Bourbaki

Proposition 11

Tag 03CK, permanent. Cite it as [Bourbaki, Tag 03CK].

Théories spectrales, chapter V, § 3. Fonctions de type positif

Let φ∈\varphi \in Pos1(G)_1(G) and let (ϱ,x)(\varrho , x) be a cyclic Hilbertian realization of φ\varphi. Then ϱ\varrho is an irreducible representation of G if and only if φ\varphi is an extreme point (EVT, II, p. 57, def. 1) of Pos1(G)_1(G).

Let E be the space of ϱ\varrho. Suppose first that ϱ\varrho is not irreducible. Let F be a nonzero closed subspace of E, stable under ϱ\varrho, and different from E. Let us write x=x1+x2x=x_1+x_2, where x1∈Fx_1\in F and x2∈F∘x_2\in F^{\circ}. Then one has 1=∥x∥2=∥x1∥2+∥x2∥21 =\|x\|^2=\|x_1\|^2+\|x_2\|^2. The subrepresentation of E generated by x1x_1 is contained in F, hence x1≠xx_1\not =x since xx is a cyclic vector of ϱ\varrho, whence x2≠0x_2\not = 0. Analogously one verifies that x1≠0x_1\not = 0.

For j=1j= 1 and j=2j= 2, let us denote by φj\varphi_j the continuous function on G such that

φj(g)=1∥xj∥2⟨xj∣ϱ(g)xj⟩ \varphi_j(g) =\frac{1}{\|x_j\|^2}\langle x_j|\varrho (g)x_j\rangle

for all g∈Gg\in G. One has φj∈\varphi_j\in Pos1(G)_1(G)(V, p. 443, example). Since φ=∥x1∥2φ1+∥x2∥2φ2\varphi =\|x_1\|^2\varphi_1+\|x_2\|^2\varphi_2, it is enough to verify that φ1≠φ2\varphi_1\not =\varphi_2 in order to prove that φ\varphi is not an extreme point of Pos1(G)_1(G); for this it is enough to prove that φ≠φ1\varphi \not =\varphi_1.

Arguing by contradiction, suppose that φ=φ1\varphi =\varphi_1. Since one has ⟨x1∣ϱ(g)x2⟩=0\langle x_1|\varrho (g)x_2\rangle = 0 for all g∈Gg\in G, it would follow that

1∥x1∥2⟨x1∣ϱ(g)x⟩=1∥x1∥2⟨x1∣ϱ(g)x1⟩=φ1(g)=φ(g)=⟨x∣ϱ(g)x⟩ \frac{1}{\|x_1\|^2}\langle x_1|\varrho (g)x\rangle =\frac{1}{\|x_1\|^2}\langle x_1|\varrho (g)x_1\rangle =\varphi_1(g) =\varphi (g) =\langle x|\varrho (g)x\rangle

for all g∈Gg\in G, whence ⟨x1∣y⟩=⟨∥x1∥2x∣y⟩\langle x_1|y\rangle =\langle \|x_1\|^2x|y\rangle for every element yy of the vector subspace of E generated by the elements ϱ(g)x\varrho (g)x for g∈Gg\in G, hence for all y∈Ey\in E since xx is a cyclic vector of ϱ\varrho. This would imply that x1=∥x1∥2xx_1=\|x_1\|^2x is also a cyclic vector of ϱ\varrho, which is a contradiction, whence the assertion.

Now suppose that ϱ\varrho is irreducible, and prove that φ\varphi is an extreme point of Pos1(G)_1(G). Let φ1≠φ2\varphi_1\not =\varphi_2 be elements of Pos1(G)_1(G) and let t1,t2∈[0,1]t_1,t_2\in [0,1] be such that t1+t2=1t_1+t_2= 1 and φ=t1φ1+t2φ2\varphi =t_1\varphi_1+t_2\varphi_2. For j∈{1,2}j\in \{1,2\}, let (ϱj,xj)(\varrho_j, x_j) denote a cyclic Hilbert realization of φj\varphi_j, and let EjE_j denote the space of ϱj\varrho_j. Let x3=t11/2x1+t21/2x2x_3=t^{1/2}_1x_1+t^{1/2}_2x_2. Then (ϱ1⊕ϱ2,x3)(\varrho_1\oplus \varrho_2, x_3) is a Hilbert realization of φ\varphi. Since (ϱ,x)(\varrho , x) is cyclic, there exists an isometric G-morphism u:E→E1⊕E2u: E\rightarrow E_1\oplus E_2 such that u(x)=x3u(x) =x_3 (Prop. 10).

Let j=1j= 1 or j=2j= 2. Since ϱ\varrho is irreducible, there exists λj⩾0\lambda_j\geqslant 0 such that the G-morphism uj=u_j= prj∘u_j\circ u of E into EjE_j satisfies

⟨uj(y)∣uj(y′)⟩=λj⟨y∣y′⟩ \langle u_j(y)|u_j(y')\rangle =\lambda_j\langle y|y'\rangle

for all yy and y′y' in E (Cor. 5 of V, p. 388, if uj≠0u_j\not = 0, and one may take λj=0\lambda_j= 0 otherwise). For all g∈Gg\in G, one has

tjφj(g)=⟨tj1/2xj∣ϱj(g)(tj1/2xj)⟩=⟨uj(x)∣ϱj(g)(uj(x))⟩ t_j\varphi_j(g) =\langle t^{1/2}_jx_j|\varrho_j(g)(t^{1/2}_jx_j)\rangle =\langle u_j(x)|\varrho_j(g)(u_j(x))\rangle
=⟨uj(x)∣uj(ϱ(g)x)⟩=λjφ(g) =\langle u_j(x)|u_j(\varrho (g)x)\rangle =\lambda_j\varphi (g)

Since φj(e)=φ(e)=1\varphi_j(e) =\varphi (e) = 1, it follows that φj=φ\varphi_j=\varphi if tj≠0t_j\not = 0. The assumption φ1≠φ2\varphi_1\not =\varphi_2 therefore implies that t1t_1 or t2t_2 must be zero, which proves that φ\varphi is an extreme point of Pos1(G)_1(G).

Bourbaki sets the proof directly after the statement, and the corpus does not separate them, so what is above is the statement and its proof together.

Cited by

Nothing.

Cites

Nothing.

Machine translation of the English, by gpt-5.4, gpt-5-6-mini, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.