Decompositions of

Modules and Comodules

Robert Wisbauer
University of D usseldorf, Germany
Abstract
It is well-known that any semiperfect A ring has a decom-
position as a direct sum (product) of indecomposable subrings
A = A
1
A
n
such that the A
i
-Mod are indecomposable
module categories. Similarly any coalgebra C over a eld can
be written as a direct sum of indecomposable subcoalgebras
C =

I
C
i
such that the categories of C
i
-comodules are inde-
composable. In this paper a decomposition theorem for closed
subcategories of a module category is proved which implies both
results mentioned above as special cases. Moreover it extends
the decomposition of coalgebras over elds to coalgebras over
noetherian (QF) rings.
1 Introduction
The close connection between module categories and comodule cate-
gories was investigated in [12] and it turned out that there are parts
of module theory over algebras which provide a perfect setting for the
theory of comodules. In a similar spirit the present paper is devoted
to decomposition theorems for closed subcategories of a module cate-
gory which subsume decomposition properties of algebras as well as of
coalgebras.
Let A be an associative algebra over a commutative ring R. For
an A-module M we denote by [M] the category of those A-modules
which are submodules of M-generated modules. This is the smallest
1
Grothendieck subcategory of A-Mod containing M. The inner proper-
ties of [M] are dependent on the module properties of M and there is
a well established theory dealing with this relationship.
We dene a -decomposition
[M] =

[N

],
for a family N

of modules, meaning that for every module L

[M], L =

,where L

[N

]. We call [M] -indecomposable if

no such non-trivial decomposition exists.
The [N

] are closely related to fully invariant submodules of a

projective generator (if there exists one) and - under certain niteness
conditions - to the fully invariant submodules of an injective cogen-
erator. Consequently an indecomposable decomposition of [M] can
be obtained provided there is a semiperfect projective generator or an
injective cogenerator of locally nite length in [M].
Such decompositions of [M] were investigated in Vanaja [10] and
related constructions are considered in Garca-Jara-Merino [4], Nast ases-
cu-Torrecillas [8] and Green [5].
Let C be a coalgebra over a commutative ring R. Then the dual C

is an R-algebra and C is a left and right module over C

. The link to
the module theory mentioned above is the basic observation that the
category of right C-comodules is subgenerated by C. Moreover, if
R
C
is projective, this category is the same as [
C
C]. This is the key to
apply module theory to comodules and our decomposition theorem for
[M] yields decompositions of coalgebras and their comodule categories
over noetherian (QF) rings. For coalgebras over elds such results were
obtained in Kaplansky [6], Montgomery [7], Shudo-Miyamoto [9].
2 Decompositions of module categories
Throughout R will denote an associative commutative ring with unit,
A an associative R-algebra with unit, and A-Mod the category of unital
left A-modules.
We write [M] for the full subcategory of A-Mod whose objects are
submodules of M-generated modules. N [M] is called a subgenera-
tor if [M] = [N].
2
2.1 The trace functor. For any N, M A-Mod the trace of M in N
is dened as
Tr(M, N) :=

Imf [ f Hom
A
(M, N),
and we denote the trace of [M] in N by
T
M
(N) := Tr([M], N) =

Imf [ f Hom
A
(K, N), K [M].
If N is M-injective, or if M is a generator in [M], then T
M
(N) =
Tr(M, N).
A full subcategory ( of A-Mod is called closed if it is closed under
direct sums, factor modules and submodules (hence it is a Grothendieck
category). It is straightforward to see that any closed subcategory is of
type [N], for some N in A-Mod.
The next result shows the correspondence between the closed sub-
categories of [M] and fully invariant submodules of an injective co-
generator of [M], provided M has locally nite length.
2.2 Correspondence relations. Let M be an A-module which is
locally of nite length and Q an injective cogenerator in [M].
(1) For every N [M], [N] = [Tr(N, Q)].
(2) The map [N] Tr(N, Q) yields a bijective correspondence be-
tween the closed subcategories of [M] and the fully invariant
submodules of Q.
(3) [N] is closed under essential extensions (injective hulls) in [M]
if and only if Tr(N, Q) is an A-direct summand of Q.
(4) N [M] is semisimple if and only if Tr(N, Q) Soc(
A
Q).
Proof. Notice that by our niteness condition every cogenerator in
[M] is a subgenerator in [M]. Moreover by the injectivity of Q,
Tr([N], Q) = Tr(N, Q).
(1) Tr(N, Q) is a fully invariant submodule which by denition be-
longs to [N]. Consider the N-injective hull

N of N (in [N]). This is
a direct sum of N-injective hulls

E of simple modules E [N]. Since
Q is a cogenerator we have (up to isomorphism)

E Q and so

E
Tr(N, Q). This implies

N [Tr(N, Q)] and so [N] = [Tr(N, Q)].
3
(2) and (4) are immediate consequences of (1).
(3) If [N] is closed under essential extensions in [M] then clearly
Tr(N, Q) is an A-direct sumand in Q (and hence is injective in [M]).
Now assume Tr(N, Q) to be an A-direct sumand in Q and let L be
any injective object in [N]. Then L is a direct sum of N-injective hulls

E of simple modules E [N]. Clearly the

Es are (isomorphic to)
direct summands of Tr(N, Q) and hence of Q, i.e., they are M-injective
and so L is M-injective, too. 2
2.3 Sum and decomposition of closed subcategories. For any
K, L [M] we write [K] [L] = 0, provided [K] and [L] have
no non-zero object in common. Given a family N

of modules in
[M], we dene

[N

] := [

].
This is the smallest closed subcategory of [M] containing all the N

s.
Moreover we write
[M] =

[N

],
provided for every module L [M], L =

T
N

(L) (internal direct

sum). We call this a -decomposition of [M], and we say [M] is
-indecomposable if no such non-trivial decomposition exists.
In view of the fact that every closed subcategory of A-Mod is of
type [N], for some A-module N, the above denition describes the
decomposition of any closed subcategory into closed subcategories.
2.4 -decomposition of modules. For a decomposition M =

,
the following are equivalent:
(a) for any distinct , , M

and M

have no non-zero isomor-

phic subfactors;
(b) for any distinct , , Hom
A
(K

, K

) = 0, where K

, K

are
subfactors of M

, M

, respectively;
(c) for any distinct , , [M

] [M

] = 0;
(d) for any , [M

] [

=
M

] = 0;
(e) for any L [M], L =

T
M

(L).
4
If these conditions hold we call M =

a -decomposition of M
and in this case
[M] =

[M

].
Proof. (a) (b) and (e) (a) are obvious.
(b) (c) This follows from the plain fact that for any A-module
N, each non-zero module in [N] contains a non-zero subfactor of N.
(c) (d) Any non-zero module in [

=
M

] contains a non-zero
subfactor of M

, for some . This implies (d).

(d) (e) It is easy (see [10]) to verify that L =

T
M

(L). 2
2.5 Corollary. Let [M] =

[N

] be a -decomposition of [M].
Then
(1) each [N

] is closed under essential extensions in [M];

(2) any L [N

] is M-injective if and only if it is N

-injective;
(3) M =

T
N

(M) is a -decomposition of M.
Proof. (1) For any L [N

], consider an essential extension LK in

[M]. Then T
N

(K) K and is a direct summand. So K = T

N

(K)
[N

].
(2) Any L [N

] is M-injective if it has no non-trivial essential

extensions in [M]. Assume L to be N

-injective. Then L has no non-

trivial essential extensions in [N

] and by (1), it has no non-trivial

essential extensions in [M].
(3) Put M

:= T
N

(M). By denition, M

[N

] and it remains
to show that N

[M

]. Let

N

denote the M-injective hull of N

is M-generated, and by (1),

N

[N

]. This implies that

N

is
M

-generated and so N

[M

]. 2
It is obvious that any -decomposition of M is also a fully invariant
decomposition. The reverse implication holds in special cases:
2.6 Corollary. Assume M to be a projective generator or an injective
cogenerator in [M]. Then any fully invariant decomposition of M is
a -decomposition.
5
Proof. Let M =

be a fully invariant decomposition, i.e.,

Hom
A
(M

, M

) = 0, for ,= .
Assume M to be a projective generator in [M]. Then clearly every
submodule of M

is generated by M

. Since the M

s are projective
in [M], any non-zero (iso)morphism between (sub)factors of M

and
M

yields a non-zero morphism between M

and M

. So our assertion
follows from 2.4.
Now suppose that M is an injective cogenerator in [M]. Then
every subfactor of M

must be cogenerated by M

. From this it follows

that for ,= , there are no non-zero maps between subfactors of M

and M

and so 2.4 applies. 2

Remark. 2.4 and 2.6 are shown in Vanaja [10, Proposition 2.2] and
related constructions are considered in Garca-Jara-Merino [4, Section
3] and [3, Theorem 5.2], Nast asescu-Torrecillas [8, Lemma 5.4] and
Green [5, 1.6c].
Following Garca-Jara-Merino [3], we call a module M -indecompos-
able if M has no non-trivial -decomposition.
2.7 Corollary. The following are equivalent:
(a) [M] is -indecomposable;
(b) M is -indecomposable;
(c) any subgenerator in [M] is -indecomposable;
(d) an injective cogenerator which is a subgenerator in [M], has no
fully invariant decomposition.
If there exists a projective generator in [M] then (a)-(d) are equivalent
to:
(e) projective generators in [M] have no fully invariant decomposi-
tions.
It is straightforward to see that a -decomposition of the ring A is
of the form
A = Ae
1
Ae
k
, for central idempotents e
i
A,
and so A-Mod is -indecomposable if and only if A has no non-trivial
central idempotent.
6
By the structure theorem for cogenerators with commutative endo-
morphism rings (see [11, 48.16]) we have:
2.8 -decomposition when End
A
(M) commutative. Let M be a
cogenerator in [M] with End
A
(M) commutative. Then M

,
where E

is a minimal representing set of simple modules in [M].

This is a -decomposition of M and
[M] =

],
where each [

] is indecomposable and contains only one simple mod-

ule.
A special case of the situation described above is the ZZ-module
I Q/ZZ =

p prime
ZZ
p
and the decomposition of the category of torsion
abelian groups as a direct sum of the categories of p-groups,
[ I Q/ZZ] =

p prime
[ZZ
p
].
Notice that although I Q/ZZ is an injective cogenerator in ZZ-Mod
with a non-trivial -decomposition, ZZ-Mod is -indecomposable. This
is possible since I Q/ZZ is not a subgenerator in ZZ-Mod.
In general it is not so easy to get -decompositions of modules. We
need some technical observations to deal with modules whose endomor-
phism rings are not commutative.
2.9 Relations on families of modules. Consider any family of A-
modules M

in [M]. Dene a relation on M

by putting
M

if there exist non-zero morphisms M

or M

.
Clearly is symmetric and reexive and we denote by the smallest
equivalence relation on M

determined by , i.e.,
M

if there exist
1
, . . . ,
k
, such that
M

= M

1
M

k
= M

.
Then M

is the disjoint union of the equivalence classes [M

.
Assume each M

, the M-injective hull of some simple module

E

[M]. Then
7

if and only if Ext

M
(E

, E

) ,= 0 or Ext
M
(E

, E

) ,= 0,
where Ext
M
denotes the extensions in [M].
Proof. For any non-zero morphism

E

, there exists a submodule

E

L

E

with a non-splitting exact sequence

0 E

L E

0.
Assume such a sequence is given. From this it is easy to construct
a non-zero morphism f :

E

. 2
A decomposition M =

is said to complement direct sum-

mands if, for every direct summand K of M, there exists a subset

such that M = K (

) (cf. [1, 12]). We observe that

such decompositions yield fully invariant indecomposable decomposi-
tions.
2.10 Lemma. Let M =

be a decomposition which complements

direct summands, where all M

are indecomposable. Then M has a

decomposition M =

A
N

, where each N

M is a fully invariant
submodule and does not decompose non-trivially into fully invariant
submodules.
Proof. Consider the equivalence relation on M

(see 2.9) with

the equivalence classes [M

and =

. Then N

:=

is a fully invariant submodule of M, for each , and

M =

.
Assume N

= K L for fully invariant K, L N

. Since the dening

decomposition of N

complements direct summands we may assume

that

is the disjoint union of subsets X, Y such that

N

X
M

Y
M

.
By construction, for any x X, y Y , we have M
x
M
y
and it is easy
to see that this implies the existence of non-zero morphisms K L
or L K, contradicting our assumption. So N

does not decompose

into fully invariant submodules. 2
The following are standard examples of module decompositions which
complement direct summands.
8
2.11 Proposition. Let M =

, where each End

A
(M

) is local.
(1) If M is M-injective the decomposition complements direct sum-
mands.
(2) If M is projective in [M] and Rad(M) << M, then the decom-
position complements direct summands.
Proof. For the rst assertion we refer to [2, 8.13].
The second condition characterizes M as semiperfect in [M] (see
[11, 42.5]) and the assertion follows from [2, 8.12]. 2
2.12 -decomposition for locally noetherian modules. Let M be
a locally noetherian A-module. Then M has a -decomposition M =

and
[M] =

[M

],
where each [M

] is -indecomposable.
(1) [M] is -indecomposable if and only if for any indecomposable
injectives K, L [M], K L (as dened in 2.9).
(2) If M has locally nite length, then [M] is -indecomposable if
and only if for any simple modules S
1
, S
2
[M],

S
1


S
2
(M-
injective hulls).
Proof. Let Q be an injective cogenerator which is also a subgenera-
tor in [M]. Then Q is a direct sum of indecomposable M-injective
modules and this is a decomposition which complements direct sum-
mands (by 2.11). By Lemma 2.10, Qhas a fully invariant decomposition
Q =

such that Q

has no non-trivial fully invariant decomposi-

tion. Now the assertions follow from Corollaries 2.6, 2.7, and 2.5.
(1) This is clear by the above proof.
(2) By our assumption every indecomposable M-injective module is
an M-injective hull of some simple module in [M]. 2
2.13 -decomposition for semiperfect generators. If M is a
projective generator which is semiperfect in [M], then M has a -
decomposition M =

, where each M

is -indecomposable.
In particular, every semiperfect ring A has a -decomposition A =
Ae
1
Ae
k
, where the e
i
are central idempotents of A which are
not a sum of non-zero orthogonal central idempotents.
9
Proof. By [11, 42.5] and 2.11, M has a decomposition which com-
plements direct summands. By Lemma 2.10, M has a fully invariant
decomposition and the assertions follow from the Corollaries 2.6, 2.7,
and 2.5. 2
3 Coalgebras and comodules
We recall some basic denitions for coalgebras and comodules.
An R-module C is an R-coalgebra if there is an R-linear map (co-
multiplication)
: C C
R
C, with (id ) = ( id) .
An R-linear map : C R is a counit if (id) and (id)
yield the canonical isomorphism C C
R
R.
Henceforth C will denote a coalgebra with counit (C, , ) and we
assume that C is at as an R-module.
The R-dual of C, C

= Hom
R
(C, R), is an associative R-algebra
with unit where the multiplication of f, g C

is dened by
(f g)(c) = (f g)((c)), for c C.
An R-submodule D C is a left coideal if (D) C
R
D, a right
coideal if (D) D
R
C, and a sub-coalgebra if (D) D
R
D and
D is pure in C.
An R-module M is called a right C-comodule if there exists an R-
linear map : M M
R
C such that (id ) = ( id) ,
and (id ) yields the canonical isomorphism M M
R
R. An
R-submodule N M is called C-sub-comodule if (N) N
R
C.
Left C-comodules are dened similarly. Clearly C is a right and
left C-comodule, and right (left) sub-comodules of C are right (left)
coideals.
An R-linear map f : M M

between right comodules is a comod-

ule morphism provided

f = (f id) .
The right (left) C-comodules and the comodule morphisms form a
category which we denote by Comod-C (C-Comod). These are Grothen-
dieck categories (remember that we assume
R
C to be at). The close
connection between comodules and modules is based on the following
facts which are proved in [12, Section 3,4].
10
3.1 C-comodules and C

-modules. Assume
R
C to be projective and
let : M M
R
C be any right C-comodule. Then M is a left C

-
module by
: C

R
M M, f m ((idf) )(m).
(1) An R-submodule U M is a sub-comodule if and only if it is a
C

-submodule.
(2) Any R-linear map between right comodules is a comodule mor-
phism if and only if it is C

-linear.
(3) The category of right C-comodules can be identied with [
C
C],
the full subcategory of C

-Mod, subgenerated by
C
C.
(4) C is a balanced (C

, C

)-bimodule and the subcoalgebras of C cor-

respond to the (C

, C

)-sub-bimodules.
The properties of the comodule C are strongly inuenced by the
properties of the ring R (see [12, 4.9]).
3.2 Coalgebras over special rings. Let
R
C be projective.
(1) If R is noetherian, then C is a locally noetherian C

-module and
direct sums of injectives are injective in [
C
C].
(2) If R is artinian, then every nitely generated module in [
C
C]
has nite length.
(3) If R is injective, then C is injective in [
C
C].
Applying our results on decompositions of closed subcategories we
obtain
3.3 -decomposition of coalgebras. Let C be a coalgebra over a
noetherian ring R with C
R
projective.
(1) There exist a -decomposition C =

, and a family of or-

thogonal central idempotents e

in C

, with C

= C e

, for
each .
(2) [
C
C] =

[
C
C

].
(3) Each C

is a sub-coalgebra of C, C

, and [
C
C

] =
[
C

].
11
(4) [
C
C] is indecomposable if and only if, for any two injective
uniform L, N [
C
C], we have L N.
(5) Assume R to be artinian. Then [
C
C] is indecomposable if and
only if for any two simple E
1
, E
2
[
C
C], we have

E
1


E
2
.
Proof. (1),(2) By 3.2, C is a locally noetherian C

-module. Now the

decomposition of
C
[C] follows from 2.12. Clearly the resulting -
decomposition of C is a fully invariant decomposition and hence it can
be described by central idempotents in the endomorphism ring (= C

,
see 3.1).
(3) The fully invariant submodules C

C are in particular R-
direct summands in C and hence are sub-coalgebras (by [12, 4.4]). It
is straightforward to verify that Hom
R
(C

, R) = C

is an
algebra isomorphism.
(4) is a special case of 2.12(2).
(5) follows from 2.12(3). Notice that

E
1


E
2
can be described by
extensions of simple modules (see 2.9). The assertion means that the
Ext-quiver of the simple modules in
C
[C] is connected. 2
3.4 Corollary. Let R be a QF ring and C an R-coalgebra with C
R
projective. Then:
(1) C has fully invariant decompositions with -indecomposable sum-
mands.
(2) Each fully invariant decomposition is a -decomposition.
(3) C is -indecomposable if and only if C has no non-trivial fully
invariant decomposition.
(4) If C is cocommutative then C =

is a fully invariant de-

composition, where E

is a minimal representing set of simple

modules.
Proof. By [12, 6.1], C is an injective cogenerator in
C
[C] and so the
assertions (1)-(3) follow from 2.6 and 3.3.
(4) Our assumption implies that C

is a commutative algebra and

so the assertion follows by 2.8. 2
12
For coalgebras C over QF rings we have a bijective correspondence
between closed subcategories of [
C
C] and (C

, C

)-submodules in C.
However the latter need not be pure R-submodules of C and hence they
may not be sub-coalgebras.
3.5 Correspondence relations. Let R be a QF ring and C an R-
coalgebra with C
R
projective. Then
(1) for every N [
C
C], [N] = [Tr(N, C)];
(2) the map [N] Tr(N, C) yields a bijective correspondence be-
tween the closed subcategories of [
C
C] and the (C

, C

)-submod-
ules of C;
(3) [N] is closed under essential extensions (injective hulls) in [
C
C]
if and only if Tr(N, C) is a C

-direct summand of
C
C. In this
case Tr(N, C) is a sub-coalgebra of C.
(4) N [
C
C] is semisimple if and only if Tr(N, C) Soc
C
(Q);
(5) If R is semisimple, then all Tr(N, C) are sub-coalgebras of C.
Proof. Since R is a QF ring,
C
C has locally nite length and is an
injective cogenerator of [
C
C]. Hence (1)-(4) follow from 2.2.
(5) For R semisimple all (C

, C

)-submodules Tr([N], C) are direct

summands as R-modules in C and so they are sub-coalgebras by [12,
4.4]. 2
Remarks. 3.3 and 3.4 extend decomposition results for coalgebras
over elds to coalgebras over noetherian (QF) rings. It was shown in
Kaplansky [6] that any coalgebra C over a eld K is a direct sum of
indecomposable coalgebras, and that for C cocommutative, these com-
ponents are even irreducible. In Montgomery [7, Theorem 2.1], a direct
proof was given to show that C is a direct sum of link-indecomposable
components. It is easy to see that the link-indecomposable components
are just the -indecomposable components of C (see remark in proof
of 3.3(5)). As outlined in [7, Theorem 1.7] this relationship can also be
described by using the wedge. In this context another proof of the de-
composition theorem is given in Shudo-Miyamoto [9, Theorem]. These
techniques are also used in Zhang [13]. We refer to Garca-Jara-Merino
[3, 4] for a detailed description of the corresponding constructions.
13
In Green [5], for every C-comodule M, the coecient space C(M)
was dened as the smallest sub-coalgebra C(M) C such that M is a
C(M)-comodule. The denition heavily relies on the existence of a K-
basis for comodules. In the more general correspondence theorem 3.5,
the C(M) are replaced by Tr(M, C). For coalgebras over elds, C(M)
and Tr(M, C) coincide and 3.5 yields [5, 1.3d], [4, Proposition 7], and
[7, Lemma 1.8]. Notice that in [5] closed subcategories in [
C
C] are
called pseudovarieties.
References
[1] Anderson, F., Fuller, K., Rings and Categories of Modules,
Springer (1973)
[2] Dung, N.V, Huynh, D.V., P.F. Smith, P.F., Wisbauer, R., Extend-
ing Modules, Pitman Research Notes 313, London (1994)
[3] Garca, J.M., Jara, P., Merino, L.M., Decomposition of locally
nite modules, Rings, Hopf algebras, and Brauer groups, ed.
Caenepeel, S., Verschoren, A., Marcel Dekker LNPAM 197 (1998)
[4] Garca, J.M., Jara, P., Merino, L.M., Decomposition of comodules,
Comm. Algebra 27(4), 1797-1805 (1999)
[5] Green, J.A., Locally nite representations, J. Algebra 41, 137-171
(1976)
[6] Kaplansky, I., Bialgebras, Lecture Notes in Math., University of
Chicago (1975)
[7] Montgomery, S., Indecomposable coalgebras, simple comodules, and
pointed Hopf algebras, Proc. Amer. Math. Soc. 123, 2343-2351
(1995)
[8] Nastasescu, C., Torrecillas, B., Colocalization on Grothendieck cat-
egories with applications to coalgebras, J. Algebra 185, 108-124
(1996)
[9] Shudo, T., Miyamoto, H., On the decomposition of coalgebras,
Hiroshima Math. J. 8, 499-504 (1978)
14
[10] Vanaja, N., All nitely generated M-subgenerated modules are ex-
tending, Comm. Alg. 24(2), 543-572 (1996)
[11] Wisbauer, R., Foundations of Module and Ring Theory,
Gordon and Breach, Reading, Paris (1991)
[12] Wisbauer, R., Semiperfect coalgebras over rings, Algebras and
Combinatorics. An International Congress ICAC97, Hongkong,
Springer Singapore (1999)
[13] Zhang, Shouchuan, Relation between decomposition of comodules
and coalgebras, Science in China (Series A) 40(1), 24-29 (1997)
Address:
Department of Mathematics
Heinrich-Heine-University
40225 D usseldorf, Germany
e-mail: [email protected]
15