ASYMPTOTIC OF FOURIER COEFFICIENTS OF HECKE

EIGENFORMS AT INTEGERS REPRESENTED BY BINARY

QUADRATIC FORM OF FIXED DISCRIMINANT

MANISH KUMAR PANDEY AND LALIT VAISHYA

Abstract. In the article, we establish the general asymptotic behaviour of the Fourier coef-
ficients of the Hecke eigenforms supported at the integers represented by a primitive integral
positive-definite binary quadratic form of fixed discriminant D < 0 under the assumption
that the class number h(D) = 1.
We also obtain a quantitative result for the number of sign changes of the sequence of the
normalised Fourier coefficients λf (n) of the Hecke eigenforms f over the indices represented
by a primitive integral positive definite binary quadratic form of fixed discriminant D < 0
when the class number h(D) = 1 in the interval (x, 2x], for sufficiently large x. Moreover,
1
under the assumption of Lindelöf hypothesis, the above said sequence has at least x 2 −ϵ
many sign changes.

1. Introduction and statements of the results

  
a b
Let SL2 (Z) := : a, b, c, d ∈ Z, ad − bc = 1 be the full modular group,
c d
  
a b
Γ0 (N ) := ∈ SL2 (Z) : c ≡ 0 (mod N ) be a congruence subgroup of level N and
c d
H := {z ∈ C : ℑ(z) > 0} denote the complex upper half plane.
A holomorphic function f : H −→ C is said to be a modular form of integral weight k
with nebentypus χ for Γ0 (N ), if f satisfies the following conditions:
 
aτ + b k

a b

• f = χ(d)(cτ + d) f (τ ) ∀ c d ∈ Γ0 (N ) and τ ∈ H.
cτ + d
• f is holomorphic at each cusp of Γ0 (N ).
Moreover, from the above two conditions, f has the following Fourier series expansion at
each cusp s of Γ0 (N ):
∞
X
f (τ ) = af,s (n)q n where q = e2πiτ .
n=0

A modular form f is said to be a cusp form if it vanishes at each cusp s of Γ0 (N ), i.e.,

af,s (0) = 0. Let Mk (Γ0 (N ), χ) and Sk (Γ0 (N ), χ) denote the C- vector space of modular
forms and cusp forms respectively of weight k on the congruence subgroup Γ0 (N ) with the
nebentypus (Dirichlet) character χ. A cusp form f ∈ Sk (Γ0 (N ), χ) is said to be a Hecke
eigenform if f is a common eigenform for all the Hecke operators and the Atkin-Lehner
W -operators. In this article, we state our results for the full modular group SL2 (Z) but

Date: January 5, 2023.

2010 Mathematics Subject Classification. Primary 11F30, 11F11, 11M06; Secondary 11N37.
Key words and phrases. Fourier coefficients of cusp form, Rankin-Selberg L function, Symmetric power
L functions, Asymptotic behaviour, Binary quadratic form.
1
2 MANISH KUMAR PANDEY AND LALIT VAISHYA

results are also valid for the congruence subgroup Γ0 (N ). Let Sk (SL2 (Z)) denote the C-
vector space of cusp forms of weight k for the full modular group. Let f ∈ Sk (SL2 (Z)) be a
Hecke eigenform with the Fourier series expansion at the cusp ∞;
∞
k−1
X
(1) f (τ ) = λf (n)n 2 q n ,
n=1
2πiτ
where q := e . f is said to normalised if λf (1) = 1. The Fourier coefficient λf (n) is known
as the normalised nth -Fourier coefficient of f . The normalised Fourier coefficient λf (n) is a
multiplicative function and satisfies the following recursive relation [15, Eq. (6.83)]:
X  mn 
(2) λf (m)λf (n) = λf ,
d2
d| gcd(m,n)

for all positive integers m and n. It also satisfies Deligne bound

(3) |λf (n)| ≤ d(n) ≪ϵ nϵ ,
for any arbitrary small ϵ > 0, where d(n) denotes the number of positive divisors of n.
The Fourier coefficients of cusp forms are fascinating but quite mysterious objects in the
world of automorphic forms and contain a lot of interesting arithmetic informations. In
analytic number theory, it is a well known approach to look at summatory function of these
Fourier coefficients arithmetical over certain sequences. This leads to certain interesting
and important result about the modular forms such as distribution of signs of their Fourier
coefficients . R. A. Rankin [39] and A. Selberg [40], independently developed a method (now
referred to as the Rankin-Selberg method) and obtained the following estimates:
3
X
(λf (n))2 = Cx + Of,ϵ (x 5 +ϵ ).
n≤x

This remain the best known estimate for more than 80 years. Recently, B. Huang [12] has
improved this estimates only at the 3rd decimal places. C. J. Moreno and F. Shahidi [29]
has obtained the following estimates using the properties of symmetric power L-functions:
for sufficiently large x,
X
(τ0 (n))4 ≪ x log x,
n≤x
11
where τ0 (n) = τ (n)n− 2 is the normalised Ramanujan tau function. The result of Moreno
and Shahidi also hold true for the Fourier coefficients λf (n)’s of a normalised
X Hecke eigenform
f of any integral weight. O. M. Fomenko [8] has established estimates for (λf (n))r , when
n≤x
r = 3, 4 and improved the result of Moreno and Shahidi [29]. G. Lü [[26], [27], [28]] improved
Fomenko’s results and generalised the results for higher moments, i.e., for r ≥ 2. Y. K. Lau et
al. [23] considered more general cases and obtained better estimates. The problem becomes
difficult if one consider the summatory function over sparse sequence. In this direction, S.
Zhai [47] has obtained the estimates of the following sums; for 2 ≤ r ≤ 8 and x ≥ 1,
X
(λf (a2 + b2 ))r ,
(a,b)∈Z2
a2 +b2 ≤x

by decomposing the associated Dirichlet series in terms of symmetric power L-functions and
their Rankin-Selberg convolutions. Recently, C. Xu [43] has extended the work of Zhai [47]
for more general r ∈ N. Using the analytic properties of symmetric power L-functions, C.
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 3

Xu has obtained the estimates for the power sum of the Fourier
X coefficients λf (n) over the
sum of two squares, i.e., the estimates of following sums (λf (a2 + b2 ))r for general
(a,b)∈Z2
a2 +b2 ≤x

r ∈ N. In this article, we generalise the work of C. Xu [43] and obtain the estimates for the
more general power moments of the Fourier coefficients λf (n) where n is represented by a
primitive integral positive definite binary quadratic form of fixed negative discriminant D
with the class number h(D) = 1. Precisely, we obtain the estimates for the following sums:
for each r ∈ N and sufficiently large x ≥ 1,
X
(λf (Q(x)))r ,
(4) x∈Z2
Q(x)≤x

where λf (n) is the nth normalised Fourier coefficients of a Hecke eigenform f and Q(x) is
a primitive integral positive definite binary quadratic form (reduced form) of fixed negative
discriminant D with the class number h(D) = 1.
Moreover, these estimates lead to a beautiful result of distribution of signs of sequence
of the Fourier coefficients of the Hecke eigenforms supported at integers represented by
a primitive integral binary quadratic form Q(x) of fixed discriminant D < 0 with class
number h(D) = 1. In particular, D. L. Duda [7] improved the previous obtained result on
the number of sign changes of Fourier coefficients only for the subsequences indexed by sum
of two squares. In this work, we also improve the result obtained in [1, 44, 45]. We refer
to the introduction in [1, 44] for the history on the problem in sign changes of the Fourier
coefficients of (classical) Hecke eigenforms. Now, we fix some notations and state our results.
Throughout the paper, Q(x) denotes a primitive integral positive definite binary quadratic
(reduced) form given by; Q(x) = ax21 + bx1 x2 + cx22 , where x = (x1 , x2 ) ∈ Z2 , a, b, c ∈ Z with
gcd(a, b, c) = 1 and fixed discriminant D = b2 − 4ac < 0. Further, we assume that for such
discriminant D, the class number h(D) is 1. For details, we refer to [6, chapter 2].
Remark 1.1. For the primitive integral positive definite binary quadratic form of the fol-
lowing fundamental discriminants D = −3, −4, −8, −7, −11, −19, −43, −67, −163 and non-
fundamental discriminants D = −12, −16, −27, −28, the class number h(D) = 1. The list
of reduced form corresponding to these discriminants are given in [3, Page 19- 20].
Let rQ (n) denote the number of representations of a positive integer n by the quadratic
form Q(x), i.e., rQ (n) = #{x ∈ Z2 : n = Q(x)}. We define the generating function θQ (τ )
associated to binary quadratic form Q(x) as follows:
X ∞
X
(Q(x))
θQ (τ ) := q = rQ (n)q n , q = e2πiτ ,
x∈Z2 n=0

The generating function θQ (τ ) ∈ M1 (Γ0 (|D|), χD ) (see [15, Theorem 10.9]) where χD is the
D


Dirichlet character modulo |D| which is given by Jacobi symbol χD (d) := d . From Weil’s
bound, we have rQ (n) ≪ nϵ , for any arbitrarily small ϵ > 0.
We define the character sum r(n; D) (given explicitly) in terms of Jacobi symbol χD (see
[15, Eqs.(11.9), (11.10)]), i.e.,

X 6 if D = −3,

(5) r(n; D) = wD χD (d), where wD = 4 if D = −4,

2 if D < −4.
d|n
4 MANISH KUMAR PANDEY AND LALIT VAISHYA

Remark 1.2. The formula for r(n; D) depends only on the discriminant D but not on the
choice of a, b and c. It is known that r(n; D) counts the number of representations of a
positive integer n by all the reduced forms of fixed discriminant D.
In this work, we are considering the quadratic form Q(x) of discriminant D < 0 such that
the class number h(D) is 1. So the formula r(n; D) given in (5) agrees with the formula for
number of representations rQ (n) of n by the quadratic form Q(x) [15, section 11.2]. Thus,
we have
X
(6) rQ (n) = r(n; D) = wD χD (d).
d|n

Now, for each r ∈ N and x ≥ 1, we define the following power sum:

X
Sr (f, D; x) := (λf (Q(x)))r ,
(7) x∈Z2
Q(x)≤x

where λf (n) denote the nth normalised Fourier coefficient of a normalised Hecke eigenform
f ∈ Sk (SL2 (Z)) and Q(x) is a primitive integral positive definite binary quadratic form
(reduced form) with fixed discriminant D < 0 with the class number h(D) = 1.
Now, we state our results.
Theorem 1.1. For sufficiently large x, we have
(8) S1 (f, D; x) = Of,D,ϵ (x7/10+ϵ ),
and
(9) S2 (f, D; x) = Cx + Of,D,ϵ (x37/51+ϵ ),
where ϵ > 0 is an arbitrarily small real number and C > 0 is constant depends on f, ϵ, D.
Using a method Landau on the Dirichlet series with the functional equation as given in
[4] and following the arguments of [7, Section 4], we improve the estimate for S1 (f, D; x).
Theorem 1.2. Let ϵ > 0 be arbitrarily small constant. For sufficiently large x, we have
(10) S1 (f, D; x) = Of,D,ϵ (x3/5+ϵ ).
As mentioned in [7, Remark 12], An estimate for S1 (f, D; x) can be given without ϵ term
in (1.1). Throughout the paper, ϵ denotes an arbitrary small positive constant but not
necessarily the same one at each places.
Under the assumption of Lindelöf hypothesis, we have the following result.
Theorem 1.3. For sufficiently large x, assuming Lindelöf hypothesis, we have
(11) S1 (f, D; x) = Of,D,ϵ (x1/2+ϵ ),
and
(12) S2 (f, D; x) = C ′ x + Of,D,ϵ (x1/2+ϵ ),
where C ′ > 0 is constant depends on f, ϵ, D.
Theorem 1.4. Let ϵ > 0 be an arbitrarily small constant. For each r ∈ N and sufficiently
large x, we have the following estimates for Sr (f, D; x) :
(13) Sr (f, D; x) = xPr (log x) + Of,D,ϵ (x1−γr +ϵ ),
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 5

where for each r = 2m (m ≥ 2), Pr (t) is polynomial of degree

   
r r
dr = − −1 and γr = (θr,m + θr,m,χ )−1
m m−1
with
m−2
!
(r − 2n + 1)2
     
13 r 15 r 1 X r
θr,m = + + +r+1
42m m − 1 4(m − 1) m − 2 2 n=1
n n−1
and
m−2
!
(r − 2n + 1)2
     
1 r 15 r 1 X r
θr,m,χ = + + +r+1 ,
3m m − 1 4(m − 1) m − 2 2 n=1
n n−1
and for each r = 2m + 1 (m ≥ 1), Pr (t) is polynomial of degree 0 and γr = αr,m −1 with
m−1
!
8

r
 X (r − 2n + 1)2  r 
αr,m = + +r+1 .
3m m − 1 n=1
n n − 1
Corollary 1.5. Let ϵ > 0 be an arbitrarily small real number. For 1 ≤ r ≤ 8 and for
sufficiently large x, we have
(14) Sr (f, D; x) = xPr (log x) + Of,D,ϵ (xγr +ϵ ),
where P2 (x), P4 (x), P6 (x) and P8 (x) are polynomials of degree 0, 1, 4 and 13 respectively,
Pr (t) ≡ 0 for r = 1, 3, 5, 7 and
7
γ1 = 10 = 0.35, γ2 = 37
51
= 0.7254, 17
γ3 = 20 = 0.85, γ4 = 1112
1196
= 0.9297,
83
γ5 = 86 = 0.9651, γ6 = 1212 = 0.9760, γ7 = 356 = 0.9915 and γ8 = 708
1183 353
711
= 0.9957.
Remark 1.3. In previous work of second author [45], the idea of Zhai [47] was used to decom-
pose the associated Dirichlet series in terms of symmetric power L-functions and Rankin-
Selberg L-functions associated to symm f and symn f as the automorphy of L(s, symm f ) was
available only for m = 2, 3 and 4 due to H. H. Kim and F. Shahidi. For details, we refer
[17, 18, 19]. Recently, C. Xu [43] utilised the automorphy of L(s, symm f ) (which is avail-
able for each m) to decompose the associated Dirichlet series in terms of symmetric power
L-functions. Using analytic technique and properties of symmetric power L-functions, Xu
has obtained his results.
As a consequence of these result, we obtain the following improved result on sign change
of the Fourier coefficients λf (n) of f ∈ Sk (SL2 (Z)) supported at integers represented by a
primitive integral positive definite binary quadratic form Q(x) of fixed negative discriminant
D with the class number h(D) = 1.
∞
k−1
X
Theorem 1.6. Let f (τ ) = λf (n)n 2 q n ∈ Sk (SL2 (Z)) be a normalised Hecke eigenform.
n=1
Then, the sequence {λf (Q(x))}x∈Z2 has infinitely many sign changes. Moreover, for any
27
arbitrarily small constant ϵ > 0, there are at least x 102 −ϵ many sign changes of the sequence
{λf (Q(x))}x∈Z2 with Q(x) lying in the interval (x, 2x], for sufficiently large x.
Remark 1.4. In [1], first author obtained the quantitative result on the number of sign of
sign changes which is generalised to any binary quadratic forms of fixed discriminant having
class number 1 and the quantitative result was improved in second author’s works [44, 45].
D. L. Duda [7] improved the previous obtained result on the number of sign changes only for
6 MANISH KUMAR PANDEY AND LALIT VAISHYA

the subsequences indexed by sum of two squares. In this work, we generalised and improved
the results to any binary quadratic forms of fixed negative discriminant having class number
1.
Theorem 1.7. Let ϵ > 0 be an arbitrarily small constant. Under the assumption of Lindelöf
1
hypothesis, the sequence {λf (Q(x))}x∈Z2 has at least x 2 −ϵ many sign changes with Q(x) lying
in the interval (x, 2x], for sufficiently large x.
1.1. General estimate. The bounds for the partial sums Sr (f, D; x) (for r = 1, 2) are
related with the convexity or sub-convexity bound and integral moment of associated L-
functions which is given in next theorem. Let g = f or f ⊗ χD . Suppose α, α′ and α′′
satisfy
X
(15) S1 (f, D; x) = λf (n)rQ (n) ≪ xα+ϵ ,
n≤x
′
(16) |L (1/2 + ϵ + it, g)| ≪f,D,ϵ T α +ϵ
and
Z 2T
1 ′′ +ϵ
(17) |L (1/2 + ϵ + it, g)|2 dt ≪f,D,ϵ T α
T T

for all ϵ > 0 as T → ∞. Similarly, suppose η, η ′ and η ′′ , satisfy

X
(18) S 2 (f, D; x) = (λf (n))2 rQ (n) = Cf x + T η+ϵ ,
n≤x
′
(19) |L (1/2 + ϵ + iT, f ⊗ g)| ≪f,D,ϵ (1 + T )η +ϵ
and
Z 2T
1 ′′ +ϵ
(20) |L (1/2 + ϵ + it, f ⊗ g)|2 dt ≪f,D,ϵ (1 + T )η .
T T

These values of α, α′ and α′′ , and η, η ′ , and η ′′ are related with each others, respectively.
More precisely, we have following relations.
Theorem 1.8. With the above notations, we have the following inequalities:
1
α≤1−
2(1 + max{α′′ , 2α′ − 1})
and
1
η ≤1− .
2(1 + max(η ′′ , 2η ′ − 1))
The proof of this theorem follows exactly same argument as [7, Lemma 5]. We give a brief
proof this theorem in §3.

2. Key ingredients of Proof.

The partial sums defined in (7) can be expressed in terms of known arithmetical functions,
i.e.,
  
X X X
(21) Sr (f, D; x) = (λf (n))r  1 = (λf (n))r rQ (n),
n≤x n=Q(x) n≤x
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 7

where rQ (n) denotes the number of representations of n by the quadratic from Q(x). We
define the following Dirichlet series associated to the sum Sr (f, D; x):
X (λf (n))r rQ (n)
(22) Rr (f, D; s) = , ℜ(s) > 1.
n≥1
ns

In [43], Xu has given the decomposition of Rr (f, D; s) into well-known symmetric power
L-functions when Q(x) = x21 + x22 , (x1 , x2 ) ∈ Z2 (see [43, Lemma 6]). We use the same
decomposition method to obtain the decomposition of Rr (f, D; s) into symmetric power
L-functions. Now, we define the L-functions associated to a normalised Hecke eigenform
∞
k−1
X
f (τ ) = λf (n)n 2 q n ∈ Sk (SL2 (Z)). The Hecke L-function associated to f is given by
n=1

X λf (n) Y  −1  −1

αp βp
(23) L(s, f ) = = 1− s 1− s , ℜ(s) > 1,
n≥1
ns p−prime
p p

where αp +βp = λg (p) and αp βp = 1. The Hecke L-function satisfies a nice functional equation
and it has analytic continuation to whole C-plane [15, Section 7.2]. For a given Dirichlet
character χ of modulus N, we define the twisted Hecke L- function as follows:
X λf (n)χ(n)
(24) L(s, f × χ) = .
n≥1
ns

It also converges absolutely in the half-plane ℜ(s) > 1. The twisted Hecke L-function
∞
k−1
X
L(s, f × χ) is associated to the cusp form fχ (τ ) = λf (n)χ(n)n 2 q n ∈ Sk (Γ0 (N 2 )).
n=1
It also satisfies a nice functional equation and it has analytic continuation to whole C-plane
[15, Section 7.2].
For m ≥ 2, the mth symmetric power L-function is defined as
m ∞
Y Y
−1 X λsymm f (n)
(25) L(s, symm f ) := 1 − αp m−j βp j p−s = ,
p−prime j=0 n=1
ns

where λsymm f (n) is multiplicative arithmetical function. At prime values, it is given by

m
X
(26) λsymm f (p) = αp m−j βp j .
j=0

From Deligne’s bound, we have

|λsymm f (n)| ≤ dm+1 (n) ≪ϵ nϵ
for any real number ϵ > 0 and dm (n) denotes the number of m factors of n.
Similar to the twisted Hecke L-function, for each m ≥ 2, we also define the twisted mth
symmetric power L-functions as follows:
X λsymm f (n)χ(n)
(27) L(s, symm f × χ) = .
n≥1
ns

It also inherits the property similar to the mth symmetric power L-functions.
8 MANISH KUMAR PANDEY AND LALIT VAISHYA

Remark 2.1. For a classical holomorphic Hecke eigenform f , J. Cogdell and P. Michel [5]
have given the explicit description of analytic continuation and functional equation for the
function L(s, symm f ), 2 ≤ m ≤ 4. Newton and Thorne [33, 34] have proved that symm f is
also an automorphic function for more general m ∈ N.

We assume the following convention:

(
L(s, sym0 f ) = ζ(s), L(s, sym0 f × χ) = L(s, χ),
L(s, sym1 f ) = L(s, f ), L(s, sym1 f × χ) = L(s, f × χ),
X X
where ζ(s) = n−s and L(s, χ) = χ(n)n−s are the Riemann zeta function and Dirichlet
n≥1 n≥1
L- function respectively. Now, we state the decomposition of Rr (f, D; s), r ∈ N into well
known L-functions.
Lemma 2.1. Let r ∈ N. we have the following decomposition:
(28) Rr (f, D; s) = Lr (s) × Ur (s),

where L1 (s) = L(s, f )L(s, f × χD ) and L2 (s) = ζ(s)L(s, χD )L(s, sym2 f )L(s, sym2 f × χD ).
and for each r = 2m (m ≥ 2),
m 
r
− r
L(s, symr−2n f )((n) (n−1)) L(s, symr−2n f × χD )((n)−(n−1)) ,
Y r r

Lr (s) =
n=0

and for each r = 2m + 1 (m ≥ 1),

m 
r
− r
L(s, symr−2n f )((n) (n−1)) L(s, symr−2n f × χD )((n)−(n−1)) ,
Y r r

Lr (s) =
n=0
r
and n is the binomial coefficient with the convention nr = 0 if n < 0, and χD is the



Dirichlet character modulo |D| and Ur (s) is a Dirichlet series given by
r ∗
!
Y (λf (p2 ) rQ (p2 ) − A(p2 ))
wD 1+ + ··· ,
p
p2s
1
which converges absolutely and uniformly for ℜ(s) > 2
and Ur (s) ̸= 0 for ℜ(s) = 1.
Before proving above Lemma, we state and give a brief proof of the following result which
explicitly lead to the proof of Lemma 2.1.
Lemma 2.2. [23, Lemma 2.2] Let l ∈ N. For each j with 0 ≤ j ≤ l, we define
   
l l
Al,j := l−j − l−j if j ≡ l (mod 2).
2 2
−1
Then
l
X
l
x = Al,j Tl−2j (x),
j=0

where Tm (x) = Um (2x) and Um (x) is the m-th Chebyshev polynomial of second kind.
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 9

Proof. Here, we give a brief proof. Since, the set of Chebyshev polynomials {Um (x)}m≥0
forms an orthogonal basis for the the set of all polynomial P(x) over R with respect to
scalar product given by
Z 1 √
⟨f, g⟩ := f (x)g(x) 1 − x2 dx.
−1

Hence, any f (x) ∈ P(x) can be expressed uniquely as

∞
X
f (x) = ⟨f, Um (x)⟩Um (x)
m=0

such that ⟨f, Um (x)⟩ = 0 for each m > deg(f ) where deg(f ) is the degree of the polynomial
f (x). We know that
   
l l l
⟨(2x) , Uj (x)⟩ = l−j − l−j if j ≡ l (mod 2).
2 2
− 1
This gives our result. □

2.1. Proof of Lemma 2.1. Let Tn (x) be the polynomial which gives trace of nth symmetric
power of an element of Sl2 (C) whose trace is x [23, Proof of Lemma 2.1]. Let Ur be the
rth Chebyshev polynomial of second kind given by; Ur (cos θ) = sin((r+1)θ)sin(θ)
. Then Tr (x) =
r
Ur (x/2). From Deligne’s estimate, λf (p) = 2 cos θ, and λf (p ) = Tr (2 cos θ) = Ur (cos θ).
Then the expression for λf (p)r is given in term of the j th - symmetric power Fourier coefficient
∗
λsymj f (p) as in [23, proof of Lemma 2.2] using Lemma 2.2. Let us write rQ (n) = rQ (n)/wD =
X
∗
χD (d). Then the function rQ (n) is a multiplicative function. We know that λf (n) is a
d|n
multiplicative function. This gives that the function Rr (f, D; s) is given as an Euler product,
i.e.,
X (λf (n))r rQ∗ ∗ ∗
(n) Y (λf (p))r rQ (p) (λf (p2 ))r rQ (p2 )

Rr (f, D; s) = wD s
= wD 1+ s
+ 2s
+ ··· .
n≥1
n p
p p

We take x = λf (p) and Tm (x) = λf (pm ) = λsymm

f
(p) and apply the Lemma 2.2 to get
 
r/2    
X r r
λf (p)r rQ
∗
(p) = λf (p)r (1 + χD (p)) =  − λsymr−2n (p) (1 + χD (p))
n=0
n n−1 f

r/2     r/2    
X r r X r r
= − λsymr−2n (p) + − λsymr−2n (p)χD (p).
n=0
n n−1 f
n=0
n n − 1 f

Now, for ℜ(s) > 1, we express the function

m 
r
− r
L(s, symr−2n f )((n) (n−1)) L(s, symr−2n f × χD )((n)−(n−1))
Y r r


n=0

as an Euler product of the form

Y A(p) A(p2 )

1 + s + 2s + · · · , where A(p) = λf (p)r rQ
∗
(p).
p
p p
10 MANISH KUMAR PANDEY AND LALIT VAISHYA

Then, we have
m
Y ((nr )−(n−1
r
)) L(s, symr−2n f × χ )((nr )−(n−1
r


 L(s, sym r−2n
f ) ))


 D
n=0
Rr (f, D; s) = r ∗
!
 Y (λf (p2 ) rQ (p2 ) − A(p2 ))
 ×w
 D
 1 + 2s
+ ··· .
p

p

Thus, we have the required result. Similarly, we can prove the result for odd positive integers.
2.2. Sub-convexity bound, convexity bound and mean square integral estimates
of the general L-functions.
X 1
Lemma 2.3. Let ζ(s) = s
be the Riemann zeta function. Then for any ϵ > 0, we have
n≥1
n

ζ(σ + it) ≪ϵ (1 + |t|)max{ 42 (1−σ),0}+ϵ

13
(29)
1
uniformly for 2
≤ σ ≤ 1 and |t| ≥ 1, and
Z T   12
5
(30) ζ + it dt ≪ϵ T 1+ϵ
0 7
uniformly for T ≥ 1.
Proof. An estimate in [13, Theorem 8.4] and [2, Theorem 5] gives the required result. □
X χ(n)
Lemma 2.4. Let L(s, χ) = be the Dirichlet L-function where χ is a Dirichlet
n≥1
ns
character modulo N . Then for any ϵ > 0, we have
1
(31) L(σ + it, χ) ≪ϵ,N (1 + |t|) 3 (1−σ)+ϵ
1
uniformly for 2
≤ σ ≤ 1 and |t| ≥ 1.
Proof. The result follows from [11, eq. (1.1)] and Phragmen-Lindelöf convexity principle. □
Lemma 2.5. For any ϵ > 0, the sub-convexity bound of Hecke L-function is given by:
L(σ + it, f ) ≪f,ϵ (1 + |t|)max{ 3 (1−σ),0}+ϵ
2
(32)
1
uniformly for 2
≤ σ ≤ 1 and |t| ≥ 1, and the integral moments of Hecke L-function is given
by:
Z T   2
1
(33) L + it, f dt ≪f,ϵ T log T
0 2
and
Z T   4
5
(34) L + it, f dt ≪f,ϵ T 1+ϵ
0 8
uniformly for T ≥ 1. The results also hold for f × χD in place of f but the absolute constant
depends on f , ϵ and |D|.
Proof. Proof the sub-convexity bound of Hecke L-function follows from standard argument
of Phragmen - Lindelöf convexity principle and a result of A. Good [10, Corollary]. For the
integral estimate, we refer to [14, Theorem 2]. □
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 11

Lemma 2.6. [35, Corollary 2.1] For any arbitrarily small ϵ > 0, we have

L(σ + it, sym2 f ) ≪f,ϵ (1 + |t|)max{ 4 (1−σ),0}+ϵ

5
(35)
uniformly for 12 ≤ σ ≤ 1 and |t| ≥ 1. The result also holds when sym2 f is twisted by a
Dirichlet character χD and the absolute constant depends on f , ϵ and |D|.
Lemma 2.7. [16, pp. 100] Let L(s, F ) be an L- function of degree m ≥ 2, i.e.,
X λF (n) m  −1
Y Y αp,f,j
(36) L(s, F ) = s
= 1− s ,
n≥1
n p−prime j=1
p

where αp,f,j , 1 ≤ j ≤ m; are the local parameter of L(F, s) at prime p and λF (n) = O(nϵ )
for any ϵ > 0. We assume that the series and Euler product converge absolutely for ℜ(s) > 1
and L(s, F ) is an entire function except possibly for pole at s = 1 of order r and satisfies a
nice functional equation (s → 1 − s). Then for any ϵ > 0, we have
 r
s−1 m
(37) L(σ + it, F ) ≪F,ϵ (1 + |t|) 2 (1−σ)+ϵ
s+1
uniformly for 0 ≤ σ ≤ 1. and |t| ≥ 1 where s = σ + it.
Lemma 2.8. [27, Lemma 2.6] Let L(s, F ) be an L-function of degree m ≥ 2. Then for
T ≥ 1, We have
Z 2T  2
1 m
(38) L + ϵ + it, F dt ≪ T 2 +ϵ .
T 2
P
The partial sum a(n) of arithmetical function a(n) is related with its Dirichlet series
P an n≥x
D(s) = ns
. It is given by following Lemma.
n≥1
an
P
Lemma 2.9. (Truncated Perron’s formula) [31, pp. 67] : Let D(s) = ns
be a
n≥1
Dirichlet series. Let an = O(nϵ ) for an arbitrary small ϵ > 0. Then for a positive non-
integral x, the truncated Perron’ s formula is given by:
Z σ+iT
xs
 σ+ϵ 
X 1 x
(39) an = D(s) ds + O ,
n≤x
2πi σ−iT s T

where σ = ℜ(s) > 1 is a real number. This result also holds when x is a positive integer.
Now, we state a result on the sign change which is a generalisation of the result of Meher
- Murty [30, Theorem 1.1].
Lemma 2.10. [45, Lemma 2.9] Let {a(n)}n∈N be a sequence of real numbers and g : Sr → N
ba a function such that g −1 (n) is a finite set for each n where S is a countable set and r ≥ 1.
Suppose that
α
(1) a(n)
X = O(n ) for each n,
(2) a(g(s)) = O(xβ ),
s∈Sr
g(s)≤x
X
(3) a2 (g(s)) = Cx + O(xγ ).
s∈Sr
g(s)≤x
12 MANISH KUMAR PANDEY AND LALIT VAISHYA

with α, β, γ > 0 and C > 0. Let α+β < 1. Then for any δ satisfying max {α+β, γ} < δ < 1,
the sequence {a(g(s))}s∈Sr has at least one sign change at s ∈ Sr with g(s) ∈ (x, x + xδ ].
Consequently, there are at least x1−δ sign changes for g(s) ∈ (x, 2x], for sufficiently large x.

3. Proof of results.
In order to obtain an estimate of Sr (f, D; x), we apply the truncated Perron’s formula
(Lemma 2.9). We choose ϵ > 0 for which λf (n) ≪ϵ,r nϵ/2r and rQ (n) ≪ϵ nϵ/2 so that we have

(λf (n))r rQ (n) ≪ϵ,r nϵ(1/2+1/2) ≪ϵ,r nϵ ,

for each fixed r ≥ 1. Now, we apply the truncated Perron’s formula (Lemma 2.9) by taking
σ = 1 + ϵ to obtain an estimate for Sr (f, D; x) for each r ∈ N, i.e.,
Z 1+ϵ+iT
xs
 1+2ϵ 
X
r 1 x
(40) Sr (f, D; x) = (λf (n)) rQ (n) = Rr (f, D; s) ds + O ,
n≤x
2πi 1+ϵ−iT s T

where Rr (f, D; s) is given in (22). Now, we shift the line of integration from 1 + ϵ to
σ0 ∈ [1/2 + ϵ, 1 − ϵ) (where ϵ > 0) and apply the Cauchy residue theorem to get:

xs
   1+2ϵ 
x
(41) Sr (f, D; x) = Res Rr (f, D; s) + |Vr | + |Hr | + O ,
s=1 s T

where the first term of right-hand side does not appear when r is odd. In this case (r-odd),
the function Rr (f, D; s) does not have a pole at s = 1 and
Z T1 !
1 1
|Vr | ≪f,ϵ xσ0 + xσ0 max |Lr ( + ϵ + it)|dt
2≤T1 ≤T T1 T21 2
(42)  σ 
x
and |Hr | ≪ max |Lr (σ + iT )| .
σ0 ≤σ≤1+ϵ T

For details, we refers to [45, 46]. we apply the convexity or sub-convexity bound and mean
square integral estimate for the L-functions to obtain an estimate of Sr (f, D; x).
Proof of Theorem 1.1
Case r = 1 :
From Lemma 2.1, we have L1 (s) = L(s, f )L(s, f × χD ). L1 (s) does not have a pole as
L(s, f ) and L(s, f ×χD ) are entire functions. Now, we substitute L1 (s) = L(s, f )L(s, f ×χD )
to obtain an estimate for H1 and V1 which gives the required estimate for S1 (f, D; x), i.e.,
7
S1 (f, D; x) ≪ x 10 +2ϵ .

This is obtained in [45].

Case r = 2 :

L2 (s) = ζ(s)L(s, χD )L(s, sym2 f )L(s, sym2 f × χD ).

 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 13

We have
xσ
 
2 2
|H2 | ≪ max |ζ(σ + iT )L(σ + iT, χD )L(σ + iT, sym f )L(σ + iT, sym f × χD )|
1
2
+ϵ≤σ≤1+ϵ T
x1+ϵ xσ
≪ + ζ(σ + iT )L(σ + iT, χD )L(σ + iT, sym2 f )L(σ + iT, sym2 f × χD )
T T σ= 21 +ϵ
1
!
x1+ϵ x 2 +ϵ ( 13 + 1 +ϵ)+2×( 5 +ϵ) x1+ϵ 1
≪ + T 84 6 8 ≪ + x 2 +ϵ T 3/7+ϵ
T T T
and
Z  T  
1
+ϵ 1
+2ϵ 1 1
|V2 | ≪ x 2 +x 2 L2 + ϵ + it dt
1 t 2
 
R T1  12
1 2 2
T1 |ζ(1/2 + ϵ + it)L(1/2 + ϵ + it, sym f )| dt


1 1 T1
V2 ≪ x 2 +ϵ + x 2 +2ϵ max R 2 1
2≤T1 ≤T 
× TT11 |L(1/2 + ϵ + it, χD )L(1/2 + ϵ + it, sym2 f × χD )|2 dt 2
2
 ! 12


1 R T 2


 T1 T1
max |ζ(1/2 + ϵ + it)|2 T11 |L(1/2 + ϵ + it, sym2 f )| dt ×
1 1 ≤t≤T1 2
V2 ≪ x 2 +ϵ + x 2 +2ϵ max 2
! 21
2≤T1 ≤T
 1 2 T1
R 1 2 2
 T1max |L( 2 + ϵ + it, χD )| T21 L( 2 + ϵ + it, sym f × χD ) dt .



2
≤t≤T1

From Lemmas 2.3, 2.4, 2.5, 2.7 and 2.8, we have

 
1 13 3  12  2 3  12 
1
+ϵ 1
+2ϵ +ϵ +ϵ 2( +ϵ) +ϵ
V2 ≪ x 2 + x 2 max T 42 × T12 × T1 6 × T12
2≤T1 ≤T T1 1
1 1 23 1 23
≪ x 2 +ϵ + x 2 +2ϵ T 28 +ϵ ≪ x 2 +2ϵ T 28 +ϵ .
So, we have
x1+ϵ
   1+2ϵ 
1
+ϵ 3
+ϵ
 1
+2ϵ 23
+ϵ
 x
(43) S2 (f, D; x) = Cx + O x 2 T 7 + + O x 2 T 28 +O .
T T
14
Now, we choose T = x 51 to get
 37 
S2 (f, D; x) = Cx + O x 51 +2ϵ .
Since ϵ > 0 is arbitrary, we have the required result. This completes the proof of Theorem 1.1.

Proof of Theorem 1.2:

Following the argument of Duda [7], we obtain this result as an application of Rankin-
Selberg convolution technique and a method of Landau. We briefly, introduce Rankin-
Selberg convolution technique. It is known that the following Dirichlet series
∞
X λf (n)rQ (n)
R1 (f, D, s) := appear as Rankin-Selberg convolution of f and θQ , i.e.,
n=1
ns
Z Z
k+1 k+1 dxdy
⟨y 2 f (τ )θQ (τ ), Ek+1 (τ, s)⟩ = y 2 f (τ )θQ (τ )Ek+1 (τ, s) 2
Γ∞ \Γ0 (|D|) y
k−1 X ∞
Γ(s + 2 ) λf (n)rQ (n)
= ,
s+ k−1 ns
4π 2 n=1
14 MANISH KUMAR PANDEY AND LALIT VAISHYA

where Ek (τ, s) denotes the weight k real-analytic Esienstein series for the congruence sub-
group Γ0 (|D|) given by;
X
Ek (τ, s) = χD (γ)J(γ, τ )k ℑ(γτ )s
γ∈Γ∞ \Γ0 (|D|)
 
a b
and for a γ = ∈ Γ0 (|D|), the normalised automorphy factor J(γ, τ ) is given by
c d
cτ +d
J(γ, τ ) = |cτ +d|
and χD (γ) = χD (d). The completed Eisenstein series is defined as:
 s/2  
∗ 4 s+1
Ek (τ, s) = Γ L(s, χD )Ek (τ, s).
π 2
The completed Eisenstein series Ek∗ (τ, s) is an entire function and satisfies a nice functional
equation Ek∗ (τ, 1 − s) = ηEk∗ (τ, s), where η is a root number with |η| = 1. For detail, we refer
to [36].
Using these notation, we define the completed Rankin-Selberg L-function
 s/2  ∞
Γ(s + k−1

k−1 4 s+1 2
) X λf (n)rQ (n)
Λ(f ⊗ θQ , s) = (4π) 2 Γ L(2s, χD ) k−1
π 2 (4π)s+ 2 n=1 ns
    ∞
1 1 k−1 X λf (n)rQ (n)
= 2s Γ s + Γ s+ L(2s, χD ) .
π 2 2 n=1
ns
It inherits a nice functional equation Λ(f ⊗ θQ , 1 − s) = Λ(f ⊗ θQ , s) from the functional
equation of Ek∗ (τ, s). Now we define the arithmetical function B(n) as follows:
∞ ∞
X B(n) X λf (n)rQ (n)
s
:= L(2s, χD ) .
n=1
n n=1
ns
Then, B(n) is given by
X
(44) B(n) = χD (d2 )λf (n/d2 )rQ (n/d2 ).
d2 |n

From the Deligne bound λf (n) ≪ nϵ and Weil bound rQ (n) ≪ nϵ , we get that B(n) ≪ nϵ .
We apply Landau’s method (as applicable to general Dirichlet series with functional equation
in [4] ) to get a bound on partial sum of
B(n). In the view of notation of [4], we have δ = 1,
A = 2, and ∆(s) = Γ s + 21 Γ s + k−1


2
which gives that
3 3 3 3
X X
B(n) = |B(n)| + X 2 +ϵ y − 2 ≪ y 1+ϵ + X 2 +ϵ y − 2 .
n≤X X<n<X+O(y)
3
By choosing y = X , we have5

3
X
B(n) ≪ X 5 +ϵ .
n≤X

Now we apply following fact ( a variant of mobius inversion formula (Let k be a fixed natu-
ral number. Suppose f (n) and X
g(n) are two arithmetical functions andX χ be a Dirichlet char-
k k
acter modulo N , Then f (n) = χ(d )g(n/d ), if, and only if g(n) = µ(d)χ(dk )f (n/dk ).)
dk |n dk |n
on (44) to get
X
λf (n)rQ (n) := µ(d)χD (d2 )B(n/d2 ).
d2 |n
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 15

Thus, we have
X XX X X
S1 (f, D, x) = λf (n)rQ (n) = µ(d)χD (d2 )B(n/d2 ) = µ(d)χD (d2 ) B(m)
√
n≤x n≤x d2 |n d≤ x m≤ x2
d
X  x 3/5+ϵ X 1
= µ(d)χD (d2 ) = x3/5+ϵ µ(d)χD (d2 ) 6
√ d2 √ d 5
+2ϵ
d≤ x d≤ x
 
6
≪ζ + 2ϵ x3/5+ϵ ≪ x3/5+ϵ
5

as the last sum is absolutely bounded by ζ( 65 + 2ϵ). This completes the proof of this result.
Proof of Theorem 1.3: Under the assumption of Lindelöf hypothesis, we have α′ = α′′ = 0
and β ′ = β ′′ = 0 appearing in §1.1. From Theorem 1.8, we have α ≤ 21 and β ≤ 12 . This gives

S1 (f, D; x) = Of,D,ϵ (x1/2+ϵ ) and S2 (f, D; x) = Cf x + Of,D,ϵ (x1/2+ϵ ).

3.1. Proof of Theorem 1.4. We prove the estimates of this theorem using the arguments
given in section §3.
Case (a) : For r = 2m (m ≥ 2),

m 
r
− r
L(s, symr−2n f )((n) (n−1)) L(s, symr−2n f × χD )((n)−(n−1)) .
Y r r

Lr (s) =
n=0

1
Let σ0 = 2
+ ϵ.

xσ
 
|Hr | ≪ max |Lr (σ + iT )|
σ0 ≤σ≤1+ϵ T
m
!
xσ Y  r r
L(σ + iT, symr−2n f )((n) (n−1)) L(σ + iT, symr−2n f × χD )((n)−(n−1))
r r

−
≪ max
σ0 ≤σ≤1+ϵ T n=0
 σ
x


 T
|L(σ + iT, symr f )L(σ + iT, symr f × χD )|
r r r r
×|L(σ + iT, χD )((m)−(m−1)) ζ(σ + iT )((m)−(m−1)) |



 r r r r
≪ max ×| (L(σ + iT, sym2 f ))((m−1)−(m−2)) (L(σ + iT, sym2 f × χD ))((m−1)−(m−2)) |
σ0 ≤σ≤1+ϵ
 m−2


 ×
Y
L(σ + iT, sym r−2n
f ) ((nr )−(n−1
r
)) L(σ + iT, symr−2n f × χ )((nr )−(n−1
r
)) .


 D
n=1

Now, we apply the convexity bound and subconvexity bound to get

x1+2ϵ xσ0 (1−σ0 )(θr,m +θr,m,χ ) x1+2ϵ

|Hr | ≪ + ×T ≪
T T T
1
provided T ≤ x θr,m +θr,m,χ , where θr,m and θr,m,χ are given in Theorem 1.4.
16 MANISH KUMAR PANDEY AND LALIT VAISHYA

!
Z T1
σ0 σ0 1
|Vr | ≪f,ϵ x +x max |Lr (σ0 + it)|dt
2≤T1 ≤T T1 T1
2
m
!
Z T1 Y
σ0 σ0 1 r−2n r−2n

((nr )−(n−1
r
))
≪f,ϵ x +x max L(σ0 + it, sym f )L(σ0 + it, sym f × χD ) dt
2≤T1 ≤T T1 T1
2 n=0

m
! 12
r r
2((n )−(n−1))
 R T1 Y
1
L(σ0 + it, symr−2n f )

dt

T1

 T1

2
σ0 σ0 n=0
≪f,ϵ x +x max ! 12
2≤T1 ≤T m r r
2((n )−(n−1))

 R T1 Y
r−2n
× L(σ0 + it, sym f × χD ) dt .


 T1
2
n=0

Let us denote the function inside the the max function by A(T ). Then
 
  12
1  r r 2 r
− r
((m−1) (m−2)) 
|A(T1 )| = sup |ζ(σ0 + it)|2((m)−(m−1)) L(σ0 + it, sym2 f )
T1 T1 ≤t≤T1
2

! 21
Z T1 m−2 r r
Y
r−2n 2((n )−(n−1))
× L(σ0 + it, sym f) dt
T1
2 n=0
 
 r r
1/2
×  sup |L(σ0 + it, χD )|2(( )−(
r
m
r
m−1 )) L(σ + it, sym2 f × χ ) 2(( m−1 )−(m−2 ))
0 D

T1
2 ≤t≤T1
! 12
Z T1 m−2 r r
Y 2((n )−(n−1))
× L(σ0 + it, symr−2n f × χD ) dt
T1
2 n=0
xσ0 (1−σ0 )(θr,m +θr,m,χ )
≪ T1 ,
T1
where θr,m and θr,m,χ are given in Theorem 1.4. Thus, we have
xσ0 (1−σ0 )(θr,m +θr,m,χ )
|Vr | ≪f,ϵ xσ0 + T .
T
Hence, we have
x1+2ϵ xσ0 (1−σ0 )(θr,m +θr,m,χ ) x1+2ϵ
 
Sr (f, D; x) = xPr (log x) + O + xσ 0 + T + ,
T T T
1
Now, we choose T = x θr,m +θr,m,χ to get required estimate of Sr (f, D; x) when r is even.
Case ( b) : when r = 2m + 1.
m 
r
− r
L(s, symr−2n f )((n) (n−1)) L(s, symr−2n f × χD )((n)−(n−1)) ,
Y r r

Lr (s) =
n=0

Take σ0 = 5/8. Then

xσ
 
|Hr | ≪ max |Lr (σ + iT )|
σ0 ≤σ≤1+ϵ T
m  !
xσ Y r
r−2n ((n)−(n−1))
r
r−2n (( r
) r
−(n−1 ))
≪ max L(σ0 + iT, sym f) L(σ0 + iT, sym f × χD ) n
σ0 ≤σ≤1+ϵ T n=0
m−1 
x1+2ϵ xσ0 Y

(( r )−( r )) r r
≪ + L(σ0 + iT, symr−2n f ) n n−1 L(σ0 + iT, symr−2n f × χD )((n)−(n−1))
T T n=0
r
− r r
− r
× L(σ0 + iT, f )((m) (m−1)) (L(σ0 + iT, f × χD ))((m−1) (m−2)) .
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 17

Now, we apply the convexity bound and subconvexity bound to get

x1+2ϵ xσ0 x1+2ϵ
|Hr | ≪
+ × T (1−σ0 )αr,m ≪
T T T
"m−1 #
1 X (r − 2n + 1)2  r 
8 r


provided T ≤ x αr,m , where αr,m = 3m m−1
+ + r + 1 . For the
n=1
n n−1
estimate of Vr (r is odd), we follow the same process in the case when r is even. We get
  !1/2 
m−1
r r

|L(σ0 + it, symr−2n f )|2((n)−(n−1))
Y
1 



 T1 sup 
T1



 2 ≤t≤T1 n=0

 R r r
1
 TT11 |L(σ0 + it, f )|2((n)−(n−1)) dt 2


|Vr | ≪f,ϵ xσ0 + xσ0 max 2 !1/2 
2≤T1 ≤T  m−1
r r
|L(σ0 + it, symr−2n f × χD )|2((n)−(n−1))
 Y
×  sup

 

 T1
≤t≤T


 2 1 n=0

 1
R T1 r r
2((n )−(n−1)) dt 2 .

T1 |L(σ0 + it, f × χD )|

2

Let Ir denote the term inside max. Then, we have

  !1/2 
 m−1
r r
|L(σ0 + it, symr−2n f )|2((n)−(n−1))
 Y
1 
sup

 

 T1
T1
  2 ≤t≤T1 n=0




 1/2

  r r
 R  12
|L(σ0 + it, f )| ((m) (m−1) )  × T11 |L(σ0 + it, f )| dt
 2 − −2 T 4
×  sup


 T1 2
≤t≤T1

Ir = 2 !1/2 
 m−1
r r
|L(σ0 + it, symr−2n f × χD )|2((n)−(n−1))
 Y
×  sup

 

 T1
≤t≤T


 2 1 n=0

  1/2


  r r
 R  12
2((m)−(m−1)−2)  T1 4

×  sup

 |L(σ 0 + it, f × χ D )| × T 1 |L(σ0 + it, f × χ D )| dt .
 T1 2
≤t≤T

2 1

Now, we apply the convexity bound or sub-convexity bound and fourth integral moment of
Hecke L-function to get
 "m−2 #
X (r − 2n + 1)2  r 
r
8  3m (m−1)− 3 +
3 8 8
+r+1 
n n − 1

Ir ≪f,ϵ T n=1 .

Now, we substitute the value of Ir to get

 "m−2 #
X (r − 2n + 1)2  r 
3 8
(r
)+ +r+1 
xσ0 8  3m m−1
n n − 1

|Vr | ≪f,ϵ T n=1 .
T
"m−2 #!
X (r − 2n + 1)2  r 
8 r


Now, we choose T ≤ x1/αr,m ( where αr,m = 3m m−1
+ +r+1 )
n=1
n n−1
to get the required estimate, i.e.,

Sr (f, D; x) = O x1−1/αr,m +ϵ .

Since ϵ > 0 is arbitrary, we have the required result. This completes the proof.

Now, we apply Lemma 2.10 to get our result on sign change.

18 MANISH KUMAR PANDEY AND LALIT VAISHYA

Proof of Theorem 1.6 Let ϵ > 0 be a fixed arbitrary small real number. We as-
sume that λf (n) = O(nϵ/2 ) (Deligne’s bound) and rQ (n) = O(nϵ/2 ) (Weil’s bound). From
Theorem 1.1 and Theorem 1.2 and Lemma 2.10, we have
3 ϵ
X
S1 (f, D; x) = λf (Q(x)) = Of,D,ϵ (x 5 + 2 )
x∈Z2
Q(x)≤x
37
X
and S2 (f, D; x) (λf (Q(x))2 = Cx + Of,D,ϵ (x 51 +ϵ ),
x∈Z2
Q(x)≤x

where C > 0 is a constant. Now, we chose δ > 37 51

+ ϵ and apply Lemma 2.10 (for r = 2,
S = Z and g = Q(x), where Q(x) is a primitive integral positive definite binary quadratic
form of fixed discriminant D < 0 with the class number 1) to obtain at least one sign change
at x ∈ Z2 with Q(x) ∈ (x, x + xδ ]. Thus, we have infinitely many sign changes. Moreover,
we obtain at least x1−δ many sign changes in the interval (x, 2x], for sufficiently large x.
27
Take δ = 10275
+ ϵ, Then we have at least x 102 −ϵ many sign changes in the interval (x, 2x], for
sufficiently large x. This completes the proof.

Proof of Theorem 1.7: From Theorem 1.3, under the assumption of Lindelöf hypoth-
esis, we have
S1 (f, D; x) = Of,D,ϵ (x1/2+ϵ ) and S2 (f, D; x) = C ′ x + Of,D,ϵ (x1/2+ϵ ),
where C ′ > 0 is constant depends on f, ϵ, D. Moreover, we have λf (n) ≪ nϵ (Ramanujan-
1
Petersson bound). Now, we apply Lemma 2.10 by taking δ = 21 + ϵ to get x 2 −ϵ many sign
changes in the interval (x, 2x], for sufficiently large x. This completes the proof.

Proof of Theorem 1.8:

From (41), we have
xs x1+2ϵ
   
Sr (f, D; x) = Res Rr (f, D; s) + |Vr | + |Hr | + O ,
s=1 s T
where
 1/2
R
1 T1
|Lr,1 ( 12 + ϵ + it)|2 dt

T1

T1
|Vr | ≪f,ϵ xσ0 + xσ0 max R 2
1/2
2≤T1 ≤T  T
 × T11 |Lr,2 ( 21 + ϵ + it)|2 dt
 σ  2
x
and |Hr | ≪ max |Lr (σ + iT )| .
σ0 ≤σ≤1+ϵ T
The estimates for |H1 | and |V1 | are as follows:
1
x1+ϵ x 2 +ϵ 2α′ +ϵ
 σ 
x
H1 ≪ 1 max |L(σ + iT, f )L(σ + iT, f ⊗ χD )| ≪ + T ,
2
+ϵ≤σ≤1+ϵ T T T
 1/2
R T1 1 2
|L( + ϵ + it, f )| dt

1  T 1 2
|V1 | ≪f,ϵ x1/2+ϵ + x1/2+ϵ max R 2 1/2
2≤T1 ≤T T1 
 T1 |L( 1 + ϵ + it, f ⊗ χ )|2 dt T1
2 D
2
 ′′ +ϵ

≪f,ϵ x1/2+ϵ + x1/2+ϵ max T1α
2≤T1 ≤T
1/2+ϵ 1/2+ϵ ′′ +ϵ
≪f,ϵ x +x Tα .
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 19

The estimates are obtained by using (15) and (16). Thus, we have
1
1/2+ϵ 1/2+ϵ α′′ +ϵ x1+ϵ x 2 +ϵ 2α′ +ϵ x1+2ϵ
S1 (f, D; x) ≪f,ϵ x +x T + + T +
T T T
1 x1+2ϵ
≪f,ϵ x 2 +ϵ T λ+ϵ + ,
T
1
where λ = max{α′′ , 2α′ − 1}. Now, chose T = x 2(1+λ) to get
1
S1 (f, D; x) ≪f,ϵ x1− 2(1+λ) +ϵ .

We compare this estimate with (15) to get

1
α≤1− .
2(1 + max{α′′ , 2α′ − 1})
The estimates for |H2 | and |V2 | are as follows which are obtained by using (18) and (19):
1
x1+ϵ x 2 +ϵ 2η′ +ϵ
 σ 
x
H2 ≪ 1 max |L(σ + iT, f ⊗ f )L(σ + iT, f ⊗ f ⊗ χD )| ≪ + T
2
+ϵ≤σ≤1+ϵ T T T

and
 R 1/2
1 1 T1 2
|L( + ϵ + it, f ⊗ f )| dt

T1

2 T1
|V2 | ≪f,ϵ x1/2+ϵ + x1/2+ϵ max R 2 1/2
2≤T1 ≤T 
 TT11 |L( 1 + ϵ + it, f ⊗ f ⊗ χD )|2 dt
2
 ′′ 2 
≪f,ϵ x1/2+ϵ + x1/2+ϵ max T1η +ϵ
2≤T1 ≤T
1/2+ϵ 1/2+ϵ ′′ +ϵ
≪f,ϵ x +x Tη .

Thus, we have
1
!
η ′′ +ϵ x1+ϵ x 2 +ϵ 2η′ +ϵ x1+2ϵ
S2 (f, D; x) = C ′ x + O x1/2+ϵ + x1/2+ϵ T + + T + .
T T T
x1+2ϵ
 
1
′ +ϵ λ+ϵ
=C x+O x 2 T + ,
T
1
where λ′ = max{η ′′ , 2η ′ − 1} and C ′ > 0 is some constant. Now, we chose T = x 2(1+λ′ ) to get
1
1− 2(1+λ′ ) +ϵ
S2 (f, D; x) = C ′ x + x .

We compare this estimate with (18) to get

1
η ≤1− .
2(1 + max{η ′′ , 2η ′ − 1})
Acknowledgement : We would like to thank Prof. B. Ramakrishnan for encouraging to
work on this problem and also for his guidance and suggestions during the preparation of the
manuscript. The first author thanks SRM, Amarawati. The second author thank NISER,
Bhubaneswar for providing financial support through an institute fellowship.
20 MANISH KUMAR PANDEY AND LALIT VAISHYA

References
[1] S. Banerjee and M. K. Pandey, Signs of Fourier coefficients of cusp form at sum of two squares, Proc.
Indian Acad. Sci. Math. Sci., Vol 130, no 2 (2020). DOI: https://doi.org/10.1007/s12044-019-0534-4.
[2] J. Bourgain, Decoupling, exponential sums and the Riemann zeta function, J. Amer. Math. Soc. (2017),
205–224.
[3] D. A. Buell, Binary Quadratic Forms: Classical Theory and Modern Computations, springer - Verlag,
New york, 1989.
[4] K. Chandrasekharan and Raghavan Narasimhan. Functional equations with multiple gamma factors and
the average order of arithmetical functions. Ann. of Math. (2), 1962, 93–136.
[5] J. Cogdell and P. Michel, On the complex moments of symmetric power L-functions at s = 1, Int. Math.
Res. Not. 31 (2004), 1561–1617.
[6] D. A. Cox, Primes of the form x2 + ny 2 : Fermat, Class Field Theory, and Complex Multiplication, Pure
and Applied Mathematics: A Wiley Series of Texts, Monographs and Tracts 119, John Wiley & Sons,
2014.
[7] David Lowry-Duda, Sign changes of cusp form coefficients on indices that are sums of two squares,
Preprint 2021. https://arxiv.org/abs/2108.12520
[8] O. M. Fomenko, Fourier coefficients of parabolic forms and automorphic L-functions, J. Math. Sci. (N.
Y.) 95 (3) (1999), 2295–2316.
[9] S. Gelbart and H. Jacquet,A relation between automorphic representations of GL(2) and GL(3), Ann.
Sci. École Norm. Sup. 11 (1978), 471–552.
[10] A. Good, The square mean of Dirichlet series associated with cusp forms, Mathematika 29 (2) (1982),
278–295.
[11] D. R. Heath-Brown, The growth rate of the Dedekind zeta-function on the critical line, Acta Arith., 49
(4) (1988), 323–339.
[12] B. Huang, On the Rankin-Selberg Problem, Preprint-2021. https://arxiv.org/abs/2002.00591
[13] [7] A. Ivı́c, Exponent pairs and the zeta function of Riemann, Studia Sci. Math. Hungar. (1980), 157–181.
[14] [8] A. Ivı́c, On zeta-functions associated with Fourier coefficients of cusp forms, Proceedings of the
Amalfi Conference on Analytic Number Theory (Maiori, 1989), 231–246.
[15] H. Iwaniec, Topics in Classical Automorphic Forms, Grad. Stud. Math., vol. 17, American Mathemat-
ical Society, Providence, RI, 1997.
[16] H. Iwaniec and E. Kowalski, Analytic Number Theory, Amer. Math. Soc. Colloq. Publ., vol. 53,
American Mathematical Society, Providence, RI, 2004.
[17] H. H. Kim, Functoriality for the exterior square of GL(4) and symmetric fourth of GL(2), Appendix 1
by D. Ramakrishnan, Appendix 2 by H. Kim and P. Sarnak, J. Amer. Math. Soc. 16 (2003), 139–183.
[18] H. H. Kim and F. Shahidi, Functorial products for GL(2) × GL(3) and functorial symmetric cube for
GL(2) (with an appendix by C.J. Bushnell and G. Henniart), Ann. of Math. 155 (2002), 837–893.
[19] H. H. Kim and F. Shahidi, Cuspidality of symmetric power with applications, Duke Math. J. 112 (2002),
177–197.
[20] Winfried Kohnen and Yves Martin. Sign changes of Fourier coefficients of cusp forms supported on
prime power indices, Int. J. Number Theory, 10(8), 2014, 1921–1927.
[21] C. I. Kuan, Hybrid bounds on twisted L-functions associated with modular forms, J. Number Theory
(2018), 380–416.
[22] Y. K. Lau and G. Lü, Sums of Fourier coefficients of cusp forms, Q. J. Math. 62 (3) (2011), 687–716.
[23] Y. K. Lau, G. Lü and J. Wu, Integral power sums of Hecke eigenvalues, Acta Arith. 150 (2) (2011),
193 – 207.
[24] Y. K. Lau and J. Wu, A density theorem on automorphic L-functions and some applications, Trans.
Amer. Math. Soc. 359 (2006), 441–472.
[25] H. Lao and A. Sankaranarayanan, The average behaviour of Fourier coefficients of cusp forms over
sparse sequences, Proc. Amer. Math. Soc. 8 (2009), 2557-2565.
[26] G. Lü, Average behaviour of Fourier coefficients of cusp forms, Proc. Amer. Math. Soc. 137 (6) (2009),
1961–1969.
[27] G. Lü, The sixth and eighth moments of Fourier coefficients of cusp forms, J. Number Theory 129 (11)
(2009), 2790–2800.
[28] G. Lü, On higher moments of Fourier coefficients of holomorphic cusp forms, Canad. J. Math. 63 (3)
(2011), 634–647.
 GENERAL ASYMPTOTIC AND SIGN CHANGE OF FOURIER COEFFICIENTS 21

[29] C. J. Moreno and F. Shahidi, The fourth moment of the Ramanujan τ -function, Math. Ann. 266 (1983),
233–239.
[30] J. Meher and M. R. Murty, Sign change of Fourier coefficient of half-integral weight cusp forms, Int. J.
Number Theory 10 (2014), 905-914.
[31] M. R. Murty, Problems in Analytic Number Theory, Graduate Texts in Mathematics, Readings in
Mathematics, 2nd ed. Springer, New York, vol. 206, 2008.
[32] M. R. Murty, Oscillations of the Fourier coefficients of Modular forms, Math. Ann., 262 (1983), 431-446.
[33] J. Newton and J.A. Thorne, Symmetric power functoriality for holomorphic modular forms, Publ.
math. IHES 134 (2021), 1–116. https://doi.org/10.1007/s10240-021-00127-3
[34] J. Newton and J.A. Thorne, Symmetric power functoriality for holomorphic modular forms, II, Publ.
math. IHES 134 (2021), 117–152. https://doi.org/10.1007/s10240-021-00126-4
[35] R. M. Nunes, Subconvexity for GL(3) L-functions. arXiv e-prints, 2017. arXiv:1703.04424.
[36] Cormac O’Sullivan, Formulas for non-holomorphic Eisenstein series and for the Riemann zeta function
at odd integers, Res. Number Theory, 4(3), Paper No. 36, 38, 2018.
[37] A. Perelli, General L-functions, Ann. Mat. Pura Appl. 4 (1982), 287–306.
[38] R. A. Rankin, Sums of cusp form coefficients, Automorphic forms and analytic number theory (Montreal,
PQ, 1989), 115-121, Univ. Montreal, QC, (1990).
[39] R. A. Rankin, Contributions to the theory of Ramanujan’s function τ (n) and similar arithmetical func-
tions II. The order of the Fourier coefficients of the integral modular forms, Math. Proc. Cambridge
Philos. Soc. 35 (1939), 357–372.
[40] A. Selberg, Bemerkungen über eine Dirichletsche Reihe, die mit der Theorie der Modulformen nahe
verbunden ist, Arch. Math. Naturvid. 43 (1940), 47–50.
[41] F. Shahidi, Third symmetric power L-functions for GL(2), Compos. Math. 70 (1989) 245–273.
[42] E. C. Titchmarsh, The Theory of the Riemann Zeta-function, Oxford University Press, Oxford (1953).
[43] C. Xu, General asymptotic formula of Fourier coefficients of cusp forms over sum of two squares, J.
Number Theory (2021). https://doi.org/10.1016/j.jnt.2021.07.017
[44] Lalit Vaishya, Signs of Fourier coefficients of cusp forms at integers represented by an integral binary
quadratic form, Proc. Indian Acad. Sci.( Math. Sci.), Vol 131, No. 41 (2021).
https://doi.org/10.1007/s12044-021-00630-x
[45] Lalit Vaishya, Average estimates and sign change of Fourier coefficients of Hecke eigenforms at integers
represented by binary quadratic forms of fixed discriminant, Submitted (2021).
[46] Lalit Vaishya, Some problems on sign change and shifted convolution sums of Fourier coefficients of
certain automorphic forms, Ph.D. thesis, Homi Bhabha National Institute, India, 2021.
[47] S. Zhai, Average behaviour of Fourier coefficients of cusp forms over sum of two squares, Journal of
Number Theory, 133 (2013), 3862–3876.

SRM University AP, Andhra Pradesh, India.

Email address: [email protected], [email protected]

National Institute of Science Education and Research, HBNI, Bhubaneswar -752050 Odisha,
India.
Email address: [email protected], [email protected]