This shows that e−2πizκρ/tρ×f|kγρhas a Fourier expansion at i∞with period tρas a function of x. Using Lemma 8, this can be written by
n−∞
∑
c+f(n)qn/tρ+c−f(0)y1−k+
∑
n∞n6=0
c−f (n)Γ(1−k,−4πny/tρ)qn/tρ.
Multiplying e−2πizκρ/tρ =qκρ/tρ both sides, we have a desired result.
Lemma 10 Let f ∈Hk!(N,χ)andα∈GL+(2,Q). Then f|kα satisfies similar growth condition as f at every cusp.
Proof 20 Letρ∈Q∪ {i∞}be a cusp ofΓ0(N)andβ ∈SL(2,Z)such thatβ(i∞) =ρ. Without loss of generality, we assume thatα β= p q
r s
!
has integer entries; if not, multiply a suitable integer toα β without affecting f|kα β. Let p=ap0and r=ar0where a= (p,r). Then since(p0,r0) =1, there are two integers x,y such that xp0+yr0=1and we obtain x y
−r0 p0
! p q r s
!
= a b
0 d
!
where b=xq+ys and d=−r0q+p0s are both integers. Since xp0+yr0=1, we observe that x y
−r0 p0
!
∈SL(2,Z).
Denote the inverse of this matrix byγ. Then we can write α β=γ
a b 0 d
! .
Now using(51)with width tρand cusp parameterκρ, we obtain f|kα β(z) =f|kγ
a b 0 d
!
(z) = (f|kγ) k
a b 0 d
! (z)
=
∑
n−∞
c+f(n)q
n+κρ
tρ +c−f(0)y1−kq
κρ
tρ +
∑
n∞n6=0
c−f(n)Γ(1−k,−4πny/tρ)q
n+κρ tρ
k
a b 0 d
! (z)
=e2πi
bκρ dtρ
a d
k/2
∑
n−∞
d+f (n)q
a(n+κρ)
dtρ +
a d
1−k
c−f(0)y1−kq
aκρ
dtρ +
∑
n∞n6=0
d−f (n)Γ(1−k,−4πany/d)q
a(n+κρ) dtρ
,
where d±f (n) =e2πibn/dtρc±f (n). Since a,d>0, the lemma follows.
(2) The weight k hyperbolic Laplacian operator can be express as
−∆k=Lk+2◦Rk+k=Rk−2◦Lk. (3) If f is an eigenfunction of∆kwith eigenvalueλ, then
∆k+2(Rk(f)) = (λ+k)Rk(f), ∆k−2(Lk(f)) = (λ−k+2)Lk(f).
Proof 21
(1) Letα= a b c d
!
∈SL(2,Z), and denote f0=∂f
∂z. Then we have Rk(f|kα)(z) =2i∂
∂z((cz+d)−kf(αz)) +k
y((cz+d)−kf(αz))
=2i
(cz+d)−k−2f0(αz)−ck(cz+d)−k−1f(αz)
+k
y((cz+d)−kf(αz))
=2i(cz+d)−(k+2)f0(αz) + (cz+d)−(k+2)f(αz)
−2ick(cz+d) +k
y(cz+d)2
=2i(cz+d)−(k+2)f0(αz) + (cz+d)−(k+2)f(αz)k
y(cz+d)(cz+d)
=2i(cz+d)−(k+2)f0(αz) + (cz+d)−(k+2)f(αz)k
y|cz+d|2. On the other hand,
(Rk(f)|k+2α)(z) =2i(f0|k+2α)(z) + k
Im(z)f k+2
α
(z)
=2i(cz+d)−(k+2)f0(αz) + (cz+d)−(k+2) k
Im(αz)f(αz)
=2i(cz+d)−(k+2)f0(αz) + (cz+d)−(k+2) k
Im(z)|cz+d|2f(αz).
Comparing two equations, we see that Rk(f|kα) =Rk(f)|k+2α.
Next, note that f is also dependent on z and so we will abuse the notation of f(z) as f(z,z).
Then we have
Lk(f|kα)(z) =−2iy2 ∂
∂z((cz+d)−kf(αz))
=−2iy2
f(αz)∂
∂z(cz+d)−k+ (cz+d)−k ∂
∂zf(αz)
=−2iy2(cz+d)−k ∂
∂zf(αz,αz)
=−2iy2(cz+d)−k∂f
∂z(αz)(cz+d)−2.
On the other hand,
(Lk(f)|k−2α)(z) =−2i(cz+d)−(k−2)(Im(αz))2∂f
∂z(αz)
=−2i(cz+d)−(k−2)
Im(z)
|cz+d|2 2
∂f
∂z(αz)
=−2i(cz+d)−k(Im(z))2 (cz+d)2
∂f
∂z(αz).
Comparing two equations, we see that Lk(f|kα) =Lk(f)|k−2α.
Finally, by using (2) which we will prove below, we have
∆k(f|kα) = (Rk−2◦Lk)(f|kα) =Rk−2(Lk(f)|k−2α) = (Rk−2◦Lk)(f)|kα=∆k(f)|kα. (2) By definition, we obtain
(Lk+2◦Rk+k)(f) = (Lk+2◦Rk)(f) +k f =
−2iy2 ∂
∂z 2i∂f
∂z +k yf
+k f
=4y2∂
∂z
∂
∂zf−2iy2 ∂
∂z k
yf
+k f
=4y2∂
∂z
∂
∂zf−2iy2
−1 2ik
y2f+k y
∂f
∂z
+k f
=4y2∂
∂z
∂
∂zf−2kiy∂f
∂z
=−∆kf.
(Rk−2◦Lk)(f) =
2i∂
∂z+k−2
y −2iy2∂f
∂z
=4∂
∂z
y2∂f
∂z
−2(k−2)iy2∂f
∂z
=
4y2∂
∂z
∂
∂zf−4iy∂f
∂z
−2(k−2)iy2∂f
∂z
=4y2∂
∂z
∂
∂zf−2kiy∂f
∂z
=−∆kf.
(3) Using∆kf =λf and (1), we have−(Rk−2◦Lk)(f) =λf . Then−(Rk−2◦Lk)(Lk(f)) =λLk(f) and so
−(Rk−2◦Lk+k−2)(Lk(f)) = (λ−k+2)Lk(f).
Similarly, we have−(Lk+2◦Rk+k)(Rk(f)) =λRk(f)and so
−(Lk+2◦Rk)(Rk(f)) = (λ+k)Rk(f).
Now we introduce two differential operators acting on the spaceHk!(N,χ), which are associated with Maass rasing and lowering operators and play an important role in studying the Fourier expansion of
harmonic Maass forms.
In order to understand the motivation for which these operators were designed, define firstD-operator asD:= 1
2πi
∂
∂z. In general, the derivative of a modular form is not modular, but the Maass raising op- eratorRk=−4πD+k
y does. On the other hand,Rkdoes not preserve meromorphicity in general. From these observations, we want to get operators which preserves both modularity and meromorphicity. For weight 0,R0=−4πDdoes. But for general weights, we need to modifyD-operator and raising operator as follows.
For any positive integern, we define thek-iteration of Maass raising operatorsas Rnk:=Rk+2(n−1)◦ · · · ◦Rk+2◦Rk,
and forn=0, defineR0k:=1.
Lemma 12 For k∈Zand n∈N∪ {0}, we have Rnk=
n
∑
m=0
(−1)m n
m
(k+m)(n−m)ym−n(4πD)m, where x(l)=x(x+1)· · ·(x+l−1)is a rising factorial.
Proof 22 We will use induction on n. For n=0, it is trivial. Suppose that the above equation holds for natural n. By the chain rule and properties of binomial coefficient, we have
Rn+1k =Rk+2n◦Rnk=−4πD◦Rnk+ (k+2n)y−1◦Rnk
=
n m=0
∑
(−1)m n
m
(k+m)(n−m){(n−m)ym−n−1(4πD)m−ym−n(4πD)m+1} +
n
∑
m=0
(−1)m n
m
(k+m)(n−m)(k+2n)ym−n−1(4πD)m
=
n
∑
m=0
(−1)m n
m
(k+m)(n−m+1)ym−n−1(4πD)m +
n+1
∑
m=1
(−1)m n
m−1
(k+m−1)(n−m+1)ym−n−1(4πD)m
=k(n+1)y−n−1+ (−1)n+1(4πD)n+1 +
n
∑
m=1
(−1)m n
m
(k+m)(n−m+1)+ n
m−1
(k+m−1)(n−m+1)
ym−n−1(4πD)m
=
n+1
∑
m=0
(−1)m n+1
m
(k+m)(n+1−m)ym−(n+1)(4πD)m. This proves the lemma.
Now we can achieve our goals by relating the iteration of raising operatorRnkwhich preserves modu- larity and the(1−k)-th order differential operatorD1−kwhich preserves meromorphicity. This relation is called ‘Bol’s identity’, and the operatorD1−k is often calledBol’s operator.
Lemma 13 For k≤0, we have
D1−k= 1
(−4π)1−kR1−kk . In particular, we have that
D1−k:Mk!(N,χ)→M2−k! (N,χ).
Proof 23 Putting n=1−k≥1in Lemma 12, we observe that the rising factorial(k+m)(1−k−m)is0 unless m=1−k. Thus we have R1−kk = (−1)1−k(4πD)1−k.
Now, using Theorem 11 (1), we see that if f satisfies modularity of weight k then R1−kk satisfies modular- ity of weight k+2(1−k) =2−k. Meromorphicity at the cusps follows from the fact that D1−kpreserves meromorphicity.
Proposition 7 (1) If f ∈Hk!(N,χ)with Fourier series as in(48), then D1−k(f(z)) =−(4π)k−1(1−k)!c−f (0) +
∑
n−∞
c+f (n)n1−kqn. (52) (2) We have
D1−k:Hk!(N,χ)→M2−k! (N,χ).
Proof 24
(1) By the definition of incomplete gamma function, Γ(s,z)ez=
Z ∞
z
e−t+zts−1dt= Z ∞
0
e−t(t+z)s−1dt.
So we have
Γ(1−k,−4πny)e−4πny= Z ∞
0
e−t(t−4πny)−kdt, and hence
D1−k(Γ(1−k,−4πny)e−4πny) = 1
2πi
1−kZ ∞
0
∂1−k
∂z1−ke−t(t−4πny)−kdt=0.
On the other hand, note that
qn=e−4πnye4πnye2πin(x+iy)=e−4πnye2πin(x−iy)=e−4πnye2πinz. Hence we have
D1−k(Γ(1−k,−4πny)qn) =D1−k(Γ(1−k,−4πny)e−4πny)e2πinz=0.
Therefore we have
D1−k(f(z)) =D1−k(c−f(0)y1−k) +D1−k(f+(z))
=c−f (0) 1
2πi 1−k
∂
∂z 1−k
y1−k+ 1
2πi 1−k
n−∞
∑
c+f(n) ∂
∂z 1−k
qn
=c−f (0) 1
2πi 1−k
1 2
1−k
(−i)1−k ∂
∂y 1−k
y1−k
!
+ 1
2πi 1−k
n−∞
∑
c+f(n)(2πin)1−kqn
=−(4π)k−1(1−k)!c−f(0) +
∑
n−∞
c+f(n)n1−kqn. This proves (1).
(2) To show that D1−k(f(z))∈M2−k! (N,χ), we need to prove that it is meromorphic at the cusps of Γ0(N). By the proof of (1), we know that D1−k(f(z))is holomorphic onH and meromorphic at i∞. For any cusp ρ ofΓ0(N), the Fourier expansion of f at the cuspρ is given in Lemma 9. Through a process similar to the proof of (1), we can calculate that the second series in(51) annihilates, and thus meromorphicity at other cusps follows directly.
For a lowering operators, an analogous modifications can be made to have properties similar toD1−k. Definition 20 Theshadow operatorξk is defined by
ξk=2iyk ∂
∂z.
The shadow operator can be express by using lowering operator and hyperbolic Laplacian operator as
ξk=yk−2Lk, and
−ξ2−k◦ξk=−2iy2−k ∂
∂z 2iyk ∂
∂z
!
=−2iy2−k ∂
∂z
−2iyk ∂
∂z
=−4y2 ∂
∂z
∂
∂z+2kiy∂
∂z
=∆k. (53)
Proposition 8 Let k be an integer not equal to1.
(1) If f ∈Hk!(N,χ)with Fourier series as in(48), then
ξk(f(z)) = (1−k)c−f(0)−(4π)1−k
∑
n−∞
c−f (−n)n1−kqn. (54)
(2) We have
ξk:Hk!(N,χ)→M2−k! (N,χ).
Moreover, this map is surjective.
Proof 25 (1) First, we have ξk(f+) =0 since f+ is holomorphic and so ∂
∂zf+(z) =0. Next, we observe that
∂
∂zΓ(1−k,−4πny) =1 2
∂
∂x+i∂
∂y
Γ(1−k,−4πny)
= i 2
∂
∂y Z ∞
−4πny
e−tt−kdt
= i 2
∂
∂y
−4πn Z ∞
y
e4πnt(−4πnt)−kdt
=−2πin∂
∂y
Z ∞
0
− Z y
0
e4πnt(−4πnt)−kdt
=2πine4πny(−4πny)−k. Then we have
ξk
∑
n∞n6=0
c−f(n)Γ(1−k,−4πny)qn
=2iyk
∑
n∞n6=0
∂
∂z
c−f(n)Γ(1−k,−4πny)qn
=2iyk
∑
n∞n6=0
c−f(n)2πin(−4πny)−ke4πnye2πinz
=−(4π)1−k
∑
n∞n6=0
c−f (n)(−n)1−ke4πnye−2πinz
=−(4π)1−k
∑
n−∞
c−f(−n)n1−kqn, and
ξk(c−f(0)y1−k) =2iyk ∂
∂zc−f (0)y1−k=2iykc−f(0)i 2
∂
∂yy1−k= (1−k)c−f(0).
This proves (1).
(2) Let f ∈Hk!(N,χ). By the automorphy of f , we have f|kγ=χ(d)f for allγ= a b c d
!
∈γ0(N).
By definition of shadow operator, observe that
(ξkf|2−kγ)(z) = ((yk−2Lk)|2−kγ)(z) = yk−2
|cz+d|2(k−2)Lk(f)(γz)(cz+d)−(2−k). By Lemma 11, we have
Lk(f)|k−2γ=Lk(f|kγ) =Lk(χ(d)f) =χ(d)Lk(f).
Taking complex conjugate on both sides, we obtain
χ(d)Lk(f)(z) =χ(d)Lk(f)(z) = (Lk(f)|k−2γ)(z) = (cz+d)−(k−2)Lk(f)(γz)
⇔Lk(f)(γz) =χ(d)(cz+d)k−2Lk(f)(z).
Thus we get
(ξkf|2−kγ)(z) = yk−2
|cz+d|2(k−2)
χ(d)(cz+d)k−2Lk(f)(z)
(cz+d)−(2−k)
=χ(d)yk−2Lk(f)(z)
=χ(d)(ξkf)(z).
This provesξk(f)∈M2−k! (N,χ).
To showξk(f(z))∈M2−k! (N,χ), we need to prove thatξk(f(z))is meromorphic at the cusps of Γ0(N). By the proof of (1), we know thatξk(f(z))is holomorphic onH and meromorphic at i∞.
For any cuspρ ofΓ0(N), the Fourier expansion of f at the cuspρis given in Lemma 9. Through a process similar to the proof of (1), we can calculate that the second series in(51)changes to the holomorphic part, and thus meromorphicity at other cusps follows directly.
Proving surjectivity uses the theory of algebraic geometry, especially the Serre duality. We omit this proof here, and we refer [36] for a detailed proof and [38] for more informations of Serre duality.