )
ϕ(q1∗ ) χ mod q∗ n≤x
d|q 1
χ4 =χ0 (n,q)=d
X 1 X
∗
X
+ ∗
χ(−ad )τ (χ) χ4 (n∗ )e(−βn4 )
ϕ(q1 ) χ mod q∗ n≤x
d|q 1
χ4 6=χ0 (n,q)=d

= T2 (α) + E2 (α), say.

8 KAM HUNG YAU

We consider the main term T2 (α). Observe that for G = (Z/q1∗ Z)∗
and G4 = {g 4 : g ∈ G}, we have by orthogonality
X X X
χ(−ad∗ )τ (χ) = e(b/q1∗ ) χ(−(ad∗ )3 b)
χ mod q1∗ b mod q1∗ χ mod q1∗
χ4 =χ0 (b,q1∗ )=1 χ4 =χ0
#G X
= e(b/q1∗ ),
#G4
−(ad∗ )3 b≡♥ mod q1∗
(b,q1∗ )=1

since χ3 = χ. Here r ≡ ♥ mod q1∗ means that n is congruent to a

fourth power of an integer modulo q1∗ . Moreover,
q1 ∗
X #G4 X
e(b/q1∗ ) = e(−(ad∗ )3 ℓ4 /q1∗ )
#G l=1
−ad∗ b≡♥ mod q1∗
(b,q1∗ )=1 (ℓ,q1∗ )=1

q1 ∗
#G4 X
= e(−ad∗ ℓ4 /q1∗ ),
#G ℓ=1
(ℓ,q1∗ )=1

since
(ad∗ )3 ℓ4 ≡ ad∗ (ad∗ ℓ)4 mod q1∗ ,
where u is the multiplicative inverse of u modulo q1∗ . It follows
q1 ∗
X 1 X
∗ 4 ∗
X
T2 (α) = e(−ad ℓ /q1 ) e(−βn4 ).
ϕ(q1∗ ) ℓ=1 n≤x
d|q
(ℓ,q1∗ )=1 (n,q)=d

In total, we obtain
Z X X
Λ(m)e(αm) e(−α(n4 + k)) dα
M m≤z n≤x
Z
= {T1 (α) + E1 (α) + O(log2 x)}{T2 (α) + E2 (α)}e(−αk) dα.
M

5. Computing the singular series

The main term contributing in the major arc is
Z
T1 (α)T2 (α)e(−αk) dα
M
q1 ∗
X µ(q) X  −ak  X 1 −ad∗ ℓ4
X  Z
= e ∗
e ∗
Θd,q (β) dβ,
q≤Q
ϕ(q) q ϕ(q1 ) q1 |β|< 1
1 a mod q d|q qQ ℓ=1 2
(a,q)=1 (ℓ,q1∗ )=1
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 9

where
X
X
Θd,q (β) = e(−βk) e −βn4 e (−βm) .
n≤x m≤z
(n,q)=d

As in [6], we get that

Z
T1 (α)T2 (α)e(−kα) dα
M
q1 ∗
X µ(q) X  −ak  X 1 −ad∗ ℓ4
X  
= e ∗
e
q≤Q
ϕ(q) a mod q
q ϕ(q1 ) ℓ=1
q1∗
1 d|q
(a,q)=1 (ℓ,q1∗ )=1
(   1 !)
ϕ(q/d) qQ2 x 2
× x+O .
q d

The presence of µ(q) allows us to assume that q is square-free and

as a consequence we have d∗ = d3 , q1∗ = q ∗ = q/d. Thus we further
simplify the above to
q/d
X µ(q) X  −ak  X X −a(dℓ)4
 
+ O (xQ2 )1/2 (log x)c2


x e e
q≤Q
ϕ(q)q a mod q q ℓ=1
q
1 d|q
(a,q)=1 (ℓ,q/d)=1

for some fixed constant c2 > 0.

Denote
q/d
X  −ak  X X −a(dℓ)4
 
Σ(q) = e e
q q
a mod q d|q ℓ=1
(a,q)=1 (ℓ,q/d)=1
q
−am4
 X  
X −ak
= e e .
a mod q
q q
m=1
(a,q)=1

First we suppose q = p, a prime. A property of Ramanujan sum

gives
p
X X  −a(k + m4 ) 
Σ(p) = e
m=1 a mod p
p
(a,p)=1

= (p − 1)nk,p − (p − nk,p )
= p(nk,p − 1),
10 KAM HUNG YAU

where nk,p is the number of solutions to

m4 + k ≡ 0 mod p, m ∈ Z/pZ.

Now let χ1,p , χ2,p , χ3,p be three quartic characters of modulus p prime
such that p ≡ 1 mod 4 and χ21,p , χ22,p , χ23,p 6= χ0 . Then
(
p{χ1,p (−k) + χ2,p (−k) + χ3,p (−k)} if p ≡ 1 mod 4,
Σ(p) =
0 if p = 2 or p ≡ 3 mod 4.

Moreover, Σ(q) is a multiplicative function in q . Indeed for square-free

positive integers q1 , q2 with all primes factors congruent to 1 modulo
4, and (q1 , q2 ) = 1, we have
p p
X X X X
Σ(q1 )Σ(q2 ) = e(Υk (a1 , r1 , q1 , a2 , r2 , q2 )),
m1 =1 a1 mod p m2 =1 a2 mod p
(a1 ,p)=1 (a2 ,p)=1

where
a1 q2 + a2 q1 a1 q2 (q2 r1 )4 + a2 q1 (q1 r2 )4
Υk (a1 , r1 , q1 , a2 , r2 , q2 ) = −k −
q1 q2 q1 q2
a1 q2 + a2 q1 (a1 q2 + a2 q1 )(q1 r2 + q2 r1 )4
≡ −k − mod 1.
q1 q2 q1 q2
Consequently
Σ(q1 q2 ) = Σ(q1 )Σ(q2 ).
Therefore for some fixed constant c2 > 0, we get
Z X µ(q)
Σ(q) + O (xQ2 )1/2 (log x)c2


T1 (α)T2 (α)e(−kα) dα = x
M q≤Q
ϕ(q)q
1
∞
Xµ(q) Y
(nk,p − 1) − xΨ(k) + O (xQ2 )1/2 (log x)c2


=x
q=1
ϕ(q)
p|q

= S(k)x + O x|Ψ(k)| + (xQ2 )1/2 (log x)c2 ,

(5.1)

where
X µ(q) Y
Ψ(k) = (nk,p − 1).
q>Q
ϕ(q)
1 p|q

Note that we can restrict q to be square-free with prime factors that

are congruent to 1 modulo 4.
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 11

6. Bounding the second moment of Ψ(k)

First we partition the second moment into three pieces
X
|Ψ(k)|2 ≪ Ψ1 + Ψ2 + Ψ3 ,
k≤y
′
κ(k)≤y 1/2−ε

where
 2
X  X µ(q) Y 

Ψ1 =  {χ1,p (−k) + χ2,p (−k) + χ3,p (−k)} ,
 ϕ(q)
k≤y Q1 <q≤U p|q 
 2
 
X  X µ(q) Y 
Ψ2 =  {χ1,p (−k) + χ2,p (−k) + χ3,p (−k)}  ,
k≤y

U <q≤2v U ϕ(q) 
p|q 
′
κ(k)≤y 1/2−ε
 2
 
X  X µ(q) Y 
Ψ3 =  {χ1,p (−k) + χ2,p (−k) + χ3,p (−k)}  ,

v
k≤y 2 U <q
ϕ(q) 
 p|q 

and U, v are parameters to be chosen later.

We deal with Ψ1 first. Expanding the square, we obtain
X µ2 (q) X µ(q1 )µ(q2 ) X
ω(q)
Ψ1 ≤ y 2
9 + Pq1 ,q2 (k),
Q <q≤U
ϕ(q) Q <q ,q ≤U
ϕ(q1 )ϕ(q2 ) k≤y
1 1 1 2
q1 6=q2

where
Pq1 ,q2 (k)
Y Y
= {χ1,p (−k) + χ2,p (−k) + χ3,p (−k)} {χ1,p (−k) + χ2,p (−k) + χ3,p (−k)}.
p|q1 p|q2

The first term is no more than

X 9ω(q) (log log 10q)2 X 1 y
ε
y 2
≪ yQ1 2
≪ 1−ε ,
Q <q≤U
q Q <q≤U
q Q1
1 1

by using the well-known bounds

q
(6.1) ≪ ϕ(q),
log log 10q
and
log q
ω(q) ≪ .
log log q
12 KAM HUNG YAU

The second term is bounded by

!2
X (q1 q2 )1/2 log(q1 q2 )3ω(q1 )+ω(q2 ) X q 1/2 3ω(q)
≪ (log U)
Q1 <q1 ,q2 ≤U
ϕ(q1 )ϕ(q2 ) Q <q≤U
ϕ(q)
1
q1 6=q2

≪ U 1+ε .
Therefore
y 1+ε
(6.2) 1−ε + U Ψ1 ≪
.
Q1
Partitioning the summation inside and applying the Cauchy’s in-
equality, we have
 2
⌊v+1⌋  
X X  X µ(q) Y 
Ψ2 ≪ v  {χ 1,p (−k) + χ2,p (−k) + χ3,p (−k)}  .
r=1 k≤y

 r−1 r
ϕ(q) 
2 U <q≤2 U p|q 
′
κ(k)≤y 1/2−ε

Expanding the term in the square, we have

X µ(q) Y
{χ1,p (−k) + χ2,p (−k) + χ3,p (−k)}
r−1 r
ϕ(q)
2 U <q≤2 U p|q
X X µ(q)
= χ(k).
ϕ(q)
2r−1 U <q≤2r U χ mod q
χ4 =χ0 ,χ2 6=χ0

Note that by Lemma 3.1, we get by Cauchy’s inequality

 2
 
 
X X  X 
ak χ(k)
 

 
2r−1 U <q≤2r U χ mod q  k≤y ′ 
4 2
χ =χ0 ,χ 6=χ0 κ(k)≤y
 1/2−ε 
 2
 
X X  X X
2 2

= 
 µ (m)aℓ 2 m χ(ℓ )χ(m)

2r−1 U <q≤2r U χ mod q ℓ2 ≤y1/2−ε′ m≤y/ℓ2 
χ4 =χ0 ,χ2 6=χ0
 2
 
′
X X X  X 2
≪ y 1/4−ε /2


 µ (m)a 2
ℓ m χ(m) 

2r−1 U <q≤2r U ′ 
χ mod q ℓ2 ≤y 1/2−ε m≤y/ℓ2

χ4 =χ0 ,χ2 6=χ0
′
X
≪ y 1/4−ε /2 (2r Uy)ε ((2r U)5/4 + (2r U)2/3 y) |ak |2 .
k≤y
′
κ(k)≤y 1/2−ε
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 13

Hence by the Duality principle (Lemma 3.4), we have

 2
 
 
X  X X µ(q) 
 χ(k)
2r−1 U <q≤2r U χ mod q ϕ(q)

k≤y 
′
χ4 =χ ,χ2 6=χ
1/2−ε
 
κ(k)≤y 0 0

′
X 1
≪ y 1/4−ε /2 (2r Uy)ε ((2r U)5/4 + (2r U)2/3 y)
ϕ(q)2
2r−1 U <q≤2r U
1/4−ε′ /2 r ε r 1/4 r −1/3
≪y (2 Uy) ((2 U) + (2 U) y).
Summing over r = 1, . . . , R = ⌊log2 (y 3+ε /U)⌋ (logarithm base 2),
we obtain
 2
 
R
y 1+ε
   
X X  X X µ(q)  1/4−ε′ 3/4+6ε
 χ(k) ≪ y
 y + 1/3−ε .
r=1 k≤y

r−1 r χ mod q
ϕ(q) U
 2 U <q≤2 U 
′ 4 2
χ =χ0 ,χ 6=χ0
1/2−ε
 
κ(k)≤y

Note that primitive quartic character of conductor q coprime to 4

m


can be realised as quartic residue symbols n 4 for some square-free
n ∈ Z[i], n ≡ 1 mod 4 and not divisible by any rational primes with
norm q . See [15] for a thorough background.
For large values of r ’s with R < r ≤ ⌊v + 1⌋, it is enough to bound
 2
 
 
X  X 
(6.3) 
 aπ χπ (n)  ,

n∈Z[i]  π∈Q[i] 
N (n)≤y 2 2r−1 <N (π)=q≤2r U
 
n≡1 mod 4
′
κ(n)≤y 1/2−ε

n


where aπ = µ(q)/ϕ(q), χπ (n) = π 4
. Note that by the quartic reci-
procity, we assert
N (n)−1 N (π)−1
χπ (n) = χn (π)(−1) 4 4 ,
therefore in view of [9, Lemma 3.2], applying Lemma 3.3 to (6.3) and
recalling (6.1), we majorise by
X 1
≪ (y 4 + 2r U)
ϕ(q)2
N (π)=q
2r−1 U <q≤2r U

log log(2 U) log(2r U)

r
≪ (y 4 + 2r U) .
2r U
14 KAM HUNG YAU

Hence
 2
 
r r
 
X  X X µ(q)  ≪ (y 4 +2r U) log log(2 U) log(2 U) .

 χ(k)
2r−1 U <q≤2r U χ mod q ϕ(q)
  2r U
k≤y 
1/2−ε′ χ4 =χ0 ,χ2 6=χ0
 
κ(k)≤y

Summing over r from R to ⌊v + 1⌋, we have

 2
 
⌊v+1⌋  
X X  X X µ(q) 
 χ(k)
2r−1 U <q≤2r U χ mod q ϕ(q)

r=R k≤y 
′
χ4 =χ0 ,χ2 6=χ0
1/2−ε
 
κ(k)≤y
⌊v+1⌋ ⌊v+1⌋
4
X 1 X
≪ y (v + log U) log(v + log U) + (r + log U)2
r=R
2r U r=R
1−ε
≪y (v + log U) log(v + log U) + v 3 + v 2 log U + v(log U)2 .
Therefore
(6.4)
y 1+ε
 
1/4−ε′ /2 3/4+6ε
Ψ2 ≪ v y y + 1/3−ε + y 1−ε (v + log U) log(v + log U)
U
!
+ v 3 + v 2 log U + v(log U)2 .

Lastly, we bound Ψ3 . For primes p ≡ 1 mod 4, p splits in Q[i] as

p = πp,1 πp,2 . Consider the function
Y  χ1,p (−k) + χ2,p (−k) + χ3,p (−k)

f (s, k) = 1−
p≡1 mod 4
(p − 1)ps
Y  nk,p − 1
 X
µ(q) Y
= 1− s
= s
(nk,p − 1)
p≡1 mod 4
(p − 1)p q
ϕ(q)q
p|q
     
k k
Y πp,1
+ πp,2
1 − 4 4
= s
.
p≡1 mod 4
(p − 1)p
p=πp,1 πp,2

Clearly if s = 0 then f (s, k) = S(k), and also f (s, k) has no poles

with ℜ(s) > 0. If we take
µ(q) Y
bq = (nk,p − 1),
ϕ(q)
p|q
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 15

then the Dirichlet series associated with bq is f (s, k).

Let us next consider the following Hecke L-function
   Y k

!−1
k π 4
L s + 1, = 1− s+1
· 4 π
N(s)
   −1
k

!−1 k
1−i 4
Y p
4
= 1 − s+1 1 −
s+1
2 p≡3 mod 4
p
   −1    −1
k k
Y πp,1 πp,2
4 4
× 1 −  1 − .
ps+1 ps+1
p≡1 mod 4
p=πp,1 πp,2

It is natural to approximate f by L−1 . Denote

  
k
h(s, k) = L s + 1, f (s, k),
· 4
so that we can write
  
−1 k
f (s, k) = L s + 1, h(s, k).
· 4
It can be shown that h is absolutely bounded for all ℜ(s) > −1/2 + ε
for any fixed ε > 0. Applying the Perron formula (Lemma 3.5), we
have
Z C+iT
y s − y1s
  
X 1 −1 k
bq = L s + 1, h(s, k) 2 ds
y1 ≤q≤y2
2πi C−iT · 4 s
2 X  C (   −1 )!
X yj yj 
min 1, T −1 log

+O |bq | .
j=1 q
q q 

for any C, T > 0. By applying Lemma 3.10 as in [6], we can bound

 √ 
(6.5) Ψ3 ≪ y exp −c3 2v U ,
for some fixed constant c3 > 0.
Setting
exp(y ε/3 ) ε′
 
3/4
U =y , v = log2 , ε=
U 28
(logarithm base 2) and recalling (6.2), (6.4), and (6.5), we get
X y
(6.6) |Ψ(k)|2 ≪ ,
k≤y
(log x)c4
′
κ(k)≤y 1/2−ε
16 KAM HUNG YAU

for some fixed constant c4 > 0.

7. Error terms from the major arc

There are three terms we need to bound, namely
 2
X Z 
E1 =  T1 (α)E2 (α)e(−αk) dα ,
 
k≤y M
 2
X Z 
E2 = 
 T2 (α)E1 (α)e(−αk) dα  ,

k≤y M
 2
X Z 
E3 =  E1 (α)E2 (α)e(−αk) dα .
 
k≤y M

Now
X X Z
E1 ≪ |T1 (a/q + β)E2 (a/q + β)|2 dβ
q<Q1 a mod q |β|<δ
(a,q)=1
Z !
+ |T1 (a/q + β)E2 (a/q + β)|2 dβ .
1
δ<|β|< qQ
2

By the summing the geometric series we can bound by

T1 (a/q + β) ≪ min(z, β −1 ),
and therefore
(7.1)
X X Z 1 2
Z
2

2
E1 ≪ z |E2 (a/q + β)| dβ + 2 |E2 (a/q + β)| dβ .
q<Q1 a mod q |β|<δ δ M
(a,q)=1

First consider
X X Z
Ωq (β) dβ,
1
q<Q1 a mod q |β|< qQ
2
(a,q)=1

where
 2
 
 
X 1 X
∗
X
4 ∗ 4 

Ωq (β) = 

∗
χ(−ad )τ (χ) χ (n )e(−βn ) .
 d|q ϕ(q1 ) χ mod q1∗ n≤x 
 χ4 6=χ 0
(n,q)=d 
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 17

Partition the summation over n into dyadic intervals and let N < n ≤
N ′ ≤ 2N ≤ x. Applying the Cauchy’s inequality, it is enough to bound
 2
 
 
XX X Z X
c5 4 ∗ 4 
 
(log x)  χ (n )e(−βn ) dβ,
1

q≤Q1 d|q χ mod q1∗ |β|≤ qQ  n∼N ′ 
2 (n,q)=d
χ4 6=χ0


which is no more than

 2
 
c5 X Z 2x  
(log x) 1 X X  X
4 ∗ 

 χ (n )  dt
Q22 q≤Q q 2 χ mod q ∗ −x

′
1 d|q max{t,N }<n≤min{t+qQ 2 /2,N }
1
 
χ4 6=χ (n,q)=d

0

by Lemma 3.8. The character sum is

X X
χ4 (n∗ ) = χ4 (n∗ )
max{t,N }<n≤min{t+qQ2 /2,N ′ } max{t,N }/d<n∗≤min{t+qQ2 /2,N ′ }/d
(n,q)=d (n∗ ,q ∗ )=1
X
= χ′ (n∗ )
max{t,N }/d<n∗≤min{t+qQ2 /2,N ′ }/d

q1∗ log q1∗ ,

p
≪

where χ′ is the non-trivial character modulo q1∗ induced by the charac-

ter χ2 modulo q1∗ . The last line follows by Pólya-Vinogradov (Lemma 3.2).
The first term in the upper bound of (7.1) can be estimated similarly.
Therefore
2 2 x(log x)c6c5
E1 ≪ z xδ (log x) +
δ 2 Q22
for some fixed constants c5 , c6 > 0.
Now by Lemma 3.9, we get
Z
E2 ≪ |T2 (α)E1 (α)|2 dα
M
Z
2
≪ sup |T2 (α)| |E1 (α)|2 dα
α∈M M
Z
≪ x2 |E1 (α)|2 dα.
M
18 KAM HUNG YAU

As exactly in [6], we get using Lemma 3.7 to get

Z X q
|E1 (α)|2 dα ≪ (qQ2 )−2 J(q, Q2 /2) + Q31 Q2 (log x)2
M q<Q1
ϕ(q)
X q
(7.2) ≪ z(log z)−A ,
q<Q
ϕ(q)
1

for any A > 0. Thus

x2 z
E2 ≪
.
(log x)c7
Again by Lemma 3.9 and (7.2) we obtain
Z
E3 ≪ |E1 (α)E2 (α)|2 dα
M
Z
2
≪ sup |E2 (α)| |E1 (α)|2 dα
α∈M M
x2 z
≪ .
(log x)c8
√
Therefore by taking δ = 1/ z , and collecting the bounds, we obtain
x(log x)c6 x2 z
E1 + E2 + E3 ≪ z 2 xδ 2 (log x)c5 + +
δ 2 Q2 (log x)c7
x2 z
(7.3) ≪ zx(log x)c5 + .
(log x)c7

8. The minor arc

Lastly we estimate the minor arc
Z 2
X  
 S1 (α)S2 (α)e(−αk) dα
 
k≤y m
′
κ(k)≤y 1/2−ε
Z 2
X  X X 
4
= Λ(m)e(αm) e(−α(n + k)) dα .
 

 m 
k≤y m≤z n≤x
′
κ(k)≤y 1/2−ε

Here the minor arc is given by

m = [1/Q2 , 1 + 1/Q2 ]\M.
By Lemma 3.9, the above is majorised by
Z 1
2
sup |S2 (α)| |S1 (α)|2 dα ≪ (z log z) sup |S2 (α)|2 .
α∈m 0 α∈m
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 19

Now by Weyl shift (Lemma 3.6), we get

 
8 4
X 1
|S2 (α)| ≪ x min x, .
−x<ℓ1 ,ℓ2 ,ℓ3 <x
k24αℓ1ℓ2 ℓ3 k
By Dirichlet’s theorem, we have an approximation
 
α − a  ≤ 1 ,
 
 q  48x3 q
where (a, q) = 1 and 1 ≤ q ≤ 48x3 . Since α ∈ m, we further have that
q > Q1 . Therefore for −x < ℓ1 , ℓ2 , ℓ3 < x, we have
  

24ℓ1ℓ2 ℓ3 α − a ≤ 1,

 q  2q
which implies
1 2
≤ .
k24ℓ1 ℓ2 ℓ3 αk k24ℓ1 ℓ2 ℓ3 a/qk
In a first step
 
X 1
min x,
−x<ℓ1 ,ℓ2 ,ℓ3 <x
k24αℓ1 ℓ2 ℓ3 k
X X 2
≪x 1+ .
k24ℓ1 ℓ2 ℓ3 a/qk
−x<ℓ1 ,ℓ2 ,ℓ3 <x −x<ℓ1 ,ℓ2 ,ℓ3 <x
q|24ℓ1 ℓ2 ℓ3 q∤24ℓ1 ℓ2 ℓ3

Now let q ′ = (q, 24) then

X X X X X
1≪ 1
−x<ℓ1 ,ℓ2 ,ℓ3 <x q1 q2 |q ′ −x<ℓ1 <x −x<ℓ2 <x −x<ℓ3 <x
q|24ℓ1 ℓ2 ℓ3 q1 |ℓ1 q2 |ℓ2 q′
|ℓ
q1 q2 3

X x 
x

x

≪ +1 +1 +1
′
q1 q2 q ′ /(q1 q2 )
q1 q2 |q

x3
≪ τ (q ′ )2 + τ (q ′ )2 x
q′
≪ x3 (log x)−c9 ,
for some fixed constant c9 > 0, and also
X 2 X q
≪ ≪ x3+ε .
−x<ℓ ,ℓ ,ℓ <x
k24ℓ1 ℓ2 ℓ3 a/qk 1≤ℓ ,ℓ ,ℓ <x
ℓ ℓ ℓ
1 2 3
1 2 3 1 2 3
q∤24ℓ1 ℓ2 ℓ3

Therefore
x2
sup |S2 (α)|2 ≪ ,
α∈m (log x)c9
20 KAM HUNG YAU

and hence recalling (4.1), we obtain

2
x6
Z
X  
(8.1)  S1 (α)S2 (α)e(−αk) dα ≪ (log x)c9 .
 
k≤y m
′
κ(k)≤y 1/2−ε

9. Proof of Theorem 2.1

By Cauchy’s inequality and recalling (5.1), (6.6), (7.3), and (8.1), we
get
 2
X X
4
 yx2 c5 x2 z x6
Λ(n + k) − S(k)x ≪ + zx(log x) + +
 
(log x)c4 (log x)c7 (log x)c9

 
k≤y n≤x
′
κ(k)≤y 1/2−ε

yx2
≪
(log x)c4
since x4 (log x)−A ≤ y ≤ x4 and by taking c4 to be sufficiently large.

Acknowledgement
The author thanks L. Zhao for the problem and many helpful conver-
sations and comments, and also I. E. Shparlinski for helpful comments.
This work is supported by an Australian Government Research Train-
ing Program (RTP) Scholarship.

References
[1] S. Baier, L. Zhao, Primes in quadratic progressions on average. Math. Ann.
338 (4) (2007), 963–982. 2
[2] S. Baier, L. Zhao, On primes represented by quadratic polynomials. Anatomy
of integers, 159–166, CRM Proc. Lecture Notes 46, Amer. Math. Soc., 2008. 2
[3] S. Baier, L. Zhao, On primes in quadratric progressions. Int. J. Number Theory
5 (6) (2009), 1017-1035. 2
[4] P. T. Bateman, R. A. Horn, A heuristic asymptotic formula concerning the
distribution of prime numbers. Math. Comp. 16, 1962, 363–367. 1
[5] H. Davenport, Multiplicative number theory. Third edition. Graduate Texts in
Mathematics, 74. Springer-Verlag, New York, 2000. 4
[6] T. Foo, L. Zhao, On primes represented by cubic polynomials. Math. Z. 274
(1-2) (2013), 323–340. 2, 9, 15, 18
[7] J. B. Friedlander, H. Iwaniec, The polynomial X 2 + Y 4 captures its primes.
Ann. of Math. (2) 148 (3) (1998), 945–1040. 1
[8] P. X. Gallagher, A large sieve density estimate near σ = 1. Invent. Math. 11,
1970, 329–339. 5
[9] P. Gao, L. Zhao, Large sieve inequalities for quartic characters. Q. J. Math.
63 (4) (2012), 891–917. 3, 13
 PRIMES REPRESENTED BY QUARTIC POLYNOMIALS ON AVERAGE 21

[10] P. R. Halmos, Finite-Dimensional Vector Spaces. D. Van Nostrand, New York

(1958). 3, 5
[11] G. H. Hardy, J. E. Littlewood, G. Pólya, Inequalities. Cambridge University
Press, Cambridge (1964). 4
[12] D. R. Heath-Brown, Primes represented by x3 + 2y 3 . Acta Math. 186 (1)
(2001), 1–84. 1
[13] D. R. Heath-Brown, X. Li, Prime values of a2 + p4 . Invent. Math. 208 (2)
(2017), 441–499. 1
[14] M. N. Huxley, The large sieve inequality for algebraic number fields. II. Means
of moments of Hecke zeta-functions. Proc. London Math. Soc. (3) 21 (1970),
108–128. 3, 5
[15] K. Ireland, M. Rosen, A classical introduction to modern number theory. Sec-
ond edition. Graduate Texts in Mathematics 84. Springer-Verlag, New York,
1990. 13
[16] H. Iwaniec, The half dimensional sieve. Acta Arith. 29 (1) (1976), 65–95. 1
[17] H. Iwaniec, E. Kowalski, Analytic number theory. American Mathematical
Society Colloquium Publications 53. American Mathematical Society, Prov-
idence, RI, 2004. 3, 4, 5
[18] P. Kuhn, Über die Primteiler eines Polynomsw. Proceedings of the Interna-
tional Congress of Mathematicians. Vol. 2 (Amsterdam, 1954), Erven P. No-
ordhoff N. V., Groningen; North-Holland, Amsterdam, 1954, 35–37. 1
[19] Ju. V. Linnik, The dispersion method in binary additive problems. American
Mathematical Society, Providence, R.I. 2
[20] H. Mikawa, On prime twins. Tsukuba J. Math. 15 (1) (1991), 19–29. 5

Department of Pure Mathematics, University of New South Wales,

Sydney, NSW 2052, Australia
E-mail address: kamhung.yau@unsw.edu.au