ON DRINFELD CUSP FORMS OF PRIME LEVEL

ANDREA BANDINI AND MARIA VALENTINO

Abstract. Let (Pd ) be any prime of Fq [t] of degree d and consider the space of Drinfeld cusp forms of level Pd ,
i.e. for the modular group Γ0 (Pd ). We provide a definition for oldforms and newforms of level Pd . Moreover,
when the dimension of the vector space of oldforms is one and P1 = t we prove that the space of cuspforms of level
t is the direct sum of oldforms and newforms and that the Hecke operator Tt acting on Drinfeld cusp forms of
level 1 is injective, thus providing more evidence for the conjectures presented and stated in [2] and [3].
arXiv:1908.09768v1 [math.NT] 26 Aug 2019

1. Introduction
Let K be the global function field Fq (t), where q is a power of a fixed prime p ∈ Z, fix the prime 1t at ∞ and
denote by O ∶= Fq [t] its ring of integers (i.e., the ring of functions regular outside ∞). Let K∞ = Fq (( 1t )) be
the completion of K at 1t with ring of integers O∞ = Fq J 1t K and denote by C∞ the completion of an algebraic
closure of K∞ .
The Drinfeld upper half-plane is the set Ω ∶= P1 (C∞ ) − P1 (K∞ ) together with a structure of rigid analytic space
(see [7]). The group GL2 (K∞ ) acts on Ω via Möbius transformation
a b az + b
( ) (z) = .
c d cz + d
Let Γ be an arithmetic subgroup of GL2 (O), then Γ has finitely many cusps, i.e. equivalence classes for the
a b
action of Γ on P1 (K). For γ = ( ) ∈ GL2 (K∞ ), k, m ∈ Z and ϕ ∶ Ω → C∞ , we define the ∣k,m γ operator by
c d

(1) (ϕ ∣k,m γ)(z) ∶= ϕ(γz)(det γ)m (cz + d)−k .

Since for any γ ∈ GL2 (O) one has det(γ) ∈ F∗q , the integers m can be cosidered modulo q − 1.
Definition 1.1. A rigid analytic function ϕ ∶ Ω → C∞ is called a Drinfeld modular function of weight k and
type m for Γ if
(2) (ϕ ∣k,m γ)(z) = ϕ(z) ∀γ ∈ Γ.
A Drinfeld modular function ϕ of weight k ⩾ 0 and type m ∈ Z/(q − 1)Z for Γ is called a Drinfeld modular form
if ϕ is holomorphic at all cusps.
A Drinfeld modular form ϕ is called a cusp form if it vanishes at all cusps.
The space of Drinfeld modular forms of weight k and type m for Γ will be denoted by Mk,m (Γ). The subspace
1
of cuspidal modular forms is denoted by Sk,m (Γ).
The above definition coincides with [4, Definition 5.1], other authors require the function to be meromorphic
(in the sense of rigid analysis, see for example [5, Definition 1.4]) and would call our functions weakly modular.
We shall deal only with the arithmetic subgroups
a b
Γ = Γ0 (m) ∶= {( ) ∈ GL2 (O) ∶ c ≡ 0 (mod m)} ,
c d
where m is an ideal of O, and we shall focus mainly on the cases m = 1 (so that Γ0 (1) = GL2 (O) ) and m a
prime ideal. When m is prime we fix the monic irreducible generator Pd of m and will use simply Pd or (Pd )
1
to denote the ideal. The spaces Sk,m (Γ0 (m)) denote cusp forms of level m. We recall that spaces of Drinfeld
1
2 ANDREA BANDINI AND MARIA VALENTINO

modular forms of fixed weight and type are finite dimensional vector space over C∞ ; for details on dimensions
the reader is referred to [8].
Fix an ideal m and a monic irreducible element Pd of degree d in O. Assume (Pd ) does not divide m (which is
1
the case we shall usually work with): we have the following Hecke operators acting, respectively, on Sk,m (Γ0 (m))
1
and Sk,m (Γ0 (mPd )):

Pd 0 1 Q
TPd (ϕ)(z) ∶= Pdk−m (ϕ ∣k,m ( ))(z) + Pdk−m ∑ (ϕ ∣k,m ( 0 P ))(z)
0 1 Q∈O d
deg Q<d

and
1 Q
UPd (ϕ)(z) ∶= Pdk−m ∑ (ϕ ∣k,m ( 0 P ))(z).
Q∈O d
deg Q<d

We recall that the operator UPd is commonly called Atkin-Lehner operator, or simply Atkin-operator.
Using Teitelbaum’s representation of cusp forms as cocycles (see [14] or [4], a brief account of the formulas
relevant for our computations is in [2, Sections 2.3 and 2.4]), in [2] we were able to compute the matrix associated
1
with the Atkin operator Ut acting on Sk,m (Γ1 (t)) (where, as usual,

a b
Γ1 (t) ∶= {( ) ∈ GL2 (O) ∶ a, d ≡ 1 (mod t) and c ≡ 0 (mod t)} )
c d
1
and to isolate inside it the blocks referring to the action on the subspace Sk,m (Γ0 (t)) (see [2, Section 4]). In [1]
(for Γ1 (t)) and [3] (for Γ0 (t)) we studied the properties of such matrix as a tool to investigate the analogue of
several classical (characteristic zero setting) issues related to Drinfeld cusp forms. In particular, we considered
problems like the structure of cusp forms of level t, the injectivity of Tt , diagonalizability and slopes for Ut , i.e.
t-adic valuation of eigenvalues of Ut . Moreover, we collected data on the distribution of slopes (available on
the web page https://sites.google.com/site/mariavalentino84/publications) as the weight varies, which led us
to formulate various conjecture à la Gouvêa-Mazur (see [10]) and on the existence of families of Drinfeld cusp
forms. For details see [2, Section 5] and [3, Section 6].
We would like to mention that, building on such results, Hattori has recently proved a function field analogue of
Gouvêa-Mazur’s conjecture (see [11]) and has made relevant progresses in the construction of (p-adic) families of
Drinfeld modular forms (see [12]). It is worth mentioning that, following a completely different (more geometric)
approach, Nicole and Rosso in [13] have provided deep results on the existence of families of modular forms in
characteristic p.
In the present paper we shall address the following issues.
1
i) Structure of Sk,m (mPd ). A major and basic topic in the study of classical modular forms is the splitting
of Sk (Γ0 (N )), for a general level N ∈ Z, as oldforms, those coming from a lower level M ∣N , and
newforms, i.e. the orthogonal complement of the space of oldforms with respect to the Petersson inner
product (see [6, Chapter 5]). In the positive characteristic setting we do not have an analogue of such
product, therefore we need a different approach. In [2, Section 3] we defined oldforms and newforms of
1
level t and we also conjectured, and proved in some particular cases, that Sk,m (Γ0 (t)) is direct sum of
newforms and oldforms. Here we generalize all definitions to a general prime level Pd and also prove
some further results for the case P1 = t.
ii) Injectivity of Tt . Building on the data mentioned above, we observed a phenomenon that has no
1
analogue in the characteristic zero setting, namely that the Hecke operator Tt acting on Sk,m (GL2 (O))
seems to be injective, and this would have consequences also on the diagonalizability of Ut acting on
the space of oldforms (see [2, Section 3.2]). In the paper [3] we already gave evidence of this conjecture
for some special cases, here we shall extend the cases in which we can prove the injectivity of Tt .
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 3

The paper is organized as follows.

In Section 2 we supply definitions of oldforms and newforms. We consider the maps δ1 , δPd , called degeneracy
1 1
maps, from a lower level Sk,m (Γ0 (m)) to an upper one Sk,m (Γ0 (mPd )) (Section 2.1) and use them to define
oldforms. On the other side we have trace maps which go the other way around and use them, together with
the crucial ingredient of Fricke involution, to define newforms (Section 2.2). Two main issues appear here:
● we define newforms only for prime level Pd (hence for m = 1), the definition seems easily generalizable
for traces but we lack an involution of level m to extend it in general;
● as mentioned above, we do not have the analog of Petersson inner product in our setting, hence we need
to prove that cusp forms are direct sum of our oldforms and newforms to confirm that our definitions
are the “right” ones.
We use the interaction between degeneracy maps, trace maps and Hecke operators to provide a description of
the kernels of TPd and UPd (Propositions 2.5 and 2.7): in particular, the criterion
1
ϕ ∈ Sk,m (GL2 (O)) is in ∈ Ker(TPd ) if and only if δ1 (ϕ) ∈ Ker(U2Pd )
will be useful to prove the injectivity of Tt in the case presented in Section 3. Moreover, in Theorem 2.12,
we show an important criterion, which is a generalization of [3, Theorem 5.1], to get the direct sum between
oldforms and newforms by proving that it is equivalent to the invertibility of the map D ∶= Id − Pdk−2m (T r′ )2 .
In Section 3 we specialize to the case P1 = t. Exploiting the linear algebra translation of our conjectures provided
in [3] and using the criterions above we shall prove the following.
1
Theorem 1.2. Assume that dimC∞ Sk,m (GL2 (O)) = 1, then we have:
1
● the operator Tt acting on Sk,m (GL2 (O)) is injective (Theorem 3.2);
1
● the space Sk,m (Γ0 (t)) is direct sum of newforms and oldforms (Theorem 3.3).

2. Newforms and oldforms

Here we define oldforms and newforms for a general prime level Pd ; most of the formulas and definitions
are straightforward and come from computations on Hecke operators and trace maps (defined in [15, Section
3]) similar to the ones presented in [2], hence we often only provide the outcome and refer the reader to those
papers for the missing details.

2.1. Degeneracy maps and oldforms. Let m be any ideal in O and consider the spaces of Drinfeld cusp
1 1
forms Sk,m (Γ0 (m)) and Sk,m (Γ0 (mPd )) of levels m and mPd respectively. We have two maps which produce
1
oldforms in Sk,m (Γ0 (mPd )):
1 1
Sk,m (m) → Sk,m (Γ0 (mPd ))
δ1 ϕ = ϕ
Pd 0
δPd ϕ = (ϕ ∣k,m ( ))(z) = Pdm ϕ(Pd z)
0 1
Proposition 2.1. Assume that (Pd ) does not divide m, then the map
1
(δ1 , δPd ) ∶ Sk,m (Γ0 (m))2 → Sk,m
1
(Γ0 (mPd ))
(ϕ1 , ϕ2 ) ↦ δ1 ϕ1 + δPd ϕ2
is injective.
Proof. The proof works exactly as in [2, Proposition 3.1], just replace the tree Tt used there with the Bruhat-Tits
tree TPd at Pd associated with GL2 (KPd ) (KPd being the completion of K at the prime Pd ). 
1,old 1
Definition 2.2. The space of oldforms of level m, denoted by Sk,m (Γ0 (m)), is the subspace of Sk,m (Γ0 (m))
1 2
generated by the set {(δ1 , δPd )(ϕ1 , ϕ2 ) ∶ (ϕ1 , ϕ2 ) ∈ Sk,m (Γ0 (m/(Pd )) , f or all (Pd )∣m}.
4 ANDREA BANDINI AND MARIA VALENTINO

1
Let ϕ ∈ Sk,m (Γ0 (m)) and assume that Pd does not divide m so that we have “different” Hecke operators TPd
and UPd on the levels m and mPd respectively. Then the relations between the maps δ1 and δPd and the Hecke
operators are the following:

(3) δ1 (TPd ϕ) = Pdk−m δPd ϕ + UPd (δ1 ϕ)

Pd 0 1 Q
(4) UPd (δPd ϕ) = Pdk−m ∑ (ϕ ∣k,m ( 0 1 ) ( 0 P ))(z)
Q∈O d
deg Q<d

1 Q Pd 0
= Pdk−m ∑ (ϕ ∣k,m ( 0 1 ) ( 0 P ))(z)
Q∈O d
deg Q<d

= Pdm ∑ ϕ(z) = 0.
Q∈O
deg Q<d

Proposition 2.3. Assume that (Pd ) does not divide m, then

{Eigenvalues of UPd ∣S 1,old (Γ0 (mP } = {Eigenvalues of TPd } ∪ {0}.
k,m d ))

Proof. Let (δ1 , δPd )(ϕ, ψ) be an old eigenform for UPd of eigenvalue λ. Then
λ(δ1 , δPd )(ϕ, ψ) = UPd ((δ1 , δPd )(ϕ, ψ))
= UPd (δ1 ϕ)
= δ1 (TPd ) − Pdk−m δPd (ϕ)
= (δ1 , δPd )(TPd ϕ, −Pdk−m ϕ)

implies TPd ϕ = λϕ because of the injectivity of (δ1 , δPd ).

If TPd ϕ = λϕ with λ ≠ 0 we have

Pdk−m
UPd ((δ1 , δPd )(ϕ, − ϕ)) = UPd (δ1 ϕ)
λ
= δ1 (TPd ϕ) − Pdk−m δPd ϕ
= λδ1 ϕ − Pdk−m δPd ϕ
Pdk−m
= λ(δ1 , δPd )(ϕ, − ϕ). 
λ
We have just seen that the behaviour of UPd on oldforms is analogous to the classical case: the eigenvalues for
UPd verify equations like X 2 − λX = 0 where λ is a nonzero eigenvalue for TPd (in the classical case the equation
was X 2 − λX + pk−1 = 0 which reduces to our one modulo p, see [10, Section 4]).
Remark 2.4. Let ϕ be an eigenvector for TPd of eigenvalue λ, then the matrix for the action of UPd on the
λ −Pdk−m
couple {δ1 ϕ, δPd ϕ} is ( ). Hence it is easy to see that, assuming (Pd ) does not divide m, the operator
0 0
UPd is diagonalizable on oldforms if and only if the operators TPd are diagonalizable at lower levels and are
injective. We believe UPd is diagonalizable in odd characteristic (and, for P1 = t, we provided evidence for it in
[1] and [3]) and this motivates our investigation on the injectivity of the Hecke operators TPd .
The next proposition describes Ker(TPd ) and will be crucial in the proof of Theorem 3.2.
1
Proposition 2.5. Let ϕ ∈ Sk,m (Γ0 (m)) such that Pd ∤ m, then ϕ ∈ Ker(TPd ) if and only if δ1 (ϕ) ∈ Ker(U2Pd ).
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 5

Proof. By (3), for any ϕ ∈ Ker(TPd ) one has U2Pd (δ1 ϕ) = −Pdk−m UPd (δPd ϕ) = 0.
1
Now let ϕ ∈ Sk,m (Γ0 (m)) be such that δ1 ϕ ∈ Ker(U2Pd ). Then
0 = U2Pd (δ1 ϕ) = UPd (δ1 (TPd ϕ) − Pdk−m δPd ϕ)
= δ1 (T2Pd ϕ) − Pdk−m δPd (TPd ϕ)
= δ(T2Pd ϕ, −Pdk−m (TPd ϕ)).
Since δ is injective we have TPd ϕ = 0. 
1
2.2. Trace maps and newforms. From now on we take m = 1 and denote (δ1 , δPd ) ∶ Sk,m (GL2 (O))2 →
1
Sk,m (Γ0 (Pd )) simply by δ. The reason for this is the crucial role played by the Fricke involution in the definition
of the twisted trace and of newforms (see below): the trace map should be easily generalizable to any level m
just considering representatives for Γ0 (mPd )/Γ0 (m) but we are still looking for the correct generalization of the
Fricke involution. We recall that a system of coset representative for Γ0 (Pd )/GL2 (O) is
0 −1
R = {Id, ( ) s.t. Q ∈ O and deg Q < d} .
1 Q
For details on some of the maps defined in this section see [15].
1
Definition 2.6. We have the following maps defined on Sk,m (Γ0 (Pd )):
1
● the Fricke involution, which preserves the space Sk,m (Γ0 (Pd )), is represented by the matrix
0 −1
γPd ∶= ( )
Pd 0
and defined by ϕF r = (ϕ ∣k,m γPd );
● the trace map is defined by
1 1
T r ∶ Sk,m (Γ0 (Pd )) → Sk,m (GL2 (O))
ϕ ↦ ∑ (ϕ ∣k,m γ)(z);
γ∈R

● the twisted trace map is defined by

T r′ ∶ Sk,m
1 1
(Γ0 (Pd )) → Sk,m (GL2 (O))
ϕ ↦ T r(ϕF r ).
We list here many useful formulas expressing the relations between these maps, the Hecke operators and the
maps δ1 and δPd , the proofs rely on matrix decomposition and on the definitions of the various maps and are
similar to those in [2, Section 3]. Please note that the first three formulas hold for cusp forms of level Pd , while
the following ones hold for cusp forms of level 1.
1
Let ψ ∈ Sk,m (Γ0 (Pd )), then we have
Fr
Fr Fr 0 −1
(5) (ψ ) = ((ψ ∣k,m ( ))(z))
Pd 0
0 −1 0 −1
= (ψ ∣k,m ( )( ))(z) = Pd2m−k ψ;
Pd 0 Pd 0

1
0 −1 1 Q 0
(6) T r(ψ) = ψ + ∑ (ψ ∣k,m ( )( ) ( Pd 1 )) (z)
Q∈O
P d 0 0 Pd 0 Pd
deg Q<d

= ψ + Pd−m UPd (ψ F r );
6 ANDREA BANDINI AND MARIA VALENTINO

(7) T r′ (ψ) = ψ F r + Pdm−k UPd (ψ).

1
Now let ϕ ∈ Sk,m (GL2 (O)), then we have

0 −1
(8) (δ1 ϕ)F r = (ϕ ∣k,m ( ))(z)
Pd 0
0 −1 P 0
= (ϕ ∣k,m ( )( d ))(z) = δPd ϕ;
1 0 0 1

Pd 0 0 −1
(9) (δPd ϕ)F r = (ϕ ∣k,m ( )( ))(z)
0 1 Pd 0
0 −1 P 0
= (ϕ ∣k,m ( )( d ))(z) = Pd2m−k δ1 ϕ;
1 0 0 Pd

(10) UPd ((δ1 ϕ)F r ) = 0;

(11) UPd ((δPd ϕ)F r ) = Pd2m−k [δ1 TPd ϕ − Pdk−m δPd ϕ];

(12) T r(δ1 ϕ) = ∑ ϕ = ϕ;
γ∈R

Pd 0
(13) T r(δPd ϕ) = ( ∑ ϕ ∣k,m ( ) γ)(z)
γ∈R
0 1
0 −Pd
= (δPd ϕ)(z) + ( ∑ ( ))(z)
Q∈O
1 Q
deg Q<d

= Pdm−k TPd ϕ.
As an application we have an explicit description of the kernel of the Hecke operator UPd .
Proposition 2.7. We have Ker(UPd ) = Im(δPd ).
Proof. We have already seen that Ker(UPd ) ⊇ Im(δPd ). Now let ϕ ∈ Ker(UPd ) and note that, by (7), T r′ (ϕ) =
ϕF r ∈ Sk,m
1
(GL2 (O)). Then it is easy to check that, with ψ ∶= Pdk−2m ϕF r ∈ Sk,m
1
(GL2 (O)), one has δPd (ψ) =
ϕ. 
1,new
Definition 2.8. The space of newforms of level Pd , denoted by Sk,m (Γ0 (Pd )) is given by Ker(T r)∩Ker(T r′ ).
Remark 2.9. From formulas (3) and (4), it is easy to see that UPd preserves the space of oldforms (of any
level). For any newform ϕ of level Pd we have T r(ϕ) = T r′ (ϕ) = 0, hence (7) yields UPd (ϕ) = −Pdk−m ϕF r . Thus
it immediately follows that T r(UPd (ϕ)) = T r′ (UPd (ϕ)) = 0, i.e. UPd preserves newforms as well.
1
Remark 2.10. The trace alone is not enough to isolate newforms: indeed let ϕ ∈ Sk,m (GL2 (O)) be such that
TPd ϕ = λϕ with λ ≠ 0. Then one can check that
Pdk−m
ψ1 ∶= δ1 ϕ − δPd ϕ ∈ Ker(T r)
λ
and
Pdk−m
ψ2 ∶= δ1 ϕ − Pdk−2m δPd ϕ ∈ Ker(T r′ )
λ
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 7

(recall that, by the proof of Proposition 2.3, ψ1 is an UPd -eigenvector of eigenvalue λ). In general, ψ1 ∉ Ker(T r′ )
k/2
and ψ2 ∉ Ker(T r) unless λ = ±Pd .
k/2
The values ±Pd (i.e. the slope k2 in the sense of [2, Definition 3.4 and Remark 3.5]) are the only possible
eigenvalues for newforms and we actually believe that they identify newforms, i.e. there are no oldforms with
such eigenvalues (this would have relevant consequences also on other conjectures like the one discussed in
Section 3, see [3, Remark 5.3]).
1 k/2
Proposition 2.11. Let ϕ ∈ Sk,m (Γ0 (Pd )) be a new UPd -eigenform of eigenvalue λ, then λ = ±Pd .
Proof. By (6) and (7)
ϕ = −Pd−m UPd (ϕF r ) and ϕF r = −Pdm−k UPd (ϕ).
It follows that
λ2 ϕ = λ(UPd ϕ) = U2Pd ϕ
= UPd (−Pdk−m ϕF r )
= −Pdk−m UPd (ϕF r ) = Pdk ϕ.
k/2
Hence λ = ±Pd . 
The following important criterion is the analog of [3, Theorem 5.1].
1 1,old 1,new
Theorem 2.12. We have a direct sum decomposition Sk,m (Γ0 (Pd )) = Sk,m (Γ0 (Pd )) ⊕ Sk,m (Γ0 (Pd )) if and
k−2m ′ 2
only if the map D ∶= Id − Pd (T r ) is bijective.
Proof. (⇐Ô) We start by proving that the intersection between newforms and oldforms is trivial.
1
Let η = δ(ϕ, ψ) ∈ Sk,m (Γ0 (Pd )) be old and new. The following facts hold:
● η = ϕ + ψ F r since ϕ and ψ are both of level 1;
● 0 = T r(η) = T r(ϕ) + T r(ψ F r ) = ϕ + T r′ (ψ), so that T r′ (ψ) = −ϕ;
● 0 = T r′ (η) = T r′ (ϕ) + T r′ (ψ F r ) = 0.
From the last two equalities we get
0 = −T r′ (T r′ ψ) + T r((ψ F r )F r ) = −(T r′ )2 (ψ) + Pd2m−k T r(ψ).
So
(T r′ )2 ψ − Pd2m−k ψ = 0
and
(Id − Pdk−2m (T r′ )2 )ψ = Dψ = 0.
Since, by hypothesis, D is invertible, this yields ψ = 0 and ϕ = −T r′ (ψ) = 0 as well.
1 1
Now we have to prove the sum condition. Given η ∈ Sk,m (Γ0 (Pd )) it is sufficient to find ϕ1 , ϕ2 ∈ Sk,m (GL2 (O))
such that η − δ(ϕ1 , ϕ2 ) is new, i.e. we need to solve the following
T r(η − δ(ϕ1 , ϕ2 )) = 0
T r′ (η − δ(ϕ1 , ϕ2 )) = 0
{ .

These equations are equivalento to

T r(η) − ϕ1 − T r(δPd ϕ2 ) = 0 T r(η) − ϕ1 − T r(ϕF2 r ) = 0
T r′ (η) − T r′ (ϕ1 ) − T r′ (δPd ϕ2 ) = 0 T r′ (η) − T r′ (ϕ1 ) − T r′ (ϕF2 r ) = 0
(14) { , i.e. { ,

which finally leads to

ϕ1 = T r(η) − T r(ϕF2 r )
(15) { .
T r(η F r ) − T r(ϕF1 r ) − Pd2m−k ϕ2 = 0
8 ANDREA BANDINI AND MARIA VALENTINO

Using the two equations of (15) we have

(16) ϕ2 = Pdk−2m [T r′ (η) − T r′ (ϕ1 )]
= Pdk−2m [T r′ (η) − T r′ (T r(η)) + (T r′ )2 (ϕ2 )] .
Then Dϕ2 = Pdk−2m [T r′ (η) − T r′ (T r(η))] and ϕ2 = Pdk−2m D−1 (T r′ (η − T r(η))).
Substituting the first expression for ϕ2 found in (16) in the first equation of (15), one has
ϕ1 = T r(η) − Pdk−2m (T r′ )2 η + Pdk−2m (T r′ )2 ϕ1 ,
which implies
Dϕ1 = T r(η) − Pdk−2m (T r′ )2 η
and finally
ϕ1 = D−1 (T r(η) − Pdk−2m (T r′ )2 η).
(Ô⇒) Let η ≠ 0 be such that η ∈ Ker(D). Then Pd2m−k η = (T r′ )2 η. Recall that T r2 = T r (as for any trace map)
and apply T r to obtain
Pd2m−k T r(η) = T r(T r′ (T r′ η))
= T r(T r((T r′ η)F r )) = (T r′ )2 (η).
Therefore T r(η) = η, so η is old and it is contained in the image of δ1 . Observe that UPd (η) ≠ 0, otherwise, by
Proposition 2.7, one would have η ∈ Im(δ1 ) ∩ Im(δPd ) = {0} (by Proposition 2.1). In particular, by Remark 2.9,
UPd (η) is old. Then
Pd2m−k η = (T r′ )2 η
= T r′ (T r′ (η)) (apply (7))
′
= T r (η + Pdm−k UPd (η))
Fr

= T r((η F r )F r ) + Pdm−k T r′ (UPd (η))

= Pd2m−k T r(η) + Pdm−k T r′ (UPd (η)).
So, T r′ (UPd (η)) = 0 (because η is old with T r(η) = η).
Finally note that, by equations (3), (12) and (13),
T r(UPd (η)) = T r(δ1 TPd (η)) − Pdk−m T r(δPd η)
= TPd (η) − Pdk−m Pdm−k TPd (η) = 0
So, UPd (η) is also new and we do not have direct sum. 
From the above proof an easy calculation leads to
1 k/2
Ker(D) = {δ1 ϕ ∶ ϕ ∈ Sk,m (GL2 (O)) and TPd ϕ = ±Pd ϕ}.
Indeed recall that for any cusp form ψ of level 1 we have δPd ψ = (δ1 ψ)F r , hence
T r′ (T r′ (δ1 ϕ)) = T r′ (T r(δPd ϕ))
= Pdm−k T r′ (TPd ϕ)
= Pdm−k T r((TPd ϕ)F r )
= Pdm−k T r(δPd TPd ϕ)
= Pd2m−2k T2Pd ϕ
Moreover, δ1 ϕ ∈ Ker(D) implies:
● UPd (δ1 ϕ) is old and new;
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 9

k/2
● if TPd ϕ = Pd ϕ then
k/2−m k/2 k/2−m
UPd (δ1 ϕ − Pd δPd ϕ) = Pd (δ1 ϕ − Pd δPd ϕ)
k/2−m
and δ1 ϕ − Pd δPd ϕ is old and new;
k/2
● if TPd ϕ = −Pd ϕ then
k/2−m k/2 k/2−m
UPd (δ1 ϕ + Pd δPd ϕ) = −Pd (δ1 ϕ + Pd δPd ϕ)
k/2−m
and δ1 ϕ + Pd δPd ϕ is old and new.

3. Special case: P1 = t.
For the level P1 = t we explicitly computed the matrices associated to the operator Ut , the Fricke involution
and the trace maps (see [2, Section 4] and [3, Sections 3 and 4]): for the convenience of the reader we are going
to briefly describe here these matrices.
1
We recall that, in order to have Sk,m (Γ0 (t)) ≠ 0, we need k ≡ 2m (mod q − 1). Moreover, it is always possible
to find a j ∈ {0, 1, . . . , q − 2} and a unique n ∈ Z⩾0 such that k = 2(j + 1) + (n − 1)(q − 1) (j is related to the type
m by the relation m ≡ j + 1 (mod q − 1), see [2, Section 4.3]). From now on, the letters j and n will always
be linked to the weight k by the previous formula, giving us information, respectively, on the type m and the
1
dimension of the matrix U associated to Ut acting on Sk,m (Γ0 (t)).
We have
s
⎛ t1 ⋯ 0 ⎞
(17) U = M D ∶= M ⎜ ⋱ ⎟
⎝ 0 ⋯ tsn ⎠
n n+1
where, for 1 ⩽ i ⩽ n, we put si = j + 1 + (i − 1)(q − 1) (so that si + sn+1−i = k for 1 ⩽ i ⩽ 2 or 1 ⩽ i ⩽ 2 according
to n being even or odd) and, for even n, the matrix M is

⎛ m1,1 m1,2 ⋯ m1, n2 (−1)j+1 m1, n2 ⋯ (−1)j+1 m1,2 (−1)j+1 (m1,1 − 1) ⎞

⎜
⎜ m2,1 m2,2 ⋯ m2, n2 (−1)j+1 m2, n2 ⋯ (−1)j+1 (m2,2 − 1) (−1)j+1 m2,1 ⎟
⎟
⎜ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⎟
⎜ ⎟
⎜ m n2 ,1 m n2 ,2 ⋯ m n2 , n2 (−1)j+1 (m n2 , n2 − 1) ⋯ (−1)j+1 m n2 ,2 (−1)j+1 m n2 ,1 ⎟
⎜ ⎟
M =⎜
⎜ m n2 +1,1 m n2 +1,2 ⋯ (−1)j 0 ⋯ (−1)j+1 m n2 +1,2 (−1)j+1 m n2 +1,1
⎟,
⎟
⎜ ⎟
⎜ . ⎟
⎜
⎜ ⋮ ⋮ .. ⋮ ⋮ ⋱ ⋮ ⋮ ⎟
⎟
⎜ ⎟
⎜ mn−1,1 (−1)j ⋯ 0 0 ⋯ 0 (−1)j+1 mn−1,1 ⎟
⎝ (−1)j 0 ⋯ 0 0 ⋯ 0 0 ⎠
n+1
while for odd n one just needs to modify the indices a bit and add the central 2 -th column
(m1, n+1 , ⋯, m n−1 , n+1 , (−1)j , 0, ⋯, 0).
2 2 2

The entries of M are the binomial coefficients in Fp

⎧ j + (n − a)(q − 1) j + (n − a)(q − 1)
⎪
⎪
⎪
⎪ − [( ) + (−1)j+1 ( )] if a ≠ b
(18)
⎪
ma,b = ⎨ j + (n − b)(q − 1) j + (b − 1)(q − 1) .
⎪
⎪ j j + (n − a)(q − 1)
⎪
⎪
⎪ (−1) ( ) if a = b
⎩ j + (a − 1)(q − 1)
The other matrices associated to the relevant maps we used to define oldforms and newforms are the following:
● the matrix for the Fricke involution is
⎛ 0 . . . (−t)sn ⎞ ⎛ 0 . . . (−1)j+1 tsn ⎞
tm−k F = tm−k ⎜ . ⎟ = tm−k ⎜ .
(19) ⎜ .. ⎟ ⎜ .. ⎟.
⎟
⎝ (−t)s1 . . . 0 ⎠ ⎝ j+1
(−1) t s 1
... 0 ⎠
10 ANDREA BANDINI AND MARIA VALENTINO

Note that, if we let A be the antidiagonal matrix

⎛ 0 ... (−1)j+1 ⎞
.
(20) A=⎜
⎜ .. ⎟,
⎟
⎝ (−1)j+1 . . . 0 ⎠
we get AF = D;
● from equation (6) we find that the trace is represented by the matrix
(21) T ∶= I + t−m M D(tm−k F ) = I + t−k M AF 2 = I + M A
where I is the identity matrix of dimension n;
● the twisted trace is represented by
(22) T ′ = tm−k T F = tm−k (F + M D).
Remark 3.1. Note that M A switches columns i and n + 1 − i in the matrix M and multiplies everything by
(−1)j+1 : looking at the description of M we see that this produces a matrix which looks just like M except for
the fact that the (−1)j on the antidiagonal disappear and are substituted by (−1)j (−1)j+1 = −1 on the diagonal.
Therefore the matrix T = I + M A is the following (for even n)
⎛ m1,1 m1,2 ⋯ m1, n2 (−1)j+1 m1, n2 ⋯ (−1)j+1 m1,2 (−1)j+1 m1,1 ⎞
⎜ m2,1
⎜ m2,2 ⋯ m2, n2 (−1)j+1 m2, n2 ⋯ (−1)j+1 m2,2 (−1)j+1 m2,1 ⎟
⎟
⎜ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⎟
⎜ ⎟
⎜ mn m n2 ,2 ⋯ m n2 , n2 (−1)j+1 m n2 , n2 ⋯ (−1)j+1 m n2 ,2 (−1)j+1 m n2 ,1 ⎟
⎜ 2 ,1 ⎟
T =⎜
⎜ m n2 +1,1 m n2 +1,2 ⋯ 0 0 ⋯ (−1)j+1 m n2 +1,2 (−1)j+1 m n2 +1,1
⎟.
⎟
⎜ ⎟
⎜ . ⎟
⎜
⎜ ⋮ ⋮ .. ⋮ ⋮ ⋱ ⋮ ⋮ ⎟
⎟
⎜ ⎟
⎜ mn−1,1 0 ⋯ 0 0 ⋯ 0 (−1)j+1 mn−1,1 ⎟
⎝ 0 0 ⋯ 0 0 ⋯ 0 0 ⎠
n+1
As before, for odd n one just needs to modify the indices a bit and add the central 2 -th column
(m1, n+1 , ⋯, m n−1 , n+1 , 0, ⋯, 0).
2 2 2
j
Hence T is basically M without the (−1) on the antidiagonal and verifies a number of equations/relations like
● T = A + M;
● T = T A (this comes directly from the previous one, to verify it via computations on the above matrix
one has to note that for odd n and even j the central column is identically 0 because of the formula (18),
while for odd j one is simply multiplying the central column by 1);
● T 2 = T , like any trace map.
From these, one can produce various relations on M (like M AT = T M = 0 or, more surprisingly, M 3 = M ) with
consequences, for example, on the diagonalizability of M , but we shall not pursue this topic any further here.
We also recall that Im(δ1 ) = Ker(T r − Id), i.e. in terms of matrices
(23) Im(δ1 ) = Ker(M A).
In [2, Section 5] we hinted at some conjectures which were stated more explicitly in [3, Conjecture 1.1]:
among other things we conjectured that for P1 = t
(1) Tt is injective;
1
(2) Sk,m (Γ0 (t)) is the direct sum of oldforms and newforms.
In [3] we proved some special cases building on the analog of Theorem 2.12 (one of the reasons which makes
us believe the conjectures should hold for any Pd ) and on the above matrices/formulas (which are not avaliable
1
for d ⩾ 2). In particular, in [3, Theorem 5.5] we proved that when dimC∞ (Sk,m (GL2 (O)) = 0 (i.e. there
are no oldforms) the matrix M is antidiagonal and the conjectures hold: we shall now approach the case
1 1
dimC∞ (Sk,m (GL2 (O)) = 1, this will include many more cases since, for example, dimC∞ (Sk,0 (GL2 (O)) = 1 if
and only if q ⩽ n < 2q − 1, by [5, Proposition 4.3] (compare with the bounds of [3, Theorems 5.8, 5.9, 5.12, 5.14]).
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 11

3.1. Injectivity of Tt .

Theorem 3.2. Assume that dimC∞ Im(δ1 ) = 1, then Tt is injective.

Proof. By Proposition 2.5, Ker(Tt ) = Ker(M A) ∩ Ker(M DM D). Thanks to our assumption on the dimen-
sion of Im(δ1 ) = Ker(M A) and to the fact that the entries of M A are in Fp , we have dimC∞ (Ker(M A) ∩
Ker(M DM D)) ⩽ 1 and we can fix a generator a = (a1 , . . . , an ) ∈ Fnp . Our goal is to prove a = 0.
We prove the even dimension case, for odd n the argument is exactly the same: the vector a satisfies the
following equations coming from M Aa = 0:
⎧
⎪
⎪ (m1,1 − 1)a1 + m1,2 a2 + ⋯ + m1, n2 a n2 + (−1)j+1 m1, n2 a n2 +1 + ⋯ + (−1)j+1 m1,1 an = 0
⎪
⎪
⎪
⎪
⎪ m2,1 a1 + (m2,2 − 1)a2 + ⋯ + m2, n2 a n2 + (−1)j+1 m2, n2 a n2 +1 + ⋯ + (−1)j+1 m2,1 an = 0
⎪
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪ m n2 ,1 a1 + m n2 ,2 a2 + ⋯ + (m n2 , n2 − 1)a n2 + (−1)j+1 m n2 , n2 a n2 +1 + ⋯ + (−1)j+1 m n2 ,1 an = 0
(24) ⎨ .
⎪
⎪
⎪ m n2 +1,1 a1 + m n2 +1,2 a2 + ⋯ + m n2 +1, n2 −1 a n2 −1 − a n2 +1 + ⋯ + (−1)j+1 m n2 +1,1 an = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ mn−1,1 a1 − an−1 + (−1)j+1 mn−1,1 an = 0
⎪
⎪
⎪
⎩ an = 0

Now put p(t) ∶= M Da ∈ Fp [t]n , then (with an = 0)

sn sn
⎛ p1 (t) ⎞ ⎛ m1,1 a1 ts1 + ⋯ + m1, n2 a n2 t 2 + (−1)j+1 m1, n2 a n2 +1 t 2 +1 + ⋯ + (−1)j+1 m1,2 an−1 tsn−1 ⎞
s1 sn j+1 s n +1 j+1 sn−1 ⎟
⎜ p2 (t) ⎟ ⎜ m 2,1 a 1 t + ⋯ + m 2, n a n t 2 + (−1) m 2, n an
+1 t 2 + ⋯ + (−1) (m 2,2 − 1)a n−1 t
⎜ ⎟ ⎜ 2 2 2 2 ⎟
⎜ ⋮ ⎟ ⎜ ⋮ ⎟
⎜ ⎟ ⎜ sn s n +1
⎟
⎜ p n (t) ⎟ ⎜ ⎜ m n
,1 a 1 t s1
+ ⋯ + m n n a n t 2 + (−1)
,
j+1
(m n n − 1)a n
, t 2 + ⋯ + (−1) j+1
m n
,2 an−1 tsn−1 ⎟
⎟
(25) p(t)= ⎜ 2 ⎟
⎜ p n +1 (t) ⎟ = ⎜ 2
s
2 2 2
j sn 2 2 +1
s2n +2 j+1
2
s ⎟.
⎜ 2 ⎟ ⎜ ⎜ m n
2 +1,1 a1 t 1
+ ⋯ + (−1) a n t 2 + mn
2 2 +1, n
2 −1 a n
2 +2 t 2 + ⋯ + (−1) m n
2 +1,2 a n−1 t n−1 ⎟
⎟
⎜ ⎟ ⎜ ⎟
⎜ ⋮ ⎟ ⎜ ⋮ ⎟
⎜ ⎟ ⎜ s1 j s2 ⎟
⎜ pn−1 (t) ⎟ ⎜ mn−1,1 a1 t + (−1) a2 t ⎟
⎝ pn (t) ⎠ ⎝ j
(−1) a1 t s1 ⎠

Since M Dp(t) = 0, we also have equations:

(26)
sn sn
⎧
⎪
⎪ m1,1 ts1 p1 (t) + ⋯ + m1, n2 t 2 p n2 (t) + (−1)j+1 m1, n2 t 2 +1 p n2 +1 (t) + ⋯ + (−1)j+1 (m1,1 − 1)tsn pn (t) = 0
⎪
⎪ sn sn
⎪
⎪
⎪ m2,1 ts1 p1 (t) + ⋯ + m2, n2 t 2 p n2 (t) + (−1)j+1 m2, n2 t 2 +1 p n2 +1 (t) + ⋯ + (−1)j+1 m2,1 tsn pn (t) = 0
⎪
⎪
⎪
⎪
⎪
⎪ ⋮
⎪ m n ,1 ts1 p1 (t) + ⋯ + m n , n ts n2 p n (t) + (−1)j+1 (m n , n − 1)ts n2 +1 p n +1 (t) + ⋯ + (−1)j+1 m n ,1 tsn pn (t) = 0
⎪
⎪
⎨ 2 2 2
s n2 2 2
sn 2 2 .
⎪
⎪
⎪ m n2 +1,1 ts1 p1 (t) + ⋯ + (−1)j t 2 p n2 (t) + (−1)j+1 m n2 +1, n2 −1 t 2 +2 p n2 +2 (t) + ⋯ + (−1)j+1 m n2 +1,1 tsn pn (t) = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ mn−1,1 ts1 p1 (t) + (−1)j ts2 p2 (t) + (−1)j+1 mn−1,1 tsn pn (t) = 0
⎪
⎪ j s
⎩ (−1) t 1 p1 (t) = 0
⎪

Note that in (26) we have polynomials in Fp [t], from now on we shall basically use the identity principle for
polynomials to solve the equations in the ai . From the last row in (26) we get p1 (t) = 0, i.e. comparing with
(25)
m1,1 a1 = m1,2 a2 = ⋯ = m1, n2 a n2 = m1, n2 a n2 +1 = ⋯ = m1,2 an−1 = 0 .
Substituting in the first and second-last equations in (24) we obtain

a1 = an−1 = 0

which also means that pn (t) = 0.

We can rewrite (24), (25) and (26) as
12 ANDREA BANDINI AND MARIA VALENTINO

⎧
⎪
⎪ (m2,2 − 1)a2 + ⋯ + m2, n2 a n2 + (−1)j+1 m2, n2 a n2 +1 + ⋯ + (−1)j+1 m2,3 an−2 = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ m n2 ,2 a2 + ⋯ + (m n2 , n2 − 1)a n2 + (−1)j+1 m n2 , n2 a n2 +1 + ⋯ + (−1)j+1 m n2 ,3 an−2 = 0
⎪
(27) ⎨ m n2 +1,2 a2 + ⋯ + m n2 +1, n2 −1 a n2 −1 − a n2 +1 + ⋯ + (−1)j+1 m n2 +1,3 an−2 = 0 ,
⎪
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪ mn−2,2 a2 − an−2 = 0
⎪
⎪
⎪
⎪
⎩ a1 = an−1 = an = 0

⎛ p1 (t) ⎞ ⎛ 0 ⎞
s2 sn j+1 sn
⎜ p2 (t) ⎟ ⎜ ⎜ m a
2,2 2 t + ⋯ + m n a
2, 2 2 n t 2 + (−1) m 2, n a n2 +1 t 2 +1 + ⋯ + (−1)j+1 m2,3 an−2 tsn−2 ⎟
⎜ ⎟ ⎜ 2 ⎟
⎜ ⋮ ⎟ ⎜ ⋮ ⎟
⎜ ⎟ sn s n +1
⎟
⎜ p n (t) ⎟ ⎜ m 2 ,2 a2 t + ⋯ + m 2 , 2 a 2 t 2 + (−1) (m 2 , n2 − 1)a n2 +1 t 2 + ⋯ + (−1)j+1 m n2 ,3 an−2 tsn−2
n
s2 n n n
j+1 n
⎟
(28) ⎜ 2 ⎟=⎜
⎜ sn sn
⎟
⎟
⎜ p n +1 (t) ⎟ ⎜ m n2 +1,2 a2 ts2 + ⋯ + (−1)j a n2 t 2 + m n2 +1, n2 −1 a n2 +2 t 2 +2 + ⋯ + (−1)j+1 m n2 +1,3 an−2 tsn−2 ⎟
⎜ 2 ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⋮ ⎟ ⎜ ⋮ ⎟
⎜ ⎟
⎜ pn−1 (t) ⎟ ⎜ ⎜ (−1)j a2 ts2
⎟
⎟
⎝ pn (t) ⎠ ⎝ 0 ⎠

and
(29)
sn sn
⎧
⎪
⎪ m1,2 ts2 p2 (t) + ⋯ + m1, n2 t 2 p n2 (t) + (−1)j+1 m1, n2 t 2 +1 p n2 +1 (t) + ⋯(−1)j+1 m1,2 tsn−1 pn−1 (t) = 0
⎪
⎪ sn sn
⎪
⎪
⎪ m2,2 ts2 p2 (t) + ⋯ + m2, n2 t 2 p n2 (t) + (−1)j+1 m2, n2 t 2 +1 p n2 +1 (t) + ⋯ + (−1)j+1 (m2,2 − 1)tsn−1 pn−1 (t) = 0
⎪
⎪
⎪
⎪
⎪
⎪ ⋮
⎪ sn sn
⎪ m n ,2 ts2 p2 (t) + ⋯ + m n , n t 2 p n (t) + (−1)j+1 (m n , n − 1)t 2 +1 p n +1 (t) + ⋯ + (−1)j+1 m n ,2 tsn−1 pn−1 (t) = 0
⎪
⎨ 2 2 2
sn 2 2 2
sn 2 2 .
⎪
⎪
⎪ m n2 +1,2 ts2 p2 (t) + ⋯ + (−1)j t 2 p n2 (t) + (−1)j+1 m n2 +1, n2 −1 t 2 +2 p n2 +2 (t) + ⋯ + (−1)j+1 m n2 +1,2 tsn−1 pn−1 (t) = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ (−1)j ts2 p2 (t) = 0
⎪
⎪
⎩ p1 (t) = pn (t) = 0
⎪

We repeat the same argument starting now from the second-last equation in (29), which yields p2 (t) = 0. This
means

m2,2 a2 = ⋯ = m2, n2 a n2 = m2, n2 a n2 +1 = ⋯ = m2,3 an−2 = 0,

which, substituted in the first equation of (27), gives a2 = 0. Thus (second-last equations in (27) and (28))
an−2 = 0 and pn−1 (t) = 0 as well.
Iterating the process we see that the specular symmetries between M D ((−1)j on the antidiagonal) and M A
(−1 on the diagonal) lead to a = 0. 

3.2. Direct sum.

1 1,old 1,new
Theorem 3.3. Assume that dimC∞ Im(δ1 ) = 1. Then Sk,m (Γ0 (t)) = Sk,m (Γ0 (t)) ⊕ Sk,m (Γ0 (t)).

Proof. We use the criterion of Theorem 2.12, where we noted that an element η ∈ Ker(D) must be in Im(δ1 ) =
Ker(M A) as well, and that Ut (η) is both old and new. We take a ∈ Fnp which verifies M Aa = 0 and represents
an element η = δ1 ϕ ∈ Ker(D), then T r′ (Ut (δ1 ϕ)) = 0, i.e. T F (M Da) = 0 and we prove that these two relations
yield a = 0, so that Ker(D) = 0 and D is invertible. As before we only treat the case of even n.
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 13

The equation M Aa = 0 gives again the system (24) (in particular an = 0), then, writing p(t) = M Da as in (25),
from T F (M Da) = 0 we get
sn sn
⎧
⎪
⎪ m1,1 ts1 p1 (t) + ⋯ + m1, n2 t 2 p n2 (t) + m1, n2 (−t) 2 +1 p n2 +1 (t) + ⋯ + m1,1 (−t)sn pn (t) = 0
⎪
⎪ sn sn
⎪
⎪
⎪
⎪ m2,1 ts1 p1 (t) + ⋯ + m2, n2 t 2 p n2 (t) + m2, n2 (−t) 2 +1 p n2 +1 (t) + ⋯ + m2,1 (−t)sn pn (t) = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪ sn sn
⎪
⎪ m n2 ,1 ts1 p1 (t) + ⋯ + m n2 , n2 t 2 p n2 (t) + m n2 , n2 (−t) 2 +1 p n2 +1 (t) + ⋯ + m n2 ,1 (−t)sn pn (t) = 0
(30) ⎨
⎪
⎪
⎪ m n2 +1,1 ts1 p1 (t) + ⋯ + m n2 +1,1 (−t)sn pn (t) = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ mn−2,1 ts1 p1 (t) + mn−2,2 ts2 p2 (t) + mn−2,2 (−t)sn−1 pn−1 (t) + mn−2,1 (−t)sn pn (t) = 0
⎪
⎪
⎪
⎩ mn−1,1 ts1 p1 (t) + mn−1,1 (−t)sn pn (t) = 0

(the matrix T can be taken from Remark 3.1).

In the last equation of (30) the term with the highest degree in t is mn−1,1 (−t)sn (−1)j a1 ts1 = −mn−1,1 a1 tk (note
that p1 (t) has degree at most sn−1 because an = 0): therefore mn−1,1 a1 = 0 and the second-last equation in (24)
tell us that an−1 = 0. Now (24) and (25) turn into

⎧
⎪
⎪ (m1,1 − 1)a1 + m1,2 a2 + ⋯ + m1, n2 a n2 + (−1)j+1 m1, n2 a n2 +1 + ⋯ + (−1)j+1 m1,3 an−2 = 0
⎪
⎪
⎪
⎪
⎪ m2,1 a1 + (m2,2 − 1)a2 + ⋯ + m2, n2 a n2 + (−1)j+1 m2, n2 a n2 +1 + ⋯ + (−1)j+1 m2,3 an−2 = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ m n2 ,1 a1 + m n2 ,2 a2 + ⋯ + (m n2 , n2 − 1)a n2 + (−1)j+1 m n2 , n2 a n2 +1 + ⋯ + (−1)j+1 m n2 ,3 an−2 = 0
⎪
(31) ⎨
⎪ m n2 +1,1 a1 + m n2 +1,2 a2 + ⋯ + m n2 +1, n2 −1 a n2 −1 − a n2 +1 + ⋯ + (−1)j+1 m n2 +1,3 an−2 = 0
⎪
⎪
⎪
⎪
⎪ ⋮
⎪
⎪
⎪
⎪
⎪ mn−2,1 a1 + mn−2,2 a2 − an−2 = 0
⎪
⎪
⎪
⎪
⎪ mn−1,1 a1 = 0
⎪
⎪ an−1 = an = 0
⎩

sn sn
⎛ p1 (t) ⎞ ⎛ m1,1 a1 ts1 + ⋯ + m1, n2 a n2 t 2 + (−1)j+1 m1, n2 a n2 +1 t 2 +1 + ⋯ + (−1)j+1 m1,3 an−2 tsn−2 ⎞
s1 sn j+1 s n +1 j+1 sn−2
⎜ p2 (t) ⎟ ⎜ m 2,1 a1 t + ⋯ + m 2, n a n t 2 + (−1) m 2, n an
+1 t 2 + ⋯ + (−1) m 2,3 an−2 t ⎟
⎜ ⎟ ⎜ ⎜
2 2
⋮
2 2 ⎟
⎟
⎜ ⋮ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ s s
⎜ p n (t) ⎟ ⎜ m n ,1 a1 ts1 + ⋯ + m n , n a n t 2 + (−1)j+1 (m n , n − 1)a n +1 t 2 + ⋯ + (−1)j+1 m n ,3 an−2 tsn−2 ⎟
n n +1
⎟
(32) p(t)= ⎜ 2 ⎟
⎜ p n +1 (t) ⎟ = ⎜ 2
s1
2 2 2
j n sn
2 2
s2n +2 j+1 n
2
sn−2 ⎟.
⎜ 2 ⎟ ⎜ m n
+1,1 a1 t + ⋯ + (−1) a t 2 + mn
+1, n
−1 a n
+2 t 2 + ⋯ + (−1) m +1,3 a n−2 t ⎟
⎜ ⎟ ⎜ ⎜ 2 2 2 2 2 2 ⎟
⎟
⎜ ⋮ ⎟ ⎜ ⋮ ⎟
⎜ ⎟ ⎜ j s2 ⎟
⎜ pn−1 (t) ⎟ ⎜ (−1) a2 t ⎟
⎝ pn (t) ⎠ ⎝ j
(−1) a1 t s1 ⎠

Now consider the second-last equation in (30)

mn−2,1 ts1 p1 (t) + mn−2,2 ts2 p2 (t) + mn−2,2 (−t)sn−1 pn−1 (t) + mn−2,1 (−t)sn pn (t) = 0.

The term with the highest possible degree s1 + sn = s2 + sn−1 = k is

mn−2,2 (−t)sn−1 (−1)j a2 ts2 + mn−2,1 (−t)sn (−1)j a1 ts1 = −(mn−2,2 a2 + mn−2,1 a1 )tk ,

hence mn−2,1 a1 + mn−2,2 a2 = 0. Looking at the system (31) we obtain an−2 = 0 and pn−3 (t) = mn−3,1 a1 ts1 + ⋯ +
(−1)j a4 ts4 (it is relevant that it has degree at most s4 = sn+1−(n−3) , i.e. that the terms of higher degree vanish).
The proof goes on in the same way: it may be less evident than the one of Theorem 3.2 (where the ai vanished
in couples), but looking always at the terms of degree k of the (n − i)-th equation of (30) we are able to prove
that an−i = 0 and, as an immediate consequence from (25), that pn−i−1 (t) has degree at most si+2 . For example
14 ANDREA BANDINI AND MARIA VALENTINO

midway through the proof we get

sn
⎛ m1,1 a1 ts1 + ⋯ + m1, n2 a n2 t 2 ⎞
⎛ a1 ⎞ ⎜ ⋮ ⎟
⎜ ⋮ ⎟ ⎜ s sn ⎟
⎜ ⎟ ⎜ m n a1 t 1 + ⋯ + m n n a n t 2
,1 ,
⎟
⎜ a n2 ⎟ ⎜ 2 2 2 2 ⎟
⎜ sn ⎟
a=⎜
⎜
⎟
⎟ and p(t) = ⎜ m n +1,1 a1 ts1 + ⋯ + (−1)j a n t 2 ⎟.
⎜ 0 ⎟ ⎜
⎜ mn
2
s j
2
s n −1 ⎟
⎟
⎜
⎜ ⋮
⎟
⎟ ⎜ 2 +2,1 a1 t 1 + ⋯ + (−1) a n2 −1 t 2 ⎟
⎜ ⎟
⎝ 0 ⎠ ⎜ ⋮ ⎟
⎝ (−1) a1 ts1
j ⎠

Therefore, what remains of (24) is

⎧
⎪ (m1,1 − 1)a1 + m1,2 a2 + ⋯ + m1, n2 a n2 = 0
⎪
⎪
⎪
⎪
⎪
⎪ m2,1 a1 + (m2,2 − 1)a2 + ⋯ + m2, n2 a n2 = 0
⎪
(33) ⎨ ⋮ .
⎪
⎪
⎪
⎪
⎪ m 2 ,1 a1 + m 2 ,2 a2 + ⋯ + (m 2 , 2 − 1)a 2 = 0
n n n n n
⎪
⎪
⎪
⎩ a n2 +1 = ⋯ = an = 0
n
Finally, we observe that the 2 -th equation of (30) is
sn s n +1
m n2 ,1 ts1 p1 (t) + ⋯ + m n2 , n2 t 2 p n2 (t) + m n2 , n2 (−t) 2 p n2 +1 (t) + ⋯ + m n2 ,1 (−t)sn pn (t) = 0 .
s n +1
As before, the term of degree k must have coefficient 0 and it appears only from m n2 , n2 (−t) 2 p n2 +1 (t) on, so
we get
m n2 , n2 a n2 + m n2 , n2 −1 a n2 −1 + ⋯ + m n2 ,1 a1 = 0

and, by (33), a n2 = 0 as well.

Iterating we get a = 0 and so our claim. 

References
[1] A. Bandini - M. Valentino On the Atkin Ut operator for Γ1 (t)-invariant Drinfeld cusp forms, Int. J. Number Theory 14
(2018), no. 10, 2599-2616.
[2] A. Bandini - M. Valentino On the Atkin Ut -operator for Γ0 (t)-invariant Drinfeld cusp forms, Proc. Amer. Math. Soc. 147
(2019), no. 10, 4171–4187.
[3] A. Bandini - M. Valentino On the structure and slopes of Drinfeld cusp forms, arXiv:1812.02032 [math:NT] (2018), submitted.
[4] G. Böckle An Eichler-Shimura isomorphism over function fields between Drinfeld modular forms and cohomology classes of
crystals, avaliable at
www1.iwr.uni-heidelberg.de/groups/arith-geom/home/members/gebhard-boeckle/publications/.
[5] G. Cornelissen A survey of Drinfeld modular forms, in [9], 167–187.
[6] F. Diamond, J. Shurman A first course in Modular Forms, GTM 228 (Springer, 2005).
[7] J. Fresnel - M. van der Put, Géométrie Analytique Rigide et Applications, Progress in Mathematics 18, (Birkhäuser, 1981).
[8] E.U. Gekeler Drinfeld modular curves, Lect. Notes Math. 1231 (Springer, 1986).
[9] Ed. E.U. Gekeler - M. van der Put - M. Reversat - J. van Geel Drinfeld modules, modular schemes and applications,
Proceedings of the workshop held in Alden-Biesen, September 9–14, 1996 (World Scientific Publishing Co., Inc., River Edge,
NJ, 1997).
[10] F. Gouvêa - B. Mazur, Families of modular eigenforms, Math. Comput. 58 (1992), no. 198, 793–805.
[11] S. Hattori Dimension variation of Gouvêa-Mazur type for Drinfeld cuspforms of level Γ1 (t), to appear in Int. Math. Res.
Not., https://doi.org/10.1093/imrn/rnz104
[12] S. Hattori ℘-adic continuous families of Drinfeld eigenforms of finite slope, arXiv:1904.08618 [math.NT] (2019).
[13] M.-H. Nicole - G. Rosso Familles de formes modulaires de Drinfeld pour le groupe général linéaire, arXiv:1805.08793
[math.NT] (2018).
[14] J. Teitelbaum The Poisson kernel for Drinfeld modular curves, J. Amer. Math. Soc. 4 (1991), no. 3, 491–511.
[15] C. Vincent On the trace and norm maps from Γ0 (m) to GL2 (A), J. Number Theory 142 (2014), 18–43.
 ON DRINFELD CUSP FORMS OF PRIME LEVEL 15

Andrea Bandini: Università di Pisa, Dipartimento di Matematica, Largo Bruno Pontecorvo 5, 56127 Pisa, Italy
E-mail address: [email protected]

Maria Valentino: Università della Calabria, Dipartimento di Matematica e Informatica, Ponte P. Bucci, Cubo
30B 87036 Rende (CS), Italy
E-mail address: [email protected]