Bourbaki

Proposition 10

Tag 0610, permanent. Cite it as [Bourbaki, Tag 0610].

Algebra, chapter III, § 4. Tensor products of algebras

Let A\mathbf{A} be a commutative ring and (Δi)i∈I(\Delta_i)_{i \in I} a finite family of commutative monoids written additively; for each i∈Ii \in I let EiE_i be a graded A\mathbf{A}-algebra of type Δi\Delta_i. Finally, let (εij)(\varepsilon_{ij}) be a system of commutation factors over the Δi\Delta_i with values in A\mathbf{A}. Then there exist a graded A\mathbf{A}-algebra EE of type Δ=∏i∈IΔi\Delta = \prod_{i \in I} \Delta_i and for each i∈Ii \in I an algebra homomorphism hi:Ei→Eh_i : E_i \to E, with the following properties:

(i) If ϕi:Δi→Δ\phi_i : \Delta_i \to \Delta is the canonical homomorphism, then hih_i is a graded homomorphism (II, § 11, no. 2), in other words, hi(Eiαi)⊂Eϕi(αi)h_i(E_i^{\alpha_i}) \subset E^{\phi_i(\alpha_i)}, where (Eiαi)(E_i^{\alpha_i}) and (Eα)(E^{\alpha}) denote the respective graduations on EiE_i and EE.

(ii) If i≠ji \neq j and xix_i (resp. xjx_j) is a homogeneous element of EiE_i (resp. EjE_j) of degree αi∈Δi\alpha_i \in \Delta_i (resp. βj∈Δj\beta_j \in \Delta_j), then

hi(xi)hj(xj)=εij(αi,βj)hj(xj)hi(xi). h_i(x_i) h_j(x_j) = \varepsilon_{ij}(\alpha_i, \beta_j) h_j(x_j) h_i(x_i).

(iii) For every A\mathbf{A}-algebra FF and every system of homomorphisms fi:Ei→Ff_i : E_i \to F satisfying the conditions

fi(xi)fj(xj)=εij(αi,βj)fj(xj)fi(xi), f_i(x_i) f_j(x_j) = \varepsilon_{ij}(\alpha_i, \beta_j) f_j(x_j) f_i(x_i),

where i,j,xi,xj,αi,βji, j, x_i, x_j, \alpha_i, \beta_j are as in (ii), then there exists one and only one algebra homomorphism f:E→Ff : E \to F such that fi=f∘hif_i = f \circ h_i for all i∈Ii \in I. Moreover the underlying A\mathbf{A}-module of EE is the tensor product ⨂i∈IEi\bigotimes_{i \in I} E_i.

Consider the A-module\mathbf{A}\text{-module} E=⨂i∈IEiE = \bigotimes_{i \in I} E_i; it is identified with the direct sum of the submodules EαE^{\alpha}, where, for each α=(αi)∈Δ\alpha = (\alpha_i) \in \Delta, we write Eα=⨂i∈IEiαiE^{\alpha} = \bigotimes_{i \in I} E_i^{\alpha_i}; the EαE^{\alpha} therefore form a graduation of type Δ\Delta on the A\mathbf{A}-module EE. We shall define on EE a graded A\mathbf{A}-algebra structure of type Δ\Delta. For this let II be given a total ordering; for every ordered pair of elements α=(αi),β=(βi)\alpha = (\alpha_i), \beta = (\beta_i) of Δ\Delta, we must first define an A\mathbf{A}-bilinear mapping of Eα×EβE^{\alpha} \times E^{\beta} into Eα+βE^{\alpha + \beta}, or alternatively an A\mathbf{A}-linear mapping mαβm_{\alpha \beta} of Eα⊗AEβE^{\alpha} \otimes_{\mathbf{A}} E^{\beta} into Eα+βE^{\alpha + \beta}. We shall define mαβm_{\alpha \beta} by the condition

mαβ((⨂i∈Ixi)⊗(⨂i∈Iyi))=ε(α,β)⨂i∈I(xiyi) m_{\alpha \beta} \left( \left( \bigotimes_{i \in I} x_i \right) \otimes \left( \bigotimes_{i \in I} y_i \right) \right) = \varepsilon(\alpha, \beta) \bigotimes_{i \in I} (x_i y_i)

for xi∈Eiαi,yi∈Eiαix_i \in E_i^{\alpha_i}, y_i \in E_i^{\alpha_i}, where

ε(α,β)=∏i>jεij(αi,βj). \varepsilon(\alpha, \beta) = \prod_{i > j} \varepsilon_{ij}(\alpha_i, \beta_j).

The right hand side of (27) obviously belongs to Eα+βE^{\alpha + \beta} and the mapping (x1,…,xn,y1,…,yn)↦ε(α,β)⨂i∈I(xiyi)(x_1, \ldots, x_n, y_1, \ldots, y_n) \mapsto \varepsilon(\alpha, \beta) \bigotimes_{i \in I} (x_i y_i) is A\mathbf{A}-multilinear in the product of the EiαiE_i^{\alpha_i} and the EiβiE_i^{\beta_i} (1≤i≤n1 \leq i \leq n). Then it must be proved that the multiplication thus defined on E is associative; now, if γ=(γi)\gamma = (\gamma_i) is a third element of Δ\Delta and zi∈Eiγi′z_i \in E_i^{\gamma_i'} for 1≤i≤n1 \leq i \leq n, then

((⨂ixi)(⨂iyi))(⨂izi)=ε(α+β,γ)ε(α,β)⨂i(xiyizi) \left( \left( \bigotimes_i x_i \right) \left( \bigotimes_i y_i \right) \right) \left( \bigotimes_i z_i \right) = \varepsilon(\alpha + \beta, \gamma) \varepsilon(\alpha, \beta) \bigotimes_i (x_i y_i z_i)
(⨂ixi)((⨂iyi)(⨂izi))=ε(α,β+γ)ε(β,γ)⨂i(xiyizi) \left( \bigotimes_i x_i \right) \left( \left( \bigotimes_i y_i \right) \left( \bigotimes_i z_i \right) \right) = \varepsilon(\alpha, \beta + \gamma) \varepsilon(\beta, \gamma) \bigotimes_i (x_i y_i z_i)

and it reduces to verifying the identity

ε(α+β,γ)ε(α,β)=ε(α,β+γ)ε(β,γ). \varepsilon(\alpha + \beta, \gamma) \varepsilon(\alpha, \beta) = \varepsilon(\alpha, \beta + \gamma) \varepsilon(\beta, \gamma).

But the latter follows immediately from the relations

ε(α+β,γ)=ε(α,β)ε(β,γ) \varepsilon(\alpha + \beta, \gamma) = \varepsilon(\alpha, \beta) \varepsilon(\beta, \gamma)
ε(α,β+γ)=ε(α,β)ε(α,γ) \varepsilon(\alpha, \beta + \gamma) = \varepsilon(\alpha, \beta) \varepsilon(\alpha, \gamma)

themselves immediate consequences of the definition (28) and (22) and (23).

If, for all i∈Ii \in I, eie_i denotes the unit element of EiE_i, we know that eie_i is homogeneous of degree 0 (§ 3, no. 1), hence e=⨂i∈Ieie = \bigotimes_{i \in I} e_i is homogeneous of degree 0 and it follows from (27), (28) and the relations

εij(αi,0)=εij(0,βj)=1 \varepsilon_{ij}(\alpha_i, 0) = \varepsilon_{ij}(0, \beta_j) = 1

that ee is unit element of EE, which completes the definition on EE of the desired graded A-algebra structure. Then take hi(xi)=xi⊗⨂j≠iejh_i(x_i) = x_i \otimes \bigotimes_{j \neq i} e_j; to verify that hi(xixi′)=hi(xi)hi(xi′)h_i(x_i x_i') = h_i(x_i) h_i(x_i') for xi,xi′x_i, x_i' in EiE_i, attention may be confined to the case where xix_i and xi′x_i' are homogeneous and then this relation follows immediately from (27) and the relations εij(αi,0)=εij(0,βj)=1\varepsilon_{ij}(\alpha_i, 0) = \varepsilon_{ij}(0, \beta_j) = 1; the same relations and (24) prove also that the hih_i satisfy conditions (i) and (ii) of the statement and that

⨂i∈Ixi=∏i∈Ihi(xi) \bigotimes_{i \in I} x_i = \prod_{i \in I} h_i(x_i)

where the right hand side is the product of the ordered sequence (hi(xi))i∈I(h_i(x_i))_{i \in I} in E with the given total ordering on I (I, § 1, no. 2) (it suffices to argue by induction on the number of xix_i (assumed homogeneous) distinct from the eie_i).

It remains to prove condition (iii); note that the mapping

(xi)i∈I↦∏i∈Ifi(xi), (x_i)_{i \in I} \mapsto \prod_{i \in I} f_i(x_i),

where the right hand side is the product of the ordered sequence (fi(xi))i∈I(f_i(x_i))_{i \in I} with the given total ordering on II, is A-multilinear. Then there exists one and only one A-linear mapping f:E→Ff : E \to F such that

f(⨂i∈Ixi)=∏i∈Ifi(xi). f\left( \bigotimes_{i \in I} x_i \right) = \prod_{i \in I} f_i(x_i).

Clearly f(e)f(e) is the unit element of FF and f∘hi=fif \circ h_i = f_i; to verify that ff is an algebra homomorphism, in other words that f(x)f(y)=f(xy)f(x)f(y) = f(xy) for x,yx, y in EE, attention may be confined, by linearity, to the case where x=⨂i∈Ixix = \bigotimes_{i \in I} x_i and y=⨂i∈Iyiy = \bigotimes_{i \in I} y_i, xix_i (resp. yiy_i) being homogeneous of degree αi\alpha_i (resp. βi\beta_i) for all i∈Ii \in I. The relation to be verified then reduces, by (27), to

(∏i∈Ifi(xi))(∏i∈Ifi(yi))=ε(α,β)∏i∈I(fi(xi)fi(yi)). \left( \prod_{i \in I} f_i(x_i) \right) \left( \prod_{i \in I} f_i(y_i) \right) = \varepsilon(\alpha, \beta) \prod_{i \in I} (f_i(x_i) f_i(y_i)).

But, taking account of relations (26), this is a consequence of Lemma 2 of no. 6.

Clearly the algebra EE and the canonical mapping ψ:⨂i∈IEi→E\psi : \bigotimes_{i \in I} E_i \to E constitute a solution of the universal mapping problem (Set Theory, IV, § 3, no. 1), where Σ\Sigma is the species of A-algebra structure and the α\alpha-mappings ∏ifi\prod_i f_i from ∏iEi\prod_i E_i to an A-algebra, satisfying conditions (26).

For a fixed total ordering on II, the graded algebra EE defined in the proof of Proposition 10 will be called a graded tensor ε\varepsilon-product of type Δ\Delta of the family (Ei)i∈I(E_i)_{i \in I} of graded algebras of type Δi\Delta_i and will be denoted by ε⨂i∈IEi^{\varepsilon} \bigotimes_{i \in I} E_i (if no confusion can arise over the ordering on II); similarly, the homomorphism f:E→Ff : E \to F defined in the proof of Proposition 10 will be denoted by ε⨂i∈Ifi^{\varepsilon} \bigotimes_{i \in I} f_i. The homomorphisms hih_i are called canonical. We also write εG⊗n^{\varepsilon} G^{\otimes n} when I={1,n}I = \{1, n\} and all the EiE_i are equal to the same algebra GG.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Cites

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