Bourbaki

Example 8

Tag 09HK, permanent. Cite it as [Bourbaki, Tag 09HK].

Integration, VII, p. 72, § 3. Applications and examples

Spaces of hermitian forms.

In this example, K always denotes one of the fields R,C,H\mathbf{R}, \mathbf{C}, \mathbf{H}. We write δ=dim⁡RK\delta = \dim_{\mathbf{R}} K (thus δ=1,2\delta = 1, 2 or 4). A hermitian form Φ\Phi on the right vector space KnK^n may be written

Φ(x,y)=Φ(x1,…,xn,y1,…,yn)=∑i,j=1nxi‾hijyj \Phi(x, y) = \Phi(x_1, \ldots, x_n, y_1, \ldots, y_n) = \sum_{i,j=1}^n \overline{x_i} h_{ij} y_j

with hij=hji‾h_{ij} = \overline{h_{ji}} for all ii and jj. We denote by H\mathfrak{H} the vector space over R\mathbf{R} formed by the hermitian matrices of Mn(K)M_n(K). The mapping (hij)↦Φ(h_{ij}) \mapsto \Phi is an isomorphism of H\mathfrak{H} onto the vector space of hermitian forms on KnK^n, by means of which we shall identify these two spaces. Let H+∗⊂H\mathfrak{H}_+^* \subset \mathfrak{H} be the set of nondegenerate positive hermitian forms on KnK^n. The set H+∗\mathfrak{H}_+^* is convex in H\mathfrak{H}; for, if Φ1,Φ2\Phi_1, \Phi_2 are in H+∗\mathfrak{H}_+^* and if λ,μ\lambda, \mu are two numbers >0> 0 such that λ+μ=1\lambda + \mu = 1, it is clear that λΦ1+μΦ2\lambda \Phi_1 + \mu \Phi_2 is a positive hermitian form; on the other hand, if (λΦ1+μΦ2)(x,x)=0(\lambda \Phi_1 + \mu \Phi_2)(x, x) = 0, then Φ1(x,x)=Φ2(x,x)=0\Phi_1(x, x) = \Phi_2(x, x) = 0, therefore x=0x = 0, so that λΦ1+μΦ2\lambda \Phi_1 + \mu \Phi_2 is nondegenerate. Let us now show that H+∗\mathfrak{H}_+^* is an open subset of H\mathfrak{H}. Let S be the set of x=(x1,…,xn)∈Knx = (x_1, \ldots, x_n) \in K^n such that x1x‾1+⋯+xnx‾n=1x_1 \overline{x}_1 + \cdots + x_n \overline{x}_n = 1; this is a compact subset of KnK^n; if Φ∈H+∗\Phi \in \mathfrak{H}_+^*, the function x↦Φ(x,x)x \mapsto \Phi(x, x) is continuous and >0> 0 on S, hence its infimum is >0> 0; if Φ′∈H\Phi' \in \mathfrak{H} is sufficiently near Φ\Phi, it follows that Φ′(x,x)>0\Phi'(x, x) > 0 for all x∈Sx \in S, so that Φ′\Phi' is positive and nondegenerate.

The general linear group GL(n,K)\mathbf{GL}(n, K) operates continuously on the right in H\mathfrak{H} by (X,Φ)↦Φ∘X(X, \Phi) \mapsto \Phi \circ X, that is, by (X,H)↦tX‾⋅H⋅X(X, H) \mapsto t \overline{X} \cdot H \cdot X, where HH denotes the hermitian matrix corresponding to Φ\Phi. It is clear that H+∗\mathfrak{H}_+^* is stable under GL(n,K)\mathbf{GL}(n, K). More precisely, by Alg., Ch. IX, §6, No. 1, Cor. 1 of Th. 1,1^1 H+∗\mathfrak{H}_+^* is the orbit under GL(n,K)\mathbf{GL}(n, K) of the form ∑i=1nxi‾yi\sum_{i=1}^n \overline{x_i} y_i corresponding to the identity matrix InI_n. The stabilizer of this form is U(n,K)\mathbf{U}(n, K). By Lemma 2 of App. I, H+∗\mathfrak{H}_+^* may be identified, as a topological homogeneous space, with GL(n,K)/U(n,K)\mathbf{GL}(n, K)/\mathbf{U}(n, K).

For every X∈GL(n,K)X \in \mathbf{GL}(n, K), let X~\widetilde{X} be the automorphism H↦tX‾⋅H⋅XH \mapsto t \overline{X} \cdot H \cdot X of the real vector space H\mathfrak{H}. If μ\mu denotes the Haar measure of the additive group H\mathfrak{H}, one has X~−1(μ)=∣det⁡X~∣⋅μ\widetilde{X}^{-1}(\mu) = |\det \widetilde{X}| \cdot \mu (§ 1, No. 10, Cor. 1 of Prop. 15). Let us show that

∣det⁡X~∣=∣N(X)∣λ, | \det \widetilde{X} | = | N(X) |^\lambda ,

where N denotes the norm in Mn(K)M_n(K) regarded as an R-algebra\mathbf{R}\text{-algebra}, and where λ=1−δ−2δn\lambda = 1 - \frac{\delta - 2}{\delta n}. It suffices to verify (15) for XX running over a system of

(1) Cf. TVS, V, §2, No. 4, Cor. 1 of Th. 2.

generators of GL(n,K)\mathbf{GL}(n, K), hence (A, II, §10, No. 13, Cor. 2 of Prop. 14) for XX of the following types:

a) XX is the matrix of a mapping of the form

(x1,…,xn)↦(xσ(1),…,xσ(n)), (x_1, \ldots, x_n) \mapsto (x_{\sigma(1)}, \ldots, x_{\sigma(n)}),
where σ∈Sn\sigma \in \mathfrak{S}_n. In this case, a power of XX is equal to 1, therefore ∣det⁡X~∣=∣N(X)∣=1|\det \widetilde{X}| = |N(X)| = 1.

b) XX is the matrix of a mapping of the form

(x1,…,xn)↦(ax1,x2,…,xn). (x_1, \ldots, x_n) \mapsto (ax_1, x_2, \ldots, x_n).
Then, if (hij)∈H(h_{ij}) \in \mathfrak{H}, one has X~((hij))=(hij′)\widetilde{X}((h_{ij})) = (h'_{ij}) with h11′=a‾h11a=∣a∣2h11h'_{11} = \overline{a} h_{11} a = |a|^2 h_{11}, h1i′=a‾h1ih'_{1i} = \overline{a} h_{1i} for i>1i > 1, hij′=hijh'_{ij} = h_{ij} for i>1,j>1i > 1, j > 1; therefore
∣det⁡X~∣=∣a∣2∣a∣δ(n−1)=∣a∣2+δ(n−1). |\det \widetilde{X}| = |a|^2 |a|^{\delta(n-1)} = |a|^{2+\delta(n-1)}.
On the other hand, if Y=(yij)∈Mn(K)Y = (y_{ij}) \in M_n(K), then XY=(yij′)XY = (y'_{ij}) with y1j′=ay1jy'_{1j} = ay_{1j} and yij′=yijy'_{ij} = y_{ij} for i>1i > 1; therefore ∣N(X)∣=∣a∣δn|N(X)| = |a|^{\delta n}. The formula (15) is again verified.

c) XX is the matrix of a mapping of the form

(x1,…,xn)↦(x1+bx2,x2,…,xn). (x_1, \ldots, x_n) \mapsto (x_1 + bx_2, x_2, \ldots, x_n).
Then X~((hij))=(hij′)\widetilde{X}((h_{ij})) = (h'_{ij}) with h11′=h11h'_{11} = h_{11}, h12′=h12+h11bh'_{12} = h_{12} + h_{11} b, h1i′=h1ih'_{1i} = h_{1i} for i>2i > 2, h22′=h22+b‾h12+h‾12b+b‾h11bh'_{22} = h_{22} + \overline{b} h_{12} + \overline{h}_{12} b + \overline{b} h_{11} b, h2i′=h2i+b‾h1ih'_{2i} = h_{2i} + \overline{b} h_{1i} for i>2i > 2, hij′=hijh'_{ij} = h_{ij} for i>2,j>2i > 2, j > 2. Taking into account Lemma 6, one sees that ∣det⁡X~∣=1|\det \widetilde{X}| = 1. One verifies similarly that ∣N(X)∣=1|N(X)| = 1, which completes the proof of formula (15).

This established, the measure ∣N(H)∣−λ/2dμ(H)|N(H)|^{-\lambda/2} d\mu(H) on H\mathfrak{H} is invariant under GL(n,K)\mathbf{GL}(n, K), since

X~−1(∣N(H)∣−λ/2dμ(H))=∣N(X~(H))∣−λ/2⋅∣det⁡X~∣dμ(H)=∣N(H)∣−λ/2∣N(X)∣−λ∣det⁡X~∣dμ(H)=∣N(H)∣−λ/2dμ(H). \widetilde{X}^{-1}(|N(H)|^{-\lambda/2} d\mu(H)) = |N(\widetilde{X}(H))|^{-\lambda/2} \cdot |\det \widetilde{X}| d\mu(H) = |N(H)|^{-\lambda/2} |N(X)|^{-\lambda} |\det \widetilde{X}| d\mu(H) = |N(H)|^{-\lambda/2} d\mu(H).

If H∈H+∗H \in \mathfrak{H}_+^*, then H=X~(In)=tX‾XH = \widetilde{X}(I_n) = {}^t \overline{X} X for some X∈GL(n,K)X \in \mathbf{GL}(n, K), therefore N(H)=N(X)‾N(X)>0N(H) = \overline{N(X)} N(X) > 0. Consequently, on H+∗\mathfrak{H}_+^*, the unique (up to a constant factor) measure invariant under GL(n,K)\mathbf{GL}(n, K) (cf. §2, No. 6, Th. 3) is the measure

dγ(H)=N(H)−λ/2dμ(H). d\gamma(H) = N(H)^{-\lambda/2} d\mu(H).
In particular,
dγ(H)=(det⁡H)−(n+1)/2dμ(H)when K=R, d\gamma(H) = (\det H)^{-(n+1)/2} d\mu(H) \quad \text{when } K = \mathbf{R},
dγ(H)=(det⁡H)−ndμ(H)when K=C. d\gamma(H) = (\det H)^{-n} d\mu(H) \quad \text{when } K = \mathbf{C}.

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 volume. Not translated.