A Hopf Algebra Structure on Hall Algebras

DRAFT VERSION

Christopher D. Walker
Department of Mathematics, University of California
Riverside, CA 92521 USA
September 26, 2010

Abstract
One problematic feature of Hall algebras is the fact that the standard
multiplication and comultiplication maps do not satisfy the bialgebra com-
patibility condition in the underlying symmetric monoidal category Vect.
In the past this problem has been resolved by working with a weaker
structure called a ‘twisted’ bialgebra. In this paper we solve the problem
differently by first switching to a different underlying category VectK of
vector spaces graded by a group K called the Grothendieck group. We
equip this category with a nontrivial braiding which depends on the K-
grading. With this braiding, we find that the Hall algebra does satisfy the
bialgebra condition exactly for the standard multiplication and comulti-
plication, and will also become a Hopf algebra object.

1 Introduction
Hall algebras have been a popular topic in recent years because of their con-
nection to quantum groups. It is a well known fact, due to Ringel [12], that
the Hall algebra constructed from the representations of a Dynkin quiver over
a finite field Fq is isomorphic to ‘half’ of the quantum group associated to the
same Dynkin diagram, namely Uq+ (g) where g is the lie algebra generated by
the quiver. This construction provides interesting insight into many structures
on the quantum group, but unfortunately does not do everything we hope.
One of the fundamental problems of Hall algebras arises when we try to make
the algebra into a Hopf algebra. In the initial definitions of the Hall algebra, we
start with a nice associative multiplication. We also find that the Hall algebra is
a coalgebra with an equally nice coassociative comultiplication. However, when
we try to check that the algebra and coalgebra fit together to form a bialgebra,
we see this fails in the standard underlying category Vect with its usual braiding.
Instead, the combination of these maps obeys “Green’s Formula”, a relationship
between the multiplication and comultiplication which we describe in detail later
in the paper (Proposition 8). This formula basically says that the Hall algebra

1
is ‘almost’ a bialgebra in the standard category Vect. Specifically, we only miss
being a bialgebra by a coefficient. To see where this extra coefficient comes from,
consider the string diagrams which describe the general bialgebra compatibility
axiom. As is standard, we will write multiplication of two elements as:



and comultiplication of an element as:



We can then draw the bialgebra axiom as follows. We first to draw multiplica-
tion, follow by comultiplication, which looks like:





This should equal the result of comultiplying each element and then multiplying
the resulting tensor product of elements. This will look like:
 
 

 

But this diagram has a wrinkle in it, namely the braiding of the strings halfway
down the diagram. This means we must be working in a braided monoidal
category. For the Hall algebra, the seemingly natural choice to work in would
be Vect. In Vect this braiding would simply swap elements with no coefficient.
However we have already noted that in Vect the Hall algebra does not satisfy
the bialgebra condition as desired.
To ‘fix’ this, a new structure is usually defined called a ‘twisted’ bialgebra,
where swapping the order of elements can still be done, but at the price of an
extra coefficient. This coefficient becomes q −hA,Di when swapping elements A
and D, where hA, Di is a bilinear form on a group K (called the Grothendieck
group) related to the underlying category of the Hall algebra.
The process that is then used to obtain an true (untwisted) bialgebra is to
extend the Hall algebra to some larger algebra. This process is interesting in it
own right, because the result is isomorphic to a larger piece of a quantum group,
namely the universal enveloping algebra of the Borel b. However, we want to
take a different direction to avoid the artificial nature of this fix.
In this paper, we will approach the problem directly. Instead of describing
the Hall algebra as a ‘twisted’ bialgebra, we will find a braided monoidal cat-
egory other than Vect where the Hall algebra is a true bialgebra object. We

2
accomplish this giving the category of K-graded vector spaces, VectK , a braid-
ing that encodes the twisting in the Hall algebra. This works since the extra
coefficient q −hA,Di from Green’s Formula depends on the crossing strands in the
bialgebra diagram. We then accomplishes the same task as before, but in a
way that accounts for the correction factor in the underlying structure, rather
than including it later. We will then round out the paper by also providing
the antipode for this bialgebra to show that the Hall algebra is a Hopf algebra
object in our new category.

2 Hall Algebras
In this section we will describe the construction of the Ringel-Hall algebra. We
begin with a quiver Q (i.e. a directed graph) whose underlying graph is that
of a simply-laced Dynkin diagram. We will then consider the abelian category
Rep(Q) of all finite dimensional representation of the quiver Q over a fixed finite
field Fq .
We start by fixing a finite field Fq and a directed graph D, which might look
like this:
•
~?
~~
: •` • /•
> @@
@
•
We shall call the category Q freely generated by D a quiver. The objects of Q
are the vertices of D, while the morphisms are edge paths, with paths of length
zero serving as identity morphisms.
By a representation of the quiver Q we mean a functor

R : Q → FinVectq ,

where FinVectq is the category of finite-dimensional vector spaces over Fq . Such

a representation simply assigns a vector space R(d) ∈ FinVectq to each vertex
of D and a linear operator R(e) : R(d) → R(d0 ) to each edge e from d to d0 . By
a morphism between representations of Q we mean a natural transformation
between such functors. So, a morphism α : R → S assigns a linear operator
αd : R(d) → S(d) to each vertex d of D, in such a way that

R(e)
R(d) / R(d0 )

αd αd0
 
S(d) / S(d0 )
S(d)

commutes for any edge e from d to d0 . There is a category Rep(Q) where

the objects are representations of Q and the morphisms are as above. This is

3
an abelian category, so we can speak of indecomposable objects, short exact
sequences, etc. in this category.
In 1972, Gabriel [6] discovered a remarkable fact. Namely: a quiver has
finitely many isomorphism classes of indecomposable representations if and only
if its underlying graph, ignoring the orientation of edges, is a finite disjoint union
of Dynkin diagrams of type A, D or E. These are called simply laced Dynkin
diagrams.
Henceforth, for simplicity, we assume the underlying graph of our quiver
Q is a simply laced Dynkin diagram when we ignore the orientations of its
edges. Let X be the underlying groupoid of Rep(Q): that is, the groupoid with
representations of Q as objects and isomorphisms between these as morphisms.
We will use this groupoid to construct the Hall algebra of Q.
As a vector space, the Hall algebra is just R[X]. Recall that this is the vector
space whose basis consists of isomorphism classes of objects in X. In fancier
language, it is the zeroth homology of X.
We now focus our attention on the Hall algebra product. Given three quiver
representations M, N, and E, we define set:
E f g
PMN = {(f, g) : 0 → N → E → M → 0 is exact}.
E
and we call its set cardinality PMN The Hall algebra product counts these exact
sequences, but with a subtle ‘correction factor’:
X E
PMN
[M ] · [N ] = [E] .
aut(M ) aut(N )
[E]∈X

Where we call aut(M ) the set cardinality of the group Aut(M )

Somewhat surprisingly, the above product is associative. In fact, Ringel [12]
showed that the resulting algebra is isomorphic to the positive part Uq+ g of the
quantum group corresponding to our simply laced Dynkin diagram! So, roughly
speaking, the Hall algebra of a simply laced quiver is ‘half of a quantum group’.
This isomorphism also relates to a coalgebra structure on the Hall algebra.
Using the same ideas from the multiplication formula, we can define a comulti-
plication on the Hall algebra to be a carefully weighted sum on ways to ‘factor’
a representation via short exact sequences. Formulaically this becomes:
X E
|PMN |
∆(E) = [N ] ⊗ [M ] .
aut(E)
[M],[N ]∈X

Again, Ringel found that these are the correct factor to make the comultipli-
cation coassociative. However, we immediately run into a problem; these two
maps do not satisfy the compatibility condition for a bialgebra.

2.1 Twisted Tensor Multiplication and Compatibility

It is interesting to note that the standard multiplication and comultiplication on
Uq+ g (Which the Hall algebra is isomorphic to) do not satisfy the compatibility

4
axiom of a bialgebra, so we should not expect the Hall algebra to either. This
does not mean there is not an interesting relationship between the multiplication
and comultiplication in the Hall algebra. This relationship is often described as
being a ‘twisted’ bialgebra, where we do not use the standard extension of the
multiplication to the tensor product. We would like to take a different point of
view here. It turns out that the bialgebra axiom can be satisfied if we change
the category in which we ask for them to be compatible.
In order to describe this new category, we will start with a definition of the
Grothendieck group of a general abelian category.
Definition 1. Let A be an abelian category. We can define an equivalence
relation on isomorphism classes of object in A by [A] + [B] = [C] if there exist
a short exact sequence 0 → A → C → B → 0. The set of equivalence classes
under this relation form a group K0 (A) called the Grothendieck group.
K0 (A) has a universal property in the following sense. Given any abelian
group G, any additive function f from isomorphism classes of A to the group G
will give a unique abelian group homomorphism f˜: K0 (A) → G such that the
following diagram commutes:

A? / K0 (A)
?? x x
?? xx
?
f ?? x
x ˜
 |xx ∃!f
G
The original purpose of the Grothendieck group was to study Euler charac-
teristics, and this is precisely why we are interested in them here.
In many of the standard references for Hall algebras [7, 15] the characteristics
of the Grothendieck group of Rep(Q) are explained explicitly. Many of these
characteristics follow from the fact that Rep(Q) is hereditary, which is a special
case of the abelian category having finite homological dimension. In the general
case of an abelian category A which has finite homological dimension, we can
describe much of the same work. First, we show that the Grothendieck group
admits a well defined bilinear form which comes from the Euler characteristic.
Proposition 2. If K = K0 (A) is the Grothendieck group of an abelian category
A with finite homological dimension d, then K admits a bilinear form h·, ·i : K ×
K → C given by:
Xd
hm, ni = (−1)i dim Exti (M, N )
i=0

Proof. We will show the theorem for d = 1 (i.e. the category is hereditary)
since this is the main case we will use. The case when d = 0 is simply bilin-
earity of Hom, and the cases where d > 1 follows by a similar argument to d = 1.

For d = 1 we need to show that:

dim Hom(M, N1 ⊕ N2 ) − dim Ext1 (M, N1 ⊕ N2 ) =

5
 dim Hom(M, N1 ) − dim Ext1 (M, N1 ) + dim Hom(M, N2 ) − dim Ext1 (M, N2 )
we begin with the short exact sequence:
1 i π
0 → N1 → N1 ⊕ N2 →2 N2 → 0

which, since d = 1, gives rise to the long exact sequence:

h
0 → Hom(M, N1 ) → Hom(M, N1 ⊕ N2 ) → Hom(M, N2 ) →

Ext1 (M, N1 ) → Ext1 (M, N1 ⊕ N2 ) → Ext1 (M, N2 ) → 0

Using a variety of basic equations from the fact that this sequence is exact, as
well as some dimension arguments, The left hand sides becomes:

dim Hom(M, N1 ⊕ N2 ) − dim Ext1 (M, N1 ⊕ N2 )

= dim imπ˜2 + dim ker π˜2 − dim imπˆ2 − dim ker πˆ2
and the right hand side turns into:

dim Hom(M, N1 ) − dim Ext1 (M, N1 ) + dim Hom(M, N2 ) − dim Ext1 (M, N2 )

= dim imh + dim ker h + dim imi˜1 − dim imiˆ1 − dim ker iˆ1 − dim imπˆ2
= dim ker iˆ1 + dim imπ˜2 + dim ker π˜2 − dim ker πˆ2 − dim ker iˆ1 − dim imπˆ2
= dim imπ˜2 + dim ker π˜2 − dim ker πˆ2 − dim imπˆ2
And thus LHS=RHS
In general, it is possible to construct a braided monoidal category VectG from
any abelian group G equipped with a bilinear form h·, ·i. One common example
is the category of super-algebras, which can be thought of in this context in
terms of the group Z2 with its unique non-trivial bilinear form. In the next
Theorem, we will describe how this braiding works in detail for our desired case
of the Grothendieck group K = K0 (A) with the previously described bilinear
form.
Theorem 3. Let A be an abelian, k-linear category with finite homological
dimension. Also, let K = K0 (A) be its Grothendieck group. The category VectK
of K-graded vector spaces and grade preserving linear operators is a braided
monoidal category, with the braiding given by:

BV,W : V ⊗ W → W ⊗ V

v ⊗ w 7→ q −hn,mi w ⊗ v

6
Proof. The monoidal structure on this category is simply just the tensor product
in the category Vect. To define a braiding on this category, we first note that the
braiding is define by isomorphisms in the category, which are grade preserving
linear operators. Because of linearity, it is enough to define these isomorphisms
on a single graded piece. Also note that for any two K-graded vector spaces V
and W , a graded piece of the tensor product V ⊗ W can be written as the sum
of tensor products of graded pieces from V and W , or more precisely:
M
(V ⊗ W )d = Vn ⊗ Wd−n .
n∈K

This lets us define the braiding BV,W : V ⊗ W → W ⊗ V only on the tensor

product of the graded piece Vn ⊗ Wm . We thus define the map:

Bn,m : Vn ⊗ Wm → Wm ⊗ Vn

v ⊗ w 7→ q −hn,mi w ⊗ v
which is easily seen to be an isomorphism. We only need to check the hexagon
equations, i.e. ones of the form:
α / W ⊗ (V ⊗ U )
(W ⊗ V ) ⊗ U
BV,W ⊗1mmmm
m6 QQQ
QQ1⊗B
m QQQV,U
mmm QQQ
mmm Q(
(V ⊗ W ) ⊗ U W ⊗ (U ⊗ V )
QQQ m6
QQQ
QQ mmmmm
α QQQ mmmα
Q( mmm
V ⊗ (W ⊗ U ) / (W ⊗ U ) ⊗ V
BV,W ⊗U

We will make the argument for the above hexagon identity, noting the the other
versions follow by a similar argument. Now, since we have restricted ourselves
to vector spaces with a single grade, it is enough to chase a general element
around this diagram. let v ∈ Vn , w ∈ Wm , and u ∈ Up . The top path of the
hexagon diagram yields the composite:

(v ⊗ w) ⊗ u 7→ q −hn,mi−hn,pi w ⊗ (u ⊗ v).

For the bottom path, we note that v ⊗w ∈ (V ⊗W )m+p , so we get the composite
:
(v ⊗ w) ⊗ u 7→ q −hn,m+pi w ⊗ (u ⊗ v).
Hence, commutativity of the diagram will follow from the equality −hm, ni −
hm, pi = −hm, n + pi, which is precisely bilinearity of the form h·, ·i.
Now we will focus back on the specific abelian category Rep(Q) and the
category of K0 (Rep(Q))−graded vector spaces. We can now state the main
theorem of this paper.

7
Theorem 4. For a Dynkin quiver Q, the Hall algebra of Rep(Q) is a Hopf
algebra object in the category V ectK .
To prove this theorem we need to work through the following lemmas. For
what follows, we will set X to be the underlying groupoid of Rep(Q), X to be
the set of isomorphism classes in X, and K = K0 (Rep(Q)). Recall that R[X]
is the vector space of all finite linear combinations of elements of X.
Lemma 5. The vector space H = R[X] is a K-graded vector space, with the
grading on each isomorphism class [M ] ∈ X given by its image in K.
For the next two lemmas, we note that the multiplication and comultipli-
cation described were shown to be associative and coassociative in the original
category Vect by Ringel. This fact passes to our new category since neither
axiom requires or depends on the particular braiding on vector spaces, so we
will not repeat the argument. After stating both lemmas, we will provide a brief
description of why each on is a morphism in the new category VectK .
Lemma 6. The multiplication map m : H ⊗ H → H defined on basis elements
by:
X E
PMN
m([M ] ⊗ [N ]) = [E]
aut(M ) aut(N )
[E]
K
is a morphism in Vect .
Lemma 7. The comultiplication map ∆ : H → H ⊗ H defined on basis elements
by:
X E
PMN
∆(E) = [N ] ⊗ [M ]
aut(E)
[M],[N ]
K
is a morphism in Vect .
Both of these lemmas are true for a similar reason. The important fact to
note here is that for a fixed M , N , and E in either sum, there is a short exact
sequence 0 → N → E → M → 0. So by the definition of the Grothendieck
group K, we have that their images hold to the identity [M ] + [N ] = [E]. These
images determine the grade of the corresponding graded piece they sit in, so the
grade is clearly preserved by both maps.
Now we can focus on the compatibility of the new maps, which was the main
reason for constructing this new category. We first need an important identity
for the multiplication and comultiplication known as Green’s Formula.
Proposition 8. (Green’s Formula). For all M , N , X, and Y in Rep(Q) we
have the identity:
X PE PE X M
PAB N
PCD X
PAC Y
PBD
MN XY
= q −hA,Di .
aut(E) aut(A)aut(B)aut(C)aut(D)
[E] [A],[B],[C],[D]

8
 The proof of Green’s formula is quite complex, and involves a large amount
of homological algebra. It was first presented by Ringel [14], and also appears
in [7] and [15] with a good amount of explaination. If you have never done so,
it is definitely worth reading. What we are interested in is the consequence of
Green’s formula.
We observe in Green’s formula the presence of our braiding coefficient
q −hA,Di . It is important to note that this coefficient depends on what some
might view as the “outside” objects A and D, and not the “inside” objects B
and C. This detail is compensated for in this paper by the use of our chosen
∆, instead of the a similar comultiplication described in other literature [7, 15].
In fact, in the category Vect our chosen comultiplication is the opposite of the
standard choice.
Lemma 9. in the category VectK the multiplication m and comultiplication ∆
satisfy the bialgebra condition, and thus H is a bialgebra object in VectK .
Proof. All the hard work for this proof was done in proving Green’s Formula. We
now just need to check that Green’s Formula gives us the bialgebra compatibility.
First we will multiply two objects, then comultiply the result to get:
X E
PMN
∆([M ] · [N ]) = ∆([E])
aut(M )aut(N )
[E]
X X PE PE MN XY
= [Y ] ⊗ [X]
aut(M )aut(N )aut(E)
[X],[Y ] [E]

On the other hand, if we first comultiply each object, then multiply the resulting
tensor products we have:
X M
PAB N
PCD
∆([M ]) · ∆([N ]) = ([B] ⊗ [A]) · ([D] ⊗ [C])
aut(M )aut(N )
[A],[B],[C],[D]

To continue, we need to remember the in our category VectG the braiding is

non-trivial. This means that if we want to extend the multiplication on H to
H ⊗ H we must include the braiding coefficient. Specifically, we get the formula:
([B] ⊗ [A]) · ([D] ⊗ [C]) = q −hA,Di [B] · [D] ⊗ [A] · [C]
Which when substituted above yields:
X M
PAB N
PCD
([B] ⊗ [A]) · ([D] ⊗ [C])
aut(M )aut(N )
[A],[B],[C],[D]
X M
PAB N
PCD
= q −hA,Di [B] · [D] ⊗ [A] · [C]
aut(M )aut(N )
[A],[B],[C],[D]
X X q −hA,Di PAB
M N
PCD X
PAC Y
PBD
= [Y ] ⊗ [X]
aut(M )aut(N )aut(A)aut(B)aut(C)aut(D)
[X],[Y ] [A],[B],[C],[D]

Thus, Green’s formula give the equality of the two side.

9
 For completeness, we will also define an antipode for this bialgebra object
to make it a Hopf algebra.
Lemma 10. The map S : H → H defined on generators by:

S([M ]) = −[M ]

is a K-grade preserving linear operator, and is an antipode for H. Thus H is a

Hopf algebra object in VectK .

References
[1] J. Baez, Groupoidification. Available at
http://math.ucr.edu/home/baez/groupoidification/.
[2] J. Baez and J. Dolan, Categorification, in Higher Category Theory, eds.
Ezra Getzler and Mikhail Kapranov, Contemp. Math. 230, American Math-
ematical Society, Providence, Rhode Island, 1998, pp. 1–36.
[3] J. Baez and J. Dolan, From finite sets to Feynman diagrams, in Mathemat-
ics Unlimited—2001 and Beyond, eds. Björn Engquist and Wilfried Schmid,
Springer, Berlin, 2001, pp. 29–50. Also available as arXiv:math/0004133.
[4] J. Baez, A. Hoffnung, and C. Walker, Groupoidification Made Easy. Avail-
able at arXiv:0812.4864v1.
[5] J. Baez, A. Hoffnung, and C. Walker, Higher Dimensional Algebra VII:
Groupoidification.
[6] P. Gabriel, Unzerlebare Darstellungen I, Manuscr. Math. 6 (1972), 71–103.
[7] A. W. Hubery, Ringel–Hall algebras, lecture notes available at
hhttp://www.maths.leeds.ac.uk/ ahubery/RHAlgs.htmli.
[8] A. Lauda, A categorification of quantum sl(2). Available at
arXiv:0803.3652v3.
[9] A. Lauda, Categorified quantum sl(2) and equivariant cohomology of iter-
ated flag varieties. Available at arXiv:0803.3848.
[10] S. Mac Lane, Categories for the Working Mathematician, Springer, Berlin,
1998.
[11] J. Morton, Categorified algebra and quantum mechanics, Theory Appl.
Categ.16, (2006), 785-854. Also available as arXiv:math/0601458.
[12] C. Ringel, Hall algebras and quantum groups, Invent. Math. 101 (1990),
583–591.

10
[13] C. Ringel, Hall algebras revisited, Israel Math. Conf. Proc. 7 (1993), 171–
176. Also available at
hhttp://www.mathematik.uni-bielefeld.de/∼ringel/opus/hall-rev.pdfi.
[14] C. Ringel, Green’s Theorem on Hall Algebras, Representation theory of
algebras and related topics (Mexico City, 1994), 185-225, CMS Conf. Proc.,
19, Amer. Math. Soc., Providence, RI, (1996)
[15] O. Schiffman, Lectures on Hall algebras, available as arXiv:math/0611617.

11