Bourbaki

Proposition 16

Tag 068K, permanent. Cite it as [Bourbaki, Tag 068K].

Algebra, chapter III, § 11. Cogebras, products of multilinear forms, inner products and duality

*For an element z≠0z \neq 0 of ⋀p(E)\bigwedge^p(E) to be pure, it is necessary and sufficient that, for all u∗∈⋀p−1(E∗)u^* \in \bigwedge^{p-1}(E^*),

(u∗⊥z)∧z=0. (u^* \perp z) \wedge z = 0.

The case p=0p = 0 is trivial and we assume p≥1p \geq 1. If z=x1∧⋯∧xpz = x_1 \wedge \cdots \wedge x_p, formula (68) (no. 9) with n=p−1n = p - 1 shows that u∗⊥zu^* \perp z is a linear combination of the xix_i (1≤i≤p1 \leq i \leq p), whence (81). If on the other hand the subspace MzM_z associated with zz is of dimension >p> p, consider a basis (ej)1≤j≤n(e_j)_{1 \leq j \leq n} of this subspace with n>pn > p. It follows from no. 11, Proposition 13 that each of the eje_j is of the form u∗⊥zu^* \perp z for some u∗∈⋀p−1(E∗)u^* \in \bigwedge^{p-1}(E^*) and relation (81) therefore implies ej∧z=0e_j \wedge z = 0 for 1≤j≤n1 \leq j \leq n. It follows that in the expression z=∑HaHeHz = \sum_H a_H e_H (where H runs through the set of subsets of {1,n}\{1, n\} with pp elements) all the coefficients aHa_H are zero, whence z=0z = 0, contrary to the hypothesis.

The criterion of Proposition 16 is equivalent to writing conditions (81) when u∗u^* runs through a basis of ⋀p−1(E∗)\bigwedge^{p-1}(E^*). In particular, suppose that E is of finite dimension nn and let (ei)1≤i≤n(e_i)_{1 \leq i \leq n} be a basis of E. Conditions (81) are then equivalent to the conditions

(82−(J,H))(82-(J, H))
⟨eJ∗,(eH∗⊥eI)∧z⟩=0 \langle e_J^*, (e_H^* \perp e_I) \wedge z \rangle = 0
for all subsets J, H of {1,n}\{1, n\} such that Card(J)=p+1\mathrm{Card}(J) = p + 1 and Card(H)=p−1\mathrm{Card}(H) = p - 1. Now, if I and I' are two subsets of {1,n}\{1, n\} with pp elements, formulae (76) of no. 10 and multiplication table (20) of § 7, no. 8 show that

⟨eJ∗,⟨eH∗⊥eI⟩∧eI′⟩=0 \langle e_J^*, \langle e_H^* \perp e_I \rangle \wedge e_{I'} \rangle = 0
unless there exists an i∈{1,n}i \in \{1, n\} such that I−H={i}I - H = \{i\} and J−I′={i}J - I' = \{i\}, in which case

(83)(83)
⟨eJ∗,(eH∗⊥eI)∧eI′⟩=(−1)(p−1)(p−2)/2εi,J,H \langle e_J^*, (e_H^* \perp e_I) \wedge e_{I'} \rangle = (-1)^{(p-1)(p-2)/2} \varepsilon_{i, J, H}
where εi,J,H=ρ{i},Hρ{i},I′\varepsilon_{i, J, H} = \rho_{\{i\}, H} \rho_{\{i\}, I'}; it can then be said that for i∈J∩CHi \in J \cap C H, εi,J,H\varepsilon_{i, J, H} is equal to +1+1 if the number of element of J which are <i< i and the number of elements of H which are <i< i have the same parity, and −1-1 otherwise.

It follows immediately that if we write z=∑IaIeIz = \sum_I a_I e_I, where I runs through the set of subsets of {1,n}\{1, n\} with pp elements, relation (82–(J, H)) is equivalent to the relation

∑i∈J∪CHεi,J,HaJ−{i}aH∪{i}=0. \sum_{i \in J \cup C_H} \varepsilon_{i, J, H} a_{J-\{i\}} a_{H \cup \{i\}} = 0.

Relations (84) are called Grassman’s relations; these are therefore necessary and sufficient conditions (when J describes the set of subsets with p+1p + 1 elements and H the set of subsets with p−1p - 1 elements of {1,n}\{1, n\}) for an element z≠0z \neq 0 of ⋀p(E)\bigwedge^p(E) to be pure.

Note that relations (84) are not independent. For example, for n=4n = 4 and p=2p = 2, Grassmann’s relations reduce to the single relation

a12a34−a13a24+a14a23=0. a_{12} a_{34} - a_{13} a_{24} + a_{14} a_{23} = 0.

Let Dp(E)D_p(E) be the subset of ⋀p(E)\bigwedge^p(E) consisting of the pure pp-vectors; clearly Dp(E)D_p(E) is saturated with respect to the equivalence relation between uu and vv: “there exists λ∈K∗\lambda \in K^* such that v=λuv = \lambda u” and two elements u,vu, v of Dp(E)D_p(E) are equivalent under this relation if and only if the subspaces MuM_u and MvM_v of E which are associated with them are the same. Therefore we thus obtain a canonical bijection of the set of p-dimensional vector subspaces of E onto the image Gp(E)G_p(E) of Dp(E)D_p(E) in the projective space P(⋀p(E))\mathbf{P}(\bigwedge^p(E)) associated with ⋀p(E)\bigwedge^p(E). The subset Gp(E)G_p(E) of P(⋀p(E))\mathbf{P}(\bigwedge^p(E)) is called the Grassmannian of index pp of the vector space E. When E is finite-dimensional and (ei)1≤i≤n(e_i)_{1 \leq i \leq n} is a basis of E, the Grassmannian of index pp is the set of points of P(⋀p(E))\mathbf{P}(\bigwedge^p(E)) for which a system of homogeneous coordinates (aI)(a_I) (relative to the basis (eI)(e_I) of ⋀p(E)\bigwedge^p(E)) satisfies Grassmann’s relations (84).

When E=KnE = K^n, we sometimes write Gn,p(K)G_{n,p}(K) instead of Gp(Kn)G_p(K^n), so that Gn,1(K)=Pn−1(K)G_{n,1}(K) = P_{n-1}(K). The mapping M↦M0M \mapsto M^0, which associates with every pp-dimensional subspace of KnK^n the orthogonal subspace in E∗E^* (identified with KnK^n under the choice of basis dual to the canonical basis of KnK^n) therefore defines a canonical bijection of Gn,p(K)G_{n,p}(K) onto Gn,n−p(K)G_{n,n-p}(K); Proposition 15 shows that this bijection is the restriction to Gn,p(K)G_{n,p}(K) of a canonical isomorphism of the projective space P(⋀p(Kn))\mathbf{P}(\bigwedge^p(K^n)) onto the projective space P(⋀n−p(Kn))\mathbf{P}(\bigwedge^{n-p}(K^n)).

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

Cited by

Nothing.

Cites

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