Bourbaki

§ 2. Discriminant of a sesquilinear form

for example A=Fp(X)A = \mathbf{F}_p(X), where pp is prime ; E′′=A′′⊗AEE'' = A'' \otimes_A E is then a vector space of the same dimension over A′′A'' as EE over AA; consider the mapping M→A′′⊗AMM \to A'' \otimes_A M.

c) Assume A′=AA' = A. Let ω\omega be a bijective mapping of F(E)\mathfrak{F}(E) onto the set F′(E′)\mathfrak{F}'(E') of subspaces of finite codimension of E′E', such that codim⁡ω(M)=dim⁡M\operatorname{codim} \omega(M) = \dim M, and ω(M+N)=ω(M)∩ω(N)\omega(M + N) = \omega(M) \cap \omega(N). Show that there exists a sesquilinear form Φ\Phi on E×E′E \times E', non-degenerate on the right, and such that ω(M)=M0\omega(M) = M^0 for every M∈F(E)M \in \mathfrak{F}(E). (Use Theorem 1 of Chapter II, § 4, No. 6).

¶ 11) a) Let AA be a left and right artinian ring (Chapter VIII, § 2, No. 3). Show that the following conditions are equivalent: 1° the right annihilator (resp. left annihilator) of every left ideal (resp. right ideal) ≠A\neq A is not reduced to 0 ; 2° the dual of every simple left (resp. right) AA-module is not reduced to 0 ; 3° the dual of every left (resp. right) AA-module of finite type is not reduced to 0. A ring AA is said to satisfy the condition (Ns)(N_s) (resp. (Nd)(N_d)) if it satisfies these conditions for the left (resp. right) AA-modules.

b) Let AA be a left and right artinian ring satisfying the condition (Nd)(N_d), EE (resp. FF) a left (resp. right) free AA-module, having a countable basis over AA, Φ\Phi a bilinear form on E×FE \times F such that sΦs_\Phi is injective. Let MM be a free submodule of EE; show that there exists a free submodule NN of FF and a basis (en)(e_n) (resp. (fn)(f_n)) of MM (resp. NN) such that Φ(ei,fj)=δij\Phi(e_i, f_j) = \delta_{ij}; one can moreover take for e1e_1 any free element of MM. (Notice that if xx is a free element of MM, the image of FF by the mapping y→Φ(x,y)y \to \Phi(x, y) is the ring AA itself; proceed then by induction to construct the bases (en)(e_n) and (fn)(f_n)). Deduce that if the basis (en)(e_n) of MM is finite, EE is the direct sum of MM and N0N^0, and that one has M00=MM^{00} = M.

c) Retain the hypotheses of b), and suppose in addition that AA satisfies condition (Ns)(N_s) and that dΦd_\Phi is injective. Show that there exists a basis (en)(e_n) in EE and a basis (fn)(f_n) in FF such that Φ(ei,fj)=δij\Phi(e_i, f_j) = \delta_{ij}. (Use b) by determining ene_n and fnf_n alternately by induction).

12) Let EE be a real Hilbert space of countable type, Φ(x,y)\Phi(x, y) the scalar product in EE. Show that there does not exist a system of two algebraic bases (eλ),(fμ)(e_\lambda), (f_\mu) of the vector space EE over R\mathbf{R} such that Φ(eλ,fμ)=δλμ\Phi(e_\lambda, f_\mu) = \delta_{\lambda \mu} for every pair of indices. (First observe that the set of indices of these bases would have the power of the continuum (Esp.vect.top.Esp. vect. top., chap. II, § 3, exerc. 15)); then consider a (countable) orthonormal basis (an)(a_n) of EE and observe that the subspace generated by the ana_n is contained in the subspace generated by a countable subfamily of (eλ)(e_\lambda).

Definition 1

Given a sesquilinear form Φ\Phi for JJ on EE and a system S=(x1,…,xn)S = (x_1, \ldots, x_n) of nn elements of EE, the discriminant of Φ\Phi with respect to this system is called, and is denoted by DΦ(x1,…,xn)D_\Phi(x_1, \ldots, x_n) or DΦ(S)D_\Phi(S), the element det⁡(Φ(xi,xj))\det(\Phi(x_i, x_j)) of AA.

If (e1,…,en)(e_1, \ldots, e_n) is a basis of EE, the discriminant of Φ\Phi with respect to this basis is none other than the determinant of the matrix of Φ\Phi with respect to this basis.

It follows from the definition of the extension of Φ\Phi to ∧nE\wedge^n E (§1§ 1, no 9) that we have

DΦ(x1,…,xn)=Φ(n)(x1∧…∧xn,x1∧…∧xn), D_\Phi(x_1, \ldots, x_n) = \Phi_{(n)}(x_1 \wedge \ldots \wedge x_n, x_1 \wedge \ldots \wedge x_n),

where Φ(n)\Phi_{(n)} denotes the extension of Φ\Phi to ∧nE\wedge^n E. For every permutation σ∈Sn\sigma \in \mathfrak{S}_n, we therefore have

DΦ(xσ(1),…,xσ(n))=DΦ(x1,x2,…,xn). D_\Phi(x_{\sigma(1)}, \ldots, x_{\sigma(n)}) = D_\Phi(x_1, x_2, \ldots, x_n).

Example

Let BB be an algebra over the ring AA, such that BB is a free AA-module of finite dimension nn. Then the mapping (x,y)→TrB/A(xy)(x, y) \to \mathrm{Tr}_{B/A}(xy) (Chapter VIII, § 12, no 2) is a bilinear form on BB. Given a system (x1,…,xn)(x_1, \ldots, x_n) of nn elements of BB, the discriminant of this form with respect to this system is called the discriminant of the system (x1,…,xn)(x_1, \ldots, x_n) over AA, and is denoted by DB/A(x1,…,xn)D_{B/A}(x_1, \ldots, x_n). Thus

DB/A(x1,…,xn)=det⁡(TrB/A(xixj)). D_{B/A}(x_1, \ldots, x_n) = \det (\mathrm{Tr}_{B/A}(x_ix_j)).

Remark

Soient (e1,…,en)(e_1, \ldots, e_n) a basis of BB over AA, and

eiej=∑k=1ncijkek(cijk∈A) e_ie_j = \sum_{k=1}^n c_{ijk}e_k \quad (c_{ijk} \in A)

(chap. II, § 7, no 2). Since the matrix of the AA-linear endomorphism x→ekxx \to e_kx of BB with respect to (er)(e_r) is (cksr)(c_{ksr}), we have TrB/A(ek)=∑r=1nckrr\mathrm{Tr}_{B/A}(e_k) = \sum_{r=1}^n c_{krr}, whence TrB/A(eiej)=∑k,rcijkckrr\mathrm{Tr}_{B/A}(e_ie_j) = \sum_{k,r} c_{ijk}c_{krr}. It follows that

DB/A(e1,…,en)=det⁡i,j(∑k,rcijkckrr). D_{B/A}(e_1, \ldots, e_n) = \det_{i,j} (\sum_{k,r} c_{ijk}c_{krr}).

Proposition 1

Let Φ\Phi be a sesquilinear form for JJ on EE, (x~1,…,xn)(\tilde{x}_1, \ldots, x_n) a system of nn elements of EE, and (aij)(i,j=1,…,n)(a_{ij})_{(i,j = 1, \ldots, n)} a family of n2n^2 elements of AA; put yi=∑i=1najixiy_i = \sum_{i=1}^n a_{ji}x_i. We then have

DΦ(y1,…,yn)=det⁡(aij)⋅det⁡(aij)j⋅DΦ(x1,…,xn). D_\Phi(y_1, \ldots, y_n) = \det (a_{ij}) \cdot \det (a_{ij})^j \cdot D_\Phi(x_1, \ldots, x_n).

Indeed, since Φ\Phi is a sesquilinear form, we have Φ(yi,yj)=∑k,maikΦ(xk,xm)ajmj\Phi(y_i, y_j) = \sum_{k,m} a_{ik} \Phi(x_k, x_m) a_{jm}^j. Hence, if we denote by AA the matrix (aij)(a_{ij}), the matrix (Φ(yi,yj))(\Phi(y_i, y_j)) is equal to A.(Φ(xi,xj)).tAjA . (\Phi(x_i, x_j)). {}^t A^j. Since det⁡(tA)=det⁡(A)\det({}^t A) = \det(A), and det⁡(Aj)=det⁡(A)j\det(A^j) = \det(A)^j, our assertion is proved.

In particular, if (ei)(e_i) and (ei′)(e'_i) are two bases of E,DE, D and D′D' the discriminants of Φ\Phi with respect to these bases, and aa the determinant of the passing matrix from the basis (ei)(e_i) to the basis (ei′)(e'_i), we have

(4)D′=aajD. (4) \qquad D' = aa^j D.

It follows from prop. 1 that, if (ei)(e_i) is a basis of EE and (xi)(x_i) an arbitrary system of nn elements of EE, DΦ(e1,…,en)D_\Phi(e_1, \ldots, e_n) divides DΦ(x1,…,xn)D_\Phi(x_1, \ldots, x_n). In particular the discriminants of Φ\Phi with respect to any two bases of EE generate the same principal ideal of AA.

Let (Ei)i∈I(E_i)_{i \in I} be a finite family of finite-dimensional free AA-modules, Φi\Phi_i a sesquilinear form for JJ on EiE_i, and BiB_i a basis of EiE_i. If Φ\Phi denotes the direct sum of the Φi\Phi_i (§ 1, no 3) and BB the basis of ∏i∈IEi\prod_{i \in I} E_i obtained by joining the BiB_i, then obviously

(5)DΦ(B)=∏i∈IDΦi(Bi). (5) \qquad D_\Phi(B) = \prod_{i \in I} D_{\Phi_i}(B_i).

Let Φ\Phi be a sesquilinear form for JJ on EE, hh a homomorphism of AA into a commutative ring A′A', Φ′\Phi' the sesquilinear form on A′⊗AEA' \otimes_A E obtained by extension of Φ\Phi (§ 1, no 4) and (x1,…,xn)(x_1, \ldots, x_n) an arbitrary system of elements of EE. Since A′⊗AEA' \otimes_A E is a free A′A'-module, DΦ′(1⊗x1,…,1⊗xn)D_{\Phi'}(1 \otimes x_1, \ldots, 1 \otimes x_n) is defined, and we have obviously

(6)DΦ′(1⊗x1,…,1⊗xn)=h(DΦ(x1,…,xn)). (6) \qquad D_{\Phi'}(1 \otimes x_1, \ldots, 1 \otimes x_n) = h(D_\Phi(x_1, \ldots, x_n)).

Example

Let BB be an algebra over AA which is a free AA-module of finite dimension nn, (x1,…,xn)(x_1, \ldots, x_n) a basis of BB over AA, and mm an ideal of AA. If hh denotes the canonical homomorphism of BB onto B/mBB/mB, (h(x1),…,h(xn))(h(x_1), \ldots, h(x_n)) is a basis of B/mBB/mB over A/mA/m (chap. I, § 6, no 5, prop. 5), and B/mBB/mB is isomorphic to (A/m)⊗AB(A/m) \otimes_A B. We therefore have

D(B/mB)/(A/m)(h(x1),…,h(xn))=h(DB/A(x1,…,xn)). D_{(B/mB)/(A/m)}(h(x_1), \ldots, h(x_n)) = h(D_{B/A}(x_1, \ldots, x_n)).

Proposition 2

Suppose that AA is an integral domain. Let Φ\Phi be a sesquilinear form for JJ on EE and (e1,…,en)(e_1, \ldots, e_n) a basis of EE, such that DΦ(e1,e2,…,en)≠0D_\Phi(e_1, e_2, \ldots, e_n) \neq 0.

a) For a system (x1,…,xn)(x_1, \ldots, x_n) of nn elements of EE to be free, it is necessary and sufficient that DΦ(x1,…,xn)D_\Phi(x_1, \ldots, x_n) be ≠0\neq 0.

b) For a system (x1,…,xn)(x_1, \ldots, x_n) of nn elements of EE to be a basis of EE, it is necessary and sufficient that DΦ(x1,…,xn)D_\Phi(x_1, \ldots, x_n) and DΦ(e1,…,en)D_\Phi(e_1, \ldots, e_n) be associated elements in AA (cf. Chapter VI, § 1, No. 5).

Let xj=∑i=1najieix_j = \sum_{i=1}^n a_{ji} e_i (aji∈Aa_{ji} \in A). Let us first prove a). If DΦ(x1,…,xn)=0D_\Phi(x_1, \ldots, x_n) = 0, we have det⁡(aji)⋅det⁡(aji)′=0\det(a_{ji}) \cdot \det(a_{ji})' = 0 (Prop. 1) since DΦ(e1,…,en)≠0D_\Phi(e_1, \ldots, e_n) \neq 0 and AA is integral; we therefore have det⁡(aji)=0\det(a_{ji}) = 0, and the vectors xjx_j are linearly dependent (Chapter III, § 7, No. 1, Th. 1, applied to the vector space K⊗AEK \otimes_A E, where KK denotes the field of fractions of AA). Conversely, if these vectors are linearly dependent we have det⁡(aji)=0\det(a_{ji}) = 0 (ibid.), whence DΦ(x1,…,xn)=0D_\Phi(x_1, \ldots, x_n) = 0 (Prop. 1).

Let us now prove b). If DΦ(x1,…,xn)D_\Phi(x_1, \ldots, x_n) and DΦ(e1,…,en)D_\Phi(e_1, \ldots, e_n) are associated in AA, Prop. 1 shows that det⁡(aij)⋅det⁡(aij)′\det(a_{ij}) \cdot \det(a_{ij})' is invertible in AA. Thus det⁡(aij)\det(a_{ij}) is itself invertible in AA; hence the matrix (aij)(a_{ij}) over AA is invertible (Chapter III, § 6, No. 5, Th. 2), and the endomorphism gg of EE defined by g(ei)=xig(e_i) = x_i (i=1,…,ni = 1, \ldots, n) is an automorphism; consequently (x1,…,xn)(x_1, \ldots, x_n) is a basis of EE. The converse follows immediately from Prop. 1.

Proposition 3

Let Φ\Phi be a sesquilinear form for JJ on EE, and let SS be a basis of EE. The following conditions are equivalent:

a) The mapping sΦs_\Phi from EE into E∗E^* associated with Φ\Phi is bijective.

b) The mapping dΦd_\Phi from EE into E∗E^* associated with Φ\Phi is bijective.

c) The element DΦ(S)D_\Phi(S) is invertible in AA.

Indeed condition c) expresses that the matrix of Φ\Phi with respect to SS is invertible (Chapter III, § 6, No. 5, Th. 2). Thus c) is equivalent to a) (§1,§ 1, No. 10); similarly c) is equivalent to b).

Proposition 4

Suppose that AA is integral. Let SS be a basis of EE. A necessary and sufficient condition for a sesquilinear form Φ\Phi on EE to be nondegenerate is that one has DΦ(S)≠0D_\Phi(S) \neq 0.

Indeed, let K be the field of fractions of A, and let Φ′\Phi' be the extension of Φ\Phi to the K-vector space K⊗AEK \otimes_A E; identify E with a subset of this vector space. The relation DΦ(S)≠0D_{\Phi}(S) \neq 0 is then equivalent to DΦ′(S)≠0D_{\Phi'}(S) \neq 0 (formula (6)), which itself expresses that sΦ′s_{\Phi'} is bijective (prop. 3), that is to say that Φ′\Phi' is nondegenerate (§ 1, No. 6, prop. 6). Now, for every x∈K⊗AEx \in K \otimes_A E, there exists a∈Aa \in A such that ax∈Eax \in E; consequently, for Φ\Phi to be degenerate, it is necessary and sufficient that Φ′\Phi' be so. This proves our assertion.

Proposition 5

Let A be a field, B a commutative algebra of finite dimension n over A, and S a basis of B. In order that B be separable (chap. VIII, § 7, No. 5, def. 1), it is necessary and sufficient that one have* DB/A(S)≠0D_{B/A}(S) \neq 0.

Indeed, let A′A' be the algebraic closure of A, and let B′B' be the algebra A′⊗ABA' \otimes_A B over A′A'. If B is separable, B′B' is semisimple (chap. VIII, § 7, No. 5, cor. of prop. 7) and is therefore the direct sum of n fields isomorphic to A′A' (chap. VIII, § 6, No. 4, cor. of prop. 9). If S′S' denotes the canonical basis of B′B' (identified with A′n{A'}^n), one has DB′/A′(S′)=1D_{B'/A'}(S') = 1, whence DB′/A′(S)≠0D_{B'/A'}(S) \neq 0 (prop. 1) and DB/A(S)≠0D_{B/A}(S) \neq 0 (formula (6)).

Conversely suppose that one has DB/A(S)≠0D_{B/A}(S) \neq 0. To prove that B is separable, it is enough to prove that B′B' is semisimple, that is to say that it admits no nonzero nilpotent element. Now, if x′x' were a nonzero nilpotent element of B′B', it could be taken as the first element of a basis S′S' of B′B', and one would then have TrB′/A′(x′y′)=0\mathrm{Tr}_{B'/A'}(x'y') = 0 for every y′∈S′y' \in S' since a nilpotent endomorphism has its eigenvalues zero (chap. VII, § 5, No. 3, cor. 3 of prop. 8), and hence a zero trace. It would follow that DB′/A′(S′)=0D_{B'/A'}(S') = 0, whence DB′/A′(S)=0D_{B'/A'}(S) = 0 (prop. 1) and DB/A(S)=0D_{B/A}(S) = 0 (formula (6)), contrary to the assumption.

Remark

Suppose that B is an overfield of A. Let S=(x1,…,xn)S = (x_1, \ldots, x_n) be a basis of B, and (s1,…,sn)(s_1, \ldots, s_n) the A-isomorphisms of B into the algebraic closure A′A' of A (each sjs_j being repeated [B:A]i[B : A]_i times). Recall that, for every z∈Bz \in B, one has TrB/A(z)=∑j=1nsj(z)\mathrm{Tr}_{B/A}(z) = \sum_{j=1}^n s_j(z) (chap. VIII, § 12, No. 2, prop. 4).

It follows then from the formula for multiplication of determinants that one has

(7)(det⁡(sj(xi)))2=DB/A(x1,…,xn).(7)\quad (\det(s_j(x_i)))^2 = D_{B/A}(x_1, \ldots, x_n).

This formula shows that Proposition 5 generalizes the separability condition given in chap. V, § 7, No. 2, Remark.

Proposition 6

Let Φ\Phi be an A-bilinear form on E, and K a subring of A such that A is a K-module free of finite dimension q. If (ei)i=1,…,n(e_i)_{i=1,\ldots,n} is a basis of E over A and (aj)j=1,…,q(a_j)_{j=1,\ldots,q} a basis of A over K, then (ajei)(a_j e_i) is a basis of E over K. The mapping Φ′\Phi' of E×EE \times E into K defined by Φ′(x,y)=TrA/K(Φ(x,y))\Phi'(x, y) = \mathrm{Tr}_{A/K}(\Phi(x, y)) is a K-bilinear form on E, and one has

(8)DΦ′(ajei)=NA/K(DΦ(e1,…,en))⋅(DA/K(a1,…,aq))n.(8)\quad D_{\Phi'}(a_j e_i) = N_{A/K}(D_\Phi(e_1, \ldots, e_n)) \cdot (D_{A/K}(a_1, \ldots, a_q))^n.

The first two assertions being evident, it is enough to prove (7). By definition the first member is the determinant of the K-linear endomorphism uu of E defined by

u(ajei)=∑r,sTrA/K(ajarΦ(ei,es))ares.u(a_j e_i) = \sum_{r,s} \mathrm{Tr}_{A/K}(a_j a_r \Phi(e_i, e_s)) a_r e_s.

Consider the A-linear endomorphism vv of E defined by v(ei)=∑sΦ(ei,es)esv(e_i) = \sum_s \Phi(e_i, e_s) e_s, and the K-linear endomorphism ww of E defined by w(ajei)=(∑rTrA/K(ajar)ar)eiw(a_j e_i) = (\sum_r \mathrm{Tr}_{A/K}(a_j a_r) a_r) e_i. We have

w(v(ajei))=w(∑sajΦ(ei,es)es)=∑r,sTrA/K(ajΦ(ei,es)ar)aresw(v(a_j e_i)) = w(\sum_s a_j \Phi(e_i, e_s) e_s) = \sum_{r,s} \mathrm{Tr}_{A/K}(a_j \Phi(e_i, e_s) a_r) a_r e_s

since w(aes)=∑rTrA/K(aar)aresw(a e_s) = \sum_r \mathrm{Tr}_{A/K}(a a_r) a_r e_s for all a∈Aa \in A; thus uu is the composite mapping w∘vw \circ v. Therefore, denoting by vKv_K the mapping vv considered as a K-linear mapping, we have det⁡(u)=det⁡(vK)det⁡(w)\det(u) = \det(v_K) \det(w). Now det⁡(vK)=NA/K(det⁡(v))\det(v_K) = N_{A/K}(\det(v)) (chap. VIII, § 12, no 2, prop. 7), and it is clear that det⁡(v)=DΦ(e1,…,en)\det(v) = D_\Phi(e_1, \ldots, e_n). On the other hand, since each of the A-modules Aei(i=1,…,n)Ae_i (i = 1, \ldots, n) is stable under ww, and the determinant of the restriction of ww to AeiAe_i is det⁡(TrA/K(ajar))=DA/K(a1,…,aq)\det(\mathrm{Tr}_{A/K}(a_j a_r)) = D_{A/K}(a_1, \ldots, a_q), we have det⁡(w)=(DA/K(a1,…,aq))n\det(w) = (D_{A/K}(a_1, \ldots, a_q))^n. Formula (8) therefore reduces to det⁡(u)=det⁡(vK)det⁡(w)\det(u) = \det(v_K) \det(w), a formula proved above.

Corollary (« Transitivity Formula for Discriminants »)

Let K be a commutative ring, A a commutative algebra admitting a finite basis (aj)j=1,…,q(a_j)_{j=1,\ldots,q} over K, and E an algebra over A admitting a finite basis (ei)i=1,…,n(e_i)_{i=1,\ldots,n}. Then (ajei)(a_j e_i) is a basis of E over K, and we have

DE/K(ajei)=NA/K(DE/A(e1,…,en))⋅(DA/K(a1,…,aq))n. D_{E/K}(a_j e_i) = N_{A/K}(D_{E/A}(e_1, \ldots, e_n)) \cdot (D_{A/K}(a_1, \ldots, a_q))^n.

Indeed, if we put Φ(x,y)=Tr⁡E/A(xy)\Phi(x, y) = \operatorname{Tr}_{E/A}(xy), the K-bilinear form Φ′\Phi' of Prop. 6 is Φ′(x,y)=Tr⁡E/K(xy)\Phi'(x, y) = \operatorname{Tr}_{E/K}(xy) by the transitivity formula for traces (Chap. VIII, § 12, no 2, cor. of Prop. 7).

Exercises. — 1) Let A be an algebra of finite rank over a commutative field K, having a unit element. a) Show that if the radical of A is not zero, the bilinear form (x,y)→Tr⁡A/K(xy)(x, y) \to \operatorname{Tr}_{A/K}(xy) on A is degenerate. b) Suppose that K is of characteristic 0. Show that if A is a matrix algebra Mn(K)M_n(K), S the canonical basis of A over K, then DA/K(S)≠0D_{A/K}(S) \neq 0. c) Deduce from a) and b) that, for an algebra A of finite rank over a field K of characteristic 0 to be absolutely semisimple, it is necessary and sufficient that the bilinear form (x,y)→Tr⁡A/K(xy)(x, y) \to \operatorname{Tr}_{A/K}(xy) be non-degenerate (or, what amounts to the same thing, that DA/K(S)≠0D_{A/K}(S) \neq 0 for every basis S of A over K).

¶ 2) Let B be a ring, A a subring of B containing the unit element of B; B is therefore a (A, A)-bimodule; denote by sB^sB (resp. dB^dB) the set B considered as a left (resp. right) A-module, by sB∗^sB^* (resp. dB∗^dB^*) the right (resp. left) A-module dual of sB^sB (resp. dB^dB). For every x′∈sB∗x' \in ^sB^* and every b∈Bb \in B, x→⟨xb,x′⟩x \to \langle xb, x' \rangle is an A-linear form on sB^sB, hence an element of sB∗^sB^* which is denoted by bx′bx'; the mapping (b,x′)→bx′(b, x') \to bx' defines on sB∗^sB^* a left B-module structure (cf. chap. III, 2nd ed., App. II, No. 7). a) Let φ\varphi be a homomorphism of the (A, A)-bimodule B into the (A, A)-bimodule A; in order that the A-bilinear mapping Φ:(x,y)→φ(xy)\Phi : (x, y) \to \varphi(xy) from sB×dB^sB \times ^dB into A be non-degenerate, it is necessary and sufficient that φ(0)\varphi(0) contain no ideal (left or right) of B distinct from {0}\{0\}. One then says that φ\varphi is a Frobenius homomorphism of B into A. b) Let φ\varphi be a Frobenius homomorphism of B into A; show that the mapping dφd_\varphi associated on the right with Φ\Phi is an isomorphism of the left B-module BsB_s onto a submodule of the left B-module sB∗^sB^*. Show that dφd_\varphi is bijective in each of the following two cases: 1° A is a left and right artinian ring satisfying the conditions (N_s) and (N_d) (§ 1, exerc. 11), and sB^sB and dB^dB are A-modules of finite length (use exerc. 11 b) of the § 1); 2° A is an artinian commutative and involutive ring (chap. VIII, § 3, exerc. 11), contained in the center of B, and sB^sB is an A-module of finite length (use exerc. 11 of chap. VIII, § 3).

c) Conversely, show that if BsB_s and sB∗^sB^* are isomorphic, there exists a Frobenius homomorphism of B into A when one is in one of the two cases considered in b) and sB^sB and dB^dB have the same length (use exerc. 11 b) of the § 1).

d) With the hypotheses and notations of a), show that the right annihilator (resp. left annihilator) of a left ideal I (resp. of a right ideal r) of B, is the orthogonal l0l^0 (resp. r0r^0) for the form Φ\Phi of the submodule I (resp. r) of sB^sB (resp. dB^dB).

e) Let φ\varphi be a Frobenius homomorphism of B into A. Show that if A is an involutive artinian ring (chap. VIII, § 3, exerc. 11) the same is true of B, when one supposes in addition that one or the other of the conditions of b) is satisfied (use b) and d)).

¶ 3) Let A be a commutative field, B a finite rank algebra over A, having a unit element.

a) In order that B be a Frobenius algebra, it is necessary and sufficient that there exist a Frobenius homomorphism of B into A (exerc. 2). (Use exerc. 2 c) and e) above, and exerc. 6 b) of chap. VIII, § 13).

b) Let φ\varphi be a Frobenius homomorphism of B into A; every A-linear form on B can then be written in one and only one way x→φ(b′x)x \to \varphi(b'x) (resp. x→φ(xb′′)x \to \varphi(xb'')) where b′b' and b′′b'' belong to B; in order that this form be a Frobenius homomorphism, it is necessary and sufficient that b′b' (resp. b′′b'') be invertible in B.

c) For every x∈Bx \in B, let xσx^\sigma be the unique element (cf. b)) such that φ(xy)=φ(yxσ)\varphi(xy) = \varphi(yx^\sigma) for every y∈By \in B. Show that x→xσx \to x^\sigma is an A-automorphism of B. One says that the Frobenius algebra B is symmetric if σ\sigma is an inner automorphism of B; there is then a Frobenius homomorphism of B into A for which σ\sigma is the identity (cf. b)). It amounts to the same thing to say that the (B, B)-bimodules B and sB∗=dB∗^sB^* = ^dB^* (which is written B*) are isomorphic (exerc. 2 c)).

d) Let E be a left b-module of finite length, E’ its dual, E’* the dual of E’ considered as vector space over A; E’* is endowed with a structure of left B-module by putting, for x′∈E′x' \in E', x′′∈E′∗x'' \in {E'}^*, b∈Bb \in B, ⟨x′,bx′′⟩=⟨x′b,x′′⟩\langle x', bx'' \rangle = \langle x'b, x'' \rangle (Chapter III, 2nd ed., App. II, No. 7). For every x∈Ex \in E, let fE(x)f_E(x) (or simply f(x)f(x)) be the element of E’* such that ⟨x′,f(x)⟩=φ(⟨x,x′⟩)\langle x', f(x) \rangle = \varphi(\langle x, x' \rangle) for every x′∈E′x' \in E'; now prove that ff is a semilinear bijection for the automorphism σ\sigma, of the left B-module E onto the left B-module E’* (use Exercise 10 of Chapter VIII, § 4). For E=BsE = B_s, one has (with the notations of Exercise 2 b)) dΦ(xσ)=fBs(x)d_\Phi(x^\sigma) = f_{B_s}(x) for every x∈Bx \in B.

4) a) Let G be a finite group. Show that the algebra B of the group G over any commutative field A is a symmetric Frobenius algebra (Exercise 3). (Consider the mapping φ\varphi of B into A which associates, with every element x=∑s∈Gξs.sx = \sum_{s \in G} \xi_s . s, the element φ(x)=ξe\varphi(x) = \xi_e, e denoting the identity element of G).

b) Let E be a finite-dimensional vector space of dimension nn over a commutative field A, and B the exterior algebra ∧E\wedge E. Show that B is a Frobenius algebra. (If (ei)1≤i≤n(e_i)_{1 \leq i \leq n} is a basis of E, consider the mapping which associates, with every element xx of B, the coefficient of e1∧e2∧…∧ene_1 \wedge e_2 \wedge \ldots \wedge e_n in the expression of xx using the basis of B corresponding to (ei)(e_i)). For the Frobenius algebra B to be symmetric, it is necessary and sufficient that nn be even or A be of characteristic 2.

¶ 5) a) Show that the tensor product of two Frobenius algebras (resp. Frobenius and symmetric algebras) of finite rank over a commutative field K is a Frobenius algebra (resp. Frobenius and symmetric) (cf. § 1, No. 9, Prop. 9).

b) Let B be an algebra of finite rank over K, L an extension of K of finite degree over K. Show that if the algebra B(L)=B⊗KLB_{(L)} = B \otimes_K L over L is Frobenius (resp. Frobenius and symmetric), the same is true of B. (Use Exercises 2 c) and 3 c) above and Exercise 2 of Chapter VIII, § 2).

6) Show that every absolutely semisimple algebra B of finite rank over a commutative field A is a symmetric Frobenius algebra. (Reduce to the case where B is simple; use Exercise 5 b) above, as well as Proposition 9 of Chapter VIII, § 12, No. 3).

Machine translation of the English, by gpt-5-6-mini, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.