Bourbaki

§ 14. Central simple algebras

In this section, K is a commutative field.

1. Central Simple Algebras

Definition 1 00GE

We say that an algebra A over the field K is central if the mapping λ↦λ1\lambda \mapsto \lambda 1 is a bijection from K to the center of A.

A central algebra is not reduced to 0. For every integer n⩾1n\geqslant 1, the matrix K-algebra Mn(K)\mathbf{M}_n(K) is central (VIII, p. 83, Corollary 2) and simple (VIII, p. 120, Theorem 1). More generally, let D be a central K-algebra of finite degree; then Mn(D)\mathbf{M}_n(D) is also central. Let A be a simple ring; its center Z is a field (VIII, p. 121, Corollary 1), and A is therefore a central simple algebra over Z. If the field K is algebraically closed, then a simple algebra of finite degree over K is central (VIII, p. 122, Corollary 3). The opposite algebra of a central simple algebra is central simple.

Remark 1 00GF

Let A and B be K-algebras. If A⊗KBA\otimes_KB is a central simple algebra, then so are A and B (III, §4, No. 4, p. 468, Corollary of Proposition 6 and VIII, p. 221, Proposition 6). Conversely, if A and B are central simple algebras and if one of them has finite degree over K, then A⊗KBA\otimes_KB is a central simple algebra (III, §4, No. 4, p. 468, Corollary of Proposition 6 and VIII, p. 222, Corollary 2).

Remark 2 00GG

Let A be a K-algebra and L an extension of the field K. If the L-algebra A(L)A_{(L)} is central simple, then the K-algebra A is central simple. Conversely, if one of the degrees [A : K] and [L : K] is finite and if A is a central simple K-algebra, then the L-algebra A(L)A_{(L)} is central simple. This follows from Corollary 2 of VIII, p. 222.

Remark 3 00GH

Let A and B be Morita-equivalent K-algebras. The algebra A is central simple if and only if B is (VIII, p. 105, Proposition 6; p. 111, Corollary; and p. 102, Corollary 1).

Remark 4 00GI

In particular, if A is a central simple K-algebra and n⩾1n\geqslant 1, then Mn(A)\mathbf{M}_n(A) is a central simple K-algebra (VIII, p. 102, Example 1).

Theorem 1 00GJ

Let A be a K-algebra of finite degree. The following properties are equivalent:

(i) The algebra A is central simple.

(ii) The algebra A is central and without radical.

(iii) The canonical homomorphism from the K-algebra A⊗KAoA\otimes_KA^o to the K-algebra EndK(A)_K(A) that sends a⊗a′a\otimes a' to the K-linear mapping x↦axa′x\mapsto axa' from A to A is bijective.

(iv) There exist an extension L of the field K and an integer n⩾1n\geqslant 1 such that the L-algebras A(L)A_{(L)} and Mn(L)\mathbf{M}_n(L) are isomorphic.

(v) For every separable closure K′K' of K, there exists an integer n⩾1n\geqslant 1 such that the K′K'-algebras A(K′)A_{(K')} and Mn(K′)\mathbf{M}_n(K') are isomorphic.

(vi) There exist a Galois extension L of the field K of finite degree and an integer n⩾1n\geqslant 1 such that the L-algebras A(L)A_{(L)} and Mn(L)\mathbf{M}_n(L) are isomorphic.

(vii) There exist a K-algebra of finite degree D that is a field with center K and an integer n⩾1n\geqslant 1 such that the algebra A is isomorphic to the algebra Mn(D)\mathbf{M}_n(D).

A ring is simple if and only if it is semisimple and its center is a field (VIII, p. 143, Corollary of Proposition 10). Since A is an algebra of finite degree over the field K, it is a left Artinian ring; it is therefore semisimple if and only if it is without radical (VIII, p. 154, Proposition 4). The equivalence of (i) and (ii) follows.

Set E=A⊗KAoE = A\otimes_KA^o and F = EndK(A)_K(A); denote by φ\varphi the canonical homomorphism from E to F defined by the relation φ(a⊗a′)(x)=axa′\varphi (a\otimes a')(x) =axa' for x,a,a′x, a, a' in A. If the algebra A is central simple, then so are AoA^o and, therefore, E (Remark 1), so φ\varphi is injective. Now, we have [E:K]=[A:K]2=[F:K][E : K] = [A : K]^2= [F : K], so φ\varphi is bijective. Conversely, suppose that φ\varphi is bijective; since the algebra F is central simple (because it is isomorphic to a matrix algebra Mm\mathbf{M}_m(K)), so is E, and therefore so is A (Remark 1). We have thus proved the equivalence of (i) and (iii).

By Remark 4, assertion (vii) implies assertion (i). The converse implication follows from Corollary 3 of VIII, p. 122 and Corollary 2 of VIII, p. 83.

It is clear that (vi) implies (iv), and (iv) implies (i) by Remark 2.

It remains to prove the implications (i)⇒\Rightarrow(v)⇒\Rightarrow(vi). Suppose that A is central simple, and let K′K' be a separable closure of K (V, §7, No. 8, p. 45) . Then A(K′)A_{(K')} is a central simple algebra of finite degree over K′K' (VIII, p. 251, Remark 2). By the corollary of VIII, p. 231, there consequently exist an integer n⩾1n\geqslant 1 and an isomorphism of K′K'-algebras from A(K′)A_{(K')} to Mn(K′)\mathbf{M}_n(K'); observe that the K′K'-algebras Mn(K′)\mathbf{M}_n(K') and Mn(K)(K′)\mathbf{M}_n(K)_{(K')} are isomorphic. By Lemma 4 of VIII, p. 234, there exists a subextension L of K′K', finitely generated over K, such that the L-algebras A(L)A_{(L)} and Mn(K)(L)\mathbf{M}_n(K)_{(L)} are isomorphic. Then L is separable and of finite degree over K, so contained in a subextension L′L' of K′K' that is Galois and of finite degree over K (V, §10, No. 1, p. 57, Proposition 2). The L′L'-algebras A(L′)A_{(L')} and Mn(L′)\mathbf{M}_n(L') are then isomorphic.

Corollary 1 00GK

Let A be a central simple algebra of finite degree over a separably closed field K. There exists an integer n⩾1n\geqslant 1 such that A is isomorphic to the matrix algebra Mn(K)\mathbf{M}_n(K).

Indeed, every Galois extension of K is equal to K; it suffices to apply the equivalence of properties (i) and (v) of Theorem 1.

Corollary 2 00GL

Let A be a central simple algebra of finite degree over K (for example, a field of finite degree over K with center K). There exists an integer n⩾1n\geqslant 1 such that [A:K]=n2[A : K] =n^2.

Let L be an extension of K and nn a strictly positive integer such that the L-algebras A(L)A_{(L)} and Mn(L)\mathbf{M}_n(L) are isomorphic. We have

[A:K]=[A(L):L]=[Mn(L):L]=n2 [A : K] = [A_{(L)}: L] = [\mathbf{M}_n(L) : L] =n^2

In the notation of Corollary 2, the integer nn is called the reduced degree of A.

Remark 5 00GM

Let A be a central simple algebra of finite degree over K whose reduced degree is a prime number ℓ\ell. Then either A is a field, or A is isomorphic to Mℓ(K)\mathbf{M}_{\ell}(K). Indeed, A is isomorphic to an algebra of the form Mn\mathbf{M}_n(D), where D is a field with center K, and we have

ℓ2=[A:K]=n2[D:K] \ell^2= [A : K] =n^2[D : K]

if A is not a field, then n≠1n\not= 1, so n=ℓn=\ell and D = K.

2. Two Lemmas on Bimodules

Let A and B be rings. For any homomorphism ff from B to A, we denote by AfA^f the (B(B, A)-bimodule with underlying right A-module AdA_d and external law for its left B-module structure given by (b,a)↦f(b)a(b, a)\mapsto f(b)a.

Lemma 1 00GN

Let ff and gg be homomorphisms from B to A. The following conditions are equivalent:

(i) The (B,A)(B,A)-bimodules AfA^f and AgA^g are isomorphic.

(ii) There exists an inner automorphism (I, §8, No. 4, p. 102, Example 2) θ\theta of A such that g=θ∘fg=\theta \circ f.

The automorphisms of the right A-module AdA_d are the mappings x↦axx\mapsto ax, where aa is an invertible element of A. Such an automorphism is a B-linear mapping from AfA^f to AgA^g if and only if we have

g(b)ax=af(b)x g(b)ax=af(b)x

for every xx in A and every bb in B. This relation is equivalent to g(b)=af(b)a−1g(b) =af(b)a^{-1} for every bb in B, that is, to g=θ∘fg=\theta \circ f, where θ\theta is the inner automorphism x↦axa−1x\mapsto axa^{-1} of A.

Lemma 2 00GO

Suppose that B is a semisimple ring that is a finitely generated module over its center Z. Let M and N be (B,A)(B,A)-bimodules. Suppose that they have finite length (which is, in particular, the case when they are right A-modules of finite length). If M and N are isomorphic as (Z,A)(Z,A)-bimodules, then they are isomorphic as (B,A)(B,A)-bimodules.

A) First consider the case when B is the endomorphism ring of a vector space S of finite dimension dd over a commutative field L. We then have Z = L; we view S as a (B(B, Z)-bimodule. The ring B is simple, S is a simple B-module, and Z is the commutant of S; every B-module is isotypical of type S (VIII, p. 122, Proposition 2 a)). Let (V,α)(V, \alpha ) (resp. (W,β)(W, \beta )) be a description of the B-module M (resp. N). The set V (resp. W) is endowed with a (Z(Z, A)-bimodule structure such that α\alpha (resp. β\beta ) is an isomorphism of (B(B, A)-bimodules (VIII, p. 64, Remark 2). As (Z(Z, A)-bimodules, M is isomorphic to VdV^d and N to WdW^d, and there exists an isomorphism from the set of (Z(Z, A)-sub-bimodules of V, ordered by inclusion, to that of the (B(B, A)-sub-bimodules of M (loc. cit.). Hence V is a (Z(Z, A)-bimodule of finite length, and so is W. Since the (Z,A)(Z,A)-bimodules VdV^d and WdW^d are isomorphic, the (Z(Z, A)-bimodules V and W are isomorphic by Theorem 2, d) of VIII, p. 37 applied to the ring Z⊗ZAoZ\otimes_{\mathbf{Z}}A^o. Finally, the (B(B, A)-bimodules M and N are isomorphic.

B) Now consider the case when B is a simple ring that is finitely generated as a Z-module. Then Z is a field, and B is a central simple algebra of finite degree over the field Z. By Theorem 1 of VIII, p. 252, there exists an extension Z′Z' of Z of finite degree over Z such that the Z′Z'-algebra B′=B(Z′)B'= B_{(Z')} is isomorphic to the endomorphism algebra of a finite-dimensional Z′Z'-vector space. Set M′=M(Z′)M'= M_{(Z')} and N′=N(Z′)N'= N_{(Z')}. Then M′M' and N′N' are (B′(B', A)-bimodules of finite length; viewed as (Z′(Z', A)-bimodules, M′M' and N′N' are isomorphic. By the case treated in A),M′A), M' and N′N' are isomorphic as (B′(B', A)-bimodules and a fortiori as (B(B, A)-bimodules. Set r=[Z′:Z]r= [Z': Z]. The (B(B, A)-bimodule M′=Z′⊗ZMM'= Z'\otimes_ZM is isomorphic to MrM^r, and, likewise, the (B(B, A)-bimodule N′N' is isomorphic to NrN^r. Since M and N are (B(B, A)-bimodules of finite length, it follows from Theorem 2, d) of VIII, p. 37 that the (B(B, A)-bimodules M and N are isomorphic.

C) Finally, consider the general case, when B is a semisimple ring that is finitely generated as a Z-module. Let S\mathscr{S} be the set of classes of simple B-modules; it is finite (VIII, p. 136, Proposition 1). For any λ∈S\lambda \in \mathscr{S}, denote by MλM_{\lambda} (resp. NλN_{\lambda}) the isotypical component of type λ\lambda of the B-module M (resp. N); it is a (B(B, A)-sub-bimodule of M (resp. N) (Remark, VIII, p. 67). For λ∈S\lambda \in \mathscr{S}, denote the annihilator of the B-module λ\lambda by bλ\mathfrak{b}_{\lambda}, and set Bλ=B/bλB_{\lambda}= B/\mathfrak{b}_{\lambda}; let ZλZ_{\lambda} be the center of BλB_{\lambda}. For λ∈S\lambda \in \mathscr{S}, the (Bλ(B_{\lambda}, A)-bimodules MλM_{\lambda} and NλN_{\lambda} have finite length. We can then identify B with the product of the simple rings BλB_{\lambda} and Z with the product of the ZλZ_{\lambda} (VIII, p. 141, Proposition 8). Moreover, we can identify M with ∏λ∈SMλ\prod_{\lambda\in\mathscr{S}}M_{\lambda} and N with ∏λ∈SNλ\prod_{\lambda\in\mathscr{S}}N_{\lambda}. By assumption, M and N are isomorphic as (Z(Z, A)-bimodules; it follows that for λ∈S,Mλ\lambda \in \mathscr{S}, M_{\lambda} and NλN_{\lambda} are isomorphic (Zλ(Z_{\lambda}, A)-bimodules. By the case treated in B), the (Bλ(B_{\lambda}, A)-bimodules MλM_{\lambda} and NλN_{\lambda} are isomorphic, and so the (B(B, A)-bimodules M and N are isomorphic.

Remark 00GP

It follows from the proof of Lemma 2 that M and N are (Z,A)(Z,A)-bimodules of finite length. Consequently, if B and A are two semisimple rings that are finitely generated modules over their respective centers Z(B) and Z(A), then two (B(B, A)-bimodules of finite length that are isomorphic as (Z(B),Z(A))-bimodules are isomorphic.

3. Conjugacy Theorems

Theorem 2 00GQ

Let B be a semisimple ring and Z its center; suppose that B is a finitely generated Z-module. Let A be a right Artinian ring, and let ff and gg be ring homomorphisms from B to A; write fZf_Z and gZg_Z for the restrictions of ff and gg to Z. The following properties are equivalent:

(i) There exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

(ii) There exists an inner automorphism θ\theta of A such that gZ=θ∘fZg_Z=\theta \circ f_Z.

Since the ring A is right Artinian, AdA_d is a right A-module of finite length (VIII, p. 6, Theorem 1). Hence AfA^f and AgA^g are (B(B, A)-bimodules of finite length. By Lemma 1 (VIII, p. 254), assertion (i) means that AfA^f and AgA^g are isomorphic (B(B, A)-bimodules and assertion (ii) that they are isomorphic (Z,A)(Z,A)-bimodules. The equivalence of (i) and (ii) therefore follows from Lemma 2 (VIII, p. 254).

Corollary 00GR

Let A and B be algebras over the field K. Suppose that B is central simple and of finite degree and that A is right Artinian. Let ff and gg be K-algebra homomorphisms from B to A. There exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

In the notation of Theorem 2, we have Z = K and therefore fZ=1Z=gZf_Z= 1_Z=g_Z.

Theorem 3 (Skolem–Noether) 00S9

Let A and B be simple K-algebras and Z(A) and Z(B) their centers. Suppose that the algebra B has finite degree over K and that the algebra Z(A)⊗KZ(B)Z(A)\otimes_KZ(B) is a field (which is, in particular, the case when A or B is central). Let ff and gg be K-algebra homomorphisms from B to A. There exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

By Lemma 1 of VIII, p. 254, it suffices to prove that the (B,A)(B,A)-bimodules AfA^f and AgA^g are isomorphic. Now, we can view AfA^f and AgA^g as left modules over the algebra C=B⊗KAoC = B\otimes_KA^o, which is simple by Proposition 7 of VIII, p. 221. As right A-modules, AfA^f and AgA^g are isomorphic to AdA_d, hence have finite length because the ring A is simple (VIII, p. 121, Corollary 1). A fortiori, AfA^f and AgA^g are C-modules of finite length. Let S be a simple C-module; there exist strictly positive integers mm and nn such that AfA^f is isomorphic to SmS^m and AgA^g to SnS^n. The right A-module S therefore has nonzero finite length. Since the right A-modules underlying AfA^f and AgA^g are isomorphic, they have the same length; we therefore have m=nm=n, so that the C-modules AfA^f and AgA^g are isomorphic.

Corollary 1 00GS

Let A be a central simple algebra over K, and let L be an extension of K of finite degree. If ff and gg are K-algebra homomorphisms from L to A, then there exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

Corollary 2 00GT

Let A be a central simple algebra over K, and let L be a subalgebra of A that is a field. Every K-algebra homomorphism from L to A extends to an inner automorphism of A.

Corollary 3 00GU

Let D be a field of finite degree over K with center K. Every element of D is algebraic over K. Let xx and yy be elements of D; there exists an element aa of D∗D^* such that y=axa−1y=axa^{-1} if and only if xx and yy have the same minimal polynomial over K.

The first assertion follows from Corollary 1 of V, §3, No. 1, p. 17.

Suppose that there exists an element aa of D∗D^* such that y=axa−1y=axa^{-1}; for every polynomial P of K[X], we have P(y)=aP(x)a−1P(y) =aP(x)a^{-1}, and, in particular, we have P(x)=0P(x) = 0 if and and only if P(y)=0P(y) = 0. Consequently, xx and yy have the same minimal polynomial over K (V, §3, No. 1, p. 16, Theorem 1).

Conversely, suppose that xx and yy have the same minimal polynomial. By loc. cit., there exists a K-isomorphism uu from K[x]K[x] to K[y]K[y] such that u(x)=yu(x) =y, and K[x]K[x] is a field. By Corollary 2,u,u extends to an inner automorphism θ:z↦aza−1\theta :z\mapsto aza^{-1} of D, and we therefore have y=θ(x)=axa−1y=\theta (x) =axa^{-1}.

Proposition 1 00GV

Let A be a central simple algebra of finite degree over K. Let B be a K-algebra, and let ff and gg be algebra homomorphisms from B to A. The following properties are equivalent:

(i) There exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

(ii) As left B-modules, AfA^f and AgA^g are isomorphic.

By Lemma 1 (VIII, p. 254), property (i) is equivalent to the fact that AfA^f and AgA^g are isomorphic as (B(B, A)-bimodules. Since A is finite-dimensional over K,AfK, A^f and AgA^g are B-modules of finite length. Since the center of A is equal to K, the equivalence of (i) and (ii) follows from Lemma 2 of VIII, p. 254 applied to the (Ao,Bo(A^o,B^o)-bimodules AfA^f and AgA^g.

4. Automorphisms of Semisimple Algebras

Theorem 4 00GW

Let A be a semisimple ring, Z its center, and uu an automorphism of A. Suppose that A is a finitely generated Z-module and that we have u(z)=zu(z) =z for every zz in Z. Then uu is an inner automorphism.

This follows from Theorem 2 of VIII, p. 256 applied with f=f= IdA_A and g=ug=u.

Example 00RH

Theorem 4 applies in the following two specific cases:

a) Let D be a field and Z its center. If D has finite degree over Z, then every automorphism of D that fixes the elements of Z is an inner automorphism. The assumption that D has finite degree over Z is essential (VIII, p. 269, Exercise 4).

b) Let V be a finite-dimensional vector space over the field K. Every automorphism of the K-algebra EndK(V)_K(V) is an inner automorphism; this result extends to the case when the space V is not finite-dimensional over K (VIII, p. 272, Exercise 13).

In particular, every automorphism of a matrix algebra Mn(K)\mathbf{M}_n(K) (with n⩾n\geqslant 1) is an inner automorphism. This result admits the following generalization.

Proposition 2 00GX

Let L be a commutative ring and V a free L-module of finite dimension mm. Suppose that every L-module M such that MmM^m is isomorphic to LmL^m is isomorphic to L. Then every automorphism of the L-algebra EndL(V)_L(V) is an inner automorphism.

Set B = EndL(V)_L(V). Let uu be an automorphism of the L-algebra B. We view V as a left B-module; let u∗(V)u_*(V) be the left B-module associated with uu, with external law (b,v)↦u(b)(v)(b, v)\mapsto u(b)(v) (II, §1, No. 13, p. 221). Let (e1,...,em)(e_1, . . . , e_m) be a basis of the L-module V; given elements v1,...,vmv_1, . . . , v_m of V, there exists a unique element bb of B such that we have b(ei)=vib(e_i) =v_i for 1⩽i⩽m1\leqslant i\leqslant m. In other words, the element e=(e1,...,em)e= (e_1, . . . , e_m) of VmV^m provides a basis of the B-module VmV^m. Since uu is an automorphism, ee also gives a basis of the B-module u∗(Vm)=u∗(V)mu_*(V^m) =u_*(V)^m, which is therefore isomorphic to VmV^m. The (B(B, L)-bimodule V is invertible (VIII, p. 102, Example 1). By Theorem 2, b) of VIII, p. 103, there consequently exists an L-module M such that the B-module u∗(V)u_*(V) is isomorphic to V⊗LMV\otimes_LM. The B-modules V⊗LLmV\otimes_LL^m and V⊗LMmV\otimes_LM^m, respectively isomorphic to VmV^m and u∗(V)mu_*(V)^m, are therefore isomorphic. By loc. cit., the L-modules LmL^m and MmM^m are isomorphic. Given the assumption, M is isomorphic to L; consequently, the B-module u∗u_*(V), which is isomorphic to V⊗LMV\otimes_LM, is isomorphic to V. Let hh be a B-module isomorphism from V to u∗(V)u_*(V); it is, in particular, an automorphism of the L-module V, that is, an invertible element of B. For bb in B and vv in V, we have h(b(v))=u(b)(h(v))h(b(v)) =u(b)(h(v)) and therefore u(b)=hbh−1u(b) =hbh^{-1}.

The conditions of Proposition 2 are, in particular, satisfied when the commutative ring L is a principal ideal domain (VII, §3, p. 15, Corollary 3) or is Artinian (VIII, p. 37, Theorem 2, d)) or local (VIII, p. 36, Corollary 6).

5. Simple Subalgebras of Simple Algebras

Theorem 5 00GY

Let A be a central simple K-algebra, and let B be a semisimple subalgebra of A of finite degree over K.

a) The commutant B′B' of B in A is a simple subalgebra, and B is the commutant of B′B' in A. Moreover, the algebra B∩B′B\cap B' is a semisimple commutative algebra of finite degree over K; it is the common center of B and B′B'.

b) Suppose that B is simple. Then B′B' is simple, and we have the equalities

[A:B′]s=[B:K],[A:B]s=[B′:K],[A:K]=[B:K][B′:K] [A : B']_s= [B : K],[A : B]_s= [B': K],[A : K] = [B : K][B': K]

(See VIII, p. 124, Definition 2 for the definition of the left degree [A:B]s.[A : B]_s.)

The K-algebra AoA^o is central simple, and the K-algebra B is semisimple and of finite degree. By Corollary 1 of VIII, p. 221, the algebra C=B⊗KAoC = B\otimes_KA^o is semisimple. Let M be the C-module with the same additive group as A and with external law given by the formula (b⊗a)a′=ba′a(b\otimes a)a'=ba'a for a,a′a, a' in A and bb in B. Let uu be an element of EndZ(A)_{\mathbf{Z}}(A). Then uu belongs to the commutant CM′C'_M of the C-module M if and only if uu is right A-linear and left B-linear, in other words, if and only if uu belongs to the commutant of BMB_M in the ring of homotheties of the A-module AsA_s. We consequently define an isomorphism γ\gamma from B′B' to CM′C'_M by the relation γ(b′)(x)=b′x\gamma (b')(x) =b'x for b′b' in B′B' and xx in M. Now, the ring C is semisimple, and the C-module M is generated by the element 1 of A. By Proposition 6 of VIII, p. 139, the ring CM′C'_M is semisimple, so the algebra B′B' is semisimple.

Let φ\varphi be the K-algebra homomorphism from A⊗KAoA\otimes_KA^o to EndK(M)_K(M) that sends a⊗a′a\otimes a' to the K-linear mapping x↦axa′x\mapsto axa' from M to M. Since the K-algebras A and AoA^o are central simple, the only two-sided ideals of A⊗KAoA\otimes_KA^o are 0 and A⊗KAoA\otimes_KA^o (VIII, p. 218, Corollary). We have CM=φ(B⊗Ao)C_M=\varphi (B\otimes A^o) and CM′=φ(B′⊗K)C'_M=\varphi (B'\otimes K). The homomorphism φ\varphi is not zero; it is therefore injective. Since the ring C is semisimple, we have CM′′=CMC''_M= C_M by Proposition 5 of VIII, p. 139. It follows that the subalgebra B⊗KAoB\otimes_KA^o of A⊗KAoA\otimes_KA^o is the commutant of the subalgebra B′⊗KKB'\otimes_KK. The commutant of B′⊗KKB'\otimes_KK in A⊗KKA\otimes_KK is therefore equal to (B⊗KAo)∩(A⊗KK)(B\otimes_KA^o)\cap (A\otimes_KK), that is, to B⊗KB\otimes K by Proposition 19 of II, §7, No. 9, p. 311. Hence the commutant of B′B' in A is equal to B. The algebra L=B∩B′L = B\cap B' is the center of B. Since B is a semisimple algebra of finite degree over K, the algebra L is commutative and semisimple and has finite degree over K. Since B is the commutant of B′B' in A, the center of B′B' is also equal to L=B∩B′L = B\cap B' (VIII, p. 77). We have proved a).

Now suppose that the algebra B is simple. By Corollary 2 of VIII, p. 222, the ring C is simple. By Proposition 4 of VIII, p. 123 applied to the C-module M, whose commutant is isomorphic to B′B', the ring B′B' is simple and M is a B′B'-module of finite length. In other words, B′B' is a simple subring of the simple ring A, and the left degree [A:B′]s[A : B']_s is an integer m⩾1m\geqslant 1. Viewed as a left B′B'-module, A has a finite basis (a1,...,am)(a_1, . . . , a_m). Moreover (loc. cit.), φ\varphi induces by restriction an isomorphism from C to CM′′=C''_M= EndB′_{B'}(A), and the mapping c↦(ca1,...,cam)c\mapsto (ca_1, . . . , ca_m) from C to AmA^m is therefore bijective. Consequently, C is a free right A-module of dimension mm. Now, we have C=B⊗AoC = B\otimes A^o, so C is a free right A-module of dimension [B : K]; it follows that [A:B′]s=m=[B:K][A : B']_s=m= [B : K]. From Proposition 6 of VIII, p. 125, we deduce

[A:K]=[A:B′]s[B′:K]=[B:K][B′:K] [A : K] = [A : B']_s[B': K] = [B : K] [B': K]

since we also have

[A:K]=[A:B]s[B:K] [A : K] = [A : B]_s[B : K]

and [B : K] is finite and nonzero, we conclude that we have the equality [A:B]s=[B′:K]([A : B]_s= [B': K] (Set Theory, III, §6, No. 3, p. 188, Corollary 3). We have proved b).

Let A be a central simple K-algebra of finite degree. There can exist semisimple commutative subalgebras B of A satisfying [A:K]≠[A : K]\not= [B:K][B′:K][B : K] [B': K] (Exercise 1 of VIII, p. 269).

Theorem 6 00GZ

Let A be a central simple K-algebra, B a subalgebra of A of finite degree, and B′B' its commutant in A.

a) Suppose that B is central simple. Then B′B' is central simple, and the K-algebra homomorphism θ:B⊗KB′→A\theta : B\otimes_KB'\rightarrow A that sends b⊗b′b\otimes b' to bb′bb' is an isomorphism.

b) Suppose that B is semisimple, and let L=B∩B′L = B\cap B'. Then B′B' is a semisimple algebra. The commutant L′L' of L in A is a semisimple ring with center L, and the ring homomorphism ψ:B⊗LB′→L′\psi : B\otimes_LB'\rightarrow L' that sends b⊗b′b\otimes b' to bb′bb' is an isomorphism.

Let us prove a). If B is central simple, then B′B' is central simple by Theorem 5 of VIII, p. 259. Hence the K-algebra B⊗KB′B\otimes_KB' is simple (VIII, p. 222, Corollary 2), and the homomorphism θ:B⊗KB′→A\theta : B\otimes_KB'\rightarrow A is injective. Now, by the equality of [A:B′]s[A : B']_s and [B : K] (Theorem 5), the left B′B'-modules B⊗KB′B\otimes_KB' and A are free of the same finite dimension; they are therefore B′B'-modules of the same finite length. By Corollary 2 of II, §1, No. 10, p. 213,θ213,\theta is bijective.

Let us prove b). By Theorem 5, the algebra L is commutative, of finite degree over K, and semisimple. By loc. cit. applied to L, its commutant L′L' in A is a semisimple algebra, and L is the commutant of L′L' in A, so L is the center of L′L'. Since L is the center of the semisimple rings L′,BL', B, and B′B', we can identify L′L' with a finite product of simple rings Li′L'_i (for i∈Ii\in I), so that we have

L=∏i∈ILi,B=∏i∈IBi,B′=∏i∈IBi′ L =\prod_{i\in I}L_i,B =\prod_{i\in I}B_i,B'=\prod_{i\in I}B'_i

where LiL_i is the center of Li′L'_i and where BiB_i and Bi′B'_i are subalgebras of Li′L'_i with center LiL_i that are each other’s commutants in Li′L'_i. We view Li′L'_i as a central simple algebra over the commutative field LiL_i and BiB_i as a central simple LiL_i-algebra of finite degree. By assertion a), the canonical mapping ψi:Bi⊗LiBi′→Li′\psi_i: B_i\otimes_{L_i}B'_i\rightarrow L'_i that sends bi⊗bi′b_i\otimes b'_i to bibi′b_ib'_i is a ring isomorphism. Now, we can identify B⊗LB′B\otimes_LB' with ∏i∈I(Bi⊗LiBi′)\prod_{i\in I}(B_i\otimes_{L_i}B'_i), so that ψ\psi is the product of the family of mappings (ψi)i∈I(\psi_i)_{i\in I}. Therefore, ψ\psi is a ring isomorphism.

Corollary 00H0

Suppose that the field K is algebraically closed and that A is a simple algebra of finite degree over K. Let B be a simple subalgebra of A, and let B′B' be the commutant of B in A. Then B′B' is a simple K-algebra, B is the commutant of B′B', we have [A:K]=[B:K][B′:K][A : K] = [B : K][B': K], and the canonical homomorphism from B⊗KB′B\otimes_KB' to A is a K-algebra isomorphism.

Since every simple algebra of finite degree over K is central, the corollary follows from Theorems 5 and 6.

6. Maximal Commutative Subalgebras

We say that a subalgebra of a K-algebra A is a maximal commutative subalgebra of A if it is a maximal element of the set of commutative subalgebras of A.

Lemma 3 00H1

Let A be a K-algebra and L a subalgebra of A.

a) The algebra L is a maximal commutative subalgebra of A if and only if L is equal to its commutant L′L' in A.

b) Let K′K' be a nonzero commutative K-algebra. Then L is a maximal commutative subalgebra of A if and only if L(K′)L_{(K')} is a maximal commutative subalgebra of A(K′)A_{(K')}.

Let us prove a). First suppose that L is equal to L′L'. Then L is commutative; if M is a commutative subalgebra of A containing L, then we have xy=yxxy=yx for xx in L and yy in M, so M⊂L′M\subset L' and therefore M = L. Consequently, L is a maximal commutative subalgebra of A.

Conversely, suppose that L is a maximal commutative subalgebra of A, and let xx be an element of L′L'. The subalgebra M of A generated by L∪{x}L\cup \{x\} is then commutative and contains L. Because L is maximal, we have M = L, and so x∈Lx\in L and finally L=L′L = L', which gives a).

By Proposition 6 of III, §4, No. 4, p. 468, the commutant of L(K′)L_{(K')} in A(K′)A_{(K')} is L(K′)′L'_{(K')}. Since the equalities L=L′L = L' and L(K′)=L(K′)′L_{(K')}= L'_{(K')} are equivalent (II, §7, No. 9, p. 311, Proposition 19), assertion b) follows from a).

Proposition 3 00H2

Let A be a central simple K-algebra of finite degree, and let L be a semisimple commutative subalgebra of A. The following properties are equivalent:

(i) The algebra L is a maximal commutative subalgebra of A.

(ii) The left L-module A is free, of dimension [L : K].

(iii) We have [A:K]=[L:K]2[A : K] = [L : K]^2.

Suppose, moreover, that A is the algebra EndK(V)_K(V), where V is a vector space of nonzero finite dimension over K. Then the previous properties are also equivalent to the following:

(iv) V is a free L-module of dimension 1.

A) Suppose that A is of the form EndK_K(V), where V is a vector space of nonzero finite dimension over the field K. We will establish the equivalence of conditions (i) through (iv) following the diagram

(i) =⇒=\Rightarrow (iv) =⇒=\Rightarrow (ii) =⇒=\Rightarrow (iii) =⇒=\Rightarrow (i) .

Since L is a semisimple commutative algebra of finite degree over K, we can identify it with a finite product ∏i∈ILi\prod_{i\in I}L_i of extensions of K of finite degree (VIII, p. 137, Proposition 3). For every i∈Ii\in I, let ViV_i be the isotypical component of type LiL_i of the L-module V; it is a vector space of nonzero finite dimension over LiL_i because V is a faithful L-module (VIII, p. 144, Corollary). We can then identify V with ∏i∈IVi\prod_{i\in I}V_i. Under these conditions, the commutant L′L' of L in A, which is simply the algebra EndL_L(V), can be identified with the product ∏i∈I\prod_{i\in I} EndLi(Vi)_{L_i}(V_i).

Suppose that L is a maximal commutative subalgebra of EndK(V)_K(V). By Lemma 3, a), we have L=L′L = L', so L′L' is commutative and we have dimLi(Vi)=1_{L_i}(V_i) = 1 for every i∈Ii\in I. So (i) implies (iv).

Suppose that the L-module V is free of dimension 1. Let (e1,...,er)(e_1, . . . , e_r) be a basis of V over K. The mapping a↦(ae1,...,aer)a\mapsto (ae_1, . . . , ae_r) is an isomorphism of left A-modules and therefore of L-modules from A to VrV^r. Consequently, A is a free left L-module of dimension rr, and we have r=r= dimK(V)=[L:K]_K(V) = [L : K]. So (iv) implies (ii).

It is clear that (i) implies (iii).

Finally, suppose that we have [A:K]=[L:K]2[A : K] = [L : K]^2, in other words, dimK(V)=[L:K]_K(V) = [L : K]. We then have ∑i\sum_i dimLi(Vi)[Li:K]=∑i[Li:K]_{L_i}(V_i)[L_i: K] =\sum_i[L_i: K], so that ViV_i has dimension 1 over LiL_i for every ii. We then have EndLi(Vi)=Li_{L_i}(V_i) = L_i for every ii, and so L′=LL'= L. By Lemma 3, a), L is a maximal commutative subalgebra of A. So (iii) implies (i).

B) Let us continue with the general case. By Theorem 1 of VIII, p. 252, there exists a separable extension K′K' of K of finite degree such that the K′K'-algebra A(K′)A_{(K')} is isomorphic to an algebra EndK′(V′)_{K'}(V'), where V′V' is a vector space of nonzero finite dimension over K′K'. Then the K′K'-algebra L(K′)L_{(K')} is commutative and semisimple (VIII, p. 222, Corollary 2). By the first part of the proof, the following properties are equivalent:

(i′') The algebra L(K′)L_{(K')} is a maximal commutative subalgebra of A(K′)A_{(K')}.

(ii′') The left L(K′)L_{(K')}-module A(K′)A_{(K')} is free, of dimension [L(K′):K′][L_{(K')}: K'].

(iii′') We have [A(K′):K′]=[L(K′):K′]2[A_{(K')}: K'] = [L_{(K')}: K']^2.

By Lemma 3, b), properties (i) and (i’) are equivalent.

Set n=[L:K]n= [L : K]; then n=[L(K′):K′]n= [L_{(K')}: K']. Property (ii) means that the left L-modules A and LnL^n are isomorphic; by Theorem 3 of VIII, p. 37, this is equivalent to the L(K′)L_{(K')}-modules A(K′)A_{(K')} and (L(K′))n(L_{(K')})^n being isomorphic. The equivalence of (ii) and (ii′') follows.

Finally, we have [A:K]=[A(K′):K′][A : K] = [A_{(K')}: K'] and [L:K]=[L(K′):K′][L : K] = [L_{(K')}: K'], which give the equivalence of properties (iii) and (iii′'). We have thus proved the equivalence of properties (i), (ii), and (iii).

Corollary 00H3

Let A be a central simple algebra of finite degree over K, and let L be a semisimple commutative K-algebra such that [A : K] is equal to [L:K]2[L : K]^2. Let ff and gg be injective homomorphisms from L to A. There exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f.

Set n=[L:K]n= [L : K]. Viewed as a left module over the subring ff(L), A is free of dimension nn; this follows from the equivalence of properties (ii) and (iii) of Proposition 3. Since ff is an isomorphism from L to ff(L), the left L-module AfA^f (whose law of action is given by (x,a)↦f(x)a(x, a)\mapsto f(x)a) is free of dimension nn. The same holds for AgA^g, which is therefore isomorphic to AfA^f. We conclude using the equivalence of properties (i) and (ii) of Proposition 1 (VIII, p. 257).

Suppose that A is a central simple algebra of finite degree over K. There can exist maximal commutative subalgebras L of A that are not semisimple and for which [A:K]≠[L:K]2[A : K]\not= [L : K]^2 (VIII, p. 270, Exercise 5).

7. Maximal Étale Subalgebras

Lemma 4 00H4

Let A be a central simple algebra of finite degree over K, distinct from K. There exists an étale (V, §6, No. 3, p. 28, Definition 1) subalgebra of A distinct from K.

By Wedderburn’s theorem (VIII, p. 120, Theorem 1), we may assume that A is of the form Mn\mathbf{M}_n(D), where nn is a strictly positive integer and D a field with center K.

Suppose n>1n >1. The algebra of diagonal matrices with entries in K is an étale subalgebra of A distinct from K.

Suppose n=1n= 1. Let pp be the characteristic exponent of D. By Lemma 1 of VIII, p. 230, there exists an element aa of D such that apma^{p^m} does not belong to K for any natural number mm. So for mm sufficiently large, the element x=apmx=a^{p^m} is separable over K (V, §7, No. 7, p. 44, Proposition 13), but it does not belong to K. The subalgebra K(x)K(x) of A is a separable extension of the field K of finite degree, hence an étale subalgebra over K; it is distinct from K.

Proposition 4 00H5

Let A be a central simple algebra of finite degree over K. Let L be a subalgebra of A, and let L′L' be the commutant of L in A.

a) If L is maximal among the semisimple commutative subalgebras of A, then we have L=L′L = L', and L is a maximal commutative subalgebra of A.

b) If L is maximal among the étale subalgebras of A, then we have L=L′L = L', and L is a maximal commutative subalgebra of A.

We know that the relation L=L′L = L' means that L is a maximal commutative subalgebra of A (VIII, p. 261, Lemma 3, a)). Suppose that L is semisimple, commutative, and distinct from L′L'. By Theorem 5 of VIII, p. 259,L′, L' is semisimple, and L is the commutant of L′L' and therefore the center of L′L'; consequently, L′L' is not commutative. It suffices to prove that there exists a semisimple commutative subalgebra M of A that is distinct from L and contains L and that is étale if L is étale.

By the structure theorem for semisimple rings (VIII, p. 135, Theorem 1), there exist simple rings B1,...,BrB_1, . . . ,B_r and an isomorphism φ\varphi from L′L' to B1×⋯×BrB_1\times \cdots \times B_r. For 1⩽i⩽r1\leqslant i\leqslant r, denote the center of BiB_i by EiE_i; we then have φ(L)=E1×⋯×Er\varphi (L) = E_1\times \cdots \times E_r. Since L′L' is not commutative, we may assume that B1B_1, for example, is not commutative; we therefore have B1≠E1B_1\not= E_1, and, by Lemma 4, there exists a subalgebra M1M_1 of B1B_1 that is commutative, distinct from E1E_1, and étale over E1E_1. Set M=φ−1(M1×E2×⋯×Er)M =\varphi^{-1}(M_1\times E_2\times \cdots \times E_r); it is a semisimple commutative subalgebra of A containing L and distinct from L. Suppose that L is étale over K; let us prove that M is étale. The extensions EiE_i of K are separable (V, §6, No. 4, p. 30, Proposition 3). Moreover, since the E1E_1-algebra M1M_1 and the K-algebra E1E_1 are étale, the K-algebra M1M_1 is also étale (V, §6, No. 5, p. 32, Corollary 2). Thus the K-algebra M1×E2×⋯×ErM_1\times E_2\times \cdots \times E_r is étale, and therefore so is M.

Let A be a central simple algebra of finite degree over K. A subalgebra of A that is maximal among the semisimple commutative subalgebras of A is called a maximal semisimple commutative subalgebra of A. By Proposition 4, the term “maximal” refers to the property of being commutative or of being semisimple and commutative. A subalgebra of A that is maximal among the étale subalgebras of A is called a maximal étale subalgebra of A.

Corollary 1 00H6

Let A be a central simple K-algebra of finite degree. Every semisimple (resp. étale) commutative subalgebra of A is contained in a maximal commutative subalgebra of A that is semisimple (resp. étale).

Corollary 2 00H7

Let D be a field of finite degree over K with center K.

a) The maximal commutative subfields of D are the maximal commutative subalgebras of D and also the maximal semisimple commutative subalgebras of D. Every commutative subfield L of D is contained in a maximal commutative subfield.

b) Let L be a commutative subfield of D that is a separable extension of K; it is contained in a maximal commutative subfield of D that is a separable extension of K.

c) Let L be a commutative subfield of D. Then L is a maximal commutative subfield of D if and only if we have [D:K]=[L:K]2[D : K] = [L : K]^2.

A subalgebra of D is a field (V, §2, No. 2, p. 10, Proposition 1) and is therefore semisimple. Moreover, a maximal commutative subfield of D contains K. Assertions a) and b) then follow from Corollary 1 and assertion c) of Proposition 3 (VIII, p. 262).

Proposition 5 00H8

Let A be a central simple algebra of finite degree over K. Let B be a semisimple subalgebra of A, and let B′B' be the commutant of B.

a) The algebra B contains a maximal semisimple commutative subalgebra of A if and only if B contains B′B'.

b) Suppose that B contains B′B', and let gg be a K-algebra homomorphism from B to A. There exists an inner automorphism θ\theta of A that coincides with gg on B.

Let L be a maximal commutative subalgebra of A; by Lemma 3 of VIII, p. 261, L is equal to its commutant L′L' in A. If B contains L, then its commutant B′B' is contained in L′L' and therefore in B.

Conversely, suppose that B′B' is contained in B. Then B′B' is the center of B, and it is a semisimple commutative algebra (VIII, p. 137, Proposition 2). By Corollary 1, there exists a maximal semisimple commutative subalgebra L of A containing B′B'. The commutant of L is L (VIII, p. 261, Lemma 3, a)), and that of B′B' is equal to B (VIII, p. 259, Theorem 5). The relation L⊃B′L\supset B' therefore implies L⊂BL\subset B. We have proved a).

Let us prove b). Suppose that B contains B′B', and choose a maximal semisimple commutative subalgebra L of A contained in B, which exists by a). Let gg be a homomorphism from B to A. By Proposition 3 of VIII, p. 262, we have the equality [A:K]=[L:K]2[A : K] = [L : K]^2; by the corollary of VIII, p. 263, there exists an inner automorphism θ1\theta_1 of A that coincides with gg on L. If ff is the canonical injection of B into A, then the homomorphisms gg and θ1∘f\theta_1\circ f have the same restriction to the center B′B' of B because B′B' is contained in L. By Theorem 2 of VIII, p. 256, there exists an inner automorphism θ\theta of A such that g=θ∘fg=\theta \circ f; in other words, θ\theta extends gg.

8. Diagonalizable Subalgebras of Simple Algebras

Let D be a K-algebra that is a field, and let V be a finite-dimensional right vector space over D. Let L be a K-subalgebra of EndD(V)_D(V) that is a diagonalizable K-algebra (V, §6, No. 3, p. 28). By definition, L has finite degree over K, and there exists a basis (εi)i∈I(\varepsilon_i)_{i\in I} of L over K with the following properties:

εi2=εi,εiεj=0\varepsilon^2_i=\varepsilon_i, \varepsilon_i\varepsilon_j= 0 if i≠j,∑i∈Iεi=1i\not=j ,\sum_{i\in I}\varepsilon_i= 1.

Set Vi=εi(V)V_i=\varepsilon_i(V) for every ii in I; then (Vi)i∈I(V_i)_{i\in I} is a family of nonzero linear subspaces of V with direct sum V (II, §1, No. 8, p. 209, Proposition 12). Let uu be an endomorphism of V; then uu belongs to L if and and only if for every i∈Ii\in I, there exists an element λi\lambda_i of K such that u(x)=λixu(x) =\lambda_ix for every x∈Vix\in V_i.

Conversely, suppose that V is the direct sum of a family (Vi)i∈I(V_i)_{i\in I} of nonzero linear subspaces. For any element λ=(λi)i∈I\boldsymbol{\lambda }= (\lambda_i)_{i\in I} of KIK^I, denote by uλu_{\boldsymbol{\lambda }} the endomorphism of the D-vector space V such that uλ(x)=λixu_{\boldsymbol{\lambda }}(x) =\lambda_ix for x∈Vix\in V_i. The set L of endomorphisms uλu_{\boldsymbol{\lambda }} for λ∈KI\boldsymbol{\lambda }\in K^I is a diagonalizable subalgebra of EndD(V)_D(V) having the family (εi)i∈I(\varepsilon_i)_{i\in I} as basis, where εi\varepsilon_i is the projector with image ViV_i and kernel ∑j≠iVj\sum_{j\not=i}V_j. We say that L is the diagonalizable subalgebra of EndD(V)_D(V) associated with the direct sum decomposition V=⊕i∈IViV =\oplus_{i\in I}V_i. We have [L : K] = Card(I) ⩽\leqslant dimD(V)_D(V).

Proposition 6 00H9

Let L be the diagonalizable subalgebra of EndD(V)_D(V) associated with a direct sum decomposition V=⊕i∈IViV =\oplus_{i\in I}V_i.

a) The algebra L is maximal among the diagonalizable subalgebras of the K-algebra EndD(V)_D(V) if and only if every ViV_i has dimension 1 over D.

b) The algebra L is a maximal commutative subalgebra of EndD(V)_D(V) if and only if we have D = K and every ViV_i has dimension 1 over K.

If every vector space ViV_i has dimension 1 over D, then we have

[L : K] = Card(I) = dimD(V)_D(V),

so L is maximal among the diagonalizable subalgebras of EndD(V)_D(V). In the opposite case, there exists an index j∈Ij\in I such that dimD(Vj)⩾2_D(V_j)\geqslant 2. Choose two nonzero linear subspaces Vj′V'_j and Vj′′V''_j of VjV_j with direct sum VjV_j. The diagonalizable subalgebra of EndD(V)_D(V) associated with the direct sum decomposition V=(⊕i∈I−{j}Vi)⊕Vj′⊕Vj′′V = (\oplus_{i\in I-\{j\}}V_i)\oplus V'_j\oplus V''_j contains L and is not equal to L; assertion a) follows.

The commutant L′L' of L in EndD(V)_D(V) consists of the endomorphisms of the form (xi)↦(ui(xi))(x_i)\mapsto (u_i(x_i)), with (ui)∈∏i∈I(u_i)\in \prod_{i\in I} EndD(Vi)_D(V_i). The algebra L is a maximal commutative subalgebra of EndD(V)_D(V) if and only if we have L=L′L = L' (VIII, p. 261, Lemma 3, a)). This relation is therefore equivalent to “EndD(Vi)=K_D(V_i) = K for every i∈i\in I”; assertion b) follows.

Proposition 7 00HA

Let L be a commutative algebra of finite degree over K. The following properties are equivalent:

(i) The algebra L is étale.

(ii) There exists a separable extension of K of finite degree that diagonalizes K.

The implication (ii)⇒\Rightarrow(i) follows from V, §6, No. 3, p. 29, Proposition 2.

Let us prove the implication (i)⇒\Rightarrow(ii). Let Ω be a separable closure of K. By Theorem 4 of V, §6, No. 7, p. 34, there exist extensions L1,...,LnL_1, . . . ,L_n of K of finite degree, contained in Ω, such that L is isomorphic to the product L1×⋯×LnL_1\times \cdots \times L_n. Let N be a Galois extension of K that contains the LiL_i (V, §10, No. 1, p. 58), and let us prove that L(N)L_{(N)} is diagonalizable. By the primitive element theorem (V, §7, No. 4, p. 40, Theorem 1), for every i∈[1,n]i\in [1, n], there exists an irreducible separable polynomial Pi∈K[X]P_i\in K[X] such that LiL_i is isomorphic to K[X]/(Pi)K[X]/(P_i). Since N is a normal extension of K in which PiP_i admits a root, the polynomial PiP_i splits in N[X], with simple roots. Consequently, the N-algebra Li(N)L_{i(N)}, which is isomorphic to N[X]/(Pi)N[X]/(P_i), is isomorphic to N[Li:K]N^{[L_i:K]}. Hence L(N)L_{(N)} is diagonalizable.

Theorem 7 00HB

Let A be a central simple K-algebra of finite degree and L a subalgebra of A. The following properties are equivalent:

(i) The algebra L is a maximal étale subalgebra of A.

(ii) There exist an extension K′K' of K, an integer n⩾1n\geqslant 1, and an isomorphism θ\theta from A(K′)A_{(K')} to Mn(K′)\mathbf{M}_n(K') that sends L(K′)L_{(K')} to the set of diagonal matrices.

(iii) There exist K′,nK',n, and θ\theta as in (ii), where the extension K′K' is, moreover, assumed Galois and of finite degree.

It is clear that (iii) implies (ii).

If property (ii) holds, then L(K′)L_{(K')} is a maximal commutative subalgebra of A(K′)A_{(K')} (Proposition 6), and it is diagonalizable. The K-algebra L is then étale (V, §6, No. 3, p. 28, Definition 1), and it is a maximal commutative subalgebra of A (VIII, p. 261, Lemma 3, b)). We have proved that (ii) implies (i).

Suppose that property (i) holds. Since L is étale over K, by Proposition 7, there exists a Galois extension K1K_1 of K of finite degree such that the K1K_1-algebra L(K1)L_{(K_1)} is diagonalizable. The algebra A is central simple; by (VIII, p. 252, Theorem 1), there exist a Galois extension K2K_2, a vector space V of finite dimension nn over K2K_2, and an isomorphism θ\theta from A(K2)A_{(K_2)} to EndK2(V)_{K_2}(V). By Proposition 1 of V, p. 55, we may assume K1=K2K_1= K_2. By Proposition 4, b) of VIII, p. 264 and Lemma 3, b) of VIII, p. 261,L(K′), L_{(K')} is a maximal commutative subalgebra of A(K′)A_{(K')}, so θ(L(K′))\theta (L_{(K')}) is a maximal commutative subalgebra of EndK′(V′)_{K'}(V'). Let us apply Proposition 6 to the diagonalizable algebra θ(L(K′)):\theta (L_{(K')}): there exists a basis (e1,...,en)(e_1, . . . , e_n) of V′V' over K′K' such that θ(L(K′))\theta (L_{(K')}) consists of the endomorphisms of V′V' with diagonal matrix with respect to this basis. So (i) implies (iii).

Exercises

See the exercises for § 14.

Transcribed from the printed edition, 2023, Springer Nature. Not translated.