§ 11. Grothendieck groups
1. Additive Functions of Modules
Let A be a ring, and let be a set of classes of A-modules (VIII, p. 51); we say that an A-module is of type if its class belongs to .
Definition 1 00CS
We say that the set of classes of A-modules is additive if every zero module is of type and the direct sum of two modules of type is of type . We say that is hereditary if it is additive and the submodules and quotient modules of a module of type are of type .
Example 1 00CT
The set of classes of A-modules of finite length is hereditary (II, §1, No. 10, p. 212, Proposition 16).
Example 2 00CU
The set of classes of finitely generated A-modules is additive. If the ring A is Noetherian, then this set is hereditary (VIII, p. 3, Proposition 3 and VIII, p. 7, Proposition 4).
Example 3 00CV
The set of classes of finitely generated projective A-modules is additive but generally not hereditary.
Definition 2 00CW
Let be a mapping from to an abelian group G (written additively); set (cl(E)) for every A-module E of type . We say that is an additive function of modules (resp. a weakly additive function of modules) if we have for every exact sequence (resp. for every split exact sequence)
of modules of type .
Example 4 00CX
Let be the set of classes of A-modules of finite length. The mapping that sends a class of A-modules of finite length to its length is an additive function of modules (II, §1, No. 10, p. 213, Corollary 3). The results of this subsection are a generalization of the results on modules of finite length established in II, §1, No. 10, p. 212–214.
In the remainder of this subsection, we consider an additive set of A-modules and an additive mapping from to an abelian group .
Let E and be modules of type ; then is of type , and there exists a split exact sequence (II, §1, No. 9, p. 210)
from this, we deduce
(1)
In particular, we have .
Proposition 1 00CY
Suppose that is hereditary. Let E and F be A-modules and a linear mapping. a) If E or F is of type , then so is the image of u. b) If E is of type , then so is the kernel of u, and we have
(2)
c) If F is of type , then so is the cokernel of u, and we have
(3)
The proposition follows from the existence of exact sequences
Corollary 00CZ
Let be a finite sequence of modules of type . If there exists an exact sequence
then we have
(4)
Let us prove the corollary by induction on ; the cases and are trivial. So let , and let
be an exact sequence of modules of type . By Proposition 1, the kernel F of is a module of type , and we have
Moreover, we have an exact sequence
and the induction hypothesis gives the relation
We then immediately deduce from (5) and (6) that , and the corollary follows.
Proposition 2 00D0
Suppose that the set is hereditary. Let E be an A-module, and let M and N be submodules of E.
a) If the modules M and N are of type , then so are the modules and M + N, and we have
b) If the modules and are of type , then so are the modules and , and we have
Assertions a) and b) follow from the existence of the exact sequences
and
(II, §1, No. 7, p. 207, Proposition 10).
Proposition 3 00D1
Suppose that is hereditary. Let E be a module of type and a composition series of E (I, §4, No. 7, p. 41). For , the module is of type , and we have
Since is hereditary, the modules and are of type , and we have . Since the sequence is a composition series of , Proposition 3 follows by induction on .
2. The Grothendieck Group of an Additive Set of Modules
Let A be a ring. In this subsection, we consider an additive set of classes of A-modules; we identify with the canonical basis of the free abelian group . For any exact sequence
of modules of type , we denote by the element cl(E) cl(E) cl(E) of . Let R be the subgroup of generated by the elements of the form ; the quotient group is called the Grothendieck group of and is denoted by . For a module E of type , we denote the image of cl(E) in by (or sometimes [E] when there is ambiguity about ). We then have the following universal property.
Proposition 4 00D2
a) The mapping from to is additive.
b) Let G be an abelian group, and let be an additive function of modules. There exists a unique homomorphism such that we have for every module E of type .
Assertion a) is obvious. Let us prove b). There exists a homomorphism that extends . Since is additive, we have for every exact sequence of modules of type ; therefore, R is contained in the kernel of . Consequently, when passing to the quotient, defines a homomorphism from to G, and it is clear that we have for every A-module of type . The group is generated by the set of elements for E running through ; the uniqueness of follows.
Let and be additive sets of classes of A-modules such that . Since the mapping fromn to the Grothendieck groupnK() is additive, there exists a homomorphism , called canonical, characterized by the formula for every module E of type . It is not always injective (VIII, p. 210, Exercise 13).
Example 00D3
Let A be a Noetherian commutative ring, and let Σ be its spectrum. For any integer , denote by the set of classes of finitely generated A-modules with support of codimension in Σ. Set and . We have a sequence of homomorphisms
We can prove (AC, VIII, §1, n, p. 13, proposition 10) that in , the elements , where runs through the set of elements of Σ of height , form a basis of a -module supplementary to the image of . More precisely, for every module E of type , we have
long (mod Im )
ht()
The group is generated by the elements of the form with , and we have by relation (1) (VIII, p. 184). Every element of is therefore of the form , where E and F belong to .
An element of is called effective if it is of the form for an A-module E of type . The set of effective elements of is denoted by ; it is a submonoid of , and can be identified with the group of differences of (I, §2, No. 4, p. 20).
Proposition 5 00D4
Let E and F be modules of type . The equality holds if and only if there exist exact sequences of modules of type
such that is isomorphic to .
The stated condition is sufficient because it implies the relations
which give .
Now suppose that we have . By the construction of the group , there exist two finite families of exact sequences of modules of type
for and
for such that in , we have
cl(E) cl(F) .
More explicitly, this relation can be written as
(7) cl(E) cl(G) cl(H) cl(H)
= cl(F) cl(G) cl(G) cl(H).
Set , etc. By passing to the direct sums, we obtain the exact sequences
0 // // G // // ,
0 // // H // // 0
consisting of modules of type .
Next, let be A-modules of type . If we have cl(M) cl(N) in the group , then we have and there exists a permutation such that cl(M) cl(N) for every (I, §7, No. 9, p. 95, Proposition 11). Consequently, the modules and are isomorphic. In particular, from (7), we deduce the existence of an isomorphism from to , where we have set
We also set
The modules L, M, P, Q are of type , and the sequence
0 // L // P // M // ,
where we define and by
is exact. We construct an exact sequence
in the same way; this concludes the proof.
The set is a commutative monoid for the law of composition cl(E). We sometimes denote the group of differences of the commutative monoid (I, §2, No. 4, p. 20) by and call it the Grothendieck group of for direct sums. For any module E of type , we denote the image of cl(E) in by .
Proposition 6 00D5
a) The mapping from to is a weakly additive function of modules.
b) Let G be an abelian group, and let be a weakly additive function of modules. There exists a unique group homomorphism G such that we have for every module E of type .
c) Let E and F be modules of type . We have if and only if there exists a module M of type such that is isomorphic to .
Assertion a) is obvious. Assertion b) follows from (I, §2, No. 4, p. 19, Theorem 1) and assertion c) from (I, §2, No. 4, p. 18, Proposition 6).
Since the mapping from to is a weakly additive function of modules, we can deduce a homomorphism from it. This homomorphism is surjective but is not always an isomorphism (VIII, p. 191, Remark 2).
Let be the subgroup of generated by the elements of the form , where is a split exact sequence of A-modules of type , that is, by the elements of the form cl(E) cl(E) cl(E), where and are modules of type . The canonical mapping from to the quotient group extends to a group homomorphism . This is an isomorphism. Indeed, the canonical mapping from to extends to a group homomorphism from to whose kernel contains , and therefore to a homomorphism by passing to the quotient; it is clear that and are inverse bijections.
An element of is called effective if it is of the form for an A-module E of type . The set of effective elements of is denoted by .
3. Using Composition Series
Let A be a ring. Let E be an A-module of finite length and S a simple A-module. By the Jordan–Hölder theorem (I, §4, No. 7, p. 43, Theorem 6), the number of quotients of a Jordan–Hölder series of E that are isomorphic to S does not depend on the sequence. We denote it by and call it the multiplicity of S in E. The support of the A-module E is the set of classes of simple A-modules S such that . When E is semisimple and has finite length, the integer is the length [E : S] of the isotypical component of E of type S (VIII, p. 72), and the notion of support coincides with that introduced in VIII, p. 66.
Lemma 1 00D6
Let E, , and be A-modules of finite length and
0 // // E // // 0
an exact sequence. We have .
Let and be Jordan–Hölder series of and , respectively. There exists a Jordan–Hölder series Σ of E whose sequence of quotients can be obtained by juxtaposing the sequence of quotients of Σ and that of (I, §4, No. 8, p. 44).
Proposition 7 00D7
Let be a hereditary set of classes of modules such that every module of type has finite length. Let be the set of classes of simple modules belonging to . Then the family is a basis of the -module , and we have
for every module E of type .
Formula (8) follows from Proposition 3 applied to a Jordan–Hölder series of E. By Lemma 1, for every element S of , there exists a -linear mapping from to such that for every module E of type . In particular, we have and for every in . It follows that the elements of the form (for ) are linearly independent over ; these elements generate by formula (8).
Corollary 00D8
Let E and F be semisimple modules of type . The module E is isomorphic to F if and only if we have in .
Indeed, we have and an analogous formula for F, and E is isomorphic to F if and only if we have for every (VIII, p. 72).
Remark 00D9
The set is the submonoid of generated by the family .
4. The Grothendieck Group R(A)
Let A be a ring. Let be the set of classes of finitely generated A-modules (VIII, p. 51). The classes of the A-modules of finite length form a subset of ; we have seen that is a hereditary set of classes of modules. We denote the Grothendieck group associated with by R(A) and the image in R(A) of the class of an A-module E of finite length by [E].
The results of Subsection 3 imply the following:
a) Let be the set of classes of simple A-modules. The family is a basis of the -module R(A).
b) Let E and F be semisimple A-modules of finite length. Then E and F are isomorphic if and only if we have [E] = [F] in R(A).
c) Let E be an A-module of finite length and a Jordan–Hölder series of E. Set . Then F is a semisimple A-module of finite length, and we have [E] = [F] in R(A).
d) Let be the homomorphism characterized by for every simple A-module S. We then have long for every A-module E of finite length.
If D is a field, then the homomorphism is an isomorphism.
Remark 1 00DA
Let be the hereditary set of classes of semisimple A-modules of finite length. By Proposition 7 of VIII, p. 190, the Grothendieck group is a free -module whose elements (for ) form a basis. The canonical homomorphism (VIII, p. 186) is therefore an isomorphism.
Remark 2 00DB
Let be the Grothendieck group of for direct sums (VIII, p. 188). The Krull–Remak–Schmidt theorem (VIII, p. 37) implies that is a free -module with basis the set of classes of indecomposable A-modules of finite length, while admits the set of classes of simple A-modules as a basis.
Remark 3 00DC
Let E be an A-module of finite length. By c), there exists a semisimple A-module of finite length such that , and by b), such a module is defined up to an isomorphism; we sometimes call it a semisimplification of E.
Proposition 8 00DD
Let A be a principal ideal domain that is not a field, and let L be its field of fractions. There exists an isomorphism such that we have
for every in A.
Let P be a system of representatives consisting of irreducible elements of A (VII, §1, No. 3, p. 3). The maximal ideals of A are the ideals for ; every simple A-module is therefore isomorphic to a unique module . Moreover (VII, §1, No. 3, p. 4, Theorem 2), the abelian group is free and has the family as a basis. Hence there exists an isomorphism from R(A) to characterized by for every .
Let in A. There exist an integer , elements of P, and an element of such that we have . The module admits the composition series defined by
and the module is isomorphic to . We therefore have (VIII, p. 185, Proposition 3)
Remark 4 00DE
We keep the assumptions and notation of Proposition 8. Let E be an A-module of finite length, and let be its invariant factors (VII, §4, No. 5, p. 20). Since E is isomorphic to , we have
Remark 5 00SI
Let A be a Dedekind ring that is not a field (Comm. Alg., VIII, §2, No. 1). By reasoning as in Proposition 8, we can prove the existence of an isomorphism from R(A) to the group of fractional ideals of A, characterized by for every nonzero ideal of
Proposition 8 will be used, for example, in the following two cases:
a) Suppose that we have . The -modules of finite length are simply the finite abelian groups. Since is the direct product of and , we can deduce from Proposition 8 an isomorphism from to given by
Card(G)
for every finite abelian group G.
b) Suppose that A is the polynomial ring K[T] in one variable T with coefficients in a commutative field K. Let E be a finite-dimensional vector space over K and an endomorphism of E. As in VII, §5, No. 1, p. 29, denote by the A-module with underlying additive group E and external law . The A-module has finite length. Conversely, every simple A-module is finite-dimensional over K (VII, §4, No. 8, p. 25, Remark 4). Consequently, every A-module of finite length is finite-dimensional over K, hence of the form . Moreover (VII, §5, No. 3, p. 32, Corollary 1), the product of the invariant factors of is equal to the characteristic polynomial of . Consequently, Proposition 8 gives an isomorphism
characterized by (cf. formula (10)).
5. Change of Rings
Let A and B be rings, and let be a ring homomorphism. Let be an additive set of A-modules and an additive set of B-modules.
First, suppose that for every B-module M of type , the A-module obtained by restricting the ring of scalars to A is of type . Then the mapping from to that sends M to is an additive function of modules; from it, we deduce a group homomorphism
We define a group homomorphism likewise.
We now suppose that for every A-module E of type , the B-module derived from E by extension of scalars via (II, §5, No. 1, p. 277) is of type . The mapping from to that sends an element E of to is a weakly additive function of modules; it therefore induces a group homomorphism .
Suppose, furthermore, that for every exact sequence
of A-modules of type , the sequence of B-modules of type
is exact. This holds, in particular, in the following cases:
a) The homomorphism makes B into a projective A-module (II, §3, No. 6, p. 251, Proposition 5 and II, §3, No. 7, p. 257, Corollary )or, more
generally, a flat (X, §1, n, p. 8, définition 1) A-module.
b) The set is a set of classes of projective or, more generally, flat (X,
§4, n, p. 72, corollaire 2) A-modules.
The mapping from to that sends E to is then additive. It therefore induces a group homomorphism .
6. The Grothendieck Group
Let K be a commutative field and A a K-algebra. The set of classes of A-modules that are finite-dimensional vector spaces over K is hereditary. The corresponding Grothendieck group is denoted by . It is a free -module with basis the family , where is the set of classes of simple A-modules that are finite-dimensional over K. There exists a homomorphism
dim
characterized by dim([E]) = [E : K] for every A-module E that is finite-dimensional over K. When A = K, it is an isomorphism. The submonoid of effective elements is denoted by .
Lemma 2 00DF
Let M and be A-modules that are finite-dimensional vector spaces over K. The supports (VIII, p. 190) of M and are disjoint if and only if there exists an such that and .
Suppose that there exists an such that and . Let S be a simple A-module. If cl(S) belongs to the support of M, then the A-module S is isomorphic to one of the quotients of a Jordan–Hölder series of M and we have . Likewise, if cl(S) belongs to the support of M’, then we have . It follows that and are disjoint.
Conversely, suppose that the sets and are disjoint. They are finite because M and are finite-dimensional over K. Every simple A-module S whose class belongs to is finite-dimensional over K and a fortiori over the field End. By Corollary 1 of Proposition 4 (VIII, p. 83), there exists an element such that we have (resp. ) for every simple A-module S whose class belongs to (resp. ). Let be a Jordan–Hölder series of M. By the above, we have for and therefore . The existence of a natural number such that is proved in the same way. Set and . The polynomial P(X) is a multiple of , so we have , while the polynomial is a multiple of , so we have . This concludes the proof.
Let L be an extension of K. If M is an A-module that is finite-dimensional over K, then is an -module that is finite-dimensional over L. Moreover, for every exact sequence
of A-modules, the sequence of -modules
deduced from it by extension of scalars is exact (II, §7, No. 7, p. 308, Proposition 14). Hence there exists a unique ring homomorphism such that for every A-module M that is finite-dimensional over K (No. 5).
Theorem 1 00DG
The homomorphism defined above is injective. Let be an element of . Then is effective if and only if is.
Lemma 3 00DH
Let S and T be two nonisomorphic simple A-modules that are finite-dimensional over K. The supports of the -modules and are disjoint.
By Lemma 2, there exists an element such that and . The element of acts as 0 on and as 1 on . By Lemma 2, the supports of the -modules and are disjoint.
Let us prove Theorem 1. Let be the set of classes of simple A-modules that are finite-dimensional over K. The family is a basis of the -module . Let . The -module is not zero, so its support is not empty. Let be an element of this support. By Lemma 1 of VIII, p. 190, there exists a homomorphism such that for every -module E that is finite-dimensional over L. We have by construction and for every by Lemma 3. We have therefore proved that the elements of of the form for are linearly independent over . It follows that the homomorphism is injective.
Let , and let be an element of the support of . For every (A), the coordinate of of index [S] in the basis is . It follows that if is effective, then so is .
7. Multiplicative Structure on
Let K be a commutative ring and A a bigebra over the ring K (III, §11, No. 4, p. 585), with coproduct and counit . Unless stated otherwise, the tensor products are over K. Let E and F be (left) A-modules. The tensor product is endowed with an A)-module structure characterized by the formula
for , and . Using the homomorphism , we deduce from this A)-module an A-module (II, §1, No. 13, p. 221). More precisely, let ; if , then we have
for and . By abuse of notation, we also denote the resulting A-module by . It follows immediately from the coassociativity of that the canonical isomorphism of K-modules
is A-linear for all A-modules E, F, and G. Likewise, if the bigebra A is cocommutative, then the canonical isomorphism from to is A-linear. Finally, let be the A-module with underlying group K and external law . The canonical isomorphism from (resp. ) to E is an isomorphism of A-modules from (resp. ) to E.
Proposition 9 00DI
Let K be a commutative ring, A a bigebra over K with counit , and an additive set of classes of A-modules with the following properties:
(i) Every A-module of type is a projective or, more generally, flat K-module.
(ii) If the A-modules E and F are of type , then so is the A-module .
(iii) The A-module defined above is of type . Then there exists a unique ring structure on the additive group whose multiplication satisfies
for all A-modules E and F of type . The unit element of is . If the bigebra A is cocommutative, then the ring is commutative.
Endowed with the law of composition given by cl(E ), the set is a monoid with unit element cl(K). Consequently, is canonically endowed with the structure of a ring with multiplication characterized by the formula
(13) cl(E) cl(F) = cl(E )
and unit element cl(K) (III, §2, No. 6, p. 446).
Given an A-module F of type and an exact sequence
of A-modules of type , the sequence
deduced from is exact because F is projective or, more generally, flat over K (II, §3, No. 6, p. 251, Proposition 5 and II, §3, No. 7, p. 257, Corollary 6). In the notation of No. 2, we then have
(14) cl(F).
It follows that the subgroup R of generated by the elements of the form is a right ideal of the ring , and one proves likewise that it is a left ideal of . By definition, is the quotient group ; hence there exists a unique ring structure on whose multiplication satisfies for all A-modules E and F of type . Its unit element is .
When the bigebra A is cocommutative, the monoid is commutative; it follows that the ring and the quotient ring are commutative.
Remark 00DJ
Under only assumptions (i) and (ii) of Proposition 9, the Grothendieck group for direct sums (VIII, p. 188) has a unique ring structure whose multiplication satisfies . Its unit element is . The ring is commutative if the bigebra A is cocommutative. The proof is analogous to that of Proposition 9 because the group can be identified with , where is the subgroup of generated by the elements of the form cl(E) cl(E) cl(E) (loc. cit.).
Under assumptions (i), (ii), and (iii) of Proposition 9, the ring is called the Grothendieck ring of . These conditions are, in particular, satisfied when K is a field and is the set of classes of A-modules that are finite-dimensional over K. Consequently, we have the following.
Corollary 00DK
Let A be a bigebra with counit over a commutative field K. Then there exists a unique ring structure on the additive group whose multiplication satisfies
for all A-modules E and F that are finite-dimensional over K. The unit element of is . If the bigebra A is cocommutative, then the ring is commutative.
Example 1 00DL
Let K be a commutative field. Let G be a group, and let K[G] be the algebra of the group G. We identify G with its canonical image in K[G] (III, §2, No. 6, p. 446).
We endow K[G] with the structure of a bigebra with coproduct and counit given by
Let E and F be K[G]-modules; by formula (12), the K[G]-module structure on is given by
The K[G]-module is the vector space K endowed with the action of G defined by for and .
The ring is also denoted by . It is commutative; its multiplication is given by , and the unit element of is .
Example 2 00DM
Let be a Lie algebra over a commutative field K and its enveloping algebra; we identify with its canonical image in Lie, I, §2, No. 7, p. 39, Corollary 2). We endow with the structure of a bigebra with coproduct and counit given by
for Lie, II, §1, No. 4, p. 115).
Let E and F be )-modules; by formula (17), the )-module structure on is characterized by
for , and .
The Grothendieck ring is also denoted by in Lie, VIII,
§7, No. 6, p.
Let A be a commutative ring. We can view A as a bigebra that is cocommutative over itself, with coproduct the natural isomorphism from A to and counit Id (III, §11, No. 4, p. 585). By Proposition 9, we obtain the following result.
Proposition 10 00DN
Let A be a commutative ring, and let be an additive set of classes of A-modules satisfying the following three conditions:
(i) Every A-module of type is projective or, more generally, flat.
(ii) If E and F are A-modules of type , then the A-module is also of type .
(iii) The A-module A is of type .
Then there exists a unique ring structure on the additive group satisfying for every pair E, F of A-modules of type . The unit element of is .
8. The Grothendieck Group
Let A be a ring. The set of classes of finitely generated projective A-modules is additive; we denote the Grothendieck group by .
For the law of composition cl(E ), the set is a commutative monoid. Moreover, every exact sequence of projective A-modules
splits (II, §2, No. 2, p. 231, Proposition 4), so that E is isomorphic to . The mapping from to therefore defines an isomorphism from the group of differences of the monoid to (VIII, p. 188).
For every finitely generated projective module P, there exists a finitely generated projective module such that is free (II, §2, No. 2, p. 232, Corollary 1). Let E and F be finitely generated projective A-modules; by I, §2, No. 4, p. 18, Proposition 6, we have [E] = [F] in if and only if there exists a finitely generated free A-module L such that the A-modules and are isomorphic. We then say that E and F are stably isomorphic; this does not necessarily imply that E and F are isomorphic (VIII, p. 207, Exercise 2 and VIII, p. 210, Exercise 14).
When the ring A is commutative, there exists a commutative ring structure on the additive group whose multiplication is characterized by the formula (VIII, p. 199, Proposition 10).
Remark 00DO
Let A be a semisimple ring. Every A-module is then semisimple and projective (VIII, p. 138, Proposition 4); we therefore have the equality
(cf. No. 4 for the definitions of and ). We therefore have by the definitions of these Grothendieck groups.
Example 00DP
If every finitely generated projective A-module is free, then the rank defines an isomorphism from to . This is, in particular, the case when A is a principal ideal domain (VII, §1, No. 3, p. 15, Corollary 3) or when A is a local ring (VIII, p. 36, Corollary 6).
9. The Grothendieck Group of an Artinian Ring
Let A be a left Artinian ring. Let be its radical; it is a nilpotent two-sided ideal of A, and the ring is semisimple (VIII, p. 173, Proposition 1). By the corollary of VIII, p. 176, the mapping cl(P) is an isomorphism from the monoid to the monoid . We deduce from it a group isomorphism from to characterized by the relation for every finitely generated projective A-module P.
Since the ring is semisimple, the remark above implies the equality . The modules of finite length over the ring are simply the semisimple modules of finite length over the ring A (VIII, p. 174, Proposition 2); consequently, we can identify with and with . We denote by the homomorphism from to (VIII, p. 191, Remark 1); it is an isomorphism. Finally, we have (A), and we set \gamma_{\mathscr{LF}(A),\mathscr{P}(A)}. We have defined a diagram
We denote the (finite) set of classes of simple A-modules by ; for every , choose a module of class and a projective cover of (VIII, p. 175, Proposition 4). It follows from Proposition 6 of VIII, p. 176 that is a free -module with basis the family . Moreover, since is isomorphic to (VIII, p. 176), transforms the basis of into the basis of . The isomorphism transforms the basis of into the basis of .
The Cartan matrix of is the matrix of the homomorphism of -modules with respect to bases of and of . By definition, we have
in the group . In other words, is the number of quotients isomorphic to in a Jordan–Hölder series of the A-module .
Set ; it is an endomorphism of the group . If M is a finitely generated semisimple A-module and a projective cover of M, then we have . By formula (19), the matrix of with respect to the basis of is simply the Cartan matrix of .
10. Change of Rings for
Let A and B be rings. Let be a ring homomorphism. If P is a finitely generated projective A-module, then the B-module is projective and finitely generated (II, §5, No. 2, p. 281, Corollary). The mapping is a homomorphism from the monoid to the monoid and therefore defines a homomorphism characterized by the relation for every finitely generated projective A-module P. If is a second ring homomorphism, then it follows from the transitivity of the extension of scalars (II, §5, No. 1, p. 278, Proposition 2) that the homomorphisms and from to are equal.
Suppose that makes B into a finitely generated projective left A-module. Let Q be a finitely generated projective left B-module. Then Q is a direct factor of a finitely generated free B-module that is projective and finitely generated over A. Consequently, the A-module obtained from Q by restriction of scalars is projective and finitely generated. As above, we deduce a homomorphism characterized by the relation for every finitely generated projective B-module Q. If is a ring homomorphism that makes C into a finitely generated projective B-module, then the homomorphisms and from to are equal.
11. Frobenius Reciprocity
Let A be a semisimple ring. Let be a homomorphism from A to a semisimple ring B. Let S be a simple A-module and T a simple B-module, and let D and E be the centralizers of S and T, respectively. By Schur’s lemma (VIII, p. 47, Corollary), D and E are fields. Let H be the set of A-linear homomorphisms from S to . We endow H with an , D)-bimodule structure with external laws and for , .
Proposition 11 00DQ
a) The multiplicity of the simple A-module S in the semisimple A-module is equal to the dimension of H viewed as a right vector space over D.
b) The multiplicity is finite and equal to the dimension of H viewed as a left vector space over E.
Assertion a) follows from formula (11) of VIII, p. 72.
The B-module is semisimple and finitely generated, hence has finite length. By formula (12) of loc. cit., we have
(20) dim Hom.
Now, in II, §5, No. 1, p. 277–278, formula (2) and Remark 2, we defined an E-linear bijection from Hom to Hom. Assertion b) then follows from formula (20).
Corollary (Frobenius reciprocity) 00S7
Suppose that A and B are semisimple algebras that are finite-dimensional over a commutative field K and that is K-linear. Then the K-vector spaces S, T, D, E, and H are finite-dimensional, and we have the equalities
In particular, when K is algebraically closed, we have D = E = K and
Since A-module S is simple, it is monogenous, and D is a linear subspace of Hom; hence S and D are finite-dimensional over K. For an analogous reason, T and E are finite-dimensional over K. Finally, H is a linear subspace of Hom; it is therefore also finite-dimensional. Formula (21) then follows from Proposition 11 because the dimension of H over K is equal to [H : D][D : K] and to [H : E][E : K] (II, §1, No. 13, p. 222, Proposition 25). By Theorem 1 of VIII, p. 47, if K is algebraically closed, we have D = E = K. The second part of the corollary then follows from the first.
Let A and B be semisimple algebras that are finite-dimensional over a commutative field K, and let be a homomorphism of K-algebras from A to B.
Let be the set of classes of simple A-modules; for every (A), let be a module of class and its centralizer. Then is an algebra of finite degree over K; we denote this degree by . We define (B), , , and for in analogously. The Grothendieck group has the family as a basis, and has as a basis. Let be the matrix of and the matrix of with respect to these bases. By definition, we have
for in and in . We denote the dimension of the vector space Hom over the field K by . By the corollary above, we have
When the field K is algebraically closed, we have ; consequently, we have
In other words, the matrices of and with respect to the given bases of and are each other’s transposes.
12. The Case of Simple Rings
Let A and B be simple rings and a homomorphism from A to B. Let S be a simple A-module and T a simple B-module. We set
(26) long;
we call the cardinal the index of . When A is a subring of B and is the canonical injection of A into B, we write instead of , and we call this cardinal the index of A in B. We define the height of analogously:
(27) long.
When A is a subring of B and the canonical injection of A into B, we write for , and we call it the height of A in B.
The A-module S is monogenous, hence so is the B-module . It follows that is finite and therefore that it is an integer. Let M be an A-module. Denote its length by ; then M is isomorphic to . Consequently, the B-module is isomorphic to . By the definition of , we therefore have
(28) long long.
The -modules and are free of dimension 1, with respective bases [S] and [T], and we have
Let us take, in particular, ; then is isomorphic to (II, §3, No. 4, p. 249), so
(30) long(B) long(A).
By Wedderburn’s theorem (VIII, p. 120, Theorem 1), there exist integers 1 and and fields D and E such that A is isomorphic to and B to . By formula (30), we have ; in particular, divides .
Let N be an B-module; denote its length by . Then N is isomorphic to , so the A-module is isomorphic to ; by the definition of , we have
(31) long long.
We have seen (VIII, p. 124, Proposition 5) that makes B into a free A-module and that all bases of this module have the same cardinal, denoted by and called the (left) degree of B over A. The A-module is isomorphic to , so has length long(A). By formula (30) and formula (31) applied to the specific case , we therefore have
Now suppose that B is a finitely generated A-module, that is, that is finite. Then is finite by formula (32). We defined (VIII, p. 202) a group homomorphism from to ; we have
Suppose that A and B are finite-dimensional algebras over a commutative field K and that is K-linear. As before, there exist integers and , K-algebras D and E that are fields, and K-algebra isomorphisms from A to and from B to . Set and . We then have the relations
and, by formulas (30) and (32), the relations
When the field K is algebraically closed, we have and therefore and .
Let A, B, and C be simple rings, and let and be homomorphisms. Let S be a simple A-module. The C-modules and are isomorphic; therefore, by formulas (26) and (28), we have
long long.
We can prove the equality likewise. When A is a subring of B and B is a subring of C, and we take the canonical injections for and , these equalities give
Proposition 12 00DR
Let B be a simple ring and A a simple subring of B. Suppose that B is a finitely generated left A-module. Let M be a nonzero finitely generated left B-module. Set End and End. Then is a subring of , the rings and are simple, is a finitely generated left -module, and we have the equalities
By Proposition 4 of VIII, p. 123, the ring is simple, M is an -module of finite length, and we have
long long(A), long long(A).
For the same reasons, the ring is simple, and we have
long long(B), long long(B).
By formulas (31) and (30), we therefore have
long long long(A) long(B) ;
the equality can be established analogously. From this, we deduce
using formula (32). In particular, is a finitely generated left -module.
Exercises
See the exercises for § 11.