Bourbaki

§ 9. Norms and traces

Throughout this paragraph, KK will denote a commutative ring and AA a unital associative KK-algebra. Every AA-module will be assumed to be given the KK-module structure obtained by restricting the scalars to KK.

1. Norms and traces relative to a module

Definition 1

Let MM be an AA-module admitting a finite basis as a KK-module. For all a∈Aa \in A, let aMa_M be the endomorphism x↦axx \mapsto ax of the KK-module MM. The trace, determinant and characteristic polynomial of aMa_M are called respectively the trace, norm and characteristic polynomial of zz relative to MM.

The trace and norm of aa are therefore elements of KK, denoted respectively by Tr⁡M/K(a)\operatorname{Tr}_{M/K}(a) and NM/K(a)N_{M/K}(a); the characteristic polynomial of aa is an element of K[X]K[X], denoted by PcM/K(a;X)\mathrm{Pc}_{M/K}(a; X). We omit KK in the above notation when there is no risk of confusion.

From the properties of the trace and determinant of an endomorphism (II, § 4, no. 3 and § 8, no. 1) we obtain the relations

Tr⁡M(a+a′)=Tr⁡M(a)+Tr⁡M(a′) \operatorname{Tr}_M(a + a') = \operatorname{Tr}_M(a) + \operatorname{Tr}_M(a')

(2) TrM(aa′)=TrM(a′a)\mathrm{Tr}_M(aa') = \mathrm{Tr}_M(a'a) (3) NM(aa′)=NM(a)NM(a′)N_M(aa') = N_M(a)N_M(a')

for all a,a′a, a' in AA.

Let (ei)1≤i≤n(e_i)_{1 \leq i \leq n} be a basis of the KK-module MM and (mij(a))(m_{ij}(a)) the matrix of the endomorphism aMa_M with respect to this basis. The functions mijm_{ij} are linear forms on the KK-module AA and

(4) TrM(a)=∑i=1nmii(a)\mathrm{Tr}_M(a) = \sum_{i=1}^n m_{ii}(a) (5) NM(a)=det⁡(mij(a))N_M(a) = \det(m_{ij}(a)) (6) Pc(a;X)=det⁡(δijX−mij(a)).\mathrm{Pc}(a; X) = \det(\delta_{ij}X - m_{ij}(a)).

It follows from the method of calculating a determinant (§ 8, no. 11, formula (50)) that

(7) PcM(a;X)=Xn+c1Xn−1+⋯+cn\mathrm{Pc}_M(a; X) = X^n + c_1 X^{n-1} + \cdots + c_n

where

(8) c1=−TrM(a),cn=(−1)nNM(a).c_1 = -\mathrm{Tr}_M(a), \quad c_n = (-1)^n N_M(a).

For λ∈K\lambda \in K,

(9) TrM(λ)=n.λ,NM(λ)=λn,PcM(λ;X)=(X−λ)n.\mathrm{Tr}_M(\lambda) = n.\lambda, \quad N_M(\lambda) = \lambda^n, \quad \mathrm{Pc}_M(\lambda; X) = (X - \lambda)^n.

Let K′K' be a commutative KK-algebra. Let M′=K′⊗KMM' = K' \otimes_K M and A′=K′⊗KAA' = K' \otimes_K A, so that M′M' has an A′A'-module structure (§ 4, Example 2). As a K′K'-module M′M' admits the basis consisting of the 1⊗ei1 \otimes e_i (1≤i≤n1 \leq i \leq n) and the matrix of aMa_M with respect to (ei)(e_i) is equal to the matrix of (1⊗a)M′(1 \otimes a)_M' with respect to (ei′)(e'_i). Then

(12) TrM′(1⊗a)=TrM(a).1,NM′(1⊗a)=NM(a).1\mathrm{Tr}_{M'}(1 \otimes a) = \mathrm{Tr}_M(a).1, \quad N_{M'}(1 \otimes a) = N_M(a).1 PcM′(1⊗a;X)=PcM(a;X).1\mathrm{Pc}_{M'}(1 \otimes a; X) = \mathrm{Pc}_M(a; X).1

for all a∈Aa \in A, where 1 denotes the unit element of K′K'. If in particular we take K′=K[X]K' = K[X], then

(13) PcM/K(a;X)=NM[X]/K[X](X−a).\mathrm{Pc}_{M/K}(a; X) = N_{M[X]/K[X]}(X - a).

2. Properties of norms and traces relative to a module

If MM and M′M' are two isomorphic AA-modules with finite bases over KK, then, for all a∈Aa \in A,

(14) TrM′(a)=TrM(a),NM′(a)=NM(a),PcM′(a;X)=PcM(a;X)\mathrm{Tr}_{M'}(a) = \mathrm{Tr}_M(a), \quad N_{M'}(a) = N_M(a), \quad \mathrm{Pc}_{M'}(a; X) = \mathrm{Pc}_M(a; X)

for if ff is an isomorphism of MM onto M′M', the matrix of aMa_M with respect to a basis BB of MM over KK is the same as the matrix of aM′a_{M'} with respect to the basis f(B)f(B) of M′M'.

Proposition 1

Let M=M0⊃M1⊃…⊃Mr={0}M = M_0 \supset M_1 \supset \ldots \supset M_r = \{0\} be a decreasing sequence of submodules of an AA-module MM such that each of the KK-modules Pi=Mi−1/MiP_i = M_{i-1}/M_i (1≤i≤r1 \leq i \leq r) admits a finite basis. Then the KK-module MM admits a finite basis and

Tr⁡M(a)=∑i=1rTr⁡Pi(a),NM(a)=∏i=1rNPi(a) \operatorname{Tr}_M(a) = \sum_{i=1}^r \operatorname{Tr}_{P_i}(a), \qquad N_M(a) = \prod_{i=1}^r N_{P_i}(a)
(15)

PCM(a;X)=∏i=1rPCPi(a;X). P_{C_M}(a; X) = \prod_{i=1}^r P_{C_{P_i}}(a; X).

Let Bi′B'_i be a basis of PiP_i over KK; then a system of representatives BiB_i of Bi′B'_i (mod. MiM_i) is a basis of a supplementary submodule of the KK-module MiM_i in the KK-module Mi−1M_{i-1} (II, § 1, no. 11, Proposition 21). The union BB of the BiB_i (1≤i≤r1 \leq i \leq r) is a basis of MM over KK. Let XttX_{tt} be the matrix of the endomorphism aPia_{P_i} with respect to the basis Bi′B'_i. It is immediate that the matrix of aMa_M with respect to BB is of the form

(XrrXr,r−1⋯Xr,10Xr−1,r−1⋯Xr−1,1⋮⋮⋱⋮00⋯X11) \begin{pmatrix} X_{rr} & X_{r,r-1} & \cdots & X_{r,1} \\ 0 & X_{r-1,r-1} & \cdots & X_{r-1,1} \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \cdots & X_{11} \end{pmatrix}

and the proposition follows from formulae (4), (5) and (6) of no. 1 and formula (31) of § 8, no. 6.

Proposition 2

Let A,A′A, A' be two KK-algebras, MM an AA-module and M′M' an A′A'-module. Suppose that MM and M′M' are free KK-modules of respective dimensions nn and n′n' and consider M⊗KM′M \otimes_K M' as an (A⊗KA′)(A \otimes_K A')-module (§4§ 4, no. 3, Example 2). Then, for a∈Aa \in A and a′∈A′a' \in A',

Tr⁡M⊗M′(a⊗a′)=Tr⁡M(a)Tr⁡M′(a′) \operatorname{Tr}_{M \otimes M'}(a \otimes a') = \operatorname{Tr}_M(a) \operatorname{Tr}_{M'}(a')
(16)

NM⊗M′(a⊗a′)=(NM(a))n′(NM′(a′))n. N_{M \otimes M'}(a \otimes a') = (N_M(a))^{n'} (N_{M'}(a'))^n.
(17)

Formula (16) follows from II, § 4, no. 4, formula (26) and formula (17) from § 8, no. 6, formula (33).

3. Norm and trace in an algebra

Definition 2

Let AA be a KK-algebra which is a finite-dimensional free KK-module. For every element a∈Aa \in A, the trace (resp. norm,\footnote{† This notion should not be confused with the notion for norm in an algebra over a valued field (\emph{General Topology}), IX, § 3, no. 7.} resp. characteristic polynomial) of aa relative to AA and KK is the trace (resp. determinant, resp. characteristic polynomial) of the endomorphism x↦axx \mapsto ax of the KK-module AA.

The trace, norm and characteristic polynomial of a∈Aa \in A relative to AA and KK are denoted by TrA/K(a)\mathrm{Tr}_{A/K}(a), NA/K(a)N_{A/K}(a) and PcA/K(a;X)\mathrm{Pc}_{A/K}(a; X); we omit KK and even AA from this notation when there is no risk of confusion. Note that the trace (resp. norm, characteristic polynomial) of a∈Aa \in A is just the trace (resp. norm, characteristic polynomial) of aa relative to the AA-module AsA_s.

Suppose that AA is the product A1×A2×⋯×AmA_1 \times A_2 \times \cdots \times A_m of a finite number finite-dimensional algebras over KK which are free KK-modules. Using the above remark and Proposition 1 of no. 2, we have, for every element

a=(a1,…,am)∈A, a = (a_1, \ldots, a_m) \in A,

TrA/K(a)=∑i=1mTrAi/K(ai),NA/K(a)=∏i=1mNAi/K(ai) \mathrm{Tr}_{A/K}(a) = \sum_{i=1}^m \mathrm{Tr}_{A_i/K}(a_i), \qquad N_{A/K}(a) = \prod_{i=1}^m N_{A_i/K}(a_i)
(18)

PcA/K(a;X)=∏i=1mPcAi/K(ai;X). \mathrm{Pc}_{A/K}(a; X) = \prod_{i=1}^m \mathrm{Pc}_{A_i/K}(a_i; X).

Similarly, Proposition 2 of no. 2 shows that if AA and A′A' are two algebras, which are free KK-modules, of finite dimensions n,n′n, n' respectively over KK, then, for a∈A,a′∈A′a \in A, a' \in A'.

TrA⊗A′(a⊗a′)=TrA(a) TrA′(a′) \mathrm{Tr}_{A \otimes A'}(a \otimes a') = \mathrm{Tr}_A(a) \, \mathrm{Tr}_{A'}(a')
(19)

NA⊗A′(a⊗a′)=(NA(a))n′(NA′(a′))n. N_{A \otimes A'}(a \otimes a') = (N_A(a))^{n'} (N_{A'}(a'))^n.
(20)

Finally let AA be a finite-dimensional algebra over KK which is a free KK-module, hh a homomorphism of KK into a commutative ring K′K' and A′=A(K′)A' = A_{(K')} the K′K'-algebra derived from AA by extending the scalars by means of hh. It follows from formula (12) of no. 1 that, for all a∈Aa \in A,

TrA′/K′(1⊗a)=h(TrA/K(a)),NA′/K′(1⊗a)=h(NA/K(a)) \mathrm{Tr}_{A'/K'}(1 \otimes a) = h(\mathrm{Tr}_{A/K}(a)), \qquad N_{A'/K'}(1 \otimes a) = h(N_{A/K}(a))
(21)

PcA′/K′(1⊗a;X)=hˉ(PcA/K(a;X)) \mathrm{Pc}_{A'/K'}(1 \otimes a; X) = \bar{h}(\mathrm{Pc}_{A/K}(a; X))

where hˉ\bar{h} is the homomorphism K[X]→K′[X]K[X] \to K'[X] derived from hh. In particular, for K′=K[X]K' = K[X], we obtain, writing A[X]=A⊗KK[X]A[X] = A \otimes_K K[X],

PcA/K(a;X)=NA[X]/K[X](X−a). \mathrm{Pc}_{A/K}(a; X) = N_{A[X]/K[X]}(X - a).
(22)

More generally, if K′K' is a commutative KK-algebra and A′=A⊗KK′A' = A \otimes_K K', then, for all x∈A′x \in A',

PcA/K(a;x)=NA′/K′(x−a). \mathrm{Pc}_{A/K}(a; x) = N_{A'/K'}(x - a).

Example

(1) Let AA be a quadratic algebra over KK of type (α,β)(\alpha, \beta) and (e1,e2)(e_1, e_2) a basis of type (α,β)(\alpha, \beta) (§ 2, no. 3). For x=ξe1+ηe2x = \xi e_1 + \eta e_2, TrA/K(x)=2ξ+βη\mathrm{Tr}_{A/K}(x) = 2\xi + \beta \eta and NA/K(x)=ξ2+βξη−αη2N_{A/K}(x) = \xi^2 + \beta \xi \eta - \alpha \eta^2; these functions are therefore identical with the Cayley trace and norm of xx (§ 2, no. 24).

(2) Let AA be a quaternion algebra over KK. A direct calculation allows us to verify that TrA/K(x)=2T(x)\mathrm{Tr}_{A/\mathbf{K}}(x) = 2T(x) and NA/K(x)=(N(x))2N_{A/\mathbf{K}}(x) = (N(x))^2, where T and N are the Cayley trace and norm (§ 2, no. 4).

(3) Let A=Mn(K)A = \mathbf{M}_n(\mathbf{K}) and let the canonical basis (Eij)(E_{ij}) of A (II, § 10, no. 3) be arranged in lexicographic order. Then it is immediately seen that for every matrix X=∑i,jξijEijX = \sum_{i,j} \xi_{ij} E_{ij} the matrix (of order n2n^2) of the endomorphism Y↦XYY \mapsto XY is of the form

(X0…00X…0…………00…X) \begin{pmatrix} X & 0 & \ldots & 0 \\ 0 & X & \ldots & 0 \\ \ldots & \ldots & \ldots & \ldots \\ 0 & 0 & \ldots & X \end{pmatrix}

whence TrA/K(X)=n⋅Tr(X)\mathrm{Tr}_{A/\mathbf{K}}(X) = n \cdot \mathrm{Tr}(X) and NA/K(X)=(det⁡(X))nN_{A/\mathbf{K}}(X) = (\det(X))^n.

4. Properties of norms and traces in an algebra

Proposition 3

Let A be a K-algebra admitting a finite basis. For an element a∈Aa \in A to be invertible, it is necessary and sufficient that its NA/K(a)N_{A/\mathbf{K}}(a) be invertible in K\mathbf{K}.

If aa admits an inverse a′a' in AA, then

NA/K(a)NA/K(a′)=NA/K(aa′)=NA/K(1)=1 N_{A/\mathbf{K}}(a) N_{A/\mathbf{K}}(a') = N_{A/\mathbf{K}}(aa') = N_{A/\mathbf{K}}(1) = 1

by formula (3) of no. 1. Conversely, if NA/K(a)N_{A/\mathbf{K}}(a) is invertible, the endomorphism h:x↦axh : x \mapsto ax is bijective (§ 8, no. 2, Theorem 1). Then there exists a′∈Aa' \in A such that aa′=1aa' = 1; then h(a′a−1)=aa′a−a=(aa′−1)a=0h(a'a - 1) = aa'a - a = (aa' - 1)a = 0, whence aa′=1aa' = 1 since hh is injective. Hence a′a' is the inverse of aa.

Proposition 4

Let A be a K-algebra admitting a finite basis. For all a∈Aa \in A, PcA/K(a;a)=0\mathrm{Pc}_{A/\mathbf{K}}(a; a) = 0.

This follows immediately from the Cayley-Hamilton theorem (§ 8, no. 11, Proposition 20).

Proposition 5

Let A be a K-algebra and m a two-sided ideal of A. Suppose that A0=A/mA_0 = A/m is a free K-module of finite dimension n, that there exists an integer r>0r > 0 such that mr={0}m^r = \{0\} and that mi−1/mim^{i-1}/m^i is a free A0A_0-module of finite dimension sis_i for 1≤i≤r1 \leq i \leq r. Let s=s1+⋯+srs = s_1 + \cdots + s_r and for all a∈Aa \in A let a0a_0 denote the class of aa mod. m. Then A is a free K-module of dimension nsns and, for all a∈Aa \in A,

TrA(a)=s⋅TrA0(a0),NA(a)=(NA0(a0))s \mathrm{Tr}_A(a) = s \cdot \mathrm{Tr}_{A_0}(a_0), \qquad N_A(a) = (N_{A_0}(a_0))^s
PcA(a;X)=(PcA0(a0;X))s. \mathrm{Pc}_A(a; X) = (\mathrm{Pc}_{A_0}(a_0; X))^s.

By virtue of II, § 1, no. 13, Proposition 25, mi−1/mim^{i-1}/m^i is a free K-module of dimension nsins_i. Hence Proposition 1 of no. 2 can be applied with Pi=mi−1/miP_i = m^{i-1}/m^i;

this shows in the first place that AA is a free KK-module of dimension n(s1+⋯+sr)=nsn(s_1 + \cdots + s_r) = ns. Moreover, the hypothesis implies that the AA-module PiP_i is isomorphic to a direct sum of sis_i submodules isomorphic to the AA-module A0A_0; by Proposition 1 of no. 2, therefore NPi(a)=NA0(a)siN_{P_i}(a) = N_{A_0}(a)^{s_i}; finally therefore

NA(a)=NA0(a)s. N_A(a) = N_{A_0}(a)^s.
In this formula NA0(a)N_{A_0}(a) is defined by considering A0A_0 as a left AA-module and it is equal to the determinant of the KK-linear mapping x↦axx \mapsto ax of A0A_0 into itself; but, as ax=a0xax = a_0x for x∈A0x \in A_0, NA0(a)=NA0(a0)N_{A_0}(a) = N_{A_0}(a_0), which completes the proof of formula (23) for the norm. The two others are shown analogously.

Proposition 6

Let AA be a commutative KK-algebra admitting a finite basis over KK and VV an AA-module admitting a finite basis over AA. Then VV admits a finite basis over KK and for every AA-endomorphism uu of VV, if uKu_K is the mapping uu considered as a KK-endomorphism of VV,*

Tr⁡(uK)=Tr⁡A/K(Tr⁡(u)),det⁡(uK)=NA/K(det⁡(u))Pc⁡(uK;X)=NA[X]/K[X](Pc⁡(u;X)). \begin{align*} \operatorname{Tr}(u_K) &= \operatorname{Tr}_{A/K}(\operatorname{Tr}(u)), \qquad \det(u_K) = N_{A/K}(\det(u)) \\ \operatorname{Pc}(u_K; X) &= N_{A[X]/K[X]}(\operatorname{Pc}(u; X)). \end{align*}
(24)

Let (ai)1≤i≤m(a_i)_{1 \leq i \leq m} be a basis of AA over KK and (ej)1≤j≤n(e_j)_{1 \leq j \leq n} a basis of VV over AA; then (aiej)(a_i e_j) is a basis of VV over KK (II, § 1, no. 13, Proposition 25). On the other hand the third of formulae (24) can be deduced from the second applied to the endomorphism X−uˉX - \bar{u} of the A[X]A[X]-module A[X]⊗AVA[X] \otimes_A V (§ 8, no. 10). It will therefore suffice to show the first two formulae in (24). We shall first establish the following lemma:

*Lemma 1. Let XijX_{ij} (1≤i≤n, 1≤j≤n1 \leq i \leq n,\ 1 \leq j \leq n) be n2n^2 indeterminates, XX the square matrix (Xij)(X_{ij}) of order nn and D(X11,…,Xnn)∈Z[X11,…,Xnn]D(X_{11}, \ldots, X_{nn}) \in \mathbf{Z}[X_{11}, \ldots, X_{nn}] the determinant det⁡(X)\det(X). On the other hand let AA be a commutative ring, MijM_{ij} (1≤i≤n, 1≤j≤n1 \leq i \leq n,\ 1 \leq j \leq n) n2n^2 matrices of order mm over AA, which are pairwise permutable, and MM the square matrix of order mnmn over AA which can be expressed as a square matrix of matrices (II, § 10, no. 7)

M=(M11M12⋯M1nM21M22⋯M2n⋯⋯⋯⋯Mn1Mn2⋯Mnn) M = \begin{pmatrix} M_{11} & M_{12} & \cdots & M_{1n} \\ M_{21} & M_{22} & \cdots & M_{2n} \\ \cdots & \cdots & \cdots & \cdots \\ M_{n1} & M_{n2} & \cdots & M_{nn} \end{pmatrix}
Then the determinant of MM is equal to the determinant of the square matrix
D(M11,…,Mnn) D(M_{11}, \ldots, M_{nn})
of order mm.

We proceed by induction on nn, the cases n=0n = 0 and n=1n = 1 being trivial.

Let ZZ be a new indeterminate and NijN_{ij} the matrix Mij+δijZImM_{ij} + \delta_{ij} Z I_m (δij\delta_{ij} the Kronecker index). If Dij(X11,…,Xnn)D^{ij}(X_{11}, \ldots, X_{nn}) is the cofactor of XijX_{ij} in the matrix XX (§ 8, no. 6), then

(25)XjiDki(X11,…,Xnn)=δjkD(X11,…,Xnn) \text{(25)} \quad X_{ji} D^{ki}(X_{11}, \ldots, X_{nn}) = \delta_{jk} D(X_{11}, \ldots, X_{nn})

(§ 8, no. 6, formula (28)). We write Nij′=Dij(N11,…,Nnn)N'_{ij} = D^{ij}(N_{11}, \ldots, N_{nn}), which is a square matrix of order mm over A[Z]\mathbf{A}[Z] and consider the product N.UN . U, where

U=(N11′0⋯0N12′Im⋯0⋯⋯⋯⋯N1n′0⋯Im), U = \begin{pmatrix} N'_{11} & 0 & \cdots & 0 \\ N'_{12} & I_m & \cdots & 0 \\ \cdots & \cdots & \cdots & \cdots \\ N'_{1n} & 0 & \cdots & I_m \end{pmatrix},
N=(N11N12⋯N1nN21N22⋯N2n⋯⋯⋯⋯Nn1Nn2⋯Nnn) N = \begin{pmatrix} N_{11} & N_{12} & \cdots & N_{1n} \\ N_{21} & N_{22} & \cdots & N_{2n} \\ \cdots & \cdots & \cdots & \cdots \\ N_{n1} & N_{n2} & \cdots & N_{nn} \end{pmatrix}

Performing this product in blocks (II, § 10, no. 5) and using formulae (25), we obtain

N.U=(PN12⋯N1n0N22⋯N2n⋯⋯⋯⋯0Nn2⋯Nnn) N . U = \begin{pmatrix} P & N_{12} & \cdots & N_{1n} \\ 0 & N_{22} & \cdots & N_{2n} \\ \cdots & \cdots & \cdots & \cdots \\ 0 & N_{n2} & \cdots & N_{nn} \end{pmatrix}

where we have written P=D(N11,…,Nnn)P = D(N_{11}, \ldots, N_{nn}). Let

Q=(N22⋯N2n⋯⋯⋯Nn2⋯Nnn) Q = \begin{pmatrix} N_{22} & \cdots & N_{2n} \\ \cdots & \cdots & \cdots \\ N_{n2} & \cdots & N_{nn} \end{pmatrix}

which is a matrix of order m(n−1)m(n-1); then (§ 8, no. 6, formula (31)) (det NN) (det UU) = (det PP)(det QQ) and det U=det⁡N11′U = \det N'_{11}. But by the induction hypothesis, det Q=det⁡(D11(N11,…,Nnn))=det⁡N11′Q = \det(D^{11}(N_{11}, \ldots, N_{nn})) = \det N'_{11} and by virtue of the definition of the NijN_{ij}, clearly det QQ is a polynomial in A[Z]\mathbf{A}[Z] of degree m(n−1)m(n-1), whose term in Zm(n−1)Z^{m(n-1)} has coefficient 1 ; it follows immediately that det QQ is not a divisor of zero in the graded algebra A[Z]\mathbf{A}[Z]. We therefore conclude that det N=det⁡(D(N11,…,Nnn))N = \det(D(N_{11}, \ldots, N_{nn})) in A[Z]\mathbf{A}[Z]; if we substitute 0 for ZZ in these polynomials, then det M=det⁡(D(M11,…,Mnn))M = \det(D(M_{11}, \ldots, M_{nn})).

Having shown this lemma, the K-module V is the direct sum of the K-modules AejAe_j (1≤j≤n1 \leq j \leq n); we write u(ej)=∑k=1ncjkeku(e_j) = \sum_{k=1}^n c_{jk} e_k. For every element xej∈Aejxe_j \in Ae_j, with x∈Ax \in A, the component of u(xej)u(xe_j) in AekAe_k is xcjkekxc_{jk} e_k; it follows that the matrix of uku_k with respect to the basis (aiej)(a_i e_j) of the K-module V can be expressed in the form of a square matrix of matrices (Mjk)(M_{jk}), where MjkM_{jk} is the matrix of the K-linear mapping xej↦xcjkekxe_j \mapsto xc_{jk} e_k of AejAe_j into AekAe_k with respect to the bases (aiej)1≤i≤m(a_i e_j)_{1 \leq i \leq m} and (aiek)1≤i≤m(a_i e_k)_{1 \leq i \leq m} of these two K-modules (II, § 10, no. 5). If for all t∈At \in A, M(t)M(t) denotes the matrix, with respect to the basis (ai)1≤i≤m(a_i)_{1 \leq i \leq m} of A over K, of the endomorphism x↦xtx \mapsto xt of A, then Mjk=M(cjk)M_{jk} = M(c_{jk}); as t↦M(t)t \mapsto M(t) is a ring homomorphism the matrices MjkM_{jk} are permutable with one another. Then

det⁡uK=det⁡(D(M11,…,Mnn)) \det u_K = \det(D(M_{11}, \ldots, M_{nn}))
by Lemma 1. But as t↦M(t)t \mapsto M(t) is a ring homomorphism, D(M11,…,Mnn)D(M_{11}, \ldots, M_{nn}) is the matrix of the K-endomorphism x↦x.det⁡(cjk)x \mapsto x.\det(c_{jk}) of A with respect to basis (ai)(a_i); by definition its determinant is therefore NA/K(det⁡(u))N_{A/K}(\det(u)), which proves the second formula of (24). On the other hand,
Tr⁡(uK)=∑j=1nTr⁡(Mjj)=∑j=1nTr⁡A/K(cjj)=Tr⁡A/K(∑j=1ncjj)=Tr⁡A/K(Tr⁡(u)) \operatorname{Tr}(u_K) = \sum_{j=1}^n \operatorname{Tr}(M_{jj}) = \sum_{j=1}^n \operatorname{Tr}_{A/K}(c_{jj}) = \operatorname{Tr}_{A/K}\left( \sum_{j=1}^n c_{jj} \right) = \operatorname{Tr}_{A/K}(\operatorname{Tr}(u))
and the proof of Proposition 6 is complete.

Corollary

Let A be a commutative K-algebra admitting a finite basis over K and B an A-algebra admitting a finite basis over A. Then B admits a finite basis over K, and, for all b∈Bb \in B ("transitivity formulae")*

Tr⁡B/K(b)=Tr⁡A/K(Tr⁡B/A(b)),NB/K(b)=NA/K(NB/A(b))PCB/K(b;X)=NA[X]/K[X](PCB/A(b;X)). \begin{align*} \operatorname{Tr}_{B/K}(b) &= \operatorname{Tr}_{A/K}(\operatorname{Tr}_{B/A}(b)), \quad N_{B/K}(b) = N_{A/K}(N_{B/A}(b)) \\ P_{C_{B/K}}(b; X) &= N_{A[X]/K[X]}(P_{C_{B/A}}(b; X)). \end{align*}
(26) This follows immediately from Proposition 6, setting V=BV = B and u(x)=bxu(x) = bx.

Remark

Suppose that the homomorphism λ↦λ.1\lambda \mapsto \lambda . 1 of K into A is injective and let K be identified with its image in A; suppose that A admits a finite basis (ei)1≤i≤n(e_i)_{1 \leq i \leq n} as a K-module. Let s be an automorphism of A such that s(K)=Ks(K) = K. Let a be an element of A; then, by transporting the structure

Tr⁡A/K(s(a))=s(Tr⁡A/K(a))NA/K(s(a))=s(NA/K(a)). \begin{align*} \operatorname{Tr}_{A/K}(s(a)) &= s(\operatorname{Tr}_{A/K}(a)) \\ N_{A/K}(s(a)) &= s(N_{A/K}(a)). \end{align*}
(27) (28) Consider also a derivation D of A (§10§ 10, no. 2) such that D(K)⊂KD(K) \subset K and write D(ei)=∑j=1nejμjiD(e_i) = \sum_{j=1}^n e_j \mu_{ji} where μji∈K\mu_{ji} \in K; write
aei=∑j=1nejλjiwithλji∈K. ae_i = \sum_{j=1}^n e_j \lambda_{ji} \quad \text{with} \quad \lambda_{ji} \in K.

Then

D(a)ei+aD(ei)=D(aei)=∑j=1n(D(ej)λji+ejD(λji)). D(a)e_i + aD(e_i) = D(ae_i) = \sum_{j=1}^n (D(e_j)\lambda_{ji} + e_j D(\lambda_{ji})).
It follows that
D(a)ei=∑j=1nejvji D(a)e_i = \sum_{j=1}^n e_j v_{ji}
with vji=D(λji)+∑k=1n(μjkλki−λjkμki)v_{ji} = D(\lambda_{ji}) + \sum_{k=1}^n (\mu_{jk}\lambda_{ki} - \lambda_{jk}\mu_{ki}). As ∑i,k(μikλki−λikμki)=0\sum_{i,k} (\mu_{ik}\lambda_{ki} - \lambda_{ik}\mu_{ki}) = 0, therefore Tr⁡A/K(D(a))=∑i=1nD(λii)\operatorname{Tr}_{A/\mathbf{K}}(D(a)) = \sum_{i=1}^n D(\lambda_{ii}), in other words
Tr⁡A/K(D(a))=D(Tr⁡A/K(a)).∗ \operatorname{Tr}_{A/\mathbf{K}}(D(a)) = D(\operatorname{Tr}_{A/\mathbf{K}}(a)). *

5. Discriminant of an algebra

Definition 3

Let A be a K-algebra admitting a finite basis of n elements. The discriminant of a sequence (x1,…,xn)(x_1, \ldots, x_n) of n elements of A, with respect to K, denoted by DA/K(x1,…,xn)D_{A/\mathbf{K}}(x_1, \ldots, x_n), is the discriminant of the square matrix*

(Tr⁡A/K(xixj))1≤i≤n,1≤j≤n. (\operatorname{Tr}_{A/\mathbf{K}}(x_i x_j))_{1 \leq i \leq n, 1 \leq j \leq n}.
Consider first a basis (ei)1≤i≤n(e_i)_{1 \leq i \leq n} of A over K and write
eiej=∑k=1ncijkekwithcijk∈K. e_i e_j = \sum_{k=1}^n c_{ij k} e_k \quad \text{with} \quad c_{ij k} \in \mathbf{K}.
Then Tr⁡A/K(ei)=∑s=1nciss\operatorname{Tr}_{A/\mathbf{K}}(e_i) = \sum_{s=1}^n c_{iss}, whence Tr⁡A/K(eiej)=∑k,scijkckss\operatorname{Tr}_{A/\mathbf{K}}(e_i e_j) = \sum_{k,s} c_{ij k} c_{kss} and therefore
DA/K(e1,…,en)=det⁡((∑k,scijkckss)1≤i≤n,1≤j≤n). D_{A/\mathbf{K}}(e_1, \ldots, e_n) = \det \left( \left( \sum_{k,s} c_{ij k} c_{kss} \right)_{1 \leq i \leq n, 1 \leq j \leq n} \right).
Now let (xi)1≤i≤n,(xi′)1≤i≤n(x_i)_{1 \leq i \leq n}, (x'_i)_{1 \leq i \leq n} be two sequences of n elements of A and suppose that there exists a square matrix of order n, M=(mij)M = (m_{ij}), with coefficients in K, such that xi=∑j=1nmijxj′x_i = \sum_{j=1}^n m_{ij} x'_j for 1≤i≤n1 \leq i \leq n. We write
T=(Tr⁡A/K(xixj))1≤i≤n,1≤j≤n,T′=(Tr⁡A/K(xi′xj′))1≤i≤n,1≤j≤n. T = (\operatorname{Tr}_{A/\mathbf{K}}(x_i x_j))_{1 \leq i \leq n, 1 \leq j \leq n}, \qquad T' = (\operatorname{Tr}_{A/\mathbf{K}}(x'_i x'_j))_{1 \leq i \leq n, 1 \leq j \leq n}.
Then Tr⁡A/K(xixj)=∑p,qmipmjqTr⁡A/K(xp′xq′)\operatorname{Tr}_{A/\mathbf{K}}(x_i x_j) = \sum_{p,q} m_{ip} m_{jq} \operatorname{Tr}_{A/\mathbf{K}}(x'_p x'_q), whence T=M⋅T′⋅tMT = M \cdot T' \cdot {}^t M; the rule of multiplication of determinants therefore gives
det⁡T=det⁡M⋅det⁡T′⋅det⁡tM=(det⁡M)2det⁡T′ \det T = \det M \cdot \det T' \cdot \det {}^t M = (\det M)^2 \det T'
whence finally
DA/K(x1,…,xn)=(det⁡M)2DA/K(x1′,…,xn′). D_{A/\mathbf{K}}(x_1, \ldots, x_n) = (\det M)^2 D_{A/\mathbf{K}}(x'_1, \ldots, x'_n).

The above formula shows in particular that the discriminants of two bases of A over K differ by the square of an invertible element of K and therefore generate the same (principal) ideal of K. This ideal ΔA/K\Delta_{A/K} is called the discriminant ideal of A over K; by formula (32) the discriminant of every sequence of n elements of A which differ only in the order of the terms have the same discriminant, for the determinant of a permutation matrix is equal to ±1\pm 1.

Example

(1) If A is a quadratic algebra of type (α,β)(\alpha, \beta) over K, then (in the notation of § 2, no. 3) Tr⁡(e1)=2,Tr⁡(e2)=β\operatorname{Tr}(e_1) = 2, \operatorname{Tr}(e_2) = \beta,

Tr⁡(e22)=αTr⁡(e1)+βTr⁡(e2)=2α+β2, \operatorname{Tr}(e_2^2) = \alpha \operatorname{Tr}(e_1) + \beta \operatorname{Tr}(e_2) = 2\alpha + \beta^2,

whence DA/K(e1,e2)=β2+4αD_{A/K}(e_1, e_2) = \beta^2 + 4\alpha.

(2) Let A=K[X]/K[X]PA = K[X]/K[X]P, where P(X)=X3+pX+qP(X) = X^3 + pX + q, so that if x is the image of X in A, 1, x, x2x^2 form a basis of A over K and x3=−px−qx^3 = -px - q. It is seen immediately that Tr⁡(1)=3,Tr⁡(x)=0,Tr⁡(x2)=−2p\operatorname{Tr}(1) = 3, \operatorname{Tr}(x) = 0, \operatorname{Tr}(x^2) = -2p, taking account of the relation x3=−px−q,Tr⁡(x3)=−3qx^3 = -px - q, \operatorname{Tr}(x^3) = -3q and Tr⁡(x4)=2p2\operatorname{Tr}(x^4) = 2p^2, whence easily DA/K(1,x,x2)=−4p3−27q2D_{A/K}(1, x, x^2) = -4p^3 - 27q^2.

(3) Let A be a quaternion algebra of type (α,β,γ)(\alpha, \beta, \gamma) over K and (1,i,j,k)(1, i, j, k) a basis of A of type (α,β,γ)(\alpha, \beta, \gamma); taking account of § 3, no. 5, formula (30), it is easily found that Tr⁡(1)=4,Tr⁡(i)=2β,Tr⁡(j)=Tr⁡(k)=0\operatorname{Tr}(1) = 4, \operatorname{Tr}(i) = 2\beta, \operatorname{Tr}(j) = \operatorname{Tr}(k) = 0, then

DA/K(1,i,j,k)=−16γ2(β2+4α)2. D_{A/K}(1, i, j, k) = -16\gamma^2(\beta^2 + 4\alpha)^2.

(4) Let A=Mn(K)A = M_n(K) and consider the canonical basis (Eij)1≤i≤n,1≤j≤n(E_{ij})_{1 \leq i \leq n, 1 \leq j \leq n} of A over K (II, § 10, no. 3). It is immediate that Tr⁡A/K(Eij)=0\operatorname{Tr}_{A/K}(E_{ij}) = 0 if j≠ij \neq i and Tr⁡A/K(Eii)=n\operatorname{Tr}_{A/K}(E_{ii}) = n for all i; it follows without difficulty that the matrix (Tr⁡(EijEhk))(\operatorname{Tr}(E_{ij}E_{hk})) of order n2n^2 is of the form n.Pn.P, where P is a permutation matrix, whence DA/K((Eij))=±nn2D_{A/K}((E_{ij})) = \pm n^{n^2}.

Exercises

See the exercises for § 9.

Transcribed from the printed edition, 1998, Springer. Not translated.