The local Gan–Gross–Prasad conjecture for general spin groups: a Lean blueprint

4 Proof of the main theorem from the multiplicity formula

Thesis §6.5–6.7: summing the multiplicity formula over all inner forms and rearranging the geometric side over stable classes in \(\Gamma (G,H)\), everything cancels except the contribution of \(x=1\). For \(x\in \mathrm{GSpin}(W)(F)\) write \(W'_x=\ker (P(x)-1)\), \(W''_x\) its orthogonal complement, and \(T_x\subset \mathrm{GSpin}(W)\) the centralizer of \(x\) acting trivially on \(W'_x\); \(S_x=T_x/A_{G,H}\).

4.1 The space \(\Gamma (G,H)\) and the geometric multiplicity

Definition 61 Elements of \(\Gamma (G,H)\)
✓
#

\(x\in \mathrm{GSpin}(W)(F)\) defines an element of \(\Gamma (G,H)\) if \(x\) is semisimple, \(T_x\) is a torus and \(T_x(F)/A_{G,H}(F)\) is compact. This differs from Thesis §6.5, which requires \(T_x\) itself to be anisotropic.

Definition 62 Classes of tori in \(\Gamma (G,H)\)
✓

The index set of \(\Gamma (G,H)\): \(\mathrm{GSpin}(W)(F)\)-conjugacy classes of the tori \(T_x\), \(x\in \Gamma (G,H)\).

Lemma 63 Finiteness of the classes of tori
✓

\(\Gamma (G,H)\) meets finitely many conjugacy classes of tori (Thesis §6.5).

Proof ▶

Twisting argument (Serre, Cohomologie galoisienne I §5). Tori \(S_y\) with the same \(\dim W''_y\) are conjugate under \(\mathrm{SO}(W)(\bar F)\): \(S_y(\bar F)\) is the split torus of a hyperbolic frame of eigenvectors of \(P(y)\), and Witt’s extension theorem carries one frame to another of the same size; a conjugator \(g\) gives the cocycle \(\sigma \mapsto g^{-1}\sigma (g)\) in \(N_x=N_{\mathrm{SO}(W)}(S_x)\), and \(H^1(F,N_x)\) is finite. Equal classes give a Galois-fixed, hence rational, conjugator in \(\mathrm{SO}(W)(F)\), which lifts to \(\mathrm{GSpin}(W)(F)\) and conjugates the tori.

Definition 64 The function \(\Delta \)
✓
#

\(\Delta (x)=|\det ((P(x)-1)|_{W''_x})|_F\) (Thesis (6.5.9)).

Definition 65 The orbit \(\mathcal O_x\)
✓

The regular nilpotent orbit \(\mathcal O_x\in \mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\) selected by Luo’s recipe ( [ 17 ] §7.1) applied to \((G'_x,H'_x)\); a junk value if there is none (then germs vanish anyway). This differs from the choice in Thesis §5.4: it is Luo’s recipe applied to \((G'_x,H'_x)\).

Definition 66 The geometric multiplicity \(m_{\mathrm{geom}}\)
✓

For \(s{\gt}0\),

\[ m_{\mathrm{geom},s}(\theta )=\sum _{[T]}|W(T)|^{-1}\int _{T(F)/A_{G,H}(F)}D^G(t)^{1/2}c_{\theta ,\mathcal O_t}(t)\Delta (t)^{s-1/2}\, dt, \]

the sum over the classes of tori of \(\Gamma (G,H)\), with the probability Haar measure and \(W(T)=N_{\mathrm{GSpin}(W)(F)}(T)/Z_{\mathrm{GSpin}(W)(F)}(T)\); and \(m_{\mathrm{geom}}(\theta )=\lim _{s\to 0^+}m_{\mathrm{geom},s}(\theta )\) (Thesis §7.2, Prop. 12.1.1). This differs from Thesis (6.5.3), which uses \(N_{\mathrm{GSpin}(W'')(F)}(T)/T(F)\) for \(W(T)\).

For tempered \(\pi _1,\dots ,\pi _k\) with central characters trivial on \(A_{G,H}(F)\), \(m_{\mathrm{geom}}(\sum _i\theta _{\pi _i})=\sum _im_{\mathrm{geom}}(\theta _{\pi _i})\) (Thesis Prop. 12.1.1, used in (6.7.2)).

Proof ▶

Each \(\theta _{\pi _i}\) lies in \(\mathrm{QC}(G,\eta _i)\) for its central character \(\eta _i\), and \(m_{\mathrm{geom}}\) is linear on the span of these spaces (Chapter 7).

\(\sum _{\pi \in \Pi _\alpha }m(\pi )=m_{\mathrm{geom}}(\theta _\alpha )\) for every inner form \(\alpha \).

Proof ▶

The multiplicity formula for \((G_\alpha ,H_\alpha ,\xi _\alpha )\) and linearity of \(m_{\mathrm{geom}}\).

4.2 Germ transfer

Lemma 69 Finitely many regular nilpotent orbits
#

For semisimple \(x\in G(F)\) the set \(\mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\) of regular nilpotent \(G_x(F)\)-orbits is finite.

Proof ▶

Not yet formalized. The regular nilpotent elements form one geometric orbit, whose rational orbits are indexed by a subset of \(H^1(F,Z_{G_x}(N))\), finite by Borel–Serre (1964), Thm. 6.1.

Lemma 70 Regular germs of a stable quasi-character

If \(\theta \) is a stable quasi-character on \(G(F)\) and \(x\in G(F)\) is semisimple, then \(c_{\theta ,\mathcal O}(x)=c_{\theta ,\mathcal O'}(x)\) for all \(\mathcal O,\mathcal O'\in \mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\).

Proof ▶

Not yet formalized. As in the proof of [ 5 ] , (3.4.7): \(G_{x,\mathrm{ad}}(F)\) acts transitively on \(\mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\), stability makes \(X\mapsto \theta (xe^X)\) invariant under it, and the local expansion is unique.

(Thesis (6.1.2).) Let \(G\) be quasi-split, \(\theta \) a stable quasi-character on \(G(F)\) and \(\theta '\) a stable quasi-character on \(G_\alpha (F)\) with \(\theta '(y)=\theta (x)\) for stably conjugate regular \(x,y\). Then for semisimple, strongly stably conjugate \(x\in \mathrm{GSpin}(W)(F)\), \(y\in \mathrm{GSpin}(W_\alpha )(F)\), \(c_{\theta '}(y)=c_\theta (x)\), where \(c_\theta (x)\) is the average of the regular germs.

Proof ▶

Not verbatim in the literature (adapted from, or derived from, published results); not yet formalized. Ingredients: [ 5 ] , (12.1.2), stated there for pure inner forms; its proof does not use purity.

(Thesis (6.1.2).) Under the hypotheses of Theorem 71, \(c_{\theta ',\mathcal O'}(y)=c_{\theta ,\mathcal O}(x)\) for all regular \(\mathcal O,\mathcal O'\).

Proof ▶

Each regular germ of \(\theta \) and of \(\theta '\) equals the average (Lemmas 70 and 69); apply Theorem 71.

If \(x\in \mathrm{GSpin}(W)(F)\), \(y\in \mathrm{GSpin}(W_\alpha )(F)\) are semisimple and strongly stably conjugate, then \(c_{\theta _\alpha ,\mathcal O'}(y)=e(G_\alpha )\, c_{\theta _1,\mathcal O}(x)\) for all regular \(\mathcal O,\mathcal O'\) (Thesis §6.7 after (6.7.4)).

Proof ▶

Apply the germ transfer to \(e(G_\alpha )\theta _1\) and \(\theta _\alpha \), using (STAB) and (TRANS).

4.3 Stable conjugacy in \(\Gamma (G,H)\)

Thesis Prop. 6.6.1.

Let \(G\) be quasi-split and \(y\in \Gamma (G_\alpha ,H_\alpha )\) with \(G_{\alpha ,y}\) quasi-split. Then \(W'_{\alpha ,y}\) embeds isometrically in \(W\).

Proof ▶

By the classification of \(p\)-adic quadratic forms and Witt cancellation: if \(W''_{\alpha ,y}\) is \(0\) or a hyperbolic plane then \(W_\alpha \cong W\); \(\dim W''_{\alpha ,y}=1\) is impossible; otherwise a second class of the dimension and discriminant of \(W''_{\alpha ,y}\) exists.

Given an isometric embedding \(W'_{\alpha ,y}\hookrightarrow W\), some \(x\in \Gamma (G,H)\) is strongly stably conjugate to \(y\).

Proof ▶

\(W\) is the twist of \(W_\alpha \) by a cocycle in \(S_y\): twist \(W''_{\alpha ,y}\) into \(\iota (W'_{\alpha ,y})^\perp \) in block coordinates, a cocycle in \(SO(W_\alpha )_{W'_y}\), moved into \(S_y\) by [ 15 ] Lemma 10.2. Then the construction of Prop. 6.6.1(iii) for \(G_\alpha \) and the inverse twist \((W,\varphi _\alpha ^{-1})\), and symmetry of strong stable conjugacy.

Let \(G\) be quasi-split and \(y\in \Gamma (G_\alpha ,H_\alpha )\) with \(G_{\alpha ,y}\) quasi-split. Then some \(x\in \Gamma (G,H)\) is strongly stably conjugate to \(y\).

Proof ▶

Combine the two preceding lemmas.

For \(x\in \Gamma (G,H)\), \(y\in \Gamma (G_\alpha ,H_\alpha )\): an \(F\)-rational isomorphism of topological groups \(T_x(F)\simeq T_y(F)\), the identity on \(A_{G,H}\), with \(x\mapsto y\), matching the regular parts \(T_\natural \) and the strong stable classes, and preserving \(D^G\) and \(\Delta \) (the conclusion of Thesis Prop. 6.6.1(ii)). It need not preserve \(|W(T)|\).

If \(x\in \Gamma (G,H)\) and \(y\in \Gamma (G_\alpha ,H_\alpha )\) are strongly stably conjugate, then \(\mathrm{Ad}(g)\circ \psi _\alpha \) restricts to an \(F\)-rational isomorphism of topological groups \(T_x(F)\simeq T_y(F)\), the identity on scalars, \(x\mapsto y\), carrying \(T_{x,\natural }\) onto \(T_{y,\natural }\) with every \(t\) stably conjugate to its image.

Proof ▶

The values of \(t\mapsto gC(\varphi _\alpha )(t)g^{-1}\) are Galois-fixed, hence rational (Galois descent); they commute with \(y\) and fix \(W'_y\). The inverse comes from \(g^{-1}\) and \(\varphi _\alpha ^{-1}\). Scalars are central and \(x\mapsto y\) by the choice of \(g\). \(e\) and \(e^{-1}\) are restrictions of \(F\)-linear maps \(C(W)\to C(W_\alpha )\), hence continuous. Stable conjugacy in \(G\) is Lemma 79. On \(T_{x,\natural }\) the centralizers of \(\bar t\) and \(\bar x\) in \(\mathrm{GSpin}(W)(\overline{F})\) and \(\mathrm{GSpin}(V)(\overline{F})\) agree (by the centralizer theorem, Theorem 7, and the commutativity of the Lie algebra of \(S_t\)); hence the rationality clauses for \(x\) give those for \(t\), and \(\mathrm{Ad}(g)\circ C(\varphi _\alpha )\) carries centralizers, fixed spaces and semisimplicity, so \(T_{x,\natural }\) and \(T_{y,\natural }\) correspond.

Lemma 79 Stable conjugacy in \(G\) from \(W\)
✓

If \(\bar s=\mathrm{Ad}(g)C(\varphi _\alpha )(\bar t)\) with \(g\in \mathrm{GSpin}(W_\alpha )(\overline{F})\), then \(\mathrm{diag}(t)\) and \(\mathrm{diag}(s)\) are stably conjugate in \(G\) through \(\psi _\alpha \).

Proof ▶

On \(V\) conjugate by the image of \(g\) under \(W_{\alpha ,\overline{F}}\subseteq V_{\alpha ,\overline{F}}\): the two compositions \(C(W)\to C(V_{\alpha ,\overline{F}})\) agree on \(W\).

Lemma 80 \(\Delta \) is a stable invariant
✓

If \(P(t')=hP(t)h^{-1}\) over \(\overline{F}\) with \(h=P(g)\circ \varphi \), then \(\Delta (t')=\Delta (t)\).

Proof ▶

\(h\) conjugates \((P(t)-1)|_{W''_t}\) to \((P(t')-1)|_{W''_{t'}}\); both determinants lie in \(F\).

Lemma 81 \(D^G\) is a stable invariant
✓

If \(t\in G(F)\) and \(t'\in G_\alpha (F)\) are stably conjugate then \(D^{G_\alpha }(t')=D^G(t)\).

Proof ▶

\(\psi _\alpha \circ \mathrm{Ad}(g)\) conjugates \(1-\mathrm{Ad}t\) to \(1-\mathrm{Ad}t'\) over \(\overline{F}\); characteristic polynomials are unchanged by base change.

If \(x\in \Gamma (G,H)\) and \(y\in \Gamma (G_\alpha ,H_\alpha )\) are strongly stably conjugate, there is a torus transfer from \(x\) to \(y\). So the \(\Gamma \)-integrals over \(T_x(F)/A(F)\) and \(T_y(F)/A(F)\) match locally.

Proof ▶

The torus isomorphism, with the invariance of \(\Delta \) and \(D^G\).

4.4 The signed count

Thesis (6.7.6) and Prop. 6.6.1(iii).

Definition 83 The counts \(n_\alpha (x)\)
✓

\(n_\alpha (x)\) is the number of \(\mathrm{GSpin}(W_\alpha )(F)\)-conjugacy classes of \(y\in \Gamma (G_\alpha ,H_\alpha )\) strongly stably conjugate to \(x\) (Thesis §6.7).

Definition 84 The signed count
✓

\(\sum _\alpha e(G_\alpha )\, n_\alpha (x)\) (Thesis (6.7.6)).

Theorem 85 Hilbert’s Theorem 90 for \(\mathrm{GL}_m\)
✓
#

Every continuous cocycle \(\sigma \mapsto a_\sigma \in \mathrm{GL}_m(\overline{F})\) of \(\mathrm{Gal}(\overline{F}/F)\) is a coboundary: \(a_\sigma =M^{-1}\sigma (M)\).

Proof ▶

Speiser’s argument: average the cocycle over \(\mathrm{Gal}(L/F)\), \(L\) the normal closure of a field of definition, and use Dedekind’s independence of characters to find an invertible average \(B\) with \(a_\sigma \, \sigma (B)=B\) ( [ 21 ] , Ch. X §1 Prop. 3).

Lemma 86 The map \(q_x:H^1(F,S_x)\to \{ \alpha \} \)
✓

Every class \(\beta \in H^1(F,S_x)\) determines a unique inner form \(\alpha =q_x(\beta )\): the twist of \(W\) by a cocycle representing \(\beta \) is isometric to exactly one \(W_\alpha \).

Proof ▶

Hilbert 90 writes the cocycle as \(M^{-1}\sigma (M)\); \(v\mapsto Q(M^{-1}v)\) is an \(F\)-form of the discriminant of \(W\). Cohomologous cocycles give \(F\)-isometric forms by Galois descent.

Lemma 87 \(P(x)\) is semisimple over \(\overline{F}\)
✓

For \(x\in \Gamma (G,H)\), \(P(x)\otimes \overline{F}\) is semisimple on \(W_{\overline{F}}\).

Proof ▶

Conjugation by \(x\) is a commuting product of left and right multiplication by elements of \(F[x]\), hence semisimple; it restricts to \(P(x)\) on \(\iota (W)\). Squarefree polynomials stay squarefree over \(\overline{F}\) in characteristic \(0\).

Lemma 88 The Lie algebra of \(S_x\) is abelian
✓

For \(x\in \Gamma (G,H)\), the skew endomorphisms of \(W_{\overline{F}}\) commuting with \(P(x)\) and vanishing on \(W'_x\otimes \overline{F}\) commute.

Proof ▶

A skew \(Y\) lifts to \(\hat Y\in \mathfrak {gspin}(W_{\overline{F}})\) with \([\hat Y,v]=Y v\); \(\hat Y\) commutes with \(x\) (the difference \(x\hat Yx^{-1}-\hat Y\) is a traceless central element) and with \(W'_x\otimes \overline{F}\). These are \(F\)-linear conditions, so by flatness the \(\overline{F}\)-solutions are spanned by the \(F\)-solutions, which commute by the commutativity of the moved part of \(\mathfrak g_x\) over \(F\). Then \([[\hat Y_1,\hat Y_2],v]=[Y_1,Y_2]v\).

Lemma 89 \(S_x\) is commutative
✓

For \(x\in \Gamma (G,H)\), the group \(S_x(\overline{F})\) is commutative.

Proof ▶

\(S_x(\overline{F})\) consists of the isometries of determinant \(1\) commuting with the semisimple \(P(x)\) and fixing \(W'_x\). If the corresponding Lie algebra is abelian, the eigenvalues \(\lambda \ne \pm 1\) of \(P(x)\) are simple and \(\dim E_{-1}\le 2\); the determinant on \(E_{-1}\) is \(1\), and Cayley–Hamilton on \(E_{-1}\) gives commutativity.

Literature 90 Finiteness of \(H^1(F,S_x)\)
✓

For \(x\in \Gamma (G,H)\), \(H^1(F,S_x)\) is finite.

Assumed from: [ 20 ] , III §4.3 Thm. 4 (Borel–Serre); Tate–Nakayama.

Literature 91 Fundamental tori
✓

For \(x\in \Gamma (G,H)\), \(H^1(F,S_x)\to H^1(F,SO(W)_{W'_x})\) is surjective (\(S_x\) is an anisotropic, hence fundamental, maximal torus of \(SO(W)_{W'_x}\cong SO(W''_x)\)).

Assumed from: [ 15 ] , §10, Lemma 10.2.

Literature 92 Finiteness of \(H^1(F,N_x)\)
✓

For \(x\in \Gamma (G,H)\), \(H^1(F,N_x)\) is finite, \(N_x=N_{\mathrm{SO}(W)}(S_x)\).

Assumed from: Borel–Serre, Théorèmes de finitude en cohomologie galoisienne, Comment. Math. Helv. 39 (1964), Thm. 6.1.

For \(x\in \Gamma (G,H)\) and every \(\alpha \), \(n_\alpha (x)=|q_x^{-1}(\alpha )|\), \(q_x^{-1}(\alpha )\subset H^1(F,S_x)\). This differs from Thesis Prop. 6.6.1(iii), which uses \(H^1(F,T_x)\).

Proof ▶

Not yet formalized. Galois-cohomological parametrization of stable classes, then Witt cancellation.

Theorem 94 The Kottwitz sign is multiplicative on \(H^1(F,S_x)\)
✓

For \(G\) quasi-split and \(x\in \Gamma (G,H)\), \(\beta \mapsto e(G_{q_x(\beta )})\) is multiplicative on \(H^1(F,S_x)\).

Proof ▶

Own work, without Brauer groups (the statement is [ 5 ] , (12.1.6), with the sign of [ 14 ] ). \(e(G_\alpha )=1\) exactly when \(W_\alpha \cong W\). For a cocycle \(a\) in \(S_x\) choose lifts \(t_\sigma \in T_x(\bar F)\); the scalars \(t_\sigma \, \sigma (t_\tau )\, t_{\sigma \tau }^{-1}\) form its obstruction. Call \(a\sim b\) when they have lifts with equal obstructions; as \(T_x(\bar F)\) is commutative this is compatible with products. If \(W_\alpha \cong W_\beta \), a lift \(h\in \mathrm{GSpin}(W)(\bar F)\) of \(\varphi _a^{-1}\varphi _b\) (made proper by a rational reflection) conjugates a lift of \(a\) into one of \(b\), so \(a\sim b\); conversely \(a\sim b\) gives \(W_\alpha \cong W_\beta \) (Lemma 95). With at most two classes of the dimension and discriminant of \(W\), the sign is multiplicative.

Lemma 95 Twists with equal obstructions are isometric
✓

Let \(U,U_1,U_2\) be nondegenerate. If the cocycles of \(\varphi _i\colon U_{\bar F}\cong U_{i,\bar F}\) have continuous lifts to \(\mathrm{GSpin}(U)(\bar F)\) with equal (scalar) obstructions, then \(U_1\cong U_2\) over \(F\).

Proof ▶

Own work. The quotient \(v_\sigma =\varphi _1 s_\sigma t_\sigma ^{-1}\varphi _1^{-1}\) of the lifts is a \(1\)-cocycle in \(\mathrm{GSpin}(U_1)(\bar F)\), and \(C(\psi )\), \(\psi =\varphi _2\varphi _1^{-1}\), satisfies \(\sigma (C(\psi )z)=C(\psi )(v_\sigma \sigma (z)v_\sigma ^{-1})\). Hilbert 90 for the units of the central simple algebra \(C(U_1)\) (even dimension) or \(C^+(U_1)\) (odd dimension), deduced from Hilbert 90 for \(\mathrm{GL}_N\) and the fact that modules of equal dimension over a simple algebra are isomorphic, gives \(v_\sigma =u^{-1}\sigma (u)\); then \(z\mapsto C(\psi )(u^{-1}zu)\) is Galois-equivariant and descends to \(C(U_1)\cong C(U_2)\) (resp. \(C^+\)). The discriminants agree since \(\det \psi \) is Galois-fixed, so \(U_1\cong U_2\) by the classification of quadratic forms over \(F\) by rank, discriminant and Clifford invariant.

Lemma 96 A twist into the other form: the plane case
✓

If \(W''_x\) contains an \(x\)-stable anisotropic plane \(\Pi \) on which \(P(x)\) has determinant \(1\), some \(\beta \in H^1(F,S_x)\) twists \(W\) into a form \(W_{q_x(\beta )}\not\cong W\).

Proof ▶

Over \(\bar F\) the plane is hyperbolic with hyperbolic pair \(f,f'\); \(\mathrm{SO}(\Pi )(\bar F)\) is the split torus scaling \(f,f'\) by \(t^{-1},t\), lifted to \(\mathrm{GSpin}\) by \(ff'+t\, f'f\), which commutes with \(x\) and fixes \(W'_x\), so it lies in \(S_x\). Twisting \(\Pi \) into the other binary form of its discriminant gives a cocycle with these values; the twisted form is not isometric to \(W\) by Witt cancellation.

Lemma 97 The moved space of \(x\in \Gamma (G,H)\)
✓

For \(x\in \Gamma (G,H)\), \(W''_x\) is not a line, and if it is a plane it is anisotropic.

Proof ▶

On a line \(P(x)\) is the reflection in it, of determinant \(-1\). On a hyperbolic plane with hyperbolic pair \(f,f'\), \(P(x)\) is diagonal on \(f,f'\) (else it is a reflection), so the lifts \(h(s)=ff'+s\, f'f\) of the hyperbolic torus commute with \(x\) and lie in \(T_x(F)\), acting by \(s\) on \(f\): this contradicts compactness of \(T_x(F)/A(F)\).

Lemma 98 Twisting \(W''_x\) along \(S_x\)
✓

For \(x\in \Gamma (G,H)\) with \(W''_x\) of dimension \(\ge 3\) or an anisotropic plane, some \(\beta \in H^1(F,S_x)\) twists \(W\) into a form \(W_{q_x(\beta )}\not\cong W\).

Proof ▶

Twist the block \(W''_x\) into the other form of its dimension and discriminant; the cocycle lies in \(SO(W)_{W'_x}\), so by [ 15 ] Lemma 10.2 it is cohomologous to one in \(S_x\); Witt cancellation.

Lemma 99 A twist into the other form, for \(x\in \Gamma (G,H)\)
✓

For \(x\in \Gamma (G,H)\) not central, some \(\beta \in H^1(F,S_x)\) twists \(W\) into a form \(W_{q_x(\beta )}\not\cong W\).

Proof ▶

\(W''_x\neq 0\) since \(x\) is not central; Lemmas 97 and 98.

For \(G\) quasi-split and \(x\in \Gamma (G,H)\) not central, some \(\beta \in H^1(F,S_x)\) twists \(W\) into a form \(W_{q_x(\beta )}\not\cong W\).

Proof ▶

Lemma 99.

For \(G\) quasi-split and \(x\in \Gamma (G,H)\) not central, some \(\beta \in H^1(F,S_x)\) has \(e(G_{q_x(\beta )})=-1\).

Proof ▶

From the previous lemma: if \(W_\alpha \not\cong W\) then \(e(G_\alpha )=-1\), since two \(p\)-adic forms of the same dimension and discriminant have Witt indices of opposite parity in odd dimension and of the same parity in even dimension.

Let \(G\) be quasi-split and \(x\in \Gamma (G,H)\) with \(G_x\) quasi-split. Then \(\sum _\alpha e(G_\alpha )n_\alpha (x)=1\) if \(x\) is central and \(0\) otherwise. This replaces the Kottwitz-sign cancellation of Thesis §6.7 after (6.7.6).

Proof ▶

The sum equals \(\sum _{\beta \in H^1(F,S_x)}e(G_{q_x(\beta )})\). If \(x\) is central this is \(e(G)=1\). Otherwise the sign is a character of the finite group \(H^1(F,S_x)\) taking the value \(-1\), so the sum vanishes.

4.5 Rearrangement and the atom

Let \(f\) be a function of the tori \(T_x\), \(x\in \Gamma (G,H)\), with \(f(T_x)=f(T_{x'})\) whenever \(x,x'\) are strongly stably conjugate to a common \(y\in \Gamma (G_\beta ,H_\beta )\), and \(f=0\) on non-quasi-split tori. Then

\[ \sum _{[T']\in \Gamma (G_\alpha ,H_\alpha )/\sim }|W(T')|^{-1}f(T_{x(T')})=\sum _{[T]\in \Gamma (G,H)/\sim }|W(T)|^{-1} \frac{n_\alpha (x_T)}{n_1(x_T)}f(T), \]

where \(x(T')\) is strongly stably conjugate to a generator of \(T'\) (\(f(T_{x(T')}) := 0\) if there is none).

Proof ▶

Not yet formalized. Regroup by stable classes; \(n_\alpha (x)=|q_x^{-1}(\alpha )|\), and the rational classes of tori over \(\alpha \) are the orbits of \((N_x/S_x)(F)\) on \(q_x^{-1}(\alpha )\subseteq H^1(F,S_x)\), with stabilizers \(W(T')\) (orbit–stabilizer).

Given germ transfer and Prop. 6.6.1(i),(ii): for \(s{\gt}0\),

\[ \sum _\alpha m_{\mathrm{geom},s}(\theta _\alpha )=\int _{\Gamma (G,H)}\frac{\sum _\alpha e(G_\alpha )n_\alpha (x)}{n_1(x)} c_{\theta _1}(x)D^G(x)^{1/2}\Delta (x)^{s-1/2}\, dx \]

(Thesis (6.7.2)–(6.7.5) before the limit).

Proof ▶

Torus by torus: non-quasi-split tori contribute \(0\); a quasi-split torus \(T_y\) of \(G_\alpha \) has \(\int _{T_y/A}=e(G_\alpha )\int _{T_x/A}\) for \(x\sim _{\mathrm{stab}}y\) (the torus transfer identifies the probability Haar measures, \(D^G\) and \(\Delta \); germ transfer at the regular orbits \(\mathcal O_x\), \(\mathcal O_y\)); regroup with the torus-class count; the stable weight is constant on \(T_{x,\natural }\).

Under the same hypotheses,

\[ \sum _\alpha m_{\mathrm{geom}}(\theta _\alpha )=\lim _{s\to 0^+}\int _{\Gamma (G,H)} \frac{\sum _\alpha e(G_\alpha )n_\alpha (x)}{n_1(x)}c_{\theta _1}(x)D^G(x)^{1/2}\Delta (x)^{s-1/2}\, dx. \]
Proof ▶

The rearrangement at \(s{\gt}0\), and the finite sum over \(\alpha \) commutes with the limit.

For central \(c\in A_{G,H}(F)\), \(c_{\theta _1,\mathcal O_c}(c)=c_{\theta _1,\mathcal O_1}(1)\).

Proof ▶

\(\theta _1(cy)=\theta _1(y)\) on \(G_{\mathrm{reg}}(F)\) since the central characters are trivial on \(Z_{G,H}(F)\); \(e^X\) is regular for small regular \(X\), so the expansion at \(1\) is one at \(c\) (uniqueness of germs); \(\mathcal O_c=\mathcal O_1\).

Assuming the signed count, the contribution of the class of \(T=A_{G,H}\) to the stable integral is \(c_{\theta _1,\mathcal O_1}(1)\): \(|W(T)|=1\), \(T(F)/A(F)\) is a point, the weight is \(1\) and \(D^G=\Delta =1\).

Proof ▶

Direct computation, with central translation of the germ.

If the signed count holds on quasi-split classes, then for \(s{\gt}0\) the stable integral equals \(c_{\theta _1,\mathcal O_1}(1)\) (Thesis §6.7 after (6.7.6)).

Proof ▶

Non-central and non-quasi-split classes have weight \(0\); the atom gives the value above.

Let \(x\in \mathrm{GSpin}(W)(F)\) be semisimple with abelian moved part and \(W'_x\), \(V'_x\) quasi-split. Then some \(\mathcal O\in \mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\) satisfies Luo’s condition at \(x\) ( [ 17 ] §7.1).

Proof ▶

Luo’s split family for the descended datum \((W'_x,V'_x)\), carried into \(V\) (or \(W\)) by the isometric embedding, paired with a regular element of the other factor.

Lemma 110 \(\mathcal O_x\) is regular at quasi-split points
✓

For \(x\in \Gamma (G,H)\) with \(G'_x\) quasi-split, \(\mathcal O_x\in \mathrm{Nil}_{\mathrm{reg}}(\mathfrak g_x)\).

Proof ▶

Lemma 109.

For \(G\) quasi-split, \(\mathcal O_1\in \mathrm{Nil}_{\mathrm{reg}}(\mathfrak g)\).

Proof ▶

Luo’s recipe applied to \((G,H)\) selects a regular orbit, which exists since \(G\) is quasi-split.

Proof ▶

Sum (6.7.2) over \(\alpha \); rearrange by (6.7.5) using germ transfer and Prop. 6.6.1(i),(ii); the signed count kills every class but the atom.