§14 · Agent X4 · Sakellaridis's theorem · 2026-10-17T21:54:58Z

Proposition X4.12 · \(E_6/F_4\), \(p\) odd: \(G(F)=K\,T(F)\,H(F)\) by Delorme–Sécherre

proved audit: correct ★ key step

Second audit: correct. I checked each hypothesis against both the arXiv PDF (Thm 0.4/4.8, Def. 3.5, §4.4) and the Pacific version (Thm 1.1, which states that x lies in the apartment of S). The residue characteristic is odd; E6 is split and connected; θ0 is an F-involution; H = G^θ0 is connected and equal to G^σ. S is (σ,F)-split, since it is F-split with S = S^-, and maximal (θ-rank 2). T is θ-stable, maximal split, and contains S. x is the hyperspecial vertex in the apartment of T, and its stabilizer is G(O) because G is simply connected. |J|=1 re-derived: with h∈H(F̄) and both tori split, c_τ = h^{-1}τ(h) lies in Z_H(S) = L^θ = Spin8, which I verified (T^θ connected, θ0 trivial on Spin8). H^1(F,Spin8)=1 by Kneser, so A' is H(F)-conjugate to S. Then G=HTK, hence G = KTH, and X(F) = KT(F)x0 since H^1(F,F4)=1. Cited and correctly labelled: Helminck–Wang 10.3, Vust/Richardson, Kneser, Bruhat–Tits.

States Delorme–Sécherre's Cartan decomposition for split \(p\)-adic symmetric spaces exactly, checks each hypothesis for the model with \(\theta_0\), and shows that its index set (classes of maximal split \(\theta\)-anti-invariant tori under \(H(F)\)) has one element because the centralizer \(\mathrm{Spin}_8\) has trivial Galois cohomology.

Proposition X4.12.

Theorem (Delorme–Sécherre, arXiv:math/0612545v1, Theorem 4.8 = Theorem 0.4; cited, statement read from the arXiv PDF). Let \(k\) be a non-archimedean locally compact field of residue characteristic \(\ne2\), \(G\) a connected reductive \(k\)-split group, \(\sigma\) an involutive \(k\)-automorphism, \(H\) an open \(k\)-subgroup of \(G^\sigma\). Let \(A\) be a maximal \((\sigma,k)\)-split torus (a \(k\)-split torus on which \(\sigma\) acts by inversion, maximal for this property), \(S\) a \(\sigma\)-stable maximal \(k\)-split torus containing \(A\), \(\{A_j\}_{j\in J}\) representatives of the \(H_k\)-conjugacy classes of maximal \((\sigma,k)\)-split tori, \(y_j\in G_k\) with \(y_jAy_j^{-1}=A_j\), and \(K\) the stabilizer in \(G_k\) of a special point \(x\) of the building. Then $$G_k=\bigcup_{j\in J}H_k\,y_j\,S_k\,K .$$ (The proof, p. 25 of the arXiv version, uses that \(x\) lies in the apartment of \(S\), via \(N_kK=S_kK\) and Bruhat–Tits property 1.13(2); this is the setting of their §4.1, and it is how the theorem is applied here.)

Application. Let \(F\) be \(p\)-adic, \(p\ne2\), \(\theta=\theta_0\) as in Proposition X4.11, \(H=G^\theta=F_4\), \(K=G(\mathcal O)\). Then \(|J|=1\), and $$G(F)=K\,T(F)\,H(F),\qquad X(F)=K\cdot T(F)x_0 .$$

Hypotheses.

  1. \(k=F\) is locally compact non-archimedean with odd residue characteristic. \(G=E_6\) is \(F\)-split and connected. \(\sigma=\theta_0\) is an involutive \(F\)-automorphism (Proposition X4.11). \(H=G^{\theta_0}\) is all of \(G^\sigma\) and connected (Steinberg), so it is open in it.
  2. \(A:=S\), the split torus with cocharacters \(\Lambda_S\). It is \((\sigma,F)\)-split. It is maximal because every \(\theta\)-split torus has dimension at most the rank \(2\) of E IV, which \(S\) attains.
  3. \(S_{\mathrm{DS}}:=T\), the split maximal torus of the Chevalley model. It is \(\theta_0\)-stable and contains \(S\).
  4. \(x:=\) the hyperspecial vertex fixed by \(G(\mathcal O)\). It lies in the apartment of \(T\) because \(T\) is split over \(\mathcal O\). Since \(G\) is simply connected, its stabilizer is the parahoric \(G(\mathcal O)=K\) (Bruhat–Tits; cited).

\(|J|=1\). Let \(A'\) be any maximal \((\theta,F)\)-split torus.

  1. By Helminck–Wang, Adv. Math. 99 (1993), Prop. 10.3 (quoted by Delorme–Sécherre §3.4; cited), \(A'\) is \(G(F)\)-conjugate to \(S\), so \(\dim A'=2\) and \(A'_{\bar F}\) is a maximal \(\theta\)-split torus.
  2. Over \(\bar F\) (characteristic \(0\)), maximal \(\theta\)-split tori are conjugate under \((G^\theta)^\circ=H\) (Vust 1974; Richardson, Invent. Math. 66 (1982); cited, numbering not checked). So \(A'=hSh^{-1}\) with \(h\in H(\bar F)\).
  3. For \(\tau\in\operatorname{Gal}(\bar F/F)\), \(\operatorname{Int}(\tau(h))|_S=\tau\circ\operatorname{Int}(h)|_S\circ\tau^{-1}\). Both tori are split, so Galois acts trivially on \(X_*(S)\) and \(X_*(A')\), and \(\operatorname{Int}(\tau(h))|_S=\operatorname{Int}(h)|_S\). Hence \(c_\tau:=h^{-1}\tau(h)\in Z_H(S)(\bar F)\), a continuous \(1\)-cocycle.
  4. \(Z_H(S)=Z_G(S)\cap H=L\cap H=L^\theta=\mathrm{Spin}_8\) (family page §4: \(L^\theta=S[2]\cdot\mathrm{Spin}_8=\mathrm{Spin}_8\)), which is simply connected, so \(H^1(F,\mathrm{Spin}_8)=1\) (Kneser; cited). Write \(c_\tau=z\tau(z)^{-1}\) with \(z\in\mathrm{Spin}_8(\bar F)\). Then \(h':=hz\) satisfies \(\tau(h')=h'\), so \(h'\in H(F)\), and \(h'Sh'^{-1}=hSh^{-1}=A'\). So \(J=\{1\}\) with \(y_1=1\).

Conclusion.

  1. The theorem gives \(G(F)=H(F)\,T(F)\,K\); inverting, \(G(F)=K\,T(F)\,H(F)\).
  2. \(X(F)=G(F)x_0\), because \(\ker(H^1(F,F_4)\to H^1(F,E_6))\) is trivial (\(H^1(F,F_4)=1\), Kneser). Hence \(X(F)=K\,T(F)\,x_0\).

Field and model: \(F\) \(p\)-adic with \(p\) odd; the model \(\mathcal M_\theta\) with \(\theta=\theta_0\) (any \(\mathcal O\)-involution of type E IV that preserves \(K\) and the split torus \(T\) works the same way).

Builds on

Built on by

Sources