Automorphisms of triality groups over local rings

The elementary adjoint case is Theorem 6.6. Transport through \pi_E gives the simply connected classification with the canonical adjoint action, since G_P=E_P.

By Lemma 6.9, every \alpha\in\operatorname{Aut}(G_Q) restricts to E_Q. Write this restriction as \operatorname{Int}(g)\circ\phi_\eta. Then \operatorname{Int}(g^{-1})\circ\alpha extends \phi_\eta, so Proposition 6.8 identifies it with \Phi_\eta. Restriction gives uniqueness of (g,\eta), and the preceding constructions show that every pair defines an automorphism.

A fixed unit has a fixed inverse, so the nonunits of R_0 are exactly J_0. For \bar v\in k^{\bar\rho}, averaging any lift t gives T(t)/3\in R_0 with residue \bar v, and this lift is a unit when \bar v\ne0. Short parameters lift arbitrarily.

The character lattice of T_\Lambda has a basis permuted by \tau: the simple roots for Q and the fundamental weights for P. Lift one unit coordinate on the outer orbit and take its \rho-conjugates; average a lift of the fixed coordinate. This gives the required fixed torus lift.

The absolute multiplication map

is an open immersion [Conrad, Theorem 5.1.13]. Its inverse image under z is open in \operatorname{Spec} R and contains the closed point, hence is all of \operatorname{Spec} R. The unique absolute factors are \delta-fixed because \delta preserves the cell. Ordering absolute roots by their commuting triality orbits identifies the fixed unipotent coordinates with the relative root maps above. Thus the factors belong to U^\pm_\Lambda and H_\Lambda; uniqueness also gives compatibility with reduction.

For a fixed torus point, the three multipliers on a short orbit form a \rho-orbit, and the multiplier on a long root is fixed. Thus H_\Lambda normalizes every relative root subgroup, over both R and k.

The absolute Bruhat decomposition over an algebraic closure of k has cells and root-coordinate maps defined over k, so its unique factorizations hold over k [Conrad, Theorem 1.4.12]. The action of \tau\bar\rho sends the cell at w to that at \tau(w); a fixed point therefore has w\in W(D_4)^\tau. This Weyl group is generated by s_{\alpha_2} and s_{\alpha_1}s_{\alpha_3}s_{\alpha_4} [Steinberg, Theorem 32], so w has a fixed representative n_w given by a relative Tits word. For \bar z\in G_\Lambda(k), write its unique factorization as

where U_w=U^+_{\rm abs}\cap n_wU^-_{\rm abs}n_w^{-1}. These unipotent subgroups are \tau\bar\rho-stable, so uniqueness makes all three factors fixed. Their orbit coordinates give G_\Lambda(k)=E_\Lambda(k)T_\Lambda(k)^{\tau\bar\rho}.

Lemma 2.1 lifts the elementary and torus factors. For z\in G_\Lambda(R), choose e_0h_0 with the same residue. The remainder (e_0h_0)^{-1}z reduces to the identity and hence lies in U^+_\Lambda H_\Lambda U^-_\Lambda by the same lemma. Since H_\Lambda normalizes E_\Lambda(R), this proves G_\Lambda(R)=E_\Lambda(R)H_\Lambda and normality. Lifting any residue factorization also proves surjectivity of reduction.

For P, the simple coroots form a cocharacter basis, so every fixed torus point is h_{b,P}(r)h_{a,P}(s) with r\in R^\times, s\in R_0^\times, where

Both factors are elementary: for v\in R_\gamma^\times, put

For a this is the absolute rank-one identity; for b it is the product of three commuting rank-one identities. Hence H_P\subseteq E_P(R) and G_P(R)=E_P(R).

Every elementary target element has an orthogonal determinant-one representative. For a prepared constant, its reduction is \mu U, where U is the standard matrix and \mu^2=1 in k. Multiplication by -I_8, if necessary, gives the required reduction without changing the determinant.

Two determinant-one representatives with this reduction differ by \lambda\in1+J with \lambda^8=1. Since

one has \lambda=1. The two sides of each retained source identity have the same projective class, determinant one, and equal standard reductions. The same argument gives exact equality, including A_tA_u=A_{tu}.

Since 48\in R^\times and each summand is congruent to I_8, C_T\equiv I_8\pmod J and C_T is invertible. Reindexing by u=st gives

On the four planes,

Their distinct pairs of signs force a commuting matrix to be \operatorname{diag}(D_1,D_2,D_3,D_4). The remaining equations are

A matrix commuting with P_2 has equal diagonal and equal off-diagonal entries, giving the stated form. The determinant follows blockwise, and A(a,b,c,d)\mapsto(b+a,b-a,d+c,d-c) gives the algebra isomorphism.

Telescoping monomials gives a polynomial matrix M with

For two such solutions, M(y,y',t) reduces to the invertible Jacobian at (y_0,t_0), so M is invertible and y=y'.

For the original pair, choose C_l as above and let \kappa be the (3,4) entry of C_l^{-1}X_sC_l. Then C_{\rm root}=C_lC_\kappa^{-1} has the required properties.

In the Tits centralizer A(a,b,c,d), commutation with U_l forces a=0. For A(0,b,c,d), the entries of AU_s-U_sA are b-c-d at (3,4),(3,-4),(4,-3),(-4,-3) and zero elsewhere. Thus b=c+d; setting b=u, c=(u-v)/2, d=(u+v)/2 gives A=uP_++vP_-. The determinant follows on the two summands.

A proper similitude in the matrix centralizer has the form Z=pP_++rP_- with p,r\in B^\times. Since Z^\#=Z, its multiplier \mu satisfies p^2=r^2=\mu. Properness gives p^7r=\mu^4=p^8, hence r=p and Z is scalar.

Now lift an adjoint point centralizing [\mathcal S_B] to a proper similitude Z after a faithfully flat base change. The negative simple-root matrices U_{-s}=U_s^t, U_{-l}=U_l^t satisfy

Thus Z commutes projectively with all four simple signed root matrices. For each of them, ZUZ^{-1}=\xi_UU; taking traces gives 8=8\xi_U, hence \xi_U=1. Their Tits words give literal commutation with \mathcal S_B, so Z is scalar by the first part. Triviality of the adjoint point descends faithfully flatly. An element centralizing the frame also centralizes its simple signed elements and their Tits words. The remaining assertions follow from \mathcal F_B\subset E_B\subset G_B.

Set Y_g=C^{-1}gC. Its adjoint class has a proper-similitude representative because R is local. The matrix Y_g differs from it by a unit scalar, so Y_g^\#Y_g=m_gI_8 and \det Y_g=m_g^4. Since \det Y_g=1 and m_g\equiv1\pmod J, the factorization

has a unit second factor. Hence m_g=1 and Y_g is orthogonal.

Put N=CC^\#. Expanding Y_g^\#Y_g=I_8 gives C^\#g^{-1}N^{-1}gC=I_8, hence g^{-1}N^{-1}g=N^{-1}. The centralizer calculation and N\equiv I_8\pmod J give

For s=\det C/u^3 and r=u/s,

Thus the required ratio is obtained from \det C without choosing a square root. The self-adjoint matrix D=P_++rP_- centralizes \mathcal S and reduces to I_8. Consequently,

Invertibility also gives C'^\#C'=uI_8, so C' is a proper similitude with multiplier u. Since D centralizes \mathcal S, (3.9) holds.

Let c=[C']\in\mathbf G_Q(R). Both [g] and c^{-1}[g]c are fixed by \delta=\tau\rho. Applying \delta shows that \delta(c)c^{-1} centralizes [\mathcal S], so Lemma 3.8 gives c\in G_R. If c_1,c_2 have the same projective transport identities, then c_2c_1^{-1} centralizes [\mathcal S]; the same lemma proves uniqueness.

Every image has an orthogonal determinant-one representative Y, obtained from an elementary root word. If its projective class commutes with a constant root element, then YU_\beta Y^{-1}=\zeta U_\beta for some \zeta\in A^\times. Since \operatorname{tr}U_\beta=8\in A^\times, taking traces gives \zeta=1. Thus commutation with the fixed frame gives linear equations on Y.

The long simple root. For v\in S_0, the element y_{a,S}(v) commutes with the constants at -b,a,c,d,-d,-e,f. With M_a=M_{\alpha_2}=E_{2,3}-E_{-3,-2} and Q_0 from Section ?, the common matrix centralizer is

The elimination has 61 pivots equal to \pm1 and no residual equations, so this description is valid after arbitrary base change. Orthogonality and determinant one give

For p=z_1+z_2 and q=z_2-z_1, these imply p^2=q^2=1 and p^7q=pq=1. Hence p=q, so z_1=0 and z_2^2=1. Rescaling gives

Its adjoint class is fixed by \delta_A=\tau\rho_A, and \tau fixes the pinned root map x_{\alpha_2,Q}. Thus x_{\alpha_2,Q}(\rho_A(w))=x_{\alpha_2,Q}(w). Injectivity gives w\in A_0 and \psi(y_{a,S}(v))=y_{a,A}(w).

The short simple root. For t\in S, the element y_{b,S}(t) commutes with the constants at b,-a,e,f,-f. Each outer absolute factor satisfies these commutations separately. Their common matrix centralizer is

Here z_0,\ldots,z_6\in A are unrestricted at the linear stage: elimination has 57 pivots equal to \pm1 and no residual equations.

The (e_4,e_4)- and (e_4,e_{-4})-entries of Y^tJ_8Y=J_8, together with determinant one, give

As in the long-root case, z_4=0 and z_6^2=1. Rescale by z_6^{-1} and rename the parameters, so that z_6=1. The remaining orthogonality conditions are

They come from the (e_4,e_{-3})- and (e_{-3},e_{-3})-entries, with 2 cancelled in the latter; all other entries follow. Hence

The pinned automorphism \tau cycles these three simple root maps with positive signs. Since q_A(Y) is \delta_A-fixed, uniqueness of absolute unipotent coordinates gives

Both products have diagonal entries one, so their projective equality has no scalar ambiguity. Thus \psi(y_{b,S}(t))=y_{b,A}(z_1).

The fixed simple signed constants determine the Tits words. Equation (?) therefore transports these two recognition statements to every signed root. Uniqueness follows from injectivity of the root maps.

Recognition defines parameter maps \eta_\gamma:S_\gamma\to A_\gamma by \psi(y_{\gamma,S}(t))=y_{\gamma,A}(\eta_\gamma(t)). Root additivity makes them additive, and the fixed frame gives \eta_\gamma(1)=1. Applying \psi to (?) yields

Additivity removes the signs. Weyl transitivity gives one short map \eta:S\to A and one long map \eta_0:S_0\to A_0, both additive and taking 1 to 1.

Apply \psi to (?) and compare the c-factor in the unique ordered positive-root coordinates:

Taking t=1 gives \eta_0=\eta|_{S_0}. Thus \eta(S_0)\subseteq A_0, and multiplication is preserved when one factor belongs to S_0.

Comparison of the d-factor in (?) gives

Since t\times1=T(t)-t, setting u=1 proves trace preservation.

Multiplication of arbitrary elements is recovered from the integral identity

The trace values lie in S_0, so the mixed products are already preserved. Applying \eta gives

Cancel 2 in A. No division in S is used. A short-root coordinate determines \eta uniquely, and the root generators determine \psi.

If \psi is an automorphism, its inverse fixes the same frame. Applying recovery to \psi^{-1} gives a ring map \theta:S\to S. Evaluation of the two composites on a short root gives \theta\eta=\eta\theta=\operatorname{id}, so \eta is a ring automorphism.

Section ? gives g\in G_A with \bar g=\bar h such that \psi=\operatorname{Int}(g^{-1})\circ\Psi fixes the frame. Theorem 4.2 supplies \eta.

For two pairs, evaluating at 1 makes the quotient of their conjugators centralize the frame. Lemma 3.8 identifies the conjugators, and a short-root coordinate then identifies the ring maps. Under (?), reduction and cancellation of \operatorname{Int}(\bar h) on a short root give

proving the residue assertion.

Put C_2=C\otimes_B C, with p_1(c)=c\otimes1 and p_2(c)=1\otimes c. The two pullbacks of g_C transport the same indexed frame, so they agree by Lemma 3.8. Affine faithfully flat descent gives g\in\mathbf G_Q(B). Applying \tau\rho_B to the descended transport identities and using the same uniqueness gives g\in G_B.

For the second assertion, equivariance makes \rho_C\otimes\rho_C an automorphism of C_2. Pulling back the root rule and cancelling the common conjugator gives p_1\eta_C(t)=p_2\eta_C(t) by injectivity of the short-root map. The equalizer B\to C\rightrightarrows C_2 yields \eta:S\to B. Injectivity of f detects the ring laws, trace preservation, and the fixed-parameter condition; faithful flatness detects the root rule. The same frame and short-coordinate arguments prove uniqueness.

A regular functional defining C pulls back under D_4\to G_2 to a regular absolute functional, since no absolute root folds to zero. Thus all absolute factors of U_C lie in one absolute unipotent group. Its unique coordinates [Conrad, Theorem 5.1.13] give unique relative coordinates in any order. Commutator corrections have greater height than both inputs, so the stated relations collect every word into one factor per root. A collected word with trivial concrete image has all parameters zero. These relations therefore present U_C, whose map into the cover is injective because its composition with q_R is the concrete inclusion. Every distinct nonopposite pair lies in a positive system, giving the cover presentation.

The absolute Steinberg projection is likewise injective on each absolute unipotent subgroup, by collection [EJK, §7.2]. Hence \widehat E_R\to E_R is an isomorphism on each relative unipotent subgroup, and the Borel-wise comparison of [EJK, §7.1] identifies their graded covers, not \widehat E_R itself with either E_R or \mathcal S_R.

All residue parameters lift by Lemma 2.1. Modulo the displayed normal closure, a root generator depends only on its residue. Lifting the defining relations gives a homomorphism from \mathcal S_k to this quotient, inverse to the map induced by reduction.

First take l\in P_\alpha with toral image and \gamma noncollinear with \alpha. Choose a functional \lambda with \lambda(\alpha)=0 and \lambda(\gamma)>0. Its positive sector lies in a chamber. Conjugating a sector root \beta by a root at \pm\alpha introduces only roots i(\pm\alpha)+j\beta, j>0, still in the sector. Thus P_\alpha normalizes the lifted sector. Both sides of the asserted formula lie there and have the same concrete image, so unipotent injectivity gives equality.

For the \alpha-line use the off-axis identities

Every letter lies off the output-root line, and all lie in one unipotent group. Concrete Weyl conjugation lifts to a graded automorphism of the cover [EJK, Lemma 7.1], so these identities give such an expression for every root. This does not assume a conjugation formula for a chosen cover word. Apply the noncollinear formula to each letter and use unipotent injectivity again. The formula now holds for the generators of \widetilde L_R, hence for their products. Multipliers preserve the radical levels, and reduction at t=1 gives the final congruence.

Let M=\left(\begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right) be the evaluation. On the Cartan element of the absolute root carrying the free parameter, the other factors in a short orbit act trivially, and

A toral image fixes this element. Independence of root and coroot vectors gives ad+bc=1, ab=cd=0. Together with ad-bc=1 and 2\in R^\times, this yields ad=1, bc=0, and then b=c=0.

If the word is toral only after reduction, the same test gives b,c\in J_\gamma and d\in R_\gamma^\times. The identity M=x_+(b/d)\operatorname{diag}(d^{-1},d)x_-(c/d) shows that

has toral image. Rearranging gives the factorization with j_+=b/d and j_-=c/d. Since these parameters are radical, r_R(l)=r_R(w). Every defining generator of \widetilde L_k lifts root by root and can be corrected this way; products give r_R(\widetilde L_R)=\widetilde L_k.

Write B_C=\widetilde U_C(J)L_R^1\widetilde U_{-C}(J)\subseteq D_R. For a simple root \gamma\in C, arbitrary-order unipotent coordinates give

Move the middle factors by Lemma 5.3 and interchange the opposite root factors using (?). All denominators lie in 1+J_\gamma, and the new toral words lie in L_R^1. Since s_\gamma C=(C\setminus\{\gamma\})\cup\{-\gamma\}, the reverse interchange gives B_C=B_{s_\gamma C}. Connectivity of the chamber graph makes all B_C one set B.

Every radical root subgroup stabilizes B by left multiplication: choose a chamber containing that root. The same holds for L_R^1 by its toral action, and for inverses. Since 1\in B, the set B is exactly the subgroup generated by these factors.

This subgroup is normal in \mathcal S_R. Nonopposite conjugation of a radical root gives radical root factors: each correction contains the radical input, \rho(J)=J, and T(J)\subseteq J_0. Opposite conjugation is covered by (?). For l\in L_R^1, Lemma 5.3 gives

since the parameter is radical. The same arguments apply to inverse conjugators, proving normality. As B contains every radical root, Lemma 5.2 gives D_R\subseteq B.

Put F=k^{\bar\rho}. If \bar\rho has order three, descent of the pinned \operatorname{Spin}_{8,k} along \tau\bar\rho over the cyclic cubic Galois extension k/F gives a simply connected quasi-split group \mathbf H/F. Its relative root system is G_2, with short parameters k and long parameters F; its root maps become our orbit products over k.

If \bar\rho=\operatorname{id}, take the pinned fixed group \mathbf H=\operatorname{Spin}_{8,k}^{\tau}. The fixed-pinning results [ALRR, Theorem 5.1 and Corollary 5.8] make it connected, smooth, and split reductive: the coinvariants of the permuted fundamental weight basis are free, and no exceptional even-A component occurs. The fixed-root construction [ALRR, §4.3.1 and Proposition 6.1] gives simple roots (2,-1) and (-3,2) in the coinvariant basis of outer and middle weights. Thus \mathbf H is split G_2, both simply connected and adjoint, since this matrix has determinant one. The adjoint D_4 fixed group has the same pinned root datum.

In either case, the central isogeny \operatorname{Spin}_8(k)\to\mathbf G_Q(k) is injective on \mathbf H(F): its kernel \mu_2^2(k) has no nonidentity \tau\bar\rho-fixed point, since \tau cycles its three nonidentity points and \bar\rho fixes them. For the standard minimal parabolic P of \mathbf H, the elementary subgroup E_P(F) therefore maps isomorphically onto E_k, root by root. The inverse image of the absolute adjoint torus is the absolute Spin torus; intersecting with \mathbf H(F) gives its torus Levi subgroup T_F(F). Thus the rank-one toral-image conditions agree.

The relative Steinberg presentation [Stavrova, Definition 3.1] consists here of additive root laws and the same nonopposite relations. In the nonsplit case this follows after extension to k by absolute unipotent coordinates; in the split case it follows from the pinned orbit maps. Lemma 5.1 identifies \mathcal S_k with \operatorname{St}_P(F) and \widetilde L_k with \widetilde L_P(F) of [Stavrova, Definition 3.9]. Apply [Stavrova, Lemma 3.15] over the field F, viewed as a local ring with zero maximal ideal. The group is simply connected and simple of relative rank two; the congruence factor is trivial, leaving \ker(\operatorname{St}_P(F)\to E_P(F))\subseteq\widetilde L_P(F). No lower bound on |F| or perfectness assumption is required.

For z\in K_R, reduction gives r_R(z)\in K_k. Lemmas 5.6 and 5.4 supply l\in\widetilde L_R with r_R(l)=r_R(z). Thus l^{-1}z\in D_R. Fix opposite chambers and write their lifted groups as \widetilde U_\pm. Proposition 5.5 and the toral action give

Under q_R, this is an absolute open-cell decomposition of the identity. Uniqueness forces q_R(u_\pm)=q_R(h)=1. Unipotent injectivity gives u_+=u_-=1, so z=h\in\widetilde L_R. Lemma 5.3 then makes z commute with every lifted root, since m_\gamma(q_R(z))=1. Hence K_R\subseteq Z(\mathcal S_R). Conversely, surjectivity of q_R sends the center into Z(E_R)=1 by Lemma 3.8, proving equality.

The integral identity

gives \eta(t^\#)=\eta(t)^\# after applying \eta and cancelling 2 in A. Multiplication gives preservation of N, and polarization gives \times. If v\in S^{\rho_S}, then T_A(\eta(v))=3\eta(v). Applying \rho_A and cancelling 3 gives \rho_A(\eta(v))=\eta(v). No division in S is used.

The inverse \eta^{-1} preserves trace and induces the inverse of F_\eta. By Theorem 5.7, K_R=Z(\mathcal S_R) is characteristic, so both maps descend to E_R.

For z\in K_S, source centrality makes F_\eta(z) commute with every lifted target parameter-one root, since \eta(1)=1. Its concrete image centralizes the target frame and is therefore trivial by Lemma 3.8. Hence F_\eta(K_S)\subseteq K_A. Neither target centrality nor surjectivity of F_\eta is used.

For t\in S, put a=\eta(\rho_S(t)), b=\eta(\rho_S^2(t)), c=\rho_A(\eta(t)), d=\rho_A^2(\eta(t)). Trace preservation on t,t^2 gives a+b=c+d and a^2+b^2=c^2+d^2. Cancelling 2 yields ab=cd, hence (a-c)(a-d)=0. Since A is a domain, either a=c or a=d. The two solution sets are additive subgroups covering S. A group cannot be the union of two proper subgroups, so one orientation holds on all of S; either makes \ker\eta stable under \rho_S.

Compose with A\to k=A/\operatorname{Jac}(A) to obtain \nu:S\to k. It preserves trace, so Lemma 5.11 gives one orientation for \nu. Thus \mathfrak p=\ker\nu=\eta^{-1}(\operatorname{Jac}(A)) is a \rho_S-stable prime, and 2,3\notin\mathfrak p. The ring B=S_{\mathfrak p} inherits \rho_S and is a local coefficient pair. Every denominator maps to a unit of A, so \eta extends to \eta_{\mathfrak p}:B\to A.

For s\notin\mathfrak p, stability gives N_S(s)\in S^{\rho_S}\setminus\mathfrak p, and

Lemma 5.8, applied to \eta, gives

since the norm denominator is a fixed unit. Thus \eta_{\mathfrak p} preserves trace. Apply Proposition 5.10 to B\to A and compose with the equivariant pinned localization E_S\to E_B. The resulting map has the required rule and is unique by root generation. No centrality of K_S or surjectivity of elementary localization is used. The orientation concerns \nu, not necessarily \eta.

Root parameters lift by Lemma 2.1, proving surjectivity, and injectivity of the long-root map gives y_{a,k}(1)\ne1. If \bar\rho=\operatorname{id}, Lemma 5.6 identifies E_k with split G_2(k). The simplicity theorem [Steinberg, Theorem 5, p. 47] applies: the center is trivial and the small exceptional fields are excluded.

If \bar\rho has order three, the same root comparison gives a surjection from the universal root-generated triality group G'_\sigma=\langle U_\sigma,U^-_\sigma\rangle of [Steinberg, §11, pp. 177–178] to E_k. Its central quotient is simple by Theorem 34 on p. 187 of that work, with no exception of type {}^3D_4. Its center maps into Z(E_k)=1 by Lemma 3.8. Thus E_k is a nontrivial image of a simple group, and is simple.

Use right commutators. Transport a long unit to y_a(v), v\in R_0^\times. Relation (?) gives every y_f(vw), w\in R_0, and Weyl transport gives all long root groups. Relation (?) at v=1, after deleting its last two long factors, gives Q(t)=y_c(t)y_d(-t^\#)\in H for every t\in R. Since (-t)^\#=t^\#, adjacent cancellation gives

Weyl transport gives all short root groups. If the initial unit is short, transport it to y_d(u) and use

from (?), reducing to the long case.

Appendix ? proves both assertions by explicit commutator words. The first uses increasing root height and deletion of extracted factors in their original positions. The second produces a leading short parameter -t, removes the negative factors in two steps, and applies positive-factor extraction. Only normality under E_R is used; the individual torus and unipotent factors need not belong to H.

If H\not\subseteq N_R, simplicity of E_k gives \pi_R(H)=E_k. Choose z\in H with residue y_{a,k}(1). This residue lies in the absolute open cell, so Lemma 2.1 gives a unique relative decomposition z=u^+hu^- with \overline{u^+}=y_{a,k}(1) and \bar h=\overline{u^-}=1. Write u^+=y_a(t)X, X\in U_0. Positive coordinates give t\in1+J_0\subseteq R_0^\times. Lemma 6.3 places y_a(t) in H, and Lemma 6.2 gives H=E_R.

Thus every proper normal subgroup lies in N_R, which is proper by Lemma 6.1. This characterization makes N_R characteristic; exactness follows from surjectivity of reduction.

If \bar\rho=\operatorname{id}, use the pinned G_2 comparison and [BuninaG2, Theorem 3, p. 16], which applies to fields with 1/2,1/3 without perfectness. If \bar\rho has order three, use the equal-order triality classification [Steinberg, Theorem 36, p. 195], with the field and root conventions on pp. 158–159 and 177–178. It supplies inner, diagonal, and field factors.

In the adjoint form, a diagonal factor is a torus character \chi:Q\to k^\times. Compatibility with the triality root products gives \chi(\tau\alpha_i)=\bar\rho(\chi(\alpha_i)): the outer scalings form a \bar\rho-orbit, and the middle scaling is fixed. Hence the inner and diagonal factors combine into \bar g\in G_Q(k). The remaining map \beta=\operatorname{Int}(\bar g^{-1})\circ\bar\alpha fixes the four simple signed constants: field factors fix \pm1, and pinned diagram identifications permute the commuting outer constant factors and fix the middle one. The Tits words then give the whole fixed frame.

Theorem 4.2, applied directly to this frame-fixing automorphism over k, gives \nu\in\operatorname{Aut}(k) preserving trace. Lemma 5.11 supplies one orientation for \nu. The pinned map \mathbf G_Q(\nu) or d_0\circ\mathbf G_Q(\nu) from Section ? agrees with \beta on every root, proving (6.5).

Proposition 2.2 lifts \bar g. Since \ker(\nu\circ\kappa_R)=J, equivariance of reduction and surjectivity of \pi_R give the data in (?). No lift of \nu to R has been assumed.

Lemma 6.5 permits application of Theorem 4.3, giving g and a unital trace-preserving ring map \eta. Normality of E_R makes \psi=\operatorname{Int}(g^{-1})\circ\alpha a frame-fixing automorphism of E_R. Theorem 4.2 makes \eta invertible, and its root rule identifies \psi with \phi_\eta. The pair is unique by Theorem 4.3.

Conversely, Theorem 5.9 constructs \phi_\eta, and g\in G_Q(R) normalizes E_Q(R).

If u^2=1, then (1-u)/2 is an idempotent in the local ring R, so u=\pm1. Thus \rho fixes all four points of the central kernel \mu_2^2(R), while \tau cycles the three nonidentity points. The fixed kernel is trivial, proving injectivity. The relative root words lift every generator of E_Q(R), proving surjectivity of \pi_E.

Conjugation on \mathbf G_P factors through the central quotient \mathbf G_Q [Conrad, Theorem 7.1.9]: lifts after a faithfully flat extension differ centrally and define the same action. This action is \tau\rho-equivariant, so fixed points act on fixed points. Normality of E_Q(R) and injectivity of \pi on G_P(R) show that it preserves E_P(R). No lift of g over R is required.

For h\in E_Q\cap H_Q, equation (?) shows that \phi_\eta(h) and \eta_T(h) induce the same conjugation on \phi_\eta(E_Q)=E_Q. Lemma 3.8 therefore gives

If eh=e'h', then e'^{-1}e=h'h^{-1} lies in this intersection, so the formula is well defined. Equation (?) applied to (eh)(e'h')=e(he'h^{-1})hh' proves multiplicativity. The construction with \eta^{-1} gives the inverse.

At each g\in G_Q(R), two extensions have values inducing the same conjugation on \phi_\eta(E_Q)=E_Q, since E_Q is normal. Their quotient centralizes E_Q, so is trivial. This proves uniqueness. Compatibility with \pi holds on the root generators, hence on G_P=E_P.

The off-diagonal relative Cartan pairings are -1,-3, so the root words h_a(-1),h_b(-1) invert the other simple root parameters. With left commutators,

Weyl transport makes every root generator a commutator in E_Q. Since G_Q=E_QH_Q and H_Q is abelian, G_Q/E_Q is abelian, giving [G_Q,G_Q]=E_Q. The assertion for P follows from \pi_E.

Put D=\eta\rho-\rho\eta and E=\eta\rho-\rho^2\eta. The calculation in Lemma 5.11, using cancellation of 2 but not the domain property, gives D(t)E(t)=0 for all t. Choose t_0 so that the residue of \eta(t_0) is not fixed by \bar\rho, and hence has three distinct conjugates. Then D(t_0)-E(t_0) is a unit. By locality, one of D(t_0),E(t_0) is a unit and the other is zero. Polarization gives

forcing D=0 or E=0. Both cannot hold, since \eta is surjective and \rho\ne\rho^2. Conversely, a normalizer permutes \rho,\rho^2 and therefore preserves T.

After cancellation of 2, equations C_{15}=C_{25}=0 give

The determinant is 1 modulo J, so l_2=l_4=0. Since l_{11} is a unit, C_{19}=0 gives l_6=l_9; then C_{17}=0 gives

Equations C_{31}=C_{32}=0 now give

Its determinant is -1 modulo J, so l_{13}=l_{15}=0.

Next I_8=0 gives 2l_1l_3=0, hence l_1=0. Then I_1=0 gives (1-l_3)(1+l_3)=0, hence l_3=1. Similarly, I_{10}=I_2=0 give l_{14}=0 and l_{16}=1. Equation I_{14}=0 becomes -l_{10}(l_7+l_{12})=0, so l_{10}=0, since the second factor is 2 modulo J. Now I_{11}=0 gives l_5=0, and I_{15}=0 gives l_7=l_{12}. Using (Proof), equation I_3=0 reduces to 1-l_7^2=0. Thus l_7=l_{12}=1.

Finally, K_{11}=0 reduces to

Since 1+l_8-l_{11} is a unit, l_8-l_{11}=1. Set p=-l_{11} and z=-l_6. Equation (Proof) becomes p^2-p-z^2=0, giving the displayed matrix.

Conversely, direct multiplication gives

The correcting matrix is invertible because p^2-z^2=p is a unit. It centralizes the Tits group, so conjugation transports all standard long-root relations to X_l(p,z).