§13 · Agent X3 · Pair model at 2 · 2026-10-17T20:20:26Z

Theorem X3.4 · Corrected A5-8 for the three families, and (SG) for the pair model without \(\tau\)

proved correction ★ key step

Correction. Corrects Proposition 5.8 · The support bound of H5H5 · Nonzero top splitting: for every orbit O_μ some Satake object V_λ ∗ IC_0 is supported on its closure with a nonzero split restriction to O_μ.(a) at function level over every \(F\), and the same-points question.

The statement of Proposition 5.8 holds as stated for the spheres, the pair model and alternating forms over every \(F\), with the pair-model proof replaced by Proposition X3.2; the triangular argument for \(\mathrm{GL}_n/(\mathrm{GL}_p\times\mathrm{GL}_q)\) uses only lattice labels, the support bound and Lemma A(2), so (SG) holds for the pair model over every \(F\), and for the involution model exactly when it coincides with the pair model, i.e. in odd residue characteristic.

Theorem X3.4.

Let \(F\) be any non-archimedean local field.

(1) Proposition 5.8, corrected. For \(X\) the sphere \(Q=\{N=1\}\) of Lemma 5.2, the pair model of \(\mathrm{GL}_n/(\mathrm{GL}_p\times\mathrm{GL}_{n-p})\), or the alternating forms \(\mathrm{GL}_{2m}/\mathrm{Sp}_{2m}\) of Computation 5.5, and every \(\lambda\in\Lambda_T^+\): $$t^\lambda x_0\in O_{\varphi(\lambda)},\qquad K\,t^\lambda X(\mathcal O)\subseteq\bigcup_{\nu\le\varphi(\lambda)}O_\nu .$$ Proofs: spheres, by pole order (Theorem 5.3, step 3); alternating forms, by the route of Proposition 5.8, whose input (a) holds in every characteristic; pair model, by Proposition X3.2. The route of Proposition 5.8 through input (a) is valid for the pair model iff the residue characteristic is odd (Proposition X3.3).

(2) (SG) without \(\tau\). Corollary C of the classical page, applied to the pair model with labels \(\Lambda_S^+\), the dominance order and Hecke labels \(\lambda(b)=(b_1,\dots,b_p,0,\dots,0)\), needs exactly: (i) \(K\backslash X(F)\leftrightarrow\Lambda_S^+\) with compact open orbits (Theorem 5.4(i)); (ii) finite lower sets (Lemma X3.1(c)); (iii) \(t^{\lambda(b)}x_0\in O_b\) for the lattice label (Theorem 5.4(ii), Proposition X3.2(i)); (iv) the support bound (Theorem 5.4(iii) or Proposition X3.2(ii)); (v) \(N(\lambda(b),b)\ge1\) (Lemma A(2)). It uses no involution, no Cartan embedding and no input (a). Hence:

Theorem. For every non-archimedean local field \(F\) — \(\operatorname{char}F=0\) or \(>0\), any residue characteristic, \(2\) included — and every \(1\le p\le n-p\), (SG) holds for the pair model \(X=\{P\in M_n:P^2=P,\ \operatorname{rk}P=n-p\}\cong\mathrm{GL}_n/(\mathrm{GL}_{n-p}\times\mathrm{GL}_p)\) over \(\mathcal O\), \(\varphi_0=\mathbf 1_{X(\mathcal O)}=\mathbf 1_{Kx_0}\).

(3) Which models. If the residue characteristic is odd, the involution model of the families page (\(\theta=\operatorname{Ad}J\)) has base point \(x_0^J\in O_0=X(\mathcal O)\), hence the same \(K\)-orbits and the same basic function, and (SG) holds for it. If \(1\le e=\operatorname{val}2<\infty\), \(x_0^J\in O_{(e,\dots,e)}\), so \(\mathbf 1_{Kx_0^J}\ne\varphi_0\). If \(\operatorname{char}F=2\) there is no involution with fixed group \(\mathrm{GL}_{n-p}\times\mathrm{GL}_p\) (Proposition X3.3(i)).

  1. Spheres. For dominant \(\lambda\) (type \(D_n\)), \(|\lambda_i|\le\lambda_1\), so \(\bar t^\lambda Q(\mathcal O)\subseteq\varpi^{-\lambda_1}\mathcal O^{2n}\), which \(K\) preserves; hence \(Kt^\lambda Q(\mathcal O)\subseteq\bigcup_{m\le\lambda_1}O_m\) (Theorem 5.3, step 3, graded correct by the second audit). On the spin page \(\varphi(\lambda)=2\lambda_1\varepsilon_1\leftrightarrow\lambda_1\), and \(t^\lambda x_0=\varpi^{\lambda_1}e_1+\varpi^{-\lambda_1}e_{2n}\in O_{\lambda_1}\). No \(\tau\) is used; Lemma 5.2 and Audit 7.20 cover residue characteristic 2. (Input (a) also holds there: on \(\langle e_1,e_{2n}\rangle\) the reflection in \(x_m\) is \(\begin{pmatrix}0&-\varpi^{2m}\\-\varpi^{-2m}&0\end{pmatrix}\), of Cartan type \((2m,-2m)\), with no division by \(2\).)
  1. Alternating forms, input (a). \(g\cdot\omega=g\omega g^{\mathsf T}\), \(x_0=J=\bigoplus J_2\), \(\theta(g)=J(g^{\mathsf T})^{-1}J^{-1}\), \(G^\theta=\mathrm{Sp}(J)\) in every characteristic, \(\tau(gx_0)=gJg^{\mathsf T}J^{-1}=\omega J^{-1}\), \(J\in K\). By the symplectic Smith form over any discrete valuation ring, \(\omega=kx_bk^{\mathsf T}\) with \(x_b=\bigoplus\varpi^{b_i}J_2\), and \(x_bJ^{-1}=\operatorname{diag}(\varpi^{b_1},\varpi^{b_1},\dots,\varpi^{b_m},\varpi^{b_m})=t^{\mu(b)}\), where here \(\mu(b):=(b_1,b_1,\dots,b_m,b_m)\in\Lambda_S^+\). So \(\tau(O_b)\subseteq Kt^{\mu(b)}K\) in every characteristic.
  1. Alternating forms, transfer. For \(x=kt^\lambda y\), \(\omega_y\in X(\mathcal O)\): \(\tau(x)=k\,t^\lambda(\omega_yJ^{-1})(Jt^\lambda J^{-1})(Jk^{\mathsf T}J^{-1})\), and \(Jt^\lambda J^{-1}\) is diagonal with the entries of \(t^\lambda\) permuted, so \(\tau(x)\in Kt^\lambda Kt^\lambda K\subseteq\bigcup_{c\le2\lambda}Kt^cK\) (Cartan convolution bound, cited). Thus \(\mu(b)\le2\lambda\). At even positions \(2k\) this reads \(B_k\le\Phi_k\), where \(B_k=\sum_{i\le k}b_i\) and \(\Phi_k=\sum_{i\le2k}\lambda_i\) are the partial sums of \(b\) and of the \(m\)-tuple \((\lambda_1+\lambda_2,\lambda_3+\lambda_4,\dots)\), so that \(\varphi(\lambda)=\lambda-\theta\lambda=\mu(\lambda_1+\lambda_2,\lambda_3+\lambda_4,\dots)\); at odd positions \(2B_{k-1}+b_k=B_{k-1}+B_k\le\Phi_{k-1}+\Phi_k\); and the totals agree, \(|b|=|\lambda|\), by \(\operatorname{val}\operatorname{Pf}\). So \(\mu(b)\le\varphi(\lambda)\). Top: \(t^\lambda Jt^\lambda=\bigoplus\varpi^{\lambda_{2i-1}+\lambda_{2i}}J_2\in O_{\varphi(\lambda)}\). This confirms the second audit's statement that (a), (b) and the transfer are valid for alternating forms in every characteristic.
  1. Pair model. Proposition X3.2 proves (1) directly for every \(F\); Proposition X3.3(iv) shows that the route through (a) works iff \(e=0\). So the statement of Proposition 5.8 holds as stated; only the justification "input (a) holds for the pair model by Theorem 5.4" in its step 4 is false, and only when \(e\ge1\) (residue characteristic 2, \(\operatorname{char}F=0\)) or \(e=\infty\) (\(\operatorname{char}F=2\), where \(\tau\) does not exist). Theorem 5.4 proves the orbit classification and \(t^{\lambda(b)}x_0\in O_b\), not \(\tau\)-compatibility.
  1. (2). Corollary C's hypotheses are the five items listed; each is cited above and none involves \(\tau\). Its proof (Lemma A(2), then Proposition B by induction on \(|\{\nu\le\mu\}|\)) needs only that \(T_{\lambda(b)}\ast\varphi_0\) is supported in the lower set of \(b\) with nonzero value on \(O_b\). Theorem 5.4(iv) runs the same argument with the componentwise order, also without \(\tau\). So the triangular argument never needed input (a), and the Theorem holds over every \(F\). The model: \(X(\mathcal O)\) is the set of pairs of complementary direct summands of \(\mathcal O^n\), with reductive stabilizer \(\mathrm{GL}(L_0\cap U_0)\times\mathrm{GL}(L_0\cap V_0)\).
  1. (3). \(x_0^J=(U^J,V^J)\), \(V^J=\langle e_i-e_{n+1-i}\rangle\), has \(P_{x_0^J}=\tfrac12(1+J)\). On each plane \(\langle e_i,e_{n+1-i}\rangle\) this is \(\tfrac12\begin{pmatrix}1&1\\1&1\end{pmatrix}\), of rank \(1\) with smallest entry valuation \(-e\), and it is \(1\) on the middle; so \(\ell(x_0^J)=(e,\dots,e)\) by Lemma X3.1(b) (checked in pair_tau.py for \(e=0,1,2\)). For \(e=0\), \(Kx_0^J=O_0\); the two models have the same \(G(F)\)-set \(X(F)\), the same \(K\), \(\mathcal H_K\), \(C_K\) and \(\varphi_0\), so (SG) is the same statement.

Builds on

Built on by

  • Proposition X3.5 · Only \(\mathbb C^\times\varphi_0\) generates; J-model fails at residue characteristic 2
  • Correction X3.6 · Correction of A7-24: its two objections stand; its claim that A5-8 holds needs a new proof
  • Summary X3.7 · A7-25, A7-32 re-derived; the \(\mathrm{GL}_n/(\mathrm{GL}_p\times\mathrm{GL}_q)\) row
  • Re-check V3 · Re-check by V3
  • Summary 9.6 · Verdict after the repairs: every input held or repaired; \(E_6/F_4\) still the one divergence

Sources