Bourbaki

Theorem 4

Tag 06U7, permanent. Cite it as [Bourbaki, Tag 06U7].

Algebra, chapter V, § 15. Separable extensions

— Let Ω\Omega be an algebraically closed extension of a field K and L a subextension of Ω\Omega. Then the following conditions are equivalent: a) L is separable over K. b) The intersection of the kernels of the 0-algebra homomorphisms of Ω⊗KL\Omega \otimes_K L into Ω\Omega is reduced to 0. c) For any elements a1,…,ana_1, \ldots, a_n of L linearly independent over K there exist K-automorphisms σ1,…,σn\sigma_1, \ldots, \sigma_n of Ω\Omega such that det⁡(σi(aj))≠0\det(\sigma_i(a_j)) \neq 0. d) Let V be a vector sub-K-space of L of finite dimension. Every K-linear mapping of V into Ω\Omega is a linear combination (with coefficients in R) of the restrictions to V of K-automorphisms of Ω\Omega.

d)⇒c)d) \Rightarrow c): Let a1,…,ana_1, \ldots, a_n be elements of L linearly independent over K and let V be the vector sub-K-space of L generated by a1,…,ana_1, \ldots, a_n. The mapping f↦(f(a1),…,f(an))f \mapsto (f(a_1), \ldots, f(a_n)) is an R-linear bijection of HomK(V,R)\mathrm{Hom}_K(V, R) onto Ωn\Omega^n. Suppose that d) holds, then there exist K-automorphisms a1,…,ana_1, \ldots, a_n of Ω\Omega such that the elements (σi(a1),…,σi(an))(\sigma_i(a_1), \ldots, \sigma_i(a_n)) of Ωn\Omega^n (for 1≤i≤n1 \leq i \leq n) form a basis of Ωn\Omega^n. We thus have det⁡(σi(aj))≠0\det(\sigma_i(a_j)) \neq 0, so d) implies c).

c) ⇒\Rightarrow b): Assume that c) holds and let x be in Ω⊗KL\Omega \otimes_K L. We write x in the form ∑j=1nxj⊗aj\sum_{j=1}^n x_j \otimes a_j, with x1,…,xnx_1, \ldots, x_n in Ω\Omega and the elements a1,…,ana_1, \ldots, a_n of L linearly independent over K. Choose K-automorphisms σ1,…,σn\sigma_1, \ldots, \sigma_n of Ω\Omega such that det⁡σi(aj)≠0\det \sigma_i(a_j) \neq 0; let χ1,…,χn\chi_1, \ldots, \chi_n be the R-homomorphisms of Ω⊗KL\Omega \otimes_K L into Ω\Omega such that χi(a⊗b)=a⋅σi(b)\chi_i(a \otimes b) = a \cdot \sigma_i(b) for a∈Ωa \in \Omega and b∈Lb \in L. Suppose that χi(x)=0\chi_i(x) = 0 for 1≤i≤n1 \leq i \leq n, in other words, that ∑j=1nxj⋅σi(aj)=0\sum_{j=1}^n x_j \cdot \sigma_i(a_j) = 0 for 1≤i≤n1 \leq i \leq n. Since we have assumed that the matrix (σi(aj))(\sigma_i(a_j)) has a non-zero determinant, we have xi=0x_i = 0 for 1≤i≤n1 \leq i \leq n and so x=0x = 0.

b) ⇒\Rightarrow a): Since every extension of a field of characteristic 0 is separable (V, p. 122, Th. 1) it suffices to examine the case where K is of characteristic p≠0p \neq 0. Let X be the set of all R-algebra homomorphisms of Ω⊗KL\Omega \otimes_K L into Ω\Omega and f the homomorphism of Ω⊗KL\Omega \otimes_K L into ΩX\Omega^X defined by f(u)=(χ(u))χ∈Xf(u) = (\chi(u))_{\chi \in X} for u∈Ω⊗KLu \in \Omega \otimes_K L. Condition b) means that f is injective and so implies that the ring Ω⊗KL\Omega \otimes_K L is reduced. Hence condition b) of Th. 2 (V, p. 122) is satisfied with K′=ΩK' = \Omega, so L is separable over K.

a) ⇒\Rightarrow d): Suppose that L is separable over K. Let V be a vector sub-K-space of finite dimension of L, V(Ω)=Ω⊗KVV_{(\Omega)} = \Omega \otimes_K V the vector Ω\Omega-space derived from V by extension of scalars and f0f_0 the linear form on V(Ω)V_{(\Omega)} such that f0(x⊗y)=xyf_0(x \otimes y) = xy for x∈Ω,y∈Vx \in \Omega, y \in V. Denote by G the group of K-automorphisms of Ω\Omega; for σ∈G\sigma \in G we put σV=σ⊗IdV\sigma_V = \sigma \otimes \mathrm{Id}_V and gσ=σ∘f0∘σV−1g_\sigma = \sigma \circ f_0 \circ \sigma_V^{-1}.

For each σ∈G\sigma \in G the mapping gσg_\sigma of V(Ω)V_{(\Omega)} into Ω\Omega is Ω\Omega-linear and maps x⊗yx \otimes y to x⋅σ(y)x \cdot \sigma(y) for x∈Ω,y∈Vx \in \Omega, y \in V. The kernel NσN_\sigma of gσg_\sigma is therefore a vector subspace of V(Ω)V_{(\Omega)}, and the same holds for N=∩σ∈GNσN = \cap_{\sigma \in G} N_\sigma. If p is the characteristic exponent of K, the field of invariants of G in Ω\Omega is equal to Kp−∞K^{p^{-\infty}} (V, p. 112, Prop. 10). We clearly have σV(N)=N\sigma_V(N) = N for all σ∈G\sigma \in G; therefore (V, p. 63, Cor.) the vector Ω\Omega-space N is generated by N0=N∩(Kp−∞⊗V)N_0 = N \cap (K^{p^{-\infty}} \otimes V). Since L is separable over K, the fields Kp−∞K^{p^{-\infty}} and L are linearly disjoint over K (V, p. 123, Cor. 1); we have Kp−∞⊗KV⊂Kp−∞⊗KLK^{p^{-\infty}} \otimes_K V \subset K^{p^{-\infty}} \otimes_K L and f0(x⊗y)=xyf_0(x \otimes y) = xy for x∈Ωx \in \Omega and y∈Vy \in V. It follows that the restriction of f0f_0 to Kp−∞⊗KVK^{p^{-\infty}} \otimes_K V is injective. Now f0=g1f_0 = g_1 is zero on N and a fortiori on N0⊂Kp−∞⊗KVN_0 \subset K^{p^{-\infty}} \otimes_K V. We thus have N0=0N_0 = 0, whence N=0N = 0. Since V is of finite dimension over K, V(Ω)V_{(\Omega)} is of finite dimension over Ω\Omega; the intersection of the kernels of the linear forms gσg_\sigma is zero, hence (II, p. 301, Th. 7) the family (gσ)σ∈G(g_\sigma)_{\sigma \in G} generates the dual of V(Ω)V_{(\Omega)}. Now let uu be a K-linear mapping of V into Ω\Omega; let u~\tilde{u} be the linear form on V(Ω)V_{(\Omega)} which maps x⊗yx \otimes y to xu(y)xu(y) for x∈Ωx \in \Omega and y∈Vy \in V. By what has been said there exist elements σ1,…,σn\sigma_1, \ldots, \sigma_n of G and λ1,…,λn\lambda_1, \ldots, \lambda_n of Ω\Omega such that u~=∑i=1nλigσi\tilde{u} = \sum_{i=1}^n \lambda_i g_{\sigma_i}, whence u(y)=∑i=1nλiσi(y)u(y) = \sum_{i=1}^n \lambda_i \sigma_i(y) for all y∈Vy \in V. So we have shown that a) implies d).

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, 2003, Springer. Not translated.