Geometric Representation Theory:

Lecture notes from a class taught by

Weiqiang Wang

George H. Seelinger
 1. Quivers and Quiver Algebras
(8/22/2017) Lecture 1. This class will cover some topics in geometric
representation theory. Our general outline is
(1) Quiver representations, including Dynkin quivers and Gabriel’s clas-
sification of indecomposables.
• Non-simply laced [Geiss, Lederc, Schröer] (arXiv 1410.1403)
• “Affine” (Euclidean) quivers
(2) Hall algebras, including “classical” Hall algebras which leads to
Hall-Littlewood polynomials, and Ringel (over Fq ), which has con-
nections to half quantum groups.
(2.5) Springer theory. This will be covered in a reading course and not in-
cluded here. However, convolution algebra on “Steinberg varieties”
leads to Weyl groups and representations.
(3) Quiver varieties. From this and the above, we get the Lusztig
(semi)canonical basis, which leads to Nakajima’s work on integrable
modules of Kac-Moody semisimple Lie algebras. Type A is covered
in Chriss-Ginzburg, chapter 4.
1.1. Definition. A quiver is a tuple Q = (I, Ω, s, t) (sometimes denoted
(Q0 , Q1 , s, t)) where I is the vertex set, Ω is the edge set, and s, t : Ω → I
such that, for h ∈ Ω, h 7→ s(h), t(h), respectively. We say h starts at s(h)
and terminates at t(h).
h
s(h) t(h)

1.2. Definition. Given a quiver Q, we say |Q| is the underlying graph

with direction removed.

Throughout these lecture notes, unless otherwise stated, we will always as-
sume Q is connected and that we are working of ground field k.
1.3. Example. Examples of small quivers include:
(a)

1
(b)
1 2
(b’)
1 2 3
(c)
1 2

3
 1.4. Definition. A representation of a quiver Q consists of the data
• To each i ∈ I, assign k-vector space Vi .
• To each h ∈ Ω, assign a linear map xh : Vs(h) → Vt(h)
1.5. Definition. A morphism of representations f : V → W consists of
linear maps fi : Vi → Wi and makes the following diagram commute.
xh
Vs(h) Vt(h)
fs(h) ft(h)
yh
Ws(h) Wt(h)

1.6. Definition. Given representations V, W , we define Hom(V, W ) =

HomQ (V, W ) as the set of all morphisms from V, W . We also let AutQ (V )
be all invertible morphisms in Hom(V, V ).
We will denote the category of (not necessarily finite-dimensional) rep-
resentations of Q as Rep(Q). A basic problem of this course is to classify
(simple, indecomposable) representations of Q, up to isomorphism.
1.7. Example. (a) Consider 1.3(a). This is the same as asking the
classification of all f : kn → kn . If k = C, then this is well-known
by the Jordan canonical form for square matrices.
(b) Consider 1.3(b) and assign V1 to vertex 1 and V2 to vertex 2. Then,
ifx : V1 → V2 has rank r, it is of the form, up to a change of basis
for V1 and V2 ,  
Ir×r 0
x≈
0 0
Remember that we can make sense of subrepresentations and quotient
representations of Q.
L
1.8. Theorem. Rep(Q) is closed under , taking kernel/cokernel, and
subreprresentations and quotients. That is, (ker f )i := ker(fi : Vi → Wi ).
Hence, Rep(Q) is an abelian category.
Proof. We must show that ker f is actually a representation. If v ∈
(ker f )s(h) , then (ft(h) ◦ xh )(v) = (yh ◦ fs(h) )(v) = 0 by the definition of a
representation morphism and the fact that v is in the kernel of fs(h) . Thus
xh (v) ∈ (ker f )t(h) and xh induces a map from (ker f )s(h) → (ker f )t(h) . We
can apply a similar argument for cokernel.

If we define ⊕ in the obvious way, (V ⊕ W )i = Vi ⊕ Wi , we get a direct

sum representation. □
1.9. Definition. A path in a quiver Q (of length ℓ) is a series of edges
in Q, say (h1 , . . . , hℓ ) such that t(hi ) = s(hi+1 ) for all i. We denote such a
path as hℓ · · · h2 h1 : s(h1 ) → t(hℓ ).

4
 1.10. Definition. A path algebra kQ is given as the free module with
basis given by paths in the quiver Q endowed with a multiplication operator
on basis elements f, g given by

0 t(g) ̸= s(f )
f ·g =
 s(g) g t(g) f t(f ) t(g) = s(f )

1.11. Definition. A path of length 0 in a quiver Q is of the form ei for

i ∈ I. For p ∈ kQ, we say
( (
p t(p) = i p s(p) = i
ei · p = p · ei =
0 else 0 else
1.12. Proposition. ei is an orthogonal idempotent of kQ.
Proof. Consider that e2i = ei and ei ej = 0 for i ̸= j by definition of ei .
(Note, as stated, this is not actually true.) □
1.13. Proposition. kQ is associative with 1 := i∈I ei .
P

1.14. Proposition. kQ is Z≥0 -graded by path lengths.

For quiver Q = (I, Ω, s, t), we get (kQ)0 = i∈I kei
L
1.15. Example.
and (kQ)1 = h∈Ω kh.
L

1.16. Proposition. kQ is finite-dimensional if and only if Q has no

oriented cycles.
Proof. If Q has an oriented cycle, then kQ may be infinite-dimensional.
Consider Q as 1.3(a). Then, kQ is the algebra of all polynomials in 1
generator, k[x]. For the forwards direction, ... □
1.17. Example. Consider the path algebra of 1.3(b). This algebra has
dimension 3: 2 paths of length 0 and 1 path of length 1. Similarly, 1.3(b’)
has 3 paths of length 0, 2 paths of length 1, and 1 path of length 2, giving
us a total of 6 paths.
1.18. Proposition. Let A := kQ. Then,
M
A= Aei
i∈I
and Aei is a projective left A-module. Furthermore, given an A-module M ,
we get
HomA (Aei , M ) ∼
= ei M
Proof. The decomposition of A is a standard fact for primitive or-
thogonal idempotents of an algebra. Aei is projective because it is a direct
summand of the free A-module A.

To show the isomorphism, we first note that f ∈ HomA (Aei , M ) is

uniquely determined by f (ei ). This is because f must be A-linear, so

5
f (aei ) = a.f (ei ) for all a ∈ A. Furthermore, f (ei ) ∈ ei M because f (ei ) =
f (e2i ) = e1 .f (ei ). Thus, consider the A-module homomorphism given by
ϕ : HomA (Aei , M ) → ei M given by f 7→ f (ei ). We already know that ϕ is
injective since, if ϕ(f ) = ϕ(g), then f (ei ) = g(ei ) =⇒ f = g. Furthermore,
ϕ is injective since, given ei m ∈ ei M , there is an f ∈ Hom(Aei , M ) such
that f (ei ) = m. □
1.19. Proposition. Given 0 ̸= f ∈ Aei and 0 ̸= g ∈ ei A, it must be
that f g ̸= 0. This follows from the fact that f g must pass through i and has
nonzero length.
1.20. Proposition. The ei ’s are primitive idempotents, that is, Aei is
indecomposable.

Proof. We wish to show that the only idempotent in EndA (Aei ) is

ei since, if Aei ∼
= M ⊕ N for non-trivial M, N , then Aei ↠ M ,→ Aei
would be an idempotent in EndA (Aei ). By 1.18, EndA (Aei ) ∼ = ei Aei . Take
idempotent f ∈ ei Aei . Then, f 2 = f = f ei , so f (f − ei ) = 0 =⇒ f = ei by
the preceding proposition. □
1.21. Proposition. Aei ∼
̸ Aej as A-modules for i ̸= j
=
Proof. If Aei ∼ = Aej , then take f ∈ HomA (Aei , Aej ) ∼ = ei Aej (by
previous proof) to be an isomorphism and g ∈ ej Aei to be its inverse. Then,
f g = ei =⇒ ei ∈ Aej A. This is only possible when i = j. □
= kQ-Mod.
1.22. Proposition. Rep(Q) ∼
L
Proof. Consider the map (Vi )i∈I 7→ V = i∈I Vi where
(
xhℓ · · · xh1 (vi ) i = s(hi )
hℓ · · · hq · v =
0 else
This sends a representation (Vi ) to a kQ-module.

Conversely, given a kQ-module M , we can construct L an inverse to the

map above by letting Mi := ei M and taking M = i∈I ei M and letting
h ∈ Ω be h : Ms(h) → Mt(h) . We note that h(Mi ) ⊆ Mj , so the collection
(Mi ) satisfies the definition of a representation of Q. □

Due to this theorem, we will often use representations of Q and kQ-

modules interchangably.

(8/24/2017) Lecture 2.
1.23. Definition. Let Q be a quiver. For i ∈ I, we define the represen-
tation
k at i
(
S(i) :=
0 else

6
 1.24. Theorem. If a quiver Q has no oriented cycles, then {S(i)}i∈I is
a full list of simple kQ-modules.
Proof by Example. Let V be a simple representation of quiver
1 2
with V1 ̸= 0 ̸= V2 . Then, consider the inclusion of the representation given
below.
0 V2 = e2 V

V1 V2
Because there are no edges starting from 2, there can be no non-zero map
starting at V2 in V , so we have found a sub-representation of V , which is a
contradiction. The full proof is of the same spirit, but replace 1 and 2 with
i and j where, since Q is acyclic, there is a j such that no edges start at
j. □
1.25. Definition. Let A = kQ. Then, we define
P (i) := Aei
which is projective, as noted earlier.
Since P (i) consists of all paths starting at i, then, if there is an edge
going from i to j, P (i) has a submodule isomorphic to P (j).
1.26. Proposition. For an acyclic quiver Q, we get P (i)/ rad P (i) ∼ =
hd(P (i)) ∼
= S(i) where rad P (i) is the intersection of all maximal submodules
of P (i).
Proof. □
1.27. Example. Let Q be the quiver
1 2 3
Then, we get that P (1), P (2), P (3) are given respectively by
k ∼
k ∼
k , 0 k ∼
k , 0 0 k
One fast consequence of this is that P (3) = S(3).
1.28. Proposition. Let V ∈ Rep(Q). Then,
HomQ (P (i), V ) = Vi
Proof. Let A = kQ. Then,
HomQ (P (i), V ) ∼
= HomkQ (Aei , V ) = ei V
from above. Now, since, for v ∈ V ,
(
v if v ∈ Vi
ei .v = xei (v) =
0 else

7
it must be that ei V ⊆ Vi . However, it is also clear that xei is the identity
on Vi , os Vi ⊆ ei V . □
1.29. Proposition. Assume Q has no oriented cycles. Then, {P (i)}i∈I
is the full set of projective indecomposable modules (PIM) in Rep(Q).
1.30. Lemma. If Q has no oriented cycles, End(P (i)) = k.

Proof of Lemma. Using previous results, we have that Hom(P (i), P (i)) ∼ =
∼
ei P (i) = ei Aei since P (i) = Aei . However, since Q has no oriented cycles,
= k. □
the only path that begins and ends at i is the trivial path, so ei Aei ∼

Proof of Proposition. The lemma gives us that P (i) is indecom-

posable. Now, consider a projective module P and let P ′ be a direct
sum of P (i)’s with multiplicity ni = dim Hom(P, S(i)). Then, we get that
Hom(P, S(i)) ∼ = kni I think) from the previous proposi-
= Hom(P ′ , S(i))(∼
tion. Thus, for any representation V , we can (somehow) show Hom(P, V ) ∼
=
Hom(P ′ , V ). Thus, P ∼
= P ′. □
1.31. Definition. The Grothendieck group of a category C, denoted
K(C), is an abelian group generated by letters [M ] for M ∈ C and with the
relations that A − B + C = 0 if there exists a short exact sequence
0→A→B→C→0
1.32. Definition. For some quiver Q, we denote K(Q) := K(Rep(Q)).
1.33. Proposition. Assume that Q has no oriented cycles. Then,
(a) The Grothendieck group K(Q) ∼
= ZI with isomorphism given by
[V ] 7→ dimV
where dimV = (dim Vi )i∈I . This is called the graded dimension.
(b) {[S(i)]}i∈I form a basis for K(Q).
(c) {[P (i)]}i∈I form a basis for K(Q).

The last two statements are similar to having a basis given by a diagonal
matrix versus an upper triangular matrix.

We now move into some abstract nonsense (which we will not prove here)
to lay the groundwork for the standard resolution.
1.34. Proposition. Let A be an associative algebra with 1 and M ∈ A-
mod. Then, we have
M∼ = A ⊗A M ∼ = (A ⊗k M )/I
where I is the k-subspace spanned by ab ⊗ m − a ⊗ bm for all a ∈ A, m ∈
M, b ∈ A.

This proposition comes in two variations.

8
 1.35. Proposition (Variation 1). Replace “b ∈ A” by “b ∈ L” where L
is a generating set of A. Reformulated as an exact sequence in A-mod, we
get
d1 d0
A ⊗ L ⊗k M → A ⊗k M → M → 0
where d1 sends a ⊗ ℓ ⊗ m 7→ aℓ ⊗ m − a ⊗ ℓm and d0 sends a ⊗ m 7→ am.

Proof. The proof follows from the identity

ab1 b2 ⊗m−a⊗b1 b2 m = (ab1 b2 ⊗m−ab1 ⊗b2 m)+(ab1 ⊗b2 m−a⊗b1 b2 m) ∈ I
□
1.36. Proposition (Variation 2). Take subalgebra A0 ⊆ A and k-subspace
L ⊆ A such that A0 L ⊆ L, LA0 ⊆ L, and ⟨A0 , L⟩ = A. Then,
A ⊗A0 L ⊗A0 M → A ⊗A0 M → M → 0
is exact.
1.37. Theorem (The Standard Resolution). Let V ∈ kQ-Mod. Then,
there exists an exact sequence of kQ-modules
d1 M d0
P (t(h)) ⊗ kh ⊗ Vs(h) →
M
0→ P (i) ⊗ Vi → V → 0
h∈Ω i∈I

given by maps p ⊗ h ⊗ v 7→ ph ⊗ v − p ⊗ xh (v) and p ⊗ v 7→ xp (v).

L Proof. Using variation 2, we take A = kQ, A0 = i∈I kei , and L =

L

h∈Ω kh. Then,

M M M M
A ⊗A0 V = ( Aei ) ⊗ V = (Aei ⊗ V ) = Aei ⊗ ei V = Aei ⊗ Vi
i∈I i∈I i∈I i∈I

It remains to show that d1 is injective. Assume

d
X X
pn ⊗ hn ⊗ vn 7→1 (pn hn ⊗ vn − pn ⊗ xhn (vn )) = 0
n n
We show that ker d1 is trivial by contradiction. Let ℓ = the maximum length
of the pn . We will then show that all such pn hn ⊗ vn are actually zero. Take
the terms in the sum (image of d1 ) with length ell + 1, say
X
p n hn ⊗ v n = 0
ℓ(pn )=ℓ

Since none of these collected pn hn are equal to each other, the collection
{pn hn }ℓ(pn )=ℓ is linearly independent. Thus, it must be that vn = 0, which
means pn hn ⊗ vn = 0, so we have contradicted the maximality of ℓ and thus
d1 is injective. □
1.38. Corollary. (a) Rep(Q) has enough projectives, and so Exti
makes sense.

9
 (b) For all V, W ∈ Rep(Q), we have
ExtiQ (V, W ) = 0, i > 1

Proof. Given an arbitrary representation V , the standard resolution is

a projective resolution of the form
0 → P1 → P0 → V → 0
= kn for some n and is thus always free. Then computing long
since Vi ∼
exact sequence
· · · → Exti−1 (P1 , W ) → ExtiQ (V, W ) → ExtiQ (P0 , W ) → · · ·
However, for i ≥ 1, Exti (P, W ) = 0 for any projective module P by definition
of a projective module. Thus, starting at i = 2, our long exact sequence
degenerates to
· · · → 0 → ExtiQ (V, W ) → 0 → · · ·
and thus, by exactness, ExtiQ (V, W ) = 0 for i > 1. □
1.39. Example. The standard resolution for S(i) is given by
M
0→ P (j) → P (i) → S(i) → 0
h : i→j

since (S(i))j = 0 for j ̸= i.

1.40. Proposition. For simple representations S(i), S(j) of a quiver Q,
dim Hom(S(i), S(j)) = δij
Ext1 (S(i), S(j)) = #{edges i → j}

Proof. From the projective resolution of S(i), we get the long exact Prove first part
sequence of proposition
L
0 → Hom(S(i), S(j)) Hom(P (i), S(j)) Hom( h : i→j P (j), S(j))

Ext1 (S(i), S(j)) Ext1 (P (i), P (j)) = 0

Now, if i ̸= j, then Hom(P (i), S(j)) = S(j)i = 0 by earlier proposition.
If i = j, then Hom(P (i), S(i)) = (S(i))i ∼ = k and, by the first part of the
proposition, Hom(S(i), S(i)) ∼
= k , so an injection between them must be an
isomorphism. Thus, we get
∼ M ∼
0 → Hom(S(i), S(j)) → Hom(P (i), S(j)) → 0 → Hom( P (j), S(j)) → Ext1 (S(i), S(j)) → 0
h : i→j

thereby proving the second part of the propositon. □

1.41. Proposition. Any submodule of a projective module P ∈ Rep(Q)
is projective.

10
 Proof. It suffices to show that Exti≥1 (S, −) = 0. □
1.42. Remark. There exist “injective” counterparts using Q(i) := (ei A)∗ ,
the graded dual.
1.43. Example. Let Q be the quiver I do not see how
1 2 3 this example is
relevant
Then, we get indecomposables S(1), S(2), S(3) as well as
k ∼
k 0 , 0 k ∼ k , k ∼
k ∼ k
(8/29/2017) Lecture 3. Since we have 1.41, then Rep(Q) is heredi-
tary. In particular, it is quasi-hereditary, but the standard and costandard
modules are boring and, since we do not have any duality, BGG reciprocity
does not yield anything terribly interesting.
1.1. Variety of representions. For this section, let Q be a quiver and
k = k (eg C).
1.44. Definition. Let V ∈ kQ-Mod and (wi )i∈I = w
⃗ = dimV , that is
xh
to say, for h : i → j, we have Vi = kwi → kwj = Vj . Then, we define the
representation space of w⃗ to be
⃗ := { all representations of Q with dim = w}
Rep(w) ⃗
Homk (kws(h) , kwt(h) )
M
=
h∈Ω
M
= Homk (Vs(h) , Vt(h) )
h∈Ω

1.45. Definition. For w

⃗ = (wi )i∈I , let us define
GL(wi , k)
Y
GL(w)
⃗ :=
i∈I

1.46. Proposition. GL(w)

⃗ acts on Rep(w) ⃗ by conjugation as follows.
Let (gi ) = g ∈ GL(w).
⃗ Then, for all h : i → j, g sends
xh 7→ gj xh gi−1
1.47. Proposition. As vector spaces
X
dim Rep(w)
⃗ = ws(h) wt(h)
h∈Ω
and X
dim GL(w)
⃗ = wi2
i∈I

Proof. This is clear given the decompositions of Rep(w) ⃗ and GL(w) ⃗

when considering Hom spaces as matrices since these are linear transforma-
tions. □

11
 1.48. Proposition. For x, x′ ∈ Rep(w),
⃗ we get
{kQ-isomorphisms x → x′ } ↔ {g ∈ GL(w)
⃗ | gx = x′ }

Proof. □ Not sure why

this is true.
1.49. Corollary. Given x ∈ Rep(w),
⃗
Write out the
AutkQ (x) ∼
= StabG (x) =: Gx actual isomor-
phism diagrams.
Proof. It is clear that an element in the stabalizer of x is a kQ au-
tomorphism of x. By above, take x′ = x. Then, the set on the right is
StabG (x) and the set on the left is all kQ-automorphisms of x. □
⃗ Then, x ∼
1.50. Proposition. Let x, x′ ∈ Rep(w). = x′ if and only if x
′
and x are in the same G-orbit.

Proof. This follows directly from the correspondance between kQ-

isomorphisms between x → x′ and the set {g ∈ GL(w)
⃗ | gx = x′ }. □

Now, we list some facts about these orbits that follow from the fact that
GL(w)⃗ is a linear algebraic group (in particular, it is a variety) acting on
a finite-dimensional vector space. In the following few propositions, G :=
GL(⃗v ).
1.51. Proposition. Each GL(w)-orbit
⃗ of Rep(w)
⃗ is a variety.
1.52. Proposition. Given x ∈ Rep(w),
⃗ Gx is closed in G in the Zariski
topology.
1.53. Proposition. The orbit Ox = G/Gx . Furthermore,
Tx Ox = T1 G/T1 Gx , and dim Ox = dim GL(w)
⃗ − dim Gx
where dim is the dimension as a variety.
1.54. Proposition. There exists at most one orbit of maximum dimen-
sion equal to dim Rep(w).
⃗
1.55. Corollary. There exists at most one orbit that is dense and open
in dim Rep(w).
⃗
1.56. Proposition. Gx is connected.
1.57. Proposition. For V, W ∈ Rep(Q), there exists exact sequence
M M
0 → HomQ (V, W ) → Homk (Vi , Wi ) → Homk (Vs(h) , Wt(h) ) → Ext1Q (V, W ) → 0
i∈I h∈Ω

Proof. Apply HomQ (−, W ) to the standard resolution, use the fact
that HomQ (P (i), W ) = Wi and ExtQ (P (i), W ) = 0, as well as tensor-hom
adjointness, to get the final result. □

12
 1.58. Corollary. For V ∈ Rep(Q), and w ⃗ = dimV , there exists exact
sequence of vector spaces
M
0 → EndQ (V ) → ⃗ → ExtiQ (V, V ) → 0
End(Vi ) → Rep(w)
i∈I

Proof. Using the above proposition, take W = V . □

1.59. Proposition. Given V ∈ Rep(Q) with w
⃗ = dimV , we get
(a) dim Ext1Q (V, V ) = dim Rep(w)
⃗ − dim OV
1
⃗ := i∈I wi2 −
P
P ExtQ (V, V ) = dim EndQ (V ) − q(w)
(b) dim ⃗ where q(w)
h∈Ω ws(h) wt(h)

Proof. Given that the alternating sum of the dimensions of objects in

an exact sequence must be 0, we get from the short exact sequence in the
proposition above
M
dim Ext1Q (V, V ) − dim Rep(w)
⃗ + dim End(Vi ) − dim EndQ (V ) = 0
i∈I
Now,Lby 1.53, we know that dim OV = dim GL(w)−dim
⃗ GV = dim EndQ (V )−
dim i∈I End(Vi ) and thus the first equality is proven. understand
why L
dim GV =
From 1.47, we get that q(w)
⃗ = dim GL(w)−dim
⃗ ⃗ = dim EndQ (V )− dim i∈I End(Vi )
Rep(w)
⃗
dim Rep(W ) and thus the second equality is proven. □
1.60. Proposition. Let V ∈ Rep(w).
⃗ Then, OV is open if and only if
Ext1Q (V, V ) = 0.
Proof. This follows from the first equality in the above proposition.
From 1.55, if dim OV = dim Rep(w),
⃗ then OV is open and dense in Rep(w).
⃗
If OV is open, then dim OV = dim Rep(w) ⃗ by virtue of being open in the
Zariski topology. Thus by 1.59(a), we are done. □
⃗ ̸= 0 and q(w)
1.61. Proposition. Let w ⃗ ≤ 0. Then, there are infinity
many orbits in Rep(w)
⃗
Proof. The proof follows from 1.59 and the fact the orbits are not open,
but will not be reproduced here. □ reference?
1.62. Example. Let Q be given by
1 2
and let w
⃗ = (1, 1). Then, we have the following representations in Rep(w).
⃗
k ∼ k k 0 k
⃗ ∼
So, Rep(w) = Hom(k, k) ∼ = k by definition and GL(w) ⃗ ∼ = GL(1, k) ×
GL(1, k) ∼
= k × × k × by definition. Now, an element (λ, µ) ∈ GL(w)
⃗ acts on
x ∈ Rep(w)
⃗ by
(λ, µ)(x) = λxµ−1

13
So, it is clear that {0} is its own orbit and then k× is its own orbit, since
(yx−1 , 1)(x) = y for any x, y ∈ k× . Thus, there are two orbits corresponding
to the isomorphism classes of representations given above.

1.2. Closed orbits.

1.63. Proposition. Let
0 → V ′ → V → V ′′ → 0
be a short exact sequence of kQ-modules. Then, OV ′ ⊕V ′′ ⊆ OV

Proof. Identify V = V ′ ⊕ V ′′ as a vector space and extend a basis for

′
V to a basis for V . Then, we can view
 ′ 
x x12
xh =
0 x′′
Now, consider a 1 parameter subgroup g(t) ∈ GL(w)⃗ with t ∈ k× where
 
tIN ′ 0
g(t)i =
0 tIN ′′
Then, g(t) acts on xh via
x′ tx12
 
xh 7→ g(t)t(h) xh g(t)−1 =
s(h) 0 x′′
From here, we may take the limit t → 0 to see that the closure OV contains
the matrix  ′ 
x 0
0 x′′
This matrix exactly corresponds to the representations of V ′ ⊕ V ′′ , and so
we are done. □
1.64. Definition. Given a filtration F of a kQ-module V , say
V = V 0 ⊇ V 1 ⊇ ··· ⊇ V ℓ = 0
we define the associated graded representation by
ℓ−1
M
grF V = V m /V m+1
m=0

1.65. Proposition. Given a representation V , Ogr V ⊆ OV .

Proof. This follows by the proposition above and induction on ℓ. □ write the actual
proof.
1.66. Proposition. If closed OV ′ ≤ OV , where dimV ′ = dimV , then
V′ ∼
= grF V for some filtration F .
1.67. Proposition. Let V ∈ Rep(w). ⃗ Then, OV is closed if and only if
V is semisimple, that is to say V is a direct sum of simple representations.

14
 Proof. The proof is by the propositions above. OV is closed if and
only if V ∼
= gr V for every filtration if and only if V is semisimple. □
1.68. Corollary. The closure of every orbit OV contains a unique
closed orbit.
1.69. Definition. We denote the unique closed orbit of OV by OV ss
1.70. Corollary. Let Q be a quiver with no oriented cycles. Then, the
⃗ is {0}.
only closed orbit of Rep(w)

Proof. This follows from the fact that the complete list of simples is
{S(i)} and so any semisimple representation has xh = 0. □
1.71. Example. Let Q be

1
and let w
⃗ = (n). Then, Rep(w) ⃗ ∼= Mn (k) with action given by conjuga-
tion by GL(w)⃗ = GLn (k) on xh (the unique edge). Thus, the orbits are
conjugacy classes of matrices. By the above result, an orbit OV is closed
if and only if V is semisimple, which in this case is the same thing as xh
being given by a diagonalizable matrix. (The only simple matrices are 1 by
1). Thus, given an arbitrary matrix A ∈ Mn (k), the unique closed orbit
contained in OA is OAss where Ass is the diagonalizable matrix with the
same eigenvalues as A.

(8/31/2017) Lecture 4. In this lecture, we will discuss the Euler form

and Dynkin quivers.
1.72. Definition. For V, W ∈ Rep(Q) define
X
⟨V, W ⟩ := (−1)i dim ExtiQ (V, W ) = dim HomQ (V, W ) − dim Ext1Q (V, W )
i

Note how this resembles the Euler characteristic. Also, the second equal-
ity is due to the fact that higher Ext groups vanish in this category.
1.73. Proposition. (a) ⟨P (i), S(j)⟩ = δij and ⟨P (i), W ⟩ = dim Wi
for an arbitrary representation W ∈ Rep(Q).
(b) ⟨−, −⟩ is biadditive with respect to exact sequences, that is to say,
if we have exact sequence
0 → V ′ → V → V ′′ → 0
then ⟨V, W ⟩ = ⟨V ′ , W ⟩ + ⟨V ′′ , W ⟩. Note, this is called the “Euler-
Poincare” principle.

15
 Proof. We know that Hom(P (i), W ) = Wi by 1.18 and Ext1Q (P (i), W )
is trivial for any W since P (i) is projective. Thus,
⟨P (i), W ⟩ = dim Hom(P (i), W ) − dim Ext1Q (P (i), W ) = dim Wi − 0
If we take W = S(i), then we get
⟨P (i), S(i)⟩ = dim(S(j))i − 0 = δij .
For the second part, this follows by getting the long exact sequence by
applying Hom(−, W ) to 0 → V ′ → V → V ′′ → 0. □
1.74. Theorem. Let V, W ∈ Rep(Q), then
(a) ⟨V, W ⟩ only depends on ⃗v := dimV and w ⃗ := dimW . So, we have
a (Euler) form
⟨·, ·⟩ : ZI × ZI → Z
(b) X X
⟨⃗v , w⟩
⃗ = ⟨V, W ⟩ = v i wi − vs(h) wt(h)
i∈I h∈Ω

Proof. For the second part, recall the standard resolution

0 → P1 → P0 → V → 0
Then, by the proposition above, ⟨V, W ⟩ = ⟨P0 , W ⟩ − ⟨P1 , W ⟩. Moreover,
M
dim Hom(P1 , W ) = dim Hom( P (t(h)) ⊗ Vs(h) , W )
h∈Ω
X
= dim Hom(P (t(h)) ⊗ Vs(h) , W )
h∈Ω
vs(h)
X M
= dim Hom( P (t(h)), W )
h∈Ω j=1
X
= vs(h) dim Hom(P (t(h)), W )
h∈Ω
X
= vs(h) wt(h)
h∈Ω

and similarly for dim Hom(P0 , W ). Finally, since dim Ext1 (P0 , W ) = 0 =
dim Ext1 (P1 , W ), we get our desired identity. □
1.75. Definition. The associated quadratic form (the so-called Tits
form) is given by
X X
q(⃗v ) := ⟨⃗v , ⃗v ⟩ = vi2 − vs(h) vt(h)
i∈I h∈Ω

1.76. Definition. We define a symmetric bilinear form (·, ·) : ZI × ZI →

Z by
⃗ = ⟨⃗v , w⟩
(⃗v , w) ⃗ + ⟨w,
⃗ ⃗v ⟩

16
 1.77. Example. Consider the quiver
1 2
with representations given by
P (1) = k ∼
k , P (2) = S(2) = 0 k
Then, ⟨P (1), P (2)⟩ = dim(P (2))1 = 0 and ⟨P (2), P (1)⟩ = dim(P (1))2 = 1.
Thus, we see that ⟨·, ·⟩ is not symmetric and (P (1), P (2)) = 1
1.78. Theorem. Let nij be the number of edges between i and j in |Q|.
Then, X X
(⃗v , w)
⃗ = vi wi (2 − 2nij ) − nij vi wj
i i̸=j

Now, equipped with these forms, we move on to a discussion of Dynkin

quivers.
1.79. Theorem. Let Q be a quiver and assume |Q| is connected. Then
(a) qQ is positive definite if and only if |Q| is Dynkin, that is, it is of
types A, D, or E.
(b) qQ is positive semidefinite if and only if |Q| is Euclidean, that is, it
is an “untwisted affine ADE”.
Understand why
this theorem is
··· true.

···

···

Figure 1. Dynkin graphs An , Dn , and E6,7,8

···

···

Figure 2. Euclidean Dynkin graphs Ân and D̂n

17
 1.80. Remark. The difference from the “Lie” setting is here nij ∈ Z
√
where in the “Lie” setting, nij = aij aji .
Proof of Theorem. The second part of the theorem is a “boundary”
case. Any graph containing ADE
\ has qQ indefinite and any graph not
containing ADE is covered by the first case.
\ □
1.81. Corollary. Consider the “star” graph Γ(p, q, r)
q

···

1 2 ··· p

2
···

r
Then, Γ(p, q, r) has positive-definite qQ if and only if
1 1 1
+ + >1
p q r
which happens if and only if (p, q, r) is of the following forms

(p, q, 1) Ap+q−1

(p, 2, 2) Dp+2

(p, 3, 2) p = 3, 4, 5 Ep+3

Understand why
A natural question one may ask is about types B, C, F, G. To do this, this is a corol-
we will have to do some more work. Due to the resemblence of these con- lary and why it
structions to Lie theory, we will use some similar definitions. is true.
1.82. Definition. Assume |Q| has no edge loop, that is nii = 0. Then,
we define the Cartan matrix to be
CQ := C|Q| = 2I − A
where A = (nij ) is the adjacency matrix.
This matrix can be used to define Kac-Moody algebra g = gQ (or simple
Lie algebra if Q is Dynkin). Then, we get the following.
1.83. Definition. We define the root latice L = i∈I Zαi ∼ = ZI where
L
αi = dimS(i) are the simple roots. This gives rise to simple reflections
si : L → L defined by α 7→ α − (α, αi )αi and thus Weyl group W = ⟨si | i ∈
I⟩ ≤ GL(L ⊗Z R
1.84. Proposition. (a) W preserves (·, ·).

18
 (b) For Dynkin (or Euclidean), the root system of g, R = {α ∈ L \ 0 |
(α, α) ≤ 2} (or = 2 if Dynkin).

Now, we are ready to discuss Gabriel’s Theorem. We require k = k.

1.85. Definition. A quiver Q is of finite type if the number of indecom-
posable representations of Q of any fixed dim = w ⃗ is finite.
1.86. Theorem (Gabriel’s Theorem). A connected quiver Q is of finite
type if and only if |Q| is Dynkin (ie types ADE). Furthermore, if Q is
Dynkin then there is a one-to-one correspondance between indecomposables
and positive roots in the root system R.

Proof. For the forward direction, assume Q is of finite type. Then,

for 0 ̸= ⃗v ∈ NI , there exists finitely many representations of dim = ⃗v . So,
GL(⃗v ) acting on Rep(⃗v ) has finitely many orbits. Therefore, Rep(⃗v ) = Ox
for some x ̸= 0 . Now, dim Rep(⃗v ) = dim Ox = dim GL(⃗v ) − dim Gx by . why?
Thus, we have
why?
1 ≤ dim Gx = dim GL(⃗v ) − dim Rep(⃗v ) = q(⃗v )
since the identity matrix, I ∈ Gx =⇒ dim Gx ≥ 1 =⇒ q(⃗v ) ≥ 1. This
completes the forward direction. We still must lay more groundwork to
prove the rest of this theorem. □
1.87. Proposition. Given ⃗u ∈ ZI and ⃗v such that vi = |ui | for all i,
then q(⃗v ) ≤ q(⃗u).

Proof.
X X X X
q(⃗v ) = |ui |2 − |us(h) ||ut(h) | ≤ u2i − us(h) ut(h) = q(⃗u)
h
□
1.88. Example. Consider the quiver
1 2
finish this exam-
ple
(9/5/2017) Lecture 5.

1.3. Reflection functors. Starting off, this section is completely gen-

eral and not restricted to quivers of type ADE.
1.89. Definition. If a vertex i in a quiver Q has no edges leaving i, we
call i a sink.

19
Similarly, if a vertex i in a quiver Q has no edges entering i, we call i a
source.

1.90. Remark. Note that

i
is neither a sink nor a source.
1.91. Definition. We define reflections s↑i and s↓i to flip all arrows in-
cident to i and are defined as follows
s↑i

source s↓i
sink

1.92. Proposition. s↓i s↑i (Q) = Q for i a sink in Q.

Prove this or at
least do a basic
1.93. Proposition. Assume |Q| is a tree. Let Ω, Ω′ be 2 orientations of
example.
|Q|. Then, Q′ is obtained from Q via a sequence of s↓i and s↑i operations.
Proof. The proof proceeds by indcution on the number of vertices.
Locate a sink or a source with one incident edge, which must exist since |Q|
is a tree. Call it k. Then, by the inductive hypothesis, one can obtain the
orientation on the other vertices consistent wiht Q′ . At this point, either
the edge incident to k is in the correct orientation or it is not. If not, apply
sk and you are done. □
1.94. Definition. For a sink i of Q, let Q′ = s↓i (Q). Define
Φ↓i : Rep(Q) → Rep(Q′ )
and take V ∈ Rep(Q). Then, if V ′ = Φ↓i (V ), we have that Vj′ = Vj when
j ̸= i and !
↓
M
′
Vi = Φi (V )i = ker Vk → Vi
k→i
Thus, Φ↓i only changes the vector space at i and the maps to and from i.
1.95. Definition. For source i at Q, let Q′ = s↑i (Q). Similar to the
above, we can define
Φ↑i : Rep(Q) → Rep(Q′ )
where Φ↑i (V )i = coker (Vi → i→k Vk )
L

1.96. Remark. These functors are called the BGP reflection functors.

20
 ↓
1.97. Example. We noteL that Φi (S(i)) = 0Lwhere i is a sink in Q.
This happens because ker ( k→i Vk → Vi ) but, k→i Vk = 0. Similarly
↑ L
0 since Vi → i→k Vk = 0 so the codomain is trivial and thus
Φi (S(i)) = L
coker(Vi → i→k Vk ).
1.98. Example. Let Q be the quiver
1 2
and consider the representation

V = k 0

Then, Φ↓2 (V )2 = ker ( → V2 ) = ker (V1 → V2 ) = ker(k → 0). So,

L
k→2 Vk
we get
V ′ := Φ↓2 (V ) = k k
Now, if we apply Φ↑2 (V ′ ), we see that we get coker (k → k) = 0, so we actu-
ally recover V . Now, take Φ↑1 (V ′ ). Since ker ( k→1 Vk → V1 ) = ker (k → k) =
L
0. So,
W := Φ↓2 (V ′ ) = 0 k
Similarly, Φ↑1 (W ) = V ′ since we are taking coker(0 → k) ∼ = k. Now, notice
that these are the 3 indecomposables and the correspond to the 3 positve
roots (2 simple). To see this, note that |Q| is of type A2 , so it has 2 simple
roots β = (0, 1) and α = (1, 0). Then, we get the correspondance using
graded dimension

(k 0) (1, 0) is simple

(k k) (1, 1)

(0 k) (0, 1) is simple

1.99. Example. Take the quiver Q and V , a representation, to be

1 2 3 , 0 0 k
Then, we can apply Φ↓2 to V to get V ′ . We compute that
!
Vk → V2 = ker(k → 0) = k
M
V2′ = ker
k→2

So,
V′ = 0 k k

21
Doing similar computations, we get the following diagram
0 k k
Φ↓3
Φ↓1
k k k k k 0
Φ↓3
Φ↓1
k k 0
Φ↓2

k 0 0
1.100. Example. Let Q be a quiver and V , a representation of Q, given
by
1 2 3 0 k 0
If we apply Φ↑1 (V ) =: V ′ , we see that
!
= coker(0 → k) = k
M
V1′ = coker V1 → Vk
1→k

So, we get the following

Φ↑1 (V ) = k k 0 Φ↑3 (V ) = 0 k k
where the second equality follows from symmetry. However, note that
!
↓
Vk → V2 = ker(0 → k) = 0 =⇒ Φ↓2 (V ) = 0
M
Φ2 (V )2 = ker 0 0
k→2

Thus, we have found a situation where Φ↓i and Φ↑i are not inverses! Finish these
computations.
1.101. Proposition. For sink i of Q, s↓i is left exact. Furthermore, the
right derived functor Rn Φ↓i = 0 for n > 1. Finally,
↓
1 ↓
M
R Φi (V ) = 0 ⇐⇒ Vk → Vi is surjective. Call this property i
k→i

1.102. Definition. In the case of the above proposition, we define

i←
Rep
↓ is the full subcategory of Rep(Q) consisting of objects V that satisfy
i .

1.103. Proposition. For source i of Q, Φ↑i is right exact. Furthermore,

the left derived functor Ln Φ↑i = 0 for n > 1. Finally,
↑
1 ↑
M
L Φi (V ) = 0 ⇐⇒ Vi → Vk is injective. Call this property i .
i→k

22
 1.104. Definition. In the case of the above proposition, we define
i→
Rep
↑ is the full subcategory of Rep(Q) consisting of objects V that satisfy
i .

1.105. Remark. We will end up showing that Repi← (Q) and Repi→ (Q)
contain all indecomposables of Q except for S(i). In a sense, this encodes
the same fact that a reflection si on a positive set of roots R+ takes the set
R+ \ {αi } to R+ \ {αi }. This becomes explicitly manifest in ??. I understand
this, but get
Proof of Above Propositions. □ a much better
handle on it.
1.106. Proposition. For i a sink of Q, let Q′ = s↓ (Q). Then, Φ↓i and
And what is this
Φ↑i are an adjoint pair. That is, for W ∈ Rep(Q) and V ′ ∈ Rep(Q′ ), we get reference?
HomQ (Φ↑i (V ′ ), W ) ∼ ↓
= HomQ′ (V ′ , Φi (W )) Understand this
proof...
Proof. Consider that the cokernel is the universal object such that the
following diagram commutes
′
L
i→k Vk

0
Vi′ coker(Vi′ →
L
i→k Vk )

Similarly, the kernel is the universal object such that the following diagram
commutes
L
k→i Wk

L 0
ker( k→i Wk → Wi ) Wi

So, let f ∈ HomQ (Φ↑i (V ′ ), W ) map coker(Vi′ →

L
i→k Vk ) → Wi . Then, we
get
L
i→k fk
L
k Wk
pr
0
Vk′
L L
i→k ker( k Wk → Wi ) Wi
fi
∃!

0
Vi′ coker(Vi′ →
L
i→k Vk )
0

23
Thus, a map f ∈ HomQ (Φ↑i (V ′ ), W ) induces a unique map in HomQ′ (V ′ , Φ↓i (W )).
Similarly, given a g ∈ HomQ′ (V ′ , Φ↓i (W )), we get
L
k gk

Vk′
L
i→k

0
Vi′ coker(Vi′ →
L L
i→k Vk ) k Wk
gi
∃!
L 0
ker( k Wk → Wi ) Wi
0

So, a map g ∈ HomQ′ (V ′ , Φ↓i (W )) corresponds to a unique map in HomQ (Φ↑i (V ′ ), W ).

Therefore, Φ↑i and Φ↓i form an adjoint pair. □

1.107. Proposition. With the same situation as above, Φ↓i (W ) ∈ Repi→ (Q′ )
and Φ↑i (V ′ ) ∈ Repi← (Q).

Proof. By definition, Repi→ (Q′ ) is all representations that have Vi ,→

L ↓ L L
i→k Vk . However, Φi (W )i = ker( k→i Wk → Wi ) ,→ i→k Wk , by defi-
nition. Note that the reflection functor changes all edges that were k → i
into i → k edges.

Similarly, Repi← (Q) is all representations that have k→i Vk ↠ Vi and

L

Φ↑i (V ′ )i = coker(V ′ → ′ by definition, and since ′

L L
i k→i V k ) ↠ V i i→k Vk ↠
′ ′ ′
L
coker(Vi → k→i Vk ) ↠ Vi , we are done. □

1.108. Proposition. As functors, Φ↑i and Φ↓i are inverses between Repi→ (Q)
and Repi← Q

Proof. We want to show that Φ↑i Φ↓i (W ) ∼= W for W ∈ Repi← (Q).

i←
Now, since W ∈ Rep (Q), then, by definition, we get
M
Wk ↠ Wi
k→i
Thus, it must be that
! !
Φ↑i (Φ↓i (W ))i = coker ker
M M
Wk ↠ W i → Wk
k→i k
! !
M M
= Wk / ker Wk ↠ Wi
k k→i

24
 M
= im Wk ↠ Wi
k→i
= Wi
□
1.109. Proposition. Let Q be arbitrary quiver. Let si = sαi : ZI → ZI
be the simple reflection sending α 7→ α − (α, αi )αi , see 1.83.
(a) For W ∈ Repi← (Q),
dimΦ↓i (W ) = si (dimW )
(b) For V ′ ∈ Repi→ (Q),
dimΦ↑i (V ′ ) = si (dimV ′ )

Proof. For the first part, we know that W has the property that
L ↓ P
k→i Wk ↠ Wi so dim(Φi (W ))i = k→i (dim Wk ) − dim Wi . First, we
compute that, if j is not incident to i, then si (αj ) = αj , if k → i, then
   
X X X X
(αk , αi ) =  (αk )j (αi )j − (αk )s(h) (αi )t(h) + (αi )j (αk )j − (αi )s(h) (αk )t(h) 
j∈I h∈Ω j∈I h∈Ω

However, since (αi )j = δij , (αk )j = δkj , and i ̸= k, then the vertex indexed
sums will be 0. Furthermore, there is only one edge k → i, so the first sum
will be 1 and the other will be 0. Thus, we get
(αk , αi ) = −1 =⇒ si (αk ) = αk − (αk , αi )αi = αk + αi
We also compute that
X X
(αi , αi ) = 2 (αi )2j − 2 (αi )s(h) (αi )t(h) = 2
j∈I h∈Ω

and so si (αi ) = αi − (αi , αi )αi = −αi . Now, we compute that

 
X X
si (dimW ) = si  wj α j + (wk αk ) + wi αi 
j not incident to i k→i
X X
= wj αj + (wk (αk + αi )) − wi αi
j not incident to i k→i
X X
= wj α j + (wk αi ) − wi αi
j̸=i k→i

= dimΦ↓i (W )
A similar computation works for part (b). □
1.110. Remark. Let D(Q) := D(Rep(Q)). Then, D(Q) ∼
= D(Q′ ). Cite lecture
notes by Mili-
cic.
25 Always? Or
Q, Q′ as above?
 (9/7/2017) Lecture 6.
1.111. Proposition. Let Q be a quiver.
(a) For i a sink for Q and Q′ = s↓i Q, if V ∈ Repi← (Q), W ∈ Rep(Q),
then
↓ ↓
HomQ (V, W ) ∼
= HomQ′ (Φi V, Φi W )
(b) For i a source for Q, if V ∈ Rep(Q), W ∈ Repi→ (Q), then
↑ ↑
HomQ (V, W ) ∼
= HomQ′ (Φi V, Φi W )

Proof. For part (a), consider that V = Φ↑i (Φ↓i V ) by 1.108, then we can
apply the adjointness (1.106) to get
HomQ (Φ↑i (Φ↓i V ), W ) ∼ ↓ ↓
= HomQ′ (Φi V, Φi W )
A dual argument applies for part (b). □
1.112. Definition. Let ⃗v ∈ Z≥0 . Then, we say (vi )i∈I = ⃗v > 0 ⇐⇒
vi ≥ 0 for all i ∈ I.
1.113. Proposition. Let 0 ̸= V ∈ Rep(Q) be indecomposable, i a
sink/source. Then,
↓↑
(a) If V ∼
= S(i), then Φi (V ) = 0.
←
(b) If V ∼
̸ S(i), then V ∈ Repi→ (Q).
=
(c) Φ↓↑
i (V ) is nontrivial and indecomposable if and only if si (dimV ) ≥
0.
Proof. Part (a) follows from computations like in example 1.100. For
f
part (b), if V ̸∈ Repi← (Q), then k→i Vk → Vi is not surjective. So, we can
L
take Vi = im(f ) ⊕ Vi′′ with Vi′′ ̸= 0. Define V ′ , V ′′ ∈ Rep(Q) by
( (
′ Vj j ̸= i ′′ 0 j ̸= i
Vj = , Vj = ′′
im(f ) j = i Vi j = i
Then, V = V ′ ⊕ V ′′ , which contradicts the indecomposability of V . To show
that nontrivial, indecomposable V ∈ Repi→ for i a source, assume, as above,
L f
that k→i Vk → Vi is not surjective. Finish this part
of the proof.
For part (c), the proof follows from (a), (b), and 1.109. □ Maybe ask for
a hint!
1.114. Proposition. Let Q be a quiver and Q′ = s↑↓ Q.
Write the actual
(a) For sink i of Q, if V ∈ Rep(Q), W ∈ Repi← (Q), then
proof. I do not
↓ ↓
Ext1Q (V, W ) ∼
= Ext1Q′ (Φi V, Φi W ). see quite how
to connect these
(b) For source i of Q, if V ∈ Repi→ (Q), W ∈ Rep(Q), then
facts to get what
Ext1Q (V, W ) = Ext1Q′ (Φ↑i V, Φ↑i W ). we want. Help!

26
 1.115. Example. One could have assumed V, W ∈ Repi← to get more
restrictive theorems, but we can check that S(i), which is not in Repi← , will
work if put in the right spots. Namely,
HomQ (V, S(i)) = 0
ExtQ (S(i), W ) = 0
where the second equality follows because S(i) = P (i) when i is a sink.
Thus, we have found a useful way to remember that we can enlarge the
possible category for W in the Hom case and enlarge the possible category
of V in the Ext case.

Proof of Proposition. For part (a), if V ∼ = S(i), we just showed in

the above example that the left side is trivial and the right side is trivial
using part (a) of 1.113. Furthermore, by general homological algebra, we
know that
Vα , W ) ∼
M Y
Ext1Q ( = Ext1Q (Vα , W )
α α
and so it suffices to consider V as indecomposable. By part (b) of 1.113,
since V ∼
̸ S(i), then V ∈ Repi← (Q). Now, consider the short exact sequence
=
in Rep(Q) given by
0→W →M →V →0
representing an element in Ext1Q (V, W ). Using the right derived functor
RΦ↓i , we get long exact sequence
0 → Φ↓i W → Φ↓i M → Φ↓i V → R1 Φ↓i W → R1 Φ↓i M → R1 Φ↓i V → 0
However, R1 Φ↓i V = 0 = R1 Φ↓i W since V, W ∈ Repi← (Q) (1.101). So, this
forces R1 Φ↓i M = 0 =⇒ M ∈ Repi← (Q). Furthermore, we are left with the
short exact sequence
0 → Φ↓i W → Φ↓i M → Φ↓i V → 0
which represents an element in Ext1Q (Φ↓i V, Φ↓i W ). Thus, we have a map
Ext1Q (V, W ) → ExtQ′ (Φ↓i V, Φ↓i W ). We can then make a similar argument
with ExtiQ′ (Φ↓i V, Φ↓i W ) → Ext1Q (Φ↑i Φ↓i V, Φ↑i Φ↓i W ) = Ext1Q (V, W ). These two
maps are inverse to each other □ Actually check
this?
1.116. Proposition. Assume i is a sink (or source) of Q and Q′ = s↑↓ Q.
If, for a = 1, 2, 0 ̸= Ia ∈ Rep(Q) is indecomposable and Φ↓↑i (Ia ) ̸= 0 (thus
indecomposable), then
↓↑ ↓↑
HomQ (I1 , I2 ) ∼
= HomQ′ (Φi I1 , Φi I2 )
↓↑ ↓↑
Ext1Q (I1 , I2 ) ∼
= Ext1Q′ (Φi I1 , Φi I2 )

We now return back to the program of proving Gabriel’s theorem (1.86).

27
 1.117. Remark. For all quivers Q without edge loops, there exists a
correspondance between the indecomposable representations of Q and the
positive roots of its root system. In fact, the real positive roots have a one-
to-one correspondance, but the imaginary roots correspond to “a family” of
indecomposables. Counting these families gives rise to the Kac polynomials
Kα (q) = a0 q N + · · · + aN where N is the dimension of the “family”, aN is
the multiplicity of α as a root, and ai ∈ N. This is due to a conjecture by
Kac that has been subsequently proven.
|I|
1.118. Lemma. Assume Q is a Dynkin quiver and 0 ̸= ⃗v ∈ Z≥0 . Then,
there exists a sequence i1 , . . . , ik+1 ∈ I such that
i1 is a sink for Q
i2 is a sink for s↓i1 (Q)
i3 is a sink for s↓i2 s↓i1 (Q)
..
.
ik+1 is a sink for s↓ik · · · s↓i1 (Q)
and ⃗v ≥ 0, si1 (⃗v ) ≥ 0, . . . , sik · · · si1 (⃗v ) ≥ 0, but sik+1 sik · · · si1 (⃗v ) ̸≥ 0.
1.119. Definition. We say a sequence i1 , . . . , ik+1 ∈ I meeting only
the sink conditions (not necessarily the positivity ones), is adapted to an
orientation Ω of Q with no oriented cycles.

Now, with this lemma in hand (whose proof we will postpone), we seek
to prove the second part of Gabriel’s theorem.

Proof of Gabriel’s Theorem (continued). Given any indecompos-

able V ̸= 0 and ⃗v = dimV , we can construct a sequence i1 , . . . , ik+1 ∈ I
that is adapted to Ω. Then,
Φ↓i1 (V ) ̸= 0
Φ↓i2 Φ↓i1 (V ) ̸= 0
..
.
V ′ := Φ↓ik · · · Φ↓i1 (V ) ̸= 0

and Φ↓ik+1 (V ′ ) = 0. Thus, it must be that V ′ = S(ik+1 ) and so, we can work
backwards to get V = Φ↑i1 · · · Φ↑ik (S(ik+1 )) and thus, by 1.109,
dimV = si1 · · · sik (αik+1 ) ∈ R+
where the root is positive the lemma above. So, we have
Ext1 (V, V ) = · · · = Ext1 (S(i), S(i)) = 0
since S(i) = P (i), and so, by 1.59, OV is open in Rep(⃗v ). □

28
 Thus, our proof is complete modulo proving the lemma, which will be
our program for the next lecture.

(9/12/2017) Lecture 7. For a quiver Q, recall that we can take L :=

ZI to be a root lattice and take a set of simple roots {αi | i ∈ I} and then
get a Weyl group W = ⟨si ⟩i∈I that acts on the roots. Using this Weyl group,
we can define a total order on I = {i1 , . . . , ir }.
1.120. Definition. Let C := sir · · · si1 ∈ W . C is a Coxeter element of
W . A Coxeter element depends on a choice of simple roots and the ordering
of the simple roots.
1.121. Example. Take W = Sn and simple reflections given by {(i, i +
1)}i≤n−1 . Then a Coxeter element is (1, 2, . . . , n). So, if n = 4, then there
are 6 4-cycles, which are all Coexeter elements.
1.122. Proposition. All Coxeter elements are W -conjugate.

Proof. (Lifted from Humphrey’s “Reflection Groups and Coxeter Groups”

74–75). From basic Lie algebras, we know that all simple systems are W -
conjugate. So, it suffices to show that, for a fixed set of simples, the Coxeter
elements resulting from different orderings of the roots are conjugate. Any
cyclic permutation of indices will give a conjugate element
sn s1 · · · sn−1 = sn (s1 · · · sn )sn
Also, an interchange of an adjacent commuting pair of si , sj will not change
the Coxeter element. Using an inductive argument, one can show that all
permutations of the indices can be acheived using combinations of these two
types of permutations. □
1.123. Definition. We define |C| = h (the order of C in W ) to be the
Coxeter number.
1.124. Example. When W = Sn , we can take si = (i, i + 1) and thus
C = (1, 2, . . . , n) and so the Coxeter number is just n. When W = D2m ,
then C is a product of two generating reflections, and thus a rotation by 2kπ
m
where k is relatively prime to m, and so it has order m, which is thus the
Coxeter number.
1.125. Proposition. Let x ∈ LR = R ⊗Z L. Then, Cx = x if and only
if x ∈ Rad(·, ·), the radical of the form (·, ·).

Proof. Let Cx = x. Then,

sir · · · si1 x = x =⇒ sir−1 · · · si1 x = sir x
=⇒ sir−1 · · · si1 x − x = sir x − x = x − (x, αir )αir − x = (x, αir )αir
However, the right hand side must be a multiple of αir and the left hand
side must be a linear combination of {αi1 , . . . , αir−1 }, so the only way this

29
can happen is if both sides are 0. Thus, (x, αir ) = 0. Now, once can repeat
this process to get

(x, αir−1 ) = 0


.
sir−1 · · · si1 x = x =⇒ · · · =⇒ ..


(x, α ) = 0
i1

and thus x ∈ Rad(·, ·).

If x ∈ Rad(·, ·), then si (x) = x − (x, αi )αi = x by definition, so Cx =

sir · · · si1 x = x. □
1.126. Example. One non-trivial example is with quivers of type ADE.
\
There is only one imaginary root, δ and si (δi ) = δi for all i.
1.127. Proposition. Let Q be a Dynkin quiver. Then, C −1 is invertible
in LR .
Proof. When Q is Dynkin, Rad(·, ·) = 0 since (x, x) = 2 for all x ∈ L
(see 1.84). Thus, by 1.125, C has no fixed points in LR , so 1 is not an
eigenvalue of C. Thus, C − 1 is invertible as an endomorphism of LR . □
1.128. Proposition. Let Q be a Dynkin quiver. For all 0 ̸= α ∈ L,
there exists a k ̸= 0 such that C k α ̸≥ 0.
Proof. Let h be the order of C. Then, consider
Ch − 1
(α) = 0 =⇒ α + Cα + · · · + C n−1 α = 0
C −1
Then, C i α ̸≥ 0 for some i, otherwise the equality could not hold. □
1.129. Lemma. Given a quiver Q with no oriented cycles, there always
exist an ordering I = {i1 , . . . , ir } such that the ordering is adapted to Q.
(See 1.119).
Proof. One can take the ordering of listing j before i if there exists an
edge i → j. Since the quiver is acyclic, such an ordering exists. To show it
is adapted, □ Show it is
1.130. Example. Consider the quiver with the ordering adapted

1 4 3 5 6
Then, we can apply a series of reflections to get
2
s↓1 Q =
1 4 3 5 6

30
 2
s↓2 s↓1 Q =
1 4 3 5 6
2
s↓3 s↓2 s↓1 Q =
1 4 3 5 6
2
s↓4 s↓3 s↓2 s↓1 Q =
1 4 3 5 6
2
s↓5 s↓4 s↓3 s↓2 s↓1 Q =
1 4 3 5 6
So, the ordering is adapted.
1.131. Proposition. Using the choice of ordering from the lemma above,
C = sir · · · si1 satisfies
C ↓ Q = s↓ir · · · s↓i1 Q = Q
1.132. Example. Apply C to the example above and one recovers CQ =
Q.
Proof. For any edge h ∈ Ω, its orientation gets reversed exactly twice
when applying s↓ir · · · s↓i1 . □
Proof of Lemma 1.118. We use the above proposition. Consider the
sequence
⃗v , s↓i1 ⃗v , s↓i2 s↓i1 ⃗v , . . . , s↓ir · · · s↓i1 ⃗v = C⃗v ,
s↓i1 C⃗v , s↓i2 s↓i1 C⃗v , . . . , C 2⃗v ,
..
.
By 1.128, C ℓ v ̸≥ 0 for some ℓ. Thus, the lemma is proven. □ Why? This is
not clear to me
1.4. Longest Element and Ordering of Positive Roots. since the lemma
1.133. Definition. Let w◦ ∈ W be the largest element w◦ − siN · · · si1 states that Cv ̸≥
reduced, where N = ℓ(w◦ ) = |R+ |. 0, but we proved
there is some ℓ
1.134. Remark. w◦2 = 1 since w◦ takes all positive roots to all negative
such that C ℓ v ̸≥
roos and vice-versa, so applying it twice takes the positive roots back to the
0.
positive roots.
Why should a
1.135. Definition. Let γ1 := αi1 , γ2 := si1 αi2 , γ3 := si1 si2 αi3 , . . . ,
given positive
γN := si1 · · · siN −1 (αiN ).
root go back
to the same
positive root?
31
 1.136. Proposition. R+ = {γ1 , . . . , γN }. This is a special case of
R+ (W ). Why? Any intu-
ition at least?
1.137. Proposition. A Dynkin graph is bipartite, that is, there is an
indexing I = I0 ⊔ I1 such that each edge connects a vertex in I0 to a vertex
in I1 . (Note that Dynkin graphs are always trees, so this definition coincides
with the more general notion of a bipartite graph.)
1.138. Remark. Note that si , sj commute for all i, j ∈ I0 .
1.139. Definition. Define
Y Y
c0 := si , c1 := si
i∈I0 i∈I1
By definition, c1 c0 is a Coxeter element.
1.140. Proposition. Let h be the order of w◦ . Then,
w◦ = c1 c0 c1 c0 · · · = c0 c1 c0 c1 · · ·
| {z } | {z }
h factors h factors
.
Proof. The proof of this can be found somewhere in Humphreys “Re-
flection groups and Coxeter groups” section 3.17 (but where?!) by looking
at the action of c0 and c1 on a special two-dimensional real plane. □
1.141. Corollary. Given a root system R, |R| = h · r where r = |I|,
the rank of the quiver, and h is the Coxeter number.
1.142. Example. This makes sense with given types. SLn has |R| =
n(n − 1), but we computed earlier that h = n and since sln corresponds to
Dynkin graph An−1 , we get r = n − 1.

Given type Dn , we compute Do this!

1.143. Proposition. Given an orientation Ω of Dynkin |Q|, there exists
a reduced expression w◦ = siN · · · si1 such that {i1 , . . . , iN } is adapted to Ω.
Proof. Given w◦ = siN · · · si1 is adapted to Ω, then for some j, w◦ =
sj siN · · · si1 is adapted to s↓i Ω. Thus,
sj w◦ = w◦ si1 =⇒ sj = w◦ si1 w◦−1
Thus, j = −w◦ (i1 ) . So, it suffices to find one reduced word of w◦ adapted why the minus
to one distinguished Ω◦ . In fact, w◦ = · · · c0 c1 c0 is adapted to Ω◦ with I0 sign?
being the sinks and I1 being the sources. □
What about ver-
1.144. Corollary. Let Q be a Dynkin quiver. Then, a complete list of tices that are
indecomposable representations is given by neither sink nor
I1 =S(i1 ) source? Like in
A3 .
I2 =Φ↑i1 (S(i2 )) S(i2 ) ∈ Rep(s↓i1 Q)

32
 I3 =Φ↑i1 Φ↑i2 (S(i3 )) S(i3 ) ∈ Rep(s↓i2 s↓i1 Q)
..
.
In =Φ↑i1 · · · Φ↑in−1 (S(in ))
1.145. Example. Consider A3 given by
2 1 3
Then,
I1 = S(1) = 0 → k ← 0
I2 = Φ↑1 (S(2)) S(2) = k ← 0 → 0
=k→k←0
I3 = Φ↑1 Φ↑2 (S(3)) S(3) = 0 → 0 → k
= Φ↑1 (0 ← 0 → k)
=0→k←k

2. Hall Algebras
(9/14/2017) Lecture 8.
2.1. Classical Hall Algebras. This section is based primarily on ([Mac95]).
Consider the quiver

1
and take O = k[t] where k is a finite field with |k| = q. Then, we have
tk[t] = p ⊴ O. We can then constuct M to be a finite dimensional k-vector
space with nilpotent T a linear transformation. Then, we turn M into a
finite O-module via the action t 7→ T .
2.1. Remark. More generally, let Ô = k[[t]] ⊇ p is a discrete valuation
ring over the p-adic numbers Zp .
2.2. Theorem. Since O is a PID, we have, by the classification of finitely
generated modules over a PID, that
r
M
M= O/pλi
i=1
for natural numbers λ1 ≥ λ2 ≥ · · · ≥ λr . In other words
 
Jλ1 (0)
 Jλ2 (0) 
t=
 
.. 
 . 
Jλr (0)

33
 Note that we have tableau λ = (λ1 , λ2 , . . . , λr ).
2.3. Definition. We define the transpose of a tableau λ to be... , de- fill in
noted λ′ .
2.4. Proposition. Let µi = dimk (pi−1 M/pi M ). Then, µ = (µ1 , µ2 , . . .) =
λ′
Proof. Let xj be a generator of O/pλj . Then pi−1 M is generated by
those ti−1 x ̸= 0, i.e. λj ≥ i. Thus,
k λj ≥ i
(
i−1 (O/pλj )
 i−1
p (O/pλj )

′ p
λi = #{j | λj ≥ i} = =⇒ dimk = λ′i
pi (O/pλj ) 0 otherwise pi (O/pλj )
□
2.5. Definition. Call λ the type of M .
2.6. Definition. Define
X
|λ| := λi
i
to be the length of M for M of type λ. Note that |λ| is the length of the
composition series of M . We will sometimes denote this ℓ(M ) = |λ|.
2.7. Definition. Let N ≤ M . Then, the cotype of N in M is the type
of M/N .

2.8. Example. If λ = (r) = ··· , then M = O/pr is cyclic.

If λ = (1r ) = = kr as a vectorspace. This is referred to as

, then M ∼
.
.
.

an elementary module.
2.9. Definition. Let M be a module of type λ and let µ, ν be partitions.
Then, we define the Hall numbers to be
Gλµν := #{0 → N → M → M/N → 0 | N ≤ M has type ν and cotype µ}
we can generalize this definition as follows. Let µ(1) , . . . , µ(r) be a sequence
of partitions. Then
Gλµ(1) ,...,µ(r) := #{M = M0 ⊇ M1 ⊇ · · · ⊇ Mr = 0 | Mi−1 /Mi has type µ(i) }
2.10. Definition. The Hall algebra H is a free Z-algebra with basis uλ
where λ is a partition such that, for partitions µ, ν,
X
uµ uν = Gλµν uλ
λ

34
 2.11. Proposition. Gλµν = 0 unless |λ| = |µ| + |ν|. Thus, the sum
defined in the Hall algebra multiplication is a finite sum.

Proof. To have a short exact sequence

0 → N → M → M/N → 0
with N having type ν and cotype µ and M having type λ, it is Lnecessary
for |µ| + |ν| = |λ|. This follows from the classification of M = O/pλi as
a module over PID O. □
2.12. Proposition. A Hall algebra H is commutative (that is, Gλµν =
Gλνµ ) and associative with 1 = u∅ .

Proof. The commutativity follows from the duality (given M, M̂ of

type λ), Where does this
1-1 duality come
{N | N ≤ M, N has type ν and cotype µ} ←→ {N̂ | N̂ ≤ M̂ | N̂ has type µ and cotype
from? ν}
Trans-
The associativity follows from the fact that the coefficient of uλ in (uµ uν )uρ position of the
and in uµ (uν uρ ) will be Gλµνρ . □ JCF type argu-
ment?
2.13. Proposition. As a Z-algebra, the Hall algebra H is generated by
u(1r ) , r ≥ 1 algebraically independently.

Proof. Denote vr = u(1r ) . Then, for λ′ = (λ′1 , . . . , λ′r ),

X
vλ′ := vλ′1 vλ′2 · · · vλ′s = aλµ uµ
µ

where aλµ = #{M = M0 ⊇ M1 ⊇ · · · ⊇ Ms = 0 | M fixed type µ, type of Mi−1 /Mi =

(1λi )}. Now, if aλµ ̸= 0,
=⇒ Such (Mi ) exists with p(Mi−1 /Mi ) = 0
=⇒ p(Mi−1 ) ⊆ Mi , ∀i
=⇒ pi M ⊆ Mi , ∀i
=⇒ µ′1 + · · · + µ′i = ℓ(M/pi M ) ≥ ℓ(M/Mi ) = λ′1 + · · · + λ′i
=⇒ λ′ ⊴ µ′
=⇒ µ ⊴ λ
So, if µ = λ, then there exists only one filtration with Mi = pi M and
so aλλ = 1. Thus, (aλµ ) is strictly upper unitriangular, with aλµ ∈ Z≥0 .
Therefore, it is invertible over Z and can be solved, thus giving uµ as a
Z-linear combination of vλ′ . □
2.14. Corollary. As rings, H ∼ = Λ, the ring of symmetric function in
infinitely many variables via the isomorphism
r(r−1)
u(1r ) 7→ q − 2 er

35
 2.15. Definition. Let λ be a partition. Then, we define
X X λ′ 
i
n(λ) := (i − 1)λi =
2
i i

2.16. Example. These equalities follow from just summing the entries
of a partition over rows versus over colums. For example, take
0 0 0 0
λ= 1 1
2
Then, n(λ) = 0 ∗ 4 + 1 ∗ 2 + 2 ∗ 1 = (0 + 1 + 2) + (0 + 1) + (0) + (0) = 4
Now, our goal is to understand these structure constants Gλµν . As a
reminder, we are still working over field k with |k| = q.
2.17. Theorem (Steinitz). λ (t) ∈
(a) There exists a polynomial gµν
Z[t] such that
Gλµν = gµν
λ
(q)
(b) The degree of gµν λ is less than or equal to n(λ) − n(µ) − n(ν). Fur-

thermore, the coefficient of tn(λ)−n(µ)−n(ν) = cλµν , the Littlewood-

Richardson coefficients.
λ (t) = g λ (t)
(c) gµν νµ
λ = 0.
(d) If cλµν = 0, then gµν
P
Recall that vλ′i = µ aλµ uµ .
2.18. Proposition. (a) There exists a combinatorial formula
X
aλµ = q d(A) ∈ Z[q]
A diagrams
What kind of di-
(b) deg aλµ (t) ≤ n(µ) − n(λ) agrams? Young
(c) ãλµ (t) = tn(µ)−n(λ) aλµ (t−1 ) satisfies ã(0) = Kµ′ λ′ , the Kostka num- diagrams?
ber.
(d) aλµ (t) = 0 unless µ ⊴ λ and aλλ = 1. Where does this
t come from?
Note that part (a) above implies the following theorem
We already
2.19. Theorem. (aλµ (t)) is a unitriangular matrix over Z[t] and is thus showed this
invertible. part, or do we
2.20. Definition. Given two partitions λ and µ, we define the union mean aλλ (t) =
λ ∪ µ to be a partition of |λ| + |µ| where each row is a row of λ or a row of 1?
µ.
2.21. Lemma. vλ′ vρ′ = vλ′ ∪µ′ .
Proof. This follows from the commutativity of our algebra. □

36
 λ ∈ Z[t].
2.22. Proposition. gµν
P
Proof. Given that (aλµ (t)) is inverticle, this means that uµ , uν ∈ λ Z[t]vλ′ .
Thus, the product uµ uν is also a Z-linear combination of vλ′ ’s, so gµν λ ∈

Z[t]. □
2.23. Definition. We define Pµ (x, ; t) to be the polynomials such that,
for eλ′ ∈ ΛZ[t] ,
X
e λ′ = ãλµ (t)Pµ (x; t)
µ Finish this lec-
ture. The last
(9/19/2017) Lecture 9. page is hard to
2.2. Generic Hall Algebras. Let Q be a quiver. Then, we can make follow.
sense of the category Rep(Q, Fq ).
2.24. Definition. Let M1 , M2 , L ∈ Rep(Q, Fq ). Define
FM L
1 ,M2
:= {Subrepresentation X ⊆ L | X ∼
= M2 , L/X ∼
= M1 }
that is
f g
L
FM 1 ,M2
= {0 → X → L → L/X → 0 | X ∼
= M2 , L/X ∼
= M1 } ∼
= {(f, g)}/ AutQ (M2 )×AutQ (M1 )
Then, we have generalization for M1 , . . . , Mk , L ∈ Rep(Q, Fq ) given by
FL := {L = L0 ⊇ L1 ⊇ · · · ⊇ Lk = 0 | Li−1 /Li ∼
M1 ,...,Mk = Mi , ∀i}
2.25. Definition. Define the structure constants
L L
FM 1 ,M2
:= |FM 1 ,M2
|
similarly,
L L
FM 1 ,...,Mk
:= |FM 1 ,...,Mk
|

Thus, we have defined a generalization of the structure constants Gλµν

for the classical Hall algebra and the program of this lecture will be to prove
analogous theorems to those that we have for classical Hall algebras.
ℓ
2.26. Proposition. FM = 0 unless dimM1 + · · · + dimMk = L.
1 ,...,Mk

2.27. Definition. We define the Hall algebra H(Q, Fq ) to be the Z-

algebra with basis given by [M ], the isomorphism classes in Rep(Q) such
that X
L
[M1 ] ∗ [M2 ] = FM 1 ,M2
[L]
[L]

2.28. Proposition. H(Q, Fq ) is an associative algebra with 1 = [0].

Moreover, it is NI -graded, that is
M
H(Q, Fq ) = H⃗v (Q, Fq )
⃗v ∈NI
where H⃗v (Q, Fq ) is spanned by [M ] with dimM = ⃗v .

37
 Proof. To see associativity, note that
L
([M1 ] ∗ [M2 ]) ∗ [M3 ] = FM 1 ,M2 ,M3
= [M1 ] ∗ ([M2 ] ∗ [M3 ])
□
2.29. Remark. There is also a twisted product, given by
⟨⃗ v⟩
u,⃗
U ∗V
U ·V =q 2

for U ∈ H⃗u (Q) and V ∈ H⃗v (Q) where ⟨, ⟩ is the Euler form. This product
is also associative with [0] = 1.
2.30. Definition. Let n be a natural number. Then, we define
qn − 1
[n]q = , [n]q ! = [n]q [n − 1]q · · · [2]q [1]q
q−1
and  
m [m]q !
=
n q [n]q ![m − n]q !
2.31. Example. Take Q = and let S be the 1-dimensional representa-
tion. Then, all representations are of the form nS := S ⊕n , n ≥ 0. We then
know that
[(n − 1)S] ∗ [S] = α[nS]
for some α. To figure out what α is, we must count the number of submod-
n −1
ules of nS, which one can quickly check is equivalent to |Pn−1 (Fq )| = qq−1 .
Thus,
qn − 1
[(n − 1)S] ∗ [S] = [nS]
q−1
In fact, we can show the following proposition.
2.32. Proposition. For Q = and S as in the above example,
 
m+n
[nS] ∗ [mS] = [(n + m)S]
m q
for natural numbers m, n.
Proof. There are two proof methods for this proposition, either by
induction or counting directly. We will go by the latter route. Note that
#{m-dimensional subspaces of Fm+n
q } = # Gr(m, n + m)
However, since GL(n + m, Fq ) acts transitively on Gr(m, nm ), then
Gr(m, n + m) = GL(n + m, Fq )/Pn,m
for some block matrix Pn,m of the form
 
∗ ∗
Pn,m =
0 ∗
It is a common abstract algebra fact that
|GLn (q)| = (q n − 1)(q n − q) · · · (q n − q n−1 )

38
 n (q n − 1) · · · (q − 1)
= q ( 2 ) (q − 1)n
(q − 1) · · · (q − 1)
n
= q ( 2 ) (q − 1)n [n]q !
So,
m+n
q( ) (q − 1)m+n [n + m] !
2
q
#GL(m + n, Fq )/Pn,m = n m
( ) ( )
q (q − 1)n [n]q !q 2 (q − 1)m [m]q !q mn
2

m+n
q ( 2 ) [n + m] ! q
= n m
q ( 2 ) [n]q !q ( 2 ) [m]q !q mn
□
2.33. Remark. This gives rise toa “generic” Hall algebra over Z[t], H( )
where q is replaced by t.
2.34. Proposition. For H( ),
[S]n
[S]n = [n]q ![nS] =⇒ [nS] =
[n]q !
This suggests that the Hall algebra is isomorphic to the algebra C[x] of poly-
nomials in one variable with isomorphism given by
xn
[nS] 7→
[n]q !
2.35. Definition. Define
xn
x(n) :=
[n]t !
2.36. Proposition. H( ) ∼
= Z[t][x, x(2) , x(3) , . . .] and over Q(t),
H( ) ⊗ Q(t) ∼
= Q(t)[x]
Understand
where this is
2.37. Proposition. Given a general quiver Q, if S ∈ Rep(Q, Fq ) such coming from.
that
Ext1Q (S, S) = 0 HomQ (S, S) = Fq
[S]n
Then, [nS] = [n]q ! .

2.38. Example. Simply take S = S(i).

2.39. Remark. In general, [M1 ] ∗ [M2 ] ̸= [M1 ⊕ M2 ].
2.40. Proposition. Assume HomQ (M2 , M1 ) = 0 = Ext1Q (M1 , M2 ).
Then,
[M1 ] ∗ [M2 ] = [M1 ⊕ M2 ]

39
 Proof. Consider a short exact sequence
0 → M2 → L → M1 → 0
Then, since Ext1 (M1 , M2 ) = 0, L = M1 ⊕ M2 . Thus,
[M1 ] ∗ [M2 ] = α[M1 ⊕ M2 ]
for some α. Let N ⊆ M1 ⊕M2 be isomorphic to M2 but N ̸= {(0, m2 ) | m2 ∈
M2 }, that is, let N not be the standard embedding of M2 into M1 ⊕ M2 .
Then, we have
ι
N M1 ⊕ M 2
̸=0
π1
M1
However, this contradicts HomQ (M2 , M1 ) = 0, so the only possible embed-
ding for N is N = {(0, m2 ) | m2 ∈ M2 } and thus α = 1. □

From now on, assume Q is Dynkin. Recall that we then have

1-1
{Indecomposables} ←→ R+ (Positive roots)
Iα →7 α
Isomorphism classes in Rep(Q) ↔ NR+ := {⃗n : R+ → N}
M
Mn = nα Iα →7 ⃗n = (nα )α∈R+
α∈R+

⃗ ⃗k ∈ NR+ ,
2.41. Proposition. Let Q be Dynkin. Then, for all ⃗n, m,
m
⃗
there is a polynomial ϕ ⃗ ∈ Z[t] such that
⃗
n,k
Mm
ϕ⃗m
⃗
n,⃗k
(q) = FM⃗
⃗
n ,M⃗ k

This gives rise to the “generic” Hall algebra Ht (Q) = H(Q)Z[t] with
multiplication
X
[M⃗n ] ∗ [M⃗k ] = ϕ⃗m
⃗
n,⃗k
(t)Mm
⃗
m
⃗
We can also make use of specialization to get
H(Q)Z[t] |t=1 =: H1 (Q) is a Z-algebra
and also
H(Q)C = H(Q)Z[t] ⊗Z[t] C

2.42. Theorem. Let à = Z[t][[n]−1t , ∀n ≥ 1]. Then, the (Ringel-)Hall

algebra H(Q)Ã is generated by θi = [S(i)] for i ∈ I.

This theorem will be expanded upon in the next lecture.

40
 (9/21/2017) Lecture 10. Recall that, by 1.143, if Q is Dynkin, there
exists a reduced w◦ = sin · · · si1 such that {i1 , . . . , iN } is adapted to Q.
Thus, we have positive roots R+ = {γ1 , . . . , γN } where
γ1 = α1 , γ2 = si1 (α2 ), . . . , γN = si1 · · · siN −1 (αN )
and Gabriel’s Theorem 1.86 says that if Q is Dynkin, there is a one-to-
one correspondance with the indecomposable representations of Q given by
γi 7→ Ii . Also, recall that, in this setting, How exactly
HomQ (Ia , Ib ) = δa,b for a ≥ b does this fol-
low?
Ext1Q (Ib , Ia ) = 0

2.43. Proposition. For Q Dynkin, if V ∼

LN
= k=1 nk Ik , then
[V ] = [I1 ](n1 ) ∗ · · · ∗ [IN ](nN ) in Ht (Q)
Proof. It suffices to check in H(Q, Fq ) . By, 2.40 and above, we have why does it suf-
" D # fice to check
M
nk Ik = [n1 I1 ] ∗ [n2 I2 ] ∗ · · · ∗ [nD ID ] only this?
k=1
□
Proof of 2.42. We define a partial order on NI by
⃗v ≤ w
⃗ ⇐⇒ vi ≤ wi
This order allows us to induce a partial order on the isomorphism classes in
Rep(Q) by
V ≤ W ⇐⇒ dimV ≤ dimW or dimV = dimW and OV ⊆ OW
Let H ′ := Ã⟨Oi | i ∈ I⟩. Then, we proceed by induction on ≤ to show
[V ] ∈ H ′ . Assume for all V ′ < V , [V ′ ] ∈ H ′ . We have two cases:
(i) V is decomposable, that is V = V ′ ⊕ V ′′ . By the proposition above,
[V ] = [V ′ ]∗[V ′′ ], both of which are in H ′ by the inductive hypothesis
since dimV ′ < dimV and dimV ′′ < dimV . Thus, V ∈ H ′ .
(ii) V is indecomposable. Then, find an i such that Vi ̸= 0 and, for all
edges i → j, Vj = 0. Let vi = dim Vi . Then, we have short exact
sequence
0 → vi S(i) → V → V /(vi S(i)) → 0
So, [V /(vi S(i))] ∗ [V ′ ] = [V ] + k ck [Vk′′ ], where Vk′′ are other repre-
P
sentations with dimVk′′ = dimV . However, since V is a unique in-
decomposable with dimV by the one-to-one correspondance, the Vk′′
ck [Vk ] ∈ H ′ and [V /(vi S(i))] ∗
′′
P
must be decomposable and thus
[V ′ ] ∈ H ′ give us that [V ] ∈ H ′ .
□
2.44. Corollary. Over C, {θi | i ∈ I} generates H1 (Q)C .

41
 We now move on to a discussion of the relations among the θi ’s.
2.45. Proposition. Let Q be a Dynkin quiver. If i and j are not con-
nected, then in the Hall algebra
θi ∗ θj = [S(i) ⊕ S(j)] = θj ∗ θi
2.46. Example. Let us do some computations with the quiver
Q= 1 2
with k = Fq . Then, we have representations
S(1) = k → 0
S(2) = 0 → k
S12 := k → k
∼

S̃12 := k → k
0

Then,
θ1 ∗ θ2 = [S12 ] + [S̃1 2]
θ2 ∗ θ1 = [S̃12 ]
since θ2 is a submodule of S12 and S̃12 with multiplicity 1 and θ1 is a sub-
module of S̃12 with multiplicity 1.
k ∼
k k 0
k k 0
k
θ2 ,→ S12 = θ2 ,→ S̃12 = θ1 ,→ S̃12 =
k 0 k 0 0 k
Note that there is no way to embed θ1 ,→ S12 and have the diagram com-
mute. Now, we wish to compute θ12 ∗θ2 , θ1 ∗θ2 ∗θ1 , and θ2 ∗θ12 . To get the right
quotients, we must first write down representations of Q of dim = (2, 1):

S112 = k2 k S̃112 = k2 0
k
Then, we have
θ12 ∗ θ2 = (q + 1)[2S(1)] ∗ [S(2)] = (q + 1)([S112 ] + [S̃112 ])
θ2 ∗ θ12 = (q + 1)[S(2)] ∗ [2S(1)] = (q + 1)[S̃112 ]
θ1 ∗ θ2 ∗ θ1 = θ1 ∗ [S̃12 ] = [S112 ] + (q + 1)[S̃112 ]
To see the first equality, we compute
k2 0
θ1 ,→ 2S(1) =

k 0
and see that there are q + 1 possible maps k → k2 that make the diagram
commute, since the map could be encoded by a rank 1 2 × 1 matrix with

42
 q 2 −1
entries in Fq , up to non-zero scalar multiplication by Fq , thus q−1 = q + 1.
Then, we see

k2 k k2 0
k k2 0
k
S(2) ,→ S112 = S(2) ,→ S̃112 = 2S(1) ,→ S̃112 =

0 k 0 k k2 0
each only have 1 embedding and
k2 k k2 0
k
S̃12 ,→ S112 = S̃12 ,→ S̃112 =

k 0
k k ∼
k
q 2 −1
have 1 and q−1 = q + 1 embeddings, respectively.

From this example, we can get the following proposition

2.47. Proposition. Let i → j in a quiver Q. Then, in H(Q)Z[t]
θi2 ∗ θj − (t + 1)θi ∗ θj ∗ θi + tθ2 ∗ θi2 = 0

Proof. The proof is the same as the computations in the exercise above.
An exercise for the reader is to find a similar relationship between θ22 ∗θ1 , θ2 ∗
θ1 ∗ θ2 , and θ1 ∗ θ22 . □
2.48. Proposition. Recall the twisted product for the Hall algebra given
in 2.29. Given Q = 1 → 2, then in H(Q, Fq ), we get
1
θ1 · θ2 = q − 2 θ1 ∗ θ 2
θ2 · θ1 = θ2 ∗ θ 1
1
θ12 · θ2 = q − 2 θ1∗2 ∗ θ2
1
θ2 · θ12 = q 2 θ2 ∗ θ1∗2
θ1 · θ2 · θ1 = θ1 ∗ θ 2 ∗ θ1

Proof. These computations follow from the computations that

⟨S(i), S(i)⟩ = 1, ⟨S(1), S(2)⟩ = −1, ⟨S(2), S(1)⟩ = 0
P P
where ⟨V, W ⟩ = i vi wi − h vs(h) wt(h) is the Euler form. □
2.49. Remark. From now on, we may choose to omit the · in the Hall
algebra, and simply write
[V ][W ] := [V ] · [W ]
1
2.50. Corollary. For Q = 1 → 2 and v = t 2 , in H(Q)Z[t] , we get
(a) θ12 θ2 − (v + v −1 )θ1 θ2 θ1 + θ2 θ12 = 0
(b) θ22 θ1 − (v + v −1 )θ2 θ1 θ2 + θ1 θ22 = 0

43
 Proof. For part (a),
1 1 1 1
θ12 θ2 − (v + v −1 )θ1 θ2 θ1 + θ2 θ12 = t− 2 θ1∗2 ∗ θ2 − (t 2 + t− 2 )θ1 ∗ θ2 ∗ θ1 + t 2 θ2 ∗ θ1∗2
1
= t− 2 (θ1∗2 ∗ θ2 − (t + 1)θ1 ∗ θ2 ∗ θ1 + tθ2 ∗ θ1∗2 )
1
= t− 2 ∗ (0) by 2.47
=0
□
2.51. Theorem (Ringel’s Theorem). Let Q be a Dynkin quiver. Let
H̃(Q, v 2 ) = Ht (Q) ⊗Z[t] Q(v) (t 7→ v 2 )
be equipped with the twisted product. Then, there exists an algebra isomor-
phism
ψ : Uv n+ → H̃(Q, v 2 ), Ei 7→ θi
where
E E =E E for i̸→j,
Uv n+ = Q(v)⟨Ei (i ∈ I) | E 2 Ej −[2]vi Eji Ej Ej i +E
i
2
j E =0 for i→j
⟩
i i

2.52. Corollary. There is an algebra isomorphism

∼
U n+ → Ht (Q) ⊗Z[t] C
sending ei → θi .
Get a handle
on the quan-
Proof of Theorem. We must check the following tum groups vs
• ψ is a homomorphism. This follows from 2.50. enveloping alge-
• ψ is onto. bras here. When
• ψ respects the grading (Uv n+ )d⃗ → H̃d⃗(Q, v 2 )d∈N
⃗ ± is Ei appropri-
We also check ate and when is
n X o ei appropriate?
dim H̃d⃗(Q, v 2 ) = (nα )α∈R+ | nα α = d⃗ = dim U n+
by the PBW theorem. □

44
 Bibliography

[CG00] N. Chriss and V. Ginzburg, Representation Theory and Complex Geometry,

2000.
[DDPW08] B. Deng, J. Du, B. Parshall, and J. Wang, Finite Dimensional Algebras and
Quantum Groups, 2008.
[Hum72] J. E. Humphreys, Introduction to Lie Algebras and Representaton Theory,
1972. Third printing, revised.
[Hum90] , Reflection Groups and Coxeter Groups, 1990.
[Jan96] J. C. Jantzen, Lectures on Quantum Groups, 1996.
[KJ16] A. Kirillov Jr., Quiver Representations and Quiver Varieties, 2016.
[Mac95] I. G. Macdonald, Symmetric Functions and Hall Polynomials, 1995. Second
edition.

45