FINITE HOCHSCHILD COHOMOLOGY WITHOUT FINITE

GLOBAL DIMENSION
RAGNAR-OLAF BUCHWEITZ, EDWARD L. GREEN, DAG MADSEN, AND YVIND
SOLBERG

Introduction
In [9], Dieter Happel asked the following question: If the Hochschild coho-
mology groups HH
n
() of a nite dimensional algebra over a eld k vanish
for all suciently large n, is the global dimension of nite? We give a neg-
ative answer to this question. Indeed, consider the nite dimensional algebras
=
q
= kx, y)/(x
2
, xy + qyx, y
2
) for some eld k and q in k. These algebras,
intensely studied already by Rainer Schulz [13] to exhibit other pathologies, are
all four dimensional, of innite global dimension, and, for q ,= 0, selnjective. If q
is not a root of unity in k and non-zero, then we show that the total Hochschild
cohomology of
q
is of dimension 5, thus answering negatively Happels question as
promised. Moreover, if q is a root of unity, another interesting phenomenon occurs:
the Hochschild cohomology vanishes on large, but nite intervals, whose lengths
depend on the multiplicative order of q.
None of these phenomena can occur for commutative algebras according to the
companion paper [1]. Indeed, there it is shown, as a consequence of a more general
result, that a commutative nite dimensional algebra R is of nite global dimension
as soon as Hochschild cohomology vanishes just in one even and one odd degree, thus
giving a strong armation of Happels question in the commutative case. Let us
point out that the family
q
contains a few (graded) commutative specimens. For
q = 1, the algebra
1
is the commutative complete intersection k[x, y]/(x
2
, y
2
)
and for q = 1, the algebra
1
is the (graded commutative) exterior algebra in
two variables, isomorphic to the group ring k[Z
2
Z
2
] in characteristic two. For
q = 0, that is
0
= kx, y)/(x
2
, xy, y
2
), we discuss the structure of its Hochschild
cohomology in section 3.6.
The Hochschild cohomology ring HH

() has a ring structure given by the

Yoneda product. If R A and S A are two ring homomorphisms, we denote
the bre product (pullback) by R
A
S.
In this paper we prove the following.
Theorem. (a) If q is not a root of unity and non-zero, then HH

() is a 5-
dimensional algebra over the eld k with

i0
dim
k
HH
i
()t
i
= 2+2t +t
2
.
Moreover, HH

() is isomorphic as a ring to the bre product

k[z]/(z
2
)
k

(u
1
, u
2
)
Date: November 15, 2004.
1991 Mathematics Subject Classication. Primary: 16E40, 16G10, 16P10.

Supported by the Research council of Norway and NTNU.

1
2 BUCHWEITZ, GREEN, MADSEN, AND SOLBERG
where z is in degree 0, the elements u
1
and u
2
are in degree 1 and

(u
1
, u
2
)
is the exterior algebra on the generators u
1
and u
2
.
(b) If q is a root of unity, then HH

() is an innite dimensional algebra over

the eld k and nitely generated as a module over a polynomial subring
generated by two elements.
(c) For q = 0, then HH

() is an innitely generated algebra over k with

i0
dim
k
HH
i
()t
i
=
1+t
3
(1t)
2
.
The Hilbert series for the algebras in case (b) are found in sections 3.1 through
3.5.
If k k

is a eld extension, then HH

(k

k
) k

k
HH

() as graded k

-
algebras; see, for example, [12, Chap. X, Theorem 7.4]; whence we may, and will,
assume that k contains all roots of unity. In fact, then our computations show that
the gap of zero Hochschild cohomology groups from degree 3 to the rst non-zero
group thereafter can be arbitrarily large by choosing q to be a primitive r-th root
of unity for large r.
Let E() be the Koszul dual of ; that is, E() = Ext

(k, k). Our computations

of HH

(
q
) lead to a description of the graded centre of E(
q
), which is given in
section 2.3.
We note that the Hochschild homology of
q
for q in k, is computed in [8]. It is
shown there that dim
k
HH
n
(
q
) 2 for all n 0.
Throughout we consider right modules unless otherwise explicitly stated. The
enveloping algebra
e
of is given by
op

k
.
1. Background
In this section we try to put Happels question into a broader context. We
consider six conditions on a nite dimensional algebra and describe how they are
interrelated.
Let be a nite dimensional algebra over a eld k with Jacobson radial r. The
conditions we consider are the following.
(1) HH
n
() vanishes for n suciently large.
(1

) HH
n
() vanishes for n suciently large.
(2) The global dimension of is nite.
(3) For every eld extension k

of k, the global dimension of k

k
is nite.
(4) The projective dimension of as a
e
-module is nite.
(5) The global dimension of
e
is nite.
Happels question is: Does (1) imply (2)? Our results show that this is not true in
general, even though it is true in the commutative case [1]. There are some obvious
connections; namely, that (3) implies (2), (4) implies (1) and (1

), and (5) implies

(4). In general, the following implications hold:
(3)
ks +Q
(4)
z}
}
}
}
}
}
}
}
}
}
}
}
}
}
7
B
B
B
B
B
B
B
B
B
B
B
B
B
B
ks +Q
(5)
(1) (2) (1

)
The implication (4) implies (3) is a consequence of [4, Corollary IX.7.2]. If k is
algebraically closed, then Happel shows that (3) is equivalent to (4) [9, 1.5]. To
see that (3) implies (4) in general, consider a minimal projective resolution (P, )
FINITE HOCHSCHILD COHOMOLOGY WITHOUT FINITE GLOBAL DIMENSION 3
of over
e
. If k

is a eld extension of k, then (k

k
P, 1 ) is a minimal
projective (k

k
)
e
-resolution of k

k
, since k

k
r rad(k

k
), and for a
projective
e
-module P, the module k

k
P is a (k

k
)
e
-module. Taking k

to
be the algebraic closure of k, we see that if the projective dimension of over
e
is innite, then the projective dimension of k

k
over (k

k
)
e
is also innite.
Applying Happels result we infer that not (4) implies not (3).
The implication (4) implies (5) is found in [14]. In that paper van den Bergh
suggests that condition (4) is a reasonable non-commutative version of smoothness.
When /r is a separable algebra over k, the conditions (2), (3), (4) and (5) are
all equivalent. We have seen that (3), (4), and (5) are equivalent and (3) implies
(2) in general. To see that (2) implies (4), we use the following two observations.
First note that if /r is a separable algebra over k, then
e
/rad
e
is isomorphic to
/r
k
/r Hom
k
(D(/r), /r), where D denotes as usual the k-dual. Secondly,
Ext
i

e (, Hom
k
(D(/r), /r)) is isomorphic to Ext
i

(D(/r), /r) by [4, Proposi-

tion IX.4.3]. In particular, if /r is a separable algebra over k, (2) implies (1).
In general (2) does not imply (1) by the following example. Let k be a eld of
characteristic 2, and let = k(a) for some a not in k but a
2
in k. By [10] the
Hochschild cohomology ring of is isomorphic to k[x, y, z]/(x
2
, y
2
). Hence, the
global dimension of is zero, but (1) fails for .
If /r is not separable over k, condition (2) does not imply conditions (3), (4)
and (5) in general. For example, if is a purely inseparable nite eld extension
of k of the form k(a), then
e
is a local non-semisimple selnjective algebra over
k. Hence the global dimension of
k
k(a) is innite.
The results of this paper and [8] show that (1) does not imply (1

). Han [8] also

proves that (1

) is equivalent to (2) for monomial algebras and for commutative

algebras the equivalence was shown by Avramov and Viguee-Poirrier [2]. Han
conjectures that (1

) and (2) are equivalent.

2. The proof
Let =
q
= kx, y)/(x
2
, xy + qyx, y
2
) for some eld k and for some element
q in k 0. Order the elements by length left lexicographic order by choosing
x > y > 1. The set x
2
, xy + qyx, y
2
is then a (non-commutative) Gr obner basis
for the ideal (x
2
, xy +qyx, y
2
), and the normal forms of the elements in are given
by 1, x, y, yx (see [5]). Since has a quadratic Gr obner basis, is a Koszul
algebra (see [6]).
First we want to nd a minimal projective resolution (P, ) of over
e
. For
each n 0, we now construct certain elements f
n
i

n
i=0
in

k
n
, where

k
0
means
k. Let f
0
0
= 1, f
1
0
= x and f
1
1
= y. By letting f
n
1
= 0 = f
n
n+1
for all n 0, we
dene f
n
i
for n 2 inductively by setting
f
n
i
= f
n1
i1
y + q
i
f
n1
i
x
for i in 0, 1, . . . , n. It follows that f
n
i
is a linear combination of all tensors of xs
and ys of length n with y occurring i times. The coecient q

in front of such
a n-fold tensor w is determined by setting equal to the number of xs that the
ys have to move across starting with the n-fold tensor x x
. .
i
y y
. .
ni
to
obtain the given tensor w. Using this we infer that
f
n
i
= x f
n1
i
+ q
ni
y f
n1
i1
.
4 BUCHWEITZ, GREEN, MADSEN, AND SOLBERG
Let P
n
= H
n
i=0

k
f
n
i

k

k
(n+2)
for n 0, and let

f
n
i
= 1 f
n
i
1 for
n 1 and

f
0
0
= 1 1. Dene
n
: P
n
P
n1
by letting

n
(

f
n
i
) = x

f
n1
i
+ q
ni
y

f
n1
i1
+ (1)
n

f
n1
i1
y + (1)
n
q
i

f
n1
i
x.
Direct computations show that
2
= 0.
One can verify directly that (P, ) is a projective
e
-resolution of as follows.
Let be a Koszul k-algebra where the degree zero part is a k. Suppose that M is a
Koszul module over for which the Betti numbers b
n

n0
in a minimal projective
resolution are known, and that
G
n
d
n
G
n1
G
1
d
1
G
0
d
0
M 0
is a linear complex G of projective -modules G
i
generated by b
i
generators in
degree i, where d
0
: G
0
M is onto and d
i
[
G
i
i
: G
i
i
G
i1
i
is injective for all i 0.
Then it follows easily by induction that G is a minimal projective resolution of M.
In our situation we apply this in the following way: Let d
0
: P
0
be the
multiplication map. The algebra
e
is a Koszul algebra over k. By a classical
formula of Cartan-Eilenberg Ext
n

e (, k) Ext
n

(k, k). The Koszul dual E() =

Ext

(k, k) of is isomorphic to kx, y)/(yx qxy), so that the Betti numbers of a

minimal projective resolution of over
e
are b
n
= n + 1
n0
. Moreover by [7]
(see [3] for a proof) a graded algebra generated in degrees 0 and 1 is Koszul if
and only if is Koszul as a right module over
e
. Hence is a Koszul module over

e
. It is straightforward to check that the above conditions are satised for (P, ).
Hence (P, ) is a minimal projective resolution of over
e
.
Our next immediate goal is to obtain a multiplication formula for elements in
the Hochschild cohomology ring; see the formula (1) below. Recall that the bar
resolution B of over
e
is given by B
n
=

k
(n+2)
and dierential d
n
: B
n

B
n1
given by
d
n
(
0

1

n+1
)
=
n

i=0
(1)
i
(
0

1

i1

i

i+1

i+2

n+1
)
Dene
i
: B
n
B
n1
by

i
(
0

n+1
) = (
0

i1

i

i+1

i+2

n+1
)
for i = 0, 1, . . . , n. Then d
n
=

n
i=0
(1)
n

i
.
We view P
n
as a submodule of B
n
in the natural way. For any given i, consider
the tensors of the form wxxw

, wxyw

, wyxw

, or wyyw

occurring in a

f
n
i
, where w is an t-fold tensor. The rst and the last of these are
mapped to zero by
t
. If a tensor q
t
w x y w

occurs in some

f
n
i
, then
q
t+1
w y x w

also occurs in

f
n
i
. Then the sum of these elements is mapped
to zero by
t
. It follows from this that the element

f
n
i
is mapped to the element
x

f
n1
i
+ q
ni
y

f
n1
i1
+ (1)
n

f
n1
i1
y + (1)
n
q
i

f
n1
i
x
by the dierential in B. This shows that (P, ) is a subcomplex of (B, d).
FINITE HOCHSCHILD COHOMOLOGY WITHOUT FINITE GLOBAL DIMENSION 5
The bar resolution B admits a comultiplication : B B

B, that is a mor-
phism of complexes over the identity on , given by
(
0

n+1
) =
n

i=0
(
0

i
1)

(1
i+1

n+1
).
The (cup) product in HH

() of two cycles and from Hom

e(B, ) is given as
the composition of the maps
B

B

,
where :

is the natural map. Let : P B be the natural inclusion,

and let : B P be a chain map such that = id
P
.
Suppose that the image (P) is already contained in P

P B

B. In
that case, induces a comultiplication on P that we denote by

. In particular
= ( )

. For homogeneous elements and in HH

() we can represent
these elements as morphisms : P
m
and : P
n
. Using the bar resolution
they can be represented by
m
and
n
. Then we have
= ( )
m+n

m+n
= (
m

n
)
m+n
= (
m

n
)
m+n
= (
m

n
)(
m

n
)

= ( )

where denotes the product of and in HH

().
Now we show that indeed maps (P) to P

P. Using the description of

the terms in f
n
i
it is straightforward to verify that
f
n
i
=
min{t,i}

j=max{0,i+tn}
q
j(ni+jt)
f
t
j
f
nt
ij
for t = 0, 1, . . . , n. From this we infer that
(

f
n
i
) =
n

t=0
min{t,i}

j=max{0,i+tn}
q
j(ni+jt)

f
t
j

f
nt
ij
,
hence we see that : P P

P and we can dene

by the above formula.

This enables us to describe the multiplication in HH

(), which we do next.

Let : P
m
and : P
n
represent elements in HH
m
() and HH
n
(),
respectively, where (

f
m
i
) =
i
for i = 0, 1, . . . , m and (

f
n
i
) =

i
for i = 0, 1, . . . , n.
Then
(1) (

f
m+n
i
) =
min{m,i}

j=max{0,in}
q
j(ni+j)

ij
for i = 0, 1, . . . , m + n. We note that the product in formula (1) is written
as a map P
m+n
, and hence may be zero as an element in HH

(). In [3] we
show that this multiplicative structure can be generalized to any Koszul algebra
= kQ/I for a nite quiver Q.
Now let us consider the vector space structure of HH
n
() for all n 0.
6 BUCHWEITZ, GREEN, MADSEN, AND SOLBERG
Since P
i
(
e
)
i+1
, we see that Hom

e (P
n1
, ) is isomorphic to
n
, which we
view as an identication. Let : P
n1
be given by (
0
, . . . ,
n1
). Then for
n 1

n
= (x
0
+ (1)
n

0
x, . . . ,
x
j
+ q
nj
y
j1
+ (1)
n

j1
y + (1)
n
q
j

j
x
. .
j-th coordinate
, . . . ,
y
n1
+ (1)
n

n1
y)
The image Im(
n
)

is spanned by vectors obtained from the above by choosing

i
to be in the set 1, x, y, since the value yx only gives zero. When
j
is x or y and
all the other
i
s are zero, then the result is a standard basis vector times ayx for
some a in k. Hence this can give rise to at most n+1 linearly independent elements.
When
j
= 1 and all the other
i
s are zero, then the resulting element has the form
(0, . . . , 0, ax
..
j
, by, 0, . . . , 0) for some a and b in k and for j = 0, 1, . . . , n1. The non-
zero elements of this form are linearly independent and also linearly independent
of those obtained by choosing
j
in x, y. This shows that dimIm(
n
)

2n + 1
for n 2.
When
i
= 0 for i ,= j and
j
= y, then

n
= (0, . . . , 0, (q + (1)
n
q
j
)yx
. .
j
, 0, . . . , 0)
for j = 0, 1, . . . , n 1. If
j1
= x and all other
i
are zero, then

n
= (0, . . . , 0, (q
nj
(1)
n
q)yx
. .
j
, 0, . . . , 0)
for j = 1, 2, . . . , n. The coecients in the j-coordinate of these elements are zero if
(i) for j = 0 we have q = (1)
n
, (ii) for j = 1, 2, . . . , n 1 we have q
j1
= (1)
n
and q
nj1
= (1)
n
, and (iii) for j = n we have q = (1)
n
. For each j such that
one of these three cases occurs, the dimension of Im(
n
)

decreases by one. If q
is not a root of unity and n 3, then these conditions never hold.
When
i
= 0 for i ,= j and
j
= 1, then

n
= (0, . . . , 0, (1 + (1)
n
q
j
)x
. .
j
, (q
nj1
+ (1)
n
)y, 0, . . . , 0)
for j = 0, 1, . . . , n 1. The coecients in this vector are zero if and only if q
j
=
(1)
n+1
and q
nj1
= (1)
n+1
. If q is not a root of unity and n 2, these
equations are never satised. Hence, when q is not a root of unity and n 3, the
dimension of Im(
n
)

is 2n + 1.
Using these facts, let us rst compute the groups HH
n
() for n = 0, 1, 2. We
have that
HH
0
() = Z() =
_
span
k
1, yx, q ,= 1
, q = 1
FINITE HOCHSCHILD COHOMOLOGY WITHOUT FINITE GLOBAL DIMENSION 7
From the above computations HH
1
() is spanned as a vector space in
Ker(
2
)

/ Im(
1
)

by the following maps

_

_
(x, 0), (0, y), q ,= 1, 1
(x, 0), (yx, 0), (0, yx), (0, y), q = 1, q ,= 1
(x, 0), (y, 0), (0, x), (0, y), q ,= 1, q = 1
Hom

e (P
1
, ), q = 1, char k = 2
Similarly as above HH
2
() is spanned by the following maps
_

_
(0, yx, 0), q ,= 1, 1
(1, 0, 0), (y, 0, 0), (0, yx, 0), (0, 0, x), (0, 0, 1), q = 1, q ,= 1
(1, 0, 0), (0, 1, 0), (0, 0, 1), (yx, 0, 0), (0, yx, 0), (0, 0, yx), q ,= 1, q = 1
Hom

e (P
2
, ), q = 1, char k = 2
2.1. When q is not a root of unity and non-zero. Suppose now that q is not
a root of unity. Since dim
k
(P
n
)

= 4(n+1) and dim

k
Im(
n
)

+dim
k
Im(
n+1
)

=
4(n+1) for n 3, it follows that Im(
n
)

= Ker(
n+1
)

and HH
n
() = (0) for those
values of n. By the considerations above we have that (i) HH
0
() = span
k
1, yx,
(ii) HH
1
() = span
k
(x, 0), (0, y) and (iii) HH
2
() = span
k
(0, yx, 0). Denote
the element yx in HH
0
() by z and the two elements (x, 0) and (0, y) in HH
1
() by
u
0
and u
1
, respectively. Using the multiplication worked out earlier, it is straight-
forward to compute that the elements z
2
, zu
0
, u
0
z, zu
1
, u
1
z, u
2
0
, u
2
1
are all zero,
while u
0
u
1
= q(0, yx, 0) and u
1
u
0
= q(0, yx, 0). Hence HH

() is isomorphic to
the bre product
k[z]/(z
2
)
k

(u
0
, u
1
),
where

(u
0
, u
1
) is the exterior algebra on the generators u
0
and u
1
.
2.2. When q is a root of unity. Suppose now that q is a primitive r-th root of
unity. Fix an n. Whenever dim
k
Im(
n
)

< 2n +1 the group HH

i
() ,= (0) for i in
n 1, n. We observed above that this happened if
(a) (i) for j = 0 we have q = (1)
n
,
(ii) for some j = 1, 2, . . . , n1 we have q
j1
= (1)
n
and q
nj1
= (1)
n
,
(iii) for j = n we have q = (1)
n
,
(b) for some j = 0, 1, . . . , n 1 we have q
j
= (1)
n+1
and q
nj1
= (1)
n+1
.
When j = 1 and n 2 = mr with n even for some m 1, then condition (a)(ii) is
satised. When j = 0 and n 1 = mr with n odd for some m 1, then condition
(b) is satised. This shows that when q is a root of unity, then HH

() is innite
dimensional over k.
The Koszul dual E() of is isomorphic to kx, y)/(yx qxy). Recall that the
graded centre Z
gr
(E()) of E() is generated by z E() [ z homogeneous, z =
(1)
deg z deg
z for all homogeneous E(). Using that E() is bigraded it is
straightforward to compute Z
gr
(E()). If r is even or char k = 2, then Z
gr
(E())
is equal to H

= k[x
r
, y
r
]. Otherwise Z
gr
(E()) is equal to k[x
2r
, x
r
y
r
, y
2r
] and
therefore contains H

= k[x
2r
, y
2r
]. In either case E() is a nitely generated
module over H

. By [3] or [11] the image of the natural map HH

() E() is
Z
gr
(E()). Take homogeneous inverse images of the two generators of H

, and let H
denote the polynomial subring of HH

() they generate. Since c() HH

(, S)
8 BUCHWEITZ, GREEN, MADSEN, AND SOLBERG
where S is the only simple
e
-module, HH

() is also a nitely generated module

over H as H is noetherian and is ltered in S as a
e
-module.
2.3. The graded centres of E(
q
). By direct calculations and the remarks above
we have the following description of the graded centre of E(
q
) for all q in k.
Proposition 2.1. The graded centre of E(
q
) is as follows.
Z
gr
(E(
q
)) =
_

_
k, q is not a root of unity
k[x
r
, y
r
],
_
q is a primitive r-th unity and r even, or
char k = 2
k[x
2r
, x
r
y
r
, y
2r
], otherwise
3. Further details
In this section we give the full ring structure of HH

(
q
) and the Hilbert series
with respect to the cohomological degree, when q is an r-th primitive root of unity
and for completeness when q = 0.
The description of the ring structure of HH

() is divided into several cases.

The vector space basis for HH
i
() for i = 0, 1, 2 are given in the previous section
when q ,= 0 and below when q = 0. We rst give a vector space basis for HH
i
()
for i 3 in the dierent cases, and then the full ring structure of HH

().
The cases we have to consider are the following (a) r > 1 is odd and char k ,= 2,
(b) char k = 2 with q ,= 1, or char k ,= 2 and r > 2 is even, (c) char k = 2 and q = 1,
(d) char k ,= 2 and q = 1 and nally (e) char k ,= 2 and q = 1.
3.1. The case r > 1 is odd and char k ,= 2. In this case representing maps for
HH
n
() when n > 2 are given by
HH
n
() =
_

_
span
k
e
j

j{tr}
2s
t=0
, n = 2sr
span
k
xe
j1

j{tr+1}
2s
t=0
, ye
j

j{tr+1}
2s
t=0
, n = 2sr + 1
span
k
yxe
j

j{tr+1}
2s
t=0
, n = 2sr + 2
0, otherwise
Using the description of the multiplication given in the previous section and re-
membering that q is an r-th root of unity, one can see that
HH

() k[z]/(z
2
)
k
_

(u
0
, u
1
)[w
0
, w
1
, w
2
]/(w
0
w
2
w
2
1
)
_
where z is in degree 0, u
0
= (x, 0) and u
1
= (0, y) are in degree 1, and w
i
: P
2r

is in degree 2r and given by w
i
(e
ir
) = 1 and zero otherwise for i = 0, 1, 2. We have
that

i0
dim
k
HH
i
()t
i
= 1 +
(1+t)
2
(1+t
2r
)
(1t
2r
)
2
.
3.2. The case char k = 2 with q ,= 1, or char k ,= 2 and r > 2 is even.
Representing maps for HH
n
() when n > 2 are in this case given by
HH
n
() =
_

_
span
k
e
j

j{tr}
s
t=0
, n = sr
span
k
xe
j1

j{tr+1}
s
t=0
, ye
j

j{tr+1}
s
t=0
, n = sr + 1
span
k
yxe
j

j{tr+1}
s
t=0
, n = sr + 2
0, otherwise
Similar as above we nd that
HH

() k[z]/(z
2
)
k
(

(u
1
, u
2
)[w
0
, w
1
])
FINITE HOCHSCHILD COHOMOLOGY WITHOUT FINITE GLOBAL DIMENSION 9
where z is in degree 0, u
0
= (x, 0) and u
1
= (0, y) are in degree 1, and w
i
: P
r

is in degree r and given by w
i
(e
j
) = 1 for j = ir and zero otherwise for i = 0, 1.
The Hilbert series is given by

i0
dim
k
HH
i
()t
i
= 1 +
(1+t)
2
(1t
r
)
2
.
3.3. The case char k = 2 and q = 1. Here the dierential in the complex
Hom

e(P, ) is zero, so that all maps in Hom

e (P
n
, ) represent non-zero ele-
ments in HH

(). We have that HH

() [w
0
, w
1
] where w
i
: P
1
is
given by w
i
(e
j
) = 1 for i = j and zero otherwise for i = 0, 1. Furthermore

i0
dim
k
HH
i
()t
i
=
4
(1t)
2
.
3.4. The case char k ,= 2 and q = 1. In this case representing maps for HH
n
()
when n > 2 are given by
HH
n
() =
_
_
_
span
k
e
j

j{2t}
n
2
t=0
, ye
0
, xe
n
, yxe
j

j{2t1}
n
2
t=1
, n even
span
k
xe
j1

j{2t1}
n+1
2
t=1
, ye
j

j{2t1}
n+1
2
t=1
, yxe
0
, yxe
n
, n odd
This gives
HH

() (
k

(u
0
, u
1
)) [w
0
, w
1
]/(xu
0
, yu
1
, xw
0
, yw
1
),
where u
0
= (x, 0) and u
1
= (0, y) are in degree 1 and w
0
= (1, 0, 0) and w
1
= (0, 0, 1)
are in degree 2. The Hilbert series is given by

i0
dim
k
HH
i
()t
i
=
44t+t
2
(1t)
2
.
As
1
k[x]/(x
2
)
k
k[y]/(y
2
), there is an isomorphism of vector spaces
HH

(
1
) HH

(k[x]/(x
2
))
k
HH

(k[y]/(y
2
))
(k[x]/(x
2
)
k

(u
0
)[w
0
])
k
(k[y]/(y
2
)
k

(u
1
)[w
1
])
by [12, Chap. X, Theorem 7.4], which is in accordance with our result!
3.5. The case char k ,= 2 and q = 1. In this case representing maps for HH
n
()
when n > 2 are given by
HH
n
() =
_
span
k
e
j

n
j=0
, yxe
j

n
j=0
, n even
span
k
xe
j

n
j=0
, ye
j

n
j=0
, n odd
It follows that
HH

()
_
k[z]/(z
2
)
k

(u
0
, u
1
, u
2
, u
3
)
_
[w
0
, w
1
, w
2
]/I,
where z is in degree 0, u
0
= (x, 0), u
1
= (y, 0), u
2
= (0, x), u
3
= (0, y) are in
degree 1, and w
i
: P
2
is given by w
i
(e
j
) = 1 for j = i and zero otherwise for
i = 0, 1, 2. The ideal I is given by
(u
0
u
2
, u
1
u
3
, u
0
u
1
+ zw
0
, u
0
u
3
+ zw
1
, u
2
u
3
+ zw
2
, u
1
u
2
zw
1
,
u
0
w
1
u
2
w
0
, u
1
w
1
u
3
w
0
, u
0
w
2
u
2
w
1
, u
1
w
2
u
3
w
1
, w
0
w
2
w
2
1
)
We have that

i0
dim
k
HH
i
()t
i
=
2
(1t)
2
.
10 BUCHWEITZ, GREEN, MADSEN, AND SOLBERG
3.6. The case q = 0. For completeness we now consider the degenerate case when
q = 0. We have that the centre of is given by span
k
1, yx, so that HH
0
() =
span
k
1, yx. It is left to the reader to verify that HH
1
() = span
k
(x, 0), (0, y).
For n > 1 the representing maps for HH
n
() are
span
k
xe
i

n2
i=0
, ye
n
, yxe
i

n1
i=1
.
It follows from this that all products in HH
1
() are zero. Hence the Hochschild
cohomology ring of is isomorphic to
k[v
00
, v
10
, v
11
, v
ij

i2,j{0,1,...,2i2}
]/(v
ij
v
i

j
),
where v
ij
is in degree i. Finally

i0
dim
k
HH
i
()t
i
=
1+t
3
(1t)
2
.
References
[1] Avramov, L. L., Iyengar, S., Gaps in Hochschild cohomology imply smoothness for commu-
tative algebras, arXiv:math, to appear.
[2] Avramov, L. L., Viguee-Poirrier, M., Hochschild homology criteria for smoothness, Internat.
Math. Research Notices 1 (1992), 1725.
[3] Buchweitz, R.-O., Green, E. L., Snashall, N., Solberg, ., Multiplicative structures for Koszul
algebras, preprint 2004.
[4] Cartan, H., Eilenberg, S., Homological algebra, Princeton University Press, 1956.
[5] Green, E. L., Non-commutative Gr obner basis and projective resolutions, Progress in Math.,
vol. 173 (1999), Birkh auser Verlag, 2960.
[6] Green, E. L., Huang, R. Q., Projective resolutions of straightening closed algebras generated
by minors, Adv. Math. 110 (1995), no. 2, 314333.
[7] Green, E. L., Zacharia, D., Unpublished result.
[8] Han, Y., Hochschild (co)homology dimension, arXiv:math.RA/0408402.
[9] Happel, D., Hochschild cohomology of nite-dimensional algebras, Seminaire dAlg`ebre Paul
Dubreil et Marie-Paul Malliavin, 39`eme Annee (Paris, 1987/1988), 108126, Lecture Notes
in Math., 1404, Springer, Berlin, 1989.
[10] Holm, T., Hochschild cohomology rings of algebras k[X]/(f), Beitr age Algebra Geom. 41
(2000), no. 1, 291301.
[11] Keller, B., Derived invariance of higher structures on the Hochschild complex, preprint.
[12] MacLane, S., Homology, Classics in Mathematics, Springer-Verlag, Berlin, 1995, x+422 pp.
[13] Schulz, R., A nonprojective module without self-extensions, Arch. Math. (Basel) 62 (1994),
no. 6, 497500.
[14] van den Bergh, M., Erratum to: A relation between Hochschild homology and cohomology
for Gorenstein rings, [Proc. Amer. Math. Soc. 126 (1998), no. 5, 13451348]. Proc. Amer.
Math. Soc. 130 (2002), no. 9, 28092810.
Ragnar-Olaf Buchweitz, Department of Mathematics, University of Toronto,
Toronto, ON Canada M5S 3G3, Canada
E-mail address: ragnar@math.toronto.edu
Edward L. Green, Department of Mathematics, Virginia Tech, Blacksburg, VA 24061,
USA
E-mail address: green@math.vt.edu
Dag Madsen, Institutt for matematiske fag, NTNU, N7491 Trondheim, Norway
E-mail address: dagma@math.ntnu.no
yvind Solberg, Institutt for matematiske fag, NTNU, N7491 Trondheim, Norway
E-mail address: oyvinso@math.ntnu.no