CODIMENSION ONE INTERSECTIONS BETWEEN

COMPONENTS OF A MODULI STACK OF TWO-DIMENSIONAL

GALOIS REPRESENTATIONS

KALYANI KANSAL

Abstract. The Emerton-Gee stack for GL2 is a stack of (φ, Γ)-modules whose
reduced part X2,red can be viewed as a moduli stack of mod p representations
of a p-adic Galois group. We compute criteria for intersections of irreducible
components of X2,red in codimension 1 and relate them to extensions of Serre
weights.

Contents
1. Introduction 1
2. Extensions of Serre weights 4
3. Stack dimensions and extensions of GK characters 18
4. Computations of Serre weights 23
5. Type I intersections 29
6. Type II intersections 34
7. Conclusion 52
References 56

1. Introduction
Let p be an odd prime and let K/Qp be a finite extension, with ring of integers
OK , residue field k and absolute Galois group GK . In [EG1], Emerton and Gee
constructed and studied the stack of rank d étale (φ, Γ)-modules, denoted Xd . Over
Artinian coefficients, there exists an equivalence of categories between rank d étale
(φ, Γ)-modules and d-dimensional GK -representations that allows one to view Xd as
a moduli stack of Galois representations. The Emerton-Gee stack Xd is expected to
play a central role in the p-adic Langlands program, occupying the position played
by the moduli stack of L-parameters in the work of Fargues-Scholze on the classical
Langlands correspondence.
Specializing to d = 2, the reduced part of X2 , denoted X2,red , is an algebraic
stack defined over a finite field F. The irreducible components of X2,red are labelled
by Serre weights, which are the irreducible mod p representations of GL2 (k). By
[CEGS, Cor. 7.2], the labelling is in such a manner that if X2,red,σ is the component
labelled by σ, then its finite type points are precisely those representations that
have σ as a Serre weight, that is, they have crystalline lifts of Hodge-Tate weights
specified in a particular way by σ (see Section 1.4 for details).
1
2 KALYANI KANSAL

The main objective of this article is to compute criteria for pairs of Serre weights
σ and τ so that X2,red,σ ∩ X2,red,τ is a substack of codimension 1. Our strategy rests
on finding families of representations that have both σ and τ as Serre weights, and
therefore give points of X2,red,σ ∩X2,red,τ . The sizes of these families can then be used
to determine the dimension of X2,red,σ ∩ X2,red,τ . As employed in [EG1], a source of
families of representations is provided by extensions of fixed GK characters together
with extensions of their unramified twists. Every irreducible component of X2,red
can be obtained as the closure of such a family. Vector spaces of extensions of fixed
GK characters are typically [K : Qp ]-dimensional. Allowing various unramified
twists of the fixed characters adds 2 to the dimension, while 1 dimension is taken
away because a Gm orbit of an extension class gives the same representation and
yet another dimension is taken away because of a Gm worth of endomorphisms of
each extension. Thus a codimension 1 intersection of X2,red,σ and X2,red,τ may be
expected to correspond to the existence of a codimension 1 subfamily of extensions
of fixed GK characters (as well as their unramified twists) with both σ and τ
as their Serre weights. This line of investigation gives us the required criteria
(stated in Theorem 7.1). For pairs of non-isomorphic and weakly regular (a very
mild genericity hypothesis, see definition in Section 1.4) Serre weights, the criteria
are summarized below. The precise criteria for intersections involving components
labelled by Serre weights that are not weakly regular are significantly less succinct
and omitted from the statement below.

Theorem 1.1. If σ and τ are a pair of non-isomorphic Serre weights, then

Ext1F[GL2 (OK )] (σ, τ ) ̸= 0 =⇒ dim X2,red,σ ∩ X2,red,τ = [K : Qp ] − 1.

When σ and τ are also weakly regular, the following stronger statement is true:

Ext1F[GL2 (OK )] (σ, τ ) ̸= 0 ⇐⇒ dim X2,red,σ ∩ X2,red,τ = [K : Qp ] − 1.

This result can be motivated in terms of the conjectural categorical p-adic Lang-
lands correspondence. Specifically, it has been conjectured ([EGH, Conj. 6.1.6])
that there exists a fully faithful functor U from a derived category of smooth repre-
sentations of GL2 (K) to a derived category of coherent sheaves on X2 that witnesses
the p-adic local Langlands. The functor U is expected to satisfy properties related
to duality and support that imply the following:
GL (K)
• For σ a non-Steinberg Serre weight, the support of U(c-IndGL22 (OK ) σ) is
X2,red,σ .
GL (K)
• For σ, τ Serre weights and V ∈ Ext1F[GL2 (OK )] (σ, τ ), U ◦ c-IndGL22 (OK ) (τ →
V → σ) is a short exact sequence.
Therefore,

U ◦ c-IndGL22 (OK ) (V )|(X2,red,σ ∩X2,red,τ )c ∼

GL (K) GL (K)
= U ◦ c-IndGL22 (OK ) (σ ⊕ τ )|(X2,red,σ ∩X2,red,τ )c .

Since U is fully faithful, if the intersection of X2,red,σ and X2,red,τ is empty, then
GL (K) GL (K) GL (K)
c-IndGL22 (OK ) (V ) must be isomorphic to c-IndGL22 (OK ) σ ⊕ c-IndGL22 (OK ) τ . Thus, we
obtain the following diagram of GL2 (OK ) representations where the right downward
arrow splits:
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 3

GL (K)
σ c-IndGL22 (OK ) σ

GL (K)
V c-IndGL22 (OK ) V

The horizontal arrows split as maps of GL2 (OK ) representations by Mackey’s

decomposition theorem. The left vertical arrow must then split as well, and V must
be isomorphic to σ ⊕ τ . This shows that if the conjectured functor U exists, then
an empty intersection of X2,red,σ with X2,red,τ implies that there are no non-trivial
extensions of τ by σ as GL2 (OK ) modules. Our theorem is a finer variant of this
expectation.
In the course of our computations, we also find a cohomological criterion for
the number of components of dimension [K : Qp ] − 1 when X2,red,σ ∩ X2,red,τ is
codimension 1, along with some naturally occurring triples of Serre weights. The
theorem below summarizes the results for pairs of weakly regular Serre weights σ
and τ .
Theorem 1.2. Let σ and τ be two weakly regular Serre weights such that X2,red,σ ∩
X2,red,τ is of codimension 1. Then the following are true:
(i) When K is unramified over Qp , the number of components of dimension
[K : Qp ] − 1 in X2,red,σ ∩ X2,red,τ is 1. When K is ramified over Qp ,
this number is 2 if the GL2 (k)-extensions of τ by σ are non-trivial, and 1
otherwise.
(ii) When K is unramified over Qp , a component of dimension [K : Qp ] −
1 in X2,red,σ ∩ X2,red,τ does not lie in an intersection of three irreducible
components of X . In the ramified case, for sufficiently generic Serre weights
(c.f. Theorem 7.3), each component of dimension [K : Qp ] − 1 in X2,red,σ ∩
X2,red,τ lies in an intersection of three irreducible components of X2,red .
Note that the criterion that appears in Theorem 1.1 has to do with GL2 (OK )-
extensions, while the criterion that appears in Theorem 1.2 has to do with GL2 (k)-
extensions.

1.3. Outline of the paper. In Section 2, we compute explicit criteria for the ex-
istence of non-trivial extensions of Serre weights as GL2 (OK ) representations. In
Section 3, we relate the dimensions of families of GK -representations with both
σ and τ as Serre weights to the dimension of X2,red,σ ∩ X2,red,τ . We also relate
the number of sufficiently large families to the number of components of maximal
dimension inside X2,red,σ ∩ X2,red,τ . Section 4 recalls explicit criteria for computa-
tions of Serre weights of representations. Along with the results of Section 3, these
criteria are used to restructure the problem as that of finding σ and τ that satisfy
a precise computable relationship. Sections 5 and 6 compute the solution to the
problem laid out in Section 4. Finally, Section 7 collates all the findings.

1.4. Notation. Let p be a fixed prime and let K be a finite extension of Qp with
valuation ring OK , residue field k and uniformizer π. Eventually, p will be an odd
prime, to allow the key input of [CEGS, Cor. 7.2]. However, we will allow p = 2
for many of the intermediate steps.
4 KALYANI KANSAL

We let f := f (K/Qp ) and e := e(K/Qp ). Let GK be the absolute Galois group

of K, and IK the inertia group. F is a finite extension of Fp , with a fixed algebraic
closure F. F is taken to be sufficiently large so that all embeddings of k into F are
contained in F. i
Let T := [0, f − 1]. Fix an embedding σf −1 : k → F. Let σf −1−i := σfp−1 for
√
pf −1
i ∈ T . Let ωi be the GK character given by ωi (g) = σi ( g(pf −1√ π) ).
π
We let V⃗t,⃗s denote the irreducible GL2 (k) representation

f −1
O
(detti ⊗ Symsi k 2 ) ⊗k,σi F
i=0

where each si ∈ [0, p − 1]. All irreducible GL2 (k) representations with coefficients
in F are of this form and are called Serre weights. We can uniquely identify each
Serre weight by ⃗s and ⃗t if we demand that ti ∈ [0, p − 1] ∀i and at least one of the
ti ’s is not p − 1. Following [Gee], we say V⃗t,⃗s is weakly regular, if each si ∈ [0, p − 2].
We say that V⃗t,⃗s is Steinberg if each si equals p − 1.
Normalize Hodge-Tate weights in such a way that all Hodge-Tate weights of the
cyclotomic character are equal to −1. Consistent with the conventions in [EG1],
we say that a representation ρ : GK → GL2 (Fp ) has Serre weight V⃗t,⃗s if ρ has a
crystalline lift ρ : GK → GL2 (Qp ) that satisfies the following condition: For each
embedding σi : k ,→ F, there is an embedding σ̃i : K ,→ Qp lifting σi such that the
σ̃i labeled Hodge-Tate weights of r are {ti , si + ti + 1}, and the remaining (e − 1)f
pairs of Hodge-Tate weights of r are all {0, 1}. In this situation, we say V⃗t,⃗s ∈ W (ρ).
Let X2,red , or simply X , be the reduced part of the Emerton-Gee stack for two-
dimensional representations of GK . It is defined over F and is an algebraic stack
of pure dimension ef . The irreducible components of X are indexed by the non-
Steinberg Serre weights. For a non-Steinberg Serre weight V⃗t,⃗s , we denote the
corresponding irreducible component by XV⃗t,⃗s . If F′ is a finite field extension of F,
then XV⃗t,⃗s (F′ ) is the groupoid of representations ρ : GK → GL2 (F′ ) with V⃗t,⃗s ∈
W (ρ).
We will consider the si ’s and ti ’s associated to the Serre weight V⃗t,⃗s to have
indices in Z/f Z via the identification of the set T with a set of representatives of
Z/f Z. We will similarly consider the indexing set of the embeddings σi ’s to be
Z/f Z.

1.5. Acknowledgements. I thank my advisor David Savitt for sharing the prob-
lem with me along with his insights on it, as well as providing continual guidance
and support. I would also like to thank Karol Koziol and Brandon Levin for helpful
conversations.

2. Extensions of Serre weights

Denote by Γ the group GL2 (k), by K the group GL2 (OK ) and by Kn the group
1 + π n M2 (OK ) for n ∈ Z>0 . Our objective in this section is to compute when
extensions of Serre weights are non-trivial, for use later.
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 5

Nf −1 tj 2 f −1−j
We will sometimes write V⃗t,⃗s as j=0 (det ⊗ Symsj F )F r , where Γ acts on
sj 2
Sym F via the natural embedding Γ ,→ GL2 (F) induced by σf −1 . The exponen-
tiation by F rf −1−j denotes precomposition of the action of Γ by the (f − 1 − j)-th
power of the (arithmetic) Frobenius map.

Proposition 2.1. The conditions for non-triviality of Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) are given
as follows:
(i) If p > 2, f > 1, then Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 if and only if one of the following
two conditions are satisfied:
(a) ∃j ∈ Z/f Z such that s′i = si for i ̸= j − 1, j; s′j−1 = sj−1 − 1;
fP −1 −1
fP
s′j = p − sj − 2; and t′i pf −1−i ≡ ti pf −1−i + (sj + 1)pf −1−j
i=0 i=0
mod pf − 1.
(b) ∃j ∈ Z/f Z such that s′i = si for i ̸= j − 1, j; s′j−1 = sj−1 + 1;
−1
fP −1
fP
s′j = p − sj − 2; and t′i pf −1−i ≡ ti pf −1−i − (p − sj − 1)pf −1−j
i=0 i=0
mod pf − 1.
(ii) If p > 2, f = 1, then Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 if and only if one of the following
two conditions are satisfied:
(a) s0 < p − 2; s′0 = p − s0 − 3; and t′0 ≡ t0 + s0 + 1 mod p − 1.
(b) s0 ̸= 0, p − 1; s′0 = p − s0 − 1; and t′0 ≡ t0 + s0 mod p − 1.
(iii) If p = 2, f > 1, then Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 if and only if the central charac-
ters for V⃗t,⃗s and Vt⃗′ ,s⃗′ are the same, as well as, ∃j ∈ Z/f Z such that s′i = si
for i ̸= j − 1, j; s′j−1 = sj−1 ± 1; and s′j = p − sj − 2.
(iv) If p = 2, f = 1, then Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 if and only if s′0 = s0 = 0; and
t′0 = t0 .
Moreover, Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) always has dimension ≤ 1.

Proof. In order to compute Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ), we need to compute the second socle
layer of the injective hull of Vt⃗′ ,s⃗′ . Note that if an F-vector space V with Γ action
is injective as an SL2 (k) module, then it is also injective as a Γ module. This is
because any SL2P (k) module map ϕ can be lifted to a Γ module map by replacing it
with [Γ:SL12 (k)] g(ϕ). Therefore, we need to find a Γ module containing Vt⃗′ ,s⃗′ ,
g∈G/H
so that it is the injective hull of Vt⃗′ ,s⃗′ as an SL2 (k) module and compute its second
socle layer.
Beyond this point, the steps are precisely as in [AJL], while carefully tracking
through the twists by powers of the determinant. The final result (stated in the
proposition) is then obtained in the same manner as [AJL, Cor. 4.5].
For p = 2, note that if V is a non-trivial Γ extension of V⃗t,⃗s by Vt⃗′ ,s⃗′ , then the
central characters of V⃗t,⃗s and Vt⃗′ ,s⃗′ are the same (this holds true for any p). This
is because F[X (Γ)] is semisimple, where X (Γ) is the center of Γ. Therefore, by
twisting by a square root of the central character (possible since p = 2), V can
be assumed to be a non-trivial F[PGL2 (k)] = F[SL2 (k)] extension. And therefore,
Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 implies that ExtF[SL
1
2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0. On the other hand,
every non-trivial F[SL2 (k)] extension is a non-trivial F[PGL2 (k)] extension, and
6 KALYANI KANSAL

therefore, a non-trivial Γ extension. It follows that Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 ⇐⇒

Ext1F[SL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0. The conditions for the latter are described in the
paragraph preceding Corollary 4.5 in [AJL]. □
Remark 2.2. Ext1F[SL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0 implies Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0. This is
because any F[SL2P(k)] splitting ϕ can be upgraded to a Γ splitting by replacing it
with [Γ:SL12 (k)] g(ϕ).
g∈Γ/SL2 (k)

In order to compute Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ), we will use the Grothendieck spectral
sequence. Let σ be a F[Γ] representation, seen via inflation as a F[K] representation.
The Grothendieck spectral sequence gives us the following left exact sequence:

(2.2.1) 0 → Ext1Γ (V⃗t,⃗s , σ) → Ext1K (V⃗t,⃗s , σ) → HomΓ (V⃗t,⃗s , H 1 (K1 , σ))

By [BP, Prop. 5.1], we have the following description of H 1 (K1 , σ).
Proposition 2.3. (i)
f −1 d
f −1−i
H 1 (K1 , σ) ∼
M M
= σ ⊗ (V2 ⊗ det−1 )F r σ
i=0 i=1
2
where V2 is the subspace spanned by 2i x̃i ỹ 2−i in Sym2 F where Γ acts via

the embedding Γ ,→ GL2 (F) induced by σf −1 ;

d = dimF Hom(1 + πOK , F) for p ̸= 2 and d = dimF Hom(1 + πOK , F) − f
for p = 2.
f −1−i
(ii) Under the above isomorphism, an element of σ ⊗(V2 ⊗det−1 )F r can be
seen explicitly as a map (cocycle) K1 → σ via the following correspondence:
f −1−i
α ⊗ x̃2 ∈ σ ⊗ (V2 ⊗ det−1 )F r ↭ κli α : K1 → σ
f −1−i
α ⊗ 2x̃ỹ ∈ σ ⊗ (V2 ⊗ det−1 )F r ↭ ϵi α : K1 → σ
f −1−i
α ⊗ ỹ 2 ∈ σ ⊗ (V2 ⊗ det−1 )F r ↭ κui α : K1 → σ
where
 
1 + πa πb
κli : ∈ K1 7→ σi (c) ∈ F
πc 1 + πd
 
1 + πa πb
ϵi : ∈ K1 7→ σi (a − d) ∈ F
πc 1 + πd
 
1 + πa πb
κui : ∈ K1 7→ σi (a) ∈ F
πc 1 + πd
Ld 1
(iii) The summand i=1 σ ⊂ H (K1 , σ) corresponds to maps K1 → σ that
factor through the determinant and are not given by any of the cocyles
Lf −1 f −1−i
appearing in i=0 σ ⊗ (V2 ⊗ det−1 )F r .
Corollary 2.4. Ext1K (V⃗t∨,⃗s , Vt⃗∨′ ,s⃗′ ) ̸= 0 if and only if Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.

Proof. Using (2.2.1), Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 implies that either Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0,
or HomΓ (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ ) ̸= 0.
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 7

Either way, the central character is the same for V⃗t,⃗s and Vt⃗′ ,s⃗′ . This is auto-
matically true if Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 because the group algebra of the center of Γ
is semisimple. If HomΓ (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ )) ̸= 0, we use the description in Propo-
sition 2.3 and the fact that V2 ⊗ det−1 has trivial central character. Therefore
−1
fP −1
fP
pf −1−j (2tj + sj ) ≡ pf −1−j (2t′j + s′j ) mod pf − 1.
j=0 j=0
−1
fP
− pf −1−j (2tj +sj )
Twisting by det j=0
, we obtain:
Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0
⇐⇒ Ext1K (V−⃗t−⃗s,⃗s , V−t⃗′ −s⃗′ ,s⃗′ ) = Ext1K (V⃗t∨,⃗s , Vt⃗∨′ ,s⃗′ ) ̸= 0
□
Corollary 2.5. Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 if and only if Ext1K (Vt⃗′ ,s⃗′ , V⃗t,⃗s ) ̸= 0.
Proof.
Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0
⇐⇒ Ext1K (V⃗t∨,⃗s , Vt⃗∨′ ,s⃗′ ) ̸= 0 (by Corollary 2.4)
⇐⇒ Ext1K (Vt⃗′ ,s⃗′ , V⃗t,⃗s ) ̸= 0 (by taking duals)
□
Proposition 2.6. Let V⃗t,⃗s and Vt⃗′ ,s⃗′ be a pair of non-isomorphic, non-Steinberg
Serre weights. One shows up in the first K1 group cohomology of the other if and
only if, after interchanging V⃗t,⃗s and Vt⃗′ ,s⃗′ if necessary, there exists i ∈ {0, ..., f − 1}
such that
(i) For p > 2,
• si = s′i + 2,
• For j ̸= i, sj = s′j and
P f −1−j
tj ≡ −pf −1−i +
P f −1−j ′
• p p tj mod pf − 1.
j∈T j∈T
(ii) For p = 2,
• si = s′i + 1,
• For j ̸= i, sj = s′j and
• The central characters of the two Serre weights are the same.
Further, for a pair of such non-isomorphic, non-Steinberg Serre weights, the
multiplicity of appearance of one in the first K1 group cohomology of the other is at
most 1.
Proof. The proof for p > 2 is covered by Proposition 5.4 and Corollary 5.5 in [BP]).
For p = 2, we first make the following observation. If HomΓ (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ )) ̸=
0, we can twist both sides by the square root of the central character and obtain
an inclusion of V⃗t,⃗s into H 1 (K1 , Vt⃗′ ,s⃗′ ) as PGL2 (k) = SL2 (k) representations. On
the other hand, suppose V⃗t,⃗s ,→ H 1 (K1 , Vt⃗′ ,s⃗′ ) as SL2 (k) representations and the
central characters of V⃗t,⃗s and Vt⃗′ ,s⃗′ are the same. Then this inclusion is easily seen
to be an inclusion as Γ-representations.
Therefore, assuming the central characters of V⃗t,⃗s and Vt⃗′ ,s⃗′ are the same, we only
need to find criteria for inclusion of V⃗t,⃗s in H 1 (K1 , Vt⃗′ ,s⃗′ ) as SL2 (k) representations.
8 KALYANI KANSAL

To emphasize disregarding the determinant twists, we will denote by L(⃗r) or by

−1 2 f −1−j
L( pf −1−j rj ) the irreducible SL2 (k) representation ⊗fj=0 (Symrj F )F r
P
where
2
rj ∈ [0, p − 1] for each j and SL2 (k) acts on Symrj F via σf −1 : SL2 (k) ,→ SL2 (F).
Lf −1 f −2−i Ld
By Proposition 2.3, H 1 (K1 , σ) ∼ = i=0 (L(s⃗′ ) ⊗ V2F r ) i=1 L(s⃗′ ) as SL2 (k)
representations. As L(⃗s) ̸∼ = L(s⃗′ ), we need to understand when L(⃗s) embeds into
⃗ F r f −2−i ∼ ⃗ f −1−i
′
L(s ) ⊗ V2 = L(s ) ⊗ L(1)F r
′ for a given i.
f −1−i
(1) si = 0. Then L(s⃗ ) ⊗ L(1)
′ ′ F r
is irreducible and isomorphic to L(⃗s),
where si = 1 and sj = s′j for j ̸= i.
(2) s′i = 1. Then

f −1−i f −1−0 f −i f −1−i

L(s⃗′ ) ⊗ L(1)F r ∼
= L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗ (L(1) ⊗ L(1))F r
f −2−i 0
⊗L(s′i+1 )F r ⊗ · · · ⊗ L(s′f −1 )F r

f −1−0 f −i f −1−i
∼
= L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗ Q1 (0)F r
f −2−i 0
⊗L(s′i+1 )F r ⊗ · · · ⊗ L(s′f −1 )F r

where Q1 (0) is a self-dual representation of Loewy length 3, with composition fac-

tors L(0), L(2), L(0) by [AJL, Lem. 3.1]. In fact, [AJL, Lem. 3.1] says that Q1 (0)
2 2 f −1−i
is a direct summand of (Sym1 F ⊗ Sym1 F )F r , but by comparing dimensions,
they are equal. As SL2 (F) representations:

f −1−0 f −i f −1−i
L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗ L(0)F r
f −2−i 0
⊗L(s′i+1 )F r ⊗ · · · ⊗ L(s′f −1 )F r
f −1−0 f −i f −1−i 
L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗ Q1 (0)F r

,→ soc f −2−i 0
⊗L(s′i+1 )F r ⊗ · · · ⊗ L(s′f −1 )F r
f −1−0 f −i f −1−i 
Q1 (s′0 )F r ⊗ · · · ⊗ Q1 (s′i−1 )F r ⊗ Q1 (0)F r

,→ soc f −2−i 0
⊗Q1 (s′i+1 )F r ⊗ · · · ⊗ Q1 (s′f −1 )F r
f −1−0 f −i f −1−i
L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗ L(0)F r
∼
= f −2−i 0
⊗L(s′i+1 )F r ⊗ · · · ⊗ L(s′f −1 )F r

The isomorphism in the last step is by [AJL, Lem. 3.4]. Using [AJL, Cor. 4.2]
and the assumption that L(⃗s) is not Steinberg, we conclude that L(⃗s) embeds
f −1−i f −1−0 f −i
into L(s⃗′ ) ⊗ L(1)F r if and only if L(⃗s) ∼= L(s′0 )F r ⊗ · · · ⊗ L(s′i−1 )F r ⊗
F r f −1−i ′ F r f −2−i ′ F r0
L(0) ⊗ L(si+1 ) ⊗ · · · ⊗ L(sf −1 ) .
□

Remark 2.7. The explicit calculation of H 1 (K1 , Vt⃗′ ,s⃗′ ) in [BP, Prop. 5.1] shows
that the inflation homomorphism H 1 (K1 /K2 , Vt⃗′ ,s⃗′ ) → H 1 (K1 , Vt⃗′ ,s⃗′ ) is an isomor-
phism, since each K1 cocycle is in fact a K1 /K2 cocycle. By the construction of the
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 9

Grothendieck spectral sequence, this means exactly that

injK (Vt⃗′ ,s⃗′ )K2 /Vt⃗′ ,s⃗′ injK (Vt⃗′ ,s⃗′ )/Vt⃗′ ,s⃗′
socΓ ( ) = socΓ ( )
injK (Vt⃗′ ,s⃗′ )K1 /V t⃗′ ,s⃗′ injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′
Lemma 2.8. Let V⃗t,⃗s and Vt⃗′ ,s⃗′ be a pair of Serre weights. Then the natural map
Ext1K/K2 (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) → Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) is an isomorphism.

Proof. For a group G with an F-representation σ, let injG (σ) denote the injective
hull of σ as a smooth F[G] module. Then, for each Serre weight σ, injK (σ)Kn
is an injective K/Kn module. By injectivity of injK/Kn (σ), there exists a map
injK (σ)Kn → injK/Kn (σ). The kernel of this map must be trivial, by the hull
property of injK (σ). By the hull property of injK/Kn (σ), it is forced to be an
isomorphism. We will henceforth use injK (σ)Kn as the injective hull of σ as a
K/Kn representation.
The explicit calculation of H 1 (K1 , Vt⃗′ ,s⃗′ ) in [BP, Prop. 5.1 ] shows that the in-
flation homomorphism H 1 (K1 /K2 , Vt⃗′ ,s⃗′ ) → H 1 (K1 , Vt⃗′ ,s⃗′ ) is an isomorphism, since
each K1 cocycle is in fact a K1 /K2 cocycle. By the construction of the Grothendieck
spectral sequence, this means exactly that
 K1   K1 
K2
inj (V ⃗
K t ,s′ ⃗′ ) /V ⃗′ ⃗
t ,s ′ inj (V ⃗
K t ,s′ ⃗′ )/V ⃗′ ⃗
t ,s′
(2.8.1) socΓ   = socΓ 
   
injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′ injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′


 K
1
injK (Vt⃗′ ,s⃗′ )K2 /Vt⃗′ ,s⃗′
Now, suppose V⃗t,⃗s lives inside the Γ socle of injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′
with multiplic-
1
ity n. Equivalently, HomΓ (V⃗t,⃗s , H (K/K2 , Vt⃗′ ,s⃗′ )) is n-dimensional. Let N denote
 K !
K2 1
injK (V ⃗′ ⃗′ ) /V ⃗′ ⃗′
the preimage in injK (Vt⃗′ ,s⃗′ )K2 /Vt⃗′ ,s⃗′ of V⃗t⊕n
,⃗
s
⊂ socΓ t ,s t ,s
injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′
.
   
Suppose, further, that L ∼ = V⃗t⊕l
,⃗
s
⊂ soc Γ inj (V ⃗
K t ,s
′ ⃗′ )/V ⃗′ ⃗
t ,s′ −soc Γ inj (V ⃗
K t ,s′ ⃗′ ) K2
/V t ,s .
⃗′ ⃗′
 K !
1
injK (Vt⃗′ ,s⃗′ )/Vt⃗′ ,s⃗′
Then, the preimage inside injK (Vt⃗′ ,s⃗′ )/Vt⃗′ ,s⃗′ of V⃗t⊕n
,⃗
s
⊂ socΓ injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′

contains L + N . As L ̸⊂ N , L + N = L ⊕ N , and the multiplicity of V⃗t,⃗s in

 K !
1
injK (Vt⃗′ ,s⃗′ )/Vt⃗′ ,s⃗′
socΓ injK (Vt⃗′ ,s⃗′ )K1 /Vt⃗′ ,s⃗′
is ≥ l + n, implying l = 0 by (2.8.1). Therefore,

socΓ (injK (Vt⃗′ ,s⃗′ )K2 /Vt⃗′ ,s⃗′ ) ,→ socΓ (injK (Vt⃗′ ,s⃗′ )/Vt⃗′ ,s⃗′ ) is an equality.
□
Since K/K2 is a finite group, Ext1K/K2 (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ∼
= H 1 (K/K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ). The
Grothendieck spectral sequence gives us the following left exact sequence:

(2.8.2)
inf res
0 → H 1 (Γ, V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ) −−→ H 1 (K/K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ) −−→ H 1 (K1 /K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ )Γ
Proposition 2.9. Suppose e > 1. Then the res map in (2.8.2) is a split surjection.
Proof. e > 1 implies that p ∈ π 2 OK . Let OK
ur
be the ring of integers for the maximal
unramified subextension inside E over Qp . Therefore k ∼ = OK ur
/p ,→ OK /π 2 . This
10 KALYANI KANSAL

gives a splitting of the natural surjection GL2 (OK /π 2 ) ↠ GL2 (k) ∼ ur

= GL2 (OK /p).
We obtain the following split exact sequence:

1 K1 /K2 K/K2 Γ 1

Therefore, K/K2 = ∼ K1 /K2 ⋊ Γ. For b ∈ Γ and a ∈ K1 /K2 , denote bab−1 by ab .

Suppose σ is a Γ representation (seen via inflation as a K/K2 representation) and
ψ is a cocycle representing a nonzero element of H 1 (K1 /K2 , σ)Γ . As K1 /K2 action
is trivial on σ, H 1 (K1 /K2 , σ)Γ = Z 1 (K1 /K2 , σ)Γ . Γ-invariance means precisely that
for b ∈ Γ and a ∈ K1 /K2 , b−1 ψ(ab ) = ψ(a).
We define a function δ on K1 /K2 ⋊ Γ by setting δ((a, b)) equal to ψ(a). I claim
that δ is a cocyle, i.e., δ((a, b)(a′ , b′ )) = δ((a, b)) + (a, b) · δ((a′ , b′ )). Evaluation of
the left hand side gives us:

L.H.S. = δ((aa′b , bb′ ))

= ψ(aa′b )
= ψ(a) + ψ(a′b )

Evaluation of the right hand side gives us:

R.H.S. = ψ(a) + (a, 1)(1, b) · ψ(a′ )

= ψ(a) + (1, b) · ψ(a′ )
= ψ(a) + (1, b) · ((1, b−1 ) · ψ(a′b )) (as ψ is Γ-invariant)
′b
= ψ(a) + ψ(a )
= L.H.S.

This establishes that δ is a cocyle and therefore, res map in (2.8.2) is a split
surjection.
□

Corollary 2.10. If e > 1,

Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ⊕ HomΓ (V⃗t,⃗s , H 1 (K1 /K2 , Vt⃗′ ,s⃗′ )).

Proof. This is immediate from Proposition 2.9 and the fact that H 1 (K1 /K2 , V⃗t∨,⃗s ⊗
V⃗′ ⃗′ )Γ ∼
t ,s = HomΓ (V⃗ , H 1 (K1 /K2 , V⃗′ ⃗′ )) by the explicit description in Proposi-
t,⃗
s t ,s
tion 2.3. □

Lemma 2.11. Let p > 2, r ≤ p − 3. Then the following are true:

2 2 2
(i) Symr+2 F embeds into Sym2 F ⊗ Symr F as a direct summand of multi-
plicity 1.
2
(ii) Let the obvious basis of Symr+2 F be given by {wk z r+2−k }k∈[0,r+2] . Fur-
2 2
ther, let a basis of Sym2 F ⊗Symr F be given by {x̃j ỹ 2−j ⊗xk y r−k }(j,k)∈[0,2]×[0,r]
if r > 0, and by {x̃j ỹ 2−j ⊗ 1}j∈[0,2] if r = 0. The embedding is given
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 11

(uniquely upto scalar multiplication) as follows:

k(k − 1)
wk z r+2−k 7→ x̃2 ⊗ xk−2 y r+2−k
(r + 2)(r + 1)
k(r + 2 − k) k−1 r+1−k
+2x̃ỹ ⊗ x y
(r + 2)(r + 1)
(r + 2 − k)(r + 1 − k) k r−k
+ỹ 2 ⊗ x y for k ∈ [2, r]
(r + 2)(r + 1)
1 r r
wz r+1 7→ 2x̃ỹ ⊗ y + ỹ 2 ⊗ xy r−1 if r > 0
r+2 r+2
r 1
wr+1 z 7→ x̃2 ⊗ xr−1 y + 2x̃ỹ ⊗ xr if r > 0
r+2 r+2
wz 7→ x̃ỹ ⊗ 1 if r = 0
r+2 2 r
w 7→ x̃ ⊗ x
r+2
z 7→ ỹ 2 ⊗ y r

Proof. The first statement is from [BP, Prop. 5.4]. The second statement can be
verified by direct computation. □

f −1 kj sj −kj
Lemma 2.12. Let V⃗t,⃗s , a Serre weight. Denote by {⊗j=0 w z }(kj )j the ob-
vious basis of V⃗t,⃗s . Use the same notation to denote a basis of V−t−s,⃗
⃗ s . Then
V⃗t∨,⃗s ∼
= V−t−s,⃗
⃗ s under the following map:

V⃗t∨,⃗s → V−t−s,⃗
⃗ s
 
sj
⊗j (wkj z sj −kj )∨ 7→ ⊗j wsj −kj (−z)kj
kj

Proof. By direct computation. □

Lemma 2.13. Let p > 2. Consider a pair of non-isomorphic, non-Steinberg Serre

weights V⃗t,⃗s and Vt⃗′ ,s⃗′ satisfying the condition in Proposition 2.6, that is, si = s′i +2,
sj = s′j for j ̸= i and
P f −1−j
tj ≡ −pf −1−i +
P f −1−j ′
p p tj mod pf − 1.
j∈T j∈T
−1 f −1 kj sj −kj −1 kj sj −kj
Denote by ⊗fj=0 (wkj z sj −kj )∨ the dual of ⊗j=0 w z , where {⊗fj=0 w z }(kj )j
f −1 kj′ s′j −kj′
gives a basis of V⃗t,⃗s . Let the basis of Vt⃗′ ,s⃗′ be given by {⊗j=0 x y }(kj )j .
′

Then the Γ-invariant cocyles of H 1 (K1 /K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ) ∼

= H 1 (K1 /K2 , V−⃗t−⃗s,⃗s ⊗
Vt⃗′ ,s⃗′ ) are a 1-dimensional subspace spanned by

κli A + ϵi B + κui C

where κli , ϵi and κui are homomorphisms K1 /K2 → F defined in Proposition 2.3 and
A, B and C are elements of V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ defined below.
For s′i > 0,
12 KALYANI KANSAL

 
 
X sj ′ O
A = − (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj )
kj
(kj )j̸=i
 ′ 
si   
X si ki (ki − 1) ki −2 s′i +2−ki 
 wsi −ki (−z)ki ⊗ ′ ′ + 1) x y

ki =2 k i (si + 2)(si


 
′

 s ′

 + si w(−z)si −1 ⊗ ′ i xsi −1 y 
s +2
 
i
 
 ′
 
+ (−z)si ⊗ xsi

 
 
X s j ′ O
B =  (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj )
kj
(kj )j̸=i
 ′ 
si   ′

X s i ki (s + 2 − k i ) ′
 wsi −ki (−z)ki ⊗ ′ i xki −1 y si +1−ki 

ki =2 ki (si + 2)(s′i + 1) 

   

si −1 1 s′i 

 + si w (−z) ⊗ ′ y
si + 2
 
 
  
1 ′
+ si w(−z)si −1 ⊗ ′ xsi
 
si + 2

 
 
X sj ′ O
C =  (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj )
kj
(kj )j̸=i
 ′ 
si   ′ ′

X si (s + 2 − ki )(s + 1 − k i ) ′
 wsi −ki (−z)ki ⊗ i i
xki y si −ki 

ki =2 ki (s′i + 2)(s′i + 1) 

 
′
 

si −1 si s′i −1 

 + si w (−z) ⊗ ′ xy
si + 2
 
 
   
si s′i
+ (w ⊗ y

For s′i = 0,

 
 
X sj ′ O
A = − (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj ) ((−z)si ⊗ 1)
kj
(kj )j̸=i
 
 
X s j ′ O
B =  (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj ) (−wz ⊗ 1)
kj
(kj )j̸=i
 
 
X s j ′ O
C =  (⊗j̸=i wsj −kj (−z)kj ) ⊗ (⊗j̸=i xkj y sj −kj ) (wsi ⊗ 1)
kj
(kj )j̸=i
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 13

 Γ
Proof. The first step is to compute H 1 (K1 /K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ) . By Proposition 2.3,
Lf −1 f −1−i Ld
this group is isomorphic to HomΓ (V⃗t,⃗s , i=0 Vt⃗′ ,s⃗′ ⊗(V2 ⊗det−1 )F r i=1 V ⃗′ ⃗′ ).
Lf −1 t ,s
Using Lemma 2.11, V⃗t,⃗s has a unique (upto scalars) embedding into i=0 Vt⃗′ ,s⃗′ ⊗
f −1−i Ld
(V2 ⊗ det−1 )F r i=1 Vt⃗′ ,s⃗′ . This embedding may be written as an element of
f −1−i
V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ⊗ (V2 ⊗ det−1 )F r ⊂ H 1 (K1 /K2 , V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ ). Employing Proposi-
tion 2.3 to further write this element as an explicit map K1 /K2 → V⃗t∨,⃗s ⊗ Vt⃗′ ,s⃗′ , we
obtain the following values of A, B and C:

For s′i > 0,

 
X ′ O
A = − (⊗j̸=i (wkj z sj −kj )∨ ) ⊗ (⊗j̸=i xkj y sj −kj )
(kj )j̸=i
s′i
 
X ki (ki − 1)

ki −2 s′i +2−ki 
 (wki z si −ki )∨ ⊗ ′ ′ + 1) x y

ki =2 (si + 2)(si


 
′

 s ′

 + (wsi −1 z)∨ ⊗ ′ i xsi −1 y 
s +2
 
i
 
 ′
 
+ (wsi )∨ ⊗ xsi

 
s′j −kj
X O
B =  (⊗j̸=i (wkj z sj −kj )∨ ) ⊗ (⊗j̸=i xkj y )
(kj )j̸=i
s′i
 
X ′

k i (s + 2 − ki ) ′
 (wki z si −ki )∨ ⊗ ′ i xki −1 y si +1−ki 

ki =2 (si + 2)(s′i + 1) 

  
 1 ′

 + (wz si −1 )∨ ⊗ ′ y si 
si + 2
 
 
  
1 ′
+ (wsi −1 z)∨ ⊗ ′ xsi
 
si + 2

 
s′j −kj
X O
C =  (⊗j̸=i (wkj z sj −kj )∨ ) ⊗ (⊗j̸=i xkj y )
(kj )j̸=i
s′i
 
X ′ ′

(s + 2 − k i )(s + 1 − k i ) ′
 (wki z si −ki )∨ ⊗ i i
xki y si −ki 

ki =2 (s′i + 2)(s′i + 1) 

 
′
 

si −1 ∨ si s′i −1 

 + wz ) ⊗ ′ xy
s +2
 
i
 
  
si ∨ s′i
+ (z ) ⊗ y

For s′i = 0,
14 KALYANI KANSAL

 
s′j −kj
X O
A = − (⊗j̸=i (wkj z sj −kj )∨ ) ⊗ (⊗j̸=i xkj y ) ((wsi )∨ ⊗ 1)
(kj )j̸=i
 
O 1

s′j −kj
X
kj sj −kj ∨ kj ∨
B =  (⊗j̸=i (w z ) ) ⊗ (⊗j̸=i x y ) (wz) ⊗
2
(kj )j̸=i
 
s′j −kj
X O
C =  (⊗j̸=i (wkj z sj −kj )∨ ) ⊗ (⊗j̸=i xkj y ) ((z si )∨ ⊗ 1)
(kj )j̸=i

Using Lemma 2.12, we can rewrite elements in V⃗t∨,⃗s as elements of V−⃗t−⃗s,⃗s , giving
us the desired answer. □

Our next order of business is to check if a Γ-invariant cocycle in H 1 (K1 /K2 , V⃗t∨,⃗s ⊗
Vt⃗′ ,s⃗′ ) ∼
= H 1 (K1 /K2 , V−⃗t−⃗s,⃗s ⊗ Vt⃗′ ,s⃗′ ) is in the image of the res map in (2.8.2).
Therefore, we will try and extend such a cocyle to K. However, instead of extending
it to all of K, we will first focus our attention on extending it to the subgroup of K
generated by the upper unipotent and diagonal matrices.

Proposition 2.14. Let e = 1. Consider a pair of non-isomorphic, weakly reg-

ular Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ satisfying the conditions in Proposition 2.6. In
particular,
HomΓ (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ ) ̸= 0

Then res is the zero map and inf in (2.8.2) is an isomorphism, implying that
Ext1K (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 iff Ext1Γ (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.

Proof. We will show the following stronger result. Let p > 2, and let V⃗t,⃗s and Vt⃗′ ,s⃗′
satisfy:

si = s′i + 2,
sj = s′j for j ̸= i,
s′i+1 < p − 1, and
X X
pf −1−j tj ≡ −pf −1−i + pf −1−j t′j mod pf − 1.
j∈T j∈T

Then res is the zero map and inf in (2.8.2) is an isomorphism. (Note that the
above conditions on p, ⃗s and s⃗′ are automatically implied by the hypotheses in the
statement of the Proposition.)
We will assume without loss of generality that ⃗t′ = ⃗0.
Our proof will show that there is no way to extend a Γ-invariant cocycle ψ ∈
H 1 (K1 /K2 , V−⃗t−⃗s,⃗s ⊗ Vt⃗′ ,s⃗′ ) simultaneously to upper unipotent and diagonal ma-
trices with 1 in the bottom right  entry.
 We denote thesetwo groups by U and D
1 α t 0
respectively. We let U (α) = , and D(t) = .
0 1 0 1
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 15

First, we give a basis of V−⃗t−⃗s,⃗s ⊗ Vt⃗′ ,s⃗′ . We note that:

V−⃗t−⃗s,⃗s ⊗ Vt⃗′ ,s⃗′ ∼
=
 ′ ′ 2 f −1 ′ ′ 2 f −i

(det−s0 ⊗ Syms0 F )F r ⊗ · · · ⊗ (det−si−1 ⊗ Symsi−1 F )F r
 
′ ′ 2 f −1−i ′ ′ 2 f −2−i
⊗(det−si −1 ⊗ Symsi +2 F )F r ⊗ (det−si+1 ⊗ Symsi+1 F )F r ⊗ . . .
 
 
′ ′ 2 0
⊗(det−sf −1 ⊗ Symsf −1 F )F r
 
′ 2 f −1 ′ 2 f −2
(det0 ⊗ Syms0 F )F r ⊗ (det0 ⊗ Syms1 F )F r ⊗ ...
⊗ ′ 2 0

⊗(det0 ⊗ Symsf −1 F )F r
This can be viewed as a tensor of 2f terms, each term being a tensor of a
determinant power and a symmetric power. The first f terms correspond to those
coming from V−⃗t−⃗s,⃗s and for each such term, a basis is given by homogeneous degree
sj monomials in variables w and z. Here, w corresponds to the first standard basis
2
element of F , while z corresponds to the second standard basis element. The last
f terms correspond to those coming from Vt⃗′ ,s⃗′ and for each such term, a basis
is given by homogeneous degree s′j monomials in variables x and y. As before, x
2
corresponds to the first standard basis element of F , while y corresponds to the
second standard basis element.
f −1 kj sj −kj
Denote by W the F subspace of V−⃗t−⃗s,⃗s ⊗ Vt⃗′ ,s⃗′ spanned by {(⊗j=0 w z )⊗
′ ′ ′
y s0 ⊗ y s1 ⊗ · · · ⊗ y sf −1 }(kj )j . Evidently, W is a quotient as a ⟨U, D⟩ ⊂ K represen-
tation. We now define a further partial order on the indexing set of the basis of W .
Let (kj )j and (kj′ )j be two indices, where (kj )j corresponds to the basis element
−1 kj sj −kj ′ ′ ′
(⊗fj=0 w z ) ⊗ y s0 ⊗ y s1 ⊗ · · · ⊗ y sf −1 while (kj′ )j corresponds to the basis el-
−1 kj′ sj −kj′ ′ ′ ′
ement (⊗fj=0 w z ) ⊗ y s0 ⊗ y s1 ⊗ · · · ⊗ y sf −1 . If kj′ ≥ kj for all j, then we say
that (kj )j is a descendant of (kj′ )j . More precisely, if (kj′ − kj ) = n ≥ 0, we say
P
j
that (kj )j is a n-descendant of (kj′ )j . Alternatively, we say (kj′ )j is an n-ascendant
of (kj )j , or (kj )j is a −n-ascendant of (kj′ )j , or (kj′ )j is a −n-descendant of (kj )j .

Now, take κli A+ϵi B +κui C to be the cocycle defined in Lemma 2.13. Denote by ψ
the restriction of this cocycle to ⟨U, D⟩∩K. Then ψ = ϵi B +κui C. Suppose it has an
extension to ⟨U, D⟩. On composing the extension with the quotient map V−⃗t−⃗s,⃗s ⊗
Vt⃗′ ,s⃗′ → W , we obtain a cocycle valued in W , which we denote by q. Denote by
f −1 kj sj −kj
q(kj )j the coordinates of q corresponding to the basis vector (⊗j=0 w z )⊗
′ ′ ′
y s0 ⊗ y s1 ⊗ · · · ⊗ y sf −1 . [Cautionary note about the notation: here the exponent of
w is kj , whereas in Lemma 2.13, sj − kj is the exponent of w].
From the definition of ψ, q(kj )j |U ∩K ̸= 0 if and only if kj = sj for all j. Further,
q(kj )j |D∩K ̸= 0 if and only if ki = si − 1 = s′i + 1 and kj = sj = s′j for all j ̸= i.
−1 kj sj −kj ′ ′ ′
Each (⊗fj=0 w z ) ⊗ y s0 ⊗ y s1 ⊗ · · · ⊗ y sf −1 ∈ W is an eigenvector for D(t)
with eigenvalue tλ(kj )j , where λ(kj )j is the unique number in [0, pf − 1) that is
equivalent to j̸=i pf −1−j (kj − s′j ) + pf −1−i (ki − s′i − 1) mod pf − 1. We make
P
some observations about λ(kj )j :
(1) λ(kj )j = 0 if and only if ki = s′i + 1 and kj = s′j for all j ̸= i if and only if
q(kj )j |D∩K ̸= 0.
16 KALYANI KANSAL

(2) λ(kj )j are evidently pairwise distinct.

(3) Suppose (kj )j ̸= (sj )j . Then λ(kj )j ̸= pf −1−l for any l ∈ [0, f − 1].
To see this, suppose on the contrary that mod pf − 1,
X
pf −1−j (kj − s′j ) + pf −1−i (ki − s′i − 1) = pf −1−l .
j̸=i

Equivalently,
X X
(2.14.1) pf −1−j kj ≡ pf −1−j s′j + pf −1−i (s′i + 1) + pf −1−l .
j j̸=i

We have three subcases:

pf −1−j sj . As
P
• If l = i, then the right hand side of (2.14.1) is
j∈T −{i}
each sj is less than or equal to p − 1, and at least one sj is strictly
less than p − 1 (by assumption), kj is forced to equal sj for each j, a
contradiction.
• If l ̸= i and s′l < p − 1, s′l + 1 ≤ p − 1. Further, s′i + 1 < p − 1,
because si = s′i + 2 ≤ p − 1. Therefore, both the right and left hand
sides have all coefficients of pf −1−j less than or equal to p − 1, and at
least one coefficient strictly less than p − 1, forcing right and left hand
side coefficients to be the same. But this is a contradiction, since the
coefficient of pf −1−l clearly differs as for each j, kj ≤ sj .
• If l ̸= i and s′l = p − 1, s′l + 1 = p. By carrying over to obtain the
coefficient of each pf −1−j in the [0, p − 1] range, we see that the right
hand side (2.14.1) is equivalent to a number with the coefficient of
pf −1−m equal to sm + 1 for some m ∈ [l − 1, i + 1]. Thus both the right
and left hand sides can be made to have all coefficients of pf −1−j less
than or equal to p − 1, and at least one coefficient strictly less than
p − 1. Thus the coefficients on the two sides after carry must be the
same. However, km is necessarily ≤ sm , giving a contradiction.
For each (kj )j , since D acts diagonally on W , q(kj )j |D is a cocycle D → F(λ(kj )j ),
where F(λ(kj )j ) is a one-dimensional F-vector space with action of D(t) given by
multiplication with tλ(kj )j . Note that D ∼ = OK ∗ ∼ ∗
= k × (1 + πOK ). Therefore, when
q(kj )j |D∩K = 0, q(kj )j |D can be seen as a cocycle k ∗ → F(λ(kj )j ). For non-zero
P λ(k )
λ(kj )j , ξ j j = 0, because if ξ˜ is the generator of the cyclic group k ∗ ,
ξ∈k∗
X X X
ξ˜λ(kj )j ξ λ(kj )j = ˜ λ(kj )j =
(ξξ) ξ λ(kj )j .
ξ∈k∗ ξ∈k∗ ξ∈k∗

It follows that H 1 (k ∗ , F(λ(kj )j )) = F/(ξ ˜λ(kj )j

− 1)F = 0. Therefore, when
q(kj )j |D∩K = 0, there exists a(kj )j ∈ F such that q(kj )j (D(ξ)) = ξ λ(kj )j a(kj )j −a(kj )j .
When q(kj )j |D∩K ̸= 0, let a(kj )j = 0. Adjust the cocycle q by the coboundary
f −1 kj sj −kj ′
given by the vector whose coordinate corresponding to (⊗j=0 w z ) ⊗ y s0 ⊗
′ ′
y s1 ⊗ · · · ⊗ y sf −1 is −a(kj )j . Therefore, we may assume that when q(kj )j |D∩K = 0,
q(kj )j |D = 0. When q(kj )j |D∩K ̸= 0, since λ(kj )j = 0, q(kj )j |D is a group homomor-
phism k ∗ × (1 + πOK ) ∼ = D → F. Since order of k ∗ is prime to p, q(kj )j (D(k ∗ )) = 0.
Therefore, regardless of (kj )j , we have q(kj )j (D(k ∗ )) = 0.
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 17

Our next order of business is to understand each q(kj )j |U . Note that U ∼ = OK .

Except when kj = sj for all j, q(kj )j |U (πOK ) = 0 (as remarked earlier) and therefore,
q(kj )j |U can be seen as a map on OK /π. Since D(ξ)U (α) = U (ξα)D(ξ) for ξ ∈ k ∗
and α ∈ OK /π, we have the following for all (kj )j ̸= (sj )j :

(2.14.2)
ξ λ(kj )j q(kj )j (U (α)) = q(kj )j (D(ξ)U (α)) = q(kj )j (U (ξα)D(ξ)) = q(kj )j (U (ξα))

Therefore, replacing α with 1 and ξ with α (this covers all the cases since
q(kj )j (U (0)) is already 0 because q(kj )j |U ∩K = 0), we obtain for all (kj )j ̸= (sj )j :

(2.14.3) q(kj )j (U (α)) = αλ(kj )j q(kj )j (U (1))

Now, we do an inductive argument to show that q(kj )j |U = 0 for all (kj )j ̸= (sj )j .
PSuppose q(kj )j |U = 0 for each m-ascendant (kj )j of (0)j , where −1 ≤ m <
( sj ) − 1. The base case with m = −1 is automatic, because (0)j has no descen-
j
dants. We will show that q(kj )j |U = 0 for each m + 1-ascendant (kj )j of (0)j .
Fix an m + 1-ascendant (kj )j of (0)j . Take (kj′ )j to be a 1-ascendant of (kj )j
(therefore, an m + 2-ascendant of (0)j ).
Since q is a cocycle, we have for each 0 ̸= α ∈ OK /π:

q(kj′ )j (U (α)) + q(kj′ )j (U (1))+

  
P f −1−j ′
X Y  sj − lj  p (kj −lj )
 α j q(lj )j (U (1))
s − k ′
j j j
(lj )j ∈1-descendants
of (kj′ )j

= q(kj′ )j (U (α + 1))

= q(kj′ )j (U (1 + α))

  
X Y  sj − lj 
= q(kj′ )j (U (1)) + q(kj′ )j (U (α)) + 
′
 q(lj )j (U (α))
j
s j − k j
(lj )j ∈1-descendants
of (kj′ )j
18 KALYANI KANSAL

Therefore,
  
P f −1−j ′
X Y  sj − lj  p (kj −lj )
 α j q(lj )j (U (1))
s − k ′
j j j
(lj )j ∈1-descendants
of (kj′ )j
  
X Y  sj − lj 
=   q(lj )j (U (α))
j
sj − kj′
(lj )j ∈1-descendants
of (kj′ )j
  
X Y  sj − lj 
=   αλ(lj )j q(lj )j (U (1)) (by (2.14.3))
s − k ′
j j j
(lj )j ∈1-descendants
of (kj′ )j

Therefore, each α ∈ k ∗ satisfies the following polynomial in x:

 
X Y  sj − lj 
  q(lj )j (U (1))xλ(lj )j
s − k ′
j j j
(lj )j ∈1-descendants
of (kj′ )j
 
P f −1−j ′
X Y  sj − lj  p (kj −lj )
− 
′
 q(lj )j (U (1))x j
j
sj − kj
(lj )j ∈1-descendants
of (kj′ )j

If non-zero, this polynomial is of degree less than pf − 1, with at least |k ∗ |

(kj − lj ) = pf −1−m(lj ) for some
P f −1−j ′
distinct roots, a contradiction. Note that p
j
m(lj ) ∈ [0, f − 1].
P f −1−j ′
Since λ(kj )j does not equal any of the p (kj − lj ) terms, the coefficient of
j
Q 
sj −kj

xλ(kj )j is j sj −k′ q(kj )j (U (1)) and it must equal 0. This implies that q(kj )j |U =
j
0 by (2.14.3).
Finally, we come to the last leg of the proof. Because q(kj )j |U = 0 for each (kj )j ̸=
(sj )j , q(sj )j (U (α + β)) = q(sj )j (U (α)) + q(sj )j (U (β)). Therefore q(sj )j (U (p)) =
pq(sj )j (U (1)) = 0. However, q(sj )j |U ∩K = κui |U ∩K (from the definition of C in
Lemma 2.13). As p is the uniformizer of OK , κui (U (p)) ̸= 0, giving a contradiction.
□

3. Stack dimensions and extensions of GK characters

Let σ and τ be a pair of non-Steinberg, non-isomorphic Serre weights. We record
a fact from [DDR] and [Ste] that we will use in this section. Suppose χ1 and χ2 are
distinct GK characters such that the subspace of Ext1F[GK ] (χ2 , χ1 ) corresponding to
representations with Serre weights both σ and τ has dimension d. Suppose χ′1 and
χ′2 are unramified twists of χ1 and χ2 respectively. If χ′1 ̸= χ′2 , then the subspace
of Ext1F[GK ] (χ′2 , χ′1 ) that corresponds to representations with Serre weights σ and τ
also has dimension d. If on the other hand, χ′1 = χ′2 , then a (d + 1)-dimensional
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 19

1
subspace of ExtF[G K]
(χ′2 , χ′1 ) corresponds to representations with Serre weights σ
and τ .
We now make a few definitions, before stating our main propositions relating
dimensions of closed substacks of X to vector space dimensions of extensions of GK
characters.
Definition 3.1. Let χ1 and χ2 be a pair of F-valued GK characters. We say that
a set Fχ1 ,χ2 of GK -representations with F-coefficients is a family of representations
if each representation in Fχ1 ,χ2 is an extension of an unramified twist of χ2 by an
unramified twist of χ1 .
Definition 3.2. Consider Gm × Gm as parametrizing the unramified twists of χ1
and χ2 via the value of the unramified characters on FrobK . We say that the
family Fχ1 ,χ2 is of dimension ≤ d (resp. of dimension d) if there exists a dense
open subset W of Gm × Gm such that the following condition is satisfied: if χ′1 and
χ′2 are unramified twists of χ1 and χ2 (respectively) corresponding to an F-point
of W , then the extensions in Ext1F[GK ] (χ′2 , χ′1 ) giving elements of Fχ1 ,χ2 form a
subspace of dimension ≤ d (resp. of dimension d).
Definition 3.3. We say that two families Fχ1 ,χ2 and Fχ′1 ,χ′2 are separated if χ′1
and χ′2 are not both unramified twists of χ1 and χ2 respectively.
We now recall some constructions from [EG2, Sec. 5] before stating our main
propositions. We note first that there exist finitely many F-valued characters of
IK that admit extensions to GK . Each such character is in fact valued in F and
is described uniquely by a = (ai )i∈T with each ai ∈ [0, p − 1] and at least some
ai < p − 1. Let A be theQset of such f tuples. Then for a ∈ A, the corresponding
IK character is given by i∈T ωiai |IK . Fix an extension of such a character to GK ,
and denote it by ψa . When each ai = 0, take ψa = 1. When each ai = e and p > 2,
take ψa to be the mod p cyclotomic character, ϵ.
Let M be the rank 1 (φ, Γ)-module over Gm := Spec F[x, x−1 ] generated by
some v ∈ M such that φ(v) = xv and Γ action is trivial. By applying the functor
D defined in [EG1, Sec. 3.6], we obtain a set of (φ, Γ)-modules {D(ψa )}a∈A defined
over F. Let Ma denote the (φ, Γ)-module D(ψa ) ⊠ M defined over Gm . For a
subscheme Spec R ⊂ Gm , we will denote by Ma |Spec R the (φ, Γ)-module obtained
by changing scalars to R. Let Xa = Gm when ψa is not trivial or cyclotomic, and
let Xa = Gm ∖ {1} when ψa is trivial or cyclotomic.
The cohomology groups of C • (Ma |Xa ), the Herr complex associated to Ma |Xa by
[EG1, Sec. 5], vanish in degrees 0 and 2 (the latter by Tate local duality). Therefore,
as the cohomology group in degree 1 gives a coherent sheaf on Xa of constant rank
[K : Qp ] by the local Euler characteristic formula, it is a locally free sheaf. Denoting
the total space of this sheaf by Va , we obtain a space parameterizing extensions of
D(1) by Ma |Xa , the former viewed as a (φ, Γ)-module over Xa by extension of
scalars from F to the global sections on Xa . Thus, for each b ∈ A, twisting further
by Mb , there exists a map
fa,b : Va × Gm → X
corresponding to the universal extension of D(1) ⊠ Mb by Ma |Xa ⊠ Mb . Note that
Gm × Gm acts on extensions of D(1) ⊠ Mb by Ma |Xa ⊠ Mb , for e.g. as described in
[EG2, Sec. 7.3]. Since D(1) ̸= Ma |Xa , the induced map
f a,b : (Va × Gm )/(Gm × Gm ) → X
20 KALYANI KANSAL

is a monomorphism. There exists a map

πa,b : Va × Gm → Ba,b := Xa × Gm
induced by the structure map Va → Xa and the identity map Gm → Gm . The map
πa,b corresponds to choices of unramified twists of D(ψa ) and D(ψb ) respectively.
Now, we consider the Herr complex associated to D(1), denoted C • (D(1)). Each
of the cohomology groups is a finite dimensional vector space over F. The con-
siderations in [EG2, Sec. 5.4] show that for any F-algebra R, H 1 (C • (D(1)R )) =
H 1 (C • (D(1)) ⊗ R, where D(1)R is the (φ, Γ)-module obtained from D(1) by ex-
tending the scalars to R. Therefore, the total space of the invertible sheaf on SpecF
corresponding to H 1 (C • (D(1)) parameterizes extensions of D(1) by itself. Thus,
denoting this total space by V1 , we can define a map
f1,b : V1 × Gm → X
giving the universal extension of D(1) ⊠ Mb by D(1) ⊠ Mb . In this case, in addition
to Gm × Gm , there is an action of the upper unipotent group U on extensions of
D(1) ⊠ Mb by D(1) ⊠ Mb , thus giving a map
f 1,b : (V1 × Gm )/(Gm × Gm × U ) → X .
Denote by π1,b the projection of V1 × Gm onto the second factor.
Finally, when p > 2, we consider the Herr complex associated to D(ϵ), denoted
C • (D(ϵ)). As before, viewing the finite dimensional degree 1 cohomology group as
an invertible sheaf on a point, the total space gives a vector bundle Vϵ defined over
Spec F that parameterizes extensions of D(1) by D(ϵ). Thus, we have a map
fϵ,b : Vϵ × Gm → X
giving the universal extension of D(1) ⊠ Mb by D(ϵ) ⊠ Mb . The induced map
f ϵ,b : (Vϵ × Gm )/(Gm × Gm ) → X
is a monomorphism. Denote by πϵ,b the projection of Vϵ × Gm onto the second
factor.
By construction, each finite type point of X corresponding to a reducible repre-
sentation is in the image of one of the (finitely many) fa,b , f1,b and fϵ,b maps.
Now, consider the set of isomorphism classes of irreducible 2-dimensional rep-
resentations defined over F. For each such representation ρ, there exists a map
Spec F → X , which in turn can be used to write a map fρ : Gm → X corresponding
to D(ρ) ⊠ M . The finite type points in the image correspond to all the unramified
twists of ρ. Since the automorphisms of irreducible representations are precisely
the invertible scalars, fρ factors via the monomorphism
f ρ : [Gm /Gm ] → X .
Since there are only finitely many isomorphism classes of irreducible 2-dimensional
representations upto unramified twists, the finite type points of X corresponding
to irreducible representations lie in the image of one of fρ for finitely many ρ.
Fix non-Steinberg Serre weights σ and τ . Denoting by E the intersection of Xσ
with Xτ , for each a ∈ A, let Ya,b := E ×X2 Va,b . We also define Y1,b := E ×X2 V1,b
and Yϵ,b := E ×X2 Vϵ,b . The maps πa,b |Ya,b , π1,b |Y1,b and πϵ,b |Yϵ,b will henceforth be
written simply as πa,b , π1,b and πϵ,b .
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 21

Proposition 3.4. Let d ≥ 0. Suppose all families of representations contained

in E(F) are of dimension ≤ d, and moreover, E(F) contains at least one family of
dimension d. Then the following are true:
(i) E has dimension d.
(ii) If d > 0, the number of d-dimensional components in E equals the number
of d-dimensional pairwise separated families contained in E.
(iii) Let d = 0, and let
C := {ρ : GK → GL2 (F) | ρ is semisimple}/ ∼
where ρ ∼ ρ′ if ρ and ρ′ are isomorphic as IK representations. Then the
number of d-dimensional components in E equals |C|.
The proof of this proposition will use the following lemmas.
Lemma 3.5. Let χ1 and χ2 be fixed GK characters. Suppose E(F) contains a
family Fχ1 ,χ2 of representations of dimension d. Suppose moreover that there is no
other family of extensions of χ2 by χ1 contained in E(F) with dimension > d. Let
a, b ∈ A be such that ψb is an unramified twist of χ2 , while ψa ⊗ ψb is an unramified
twist of χ1 . Then the following are true:
(i) The dimension of the scheme-theoretic image of Ya,b is ≤ d.
(ii) The number of d-dimensional components in the scheme-theoretic image of
Ya,b is at most 1.
Proof. (i) Let q be a closed point of Ba,b , and after fixing an embedding κ(q) ,→
F, let q be the corresponding F-point of Ba,b . By the construction of Ba,b ,
representations coming from Ya,b (F) are never an extension of a character
by itself. Therefore, the hypotheses in the statement of the Lemma force
−1
πa,b (q)(F) to be a vector space of dimension ≤ d. We have:
−1
Since the F-points of πa,b (q) form a vector space, the reduced induced
−1
closed subscheme of πa,b (q) must be cut out by homogeneous linear equa-
tions in Va,b × Gm × κ(q) and thus be irreducible of dimension equal to the
−1
F-vector space dimension of πa,b (q)(F).
Let S be an irreducible component of Ya,b . Denote by fa,b (S) the scheme-
theoretic image of S. By [Sta, Tag 0DS4], there exists a dense set U ⊂ S
such that for any p ∈ U (F), the dimension of fa,b (S) is given by:
dim fa,b (S) = dim S − dimp (Sfa,b (p) ) = dimp S − dimp (Sfa,b (p) )
where,
−1
dimp S ≤ dim πa,b (πa,b (p)) + dim (πa,b (S))
Restrict U further if necessary so that it is disjoint from other irreducible
−1
components of Yj . Then for p a closed point in U , since πa,b (πa,b (p)) is
irreducible, it is contained entirely in some irreducible component of Ya,b .
−1
By the conditions on U , πa,b (πa,b (p)) ⊂ S. Therefore, dimp (Sfa,b (p) ) =
dimp (Ya,b )fa,b (p) . Since fa,b |Ya,b factors through the quotient Ya,b /(Gm ×
Gm ), we obtain:
dimp (Sfa,b (p) ) = dimp (Ya,b )fa,b (p) = 2
Therefore the dimension of scheme-theoretic image of S is ≤ d − (2 −
dim (πa,b (S))) ≤ d.
22 KALYANI KANSAL

(ii) For i ∈ {1, 2}, suppose S i is an irreducible component of Ya,b with a scheme-
theoretic image of dimension d. Let U i be the dense open subset of S i
obtained by taking the complement of all other irreducible components of
Ya,b . Therefore, πa,b (U i ) = πa,b (S i ) = Ba,b . Since πa,b (U i ) is constructible,
it contains a dense open W i of Ba,b . Let W = W 1 ∩ W 2 . If q is a closed
−1
point of W , πa,b (q) is irreducible and contained entirely in at least one
−1
irreducible component of Ya,b . But since for each i, πa,b (q) ∩ U i is non-
empty and disjoint from all irreducible components of Ya,b apart from S i ,
S 1 must be the same as S 2 . This shows that at most one irreducible
component of Ya,b can have a d-dimensional scheme-theoretic image.
□

Lemma 3.6. For each b ∈ A, the scheme theoretic images of fϵ,b and of f1,b are
strictly less than d.
Proof. The proof follows the same ideas as the proof of Lemma 3.5. The reduction
in dimension for the scheme-theoretic image of f1,b arises from the fact that π1,b :
V1 × Gm → Gm has dimension 1 less than the target of πa,b along with the fact
that f1,b factors through (V1 × Gm )/(Gm × Gm × U ). When p > 2, the reduction in
dimension for the scheme-theoretic image of fϵ,b arises from the fact that the target
of the map πϵ,b : Vϵ × Gm → Gm has dimension 1 less than the target of πa,b . □

Lemma 3.7. Let χ1 and χ2 be fixed GK characters. Suppose E(F) contains a

family Fχ1 ,χ2 of representations of dimension d. Suppose moreover that there is no
other family of extensions of χ2 by χ1 contained in E(F) with dimension > d. Let
a, b ∈ A be such that ψb is an unramified twist of χ2 , while ψa ⊗ ψb is an unramified
twist of χ1 . Then the scheme-theoretic image of Ya,b has dimension d.
Proof. By the construction of Ba,b , representations coming from Ya,b (F) are never
extensions of a character by itself. Therefore, for each unramified twist of χ1 and
χ2 coming from twisting ψa ⊗ ψb and ψb by unramified characters corresponding to
F-points of Ba,b , the space of extensions giving representations contained in E(F)
is precisely d-dimensional.
As πa,b is of finite type over an integral scheme, there exists a dense open W of
Ba,b such that over W , πa,b is flat.
Let q be a closed point of W . Fix an embedding of κ(q) into F to view q as
a F-point q. By hypothesis, (Ya,b )q (F) has the structure of a F-vector space of
dimension d. Therefore, (Ya,b )q (and hence (Ya,b )q ) is irreducible of dimension d.
By flatness over W , dim Ya,b |W = dim (Ya,b )q +2 = d+2. Therefore, there exists
an irreducible component S of Ya,b |W with dimension d + 2. Denote by fa,b (S) the
scheme-theoretic image of S. As in the proof of Lemma 3.5, there exists a dense
open subset U of S, such that for all p ∈ U ,
dim fa,b (S) = dim S − dimp (Sf (p) )
and
dimp (Sfa,b (p) ) = 2

Therefore, the dimension of fa,b (S), and of Ya,b , is precisely d (it cannot exceed
d by Lemma 3.5). □
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 23

Proof of Proposition 3.4. Recall that each reducible representation is in the literal
image of either fa,b or f1,b or f1,b for some a, b ∈ A. Moreover, irreducible rep-
resentations contribute to finitely many zero-dimensional substacks of E by the
description of the maps f ρ for ρ irreducible.
Therefore, the first statement follows from Lemmas 3.5 and 3.6, which also show
that each d-dimensional family contains precisely one d-dimensional component in
its closure.
Now assume that Y is a top dimensional component in E contained in the closure
of two separated d-dimensional families Fχ1 ,χ2 and Fχ′1 ,χ′2 . Then there exist unique
a, a′ , b, b′ ∈ A so that (ψa , ψb ) and (ψa′ , ψb′ ) are unramified twists of (χ1 , χ2 ) and
(χ′1 , χ′2 ) respectively. Therefore, Y is in the scheme-theoretic image of both Ya,b
and Ya′ ,b′ (this uses Lemma 3.6 which shows that the scheme-theoretic image of
Y1,b and Yϵ,b is necessarily of dimension less than d).
Let W (resp. W ′ ) be a dense open subset of Ba,b (resp. Ba′ ,b′ ) for each q ∈
Ba,b (F) (resp. q ∈ Ba′ ,b′ (F)), each F point of (Ya,b )q (resp. (Ya′ ,b′ )q ) corresponds
to a representation contained in Fχ1 ,χ2 (resp. Fχ′1 ,χ′2 ).
By the arguments in Lemma 3.7, Y is contained in the scheme-theoretic image of
an irreducible component S (resp. S ′ ) of Ya,b |W (resp. Ya,b ′
|W ′ ). Since the images
′
of |S| and of |S | in |Y| are constructible sets dense in Y, there exists a dense open
U of |Y| contained in both |S| and |S ′ |. If (a, b) ̸= (a′ , b′ ), then this means that
(a, b) = (b′ , a′ ) and U contains only split extensions. Therefore, families of split
extensions are dense in Y. If d > 0, this is an impossibility because the split locus
correspond to a dimension 2 closed substack of Ya,b whose scheme-theoretic image
has dimension 0 by the arguments in Lemmas 3.5 and 3.7. This settles the second
statement, and along with the fact that the image of each f ρ is 0-dimensional,
settles the third statement as well. □

4. Computations of Serre weights

4.1. Linear algebraic formulation for Serre weights of GK -representations.
In the subsequent text, we will write our f -tuples with decreasing indices. We recall
some relevant  resultsfrom [Ste] and [DDR] below. Let ρ be a GK representation
χ1 ∗
of the form ∈ Ext1GK (F(χ2 ), F(χ1 )).
0 χ2
V⃗t,⃗s is a Serre weight of ρ∨ if and only if V⃗t,⃗s is a Serre weight of ρ in the sense
of [Ste] and [DDR]. Thus V⃗t,⃗s is a Serre weight of ρ∨ iff the following conditions
are met:
(1) There exists a subset J of T , and for each i ∈ T there exists xi ∈ [0, e − 1]
such that:

Y Y Y
(4.1.1) χ 1 | IK = ωiti ωisi +1+xi ωixi
i∈T i∈J i∈J c

and
Y Y Y
(4.1.2) χ 2 | IK = ωiti ωie−1−xi ωisi +e−xi
i∈T i∈J i∈J c
24 KALYANI KANSAL

(2) ρ ∈ LV⃗t,⃗s (F(χ1 ), F(χ2 )) ⊂ Ext1GK (F(χ2 ), F(χ1 )), where LV⃗t,⃗s (F(χ1 ), F(χ2 ))
(or just LV⃗t,⃗s if χ1 and χ2 are understood) is a particular distinguished
subspace.
Assuming (4.1.1) and (4.1.2), we now note the recipe for obtaining LV⃗t,⃗s as given
in [Ste], with slight differences Q in notation.
We first write χ2 |IK = i∈T ωiti i∈T ωimi for the unique mi ∈ [0, p − 1] with
Q
not all mi equal to p − 1. Let S be Q the set ofQf -tuples of non-negative integers
(af −1 , af −2 , ..., a0 ) satisfying χ2 |IK = i∈T ωiti i∈T ωiai and ai ∈ [0, e − 1] ∪ [si +
1, si + e] for all i. Evidently, S is non-empty.
For i ̸= f − 1, let vi be the f -tuple (0, ..., 0, p, −1, 0, ..., 0) with −1 in i position,
p in i + 1 position and 0 everywhere else. Let vf −1 be (−1, 0, ..., 0, p). Then there
exists a subset A ⊂ T such that
X
(4.1.3) (mf −1 , ..., m0 ) + vi ∈ S
i∈A

Definition 4.2. Define Amin to be the minimal A satisfying (4.1.3), in the sense
that it is contained in any other subset of T satisfying (4.1.3).
Definition 4.3. Given (mf −1 , ..., m0 ) and Amin as above.
X
(4.3.1) (yf −1 , ..., y0 ) := (mf −1 , ..., m0 ) + vi ∈ S
i∈Amin
(4.3.2) zi := si + e − yi for all i
The indices of yi ’s and zi ’s will be interpreted to be elements of Z/f Z.
Q zi Q ti Q yi Q ti −1
Remark 4.4. χ1 = i∈T ωi i∈T ωi , χ2 = i∈T ωi i∈T ωi and χ2 χ1 =
Q zi −yi
i∈T ωi .
Definition 4.5. If yi ≥ si + 1, let Ii := [0, zi − 1], and if yi < si + 1, let Ii :=
{yi } ∪ [si + 1, zi − 1]. Here the interval [0, zi − 1] is interpreted as the empty set if
zi − 1 < 0. We follow similar convention for [si + 1, zi − 1] when zi − 1 < si + 1.
Remark 4.6. When e = 1, Ii = {0} if yi = 0 and Ii = ∅ if yi = si + 1.
Remark 4.7. If yi ≥ si + 1, then |Ii | ≤ e − 1 with equality if and only if yi = si + 1.
If yi < si + 1, then since zi ≤ si + e, |Ii | ≤ e with equality if and only if zi = si + e
or equivalently, yi = 0.
Suppose χ−1 ai
Q
2 χ1 = i∈T ωi for ai ∈ [1, p] and not all ai = p. We will extend the
indices of the ai to all of Z by setting aj = aj ′ if j ≡ j ′ mod f . We call the tuple
(af −1 , ..., a0 ) the tame signature of χ−1 ∼
2 χ1 . Gal(k/Fp ) = ⟨Frob⟩ = Z/f Z acts on
such tuples (af −1 , ..., a0 ) via

(4.7.1) Frob · (af −1 , ..., a0 ) = (a0 , af −1 , ..., a1 )

Let f ′ be the cardinality of the orbit of (af −1 , ..., a0 ) under the action of Gal(k/Fp ),
and let f ′′ := f /f ′ .
Definition 4.8. Let ni ∈ [0, pf − 1) be such that χ−1 ni
2 χ1 |IK = ωi |IK .

Note that ni = nj iff i ≡ j mod f ′ .

 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 25

Definition 4.9. For i ∈ T , let

f
X −1
(4.9.1) λi := (zi+j+1 − yi+j+1 )pf −1−j
j=0

(4.9.2) ξi := (pf − 1)zi + λi

Definition 4.10. Let JVAH ⃗
t,⃗
s
(χ1 , χ2 ) denote the subset of all α = (m, κ) ∈ Z ×
{0, ..., f ′′ − 1} satisfying:
(i) ∃i ∈ T and u ∈ Ii , such that if ν is the p-adic valuation of ξi − u(pf − 1),
then
ξi − u(pf − 1)
(4.10.1) m=
pν
(ii) Let im ∈ {0, ..., f ′ −1} be such that m ≡ nim mod pf − 1. It exists because
by the above, pν m ≡ ni mod pf − 1, and so, m ≡ ni−ν . We require that κ
satisfies
(4.10.2) im + κf ′ ≡ i − ν mod f
By [Ste, Prop. 3.13], for each i ∈ T and u ∈ Ii , there exists a unique α satisfying
the conditions above. By [Ste, Thm. 3.16], each α in JVAH ⃗
t,⃗s
(χ1 , χ2 ) gives a unique
basis element of LV⃗t,⃗s (χ1 , χ2 ), denoted as cα . LV⃗t,⃗s (χ1 , χ2 ) is the span of cα ’s
together with additional, distinguished basis elements cun if χ−1 2 χ1 is trivial and
−1 Q −ti
ctr if χ2 χ1 is cyclotomic, i∈T ω ⊗ χ2 is unramified and si = p − 1 for all i. A
consequence of these results is that
X
(4.10.3) dimF LV⃗t,⃗s (χ1 , χ2 ) = |Ii | + δ
i∈T

where δ depends on the situation and could be 0 or 1 for p > 2, and 0, 1 or 2 for
p = 2. It is always 0 if χ−1
2 χ1 is neither trivial nor cyclotomic.
Consider two Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ . Suppose there exist subsets J and J ′
of T , and for each i ∈ T , there exist xi , x′i ∈ [0, e − 1] such that:

t′ s′ +1+x′i x′
Y Y Y Y Y Y
(4.10.4) χ 1 | IK = ωiti ωisi +1+xi ωixi = ωi i ωi i ωi i
i∈T i∈J i∈J c i∈T i∈J ′ i∈J ′c

and

(4.10.5)
t′ e−1−x′i s′ +e−x′i
Y Y Y Y Y Y
χ2 |IK = ωiti ωie−1−xi ωisi +e−xi = ωi i ωi ωi i
i∈T i∈J i∈J c i∈T i∈J ′ i∈J ′c

Then a basis for the intersection of LV⃗t,⃗s (χ1 , χ2 ) with LVt⃗′ ,s⃗′ (χ1 , χ2 ) is given by
cα for α ∈ JVAH
⃗
t,⃗
s
(χ1 , χ2 ) ∩ JVAH
⃗′ ⃗′
(χ1 , χ2 ), together with cun and/or ctr if χ−1
2 χ1 is
t ,s
trivial and/or cyclotomic with some additional conditions.
When e = 1, there is another algorithm to specify a basis of LV⃗t,⃗s (χ1 , χ2 ), given
in [DDR]. We recall some essentials of this algorithm because it will be conve-
nient/shorter to use it for some of the calculations in the unramified case.
26 KALYANI KANSAL

Definition 4.11. Let Jmax := {i ∈ Z/f Z|yi = 0}, where yi are as defined in
Definition 4.3.
Definition 4.12. Let (af −1 , ..., a0 ) be the tame signature of χ−1 2 χ1 . The func-
tion δ : Z → Z is defined in the following way: For j ∈ Z, δ(j) = j unless
(ai−1 , ai−2 , ..., aj ) = (p, p − 1, ..., p − 1) for some j < i, in which case δ(j) = i.
When j = i − 1, the condition (ai−1 , ai−2 , ..., aj ) = (p, p − 1, ..., p − 1) is interpreted
as aj = p.
δ induces a function Z/f Z → Z/f Z, also denoted by δ.
Let J be a subset of Z/f Z. If δ(J) ⊂ J, µ(J) := J. Else choose some [i1 ] ∈
δ(J) ∖ J and let j1 be the largest integer such that j1 < i1 , [j1 ] ∈ J and δ(j1 ) = i1 .
If J = {[j1 ], ..., [jr ]} with j1 > j2 > ... > jr > j1 − f , define iκ for κ ∈ [2, r]
inductively as follows:
(
δ(jκ ), if iκ−1 > δ(jκ )
iκ =
jκ , otherwise
Then µ(J) := {[i1 ], ..., [ir ]}.
When e = 1, LV⃗t,⃗s (χ1 , χ2 ) has a basis given by certain elements of Ext1GK (F(χ2 ), F(χ1 ))
indexed by τ ∈ µ(Jmax ) along with cun and/or ctr if χ−1 2 χ1 is trivial and/or cyclo-
tomic with some additional conditions.
We now state the criterion for determining the Serre weights associated to an
irreducible GK representation, when K = Qp . It can be stated for arbitrary K,
but this paper will only need the case K = Qp .
Let ρ be an irreducible GQp representation. Let η1 and η2 be the two level 2
fundamental characters of IQp . Vt,s ∨
is a Serre weight of ρ∨ iff ρ = η1t+s η2t ⊕ η1t η2t+s .
4.13. Translation from Linear Algebra to Geometry of Stacks. Let V⃗t,⃗s and
Vt⃗′ ,s⃗′ be non-isomorphic, non-Steinberg Serre weights. The closure of irreducible
F-representations is 0-dimensional. Thus, unless K = Qp , we only need to consider
closures of families of reducible representations in order to detect codimension 1
intersections between irreducible components. On the other hand, if K = Qp , then
by Proposition 3.4, we additionally need to consider when XVt,s ∩ XVt′ ,s′ contains
irreducible finite type points. This last point is dealt with easily.
Lemma 4.14. When K = Qp , XVt,s ∩ XVt′ ,s′ contains irreducible finite type points
if and only if s′ = p − 1 − s and t′ ≡ t + s mod p − 1.
Proof. By the algorithm for computing Serre weights, we need to determine when
′ ′ ′ ′ ′ ′
η1t+s η2t ⊕ η1t η2t+s = η1t +s η2t ⊕ η1t η2t +s .
Since Vt,s and Vt′ ,s′ are non-isomorphic and non-Steinberg, the relationship between
(t, s) and (t′ , s′ ) follows immediately. □
Remark 4.15. The criterion in the statement of Lemma 4.14 is the same as that in
Proposition 2.1(ii)(b).
Using Proposition 3.4, we can state a sufficient (and necessary when K ̸= Qp )
condition for XV⃗t,⃗s ∩ XVt⃗′ ,s⃗′ to be codimension 1: there exist GK characters χ1
and χ2 so that after replacing χ1 and χ2 by generic unramified twists, the subspace
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 27

{ρ | V⃗t,⃗s , Vt⃗′ ,s⃗′ ∈ W (ρ)} ⊂ Ext1GK (F(χ−1 −1

1 ), F(χ2 )) has dimension ef −1. By generic
unramified twists we mean that if we let Gm × Gm parametrize the unramified
twists of χ1 and χ2 via the value of the unramified characters on FrobK , then the
statement is true for the points of a dense open subset of Gm × Gm . Equivalently,
LV⃗t,⃗s (χ1 , χ2 ) ∩ LVt⃗′ ,s⃗′ (χ1 , χ2 ) ⊂ Ext1GK (F(χ2 ), F(χ1 )) is spanned by ef − 1 basis
elements excluding cun and ctr .
Therefore, we must find GK characters χ1 and χ2 such that there exist subsets
J and J ′ of T , and for each i ∈ T , there exist xi , x′i ∈ [0, e − 1] such that (4.10.4)
and (4.10.5) are satisfied. We next require that |JVAH ⃗
(χ1 , χ2 ) ∩ JVAH (χ1 , χ2 )| =
t,⃗
s t⃗′ ,s
⃗′

ef − 1. This can happen in one of two ways.

Definition 4.16. We say that a pair of Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ have a type
I intersection witnessed by (χ1 , χ2 ) if |JVAH
⃗
(χ1 , χ2 )| = ef while |JVAH (χ1 , χ2 )| =
t,⃗
s t⃗′ ,s
⃗′

ef −1. The ordering of the pair of Serre weights is not important for this definition.
Definition 4.17. We say that a pair of Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ have a type
II intersection witnessed by (χ1 , χ2 ) if JVAH
⃗
(χ1 , χ2 ) = JVAH (χ1 , χ2 ) of cardinality
t,⃗
s t⃗′ ,s
⃗′

ef − 1.
We will say that the number of separated families in a type I (resp. type II)
intersection for the Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ is n if there exist exactly n pairs
of GK characters that witness the type I (resp. type II) intersection, such that
each pair is distinct from all others upon restriction to IK . By Proposition 3.4,
if K ̸= Qp , the number of irreducible components of XV⃗t,⃗s ∩ XVt⃗′ ,s⃗′ of dimension
[K : Qp ] − 1 equals the number of separated families in either a type I or a type II
intersection for the Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ .
For the remainder of this article, we may assume that ⃗s, s⃗′ do not have all com-
ponents equal to p − 1 since we are excluding Steinberg components from analysis.
Finally, since we are interested in intersections of different irreducible components,
we may assume that V⃗t,⃗s ̸= Vt⃗′ ,s⃗′ .

Lemma 4.18. |JVAH (χ1 , χ2 )| = ef if and only if χ1 = i∈T ωisi +e i∈T ωiti and
Q Q
⃗
t,⃗
s

χ2 |IK = i∈T ωiti .

Q

Proof. By Remarks 4.4 and 4.7, |JVAH (χ1 , χ2 )| = ef implies that χ1 |IK = i∈T ωizi =
Q
⃗
t,⃗
s
si +e
and χ2 |IK = i∈T ωiyi = 1. On the other hand, starting with these
Q Q
i∈T ωi
χ1 and χ2 , we can compute Amin as in Definition 4.2. In this case, we observe
that (mf −1 , ..., m0 ) = (0, ..., 0) as χ2 |IK = 1. J = ∅ satisfies the criterion for Amin ,
and we obtain that yi = 0 and zi = si + e for all i. By Remark 4.7, we get that
|JVAH
⃗
(χ1 , χ2 )| = ef , as desired. □
0,⃗
s

Definition 4.19. Let χ1 , χ2 be two characters and V⃗t,⃗s be a Serre weight satisfying
the conditions in Lemma 4.18. Then, we say that V⃗t,⃗s is the highest weight associ-
ated to the pair (χ1 , χ2 ). It is uniquely determined since not all si can be p − 1.
Moreover, knowing the highest weight determines the pair (χ1 , χ2 ) after restriction
to IK .
Let V⃗t∨,⃗s = V⃗t,⃗s . Then for any ρ∨ ∈ Ext1GK (F(χ2 ), F(χ1 )), we say that XV⃗t,⃗s is the
highest weight component containing ρ.
28 KALYANI KANSAL

Remark 4.20. The number of separated families in a type I intersection can be at

most 2, because one of the two Serre weights has to be the highest weight.

Lemma 4.21. |JVAH (χ1 , χ2 )| = ef −1 if and only if one of the following conditions
t⃗′ ,s
⃗′

is satisfied:
s +e Q tj
(i) There exists an i ∈ T such that χ1 = ωie−1 j̸=i ωj j
Q
j∈T ωj and χ2 =
t
ωisi +1 j∈T ωjj , and moreover, si ≤ p − 2, and if f = 1 then si < p − 2.
Q
s +e Q tj
(ii) e = 1, f > 1 and there exists an i ∈ T such that χ1 = ωie−1 j̸=i ωj j
Q
j∈T ωj ,
t
χ2 = ωisi +1 j∈T ωjj , si = p − 1 and si−1 > 0.
Q
s +e Q tj
(iii) e > 1 and there exists an i ∈ T such that χ1 = ωisi +e−1 j̸=i ωj j
Q
j∈T ωj ,
Q t
χ2 = ωi j∈T ωjj , and si ̸= 0.
In the first two situations above, yi = si + 1 and yj = 0 for all j ̸= i (recall
Definition 4.3). On the other hand, if yi = si + 1 and yj = 0 for all j ̸= i, then one
of the two above must be satisfied.
The third situation is equivalent to yi = 1, yj = 0 for all j ̸= i along with si ̸= 0.

Proof. By Remark 4.7, if |JVAH (χ1 , χ2 )| = ef − 1 then one of the following two
t⃗′ ,s
⃗′

conditions must be satisfied:

(i) There exists i ∈ T such that yi = si + 1 and for j ̸= i, yj = 0. This implies
s +e Q tj si +1 Q tj
that χ1 = ωie−1 j̸=i ωj j
Q
j∈T ωj and χ2 = ωi j∈T ωj . On the
Q −t
other hand, starting with such χ1 and χ2 , twisting them by j∈T ωj j and
applying the recipe to compute Amin (Definition 4.10), yj and zj (Defini-
tion 4.3), we branch into two scenarios:
(a) If si ≤ p − 2 for f > 1 and < p − 2 for f = 1, then Amin = ∅, yi = si + 1
and yj = 0 for j ̸= i, giving |JVAH ⃗
t,⃗
s
(χ1 , χ2 )| = ef − 1.
Q −tj Q m
(b) If si = p − 1, then χ2 ⊗ j∈T ωj = j∈T ωj j , where mi−1 = 1 and
mj = 0 if j ̸= i − 1. Note that this automatically implies that f > 1,
since we are assuming our Serre weights are non-Steinberg. We can
obtain the desired values of yj ’s if and only if (mf −1 , ..., m0 ) is not
already in S of Definition 4.2. In other words, if and only if e = 1 and
si−1 ̸= 0.
(ii) There exists i ∈ T such that yi = 1, si ̸= 0 (this is to enforce distinction
from the condition above) and yj = 0 when j ̸= i. Note that this auto-
s +e Q tj
matically implies that e > 1, and that χ1 = ωisi +e−1 j̸=i ωj j
Q
j∈T ωj
Q t
and χ2 = ωi j∈T ωjj . On the other hand, starting with such χ1 and χ2 ,
Q −t
twisting them by j∈T ωj j and applying the recipe to compute Amin , yj
and zj , we obtain that Amin = ∅, yi = 1 and yj = 0 for i ̸= j, and we get
the correct cardinality of JVAH⃗
t,⃗
s
(χ1 , χ2 ).
□

Remark 4.22. In the cases Lemma 4.21(i) and Lemma 4.21(ii),

Y s +e
χ2−1 χ1 = ωie−2−si ωj j
j̸=i
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 29

In the case Lemma 4.21(iii),

s +e
Y
χ2−1 χ1 = ωie−2+si ωj j
j̸=i

Remark 4.23. When e = 1, the cases Lemma 4.21(i) and Lemma 4.21(ii) are to-
gether equivalent to Jmax = Z/f Z ∖ [i].
Before launching into computations of Type I and II intersections, we introduce
some more notation. When comparing f -tuples ⃗s and s⃗′ , we will often only state
the values of si and s′i that have specific constraints or are potentially different
from each other. If the values of si or s′i are not specified, then we assume that
si = s′i . If no range is specified for si , we mean that beyond any relations that it
must satisfy with respect to s′i , the value of si can be anything in [0, p − 1]. Further,
if we say (..., si , ...) = (..., ∈ [a, b], ...), we mean that si can take any value ∈ [a, b].
Similar notational assumptions apply with the roles of si and s′i interchanged.
Finally, we say that a tuple (bf −1 , bf −2 , ..., b0 ) is equivalent to (b′f −1 , b′f −2 , ..., b′0 ) if
Pf −1 f −1−j
Pf −1
j=0 bj p ≡ j=0 b′j pf −1−j mod pf − 1.
We will retain the symbols yi , zi , Ii , λi and ξi as defined in Definitions 4.3, 4.5
and 4.9 for V⃗t,⃗s , and will replace them respectively with yi′ , zi′ , Ii′ , λ′i and ξi′ for
Vt⃗′ ,s⃗′ and with yi′′ , zi′′ , Ii′′ , λ′′i and ξi′′ for Vt⃗′′ ,s⃗′′ .

5. Type I intersections
In this section, we will compute criteria for existence of a pair of characters
(χ1 , χ2 ) witnessing a type I intersection for Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ , with
|JVAH
⃗
(χ1 , χ2 )| = ef and |JVAH (χ1 , χ2 )| = ef − 1. |JVAH (χ1 , χ2 )| = ef − 1 can
t,⃗
s t⃗′ ,s
⃗′ t⃗′ ,s
⃗′

happen via one of three ways as enumerated in Lemma 4.21. In all three situations,
we may assume without loss of generality that i in the statements of Lemma 4.21(i),
Lemma 4.21(ii) and Lemma 4.21(iii) is f − 1. We will also count the number of
families contributing to a type I intersection when V⃗t,⃗s and Vt⃗′ ,s⃗′ are both weakly
regular. (In the general case, the information can still be gleaned directly from
the computations that follow, but we omit the explicit description for the sake of
clarity).
5.1. Type I intersections when f = 1. We will omit subscripts of components
of f -tuples in this section as f = 1.

5.1.1. Case 1. : |JVAH

⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(i).
t ,s
Suppose p = 2. The non-Steinberg condition requires that s = s′ = 0. Plugging
in s and s′ in the expressions for χ2 (using Lemmas 4.18 and 4.21), we get t ≡ t′ + 1
mod p − 1. This gives t = t′ and shows that V⃗t,⃗s and Vt⃗′ ,s⃗′ are isomorphic, a
contradiction. Therefore, we may assume p > 2.
Comparing the two ways of writing χ−1 2 χ1 , we obtain:

s + e ≡ e − 2 − s′ mod p − 1 ⇐⇒
′
s ≡ −2 − s ≡ p − 3 − s mod p − 1
This gives one of the following two situations:
(1) s ≤ p − 3 =⇒ s′ = p − s − 3.
30 KALYANI KANSAL

(2) s = p − 2 =⇒ s′ = p − 2.
In both situations, comparing the two ways of writing χ2 , we obtain that t′ +
s + 1 ≡ t mod p − 1. In other words t′ ≡ t + p − s′ − 2 ≡ t + s + 1 mod p − 1.
′

The second situation therefore implies that V⃗t,⃗s = Vt⃗′ ,s⃗′ , which is a contradiction.
The first situation is equivalent to the conditions in Proposition 2.1(ii)(a) imply-
1
ing that ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0. Since it is symmetric in s and s′ , whenever
Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0, there exist two separated families witnessing a type I
intersection for V⃗t,⃗s and Vt⃗′ ,s⃗′ .

5.1.2. Case 2. : |JVAH

⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(iii). Implicit in this case
t ,s
is e > 1 and p > 2, the latter since s′ is not allowed to be 0.
Comparing the two ways of writing χ−1 2 χ1 , we obtain:

s + e ≡ e − 2 + s′ mod p − 1 ⇐⇒
′
(5.1.1) s ≡s+2 mod p − 1

This gives one of the following two situations:

(1) s < p − 3 =⇒ s′ = s + 2.
(2) s = p − 3 =⇒ s′ = 0.
(3) s = p − 2 =⇒ s′ = 1.
In both situations, comparing the two ways of writing χ2 , we obtain that t′ ≡
−1 + t ≡ p − 2 + t mod p − 1. By comparing with Proposition 2.1, we notice that
the first situation implies HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′ ,s⃗′ )) ̸= 0 (Proposition 2.6),
1
the second implies ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposition 2.1(ii)(a) and the
1
third implies ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposition 2.1(ii)(b). Notice that
the relationship between s and s′ is asymmetric in all three situations, unless p = 3
in which case the second and third situations are symmetric.
The above calculations may be summarized in the following proposition:

Proposition 5.2. Let f = 1. A Type I intersection occurs with non-isomorphic,

non-Steinberg Serre weights V⃗t,⃗s and Vt⃗′ ,s⃗′ if and only if one of the following holds
true:
1
(i) ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposition 2.1(ii)(a). In this case, 2 fam-
ilies witness the type I intersection.
(ii) e > 1 and HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′ ,s⃗′ )) ̸= 0. In this case, 1 family
witnesses the type I intersection.
(iii) e > 1 and s = p − 2, s′ = 1, d′ ≡ −1 + d mod p − 1. In this case, the
number of families witnessing the type I intersection is 1 unless p = 3, in
which case the number is 2.
Note that the non-isomorphic, non-Steinberg condition automatically forces p >
1
2. Further, the last statement implies ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposi-
tion 2.1(ii)(b).

5.3. Type I intersections when f > 1.

 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 31

5.3.1. Case 1. : |JVAH

⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(i) or via Lemma 4.21(ii).
t ,s

Comparing the two ways of writing χ−1 2 χ1 , we obtain the following equivalences
mod pf − 1 upto translating all indices by a fixed element of Z/f Z:
X f
X −2
pf −1−j (sj + e) ≡ e − 2 − s′f −1 + pf −1−j (s′j + e) ⇐⇒
j∈T j=0

X f
X −2
pf −1−j sj ≡ − 2 − s′f −1 + pf −1−j s′j
j∈T j=0
f
X −3
≡ p− s′f −1 −2+ p(s′f −2 − 1) + pf −1−j s′j
j=0

Therefore, for a fixed ⃗s, s⃗′ is forced to be unique since each s′i ∈ [0, p − 1], and
the non-Steinberg condition requires that not all s′i can be p − 1. Similarly, for a
fixed s⃗′ , ⃗s is forced to be unique.
We have a number of possible cases :
(i) Suppose s′f −1 ≤ p − 2, s′f −2 = s′f −3 = ... = s′f −i = 0 and s′f −1−i ≥ 1, where
i ≥ 1.
(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′f −i , s′f −1−i , s′f −2−i , ..., s′0 ) ≡
(p − s′f −1 − 2, −1, 0, ..., 0, s′f −1−i , s′f −2−i , ..., s′0 ) ≡
(p − s′f −1 − 2, p − 1, p − 1, ..., p − 1, s′f −1−i − 1, s′f −2−i , ..., s′0 )
Therefore, sf −1 = p − s′f −1 − 2, sf −2 = sf −3 = ... = sf −i = p − 1,
sf −1−i = s′f −1−i − 1 ≤ p − 2 and sj = s′j for all the remaining j’s.

(ii) Suppose s′f −1 ≤ p − 3, s′f −2 = s′f −3 = ... = s′0 = 0.

(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′0 ) ≡ (p − s′f −1 − 2, −1, 0, ..., 0)
≡ (p − s′f −1 − 3, p − 1, p − 1, ..., p − 1)
We get sf −1 = p − s′f −1 − 3 ≤ p − 3, sf −2 = sf −3 = ... = s0 = p − 1.

(iii) Suppose s′f −1 = p − 2, s′f −2 = s′f −3 = ... = s′0 = 0.

(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′0 ) ≡ (0, −1, 0, ..., 0)
≡ (p − 1, p − 2, p − 1, ..., p − 1)
Hence, sf −2 = p − 2 and all the other sj ’s equal p − 1.

The remaining cases require |JVAH (χ1 , χ2 )| = ef − 1 via Lemma 4.21(ii), and
t⃗′ ,s
⃗′

implicitly, e = 1.
(iv) Suppose s′f −1 = p − 1, s′f −2 > 1.
(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′0 ) ≡ (−1, s′f −2 − 1, s′f −3 , ..., s′0 )
≡ (p − 1, s′f −2 − 2, s′f −3 , ..., s′0 )
Therefore, sf −1 = p − 1, sf −2 = s′f −2 − 2 and sj = s′j for the remaining j’s.
32 KALYANI KANSAL

(v) Suppose f > 2, s′f −1 = p − 1, s′f −2 = 1, s′f −3 = s′f −4 = ... = s′f −i = 0 and
s′f −1−i ≥ 1 for some i > 2.

(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′f −i , s′f −1−i , s′f −2−i , ..., s′0 ) ≡
(−1, 0, 0, ..., 0, s′f −1−i , s′f −2−i , ..., s′0 ) ≡
(p − 1, p − 1, p − 1, ..., p − 1, s′f −1−i − 1, s′f −2−i , ..., s′0 )
Therefore, sf −1 = sf −2 = ... = sf −i = p − 1, sf −1−i = s′f −1−i − 1 and
sj = s′j for all the other j’s.

(vi) Suppose f = 2, s′f −1 = p − 1, s′f −2 = 1.

(p − s′f −1 − 2, s′f −2 − 1) ≡ (−1, 0) ≡ (p − 2, p − 1)
Therefore, sf −1 = p − 2 and sf −2 = p − 1.

(vii) Suppose f > 2, s′f −1 = p − 1, s′f −2 = 1 and s′f −3 = ... = s′0 = 0.

(p − s′f −1 − 2, s′f −2 − 1, s′f −3 , ..., s′0 ) ≡ (−1, 0, 0, ..., 0)
≡ (p − 2, p − 1, p − 1, ..., p − 1)
Therefore, sf −1 = p − 2 and sj = p − 1 for all the other j’s.
The results of the computations are summarized in the proposition below.
Proposition 5.4. Let f > 1. Consider pairs (⃗s, s⃗′ ) satisfying:
Qf −1 s −s′ −2 Qf −2 s′j ′
• j=0 ωj j = ωf −1f −1 j=0 ωj , where sj , sj ∈ [0, p − 1];
′
• Not all sj , as well as not all sj , are p − 1.
• yf′ −1 = s′f −1 + 1 and yj′ = 0 for j ̸= f − 1 (yj′ are as defined in Defini-
tion 4.3).
Below is an enumeration of all such pairs.
(i) (sf −1 , sf −2 , ..., sf −i , sf −1−i ) = (∈ [0, p−2], p−1, ..., p−1, ∈ [0, p−2]), where
i ∈ [1, f − 1];
(s′f −1 , s′f −2 , ..., s′f −i , s′f −1−i ) = (p − sf −1 − 2, 0, ..., 0, sf −1−i + 1).

(ii) (sf −1 , sf −2 , ..., s0 ) = (∈ [0, p − 3], p − 1, ..., p − 1);

(s′f −1 , s′f −2 , ..., s′0 ) = (p − 3 − sf −1 , 0, ..., 0).
This only makes sense if p ≥ 3.

(iii) (sf −1 , sf −2 , sf −3 , ..., s0 ) = (p − 1, p − 2, p − 1, ..., p − 1);

(s′f −1 , s′f −2 , s′f −3 , ..., s′0 ) = (p − 2, 0, 0, ..., 0).
When e = 1, we additionally have:
(iv) (sf −1 , sf −2 ) = (p − 1, ∈ [0, p − 3]);
(s′f −1 , s′f −2 ) = (p − 1, sf −2 + 2).
This only makes sense if p ≥ 3.

(v) f > 2,
(sf −1 , sf −2 , sf −3 , ..., sf −i , sf −1−i ) = (p − 1, p − 1, p − 1, ..., p − 1, ∈ [0, p − 2]),
where i ∈ [2, f − 1];
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 33

(s′f −1 , s′f −2 , s′f −3 ..., s′f −i , s′f −1−i ) = (p − 1, 1, 0, ..., 0, sf −1−i + 1).

(vi) f = 2,
(sf −1 , sf −2 ) = (p − 2, p − 1);
(s′f −1 , s′f −2 ) = (p − 1, 1).

(vii) f > 2,
(sf −1 , sf −2 , sf −3 , ..., s0 ) = (p − 2, p − 1, p − 1, ..., p − 1);
(s′f −1 , s′f −2 , s′f −3 , ..., s′0 ) = (p − 1, 1, 0, ..., 0).

Comparing the two ways of writing χ2 , we obtain:

X X
pf −1−j t′j + (s′f −1 + 1) ≡ pf −1−j tj mod pf − 1 ⇐⇒
j∈T j∈T
X X
(5.4.1) pf −1−j t′j ≡ −1 − s′f −1 + pf −1−j tj mod pf − 1
j∈T j∈T

5.4.1. Case 2. : |JVAH (χ1 , χ2 )| = ef − 1 via Lemma 4.21(iii). Implicit in this case
t⃗′ ,s
⃗′

is that e > 1.
Comparing the two ways of writing χ−1 2 χ1 , we obtain the following equivalences
mod pf − 1:
X X
pf −1−j (sj + e) ≡ e − 2 + s′f −1 + pf −1−j (s′j + e) ⇐⇒
j∈T j̸=f −1
X X
f −1−j
p (sj + 1) ≡ s′f −1 −1+ pf −1−j (s′j + 1) ⇐⇒
j∈T j̸=f −1
X X
f −1−j
(5.4.2) p sj ≡ s′f −1 −2+ pf −1−j s′j
j∈T j̸=f −1

Proposition 5.5. Let f > 1 and e > 1.

Consider pairs (⃗s, s⃗′ ) satisfying:
Qf −1 s s′ −1 −2 Qf −2 s′j ′
• j=0 ωj j = ωff−1 j=0 ωj , where sj , sj ∈ [0, p − 1];
• Not all sj , as well as not all sj , are p − 1. Also, s′f −1 ̸= 0
′

• yf′ −1 = 1 and yj′ = 0 for j ̸= f − 1 (yj′ are as defined in Definition 4.3).

• After reindexing if necessary, ⃗s and s⃗′ satisfy one of the below:
(i) sf −1 ≤ p − 3;
s′f −1 = sf −1 + 2.
This only makes sense if p ≥ 3.

(ii) (sf −1 , sf −2 , ..., sf −i , sf −1−i ) = (p − 1, p − 1, ..., p − 1, ∈ [0, p − 2]), where

i ≥ 1;
(s′f −1 , s′f −2 , ..., s′f −i , s′f −1−i ) = (1, 0, ..., 0, sf −i−1 + 1).

(iii) (sf −1 , sf −2 , ..., s0 ) = (p − 2, p − 1, ..., p − 1);

(s′f −1 , s′f −2 , ..., s′0 ) = (1, 0, ..., 0).

Comparing the two ways of writing χ2 , we obtain:

34 KALYANI KANSAL

X X
(5.5.1) pf −1−j t′j ≡ −1 + pf −1−j tj mod pf − 1
j∈T j∈T

It is evident that when each si and each s′i is < p − 1, the relationship between
⃗s and ⃗s′ described in Propositions 5.4 and 5.5 is asymmetric.
The calculations for type 1 intersections are summarized below.
Proposition 5.6. Let f > 1. V⃗t,⃗s and Vt⃗′ ,s⃗′ be a pair of non-isomorphic, non-
Steinberg Serre weights. Then there exist GK characters χ1 and χ2 such that
|JVAH
⃗
(χ1 , χ2 )| = ef , |JVAH (χ1 , χ2 )| = ef − 1 if and only if one of the following
t,⃗
s t⃗′ ,s
⃗′

is satisfied:
(i) Upto translating the indices by any fixed number, ⃗s and s⃗′ satisfy one of the
conditions in Proposition 5.4 while ⃗t and ⃗t′ satisfy (5.4.1) .
(ii) Upto translating the indices by any fixed number, ⃗s and s⃗′ satisfy one of the
conditions in Proposition 5.5 while ⃗t and ⃗t′ satisfy (5.5.1).
Corollary 5.7. Suppose f > 1 and V⃗t,⃗s and Vt⃗′ ,s⃗′ are two non-isomorphic weakly
regular Serre weights. Then there exists a type I intersection for the pair if and
only if one of the following holds (upto translating the indices by any fixed number
and/or interchanging V⃗t,⃗s and Vt⃗′ ,s⃗′ if necessary):
(i) ⃗s and s⃗′ satisfy Proposition 5.4(i) with i = 1; while ⃗t and ⃗t′ satisfy (5.4.1).
(ii) ⃗s and s⃗′ satisfy Proposition 5.5(i); while ⃗t and ⃗t′ satisfy (5.5.1).
In other words, if and only if one of the following is true:
(i) Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0, or
(ii) e > 1 and HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′ ,s⃗′ )) ̸= 0
Equivalently, if and only if Ext1F[GL2 (OK )] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.
Moreover, exactly 1 family witnesses the type I intersection. When f = 2, s1 = p−1 2 ,
s0 = p−3 , s′
1 = p−3
and s ′
0 = p−1
, interchanging ⃗
s and ⃗
s′ also satisfies Proposi-
2 2 2
tion 5.4(i) after shifting the indices by 1. However, in this case the computations
of ⃗t and ⃗t′ show that the situation is not symmetric, and we still have just 1 family
witnessing the intersection.
Proof. By Propositions 2.1, 2.6 and 2.14 and corollary 2.10. □

6. Type II intersections
In this section, we will compute criteria for existence of a pair of characters
(χ1 , χ2 ) witnessing a type II intersection for (non-isomorphic and non-Steinberg)
Serre weights Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ . Thus, we will determine if χ1 and χ2 exist such that
JVAH (χ1 , χ2 ) = JVAH (χ1 , χ2 ) of cardinality ef − 1. We will denote the highest
t⃗′ ,s
⃗′ t⃗
′′ ,s⃗′′

weight associated to the pair by V⃗t,⃗s . A family witnessing a type II intersection

necessarily also witnesses two type I intersections, one for the Serre weights V⃗t,⃗s
and Vt⃗′ ,s⃗′ , and the other for V⃗t,⃗s and Vt⃗′′ ,s⃗′′ , thus it gives a triple intersection of
codimension 1. On the other hand, every such triple intersection must involve a
type II intersection.
6.1. Type II intersections when f = 1. We will omit subscripts of components
of f -tuples in this section as f = 1.
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 35

6.1.1. Case 1. : |JVAH

t′ ,s′
(χ1 , χ2 )| = |JVAH
t′′ ,s′′
(χ1 , χ2 )| = ef −1, both via Lemma 4.21(i)
or both via Lemma 4.21(iii). It is immediate that this forces Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ to be
isomorphic, a contradiction.

6.1.2. Case 2. : |JVAH

t′ ,s′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(i) (Lemma 4.21(ii) is
not possible because non-Steinberg); |JVAH t′′ ,s′′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(iii).
Lemma 4.21(iii) assumes that e > 1.
Comparing ways of writing χ−1 2 χ1 using Remark 4.22, we have

e − 2 − s′ ≡ e − 2 + s′′ ≡ s + e mod p − 1
(6.1.1) ⇐⇒ s′′ ≡ p − 1 − s′ ,
s′ ≡ p − 3 − s,
s′′ ≡ s + 2
Comparing ways of writing χ2 using Lemma 4.21, we obtain:

s′ + 1 + t′ ≡ 1 + t′′ ≡ t mod p − 1
(6.1.2) ⇐⇒ t′′ ≡ t′ + s′
t′ ≡ t + s + 1
t′′ ≡ −1 + t
By stipulation in Lemma 4.21(i), s′ ̸= p − 2. Therefore, s′ ≤ p − 3, and since
(t , s′ ) ̸= (t′′ , s′′ ), s < p − 3, the equivalences in (6.1.1) are equalities and p > 3.
′

Notice the nature of type I intersection for V⃗t,⃗s and Vt⃗′ ,s⃗′ . It corresponds to
1
ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposition 2.1(ii)(a). On the other hand, the type I
intersection for V⃗t,⃗s and Vt⃗′′ ,s⃗′′ corresponds to HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′′ ,s⃗′′ )) ̸= 0.
Imposing the above conditions, we now calculate y ′ , y ′′ , z ′ , z ′′ , I ′ , I ′′ , ξ ′ and ξ ′′ ,
and compare JVAH t′ ,s′
(χ1 , χ2 ) with JVAH t′′ ,s′′
(χ1 , χ2 ).
By Lemma 4.21, y ′ = s′ + 1 and z ′ = e − 1, while y ′′ = 1 and z ′′ = s′′ + e − 1.
Therefore
I ′ = [0, e − 2]
I ′′ = {1} ∪ [s′′ + 1, s′′ + e − 2]
ξ ′ = (p − 1)(e − 1) + (e − 2 − s′ )
ξ ′′ = (p − 1)(s′′ + e − 1) + (e − 2 + s′′ )
As u′ varies in I ′ , ξ ′ − u′ (p − 1) = (p − 1)v ′ + (e − 2 − s′ ) with v ′ taking up values
in [1, e − 1]. Similarly, as u′′ varies in I ′′ ,
ξ ′′ − u′′ (p − 1) = (p − 1)v ′′ + (e − 2 + s′′ ) where v ′′ ∈ [1, e − 2] ∪ {s′′ + e − 2}
= (p − 1)v ′′ + e − 2 − s′ + (p − 1) where v ′′ ∈ [1, e − 2] ∪ {p − 3 − s′ + e}
= (p − 1)v ′′ + e − 2 − s′ , where v ′′ ∈ [2, e − 1] ∪ {p − 2 − s′ + e}

By Definition 4.10, JVAH t′ ,s′

(χ1 , χ2 ) = JVAH
t′′ ,s′′
(χ1 , χ2 ) if and only if for all v ′ ∈
[1, e − 1], there exists a v ∈ [2, e − 1] ∪ {p − 2 − s′ + e} such that:
′′
36 KALYANI KANSAL

(p − 1)v ′ + (e − 2 − s′ ) (p − 1)v ′′ + (e − 2 − s′ )
(6.1.3) ′ =
p ν pν ′′
where ν ′ is the p-adic valuation of the numerator on L.H.S, while ν ′′ is that of
the numerator on R.H.S.
The only thing to check then is that (6.1.3) holds for v ′ = 1 and v ′′ = p−2−s′ +e.
Plugging in,

p − 3 − s′ + e
(6.1.4) L.H.S. =
pν

(6.1.5)
(p − 1)(p − 2 − s′ + e) + e − 2 − s′ p(p − 3 − s′ + e)
R.H.S. = = = L.H.S.
pν ′′ pν ′ +1
Therefore, conditions (6.1.1) and (6.1.2) guarantee a type II intersection, and are
equivalent to the conditions in Proposition 2.1(ii)(b). In this case, the relationship
between the pairs (t′ , s′ ) and (t′′ , s′′ ) is symmetric except when s′ = 1 and s′′ = p−2.
Therefore the calculations show the existence of 2 separated families (because the
highest weights are distinct) witnessing the type II intersection except when s′ = 1
and s′′ = p − 2. In the special case s′ = 1 and s′′ = p − 2, there is just 1 family.
Summarizing these findings, we have the proposition below.
Proposition 6.2. Let f = 1. If Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ are a pair of non-isomorphic,
non-Steinberg Serre weights, then a type II intersection occurs for the pair if and
only if e > 1, p > 3 and Ext1GL2 (k) (Vt⃗′ ,s⃗′ , Vt⃗′′ ,s⃗′′ ) ̸= 0 via Proposition 2.1(ii)(b). In
addition, the following statements are true:
• If (χ1 , χ2 ) witness the type II intersection, then one of the two corresponding
type I intersections witnessed by (χ1 , χ2 ) arises via Proposition 5.2(i). The
other arises via Proposition 5.2(ii).
• Each type II intersection is witnessed by 2 separate families except when
s′ = 1 and s′′ = p − 2, in which case just one family witnesses it.
6.3. Type II intersections when f > 1, e = 1. We will use the algorithm in
[DDR] for this section. Our objective is to find the conditions on Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′
′ ′′ ′
so that µ(Jmax ) = µ(Jmax ) of cardinality f − 1, where Jmax is the subset of Z/f Z
′′
satisfying the conditions in Definition 4.11 for Vt⃗′ ,s⃗′ while Jmax is the corresponding
subset for Vt⃗′′ ,s⃗′′ .
We will find these intersections in two steps. First, we will find Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′
′ ′′
such that Jmax = Z/f Z − {[f − 1]}, Jmax = Z/f Z − {[f − 1 − i]} for some i ∈
−s′f −1 −1 Q s′j +1 −s′′ −1 Q s′′
j +1
[0, f −1], and ωf −1 j∈T ∖{f −1} ωj = ωi f −1−i j∈T ∖{f −1−i} ωj . The
′
assumption that Jmax = Z/f Z − {[f − 1]} does not cause any loss of generality. In
′ ′′
the second step, we will compute µ(Jmax ) and µ(Jmax ), and identify the situations
in which they are the same.
For the first step, we will use the results of Proposition 5.4. Specifically, if
′
a Vt⃗′ ,s⃗′ exists with Jmax = Z/f Z − {[f − 1]}, then there exists a non-Steinberg
⃗
V⃗t,⃗s so that the pair (⃗s, s′ ) satisfies one of the conditions enumerated in Propo-
sition 5.4. This is simply a consequence of Lemma 4.21. Similarly, we can find
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 37

a non-Steinberg V ⃗˜ ⃗ so that the pair (⃗s̃, s⃗′′ ) satisfies one of the conditions enu-
d,s̃
merated in Proposition 5.4 after adding i to each index. Since we are impos-
s −s′ −2 Q s′j −s′′
f −1−i −2 s′′
j +2
ing j∈T ωj j = ωf −1f −1
Q Q
j∈T ∖{f −1} ωj = ωi j∈T ∖{f −1−i} ωj =
s̃j
ω , we have ⃗s = ⃗s̃. Therefore, in the first step we are looking for vectors ⃗s
Q
j∈T j
that show up in more than one items of the list in Proposition 5.4 (after translating
the indices by adding some fixed integer if necessary). If such a ⃗s exits, we will
say that the two items in the list can be cycled with each other, and that one item
is a cycling of the other. The corresponding two s⃗′ ’s (in the notation of the list
in Proposition 5.4) give us our candidate (s⃗′ , s⃗′′ ) and the reindexing informs us
what i should be. In this situation, since χ−1 si +1
Q
2 χ 1 must equal ω
j∈T i , the tame
signature (af −1 , ..., a0 ) = (sf −1 + 1, ..., s0 + 1).
For instance, consider the f -tuple ⃗x with (xf −1 , xf −2 , ..., xf −m , xf −1−m ) = (∈
[0, p−2], p−1, ..., p−1, ∈ [0, p−2]) for some m ∈ [1, f −4], (xf −1−i , xf −2−i , ..., xf −i−k ) =
(p − 1, p − 1, ..., p − 1) for some i ∈ [m + 1, f − 3], k ∈ [2, f − 1 − i] and xf −1−i−k ∈
[0, p − 2]. Clearly, ⃗x satisfies the conditions required of ⃗s showing up in Proposi-
tion 5.4(i). If we reindex ⃗x, adding i to the indices mod f , then we see that ⃗x can
show up as the ⃗s in Proposition 5.4(v). That is, Proposition 5.4(i) can be cycled with
Proposition 5.4(v). The corresponding two s⃗′ (in the notation of Proposition 5.4)
that show up in Proposition 5.4(i) and Proposition 5.4(v) are our candidates for s⃗′
and s⃗′′ respectively (in the notation of this proposition). The reindexing tells us
′ ′′
that Jmax ought to be Z/f Z ∖ {[f − 1]} and Jmax ought to be Z/f Z ∖ {[f − 1 − i]}.
For the second step, we note that δ(f − 2 − i) = f − 1 − i. Similarly, δ(f −
3 − i) = f − 2 − i and so on until δ(f − i − k) = f − i − k + 1. Therefore
′′ ′′
f − i − k ̸∈ µ(Jmax ), which implies that µ(Jmax ) = Z/f Z ∖ {[f − i − k]}. Similarly,
δ(f − 2) = f − 1 and if m > 1, we observe that δ causes an increase in index right
′
until f − m, so that δ(f − m) = f + 1 − m. This forces f − 1 − m ∈ µ(Jmax ) and
′ ′
eventually, f − i − k ∈ µ(Jmax ). If m = 1, f − 2 − m is in µ(Jmax ) again forcing
′ ′ ′′
f − i − k ∈ µ(Jmax ). Hence µ(Jmax ) ̸= µ(Jmax ).
′
We repeat this process by finding all possible cyclings and computing µ(Jmax )
′′
and µ(Jmax ). Instead of showing details for all computations, we will give an
outline. Each ⃗s showing up in the list items of Proposition 5.4 has constraints for
the components positioned in some specific way relative to the indices f − 1 and
f − 1 − m for some m (In the notation of Proposition 5.4, the symbol i is used
instead of m. Here we are using i differently, to indicate the translation of indices).
If this ⃗s shows up in another list item after reindexing by adding i mod f , the
constraints for the reindexed second list item will have a description relative to
indices f − 1 and f − 1 − n for some n. After undoing the reindexing, we may
expect to see constraints on ⃗s components positioned in a specific away around the
indices given by f − 1 and f − 1 − m (as posed by the specifications of the first list
item), and f − 1 − i and f − 1 − i − n (as posed by the specifications of the second
list item). We will use this notation in the outline below.

(1) Proposition 5.4(i) can be cycled with Proposition 5.4(iv) in the following
possible ways:
′′
(a) i ∈ [m + 1, f − 2]. Then µ(Jmax ) excludes f − 1 − i. If m > 1,
′ ′
µ(Jmax ) excludes f − m. If m = 1, µ(Jmax ) excludes f − 1. Therefore,
′′ ′
µ(Jmax ) ̸= µ(Jmax ).
38 KALYANI KANSAL

′′
(b) i = m − 1 ≥ 1. Again, µ(Jmax ) excludes f − 1 − i = f − m. The same
′ ′′ ′
holds true for µ(Jmax ) and we have µ(Jmax ) = µ(Jmax ).
(2) Proposition 5.4(i) can be cycled with Proposition 5.4(v) in the following
possible ways:
(a) i ∈ [m + 1, f − 3] and f − 1 − i − n ̸≡ f − 1 mod f . The calculations
′ ′′
in the example above show that µ(Jmax ) ̸= µ(Jmax ).
′′
(b) i ∈ [m + 1, f − 3] and f − 1 − i − n ≡ f − 1 mod f . Here, µ(Jmax )
′
excludes 0, whereas µ(Jmax ) includes it, making them unequal.
′ ′′
(c) i ∈ [1, m − 2] and f − 1 − i − n = f − 1 − m. Here µ(Jmax ) and µ(Jmax )
are both of cardinality f − 1 and exclude f − m. Therefore, they are
equal.
(3) Proposition 5.4(ii) can be cycled with Proposition 5.4(iv) with i = f − 1.
′′ ′
µ(Jmax ) excludes f − 1 − i = 0. The same is true for µ(Jmax ), which is
′′
thus equal to µ(Jmax ).
(4) Proposition 5.4(ii) can be cycled with Proposition 5.4(v) with any i ∈
′ ′′
[1, f − 2]. In this case, both µ(Jmax ) and µ(Jmax ) exclude 0. They are
therefore equal.
(5) Proposition 5.4(iii) can be cycled with Proposition 5.4(v) with i ∈ [0, f −
′ ′′
1] ∖ {1}. Both µ(Jmax ) and µ(Jmax ) exclude f − 1, and are equal.
(6) Proposition 5.4(iii) can be cycled with Proposition 5.4(vi) with i = 1. Both
′ ′′
µ(Jmax ) and µ(Jmax ) exclude f − 1, and are equal.
(7) Proposition 5.4(iii) can be cycled with Proposition 5.4(vii) with i = 1. Both
′ ′′
µ(Jmax ) and µ(Jmax ) exclude f − 1, and are equal.
(8) Proposition 5.4(iv) can be cycled with Proposition 5.4(v):
′ ′′
(a) i > 1, f − 1 − i − n = f − 2. Both µ(Jmax ) and µ(Jmax ) exclude f − 1
and are equal.
′ ′′
(b) i > 1, f − 1 − i − n ̸= f − 2. µ(Jmax ) exlucdes f − 1, while µ(Jmax )
′ ′′
excludes f − i − n. Therefore, µ(Jmax ) ̸= µ(Jmax ).
(9) Proposition 5.4(v) can be cycled with Proposition 5.4(vii) with i = m. Both
′ ′′
µ(Jmax ) and µ(Jmax ) exclude f − i and are equal.

Proposition 6.4. Let f > 1, e = 1. There exists a pair of GK characters (χ1 , χ2 )

of highest weight V⃗t,⃗s witnessing a type II intersection for Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ if and
only if after translating the indices by adding some fixed integer, there exists an
i ∈ T such that the following are true:
Pf −1 Pf −1 Pf −1
(i) s′f −1 +1+ j=0 pf −1−j t′j ≡ pi (s′′f −1−i +1)+ j=0 pf −1−j d′′j ≡ j=0 pf −1−j tj
mod pf − 1.
(ii) The vectors s⃗′ , s⃗′′ and ⃗s satisfy one of the following conditions:
(a) (s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 2]) for
some i ∈ [1, f − 2];

(s′′f −1 , s′′f −2 , ..., s′′f −1−i , s′′f −2−i ) = (p − s′f −1 − 2, p − 1, ..., p − 1, s′f −2−i +
1);

(sf −1 , sf −2 , ..., sf −1−i , sf −2−i ) = (p − s′f −1 − 2, p − 1, ..., p − 1, s′f −2−i −

1).
INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 39

(b) f > 2 and (s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i , s′f −3−i , ..., s′f −m , s′f −1−m ) =
(∈ [0, p − 2], 0, ..., 0, 0, 0, ..., 0, ∈ [1, p − 1]) for some m ∈ [3, f − 1];

(s′′f −1 , s′′f −2 , ..., s′′f −1−i , s′′f −2−i , s′′f −3−i , ..., s′′f −m , s′′f −1−m ) = (p−s′f −1 −
2, p − 1, ..., p − 1, 1, 0, ..., 0, s′f −1−m ) where i ∈ [1, m − 2];

(sf −1 , sf −2 , ..., sf −1−i , sf −2−i , sf −3−i , ..., sf −m , sf −1−m ) = (p−s′f −1 −

2, p − 1, ..., p − 1, p − 1, p − 1, ..., p − 1, s′f −1−m − 1).

(c) i = f − 1 and (s′f −1 , s′f −2 , ..., s′1 , s′0 ) = (∈ [0, p − 3], 0, ..., 0, 0);

(s′′f −1 , s′′f −2 , ..., s′′1 , s′′0 ) = (p − 1 − s′f −1 , p − 1, ..., p − 1, p − 1);

(sf −1 , sf −2 , ..., s1 , s0 ) = (p − 3 − s′f −1 , p − 1, ..., p − 1, p − 1).

(d) f > 2 and (s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i , s′f −3−i , ..., s′0 ) =
(∈ [0, p − 3], 0, ..., 0, 0, 0, ..., 0);

(s′′f −1 , s′′f −2 , ..., s′′f −1−i , s′′f −2−i , s′′f −3−i , ..., s′′0 ) =
(p − 2 − s′f −1 , p − 1, ..., p − 1, 1, 0, ..., 0) where i ∈ [1, f − 2];

(sf −1 , sf −2 , ..., sf −1−i , sf −2−i , sf −3−i , ..., s0 ) =

(p − 3 − s′f −1 , p − 1, ..., p − 1, p − 1, p − 1, ..., p − 1).

(e) f > 2 and (s′f −1 , s′f −2 , ..., s′f −i , s′f −1−i , s′f −2−i , s′f −3−i , ..., s′0 ) =
(p − 2, 0, ..., 0, 0, 0, 0, ..., 0);

(s′′f −1 , s′′f −2 , ..., s′′f −i , s′′f −1−i , s′′f −2−i , s′′f −3−i , ..., s′′0 ) =
(0, p − 1, ..., p − 1, p − 1, 1, 0, ..., 0) where i ∈ [2, f − 1];

(sf −1 , sf −2 , sf −3 , ..., s0 ) =
(p − 1, p − 2, p − 1, ..., p − 1).

(f ) f = 2 and (s′f −1 , s′f −2 ) = (p − 2, 0);

(s′′f −1 , s′′f −2 ) = (1, p − 1) where i = 1;

(sf −1 , sf −2 ) = (p − 1, p − 2).

(g) f > 2 and (s′f −1 , s′f −2 , s′f −3 , s′f −4 , ..., s′0 ) =

(p − 2, 0, 0, 0, ..., 0);

(s′′f −1 , s′′f −2 , s′′f −3 , ..., s′′0 ) = (0, p − 1, 1, 0, ..., 0) where i = 1;

(sf −1 , sf −2 , sf −3 , sf −4 ..., s0 ) =
(p − 1, p − 2, p − 1, p − 1, ..., p − 1).
40 KALYANI KANSAL

(h) f > 2 and (s′f −1 , s′f −2 , s′f −3 , ..., s′f −1−i , s′f −2−i )
= (p − 1, 1, 0, ..., 0, ∈ [1, p − 2]) where i > 1;

(s′′f −1 , s′′f −2 , s′′f −3 , ..., s′′f −1−i , s′′f −2−i ) =

(p − 1, p − 1, p − 1, ..., p − 1, s′f −2−i + 1);

(sf −1 , sf −2 , sf −3 , ..., sf −1−i , sf −2−i ) =

(p − 1, p − 1, p − 1, ..., p − 1, s′f −2−i − 1).

(i) f > 2, i = f − 1 and (s′f −1 , s′f −2 , s′f −3 , ..., s′1 , s′0 ) =

(p − 1, 1, 0, ..., 0, p − 1);

(s′′f −1 , s′′f −2 , s′′f −3 , ..., s′′1 , s′′0 ) = (1, 0, 0, ..., 0, p − 1);

(sf −1 , sf −2 , sf −3 , ..., s1 , s0 ) = (p − 1, p − 1, p − 1, ..., p − 1, p − 2).

Proof. The conditions on s⃗′ , s⃗′′ and ⃗s are a consequence of the preceding discussion
along with explicit descriptions coming from the list in Proposition 5.4. The con-
dition on ⃗t′ , t⃗′′ and ⃗t follow from comparing descriptions of χ2 using Lemmas 4.18
and 4.21. □

Remark 6.5. In each triple of s⃗′ , s⃗′′ and ⃗s featuring in the list in Proposition 6.4,
at least two of the vectors have some component equal to p − 1.
6.6. Type II intersections when f > 1, e > 1. We will compute the scenarios
in which type II intersections occur for the pair Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ . In the case of
Vt⃗′ ,s⃗′ , we will assume without loss of generality that i = f − 1 in the statements of
Lemma 4.21 and that ⃗t′ = 0.
In the following calculations, we will use some extra notation and strategies for
comparing JVAH (χ1 , χ2 ) and JVAH (χ1 , χ2 ) that we now explain. Given the Serre
t⃗′ ,s
⃗′ t⃗
′′ ,s⃗′′

weights Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ , and suitable GK characters χ1 and χ2 , we may compute
yj′ , yj′′ , zj′ , zj′′ , λ′j , λ′′j , Ij′ , Ij′′ , ξj′ and ξj′′ using Definitions 4.3, 4.5 and 4.9.

Definition 6.7. Fix j ∈ T . Vj′ ⊂ Z is defined to satisfy:

{ξj′ − u(pf − 1)|u ∈ Ij′ } = {(pf − 1)v + λ′j |v ∈ Vj′ }

Vj′′ ⊂ Z is defined to satisfy:

{ξj′′ − u(pf − 1)|u ∈ Ij′′ } = {(pf − 1)v + λ′j |v ∈ Vj′′ }

The above definition of Vj′′ makes sense because λ′′j ≡ λ′j mod pf −1 , since ex-
ponentiating ωj with either gives the same character χ−1
2 χ1 .

Definition 6.8. Define P ′ , P ′′ ⊂ T × Z as follows:

P ′ := {(j, v) ∈ T × Z|v ∈ Vj′ }
P ′′ := {(j, v) ∈ T × Z|v ∈ Vj′′ }
We define two functions β and α next.
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 41

Definition 6.9.
β :T ×Z→Z
(j, v) 7→ (pf − 1)v + λ′j
and,
α : T × Z → Z × {0, 1, ..., f ′′ − 1}
(j, v) 7→ (m, κ)
β(j, v)
where m = , and κ satisfies (4.10.2).
pvalp (β(j,v))
Remark 6.10. By Definitions 4.10 and 6.9, JVAH
⃗′ ⃗′
(χ1 , χ2 ) = {α(j, v)|(j, v) ∈ P ′ }
t ,s

and JVAH (χ1 , χ2 ) = {α(j, v)|(j, v) ∈ P ′′ }.

⃗t ,s⃗′′
′′

Remark 6.11. By the comments following Definition 4.10, α|P ′ and α|P ′′ are injec-
tive functions.
Remark 6.12. An examination of Definition 4.10 shows that if v ∈ Vj′ for some
j ∈ T , then finding a pair (j̃, ṽ) ∈ P ′′ such that α(j, v) = α(j̃, ṽ) is equivalent to
finding j̃ and ṽ ∈ Vj̃′′ satisfying the following two conditions:
• (pf − 1)v − λ′j and (pf − 1)ṽ − λ′j differ by a factor of a p-power;
• the difference of p-adic valuations offsets the difference between j and j̃ in
the formula for computing κ in (4.10.2).
Remark 6.13. Let α(j, v) = α(j̃, ṽ). Then j = j̃ ⇐⇒ valp ((pf − 1)v − λ′j ) ≡
valp ((pf − 1)ṽ − λ′j̃ ) mod f .

Remark 6.14. If j ̸= j̃ and valp ((pf − 1)v − λ′j ) = valp ((pf − 1)ṽ − λ′j̃ ) = 0, then
α(j, v) ̸= α(j̃, ṽ).
Definition 6.15. We will say that a pair (j, v) ∈ P ′ matches (j̃, ṽ) ∈ P ′′ if α(j, v) =
α(j̃, ṽ).
Remark 6.16. For the purposes of our calculations, we will classify the ways a pair
(j, v) ∈ P ′ can match a pair (j̃, ṽ) ∈ P ′′ in the following manner:
(i) (j, v) = (j̃, ṽ).
′ ′
(ii) j̃ ≡ j +1 mod f and ṽ = pv +zj+1 −yj+1 ; or j ≡ j̃ +1 and v = pṽ +zj′ −yj′ .
In these cases, |valp ((p − 1)v − λj ) − valp ((pf − 1)ṽ − λ′j̃ )| = 1.
f ′

(iii) Matches not classified by either of the above.

As we will see, the first two types will be easy to spot, whereas the third will
need some verification.
We will use the notation and ideas above repeatedly in the calculations below.
Because of the repetitiveness of the arguments, we will show the calculations in
detail only for a few scenarios, and will only report the findings from the calculations
for the rest.
6.16.1. Case 1. : |JVAH
⃗′ ⃗′
(χ1 , χ2 )| = |JVAH (χ1 , χ2 )| = ef −1, both via Lemma 4.21(i).
t ,s ⃗
′′ t ,s⃗′′

Case 1a. : i = f − 1. Comparing ways of writing χ−1 2 χ1 and χ2 in terms of s , s

⃗′ ⃗′′
⃗′′
and t using Remark 4.22, we obtain that Vt⃗′ ,s⃗′ = Vt⃗′′ ,s⃗′′ , a contradiction.
42 KALYANI KANSAL

Case 1b. : i < f − 1.

Comparing ways of writing χ−1 ⃗′ ⃗′
2 χ1 in terms of s and s , we have:
X
e − 2 − s′f −1 + pf −1−j (s′j + e) ≡
j∈T ∖{f −1}
X
pf −1−i (e − 2 − s′′i ) + pf −1−j (s′′j + e)
j∈T ∖{i}

X X
⇐⇒ − 2 − s′f −1 + pf −1−j s′j ≡ pf −1−i (−2 − s′′i ) + pf −1−j s′′j
j∈T ∖{f −1} j∈T ∖{i}

X
⇐⇒ p − 2 − s′f −1 + p(s′f −2 − 1) + pf −1−j s′j ≡
j∈T ∖{f −1,f −2}
X
pf −1−i (p − 2 − s′′i ) + pf −i (s′′i−1 − 1) pf −1−j s′′j Vt⃗′ ,s⃗′
j∈T ∖{i,i−1}

⇐⇒ (p − 2 − s′f −1 , s′f −2 − 1, s′f −3 , ..., s′0 ) ≡

(s′′f −1 , ..., s′′i+1 , p − 2 − s′′i , s′′i−1 − 1, s′′i−2 ..., s′′0 )

Lemma 6.17. The above condition is satisfied if and only if (upto interchanging
s⃗′ with s⃗′′ ), one of the following pairs describe s⃗′ and s⃗′′ :
(i) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1]) for some k ∈
[i + 1, f − 2],
(s′i , s′i−1 , ..., s′l+1 , s′l ) = (∈ [0, p − 2], p − 1, ..., p − 1, ∈ [0, p − 2]) for some
l ∈ [0, i − 1];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 2 − s′f −1 , p − 1, ..., p − 1, s′k − 1),
(s′′i , s′′i−1 , ..., s′′l+1 , s′′l ) = (p − 2 − s′i , 0, ..., 0, s′l + 1).
(ii) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1]) for some k ∈
[i + 1, f − 2],
(s′i , s′i−1 , ..., s′0 ) = (∈ [0, p − 2], p − 1, ..., p − 1);
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 1 − s′f −1 , p − 1, ..., p − 1, s′k − 1),
(s′′i , s′′i−1 , ..., s′′0 ) = (p − 2 − s′i , 0, ..., 0).
(iii) (s′f −1 , s′f −2 , ..., s′i+1 , s′i , s′i−1 , ..., s′0 ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1], p −
1, ..., p − 1);
(s′′f −1 , s′′f −2 , ..., s′′i+1 , s′′i , s′′i−1 , ..., s′′0 ) = (p − 1 − s′f −1 , p − 1, ..., p − 1, p − s′i −
1, 0, ..., 0).

Proof. Easy verification upon recalling that s′f −1 , s′′i ≤ p − 2 by Lemma 4.21(i). □

Imposing the above conditions, we now calculate yj′ , yj′′ , zj′ , zj′′ , Ij′ , Ij′′ , Vj′ and
Vj′′ ,and compare JVAH
⃗′ ⃗′
(χ1 , χ2 ) with JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ) using Remark 6.10.
t ,s t ,s
For Lemma 6.17(i), we have:

(
s′j + 1 if j = f − 1
(6.17.1) yj′ =
0 if j ∈ T ∖ {f − 1}
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 43

(
p − 1 − s′j if j = i
(6.17.2) yj′′ =
0 if j ∈ T ∖ {i}
(
e−1 if j = f − 1
(6.17.3) zj′ =
e + s′j if j ∈ T ∖ {f − 1}

′ ′ ′

p + e − 2 − sj = p + (zj − yj ) if j = f − 1

′ ′
p + e − 1 = p − 1 + (zj − yj ) if j ∈ [k + 1, f − 2]




e − 1 + s′j = −1 + (zj′ − yj′ ) if j = k



′′
(6.17.4) zj = e − 1 = −p + (zj′ − yj′ ) + yj′′ if j = i
e = −(p − 1) + (zj′ − yj′ )

if j ∈ [l + 1, i − 1]




′ ′ ′




 e + 1 + sj = 1 + (zj − y j ) if j = l
′ ′ ′
e + sj = (zj − yj ) if j ∈ [i + 1, k − 1] ∪ [0, l − 1]


(
[0, e − 2] if j = f − 1
(6.17.5) Ij′ =
{0} ∪ [s′j + 1, s′j + e − 1] if j ∈ T ∖ {f − 1}

′ ′′

{0} ∪ [p − 1 − sj , zj − 1] if j = f − 1

′′
{0} ∪ [p, zj − 1] if j ∈ [k + 1, f − 2]




′ ′′
{0} ∪ [sj , zj − 1] if j = k



(6.17.6) Ij′′ = [0, e − 2] if j = i
{0} ∪ [1, zj′′ − 1]

if j ∈ [l + 1, i − 1]




′′ ′′




 {0} ∪ [sj + 2, zj − 1] if j = l
′′ ′′
{0} ∪ [sj , zj − 1] if j ∈ [i + 1, k − 1] ∪ [0, l − 1]


(
[1, e − 1] if j = f − 1
(6.17.7) Vj′ =
[1, e − 1] ∪ {s′j + e} if j ∈ T ∖ {f − 1}

[1, e − 1] ∪ {p + zj′ − yj′ }



 if j =f −1
[2, e] ∪ {p + zj′ − yj′ } ∈ [k + 1, f − 2]




 if j
′
[2, e] ∪ {sj + e} if j =k



(6.17.8) Vj′′ = [1, e − 1] if j =i

[0, e − 1] if j ∈ [l + 1, i − 1]




[0, e − 2] ∪ {s′j + e}




 if j =l
[1, e − 1] ∪ {s′j + e} ∈ [i + 1, k − 1] ∪ [0, l − 1]

if j
Recall that JVAH
⃗′ ⃗′
(χ1 , χ2 ) = {α(j, v)|(j, v) ∈ P ′ } and JVAH (χ1 , χ2 ) = {α(j, v)|(j, v) ∈
t ,s ⃗
′′ t ,s⃗′′
P }. In order to compare the two, there is no work to be done for (j, v) ∈ P ′ ∩ P ′′ .
′′

So, we must now examine the image of α when restricted to the set P ′ − P ′′ and
compare it to the image of α when restricted to the set P ′′ − P ′ .
To begin, consider {(j, 1)|j ∈ [k, f − 2]} ⊂ P ′ − P ′′ . One can immediately verify
using Remark 6.12 that α(j, 1) = α(j + 1, p + zj′ − yj′ ), where (j, 1) ∈ P ′ − P ′′ and
44 KALYANI KANSAL

(j + 1, p + zj′ − yj′ ) ∈ P ′′ − P ′ . Similarly, for j ∈ [l + 1, i], α(j, zj′ − yj′ ) = α(j − 1, 0).
Here (j, zj′ − yj′ ) ∈ P ′ − P ′′ (since zj′ − yj′ = s′j + e) and (j − 1, 0) ∈ P ′′ − P ′ . These
matches are of the type described in Remark 6.16(ii). After taking into account
all matches of the types described in Remark 6.16(i) and Remark 6.16(ii), the only
possibly unmatched pairs are (l, e − 1) ∈ P ′ − P ′′ and (k, e) ∈ P ′′ − P ′ . Now,
valp (β(l, e − 1)) = 0 as s′l ̸= p − 1. As s′k ̸= 1, valp (β(k, e)) = 0. As l ̸= k,
α(l, e − 1) ̸= α(k, e) by Remark 6.14.
Therefore, JVAH ⃗′ ⃗′
(χ1 , χ2 ) ̸= JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ).
t ,s t ,s
Calculations for Lemma 6.17(ii) are as follows:

(
s′j + 1 if j = f − 1
(6.17.9) yj′ =
0 if j ∈ T ∖ {f − 1}

(
p − 1 − s′j if j = i
(6.17.10) yj′′ =
0 if j ∈ T ∖ {i}

(
e−1 if j = f − 1
(6.17.11) zj′ =
e + s′j if j ∈ T ∖ {f − 1}



 p + e − 1 − s′j = p + (zj′ − yj′ ) + 1 if j =f −1
p + e − 1 = p − 1 + (zj′ − yj′ )

if j ∈ [k + 1, f − 2]




e − 1 + s′ = −1 + (z ′ − y ′ )

if j =k
(6.17.12) zj′′ = ′
j
′ ′
j j


 e + sj = (zj − y j ) if j ∈ [i + 1, k − 1]
′ ′ ′′
e − 1 = −p + (zj − yj ) + yj if j =i




e = −(p − 1) + (zj′ − yj′ ) if j ∈ [0, i − 1]


(
[0, e − 2] if j = f − 1
(6.17.13) Ij′ =
{0} ∪ [s′j + 1, s′j + e − 1] if j ∈ T ∖ {f − 1}



{0} ∪ [p − s′j , zj′′ − 1] if j = f − 1
{0} ∪ [p, zj′′ − 1]

if j ∈ [k + 1, f − 2]




{0} ∪ [s′ , z ′′ − 1]

if j = k
(6.17.14) Ij′′ = j j
′′ ′′
{0} ∪ [sj + 1, zj − 1] if j ∈ [i + 1, k − 1]


[0, e − 2] if j = i





′′
{0} ∪ [1, zj − 1] if j ∈ [0, i − 1]


(
[1, e − 1] if j = f − 1
(6.17.15) Vj′ = ′
[1, e − 1] ∪ {sj + e} if j ∈ T ∖ {f − 1}
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 45



[0, e − 2] ∪ {p + zj′ − yj′ } if j =f −1
[2, e] ∪ {p + zj′ − yj′ }

if j ∈ [k + 1, f − 2]




[2, e] ∪ {s′ + e}

if j =k
(6.17.16) Vj′′ = j
[1, e − 1] ∪ {s′j + e}
 if j ∈ [i + 1, k − 1]

[1, e − 1] if j =i





[0, e − 1] if j ∈ [0, i − 1]


To verify JVAH
⃗′ ⃗′
(χ1 , χ2 ) = JVAH (χ1 , χ2 ), we apply the same strategy as we
t ,s ⃗
′′ t ,s⃗′′
previously did. As before, for each (j, v) ∈ P except (f − 1, e − 1), we can get (j, v)
to match some (j̃, ṽ) with (j̃, ṽ) ∈ P ′′ via Remark 6.16(i) or Remark 6.16(ii). (k, e)
is the only pair in P ′′ not matched to anything in P − {(f − 1, e − 1)} via these two
matching strategies. By Remark 6.14, (f − 1, e − 1) cannot match (k, e) because
valp (β(f − 1, e − 1)) = 0 = valp (β(k, e)), since s′f −1 ̸= p − 1 and s′k ̸= 0. Therefore,
we do not get a type II intersection in the desired manner.
The calculations for Lemma 6.17(iii) are similar and left to the reader. The re-
sults from the calculations are also similar, and show that JVAH (χ1 , χ2 ) ̸= JVAH (χ1 , χ2 ).
t⃗′ ,s
⃗′ t⃗
′′ ,s⃗′′

The findings are summarized below.

Proposition 6.18. Let e > 1, f > 1. Suppose Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ are a pair of non-
isomorphic, non-Steinberg Serre weights. There do not exist any GK characters χ1
and χ2 such that |JVAH
⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(i), |JVAH ⃗
′′ ⃗′′
(χ1 , χ2 )| =
t ,s t ,s

ef − 1 via Lemma 4.21(i) and JVAH

⃗′ ⃗′
(χ1 , χ2 ) = JVAH (χ1 , χ2 ).
t ,s ⃗
′′ t ,s⃗′′

6.18.1. Case 2. : |JVAH

⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(iii); |JVAH (χ1 , χ2 )| =
t ,s ⃗
′′ t ,s⃗′′
ef − 1 via Lemma 4.21(iii).
Case 2a. : i = f − 1. Comparing ways of writing χ−1 2 χ1 and χ2 in terms of s , s
⃗′ ⃗′′
⃗′
and t using Remark 4.22, we obtain that Vt⃗′ ,s⃗′ = Vt⃗′′ ,s⃗′′ , a contradiction.
Case 2b. : i < f − 1.
Comparing ways of writing χ−1 ⃗′ ⃗′′
2 χ1 in terms of s , s and ⃗ s, we have:
X
e − 2 + s′f −1 + pf −1−j (s′j + e)
j∈T ∖{f −1}
X X
f −1−i
≡p (e − 2 + s′′i ) + pf −1−j (s′′j + e) ≡ pf −1−j (sj + e)
j∈T ∖{i} j∈T

(6.18.1)
⇐⇒ (−2 + s′f −1 , s′f −2 , ..., s′0 ) ≡ (s′′f −1 , ..., s′′i+1 , −2 + s′′i , s′′i−1 , ..., s′′0 )
≡ (sf −1 , sf −2 , ..., s0 )
Comparing ways of writing χ2 , we have:
X X
pf −1−i + pf −1−j d′′j ≡ 1 ≡ pf −1−j tj mod pf − 1
j∈T j∈T
X X
(6.18.2) ⇐⇒ pf −1−j d′′j ≡1−p f −1−i
, pf −1−j tj ≡ 1 mod pf − 1
j∈T j∈T
46 KALYANI KANSAL

Lemma 6.19. The condition in (6.18.1) is satisfied for some s⃗′ , s⃗′′ and ⃗s if and
only if one of the following pairs describe s⃗′ and s⃗′′ :
(i) s′f −1 ∈ [2, p − 1], s′i ∈ [0, p − 3];
s′′f −1 = s′f −1 − 2, s′′i = s′i + 2.

(ii) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, 1], 0, ..., 0, ∈ [1, p − 1]) for some k ∈ [i +
1, f − 2], s′i ∈ [0, p − 3];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 2 + s′f −1 , p − 1, ..., p − 1, s′k − 1), s′′i = s′i + 2.

(iii) (s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (1, 0, ..., 0, ∈ [1, p − 2]);
(s′′f −1 , s′′f −2 , ..., s′′i+1 , s′′i ) = (p − 1, p − 1, ..., p − 1, s′i + 1).

(iv) s′f −1 ∈ [2, p − 1], (s′i , s′i−1 ..., s′l+1 , s′l ) = (p − 1, p − 1, ..., p − 1, ∈ [0, p − 2])
for some l ∈ [0, i − 1];
s′′f −1 = s′f −1 − 2, (s′′i , s′′i−1 ..., s′′l+1 , s′′l ) = (1, 0, ..., 0, s′l + 1).

(v) s′f −1 ∈ [1, p − 1], (s′i , s′i−1 ..., s′0 ) = (p − 1, p − 1, ..., p − 1);
s′′f −1 = s′f −1 − 1, (s′′i , s′′i−1 ..., s′′0 ) = (1, 0, ..., 0);
sf −1 ∈ [1, p − 1], (si , si−1 ..., s0 ) = (p − 1, p − 1, ..., p − 1).

(vi) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (1, 0, ..., 0, ∈ [1, p − 1]) for some k ∈ [i + 1, f − 2],
(s′i , s′i−1 , ..., s′l+1 , s′l ) = (p−1, p−1, ..., p−1, ∈ [0, p−2]) for some l ∈ [0, i−1];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 2 + s′f −1 , p − 1, ..., p − 1, s′k − 1),
(s′′i , s′′i−1 , ..., s′′l+1 , s′′l ) = (1, 0, ..., 0, s′l + 1).
Proof. Easy verification upon recalling that s′f −1 , s′′i ≥ 1 byLemma 4.21(iii). □
Imposing the above conditions, we now calculate yj′ , yj′′ , zj′ , zj′′ , Ij′ , Ij′′ , Vj′ and
Vj′′ ,and compare JVAH
⃗′ ⃗′
(χ1 , χ2 ) with JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ) using Remark 6.10.
t ,s t ,s
For Lemma 6.19(i), we have:
(
1 if j = f − 1
(6.19.1) yj′ =
0 if j ∈ T ∖ {f − 1}
(
1 if j = i
(6.19.2) yj′′ =
0 if j ∈ T ∖ {i}
(
e − 1 + s′j if j = f − 1
(6.19.3) zj′ =
e + s′j if j ∈ T ∖ {f − 1}

′ ′ ′
e − 2 + sj = zj − yj
 if j = f − 1
′′
(6.19.4) zj = e + 1 + si = (zj − yj′ ) + yj′′ if j = i
′ ′

e + s′j = zj′ − yj′


if j ∈ T ∖ {f − 1, i}

(
′ {1} ∪ [s′j + 1, zj′ − 1] if j = f − 1
(6.19.5) Ij =
{0} ∪ [s′j + 1, zj′ − 1] if j ∈ T ∖ {f − 1}
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 47


′ ′′
{0} ∪ [sj − 1, zj − 1] if j = f − 1

′′
(6.19.6) Ij = {1} ∪ [s′i + 3, zj′′ − 1] if j = i
{0} ∪ [s′j + 1, zj′ − 1] if j ∈ T ∖ {f − 1, i}



(
[1, e − 2] ∪ {zj′ − yj′ } if j = f − 1
(6.19.7) Vj′ =
[1, e − 1] ∪ {zj′ − yj′ } if j ∈ T ∖ {f − 1}

′ ′
[1, e − 1] ∪ {zj − yj }
 if j = f − 1
(6.19.8) Vj′′ = [1, e − 2] ∪ {zj′ − yj′ } if j = i
[1, e − 1] ∪ {zj′ − yj′ }

if j ∈ T ∖ {f − 1, i}


The only pairs in P and P ′′ that are unmatched after applying matching strategy
Remark 6.16(i) are (i, e−1) ∈ P and (f −1, e−1) ∈ P ′′ . As s′i ̸= p−1 and s′f −1 ̸= 1,
valp (β(i, e−1)) = 0 = valp (β(f −1, e−1)). By Remark 6.14, (i, e−1) cannot possibly
match (f − 1, e − 1), and therefore, JVAH⃗′ ⃗′
(χ1 , χ2 ) ̸= JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ).
t ,s t ,s
For Lemma 6.19(ii), we have:
(
1 if j = f − 1
(6.19.9) yj′ =
0 if j ∈ T ∖ {f − 1}
(
1 if j = i
(6.19.10) yj′′ =
0 if j ∈ T ∖ {i}
(
e − 1 + s′j if j = f − 1
(6.19.11) zj′ =
e + s′j if j ∈ T ∖ {f − 1}



 p + e − 2 + s′j = p + (zj′ − yj′ ) if j = f − 1
′ ′
p + e − 1 = (p − 1) + (zj − yj ) if j ∈ [k + 1, f − 2]



(6.19.12) zj′′ = e − 1 + s′j = −1 + (zj′ − yj′ ) if j = k
e + 1 + s′i = (zj′ − yj′ ) + yj′′

if j = i



e + s′ = z ′ − y ′

if j ∈
̸ {i} ∪ [k, f − 1]
j j j
(
{1} ∪ [s′j + 1, zj′ − 1] if j = f − 1
(6.19.13) Ij′ =
{0} ∪ [s′j + 1, zj′ − 1] if j ∈ T ∖ {f − 1}

′ ′′
{0} ∪ [p + sj − 1, zj − 1] if

 j =f −1
{0} ∪ [p, zj′′ − 1] if j ∈ [k + 1, f − 2]



′′
(6.19.14) Ij = {0} ∪ [s′j , zj′′ − 1] if j =k
′ ′′

{1} ∪ [si + 3, zj − 1] if j =i




{0} ∪ [s′ + 1, z ′ − 1]

if j ̸∈ {i} ∪ [k, f − 1]
j j
(
[1, e − 2] ∪ {zj′ − yj′ } if j = f − 1
(6.19.15) Vj′ =
[1, e − 1] ∪ {zj′ − yj′ } if j ∈ T ∖ {f − 1}
48 KALYANI KANSAL


′ ′
[1, e − 1] ∪ {p + zj − yj }

 if j =f −1
[2, e] ∪ {p + zj′ − yj′ } if j ∈ [k + 1, f − 2]



(6.19.16) Vj′′ = [2, e] ∪ {zj′ − yj′ } if j =k
[1, e − 2] ∪ {zj′ − yj′ }

if j =i




[1, e − 1] ∪ {z ′ − y ′ }

if j ̸∈ {i} ∪ [k, f − 1]
j j

The only pairs in P and P ′′ that are unmatched after applying matching strate-
gies in Remark 6.16(i) and Remark 6.16(ii) are those in {(f − 1, zf′ −1 − yf′ −1 ), (i, e −
1)} ⊂ P and {(f −1, e−1), (k, e)} ⊂ P ′′ . As valp (β(i, e−1)) = 0 = valp (β(k, e)) and
i ̸= k, JVAH
⃗′ ⃗′
(χ1 , χ2 ) = JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ) if and only if (i, e − 1) matches (f − 1, e − 1),
t ,s t ,s
while (f − 1, zf′ −1 − yf′ −1 ) matches (k, e). Suppose this is true and (m, κ) =
α(f − 1, zf′ −1 − yf′ −1 ) = α(k, e). Plugging this data into the formula for κ in
(4.10.2), we get:

(6.19.17) valp (β(f − 1, zf′ −1 − yf′ −1 )) ≡ f − 1 − k mod f

Therefore, pf −1−k |β(f − 1, zf′ −1 − yf′ −1 ) = p(zf′ −2 − yf′ −2 ) + p2 (zf′ −3 − yf′ −3 ) + ... +
pf (zf′ −1 − yf′ −1 ) and we have:

p(zf′ −2 − yf′ −2 ) + p2 (zf′ −3 − yf′ −3 ) + ... + pf (zf′ −1 − yf′ −1 )

m≤
pf −1−k
(zf′ −2 − yf′ −2 ) ′
(zk+1 ′
− yk+1 )
= (zk′ − yk′ ) + p(zk−1
′ ′
− yk−1 ) + ... + pk+1 (zf′ −1 − yf′ −1 ) + f −2−k
+ ... +
p p
< (pf − 1)e + (zk′ − yk′ ) + p(zk−1 ′ ′
− yk−1 ) + ... + pf −1 (zk+1
′
− yk+1′
)
= α(k, e) = m
Contradiction. Therefore, JVAH
⃗′ ⃗′
(χ1 , χ2 ) ̸= JVAH (χ1 , χ2 ).
t ,s ⃗
′′ t ,s⃗′′
For Lemma 6.19(iii), we have:
(
1 if j = f − 1
(6.19.18) yj′ =
0 if j ∈ T ∖ {f − 1}
(
1 if j = i
(6.19.19) yj′′ =
0 if j ∈ T ∖ {i}
(
e − 1 + s′j if j = f − 1
(6.19.20) zj′ =
e + s′j if j ∈ T ∖ {f − 1}

p + e − 2 + s′j = p + (zj′ − yj′ ) if j



 =f −1
p + e − 1 = (p − 1) + (z ′ − y ′ ) if j

∈ [i + 1, f − 2]
(6.19.21) zj′′ = ′ ′ ′
j j


 e + sj = zj − yj if j =i
′ ′ ′
e + sj = zj − yj if j ̸∈ [i, f − 1]

 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 49

(
{1} ∪ [s′j + 1, zj′ − 1] if j = f − 1
(6.19.22) Ij′ =
{0} ∪ [s′j + 1, zj′ − 1] if j ∈ T ∖ {f − 1}

′ ′′

{0} ∪ [p + sj − 1, zj − 1] if j
 =f −1
{0} ∪ [p, z ′′ − 1]

if j ∈ [i + 1, f − 2]
(6.19.23) Ij′′ = ′
j
′′
{1} ∪ [sj + 2, zj − 1]
 if j =i

{0} ∪ [s′j + 1, zj′ − 1] if j ̸∈ [i, f − 1]


(
[1, e − 2] ∪ {zj′ − yj′ } if j = f − 1
(6.19.24) Vj′ =
[1, e − 1] ∪ {zj′ − yj′ } if j ∈ T ∖ {f − 1}

[1, e − 1] ∪ {p + zj′ − yj′ }



 if j =f −1
[2, e] ∪ {p + z ′ − y ′ }

if j ∈ [i + 1, f − 2]
(6.19.25) Vj′′ = j
′
j
′


 [2, e − 1] ∪ {zj − yj } if j =i
′ ′
[1, e − 1] ∪ {zj − yj } if j ̸∈ [i, f − 1]


Every pair in P matches with some pair in P ′′ upon applying matching strategies
in Remark 6.16(i) and Remark 6.16(ii). Therefore, JVAH⃗′ ⃗′
(χ1 , χ2 ) = JVAH
⃗
′′ ⃗′′
(χ1 , χ2 )!
t ,s t ,s
We omit demonstrating the calculations for Lemma 6.19(iv), Lemma 6.19(v) and
Lemma 6.19(vi). They proceed similar to the calculations above, and the findings
for all pairs described in Lemma 6.19 can be summarized as follows:

Proposition 6.20. Suppose Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ are a pair of non-isomorphic, non-
Steinberg Serre weights.
Then there exist GK characters χ1 and χ2 such that |JVAH ⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via
t ,s

Lemma 4.21(iii), |JVAH (χ1 , χ2 )| = ef −1 via Lemma 4.21(iii) and JVAH (χ1 , χ2 ) =
⃗
′′ t ,s⃗′′ ⃗′ ⃗′ t ,s

JVAH (χ1 , χ2 ) if and only if

t ,s⃗′′
⃗
′′

• After reindexing if necessary, s⃗′ and s⃗′′ satisfy either of the below for some
i<f −1 :
– (s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (1, 0, ..., 0, ∈ [0, p − 2]);
(s′′f −1 , s′′f −2 , ..., s′′i+1 , s′′i ) = (p−1, p−1, ..., p−1, s′i +1) (Lemma 6.19(iii)).
– s′f −1 ∈ [1, p − 1], (s′i , s′i−1 ..., s′0 ) = (p − 1, p − 1, ..., p − 1);
s′′f −1 = s′f −1 − 1, (s′′i , s′′i−1 ..., s′′0 ) = (1, 0, ..., 0) (Lemma 6.19(v)).
′′ f −1−i
+ j∈T t′j , and j∈T tj ≡ 1 +
P P P
• With i as above, j∈T dj ≡ 1 − p
′ f
P
j∈T tj mod p − 1.

Remark 6.21. In both the cases listed in Proposition 6.20, (sf −1 , sf −2 , ..., s0 ) ≡
(s′f −1 − 2, s′f −2 , ..., s′0 ). We leave the precise specification of the highest weight to
the reader.

6.21.1. Case 3. : |JVAH

⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(i); |JVAH (χ1 , χ2 )| =
t ,s ⃗
′′ t ,s⃗′′
ef − 1 via Lemma 4.21(iii).
50 KALYANI KANSAL

Case 3a. : i = f − 1.
Comparing ways of writing χ−1
2 χ1 in terms of ⃗ s, s⃗′ and s⃗′′ , we have:
X
e − 2 − s′f −1 + pf −1−j (s′j + e)
j∈T ∖{f −1}
X X
≡ (e − 2 + s′′f −1 ) + pf −1−j (s′′j + e) ≡ pf −1−j (sj + e)
j∈T ∖{f −1} j∈T

(6.21.1)
⇐⇒ (−2 − s′f −1 , s′f −2 , ..., s′0 ) ≡ (−2 + s′′f −1 , s′′f −2 , ..., s′′0 ) ≡ (sf −1 , sf −2 , ..., s0 ).
Comparing ways of writing χ2 , we have:
X X
1+ pf −1−j d′′j ≡ s′f −1 + 1 ≡ pf −1−j tj mod pf − 1
j∈T j∈T
X X
(6.21.2) ⇐⇒ pf −1−j d′′j ≡ s′f −1 , f −1−j
p tj ≡ s′f −1 + 1 mod pf − 1
j∈T j∈T

Lemma 6.22. The condition in (6.21.1) is satisfied for some s⃗′ , s⃗′′ and ⃗s if and
only if one of the following pairs describe s⃗′ and s⃗′′ :
(i) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [1, p − 2], 0, ..., 0, ∈ [1, p − 1]) where k ∈ [0, f −
2];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − s′f −1 , p − 1, ..., p − 1, s′k − 1).
(ii) (s′f −1 , s′f −2 , ..., s′0 ) = (∈ [1, p − 2], 0, ..., 0);
(s′′f −1 , s′′f −2 , ..., s′′0 ) = (p − s′f −1 − 1, p − 1, ..., p − 1).

Proof. Easy verification upon recalling that s′f −1 ≤ p − 2 by Lemma 4.21(i) and
s′′f −1 ≥ 1 by Lemma 4.21(iii). □

We omit the calculations for Lemma 6.22(i) which show that JVAH
⃗′ ⃗′
(χ1 , χ2 ) ̸=
t ,s

JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ). Briefly, (f − 1, e − 1) ∈ P and (k, e) ∈ P ′′ are the pairs in P and P ′′
t ,s
that don’t match using matching strategies Remark 6.16(i) and Remark 6.16(ii).
Both β(f − 1, e − 1) and β(k, e) turn out to have p-adic valuation 0 and therefore,
by Remark 6.14, α(f − 1, e − 1) ̸= α(k, e).
For Lemma 6.22(ii), all pairs in P end up matching with some pair in P ′′ via
Remark 6.16(i) or Remark 6.16(ii) (details omitted). Therefore, in this situation,
JVAH (χ1 , χ2 ) = JVAH (χ1 , χ2 ).
t⃗′ ,s
⃗′ t⃗
′′ ,s⃗′′

Case 3b. : i < f − 1.

Comparing ways of writing χ−1 ⃗′ ⃗′′
2 χ1 in terms of s , s and ⃗
s, we have:
X
e − 2 − s′f −1 + pf −1−j (s′j + e)
j∈T ∖{f −1}
X X
f −1−i
≡p (e − 2 + s′′i ) + pf −1−j (s′′j + e) ≡ pf −1−j (sj + e)
j∈T ∖{i} j∈T

(6.22.1)
⇐⇒ (−2 − s′f −1 , s′f −2 , ..., s′0 ) ≡ (s′′f −1 , ..., s′′i+1 , −2 + s′′i , s′′i−1 , ..., s′′0 )
≡ (sf −1 , sf −2 , ..., s0 )
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 51

Comparing ways of writing χ2 , we have:

X X
pf −1−i + pf −1−j d′′j ≡ s′f −1 + 1 ≡ pf −1−j tj mod pf − 1
j∈T j∈T
(6.22.2)
X X
⇐⇒ pf −1−j d′′j ≡ s′f −1 + 1 − pf −1−i , pf −1−j tj ≡ s′f −1 + 1 mod pf − 1
j∈T j∈T

Lemma 6.23. The condition in (6.22.1) is satisfied for some s⃗′ , s⃗′′ and ⃗s if and
only if one of the following pairs describe s⃗′ and s⃗′′ :
(i) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1]) for some k ∈
[i + 1, f − 2],
s′i ∈ [0, p − 3];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 2 − s′f −1 , p − 1, ..., p − 1, s′k − 1),
s′′i = s′i + 2.
(ii) (s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (∈ [0, p − 2], 0, ..., 0, ∈ [0, p − 2]);
(s′′f −1 , s′′f −2 , ..., s′′i+1 , s′′i ) = (p − 2 − s′f −1 , p − 1, ..., p − 1, s′i + 1).
(iii) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1]) for some k ∈
[i + 1, f − 2],
(s′i , s′i−1 , ..., s′l+1 , s′l ) = (p−1, p−1, ..., p−1, ∈ [0, p−2]) for some l ∈ [0, i−1];
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 2 − s′f −1 , p − 1, ..., p − 1, s′k − 1),
(s′′i , s′′i−1 , ..., s′′l+1 , s′′l ) = (1, 0, ..., 0, s′l + 1).
(iv) (s′f −1 , s′f −2 , ..., s′k+1 , s′k ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 1]) for some k ∈
[i + 1, f − 2],
(s′i , s′i−1 , ..., s′0 ) = (p − 1, p − 1, ..., p − 1);
(s′′f −1 , s′′f −2 , ..., s′′k+1 , s′′k ) = (p − 1 − s′f −1 , p − 1, ..., p − 1, s′k − 1),
(s′′i , s′′i−1 , ..., s′′0 ) = (1, 0, ..., 0).
Proof. Easy verification upon recalling that s′f −1 ≤ p − 2 by Lemma 4.21(i) and
s′′i ≥ 1 by Lemma 4.21(iii). □
For each of the pairs in the statement of Lemma 6.23, we omit the details of the
calculations comparing JVAH⃗′ ⃗′
(χ1 , χ2 ) and JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ). For pairs in Lemma 6.23
t ,s t ,s

(i), (iii) and (iv), JVAH

⃗′ ⃗′
(χ1 , χ2 ) ̸= JVAH (χ1 , χ2 ). For the pair in Lemma 6.23(ii),
t ,s ⃗
′′ t ,s⃗′′

JVAH
⃗′ ⃗′
(χ1 , χ2 ) = JVAH (χ1 , χ2 ).
t ,s ⃗t ,s⃗′′
′′

Proposition 6.24. Suppose Vt⃗′ ,s⃗′ and Vt⃗′′ ,s⃗′′ are a pair of non-isomorphic, non-
Steinberg Serre weights.
Then there exist GK characters χ1 and χ2 such that |JVAH ⃗′ ⃗′
(χ1 , χ2 )| = ef − 1 via
t ,s

Lemma 4.21(i), |JVAH

⃗
′′ ⃗′′
(χ1 , χ2 )| = ef − 1 via Lemma 4.21(iii) and JVAH
⃗′ ⃗′
(χ1 , χ2 ) =
t ,s t ,s

JVAH
⃗
′′ ⃗′′
(χ1 , χ2 ) if and only if (after reindexing if necessary) s⃗′ , s⃗′′ , ⃗s, ⃗t′ and t⃗′′ and
t ,s
⃗t satisfy either of the conditions below (we describe ⃗s only upto equivalence for the
sake of clarity.):
• (s′f −1 , s′f −2 , ..., s′0 ) = (∈ [1, p − 2], 0, ..., 0);
(s′′f −1 , s′′f −2 , ..., s′′0 ) = (p − s′f −1 − 1, p − 1, ..., p − 1);
(s , s −2 , ..., s0 ) ≡ (p − 3 − s′f −1 , p − 1, ..., p − 1);
Pf −1 ff −1−j
d′′j ≡ s′f −1 + j∈T pf −1−j t′j mod pf − 1, and
P
j∈T p
52 KALYANI KANSAL

pf −1−j tj ≡ s′f −1 +1+ pf −1−j t′j mod pf −1 (Lemma 6.22(ii)).

P P
j∈T j∈T

• There exists some i ≤ f − 2 such that:

(s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (∈ [0, p − 2], 0, ..., 0, ∈ [0, p − 2]);
(s′′f −1 , s′′f −2 , ..., s′′i+1 , s′′i ) = (p − 2 − s′f −1 , p − 1, ..., p − 1, s′i + 1) ;
(sf −1 , sf −2 , ..., si+1 , si , si−1 , ..., s0 ) ≡ (p − 2 − s′f −1 , p − 1, ..., p − 1, s′i −
1, s′i−1 , ..., s′0 );
pf −1−j d′′ ≡ s′f −1 + 1 − pf −1−i + j∈T pf −1−j t′j mod pf − 1, and
P P
Pj∈T f −1−j j
tj ≡ s′f −1 +1+ j∈T pf −1−j t′j mod pf −1 (Lemma 6.23(ii)).
P
j∈T p

Remark 6.25. For p > 2 and i = f − 2, the second condition in the proposition
above is identical to that required for Ext1F[GL2 (k)] (Vt⃗′ ,s⃗′ , Vt⃗′′ ,s⃗′′ ) to be non-zero via
Proposition 2.1(i)(b). The relationship between (⃗t′ , s⃗′ ) and (t⃗′′ , s⃗′′ ) is asymmetric
showing that only 1 family sees the type II intersection. For such a family wit-
nessing a type II intersection, the two associated type I intersections correspond to
1
ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 via Proposition 2.1(i)(b) and HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′′ ,s⃗′′ )) ̸=
0. In particular, when e > 1, f > 1,
Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0 guarantees the existence of both a type I intersection
and a type II intersection, while HomGL2 (k) (V⃗t,⃗s , H 1 (GK , Vt⃗′′ ,s⃗′′ )) ̸= 0 only guaran-
tees a type I intersection.

7. Conclusion
Theorem 7.1. Let p > 2 be a fixed prime. Let K be a finite extension of Qp , with
ring of integers OK and residue field k. Set e = e(K/Qp ), f = f (K/Qp ). Let X
be the reduced part of the Emerton-Gee stack for GL2 constructed in [EG1], defined
over a finite field F. Let V⃗t,⃗s and Vt⃗′ ,s⃗′ be a pair of non-isomorphic, non-Steinberg
Serre weights for GL2 (k). Consider the irreducible component XV⃗t,⃗s (resp. XVt⃗′ ,s⃗′ )
of X with the property that ρ ∈ X (F) is a point of XV⃗t,⃗s (resp. XVt⃗′ ,s⃗′ ) if and only
if V⃗t,⃗s (resp. Vt⃗′ ,s⃗′ ) is a Serre weight of ρ.
Then XV⃗t,⃗s and XVt⃗′ ,s⃗′ intersect in codimension 1 if and only if one of the fol-
lowing list of criteria holds. Next to each criterion we indicate in parenthesis the
type of intersection.
1
(i) V⃗t,⃗s and Vt⃗′ ,s⃗′ are both weakly regular and ExtF[GL 2 (OK )]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.
(Type I if e = 1, Type I or II or both if e > 1).
(ii) V⃗t,⃗s and Vt⃗′ ,s⃗′ are not both weakly regular, f > 1 and one of the follow-
ing is true after possibly interchanging V⃗t,⃗s and Vt⃗′ ,s⃗′ and possibly chang-
ing the indices of {sj }j , {s′j }j , {tj }j and {t′j }j by adding a fixed integer.
We also indicate when Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) is non-vanishing, or when
1
ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) is vanishing but Ext1F[GL2 (OK )] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) is non-
vanishing, or when HomGL2 (k) (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ )) is non-vanishing but it
1
is not known whether it contributes to ExtF[GL 2 (OK )]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) or not. If
1
nothing is mentioned, it means that ExtF[GL 2 (OK )]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) is vanishing.
(a) (sf −1 , sf −2 , ..., sf −i , sf −1−i ) = (∈ [0, p − 2], p − 1, ..., p − 1, ∈ [0, p − 2]),
where i ∈ [1, f − 1];
(s′f −1 , s′f −2 , ..., s′f −i , s′f −1−i ) = (p − sf −1 − 2, 0, ..., 0, sf −1−i + 1);
INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 53

pf −1−j t′j ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).

P P
j∈T
When i = 1, this implies Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.
(b) (sf −1 , sf −2 , ..., s0 ) = (∈ [0, p − 3], p − 1, ..., p − 1);
(s′f −1 , s′f −2 , ..., s′0 ) = (p − 3 − sf −1 , 0, ..., 0);
f −1−j ′
tj ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
(c) (sf −1 , sf −2 , sf −3 , ..., s0 ) = (p − 1, p − 2, p − 1, ..., p − 1);
(s′f −1 , s′f −2 , s′f −3 , ..., s′0 ) = (p − 2, 0, 0, ..., 0);
f −1−j ′
tj ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
When f = 2, this implies Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.

When e = 1, f > 1, we additionally have:

(d) (sf −1 , sf −2 ) = (p − 1, ∈ [0, p − 3]);
(s′f −1 , s′f −2 ) = (p − 1, sf −2 + 2);
f −1−j ′
tj ≡ −p + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
1
This implies ExtF[GL 2 (k)]
(V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0, HomGL2 (k) (V⃗t,⃗s , H 1 (K1 , Vt⃗′ ,s⃗′ )) ̸=
0.
(e) f > 2,
(sf −1 , sf −2 , sf −3 , ..., sf −i , sf −1−i ) = (p − 1, p − 1, p − 1, ..., p − 1, ∈
[0, p − 2]), where i ∈ [2, f − 1];
(s′ , s′ −2 , s′f −3 ..., s′f −i , s′f −1−i ) = (p − 1, 1, 0, ..., 0, sf −1−i + 1);
Pf −1 ff −1−j
t′j ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P
j∈T p
(f ) f = 2,
(sf −1 , sf −2 ) = (p − 2, p − 1);
(s′f −1 , s′f −2 ) = (p − 1, 1);
f −1−j ′
tj ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
(g) f > 2,
(sf −1 , sf −2 , sf −3 , ..., s0 ) = (p − 2, p − 1, p − 1, ..., p − 1);
(s′f −1 , s′f −2 , s′f −3 , ..., s′0 ) = (p − 1, 1, 0, ..., 0);
f −1−j ′
tj ≡ −1 − s′f −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
(h) (sf −1 , sf −2 , ..., sf −1−i , sf −2−i ) = (∈ [0, p − 2], 0, ..., 0, ∈ [1, p − 2]) for
some i ∈ [1, f − 2];
(s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i ) = (p − sf −1 − 2, p − 1, ..., p − 1, sf −2−i +
1);
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pi+1 + j=0 t′j mod pf − 1 (Type II).
(i) f > 2,
(sf −1 , sf −2 , ..., sf −1−i , sf −2−i , sf −3−i , ..., sf −m , sf −1−m )
= (∈ [0, p − 2], 0, ..., 0, 0, 0, ..., 0, ∈ [1, p − 1]) for some m ∈ [3, f − 1];
(s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i , s′f −3−i , ..., s′f −m , s′f −1−m )
= (p − sf −1 − 2, p − 1, ..., p − 1, 1, 0, ..., 0, sf −1−m ) where i ∈ [1, m − 2];
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pi (s′f −1−i + 1) + j=0 t′j mod pf − 1 (Type II).
(j) (sf −1 , sf −2 , ..., s1 , s0 ) = (∈ [0, p − 3], 0, ..., 0, 0);
(s′f −1 , s′f −2 , ..., s′1 , s′0 ) = (p − 1 − sf −1 , p − 1, ..., p − 1, p − 1);
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pf −1 (s′0 + 1) + j=0 t′j mod pf − 1 (Type II).
(k) f > 2,
(sf −1 , sf −2 , ..., sf −1−i , sf −2−i , sf −3−i , ..., s0 ) =
54 KALYANI KANSAL

(∈ [0, p − 3], 0, ..., 0, 0, 0, ..., 0);

(s′f −1 , s′f −2 , ..., s′f −1−i , s′f −2−i , s′f −3−i , ..., s′0 ) =
(p − 2 − sf −1 , p − 1, ..., p − 1, 1, 0, ..., 0) where i ∈ [1, f − 2];
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pi (s′f −1−i + 1) + j=0 t′j mod pf − 1 (Type II).
(l) (sf −1 , sf −2 , ..., sf −i , sf −1−i , sf −2−i , sf −3−i , ..., s0 ) =
(p − 2, 0, ..., 0, 0, 0, 0, ..., 0);
(s′f −1 , s′f −2 , ..., s′f −i , s′f −1−i , s′f −2−i , s′f −3−i , ..., s′0 ) =
(0, p − 1, ..., p − 1, p − 1, 1, 0, ..., 0) where i ∈ [2, f − 1];
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pi (s′f −1−i + 1) + j=0 t′j mod pf − 1 (Type II).
(m) f = 2,
(sf −1 , sf −2 ) = (p − 2, 0);
(s′f −1 , s′f −2 ) = (1, p − 1)
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ p2 + j=0 t′j mod pf − 1 (Type II).
(n) f > 2,
(sf −1 , sf −2 , sf −3 , sf −4 ..., s0 ) =
(p − 2, 0, 0, ..., 0);
(s′f −1 , s′f −2 , s′f −3 , ..., s′0 ) = (0, p − 1, 1, 0, ..., 0);
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ p(s′f −2 + 1) + j=0 t′j mod pf − 1 (Type II).
(o) f > 2,
(sf −1 , sf −2 , sf −3 , ..., sf −1−i , sf −2−i ) = (p − 1, 1, 0, ..., 0, ∈ [1, p − 2])
where i > 1;
(s′f −1 , s′f −2 , s′f −3 , ..., s′f −1−i , s′f −2−i ) = (p−1, p−1, p−1, ..., p−1, sf −2−i +
1);
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pi (s′f −1−i + 1) + j=0 t′j mod pf − 1 (Type II).
(p) f >2
(sf −1 , sf −2 , sf −3 , ..., s1 , s0 ) = (p − 1, 1, 0, ..., 0, p − 1);
(s′f −1 , s′f −2 , s′f −3 , ..., s′1 , s′0 ) = (1, 0, 0, ..., 0, p − 1);
Pf −1 Pf −1
sf −1 + 1 + j=0 tj ≡ pf −1 (s′0 + 1) + j=0 t′j mod pf − 1 (Type II).

When e > 1, we additionally have:

(q) sf −1 ≤ p − 3;
s′f −1 = sf −1 + 2;
f −1−j ′
tj ≡ −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
This implies Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0, but Ext1F[GL2 (OK )] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸=
0.
(r) (sf −1 , sf −2 , ..., sf −i , sf −1−i ) = (p − 1, p − 1, ..., p − 1, ∈ [0, p − 2]), where
i ≥ 1;
(s′ , s′ −2 , ..., s′f −i , s′f −1−i ) = (1, 0, ..., 0, sf −i−1 + 1);
Pf −1 ff −1−j
t′j ≡ −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P
j∈T p
(s) (sf −1 , sf −2 , ..., s0 ) = (p − 2, p − 1, ..., p − 1);
(s′f −1 , s′f −2 , ..., s′0 ) = (1, 0, ..., 0);
f −1−j ′
tj ≡ −1 + j∈T pf −1−j tj mod pf − 1 (Type I).
P P
j∈T p
(t) (sf −1 , sf −2 , ..., si+1 , si ) = (1, 0, ..., 0, ∈ [0, p − 2]), where i < f − 1;
(s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (p − 1, p − 1, ..., p − 1, si + 1);
′ f −1−i
+ j∈T tj mod pf − 1 (Type II).
P P
j∈T tj ≡ 1 − p
 INTERSECTIONS OF COMPONENTS OF EMERTON-GEE STACK FOR GL2 55

(u) sf −1 ∈ [1, p−1], (si , si−1 ..., s0 ) = (p−1, p−1, ..., p−1), where i < f −1;
s′f −1 = sf −1 − 1, (s′i , s′i−1 ..., s′0 ) = (1, 0, ..., 0);
′ f −1−i
+ j∈T tj mod pf − 1 (Type II).
P P
j∈T tj ≡ 1 − p
(v) (sf −1 , sf −2 , ..., s0 ) = (∈ [1, p − 2], 0, ..., 0);
(s′f −1 , s′f −2 , ..., s′0 ) = (p − sf −1 − 1, p − 1, ..., p − 1);
f −1−j ′
tj ≡ sf −1 + j∈T pf −1−j tj mod pf −1 (Type II).
P P
j∈T p
(w) (sf −1 , sf −2 , ..., si+1 , si ) = (∈ [0, p − 2], 0, ..., 0, ∈ [0, p − 2]) for some
i ≤ f − 2;
(s′f −1 , s′f −2 , ..., s′i+1 , s′i ) = (p − 2 − sf −1 , p − 1, ..., p − 1, si + 1);
f −1−j ′
tj ≡ sf −1 +1−pf −1−i + j∈T pf −1−j tj mod pf −1 (Type
P P
j∈T p
II). When i = f −2, this agrees with Item (a), corresponds to both Type
I and II intersections, and implies Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0.

Proof. By Section 4.13, we need to find the criteria for when there exist two GK
characters χ1 and χ2 such that LV⃗t,⃗s (χ1 , χ2 ) ∩ LVt⃗′ ,s⃗′ (χ1 , χ2 ) ⊂ Ext1GK (χ2 , χ1 ) has
dimension ef − 1 and the same is true for most unramified twists of χ1 and χ2 .
The criteria are covered in Propositions 5.2, 5.6, 6.2, 6.4, 6.18, 6.20 and 6.24. In
each of these, we have constraints on ⃗s and s⃗′ that do not depend on ⃗t and ⃗t′ . We
Pf −1 f −1−j ′ Pf −1
similarly have constraints on j=0 p tj − j=0 pf −1−j tj mod pf − 1 that
also do not depend on ⃗t and ⃗t′ . Collectively, the two sets of constraints define the
criteria completely.
In other words, if and only if V⃗t,⃗s and Vt⃗′ ,s⃗′ satisfy the criteria in one of Propo-
sitions 5.2, 5.6, 6.2, 6.4, 6.18, 6.20 and 6.24.
The above criteria are necessary and sufficient when K ̸= Qp . When K = Qp ,
XV⃗t,⃗s ∩ XVt⃗′ ,s⃗′ is codimension 1 if and only if either the above criteria hold or
the intersection contains an irreducible representation (by ??). The criterion for
existence of irreducible representations in XV⃗t,⃗s ∩ XVt⃗′ ,s⃗′ is given in Lemma 4.14.
Putting all the criteria together gives the list in the statement of the Theorem,
along with Proposition 2.14 and Corollary 2.10 on computations of extensions of
Serre weights as GL2 (OK )-modules.
Remark 7.2. In fact, our criteria show that when σ and τ are non-isomorphic,
non-Steinberg Serre weights, then
1
ExtF[GL 2 (OK )]
(σ, τ ) ̸= 0 =⇒ dim Xσ ∩ Xτ = [K : Qp ] − 1.
This follows from the stronger statement in the proof of Proposition 2.14 when
K/Qp is unramified and Corollary 2.10 when K/Qp is ramified.
□

Theorem 7.3. In the setup of Theorem 7.1, assume that V⃗t,⃗s and Vt⃗′ ,s⃗′ are both
weakly regular and that XV⃗t,⃗s and XVt⃗′ ,s⃗′ intersect in codimension 1. Let n be the
number of [K : Qp ] − 1 dimensional irreducible components in the intersection.
Then the following are true:
(i) If e = 1, then n = 1.
None of these components of dimension [K : Qp ] − 1 are contained in
triple intersections of irreducible components of X .
56 KALYANI KANSAL

(ii) If e > 1 and f = 1, then

(
2 if Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0
n=
1 if Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0
If s, s′ < p − 3, then each component of dimension [K : Qp ] − 1 is con-
tained in a triple intersection.

(iii) If e > 1 and f > 1, then

(
2 if Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) ̸= 0
n=
1 if Ext1F[GL2 (k)] (V⃗t,⃗s , Vt⃗′ ,s⃗′ ) = 0
Each of these components of dimension [K : Qp ] − 1 is contained in a
triple intersection.
Proof. When e = 1, the statements are a consequence of collating criteria for type
I and type II intersections in Corollary 5.7 and Proposition 6.4. When f = 1, the
relevant results are in Propositions 5.2 and 6.2. When e > 1, f > 1, they are in
Corollary 5.7 and Remark 6.25. □

References
[AJL] H. H. Andersen, J. Jørgensen, and P. Landrock. The projective indecomposable modules
of SL(2, pn ). (1983).
[BP] Christophe Breuil and Vytautas Paskunas. Towards a modulo p Langlands correspon-
dence for GL2 . American Mathematical Society, 2012.
[CEGS] Ana Caraiani, Matthew Emerton, Toby Gee, and David Savitt. The geometric Breuil-
Mézard conjecture for two-dimensional potentially Barsotti-Tate Galois representations.
arXiv preprint arXiv:2207.05235 (2022).
[DDR] Lassina Dembélé, Fred Diamond, and David P Roberts. Serre weights and wild ram-
ification in two-dimensional Galois representations. In Forum of Mathematics, Sigma,
volume 4. Cambridge University Press, 2016.
[EG1] Matthew Emerton and Toby Gee. Moduli stacks of étale (φ, Γ)-modules and the existence
of crystalline lifts. Preprint (2019).
[EG2] Matthew Emerton and Toby Gee. Moduli stacks of (phi, Gamma)-modules: a survey.
arXiv preprint arXiv:2012.12719 (2020).
[EGH] Matthew Emerton, Toby Gee, and Eugen Hellman. An introduction to the categorical
p-adic Langlands program. Notes for IHES Summer school on Langlads Program (2022).
[Gee] T. Gee. On the weights of mod p Hilbert modular forms. Inventiones Mathematicae
184(2011).
[Sta] The Stacks Project Authors. Stacks Project. https://stacks.math.columbia.edu, 2018.
[Ste] Misja FA Steinmetz. Explicit Serre weights for two-dimensional Galois representations
over a ramified base. arXiv preprint arXiv:2002.06029 (2020).