Bourbaki

Remark 3

Tag 0DJQ, permanent. Cite it as [Bourbaki, Tag 0DJQ].

Algebra, chapter IX, § 1. Formes sesquilinéaires

If MM and NN are right AA-modules having finite bases (mi)(m_i) and (nk)(n_k), Φ\Phi a left sesquilinear form for JJ on M×NM \times N (No. 6, Remark), ss and dd its associated mappings, and R=(Φ(mi,nk))R = (\Phi(m_i, n_k)) its matrix, the formulas (26) of No. 6 show that one has

M(d)=RJ′,M(s)=′R. M(d) = R^{J'}, \quad M(s) = 'R.

Suppose now that the mappings ss and dd associated to Φ\Phi are bijective and calculate the matrix R^\hat{R} of the inverse form of Φ\Phi (No. 7). When Φ\Phi is bilinear, Φ\Phi is the inverse image of Φ^\hat{\Phi} relative to the linear mappings s:E→F∗s : E \to F^* and d:F→E∗d : F \to E^*; one therefore has, by virtue of (48) and (49), R=R.R^.RR = R . \hat{R} . R, whence, since RR is invertible (dd being bijective), R^=R−1\hat{R} = R^{-1}. This formula extends to the case where Φ\Phi is sesquilinear, because, if one considers Φ\Phi as a bilinear form on E×FJE \times F^J, and if one identifies (FJ)∗(F^J)^* with (F∗)J(F^*)^J (cf. No. 9), the inverse form of this bilinear form coincides with Φ^\hat{\Phi} considered as a bilinear form on (F∗)J×E∗(F^*)^J \times E^*. In both cases the matrix of the inverse form of Φ\Phi is the inverse of the matrix of Φ\Phi.

Let finally E′E' be a left A-module, F′F' a right A-module (resp. left A-module), both admitting finite bases (es′)(e'_s) and (ft′)(f'_t); let Φ′\Phi' be a bilinear form (resp. sesquilinear form for J) on E′×F′E' \times F', and let R′R' be its matrix. Suppose sΦs_\Phi and dΦd_\Phi are bijective. Let u:E→E′u : E \to E' and ν:F→F′\nu : F \to F' be linear mappings, u∗:F′→Fu^* : F' \to F and ν∗:E′→E\nu^* : E' \to E their adjoints (No. 8, prop. 7); denote by U,V,U∗,V∗U, V, U^*, V^* the matrices of u,ν,u∗,ν∗u, \nu, u^*, \nu^* with respect to the given bases. One then has

U∗=R−1.tU.R′,tV∗=R′.V.R−1 U^* = R^{-1} . {}^t U . R', \quad {}^t V^* = R'. V . R^{-1}
(resp. U∗J=R−1.tU.R′, tV∗=R′.VJ.R−1U^{*J} = R^{-1} . {}^t U . R', \ {}^t V^* = R'. V^J . R^{-1}).

Indeed, for all x∈Ex \in E and y∈F′y \in F', we have Φ′(u(x),y)=Φ(x,u∗(y))\Phi'(u(x), y) = \Phi(x, u^*(y)) (No. 8, def. 10). Whence, when Φ\Phi is bilinear, by virtue of (47), tM(u(x)).R′.M(y)=tM(x).R.M(u∗(y)){}^t M(u(x)) . R' . M(y) = {}^t M(x) . R . M(u^*(y)); this gives, by virtue of (45 G) and (45 D), tM(x).tU.R′.M(y)=tM(x).R.U∗.M(y){}^t M(x) . {}^t U . R' . M(y) = {}^t M(x) . R . U^* . M(y), whence tU.R′=R.U∗{}^t U . R' = R . U^* and the first formula announced since, dd being bijective, RR is invertible. When Φ\Phi is sesquilinear (47) and (45 G) give tM(x).tU.R′.M(y)J=tM(x).R.t(tM(y).tU∗)J{}^t M(x) . {}^t U . R' . M(y)^J = {}^t M(x) . R . {}^t ({}^t M(y) . {}^t U^*)^J; now, by (43), we have (tM(y).tU∗)J=t(tU∗J.tM(y)J)({}^t M(y) . {}^t U^*)^J = {}^t ({}^t U^{*J} . {}^t M(y)^J), whence t(tM(y).tU∗)J=U∗J.M(y)J{}^t ({}^t M(y) . {}^t U^*)^J = U^{*J} . M(y)^J; it follows therefore that tM(x).tU.R′.M(y)J=tM(x).R.U∗J.M(y)J{}^t M(x) . {}^t U . R' . M(y)^J = {}^t M(x) . R . U^{*J} . M(y)^J, whence tU.R′=R.U∗J{}^t U . R' = R . U^{*J}, and U∗J=R−1.tU.R′U^{*J} = R^{-1} . {}^t U . R'. The verification of the formulas for V∗V^* is analogous.

Exercises. — 1) Let A be a commutative field, E a vector space over A admitting an infinite countable basis (en)n⩾1(e_n)_{n \geqslant 1}. A bilinear form Φ\Phi is defined on E by setting Φ(ei+1,ei)=1\Phi(e_{i+1}, e_i) = 1 for i⩾1i \geqslant 1, Φ(ek,ej)=0\Phi(e_k, e_j) = 0 for k≠j+1k \neq j + 1 and j⩾1j \geqslant 1. Prove that the linear mapping dΦd_\Phi associated on the right to Φ\Phi is injective, but that the linear mapping sΦs_\Phi associated on the left to Φ\Phi is not injective.

2) Let E be the Z\mathbf{Z}-module direct sum of Z\mathbf{Z} and of Z/(2)\mathbf{Z}/(2), and let E∗E^* be its dual (isomorphic to Z\mathbf{Z}). Prove that the bilinear form (x,x′)→⟨x,x′⟩(x, x') \to \langle x, x' \rangle on E×E∗E \times E^* is such that the linear mapping associated on the right is injective, but the linear mapping associated on the left is not.

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

Cited by

Nothing.

Cites

Nothing.

Translated with machine help, not yet reviewed by a person.