Introduction
On a compact torsion-free G 2 G_2 G 2 -manifold, Joyce showed that the moduli space of torsion-free G 2 G_2 G 2 -structures is locally modeled on harmonic 3-forms [37 ] . These forms decompose into irreducible G 2 G_2 G 2 -types. Through the identification with traceless symmetric tensors, the harmonic 27 -sector corresponds to traceless metric deformations, as described by Grigorian and Yau [31 ] and Karigiannis and Leung [40 ] . Lee and Leung introduced a natural symmetric cubic on the moduli space, called the Yukawa coupling [46 ] . On harmonic 27 -directions, it is obtained by integrating a pointwise cubic in the corresponding traceless symmetric tensors [31 , 40 ] .
In their 2002 preprint, Lee and Leung also proposed an exceptional algebraic interpretation of the Yukawa coupling. In [46 §1.2.1, Remark] , they suggested that the cubic structure on the 27-dimensional G 2 G_2 G 2 -module Λ 27 3 \Lambda^{3}_{27} Λ 27 3 should arise from the exceptional Jordan algebra, also known as the Albert algebra, with an underlying symmetry group of type E 6 E_6 E 6 . They further proposed that when the holonomy reduces from G 2 G_{2} G 2 to S U ( 3 ) SU(3) S U ( 3 ) , as on M = Y × S 1 \mathcal{M} = \mathcal{Y} \times S^{1} M = Y × S 1 for a Calabi–Yau threefold Y \mathcal{Y} Y , the distinguished direction associated with the Kähler form gives a splitting 27 = 1 ⊕ 26 \mathbf{27} = \mathbf{1} \oplus\mathbf{26} 27 = 1 ⊕ 26 and a corresponding reduction from E 6 E_{6} E 6 to F 4 F_{4} F 4 . In his thesis, the author also revisited this proposal and asked whether the natural product on Λ 27 3 \Lambda^{3}_{27} Λ 27 3 could be modified to satisfy the Jordan identity [24 §4.4.1, pp. 129–130] . The Albert algebra and the Jordan identity are recalled below. In § 3.3 we describe the two G 2 G_{2} G 2 -equivariant commutative products on 27 \mathbf{27} 27 .
Almost 25 years after their proposal, we resolve this question at the level of pointwise cubics, although the answer differs from the original proposal in one key respect. The pointwise Yukawa cubic is not itself an Albert norm (Proposition 4.16 ). The exceptional norms arise instead from precisely two non-real linear combinations of the Yukawa and trace cubics. We determine which members of the natural two-dimensional space of G 2 G_{2} G 2 -invariant cubics possess exceptional symmetry and describe the associated Albert algebras and bundle reductions. These cubics are defined for every G 2 G_{2} G 2 -structure, without compactness or a torsion hypothesis. Our main result is stated in Theorem A, while Theorem B describes the Albert algebra bundles associated with a choice of smooth line subbundle ℓ ⊂ T M \ell\subset TM ℓ ⊂ T M . Theorem C describes the implications for connections on the induced bundles.
Let M \mathcal{M} M be an oriented smooth 7-manifold with a G 2 G_{2} G 2 -structure, not necessarily torsion-free, given by a positive 3-form φ \varphi φ . The form induces a metric g g g and the rank-27 bundle
W φ = S 0 2 T ∗ M ≅ Λ 27 3 T ∗ M . \mathcal{W}_{\varphi}=S^{2}_{0}T^{*}\mathcal{M}\cong\Lambda^{3}_{27}T^{*}\mathcal{M}. W φ = S 0 2 T ∗ M ≅ Λ 27 3 T ∗ M .
At x ∈ M x\in\mathcal{M} x ∈ M , write V = T x M V=T_{x}\mathcal{M} V = T x M and W = S 0 2 V ∗ ≅ Λ 27 3 V ∗ W=S^{2}_{0}V^{*}\cong\Lambda^{3}_{27}V^{*} W = S 0 2 V ∗ ≅ Λ 27 3 V ∗ . We work in the symmetric-tensor model and write W φ , C = W φ ⊗ R C \mathcal{W}_{\varphi,\mathbb{C}}=\mathcal{W}_{\varphi}\otimes_{\mathbb{R}}\mathbb{C} W φ , C = W φ ⊗ R C for its complexification. Symmetric tensors describe pointwise metric variations and the corresponding metric-changing variations of φ \varphi φ . We explain this deformation interpretation in § 3.2 .
Throughout this paper we use the Einstein summation convention: an index repeated once upstairs and once downstairs is summed from 1 to 7, over all ordered values, unless another range is specified. Indices are raised and lowered with g g g . Explicit summation signs are retained for restricted sums and matrix-entry formulas. For a traceless symmetric tensor h h h , with indices raised by g g g , we define the two cubic polynomials
C 0 ( h ) = tr ( h 3 ) , and C φ ( h ) = φ a b c φ d e f h a d h b e h c f . (1) C_{0}(h)=\operatorname{tr}(h^{3}),\qquad\text{and}\qquad C_{\varphi}(h)=\varphi_{abc}\varphi_{def}h^{ad}h^{be}h^{cf}. \tag*{(1)} C 0 ( h ) = tr ( h 3 ) , and C φ ( h ) = φ ab c φ d e f h a d h b e h c f . (1)
Complexification extends all tensors complex-bilinearly, without complex conjugation. We also use the standard facts on G 2 G_{2} G 2 -structures from Bryant [3 ] ; the conventions and the G 2 G_{2} G 2 -contraction identities are collected in § 3 .
The two cubics span the G 2 G_{2} G 2 -invariant cubic space on W W W , over both R \mathbb{R} R and C \mathbb{C} C . This dimension count is standard and follows from the tensor-product decomposition in [20 Table A.121] . Both generators already occur together in [25 (4.30b)] and [31, (5.45)}, and have natural geometric interpretations. Up to normalization constants, C 0 C_{0} C 0 and C φ C_{\varphi} C φ arise as third affine variations of the pointwise volume on the spaces of metrics and positive 3-forms, respectively (Proposition 2.12 ). The cubic C φ C_{\varphi} C φ also describes the variation of the metric pairing of fixed 3-form directions (Proposition 2.11 ). The same cubic plane occurs in the expansion of the G 2 G_{2} G 2 Hitchin functional [12 §2.2] . Via the metric pairing, it also describes the two-dimensional space of G 2 G_{2} G 2 -equivariant commutative products on W W W [14 Proposition 3.8] .
On the moduli space of torsion-free G 2 G_2 G 2 -structures, the integrated cubic C φ C_\varphi C φ gives, up to normalization, the Yukawa coupling. Its relationship with the global volume functional was established by Karigiannis and Leung for irreducible G 2 G_2 G 2 -manifolds [40 ] . More recently, Langlais showed that, on harmonic 27-directions and under the hypothesis b 1 ( M ) = 0 b^1(M)=0 b 1 ( M ) = 0 , the Yukawa coupling agrees, up to a constant, with the third derivative of the logarithm of volume in period coordinates [44 Proposition 5.5] , [45 §3.2, Proposition 3.4] . We revisit the relationship between these cubics and the volume functional in § 3.4 .
We define the homogeneous pencil as
N a , b = a C 0 + b C φ , where [ a : b ] ∈ P 1 ( C ) and λ = b / a if a ≠ 0. N_{a,b}=aC_0+bC_\varphi,\qquad\text{where }[a:b]\in\mathbb{P}^1(\mathbb{C})\text{ and }\lambda=b/a\text{ if }a\ne0. N a , b = a C 0 + b C φ , where [ a : b ] ∈ P 1 ( C ) and λ = b / a if a = 0.
Fixing a = 1 a=1 a = 1 , we also define the affine pencil
N λ = C 0 + λ C φ for λ ∈ C . N_\lambda=C_0+\lambda C_\varphi\qquad\text{for }\lambda\in\mathbb{C}. N λ = C 0 + λ C φ for λ ∈ C .
We show that precisely two members of this pencil have exceptional E 6 E_6 E 6 -symmetry and are linearly equivalent to the complex Albert norm, and we determine the infinitesimal stabilizers of all remaining members. Here the exact stabilizer preserves the cubic itself, rather than only its one-dimensional span, and stab ( N ) \operatorname{stab}(N) stab ( N ) denotes its Lie algebra (Definition 3.1 ). We write E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) for the simply connected complex group of type E 6 E_6 E 6 . Its two mutually dual minuscule representations of dimension 27 are irreducible, and the weights of each form a single Weyl-group orbit; these are the representations appearing below. Set N ± = N λ ± N_\pm=N_{\lambda_\pm} N ± = N λ ± , where λ ± = ( 1 ± i 7 ) / 8 \lambda_\pm=(1\pm i\sqrt{7})/8 λ ± = ( 1 ± i 7 ) /8 .
Theorem A (The exceptional pencil). Let ( M , φ ) (M,\varphi) ( M , φ ) be a smooth G 2 G_2 G 2 -structure on a 7-manifold, with no torsion assumption, and take the cubics C 0 C_0 C 0 and C φ C_\varphi C φ in this order, with the normalization of (1) . Then the following hold:
(1) On each complexified fiber W C W_{\mathbb{C}} W C , the exact infinitesimal linear stabilizers of N a , b = a C 0 + b C φ N_{a,b}=aC_0+bC_\varphi N a , b = a C 0 + b C φ are
stab ( N a , b ) ≅ { s o ( 7 , C ) , [ a : b ] = [ 1 : 0 ] ( dim C = 21 ) , e 6 , a 2 − 2 a b + 8 b 2 = 0 ( dim C = 78 ) , g 2 C , otherwise ( dim C = 14 ) . \operatorname{stab}(N_{a,b})\cong
\begin{cases}
\mathfrak{so}(7,\mathbb{C}), & [a:b]=[1:0]\quad(\dim_{\mathbb{C}}=21),\\
\mathfrak{e}_6, & a^2-2ab+8b^2=0\quad(\dim_{\mathbb{C}}=78),\\
\mathfrak{g}_2^{\mathbb{C}}, & \text{otherwise}\quad(\dim_{\mathbb{C}}=14).
\end{cases} stab ( N a , b ) ≅ ⎩ ⎨ ⎧ so ( 7 , C ) , e 6 , g 2 C , [ a : b ] = [ 1 : 0 ] ( dim C = 21 ) , a 2 − 2 ab + 8 b 2 = 0 ( dim C = 78 ) , otherwise ( dim C = 14 ) . The two critical affine parameters are
λ ± = 1 ± i 7 8 , and we write N ± = C 0 + λ ± C φ . \lambda_\pm=\frac{1\pm i\sqrt{7}}{8},
\qquad\text{and we write}\qquad
N_\pm=C_0+\lambda_\pm C_\varphi. λ ± = 8 1 ± i 7 , and we write N ± = C 0 + λ ± C φ . The critical Lie algebras are simple of type E 6 E_6 E 6 , and each acts on W C W_{\mathbb{C}} W C through one of the two mutually dual 27-dimensional minuscule representations.
(2) For either sign, the full exact linear stabilizer is
G ± : = Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) , G_\pm:=\operatorname{Stab}_{GL(W_{\mathbb{C}})}(N_\pm)\cong E_6(\mathbb{C}), G ± := Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) , the simply connected group of type E 6 E_6 E 6 in its faithful minuscule representation. Moreover, N ± N_\pm N ± is linearly equivalent over C \mathbb{C} C to the cubic norm of the complex Albert algebra.
(3) The critical cubics N ± N_{\pm} N ± are global smooth sections of S 3 W φ , C ∗ S^{3}W^{*}_{\varphi,\mathbb{C}} S 3 W φ , C ∗ . Each defines a canonical reduction of the complex frame bundle of W φ , C W_{\varphi,\mathbb{C}} W φ , C to G ± ≅ E 6 ( C ) G_{\pm} \cong E_{6}(\mathbb{C}) G ± ≅ E 6 ( C ) , and complex conjugation exchanges the two reductions.
Theorem A follows from the critical-parameter characterization in Proposition 3.7 , the stabilizer computation in Proposition 3.31 and Theorem 3.3 , and the identification of the critical stabilizers with E 6 E_{6} E 6 in Theorems 3.32 and 4.1 . Corollary 3.33 collects the isomorphism types of all infinitesimal stabilizers. The Albert norm equivalence is shown in Theorem 4.7 , and the global reductions are constructed in Theorem 5.1 .
Theorem A has the following bundle consequences, established in § 6.1 and § 6.4 :
The complexified bundle S 3 W φ , C ∗ S^{3}W^{*}_{\varphi,\mathbb{C}} S 3 W φ , C ∗ admits two complex-conjugate trivial complex line subbundles generated by N ± N_{\pm} N ± .
The infinitesimal stabilizers of N ± N_{\pm} N ± define adjoint e 6 \mathfrak{e}_{6} e 6 -bundles associated with the G 2 G_{2} G 2 -structure bundle. With respect to the action of G 2 G_{2} G 2 , we have a decomposition e 6 ≅ g 2 ⊕ 64 \mathfrak{e}_{6} \cong\mathfrak{g}_{2} \oplus\mathbf{64} e 6 ≅ g 2 ⊕ 64 , and hence the 64 \mathbf{64} 64 -component provides the additional fiberwise infinitesimal symmetries of N ± N_{\pm} N ± beyond those of φ \varphi φ .
The full stabilizer groups of N ± N_{\pm} N ± define concrete, complex-conjugate E 6 ( C ) E_{6}(\mathbb{C}) E 6 ( C ) -principal bundles F ± \mathcal{F}_{\pm} F ± on M \mathcal{M} M , which canonically extend the original G 2 G_{2} G 2 structure group.
The metric also determines compact forms of the exceptional groups. Let ⟨ h , k ⟩ W = tr ( h k ) \langle h,k\rangle_{W}=\operatorname{tr}(hk) ⟨ h , k ⟩ W = tr ( hk ) . In each reduction, the frames that are unitary for the Hermitian form ⟨ h , k ‾ ⟩ W \langle h,\overline{k}\rangle_{W} ⟨ h , k ⟩ W form a principal bundle under the compact simply connected group of type E 6 E_{6} E 6 , and this bundle contains the adapted G 2 G_{2} G 2 frames (Proposition 4.18 and § 6.2 ).
The exceptional embedding of G 2 G_{2} G 2 in E 6 E_{6} E 6 and the corresponding branching of the adjoint representation are classical. Over C \mathbb{C} C , Dynkin identified two conjugacy classes of irreducible G 2 G_{2} G 2 subgroups in E 6 E_{6} E 6 acting irreducibly on the 27 \mathbf{27} 27 , exchanged by outer automorphisms, and for each of them, 78 = 14 ⊕ 64 \mathbf{78}=\mathbf{14}\oplus\mathbf{64} 78 = 14 ⊕ 64 [16 Theorem 11.1, Tables 24–26] . In positive characteristic, Testerman constructed the corresponding embeddings over fields containing − 7 \sqrt{-7} − 7 , under the characteristic restrictions of [58 Theorems 1(a), 2(a)] . She also reports Aschbacher’s result that, under suitable characteristic assumptions, exactly two invariant cubics up to scalar are of E 6 E_{6} E 6 -type if and only if the field contains − 7 \sqrt{-7} − 7 [58 p. 300] . In this paper, we locate these cubics explicitly in the plane generated by the geometric cubics C 0 , C φ C_{0}, C_{\varphi} C 0 , C φ , determine the infinitesimal stabilizers throughout the pencil, and express the additional symmetries through tensor operations on traceless symmetric 2 2 2 -tensors in 7 7 7 dimensions. This realization then supplies the exceptional bundle structures and compatible connections described below.
The complex Albert algebra A = H 3 ( O C ) \mathbb{A}=H_{3}(\mathbb{O}_{\mathbb{C}}) A = H 3 ( O C ) is the 27 27 27 -dimensional exceptional simple Jordan algebra of 3 × 3 3\times3 3 × 3 Hermitian matrices over the complexified octonions [22 Example 6.6 and Definition 6.7] . Its product is x ∘ y = 1 2 ( x y + y x ) x\circ y=\frac{1}{2}(xy+yx) x ∘ y = 2 1 ( x y + y x ) , with x y xy x y the octonionic matrix product. It is commutative and satisfies the Jordan identity x ∘ ( ( x ∘ y ) ∘ y ) = ( x ∘ x ) ∘ ( y ∘ y ) x\circ((x\circ y)\circ y)=(x\circ x)\circ(y\circ y) x ∘ (( x ∘ y ) ∘ y ) = ( x ∘ x ) ∘ ( y ∘ y ) , although it is not associative. Its cubic norm is the Albert determinant. The exact stabilizer of this norm is E 6 ( C ) E_{6}(\mathbb{C}) E 6 ( C ) , and the subgroup fixing the unit is F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) [22 §15] .
The critical cubic determines an Albert norm, but does not specify a Jordan product without a choice of unit. For either sign, an element e ∈ W C e\in W_{\mathbb{C}} e ∈ W C with N ± ( e ) ≠ 0 N_{\pm}(e)\ne0 N ± ( e ) = 0 determines a unital complex Albert algebra J e \mathcal{J}_{e} J e through the normalized cubic N ± / N ± ( e ) N_{\pm}/N_{\pm}(e) N ± / N ± ( e ) and the basepoint e e e (Theorem 4.9 ). This is the classical construction of a Jordan algebra from a cubic norm structure [56 , 48 ] . Its automorphism group is Stab G ± ( e ) ≅ F 4 ( C ) \operatorname{Stab}_{G_{\pm}}(e) \cong F_4(\mathbb{C}) Stab G ± ( e ) ≅ F 4 ( C ) (Proposition 4.12 ; [22 §15] ). Moreover, each nonzero norm level is a single G ± ≅ E 6 ( C ) G_{\pm} \cong E_6(\mathbb{C}) G ± ≅ E 6 ( C ) -orbit, isomorphic to E 6 ( C ) / F 4 ( C ) E_6(\mathbb{C})/F_4(\mathbb{C}) E 6 ( C ) / F 4 ( C ) (Theorem 4.9 ; [42 Proposition 12(i)] ). Thus the value of N ± N_{\pm} N ± distinguishes the orbits of basepoints under the exact norm stabilizer. Allowing a scalar adjustment, every basepoint can be transported to a fixed model basepoint, and all the resulting Albert algebras are isomorphic (Corollary 4.10 ; [23 Corollary 31.25] ).
A tangent line provides a natural geometric choice of unit. Let ℓ ⊂ T M \ell\subset TM ℓ ⊂ T M be a smooth real line subbundle, and let P ℓ P_{\ell} P ℓ be its orthogonal projection tensor. Define
E ℓ = 7 P ℓ − g . E_{\ell}=7P_{\ell}-g. E ℓ = 7 P ℓ − g .
This tensor depends only on the unoriented line and satisfies
N ± ( E ℓ ) = n ± = 42 ( 5 + 2 λ ± ) ≠ 0. N_{\pm}(E_{\ell})=n_{\pm}=42(5+2\lambda_{\pm})\ne0. N ± ( E ℓ ) = n ± = 42 ( 5 + 2 λ ± ) = 0.
This follows from (4.22). Consequently, ℓ \ell ℓ determines an F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reduction of each canonical E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) -reduction, together with a smooth bundle of unital complex Albert algebras. Their Jordan products are given explicitly in terms of g g g , φ \varphi φ , and P ℓ P_{\ell} P ℓ (Theorem 5.5). The basepoint E ℓ E_{\ell} E ℓ has residual G 2 G_2 G 2 -stabilizer S U ( 3 ) ⋊ Z / 2 SU(3)\mathbin{\rtimes}\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 (Proposition 4.13 ), realizing the E 6 E_6 E 6 -to-F 4 F_4 F 4 reduction anticipated by Lee and Leung [46 §1.2.1, Remark] . The construction also gives an explicit Euclidean real form of each based Albert algebra (Propositions 4.22 and 5.6 ).
Moreover, the residual stabilizer of any nonzero basepoint in W C W_{\mathbb{C}} W C has dimension at most 8 8 8 , and the line basepoints attain this bound (Proposition 4.15 ). Since W C W_{\mathbb{C}} W C has no nonzero G 2 G_2 G 2 -invariant element, the original G 2 G_2 G 2 -structure alone does not select a Jordan unit. On the other hand, the choice of a basepoint has no topological obstruction: E 6 ( C ) / F 4 ( C ) E_6(\mathbb{C})/F_4(\mathbb{C}) E 6 ( C ) / F 4 ( C ) is 8 8 8 -connected, and every F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reduction is homotopic to the one supplied by a line field (Proposition 5.7 ).
Theorem B (Albert algebra bundles). For either critical parameter and any smooth line subbundle ℓ ⊂ T M \ell\subset TM ℓ ⊂ T M , not necessarily orientable, the following hold:
(1) E ℓ E_{\ell} E ℓ is a smooth global section of W φ W_{\varphi} W φ , independent of the orientation of ℓ \ell ℓ , and N ± ( E ℓ ) = n ± N_{\pm}(E_{\ell})=n_{\pm} N ± ( E ℓ ) = n ± is a nonzero constant. Fixing E ℓ E_{\ell} E ℓ reduces the E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) -reduction of Theorem A to a principal F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -bundle.
(2) The based algebras at E ℓ E_{\ell} E ℓ form a smooth bundle of complex Albert algebras on W φ , C W_{\varphi,\mathbb{C}} W φ , C , with norm N ± / n ± N_{\pm}/n_{\pm} N ± / n ± and unit E ℓ E_{\ell} E ℓ . Their trace, trace form, based adjoint and Jordan product are globally defined and are invariant under u ↦ − u u\mapsto-u u ↦ − u for any local unit section u u u of ℓ \ell ℓ .
(3) There is an explicit smooth antilinear involutive automorphism σ ℓ \sigma_{\ell} σ ℓ of the Albert algebra bundle. Its fixed-point bundle is a bundle of Euclidean Albert algebras, with each fiber isomorphic to H 3 ( O ) H_3(\mathbb{O}) H 3 ( O ) , the Euclidean Albert algebra of 3 × 3 3\times3 3 × 3 Hermitian octonionic matrices.
This theorem combines the construction (5.3) of E ℓ E_{\ell} E ℓ with Lemma 5.4, Theorem 5.5 and Proposition 5.6. The reduction uses the standard associated-bundle construction. The algebra operations and the Euclidean real form realize classical Albert structures explicitly in the deformation bundle, in terms of g g g , φ \varphi φ and P ℓ P_\ell P ℓ .
The critical cubics also capture the first-order geometry of the G 2 G_2 G 2 -structure. Let T ∈ Ω 1 ( M , T M ) T \in\Omega^1(\mathcal{M},T\mathcal{M}) T ∈ Ω 1 ( M , T M ) denote its intrinsic torsion, defined by
D X L C φ = ι T ( X ) ψ , D^{LC}_X\varphi=\iota_{T(X)}\psi, D X L C φ = ι T ( X ) ψ ,
where ψ = ∗ g φ \psi=*_{g}\varphi ψ = ∗ g φ and D L C D^{LC} D L C denotes the Levi-Civita connection. Write Θ v ( w ) = w × v \Theta_v(w)=w\times v Θ v ( w ) = w × v , with the cross product of (8) , and let
D min = D L C − 1 3 Θ T ( ⋅ ) D^{\min}=D^{LC}-\frac{1}{3}\Theta_{T(\cdot)} D m i n = D L C − 3 1 Θ T ( ⋅ )
be the minimal G 2 G_2 G 2 -connection on T M T\mathcal{M} T M [7 §3.1] .
We use the same notation for connections on m a t h c a l M mathcal{M} ma t h c a l M and their induced connections on W φ , C W_{\varphi,\mathbb{C}} W φ , C , as well as, when compatible, the corresponding principal connections on the E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) -bundles F ± \mathcal{F}_{\pm} F ± . Write ρ : s o ( V ) → End ( W ) \rho:\mathfrak{so}(V)\to\operatorname{End}(W) ρ : so ( V ) → End ( W ) for the induced infinitesimal action on symmetric tensors, defined in (26) .
To understand connections on W φ , C W_{\varphi,\mathbb{C}} W φ , C with respect to the E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) -structure, we need to understand the corresponding Lie algebra. Indeed, with g ± = stab ( N ± ) \mathfrak{g}_{\pm}=\operatorname{stab}(N_{\pm}) g ± = stab ( N ± ) , we have the following decomposition, orthogonal for the complex-bilinear pairing ( X , Y ) ↦ tr W ( X Y ) (X,Y)\mapsto\operatorname{tr}_{W}(XY) ( X , Y ) ↦ tr W ( X Y ) :
End ( W C ) = C id W ⊕ g ± ⊕ c ± , g ± ≅ e 6 , c ± ≅ 650. \operatorname{End}(W_{\mathbb{C}})=\mathbb{C}\operatorname{id}_{W}\oplus\mathfrak{g}_{\pm}\oplus\mathfrak{c}_{\pm},\qquad\mathfrak{g}_{\pm}\cong\mathfrak{e}_{6},\quad\mathfrak{c}_{\pm}\cong650. End ( W C ) = C id W ⊕ g ± ⊕ c ± , g ± ≅ e 6 , c ± ≅ 650.
Theorem C (Torsion reconstruction and exceptional projection). Let ( M , φ ) (\mathcal{M},\varphi) ( M , φ ) be a smooth G 2 G_2 G 2 -structure, with no torsion assumption. Then:
(i) For either sign,
D X L C N ± = λ ± L φ ( T ( X ) ) , D^{LC}_X N_{\pm}=\lambda_{\pm}L_{\varphi}(T(X)), D X L C N ± = λ ± L φ ( T ( X )) , where
L φ : T M ⟶ S 3 W φ ∗ L_{\varphi}:T\mathcal{M}\longrightarrow S^3W_{\varphi}^{*} L φ : T M ⟶ S 3 W φ ∗ is a fiberwise injective G 2 G_2 G 2 -equivariant map. Thus the Levi-Civita derivative of either critical cubic uniquely determines the full intrinsic torsion of the G 2 G_2 G 2 -structure, and in particular, D L C N ± = 0 D^{LC}N_{\pm}=0 D L C N ± = 0 if and only if φ \varphi φ is torsion-free. More generally, a metric connection on T M T\mathcal{M} T M preserves φ \varphi φ if and only if its induced connection preserves either critical cubic, or even its line.
(ii) For either sign,
D L C − D min = 1 3 ρ ( Θ T ( ⋅ ) ) ∈ Ω 1 ( M , c ± ) . D^{LC}-D^{\min}=\frac{1}{3}\rho\bigl(\Theta_{T(\cdot)}\bigr)\in\Omega^1(\mathcal{M},\mathfrak{c}_{\pm}). D L C − D m i n = 3 1 ρ ( Θ T ( ⋅ ) ) ∈ Ω 1 ( M , c ± ) . The induced minimal connection is the unique connection on F ± \mathcal{F}_{\pm} F ± satisfying this c ± \mathfrak{c}_{\pm} c ± -valued difference condition. Equivalently, it is the orthogonal projection of the induced Levi-Civita connection onto F ± \mathcal{F}_{\pm} F ± , and the unique compatible connection pointwise closest to the induced Levi-Civita connection for the induced Hermitian norm. Both exceptional reductions give the same projection, which also preserves their compact reductions.
The torsion reconstruction follows from Lemma 5.10 and Corollary 5.12 . The exceptional projection follows from Lemma 5.14 and Proposition 5.15.
The connection D min D^{\min} D m i n is the minimal connection of the original tangent-bundle G 2 G_{2} G 2 -structure in the sense of Cleyton and Swann [8 Definition 3.1 and Proposition 3.2] . Their construction applies to a Riemannian G G G -structure on T M TM T M , with G ⊂ S O ( n ) G \subset SO(n) G ⊂ S O ( n ) , and projects the Levi-Civita connection using s o ( n ) = g ⊕ g ⊥ \mathfrak{so}(n) = \mathfrak{g} \oplus\mathfrak{g}^{\perp} so ( n ) = g ⊕ g ⊥ . The exceptional reductions considered here are instead structures on the auxiliary bundle W φ , C = S 0 2 T ∗ M ⊗ C W_{\varphi,\mathbb{C}} = S_{0}^{2}T^{*}M \otimes\mathbb{C} W φ , C = S 0 2 T ∗ M ⊗ C . Their orthogonal projections take place in End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) , with complement C id W ⊕ 650 \mathbb{C}\operatorname{id}_{W} \oplus\mathbf{650} C id W ⊕ 650 .
Theorem C shows that these two constructions are compatible: the connection induced by the minimal G 2 G_{2} G 2 -connection agrees with the projection of the induced Levi-Civita connection onto either exceptional reduction. The underlying reason is the representation-theoretic inclusion
ρ ( Λ 7 2 ) ⊂ c ± ≅ 650. \rho(\Lambda_{7}^{2}) \subset\mathfrak{c}_{\pm} \cong\mathbf{650}. ρ ( Λ 7 2 ) ⊂ c ± ≅ 650 .
Thus the original intrinsic torsion, which takes values in Λ 7 2 \Lambda_{7}^{2} Λ 7 2 , becomes a distinguished 7 7 7 -valued defect inside the exceptional 650 \mathbf{650} 650 .
This paper develops the pointwise algebra and global bundle geometry of the critical cubic pair for arbitrary smooth G 2 G_{2} G 2 -structures. Tangent line subbundles give explicit Albert products and Euclidean real forms, while the connection results relate these structures to intrinsic torsion and parallelism. In particular, the metric selects the same minimal G 2 G_{2} G 2 -connection on both exceptional reductions by projection of the induced Levi-Civita connection. Further work will investigate curvature conditions for compatible exceptional connections and the relation between this pointwise exceptional algebra and the integrated Yukawa coupling on moduli spaces of torsion-free G 2 G_{2} G 2 -structures.
Figure 1 summarizes the overall construction in the paper.
Figure 1. Overall construction of the paper.
We begin the algebraic argument with contraction identities for the two cubics C 0 C_{0} C 0 and C φ C_{\varphi} C φ . From these, we derive two universal quartic identities and use them to characterize the critical cubics through a double-adjoint identity (Proposition 3.7 ). For λ ∈ C \lambda\in\mathbb{C} λ ∈ C , define the quadratic adjoint maps F λ : W C → W C F_{\lambda} : W_{\mathbb{C}} \to W_{\mathbb{C}} F λ : W C → W C by
⟨ F λ ( h ) , k ⟩ W = N λ ( h , h , k ) , (2) \langle F_{\lambda}(h), k\rangle_{W} = N_{\lambda}(h,h,k), \tag*{(2)} ⟨ F λ ( h ) , k ⟩ W = N λ ( h , h , k ) , (2)
for h , k ∈ W C h,k \in W_{\mathbb{C}} h , k ∈ W C , where ⟨ h , k ⟩ W = tr ( h k ) \langle h,k\rangle_{W} = \operatorname{tr}(hk) ⟨ h , k ⟩ W = tr ( hk ) is the W W W -pairing, extended complex-bilinearly to W C W_{\mathbb{C}} W C . We use the same symbol N λ N_{\lambda} N λ for its symmetric trilinear polarization, normalized by N λ ( h , h , h ) = N λ ( h ) N_{\lambda}(h,h,h) = N_{\lambda}(h) N λ ( h , h , h ) = N λ ( h ) as in (34) .
Using quartic contraction identities for C 0 C_{0} C 0 and C φ C_{\varphi} C φ (Lemma 3.5 ), we reduce the double-adjoint condition F μ ( F λ ( h ) ) = κ N λ ( h ) h F_{\mu}(F_{\lambda}(h)) = \kappa N_{\lambda}(h)h F μ ( F λ ( h )) = κ N λ ( h ) h , for every h h h and some κ ∈ C \kappa\in\mathbb{C} κ ∈ C , to λ + μ = 1 4 \lambda+ \mu= \frac{1}{4} λ + μ = 4 1 and λ μ = 1 8 \lambda\mu= \frac{1}{8} λ μ = 8 1 . Thus λ , μ \lambda,\mu λ , μ must be the two roots of 8 λ 2 − 2 λ + 1 8\lambda^{2} - 2\lambda+ 1 8 λ 2 − 2 λ + 1 , which explains both the critical quadratic and the appearance of − 7 \sqrt{-7} − 7 (Proposition 3.7 and (82) ). Writing F ± = F λ ± F_{\pm} = F_{\lambda_{\pm}} F ± = F λ ± , the identity is
F ∓ ( F ± ( h ) ) = 2 3 N ± ( h ) h . (3) F_{\mp}(F_{\pm}(h)) = \frac{2}{3}N_{\pm}(h)h. \tag*{(3)} F ∓ ( F ± ( h )) = 3 2 N ± ( h ) h . (3)
A key feature is that the critical parameters are identified before the stabilizers of N ± N_{\pm} N ± are shown to be E 6 E_{6} E 6 . The double-adjoint identity is analogous to the classical identity adj ( adj A ) = det ( A ) A \operatorname{adj}(\operatorname{adj} A) = \det(A)A adj ( adj A ) = det ( A ) A for 3 × 3 3 \times3 3 × 3 matrices. Moreover, (1.2) gives F λ = 1 3 ∇ N λ F_{\lambda} = \frac{1}{3}\nabla N_{\lambda} F λ = 3 1 ∇ N λ with respect to the W W W -pairing, in analogy with the classical adjugate identity adj ( A ) = ∇ det ( A ) \operatorname{adj}(A) = \nabla\det(A) adj ( A ) = ∇ det ( A ) for symmetric matrices. Our quadratic adjoints pair the conjugate critical cubics N + N_{+} N + and N − N_{-} N − .
The double-adjoint identity also has an a posteriori explanation through the representation theory of E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) . This group has two inequivalent, mutually dual minuscule representations, 27 \mathbf{27} 27 and 27 ‾ = 27 ∗ \overline{\mathbf{27}} = \mathbf{27}^* 27 = 27 ∗ [54 §3, Table 1] . Their tensor products decompose as
S 2 27 = 27 ‾ ⊕ 351 ′ , S^2\mathbf{27} = \overline{\mathbf{27}} \oplus\mathbf{351}', S 2 27 = 27 ⊕ 351 ′ ,
27 ⊗ 27 ‾ = 1 ⊕ 78 ⊕ 650 , \mathbf{27} \otimes\overline{\mathbf{27}} = \mathbf{1} \oplus\mathbf{78} \oplus\mathbf{650}, 27 ⊗ 27 = 1 ⊕ 78 ⊕ 650 ,
S 3 27 = 1 ⊕ 650 ⊕ 3003. S^3\mathbf{27} = \mathbf{1} \oplus\mathbf{650} \oplus\mathbf{3003}. S 3 27 = 1 ⊕ 650 ⊕ 3003 .
See [20 ] for the representation data. Neither minuscule representation admits an invariant bilinear form, but the two carry a natural invariant pairing. The unique trivial summand in the symmetric cube corresponds to the Albert norm (Proposition 4.4 ; [57 12.4(ii)] ), while the occurrence of 27 ‾ \overline{\mathbf{27}} 27 in S 2 27 S^2\mathbf{27} S 2 27 gives, up to scale, a unique equivariant quadratic adjoint into the dual representation, and similarly in the reverse direction.
The composition of these adjoints is a quartic equivariant self-map of 27 \mathbf{27} 27 . It must be proportional to h ↦ N ( h ) h h \mapsto N(h)h h ↦ N ( h ) h : at a point of nonzero norm, the stabilizer F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) fixes precisely the line through that point, and E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) acts transitively on each nonzero norm level (Proposition 4.12 and Theorem 4.9 ). In our G 2 G_2 G 2 realization, the metric-induced pairing identifies the two dual modules as vector spaces, but not as E 6 E_6 E 6 -modules. The corresponding dual actions are realized by the two critical stabilizers, with quadratic adjoints F + F_+ F + and F − F_- F − (Remark 3.35). Thus representation theory explains the form of (3) once the exceptional structures have been identified. By contrast, Proposition 3.7 determines the critical parameters and the coefficient 2 3 \frac{2}{3} 3 2 directly from G 2 G_2 G 2 tensor identities, without assuming exceptional symmetry.
The double-adjoint identity (3) also allows us to construct infinitesimal symmetries through the Koecher operators, which are linear endomorphisms associated with the Tits–Kantor–Koecher construction (Remark 3.35). Explicitly, with F ± ( k , h ) F_\pm(k,h) F ± ( k , h ) denoting the halved bilinear polarization of F ± F_\pm F ± , they are
S ± ( k , l ) h = ⟨ h , l ⟩ W k − 2 F ∓ ( F ± ( k , h ) , l ) + 1 3 ⟨ k , l ⟩ W h . S_\pm(k,l)h = \langle h,l\rangle_W k - 2F_\mp(F_\pm(k,h),l) + \frac{1}{3}\langle k,l\rangle_W h. S ± ( k , l ) h = ⟨ h , l ⟩ W k − 2 F ∓ ( F ± ( k , h ) , l ) + 3 1 ⟨ k , l ⟩ W h .
By the double-adjoint identity, these operators are infinitesimal symmetries (Corollary 3.16 ). Separately, Proposition 3.13 excludes five G 2 G_2 G 2 -isotypic components of End W C \operatorname{End} W_{\mathbb C} End W C from the stabilizer of every nonzero member of the pencil. Namely, it excludes a 27 ⊕ 27 \mathbf{27} \oplus\mathbf{27} 27 ⊕ 27 -block, two blocks corresponding to the two inequivalent 77 77 77 -dimensional representations of G 2 G_2 G 2 , as well as the higher-dimensional 182 \mathbf{182} 182 and 189 \mathbf{189} 189 blocks. The next key step is to evaluate the span of the critical Koecher operators S ± ( k , l ) S_\pm(k,l) S ± ( k , l ) at a test tensor E u = 7 u b ⊗ u b − g E_u = 7u^b \otimes u^b - g E u = 7 u b ⊗ u b − g for a unit vector u u u . This evaluation maps onto E u ⊥ E_u^\perp E u ⊥ , of dimension 26. After the five exclusions, the available irreducible types other than the 64-dimensional module V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) have total dimension 1 + 7 + 14 = 22 1+7+14=22 1 + 7 + 14 = 22 . Hence a copy of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) must occur. Together with g 2 \mathfrak{g}_2 g 2 , it gives the lower bound 78 for the dimension of the critical stabilizer (Proposition 3.18 ).
Using this bound and the stabilizers of the two generators, we then identify the 64-parts of the two critical stabilizers as explicit graphs of G 2 G_2 G 2 -isomorphisms between the two copies of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in the 64 ⊗ 64 64 \otimes64 64 ⊗ 64 component of End W C \operatorname{End} W_{\mathbb C} End W C (Lemma 3.29 ). The count over the remaining isotypic components gives the exact stabilizer dimensions throughout the pencil (Proposition 3.31). Finally, we prove that the critical stabilizer Lie algebras are simple and identify their type as E 6 E_6 E 6 using the classification of simple Lie algebras and their irreducible representations (Theorem 3.32 ). By simplicity of the Lie algebra, the Koecher operators in fact span the whole critical stabilizer (Corollary 3.34), giving an explicit realization of e 6 \mathfrak{e}_6 e 6 on W C W_{\mathbb{C}} W C from the quadratic maps F ± F_{\pm} F ± . The double-adjoint characterization and the 78-dimensional lower bound precede, and do not use, the classification or the Albert identification.
Theorem A is proved in Sections 2–5, and Theorems B and C in Section 5. In Section 2, we introduce the deformation module, its invariant cubic plane and the description of the two generators as variations of the volume. In Section 3, we locate the critical pair and determine every stabilizer through the double-adjoint and Koecher argument just described. In Section 4, we determine the full stabilizer groups, identify the critical cubics with the Albert norm, discuss basepoints and their residual G 2 G_2 G 2 symmetry, compare the result with the Yukawa cubic, and construct the compact and Euclidean real forms. In Section 5, we return to the manifold: we construct the conjugate exceptional reductions and their compact forms, prove Theorem B and the homotopy uniqueness of the F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reductions, and conclude with compatible connections and parallel pointings, that is, parallel sections of nonzero critical norm.
In Appendix A, we derive the one- and two-index contraction identities for C 0 C_0 C 0 and C φ C_{\varphi} C φ (c-contraction identities) in detail, compute the contractions used in the stabilizer exclusion, and prove the corresponding quartic identities. Appendix A.4 gives the full notation inventory for this paper.
Acknowledgments
Codex (most recently, with GPT-6.1 Sol and GPT 6 Astra models) and Claude Code (most recently, with Opus 5.5 and Fable 5.1/5.2 models) were used to organize and verify the research artifacts behind this paper and to assist with typesetting. ChatGPT was used for assistance with ideation. All results have been verified by the author, who takes full responsibility for the content.
The traceless symmetric tensors and their natural cubics
The critical parameters of the pencil in Section 3, and the fields they generate, depend on the specified generators C 0 C_0 C 0 , C φ C_{\varphi} C φ and on the ground field. In this section we therefore fix explicitly the scale of the positive form, of the insertion map from symmetric tensors to 3-forms, and of the two cubics.
The standard form and its G 2 G_2 G 2 -contraction identities
Let V = R 7 V = \mathbb{R}^7 V = R 7 , with ordered basis ( e 1 , … , e 7 ) (e_1,\ldots,e_7) ( e 1 , … , e 7 ) and dual coframe ( e 1 , … , e 7 ) (e^1,\ldots,e^7) ( e 1 , … , e 7 ) . The Euclidean metric g g g , the orientation form vol \operatorname{vol} vol and the standard positive 3-form φ \varphi φ are
g = ∑ i = 1 7 e i ⊗ e i , and vol = e 1234567 , φ = e 123 + e 145 + e 167 + e 246 − e 257 − e 347 − e 356 , (4) \begin{aligned}
g &= \sum_{i=1}^{7} e^i \otimes e^i, \qquad\text{and} \qquad\operatorname{vol} = e^{1234567},\\
\varphi&= e^{123} + e^{145} + e^{167} + e^{246} - e^{257} - e^{347} - e^{356}, \tag*{(4)}
\end{aligned} g φ = i = 1 ∑ 7 e i ⊗ e i , and vol = e 1234567 , = e 123 + e 145 + e 167 + e 246 − e 257 − e 347 − e 356 , (4)
where e i 1 ⋯ i k = e i 1 ∧ ⋯ ∧ e i k e^{i_1\cdots i_k} = e^{i_1} \wedge\cdots\wedge e^{i_k} e i 1 ⋯ i k = e i 1 ∧ ⋯ ∧ e i k .
The Hodge star is determined by
α ∧ ∗ α ′ = ⟨ α , α ′ ⟩ vol , (5) \alpha\wedge*\alpha' = \langle\alpha,\alpha' \rangle\operatorname{vol}, \tag*{(5)} α ∧ ∗ α ′ = ⟨ α , α ′ ⟩ vol , (5)
where the increasing wedge monomials are orthonormal. In these conventions,
ψ = ∗ φ = e 4567 + e 2367 + e 2345 + e 1357 − e 1346 − e 1256 − e 1247 , \psi= *\varphi= e^{4567}+e^{2367}+e^{2345}+e^{1357}-e^{1346}-e^{1256}-e^{1247}, ψ = ∗ φ = e 4567 + e 2367 + e 2345 + e 1357 − e 1346 − e 1256 − e 1247 ,
∣ φ ∣ 2 = ∣ ψ ∣ 2 = 7 , and ∗ ψ = φ . (6) |\varphi|^2=|\psi|^2=7,\qquad\text{and}\qquad*\psi=\varphi. \tag*{(6)} ∣ φ ∣ 2 = ∣ ψ ∣ 2 = 7 , and ∗ ψ = φ . (6)
The signs agree with Bryant’s conventions [3 (2.1)–(2.8)] .
We use the index convention of § 1 ; in this adapted frame, g i j = δ i j g_{ij}=\delta_{ij} g ij = δ ij and g i j = δ i j g^{ij}=\delta^{ij} g ij = δ ij .
The alternating components are φ i j k = φ ( e i , e j , e k ) \varphi_{ijk}=\varphi(e_i,e_j,e_k) φ ij k = φ ( e i , e j , e k ) and ψ i j k l = ψ ( e i , e j , e k , e l ) \psi_{ijkl}=\psi(e_i,e_j,e_k,e_l) ψ ij k l = ψ ( e i , e j , e k , e l ) , so that
φ = 1 3 ! φ i j k e i j k , and ψ = 1 4 ! ψ i j k l e i j k l , with φ 123 = 1. \varphi=\frac{1}{3!}\varphi_{ijk}e^{ijk},\qquad\text{and}\qquad\psi=\frac{1}{4!}\psi_{ijkl}e^{ijkl},\qquad\text{with }\varphi_{123}=1. φ = 3 ! 1 φ ij k e ij k , and ψ = 4 ! 1 ψ ij k l e ij k l , with φ 123 = 1.
The factorials appear in the expansion of each form, not in its components.
The stabilizer
G 2 = { O ∈ G L ( V ) : O ∗ φ = φ } (7) G_2=\{O\in GL(V):O^*\varphi=\varphi\} \tag*{(7)} G 2 = { O ∈ G L ( V ) : O ∗ φ = φ } (7)
preserves g g g and vol \operatorname{vol} vol [3 (2.2)] . All the algebraic objects extend to V C = V ⊗ R C V_{\mathbb{C}}=V\otimes_{\mathbb{R}}\mathbb{C} V C = V ⊗ R C by complex multilinearity. In particular, no complex conjugation enters any bilinear or cubic formula, including the complex critical cubics of § 4 . These conventions are in characteristic zero, and the comparisons with positive characteristic cited in the text hold under their own stated hypotheses.
For u , v ∈ V u,v\in V u , v ∈ V , the cross product u × v ∈ V u\times v\in V u × v ∈ V is defined by
ι v ι u φ = ( u × v ) ♭ , equivalently ( u × v ) i = φ i a b u a v b , (8) \iota_v\iota_u\varphi=(u\times v)^\flat,\qquad\text{equivalently}\qquad(u\times v)_i=\varphi_{iab}u^av^b, \tag*{(8)} ι v ι u φ = ( u × v ) ♭ , equivalently ( u × v ) i = φ iab u a v b , (8)
so that g ( u × v , w ) = φ ( u , v , w ) g(u\times v,w)=\varphi(u,v,w) g ( u × v , w ) = φ ( u , v , w ) for w ∈ V w\in V w ∈ V .
The following contraction identities of φ \varphi φ and ψ \psi ψ are used throughout § 3 and § 4 and Appendix § 7 .
Lemma 2.1. For u , v ∈ V u,v\in V u , v ∈ V ,
( ι u φ ) ∧ ( ι v φ ) ∧ φ = 6 g ( u , v ) vol . (9) (\iota_u\varphi)\wedge(\iota_v\varphi)\wedge\varphi=6g(u,v)\operatorname{vol}. \tag*{(9)} ( ι u φ ) ∧ ( ι v φ ) ∧ φ = 6 g ( u , v ) vol . (9) The component contractions are
φ i a b φ j a b = 6 δ i j , and φ i j a φ k l a = δ i k δ j l − δ i l δ j k + ψ i j k l . (10) \varphi_{iab}\varphi_j{}^{ab}=6\delta_{ij},\qquad\text{and}\qquad\varphi_{ija}\varphi_{kl}{}^a=\delta_{ik}\delta_{jl}-\delta_{il}\delta_{jk}+\psi_{ijkl}. \tag*{(10)} φ iab φ j ab = 6 δ ij , and φ ij a φ k l a = δ ik δ j l − δ i l δ j k + ψ ij k l . (10) The mixed contraction of φ \varphi φ with ψ \psi ψ is
φ i a b ψ j k a b = 4 φ i j k . (11) \varphi_{iab}\psi_{jk}{}^{ab}=4\varphi_{ijk}. \tag*{(11)} φ iab ψ j k ab = 4 φ ij k . (11) Proof. The identities (2.6) and (2.7) follow by a componentwise check from (2.1) and (2.3). Substituting the second identity of (2.7) into the mixed contraction and then applying the first gives
φ i a b ψ j k a b = 6 φ j k i − 2 φ i j k = 4 φ i j k . \varphi_{iab}\psi_{jk}{}^{ab}=6\varphi_{jki}-2\varphi_{ijk}=4\varphi_{ijk}. φ iab ψ j k ab = 6 φ j k i − 2 φ ij k = 4 φ ij k .
In Bryant’s notation, after relabeling, the identities (2.7) and (2.8) are given in [3 (2.6)–(2.8)] . For exterior-algebra identities, see also [38 Lemma 2.2.3] . Component contractions are collected in [39 Lemmas A.12–A.14] , where the orientation convention gives the opposite sign for the ψ \psi ψ terms.
Symmetric tensors and pointwise deformations
A positive 3-form φ \varphi φ on a 7-manifold M M M is pointwise linearly equivalent to the standard model (2.1) and therefore determines a G 2 G_2 G 2 -structure, a metric g = g φ g=g_\varphi g = g φ , and an orientation [3 Section 3.1.1] . On both V V V and M M M , we write ψ = ∗ φ \psi=* \varphi ψ = ∗ φ .
Infinitesimal deformations of φ \varphi φ split into the irreducible summands Λ 1 3 \Lambda^3_1 Λ 1 3 , Λ 7 3 \Lambda^3_7 Λ 7 3 , and Λ 27 3 \Lambda^3_{27} Λ 27 3 . The first gives conformal metric variations, the second leaves the metric unchanged to first order, and the third gives traceless metric variations. Our cubics are sections of S 3 W φ ∗ S^3W_\varphi^* S 3 W φ ∗ , where
W φ = S 0 2 T ∗ M ≅ Λ 27 3 T ∗ M . W_\varphi=S^2_0T^*M\cong\Lambda^3_{27}T^*M. W φ = S 0 2 T ∗ M ≅ Λ 27 3 T ∗ M .
We first specify the deformation bundle and its frames, before describing its fibers as pointwise metric and form deformations.
Definition 2.2. Let ( M , φ ) (M,\varphi) ( M , φ ) be a 7-manifold with a G 2 G_2 G 2 -structure, with no assumption on its torsion, and let W = S 0 2 ( R 7 ) ∗ W=S^2_0(\mathbb{R}^7)^* W = S 0 2 ( R 7 ) ∗ be the model fiber.
(i) Fr φ → M \operatorname{Fr}_\varphi\to M Fr φ → M is the principal G 2 G_2 G 2 -bundle of φ \varphi φ -adapted coframes, which identify φ \varphi φ pointwise with the standard model (2.1), and Fr φ , C = Fr φ × G 2 G 2 C \operatorname{Fr}_{\varphi,\mathbb{C}}=\operatorname{Fr}_\varphi\times_{G_2}G_2^\mathbb{C} Fr φ , C = Fr φ × G 2 G 2 C is its complexification, with G 2 G_2 G 2 acting on G 2 C G_2^\mathbb{C} G 2 C by left multiplication. Here G 2 C G_2^\mathbb{C} G 2 C is the complex Lie group of type G 2 G_2 G 2 , the complexification of G 2 G_2 G 2 (recalled in Section 4).
(ii) The deformation bundle of φ \varphi φ is
W φ = S 0 2 T ∗ M = Fr φ × ρ W , W_\varphi=S^2_0T^*M=\operatorname{Fr}_\varphi\times_\rho W, W φ = S 0 2 T ∗ M = Fr φ × ρ W , where ρ ( O ) h = ( O − 1 ) ∗ h \rho(O)h=(O^{-1})^*h ρ ( O ) h = ( O − 1 ) ∗ h ( O ∈ G 2 ) (O\in G_2) ( O ∈ G 2 ) , and its complexification is
W φ , C = W φ ⊗ R C = Fr φ × ρ W C , W_{\varphi,\mathbb{C}}=W_\varphi\otimes_{\mathbb{R}}\mathbb{C}=\operatorname{Fr}_\varphi\times_\rho W_{\mathbb{C}}, W φ , C = W φ ⊗ R C = Fr φ × ρ W C , a complex vector bundle of rank 27.
(iii) For a coframe p ∈ Fr φ p\in\operatorname{Fr}_\varphi p ∈ Fr φ over x ∈ M x\in M x ∈ M , the induced frame of the fiber is
f p : W C → W φ , C , x , with f p ( h ) = [ p , h ] . f_p:W_{\mathbb{C}}\to W_{\varphi,\mathbb{C},x},\qquad\text{with}\qquad f_p(h)=[p,h]. f p : W C → W φ , C , x , with f p ( h ) = [ p , h ] .
Fiberwise, the insertion map I φ \mathcal{I}_\varphi I φ defined below identifies W φ W_\varphi W φ with Λ 27 3 T ∗ M \Lambda^3_{27}T^*M Λ 27 3 T ∗ M .
We use the tensor model S 0 2 T ∗ M S^2_0T^*M S 0 2 T ∗ M throughout; the form model Λ 27 3 T ∗ M \Lambda^3_{27}T^*M Λ 27 3 T ∗ M enters through I φ \mathcal{I}_\varphi I φ . Note that W φ W_\varphi W φ actually only depends on the metric g g g , since trace removal in S 2 T ∗ M S^2T^*M S 2 T ∗ M requires the metric. Hence, W φ W_\varphi W φ can also be regarded as an associated bundle to the orthonormal frame bundle. In our case, since we assume a G 2 G_2 G 2 -structure, which induces the metric g = g φ g=g_\varphi g = g φ , we regard W φ W_\varphi W φ as a bundle associated to Fr φ \operatorname{Fr}_\varphi Fr φ .
The cubic C 0 C_0 C 0 depends only on the metric, while C φ C_\varphi C φ also depends on φ \varphi φ , as in (2.27) below.
We now fix x ∈ M x\in M x ∈ M and identify V = T x M V=T_xM V = T x M with the model R 7 \mathbb{R}^7 R 7 by an adapted coframe. Throughout the rest of this section and Sections 3–4 we work in this fiber, with W = W φ , x ≅ S 0 2 V ∗ W=W_{\varphi,x}\cong S^2_0V^* W = W φ , x ≅ S 0 2 V ∗ .
Elements of S 2 V ∗ S^2V^* S 2 V ∗ are covariant symmetric 2-tensors, and we multiply covariant 2-tensors by contracting through the metric:
( h k ) a b = h a c g c d k d b . (12) (hk)_{ab}=h_{ac}g^{cd}k_{db}. \tag*{(12)} ( hk ) ab = h a c g c d k d b . (12)
This product is associative and has g g g as its identity, but h k hk hk is in general neither equal to k h kh k h nor symmetric. We write h n h^n h n for the n n n -fold contracted product, with h 0 = g h^0=g h 0 = g , and tr ( h ) = g a b h a b \operatorname{tr}(h)=g^{ab}h_{ab} tr ( h ) = g ab h ab , so that tr ( h n ) = g a b ( h n ) a b \operatorname{tr}(h^n)=g^{ab}(h^n)_{ab} tr ( h n ) = g ab ( h n ) ab and tr ( h k ) = tr ( k h ) \operatorname{tr}(hk)=\operatorname{tr}(kh) tr ( hk ) = tr ( k h ) . The transpose of a covariant 2-tensor is ( h ⊤ ) a b = h b a (h^\top)_{ab}=h_{ba} ( h ⊤ ) ab = h ba . The metric identifies h h h with the endomorphism
g − 1 h ∈ End ( V ) , ( g − 1 h ) a b = g a c h c b , (13) g^{-1}h \in\operatorname{End}(V), \qquad(g^{-1}h)^a{}_b=g^{ac}h_{cb}, \tag*{(13)} g − 1 h ∈ End ( V ) , ( g − 1 h ) a b = g a c h c b , (13)
which is self-adjoint with respect to g g g exactly when h h h is symmetric. This identification carries contracted products to compositions, g − 1 ( h k ) = ( g − 1 h ) ( g − 1 k ) g^{-1}(hk)=(g^{-1}h)(g^{-1}k) g − 1 ( hk ) = ( g − 1 h ) ( g − 1 k ) , preserves traces, and sends g g g to the identity endomorphism I I I of V V V . We use it when compositions, commutators or eigenspaces are needed, without a separate symbol for g − 1 h g^{-1}h g − 1 h , and write id W \operatorname{id}_W id W for the identity endomorphism of W W W . In a g g g -orthonormal frame, h h h and g − 1 h g^{-1}h g − 1 h have the same matrix, g g g is represented by the identity matrix, the contracted product by the matrix product and h ⊤ h^\top h ⊤ by the matrix transpose; coordinate calculations below use this matrix notation.
Definition 2.3. The space of traceless covariant symmetric 2-tensors and its pairing induced by g g g are
W = S 0 2 V ∗ = { h ∈ S 2 V ∗ : tr ( h ) = 0 } , ⟨ h , k ⟩ W = g a c g b d h a b k c d = tr ( h k ) . (14) W=S^2_0V^*=\{h\in S^2V^*:\operatorname{tr}(h)=0\}, \qquad\langle h,k\rangle_W=g^{ac}g^{bd}h_{ab}k_{cd}=\operatorname{tr}(hk). \tag*{(14)} W = S 0 2 V ∗ = { h ∈ S 2 V ∗ : tr ( h ) = 0 } , ⟨ h , k ⟩ W = g a c g b d h ab k c d = tr ( hk ) . (14) We call ⟨ ⋅ , ⋅ ⟩ W \langle\cdot,\cdot\rangle_W ⟨ ⋅ , ⋅ ⟩ W the W W W -pairing, and use the same notation for its complex-bilinear extension to W C W_{\mathbb{C}} W C ; no complex conjugation is involved.
We use the following tensor conventions throughout the paper. For h ∈ S 2 V ∗ h\in S^2V^* h ∈ S 2 V ∗ , define its traceless part h 0 h_0 h 0 by
h 0 = h − 1 7 tr ( h ) g . (15) h_0=h-\frac{1}{7}\operatorname{tr}(h)g. \tag*{(15)} h 0 = h − 7 1 tr ( h ) g . (15)
Let Σ ⊆ V \Sigma\subseteq V Σ ⊆ V be a subspace with orthogonal projection pr Σ ∈ End ( V ) \operatorname{pr}_{\Sigma}\in\operatorname{End}(V) pr Σ ∈ End ( V ) . For u ∈ V u\in V u ∈ V , or u ∈ V C u\in V_{\mathbb{C}} u ∈ V C with ♭ \flat ♭ extended complex-linearly, define
P u = u ♭ ⊗ u ♭ , P Σ ( v , w ) = g ( pr Σ v , w ) , (16) P_u=u^\flat\otimes u^\flat, \qquad P_{\Sigma}(v,w)=g(\operatorname{pr}_{\Sigma}v,w), \tag*{(16)} P u = u ♭ ⊗ u ♭ , P Σ ( v , w ) = g ( pr Σ v , w ) , (16)
so that g − 1 P Σ = pr Σ g^{-1}P_{\Sigma}=\operatorname{pr}_{\Sigma} g − 1 P Σ = pr Σ and P Σ ⊥ = g − P Σ P_{\Sigma^\perp}=g-P_{\Sigma} P Σ ⊥ = g − P Σ . For a unit vector u u u , g − 1 P u = pr R u g^{-1}P_u=\operatorname{pr}_{\mathbb{R}u} g − 1 P u = pr R u , so that P u P_u P u depends only on the line ℓ = R u \ell=\mathbb{R}u ℓ = R u and equals P ℓ P_{\ell} P ℓ ; we write P ℓ P_{\ell} P ℓ when a line is specified and P u P_u P u when only a vector is. For a general u u u , P u P_u P u is only a rank-one tensor, and for a complex null vector g − 1 P u g^{-1}P_u g − 1 P u is nilpotent. The calligraphic letter P \mathcal{P} P with two arguments is reserved for the bilinear operation P ( h , k ) \mathcal{P}(h,k) P ( h , k ) defined below. The traceless parts of these tensors, rescaled by 7, serve as test tensors and basepoints throughout the paper. For v ∈ V v\in V v ∈ V or v ∈ V C v\in V_{\mathbb{C}} v ∈ V C , and for a subspace Σ ⊆ V \Sigma\subseteq V Σ ⊆ V , define
E v = 7 ( P v ) 0 = 7 P v − g ( v , v ) g , E Σ = 7 ( P Σ ) 0 = 7 P Σ − ( dim Σ ) g . (17) E_v=7(P_v)_0=7P_v-g(v,v)g, \qquad E_{\Sigma}=7(P_{\Sigma})_0=7P_{\Sigma}-(\dim\Sigma)g. \tag*{(17)} E v = 7 ( P v ) 0 = 7 P v − g ( v , v ) g , E Σ = 7 ( P Σ ) 0 = 7 P Σ − ( dim Σ ) g . (17)
For a unit vector u u u this is E u = 7 P u − g E_u=7P_u-g E u = 7 P u − g . It depends only on the line ℓ = R u \ell=\mathbb{R}u ℓ = R u , and we write E ℓ = 7 P ℓ − g E_{\ell}=7P_{\ell}-g E ℓ = 7 P ℓ − g when the line is specified. For a 3-plane Σ \Sigma Σ , E Σ = 7 P Σ − 3 g = 4 P Σ − 3 P Σ ⊥ E_{\Sigma}=7P_{\Sigma}-3g=4P_{\Sigma}-3P_{\Sigma^\perp} E Σ = 7 P Σ − 3 g = 4 P Σ − 3 P Σ ⊥ . For the basis vectors e i e_i e i of (2.1) we abbreviate P i = P e i P_i=P_{e_i} P i = P e i and E i = E e i E_i=E_{e_i} E i = E e i .
For h ∈ S 2 V ∗ h\in S^2V^* h ∈ S 2 V ∗ , define the insertion map by
I φ ( h ) = d d t ∣ t = 0 ( I + t h ) ∗ φ , \mathcal{I}_{\varphi}(h)=\left.\frac{d}{dt}\right|_{t=0}(I+th)^*\varphi, I φ ( h ) = d t d t = 0 ( I + t h ) ∗ φ ,
( I φ ( h ) ) i j k = h i a φ a j k + h j a φ i a k + h k a φ i j a . (18) (\mathcal{I}_{\varphi}(h))_{ijk}=h_i{}^a\varphi_{ajk}+h_j{}^a\varphi_{iak}+h_k{}^a\varphi_{ija}. \tag*{(18)} ( I φ ( h ) ) ij k = h i a φ aj k + h j a φ iak + h k a φ ij a . (18)
We write χ h = I φ ( h ) \chi_h=I_\varphi(h) χ h = I φ ( h ) . For h = g h=g h = g , (18) gives I φ ( g ) = 3 φ I_\varphi(g)=3\varphi I φ ( g ) = 3 φ . Note that in [25 , 31 ] , the averaged insertion i φ ( h ) = I φ ( h ) / 3 i_\varphi(h)=I_\varphi(h)/3 i φ ( h ) = I φ ( h ) /3 is used. In [15 ] , the notation h ⋄ φ h\diamond\varphi h ⋄ φ is used, and is extended to not necessarily symmetric h h h . The standard G 2 G_2 G 2 -decomposition of 3-forms [3 ] , [38 (2.19)–(2.21)] is the orthogonal sum
Λ 3 V ∗ = R φ ⊕ { ι v ψ : v ∈ V } ⊕ Λ 27 3 V ∗ , where Λ 27 3 V ∗ = { χ : χ ∧ φ = 0 , χ ∧ ψ = 0 } . (19) \Lambda^3V^*=\mathbb{R}\varphi\oplus\{\iota_v\psi:v\in V\}\oplus\Lambda^3_{27}V^*,\qquad\text{where}\qquad\Lambda^3_{27}V^*=\{\chi:\chi\wedge\varphi=0,\ \chi\wedge\psi=0\}. \tag*{(19)} Λ 3 V ∗ = R φ ⊕ { ι v ψ : v ∈ V } ⊕ Λ 27 3 V ∗ , where Λ 27 3 V ∗ = { χ : χ ∧ φ = 0 , χ ∧ ψ = 0 } . (19)
We write pr 1 \operatorname{pr}_1 pr 1 , pr 7 \operatorname{pr}_7 pr 7 and pr 27 \operatorname{pr}_{27} pr 27 for the projections onto the three summands of (19) , in this order; for another positive form, they denote the projections onto the summands of its splitting. The following contractions, with χ h = I φ ( h ) \chi_h=I_\varphi(h) χ h = I φ ( h ) as in (18) , fix the normalization of I φ I_\varphi I φ .
Lemma 2.4. For h , k ∈ S 2 V ∗ h,k\in S^2V^* h , k ∈ S 2 V ∗ , in the inner product of forms (5) ,
( χ h ) i a b φ j a b = 4 h i j + 2 ( tr h ) δ i j , (20) (\chi_h)_{iab}\varphi_j^{ab}=4h_{ij}+2(\operatorname{tr}h)\delta_{ij}, \tag*{(20)} ( χ h ) iab φ j ab = 4 h ij + 2 ( tr h ) δ ij , (20) ⟨ I φ ( h ) , I φ ( k ) ⟩ = 2 tr ( h k ) + ( tr h ) ( tr k ) . (21) \langle I_\varphi(h),I_\varphi(k)\rangle=2\operatorname{tr}(hk)+(\operatorname{tr}h)(\operatorname{tr}k). \tag*{(21)} ⟨ I φ ( h ) , I φ ( k )⟩ = 2 tr ( hk ) + ( tr h ) ( tr k ) . (21) In particular, for h ∈ W h\in W h ∈ W ,
∣ χ h ∣ 2 = 2 tr ( h 2 ) , and h i j = 1 4 ( χ h ) i a b φ j a b . (22) |\chi_h|^2=2\operatorname{tr}(h^2),\qquad\text{and}\qquad h_{ij}=\frac{1}{4}(\chi_h)_{iab}\varphi_j^{ab}. \tag*{(22)} ∣ χ h ∣ 2 = 2 tr ( h 2 ) , and h ij = 4 1 ( χ h ) iab φ j ab . (22)
Proof. For (20) , expand the three terms of χ h \chi_h χ h and apply (10) . The first term contributes 6 h i j 6h_{ij} 6 h ij ; each of the other two contributes ( tr h ) δ i j − h i j (\operatorname{tr}h)\delta_{ij}-h_{ij} ( tr h ) δ ij − h ij . The contraction with ψ \psi ψ vanishes because h h h is symmetric. For the inner product, we use the component pairing with its factor 1 / 6 1/6 1/6 . By alternation, the three terms of I φ ( k ) I_\varphi(k) I φ ( k ) contribute equally. Thus we obtain one half of (20) paired with k k k , which proves (21) . The identities (22) on the traceless subspace follow.
For h ∈ W h\in W h ∈ W , the trace of (20) gives ⟨ χ h , φ ⟩ = 1 6 ( χ h ) i a b φ i a b = 3 tr h = 0 \langle\chi_h,\varphi\rangle=\frac{1}{6}(\chi_h)_{iab}\varphi^{iab}=3\operatorname{tr}h=0 ⟨ χ h , φ ⟩ = 6 1 ( χ h ) iab φ iab = 3 tr h = 0 . The three summands of χ h \chi_h χ h in (18) have the same contraction with ψ m i j k \psi_{mijk} ψ mij k after cyclic relabeling, so (11) gives
( χ h ) i j k ψ m i j k = 3 h i a φ j k a ψ m i j k = 12 h i a φ m a i = 0 , (\chi_h)_{ijk}\psi_m^{ijk}=3h_{ia}\varphi^a_{jk}\psi_m^{ijk}=12h_{ia}\varphi^{a i}_m=0, ( χ h ) ij k ψ m ij k = 3 h ia φ j k a ψ m ij k = 12 h ia φ m ai = 0 ,
by symmetry of h h h . Thus χ h \chi_h χ h is orthogonal to ι e m ψ \iota_{e_m}\psi ι e m ψ for every m m m . By the orthogonal splitting (19) , χ h ∈ Λ 27 3 V ∗ \chi_h\in\Lambda^3_{27}V^* χ h ∈ Λ 27 3 V ∗ . Directly from the explicit forms, φ , ι e 1 ψ , … , ι e 7 ψ \varphi,\iota_{e_1}\psi,\ldots,\iota_{e_7}\psi φ , ι e 1 ψ , … , ι e 7 ψ are mutually orthogonal with squared norms 7 , 4 , … , 4 7,4,\ldots,4 7 , 4 , … , 4 ; in particular, ∣ ι v ψ ∣ 2 = 4 ∣ v ∣ 2 |\iota_v\psi|^2=4|v|^2 ∣ ι v ψ ∣ 2 = 4∣ v ∣ 2 , and Λ 27 3 V ∗ \Lambda^3_{27}V^* Λ 27 3 V ∗ has dimension 35 − 8 = 27 35-8=27 35 − 8 = 27 . The identity (22) gives injectivity, and W W W also has dimension 27 27 27 , so I φ : W → Λ 27 3 V ∗ I_\varphi:W\to\Lambda^3_{27}V^* I φ : W → Λ 27 3 V ∗ is an isomorphism. We use the pullback derivative itself, rather than the isometric rescaling I φ / 2 I_\varphi/\sqrt{2} I φ / 2 .
The interpretation of W W W in terms of deformations follows from the same pullback curve. For h ∈ S 2 V ∗ h\in S^2V^* h ∈ S 2 V ∗ , the induced metric is ( I + t h ) ∗ g (I+th)^*g ( I + t h ) ∗ g , whose derivative at t = 0 t=0 t = 0 is 2 h 2h 2 h . Thus h ∈ W h\in W h ∈ W describes a pointwise traceless metric variation, and χ h \chi_h χ h is the corresponding metric-changing variation of φ \varphi φ . We call W W W the deformation representation (or module) in this sense, and W φ W_\varphi W φ its associated deformation bundle. Their elements are pointwise directions, and the terminology involves no differential equation or harmonic condition. For finite deformation formulas for the induced metric and torsion, see [27, §4; Lemma 5.1, equation (5.31)].
For a Lie-algebra element M ∈ g l ( V ) M \in\mathfrak{gl}(V) M ∈ gl ( V ) , define θ M \theta_M θ M , the infinitesimal pullback of φ \varphi φ by M M M , by
θ M ( v 1 , v 2 , v 3 ) = φ ( M v 1 , v 2 , v 3 ) + φ ( v 1 , M v 2 , v 3 ) + φ ( v 1 , v 2 , M v 3 ) , (23) \theta_M(v_1,v_2,v_3)=\varphi(Mv_1,v_2,v_3)+\varphi(v_1,Mv_2,v_3)+\varphi(v_1,v_2,Mv_3), \tag*{(23)} θ M ( v 1 , v 2 , v 3 ) = φ ( M v 1 , v 2 , v 3 ) + φ ( v 1 , M v 2 , v 3 ) + φ ( v 1 , v 2 , M v 3 ) , (23)
g 2 = Lie ( G 2 ) = { M ∈ g l ( V ) : θ M = 0 } ⊆ s o ( 7 ) . (24) \mathfrak{g}_2=\operatorname{Lie}(G_2)=\{M\in\mathfrak{gl}(V):\theta_M=0\}\subseteq\mathfrak{so}(7). \tag*{(24)} g 2 = Lie ( G 2 ) = { M ∈ gl ( V ) : θ M = 0 } ⊆ so ( 7 ) . (24)
The orthogonal group acts on covariant 2-tensors by the natural left action, which preserves W W W :
ρ ( O ) h = ( O − 1 ) ∗ h , so that g − 1 ( ρ ( O ) h ) = O ( g − 1 h ) O − 1 ( O ∈ O ( 7 ) ) . (25) \rho(O)h=(O^{-1})^*h,\qquad\text{so that}\qquad g^{-1}(\rho(O)h)=O(g^{-1}h)O^{-1}\quad(O\in O(7)). \tag*{(25)} ρ ( O ) h = ( O − 1 ) ∗ h , so that g − 1 ( ρ ( O ) h ) = O ( g − 1 h ) O − 1 ( O ∈ O ( 7 )) . (25)
Its derivative is
ρ : s o ( 7 ) ⟶ End ( W ) , \rho:\mathfrak{so}(7)\longrightarrow\operatorname{End}(W), ρ : so ( 7 ) ⟶ End ( W ) ,
given by
ρ ( M ) h = g ( [ M , g − 1 h ] , ⋅ ) . (26) \rho(M)h=g([M,g^{-1}h],\mathord\cdot). \tag*{(26)} ρ ( M ) h = g ([ M , g − 1 h ] , ⋅ ) . (26)
that is, ( ρ ( M ) h ) ( u , v ) = − h ( M u , v ) − h ( u , M v ) (\rho(M)h)(u,v)=-h(Mu,v)-h(u,Mv) ( ρ ( M ) h ) ( u , v ) = − h ( M u , v ) − h ( u , M v ) , the commutator being taken in End ( V ) \operatorname{End}(V) End ( V ) . In a g g g -orthonormal frame ρ ( M ) h = [ M , h ] \rho(M)h=[M,h] ρ ( M ) h = [ M , h ] as matrices, and we use this form in matrix calculations. Each ρ ( M ) \rho(M) ρ ( M ) is skew-adjoint for the W W W -pairing (W W W -skew-adjoint in the terminology introduced after (31) below), since cyclicity of trace gives
⟨ ρ ( M ) h , k ⟩ W + ⟨ h , ρ ( M ) k ⟩ W = tr ( [ M , g − 1 h ] g − 1 k + g − 1 h [ M , g − 1 k ] ) = 0. \langle\rho(M)h,k\rangle_W+\langle h,\rho(M)k\rangle_W=\operatorname{tr}\bigl([M,g^{-1}h]g^{-1}k+g^{-1}h[M,g^{-1}k]\bigr)=0. ⟨ ρ ( M ) h , k ⟩ W + ⟨ h , ρ ( M ) k ⟩ W = tr ( [ M , g − 1 h ] g − 1 k + g − 1 h [ M , g − 1 k ] ) = 0.
This representation is injective, and its restriction to g 2 \mathfrak{g}_2 g 2 is also an injective representation on W W W . We use the standard identification of g 2 \mathfrak{g}_2 g 2 as the compact simple Lie algebra of type G 2 G_2 G 2 , of dimension 14, and write g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C for its complexification.
Using g g g we identify s o ( 7 ) \mathfrak{so}(7) so ( 7 ) with Λ 2 V ∗ \Lambda^2V^* Λ 2 V ∗ . The standard splitting of 2-forms is
Λ 2 V ∗ = Λ 7 2 ⊕ Λ 14 2 , Λ 7 2 = { ι v φ : v ∈ V } = { α : ψ a b c d α c d = 4 α a b } , \Lambda^2V^*=\Lambda^2_7\oplus\Lambda^2_{14},\qquad
\Lambda^2_7=\{\iota_v\varphi:v\in V\}=\{\alpha:\psi_{abcd}\alpha^{cd}=4\alpha_{ab}\}, Λ 2 V ∗ = Λ 7 2 ⊕ Λ 14 2 , Λ 7 2 = { ι v φ : v ∈ V } = { α : ψ ab c d α c d = 4 α ab } ,
Λ 14 2 = { α : φ a b c α b c = 0 } = { α : ψ a b c d α c d = − 2 α a b } = g 2 , (27) \Lambda^2_{14}=\{\alpha:\varphi_{abc}\alpha^{bc}=0\}=\{\alpha:\psi_{abcd}\alpha^{cd}=-2\alpha_{ab}\}=\mathfrak{g}_2, \tag*{(27)} Λ 14 2 = { α : φ ab c α b c = 0 } = { α : ψ ab c d α c d = − 2 α ab } = g 2 , (27)
where ( ι v φ ) b c = v a φ a b c (\iota_v\varphi)_{bc}=v^a\varphi_{abc} ( ι v φ ) b c = v a φ ab c [3 ] . The identification of Λ 14 2 \Lambda^2_{14} Λ 14 2 with the Lie algebra of G 2 G_2 G 2 is given in [31 (3.33), (3.37)] , and the two eigenvalues of α ↦ ( ψ a b c d α c d ) a b \alpha\mapsto(\psi_{abcd}\alpha^{cd})_{ab} α ↦ ( ψ ab c d α c d ) ab follow from the G 2 G_2 G 2 -contraction identities (10) and (11) . In particular, the contraction α ↦ ( φ a b c α b c ) a \alpha\mapsto(\varphi_{abc}\alpha^{bc})_a α ↦ ( φ ab c α b c ) a maps ι v φ \iota_v\varphi ι v φ to 6 v 6v 6 v , so dim Λ 7 2 = 7 \dim\Lambda^2_7=7 dim Λ 7 2 = 7 and dim g 2 = 21 − 7 = 14 \dim\mathfrak{g}_2=21-7=14 dim g 2 = 21 − 7 = 14 .
Write V ( j 1 , j 2 ) V(j_1,j_2) V ( j 1 , j 2 ) for the irreducible g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C -module of highest weight j 1 ω 1 + j 2 ω 2 j_1\omega_1+j_2\omega_2 j 1 ω 1 + j 2 ω 2 , where the first simple root is short. The fundamental weights ω 1 , ω 2 \omega_1,\omega_2 ω 1 , ω 2 give the defining seven-dimensional and adjoint representations, respectively, so V ( 0 , 1 ) = g 2 C V(0,1)=\mathfrak{g}_2^{\mathbb{C}} V ( 0 , 1 ) = g 2 C and W C = V ( 2 , 0 ) W_{\mathbb{C}}=V(2,0) W C = V ( 2 , 0 ) . We also use Bryant’s identification V ( j , 0 ) = S 0 j V C V(j,0)=S^j_0V_{\mathbb{C}} V ( j , 0 ) = S 0 j V C , the traceless symmetric j j j -tensors [3 ] . The relevant decompositions are recorded in the tensor-product tables [20 Table A.121], [60, Table 1500] . A boldface number n \mathbf{n} n denotes an irreducible representation of dimension n n n of the group under consideration.
Proposition 2.5. The symmetric and alternating squares of W C W_{\mathbb{C}} W C are
S 2 W C = V ( 0 , 0 ) ⊕ 2 V ( 2 , 0 ) ⊕ V ( 1 , 1 ) ⊕ V ( 0 , 2 ) ⊕ V ( 4 , 0 ) , (28) S^2W_{\mathbb{C}}=V(0,0)\oplus2V(2,0)\oplus V(1,1)\oplus V(0,2)\oplus V(4,0), \tag*{(28)} S 2 W C = V ( 0 , 0 ) ⊕ 2 V ( 2 , 0 ) ⊕ V ( 1 , 1 ) ⊕ V ( 0 , 2 ) ⊕ V ( 4 , 0 ) , (28) Λ 2 W C = V ( 1 , 0 ) ⊕ V ( 0 , 1 ) ⊕ V ( 1 , 1 ) ⊕ V ( 3 , 0 ) ⊕ V ( 2 , 1 ) . (29) \Lambda^2W_{\mathbb{C}}=V(1,0)\oplus V(0,1)\oplus V(1,1)\oplus V(3,0)\oplus V(2,1). \tag*{(29)} Λ 2 W C = V ( 1 , 0 ) ⊕ V ( 0 , 1 ) ⊕ V ( 1 , 1 ) ⊕ V ( 3 , 0 ) ⊕ V ( 2 , 1 ) . (29) Here V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) has dimension 64, while the two modules of dimension 77 are V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) in the symmetric square and V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) in the alternating square. In particular, V ( 1 , 0 ) V(1,0) V ( 1 , 0 ) and V ( 0 , 1 ) V(0,1) V ( 0 , 1 ) occur only in the alternating square, each with multiplicity one. Moreover,
V ( 1 , 0 ) ⊗ V ( 0 , 1 ) = V ( 1 , 1 ) ⊕ V ( 2 , 0 ) ⊕ V ( 1 , 0 ) . (30) V(1,0)\otimes V(0,1)=V(1,1)\oplus V(2,0)\oplus V(1,0). \tag*{(30)} V ( 1 , 0 ) ⊗ V ( 0 , 1 ) = V ( 1 , 1 ) ⊕ V ( 2 , 0 ) ⊕ V ( 1 , 0 ) . (30)
The W W W -adjoint of X ∈ End ( W C ) X \in\operatorname{End}(W_{\mathbb{C}}) X ∈ End ( W C ) is its adjoint X T W X^{T_W} X T W with respect to the W W W -pairing:
⟨ X h , k ⟩ W = ⟨ h , X T W k ⟩ W for all h , k ∈ W C . (31) \langle Xh,k\rangle_W=\langle h,X^{T_W}k\rangle_W \quad\text{for all } h,k\in W_{\mathbb{C}}. \tag*{(31)} ⟨ X h , k ⟩ W = ⟨ h , X T W k ⟩ W for all h , k ∈ W C . (31)
We call X X X W W W -self-adjoint if X T W = X X^{T_W}=X X T W = X and W W W -skew-adjoint if X T W = − X X^{T_W}=-X X T W = − X . In a W W W -orthonormal basis, X ↦ X T W X\mapsto X^{T_W} X ↦ X T W is matrix transposition, with no complex conjugation. The W W W -pairing on W C W_{\mathbb{C}} W C identifies an endomorphism X ∈ End ( W C ) X\in\operatorname{End}(W_{\mathbb{C}}) X ∈ End ( W C ) with the bilinear form ( h , k ) ↦ ⟨ X h , k ⟩ W (h,k)\mapsto\langle Xh,k\rangle_W ( h , k ) ↦ ⟨ X h , k ⟩ W . Taking its W W W -self-adjoint and W W W -skew-adjoint parts gives the equivariant identification
End ( W C ) ≅ S 2 W C ⊕ Λ 2 W C . (32) \operatorname{End}(W_{\mathbb{C}})\cong S^2W_{\mathbb{C}}\oplus\Lambda^2W_{\mathbb{C}}. \tag*{(32)} End ( W C ) ≅ S 2 W C ⊕ Λ 2 W C . (32)
The two copies of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) are distinguished by their W W W -adjoint parity: the copy in S 2 W C S^2W_{\mathbb{C}} S 2 W C consists of W W W -self-adjoint operators, while the copy in Λ 2 W C \Lambda^2W_{\mathbb{C}} Λ 2 W C consists of W W W -skew-adjoint operators. Table 1 records the irreducible modules and their multiplicities in the W W W -self-adjoint and W W W -skew-adjoint components.
The invariant cubic plane
We show that the trace cubic and the φ \varphi φ -cubic span the invariant cubics on W W W , and introduce a product in terms of which we write the φ \varphi φ -cubic.
Definition 2.6. The trace cubic and the φ \varphi φ -cubic on W W W are, respectively,
C 0 ( h 1 , h 2 , h 3 ) = tr ( h 1 h 2 h 3 ) , C_0(h_1,h_2,h_3)=\operatorname{tr}(h_1h_2h_3), C 0 ( h 1 , h 2 , h 3 ) = tr ( h 1 h 2 h 3 ) , C φ ( h 1 , h 2 , h 3 ) = φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 . (33) C_\varphi(h_1,h_2,h_3)=\varphi_{a_1a_2a_3}\varphi_{b_1b_2b_3}h^{a_1b_1}_1h^{a_2b_2}_2h^{a_3b_3}_3. \tag*{(33)} C φ ( h 1 , h 2 , h 3 ) = φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 . (33)
We identify each symmetric trilinear form with its associated homogeneous cubic polynomial, writing C 0 ( h ) = C 0 ( h , h , h ) = tr ( h 3 ) C_0(h)=C_0(h,h,h)=\operatorname{tr}(h^3) C 0 ( h ) = C 0 ( h , h , h ) = tr ( h 3 ) and C φ ( h ) = C φ ( h , h , h ) C_\varphi(h)=C_\varphi(h,h,h) C φ ( h ) = C φ ( h , h , h ) .
The contraction in C φ C_\varphi C φ is over all ordered index values, with no prefactor. Both cubics are symmetric and G 2 G_2 G 2 -invariant. By cyclicity of the trace, C 0 C_0 C 0 takes the same value on three of the six orderings of its arguments, and since the arguments are symmetric, transposition shows that it takes this value on the other three as well. Permuting the slots of C φ C_{\varphi} C φ applies the same sign twice to the alternating tensors. Trace invariance and direct contraction give invariance under ρ ( O ) \rho(O) ρ ( O ) (25) for O ∈ G 2 O \in G_2 O ∈ G 2 . The trace cubic involves only g g g and is invariant under the action of O ( 7 ) O(7) O ( 7 ) : tr ( ( ρ ( O ) h ) 3 ) = tr ( h 3 ) \operatorname{tr}((\rho(O)h)^3)=\operatorname{tr}(h^3) tr (( ρ ( O ) h ) 3 ) = tr ( h 3 ) .
We identify a homogeneous cubic N ( h ) = N ( h , h , h ) N(h)=N(h,h,h) N ( h ) = N ( h , h , h ) on W W W or W C W_{\mathbb{C}} W C with its symmetric trilinear polarization by
N ( h 1 , h 2 , h 3 ) = 1 6 ( N ( h 1 + h 2 + h 3 ) − N ( h 1 + h 2 ) − N ( h 1 + h 3 ) − N ( h 2 + h 3 ) + N ( h 1 ) + N ( h 2 ) + N ( h 3 ) ) . (34) N(h_1,h_2,h_3)=\frac{1}{6}\bigl(N(h_1+h_2+h_3)-N(h_1+h_2)-N(h_1+h_3)-N(h_2+h_3)+N(h_1)+N(h_2)+N(h_3)\bigr). \tag*{(34)} N ( h 1 , h 2 , h 3 ) = 6 1 ( N ( h 1 + h 2 + h 3 ) − N ( h 1 + h 2 ) − N ( h 1 + h 3 ) − N ( h 2 + h 3 ) + N ( h 1 ) + N ( h 2 ) + N ( h 3 ) ) . (34)
so D 3 N ( 0 ) [ h 1 , h 2 , h 3 ] = 6 N ( h 1 , h 2 , h 3 ) D^3N(0)[h_1,h_2,h_3]=6N(h_1,h_2,h_3) D 3 N ( 0 ) [ h 1 , h 2 , h 3 ] = 6 N ( h 1 , h 2 , h 3 ) .
Using the representation counts given above, we show in Lemma 2.7 below that these two cubics exhaust the invariant cubic forms.
Lemma 2.7. The invariant cubic spaces are
( S 3 W ∗ ) G 2 = R C 0 ⊕ R C φ , and ( S 3 W C ∗ ) G 2 C = C C 0 ⊕ C C φ . (35) (S^3W^*)_{G_2}=\mathbb{R}C_0\oplus\mathbb{R}C_{\varphi},\qquad\text{and}\qquad(S^3W_{\mathbb{C}}^*)_{G_2^{\mathbb{C}}}=\mathbb{C}C_0\oplus\mathbb{C}C_{\varphi}. \tag*{(35)} ( S 3 W ∗ ) G 2 = R C 0 ⊕ R C φ , and ( S 3 W C ∗ ) G 2 C = C C 0 ⊕ C C φ . (35) Proof. We identify W C W_{\mathbb{C}} W C with its dual via the complex-bilinear W W W -pairing. As shown in Table 1 above, W C ≅ V ( 2 , 0 ) W_{\mathbb{C}}\cong V(2,0) W C ≅ V ( 2 , 0 ) occurs with multiplicity two in S 2 W C S^2W_{\mathbb{C}} S 2 W C and does not occur in Λ 2 W C \Lambda^2W_{\mathbb{C}} Λ 2 W C . Consequently, we can read off the dimensions of G 2 C G_2^{\mathbb{C}} G 2 C -invariant subspaces:
dim ( ( W C ∗ ⊗ 3 ) G 2 C ) = 2 , ( W C ∗ ⊗ Λ 2 W C ∗ ) G 2 C = 0. \dim((W_{\mathbb{C}}^*\otimes3)_{G_2^{\mathbb{C}}})=2,\qquad(W_{\mathbb{C}}^*\otimes\Lambda^2W_{\mathbb{C}}^*)_{G_2^{\mathbb{C}}}=0. dim (( W C ∗ ⊗ 3 ) G 2 C ) = 2 , ( W C ∗ ⊗ Λ 2 W C ∗ ) G 2 C = 0. The second equality implies that every invariant trilinear form is symmetric in any pair of arguments. Hence
dim ( ( S 3 W C ∗ ) G 2 C ) = 2. \dim((S^3W_{\mathbb{C}}^*)_{G_2^{\mathbb{C}}})=2. dim (( S 3 W C ∗ ) G 2 C ) = 2. To see that C 0 C_0 C 0 and C φ C_{\varphi} C φ are independent, recall that C 0 C_0 C 0 is S O ( 7 ) SO(7) S O ( 7 ) -invariant. For any orthonormal triple u , v , w u,v,w u , v , w , applying the G 2 G_2 G 2 -contraction identities (10) to the tensors (17) gives
C 0 ( E u , E v , E w ) = 14 , C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 − 63. C_0(E_u,E_v,E_w)=14,\qquad C_{\varphi}(E_u,E_v,E_w)=343\varphi(u,v,w)^2-63. C 0 ( E u , E v , E w ) = 14 , C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 − 63. Since φ ( u , v , w ) = g ( u × v , w ) \varphi(u,v,w)=g(u\times v,w) φ ( u , v , w ) = g ( u × v , w ) , with u × v u\times v u × v the cross product (8) , the second expression varies with the orthonormal triple, whereas the first does not. Thus C φ C_{\varphi} C φ is not S O ( 7 ) SO(7) S O ( 7 ) -invariant and is not proportional to C 0 C_0 C 0 .
The two cubics therefore span the complex invariant space. Since both are real, the corresponding real invariant space is their real span.
Table 1. Irreducible G 2 C G_2^{\mathbb{C}} G 2 C -modules in End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) . The S 2 W S^2W S 2 W and Λ 2 W \Lambda^2W Λ 2 W columns give their multiplicities in the W W W -self-adjoint and W W W -skew-adjoint components, respectively, after complexification.
With the generators and normalization of this section, we write the cubic pencil as
{ N a , b = a C 0 + b C φ : [ a : b ] ∈ P 1 ( C ) } , with N λ = C 0 + λ C φ , where λ = b / a . (36) \{N_{a,b}=aC_0+bC_{\varphi}:[a:b]\in\mathbb{P}^1(\mathbb{C})\},
\qquad\text{with}\qquad N_{\lambda}=C_0+\lambda C_{\varphi},\quad\text{where }\lambda=b/a. \tag*{(36)} { N a , b = a C 0 + b C φ : [ a : b ] ∈ P 1 ( C )} , with N λ = C 0 + λ C φ , where λ = b / a . (36)
In § 4 , we determine which members of the pencil have infinitesimal stabilizers larger than ρ ( g 2 ) \rho(\mathfrak{g}_2) ρ ( g 2 ) .
Remark 2.8 (Change of generators). An invertible change of generators of the cubic plane induces a projective linear change of coordinates on P 1 \mathbb{P}^{1} P 1 . Thus the members with larger stabilizer, classified in Theorem 3.3 , and their stabilizer dimensions are independent of the generators, whereas their parameter values depend on the chosen normalization. Real changes preserve real lines and conjugate pairs.
Under the standard identification V ≅ Im O V \cong\operatorname{Im}\mathbb{O} V ≅ Im O , Desmet’s model S 2 Im O S^{2}\operatorname{Im}\mathbb{O} S 2 Im O identifies with W W W . The cubic plane corresponds to his two-dimensional space of G 2 G_{2} G 2 -equivariant commutative products [14 Proposition 3.8] ; see also [5 Section 9] . We return to the product interpretation in Section 3. Interestingly, in positive characteristic, Craven gives a two-dimensional space of invariant symmetric trilinear forms for G 2 ( 11 ) G_{2}(11) G 2 ( 11 ) on its 27-dimensional module [10 §5.2.4, p. 683] .
The invariant W W W -pairing identifies C 0 C_{0} C 0 and C φ C_{\varphi} C φ with two G 2 G_{2} G 2 -equivariant symmetric bilinear products A A A and B B B . For symmetric tensors h , k ∈ S 2 V ∗ h,k\in S^{2}V^{*} h , k ∈ S 2 V ∗ , define
A ( h , k ) = 1 2 ( h k + k h ) 0 , (37) A(h,k)=\frac{1}{2}(hk+kh)_{0}, \tag*{(37)} A ( h , k ) = 2 1 ( hk + k h ) 0 , (37)
and
P ( h , k ) i j = φ i a b φ j c d h a c k b d , B ( h , k ) = P ( h , k ) + 1 7 ( tr ( h k ) − tr ( h ) tr ( k ) ) g . (38) \mathcal{P}(h,k)_{ij}=\varphi_{iab}\varphi_{jcd}h^{ac}k^{bd},\qquad B(h,k)=\mathcal{P}(h,k)+\frac{1}{7}\bigl(\operatorname{tr}(hk)-\operatorname{tr}(h)\operatorname{tr}(k)\bigr)g. \tag*{(38)} P ( h , k ) ij = φ iab φ j c d h a c k b d , B ( h , k ) = P ( h , k ) + 7 1 ( tr ( hk ) − tr ( h ) tr ( k ) ) g . (38)
Here the subscript 0 denotes the traceless part (15) , so B ( h , k ) = P ( h , k ) 0 B(h,k)=\mathcal{P}(h,k)_{0} B ( h , k ) = P ( h , k ) 0 by (42) below. We will use P \mathcal{P} P only where the trace part is needed. For h , k , l ∈ W h,k,l\in W h , k , l ∈ W , these products satisfy
⟨ A ( h , k ) , l ⟩ W = C 0 ( h , k , l ) , ⟨ B ( h , k ) , l ⟩ W = C φ ( h , k , l ) . (39) \langle A(h,k),l\rangle_{W}=C_{0}(h,k,l),\qquad\langle B(h,k),l\rangle_{W}=C_{\varphi}(h,k,l). \tag*{(39)} ⟨ A ( h , k ) , l ⟩ W = C 0 ( h , k , l ) , ⟨ B ( h , k ) , l ⟩ W = C φ ( h , k , l ) . (39)
Their quadratic maps A ( h ) = A ( h , h ) A(h)=A(h,h) A ( h ) = A ( h , h ) and B ( h ) = B ( h , h ) B(h)=B(h,h) B ( h ) = B ( h , h ) will be used to construct the quadratic adjoints of the cubic pencil in Section 3. As needed, we may also restrict the above products to traceless symmetric tensors, in which case, we get A , B : S 2 W → W A,B:S^{2}W\to W A , B : S 2 W → W .
Our product P \mathcal{P} P coincides with ( φ h h φ ) (\varphi hh\varphi) ( φ hh φ ) of Grigorian–Yau [31 (5.39)] and the operation A ⊚ B A\mathbin{\circledcirc}B A ⊚ B of Dwivedi–Gianniotis–Karigiannis [15 (2.46)] . Its contraction identities follow from the standard φ φ \varphi\varphi φφ identity [3 ] , [15 Prop. 2.48] .
The product P \mathcal{P} P satisfies the following contraction identities.
Lemma 2.9. For v ∈ V v\in V v ∈ V let P v = v ♭ ⊗ v ♭ P_{v}=v^{\flat}\otimes v^{\flat} P v = v ♭ ⊗ v ♭ be the rank-one tensor of (16) , let v × v ′ v\times v' v × v ′ be the cross product (8) , and let h , h 1 , h 2 , h 3 h,h_{1},h_{2},h_{3} h , h 1 , h 2 , h 3 be symmetric 2-tensors, not necessarily traceless. The contraction P ( h 1 , h 2 ) i j = φ i a 1 a 2 φ j b 1 b 2 h 1 a 1 b 1 h 2 a 2 b 2 \mathcal{P}(h_{1},h_{2})_{ij}=\varphi_{ia_{1}a_{2}}\varphi_{jb_{1}b_{2}}h_{1}^{a_{1}b_{1}}h_{2}^{a_{2}b_{2}} P ( h 1 , h 2 ) ij = φ i a 1 a 2 φ j b 1 b 2 h 1 a 1 b 1 h 2 a 2 b 2 of (38) satisfies
P ( P v , P v ′ ) = P v × v ′ , P ( P v , P v ) = 0 , (40) \mathcal{P}(P_{v},P_{v'})=P_{v\times v'},\qquad\mathcal{P}(P_{v},P_{v})=0, \tag*{(40)} P ( P v , P v ′ ) = P v × v ′ , P ( P v , P v ) = 0 , (40) P ( g , h ) = ( tr h ) g − h , (41) \mathcal{P}(g,h)=(\operatorname{tr}h)g-h, \tag*{(41)} P ( g , h ) = ( tr h ) g − h , (41) tr P ( h 1 , h 2 ) = ( tr h 1 ) ( tr h 2 ) − tr ( h 1 h 2 ) . (42) \operatorname{tr}\mathcal{P}(h_{1},h_{2})=(\operatorname{tr}h_{1})(\operatorname{tr}h_{2})-\operatorname{tr}(h_{1}h_{2}). \tag*{(42)} tr P ( h 1 , h 2 ) = ( tr h 1 ) ( tr h 2 ) − tr ( h 1 h 2 ) . (42) The product is symmetric in its arguments, and tr ( h 3 P ( h 1 , h 2 ) ) \operatorname{tr}(h_{3}\mathcal{P}(h_{1},h_{2})) tr ( h 3 P ( h 1 , h 2 )) is symmetric in all three symmetric tensors.
Proof. For rank-one arguments the two alternating factors are ( v × v ′ ) i (v\times v')_{i} ( v × v ′ ) i and ( v × v ′ ) j (v\times v')_{j} ( v × v ′ ) j . Permuting the arguments in the trilinear contraction permutes both copies of φ \varphi φ , so their signs cancel. Contracting (10) with h h h gives
P ( g , h ) i j = ( δ b d δ i j − δ b j δ i d + ψ b i d j ) h b d = ( tr h ) δ i j − h i j , \mathcal{P}(g,h)_{ij}=(\delta_{bd}\delta_{ij}-\delta_{bj}\delta_{id}+\psi_{bidj})h^{bd}=(\operatorname{tr}h)\delta_{ij}-h_{ij}, P ( g , h ) ij = ( δ b d δ ij − δ bj δ i d + ψ bi d j ) h b d = ( tr h ) δ ij − h ij ,
since the ψ \psi ψ term vanishes. Finally, trilinear symmetry gives tr P ( h 1 , h 2 ) = tr ( h 1 P ( h 2 , g ) ) \operatorname{tr}\mathcal{P}(h_1,h_2)=\operatorname{tr}(h_1\mathcal{P}(h_2,g)) tr P ( h 1 , h 2 ) = tr ( h 1 P ( h 2 , g )) , proving the trace identity.
Using P \mathcal{P} P , the cubic can then be written
C φ ( h 1 , h 2 , h 3 ) = tr ( P ( h 1 , h 2 ) h 3 ) = 1 2 h 3 i j ( χ h 1 ) i a b ( χ h 2 ) j a b , (43) C_{\varphi}(h_1,h_2,h_3)=\operatorname{tr}(\mathcal{P}(h_1,h_2)h_3)=\frac{1}{2}h_3^{ij}(\chi_{h_1})_{iab}(\chi_{h_2})_j^{ab}, \tag*{(43)} C φ ( h 1 , h 2 , h 3 ) = tr ( P ( h 1 , h 2 ) h 3 ) = 2 1 h 3 ij ( χ h 1 ) iab ( χ h 2 ) j ab , (43)
where χ h 1 = I φ ( h 1 ) \chi_{h_1}=\mathcal{I}_{\varphi}(h_1) χ h 1 = I φ ( h 1 ) and χ h 2 = I φ ( h 2 ) \chi_{h_2}=\mathcal{I}_{\varphi}(h_2) χ h 2 = I φ ( h 2 ) , as in (18) . These expressions thus allow the cubic to be written in terms of 3-forms. The second equality in (43) follows from the quadratic insertion identity of Lemma 2.10.
Lemma 2.10. For any h ∈ S 2 V ∗ h\in S^2V^* h ∈ S 2 V ∗ , not necessarily traceless,
( χ h ) i a b ( χ h ) j a b = 2 P ( h , h ) i j + 4 ( tr h ) h i j + 2 tr ( h 2 ) g i j . (44) (\chi_h)_{iab}(\chi_h)_j^{ab}=2\mathcal{P}(h,h)_{ij}+4(\operatorname{tr}h)h_{ij}+2\operatorname{tr}(h^2)g_{ij}. \tag*{(44)} ( χ h ) iab ( χ h ) j ab = 2 P ( h , h ) ij + 4 ( tr h ) h ij + 2 tr ( h 2 ) g ij . (44) Proof. To prove (44) , diagonalize h h h in an orthonormal frame, with eigenvalues d i d_i d i , allowing φ \varphi φ to have general components in this frame. The components are
( χ h ) i a b = ( d i + d a + d b ) φ i a b (no sum over i , a , b ) . (\chi_h)_{iab}=(d_i+d_a+d_b)\varphi_{iab}\qquad\text{(no sum over }i,a,b\text{)}. ( χ h ) iab = ( d i + d a + d b ) φ iab (no sum over i , a , b ) . We expand the contracted product by grouping the terms according to their eigenvalue factors. The terms involving only d i d_i d i , d j d_j d j contribute 6 d i 2 δ i j 6d_i^2\delta_{ij} 6 d i 2 δ ij , those linear in d a + d b d_a+d_b d a + d b contribute 4 d i ( tr h − d i ) δ i j 4d_i(\operatorname{tr}h-d_i)\delta_{ij} 4 d i ( tr h − d i ) δ ij , and those involving d a 2 + d b 2 d_a^2+d_b^2 d a 2 + d b 2 contribute 2 ( tr ( h 2 ) − d i 2 ) δ i j 2(\operatorname{tr}(h^2)-d_i^2)\delta_{ij} 2 ( tr ( h 2 ) − d i 2 ) δ ij (no sum over i i i ). The remaining 2 d a d b 2d_ad_b 2 d a d b contribution is 2 P ( h , h ) i j 2\mathcal{P}(h,h)_{ij} 2 P ( h , h ) ij . The d i 2 d_i^2 d i 2 terms cancel.
For h , k ∈ W h,k\in W h , k ∈ W , polarization of (44) yields
1 2 ( ( χ h ) i a b ( χ k ) j a b + ( χ k ) i a b ( χ h ) j a b ) = 2 P ( h , k ) i j + 2 tr ( h k ) g i j . (45) \frac{1}{2}\left((\chi_h)_{iab}(\chi_k)_j^{ab}+(\chi_k)_{iab}(\chi_h)_j^{ab}\right)=2\mathcal{P}(h,k)_{ij}+2\operatorname{tr}(hk)g_{ij}. \tag*{(45)} 2 1 ( ( χ h ) iab ( χ k ) j ab + ( χ k ) iab ( χ h ) j ab ) = 2 P ( h , k ) ij + 2 tr ( hk ) g ij . (45)
Contracting (45) with a traceless symmetric h 3 i j h_3^{ij} h 3 ij gives the second equality of (43) .
Written in terms of 3-forms, the cubic C φ C_{\varphi} C φ is, up to a constant, the pointwise integrand of the Yukawa coupling of Lee and Leung. In our notation, their parallel 3-form and its Hodge dual are φ \varphi φ and ψ \psi ψ . The operation in [46 Definition 1.1] sends a 3-form α \alpha α to the vector-valued one-form
α ^ = g a b ∗ ( α ∧ ι e a ψ ) ⊗ e b = α ^ i b e i ⊗ e b , \widehat{\alpha}=g^{ab}*(\alpha\wedge\iota_{e_a}\psi)\otimes e_b=\widehat{\alpha}_i^b e^i\otimes e_b, α = g ab ∗ ( α ∧ ι e a ψ ) ⊗ e b = α i b e i ⊗ e b ,
and the Yukawa integrand is φ ( α ^ 1 , α ^ 2 , α ^ 3 ) ∧ ψ \varphi(\widehat{\alpha}_1,\widehat{\alpha}_2,\widehat{\alpha}_3)\wedge\psi φ ( α 1 , α 2 , α 3 ) ∧ ψ . We read the evaluation of φ \varphi φ on vector-valued one-forms as evaluation on the vector components and the wedge product of the one-form components: φ ( θ 1 ⊗ v 1 , θ 2 ⊗ v 2 , θ 3 ⊗ v 3 ) = φ ( v 1 , v 2 , v 3 ) θ 1 ∧ θ 2 ∧ θ 3 \varphi(\theta_1\otimes v_1,\theta_2\otimes v_2,\theta_3\otimes v_3)=\varphi(v_1,v_2,v_3)\theta_1\wedge\theta_2\wedge\theta_3 φ ( θ 1 ⊗ v 1 , θ 2 ⊗ v 2 , θ 3 ⊗ v 3 ) = φ ( v 1 , v 2 , v 3 ) θ 1 ∧ θ 2 ∧ θ 3 . Since ψ = ∗ φ \psi=* \varphi ψ = ∗ φ , the Hodge convention (5) gives η ∧ ψ = ⟨ η , φ ⟩ vol \eta\wedge\psi=\langle\eta,\varphi\rangle\operatorname{vol} η ∧ ψ = ⟨ η , φ ⟩ vol for every 3-form η \eta η . We wedge the Leibniz rule α ∧ ι v ψ = ι v α ∧ ψ − ι v ( α ∧ ψ ) \alpha\wedge\iota_v\psi=\iota_v\alpha\wedge\psi-\iota_v(\alpha\wedge\psi) α ∧ ι v ψ = ι v α ∧ ψ − ι v ( α ∧ ψ ) with a one-form and apply this identity to both resulting terms; we also apply it to η = e i ∧ e j ∧ e k \eta=e^i\wedge e^j\wedge e^k η = e i ∧ e j ∧ e k . In components, this gives
α ^ i a = 1 2 α j k a φ i j k − ⟨ α , φ ⟩ δ i a , \widehat{\alpha}_i^a=\frac{1}{2}\alpha^a_{jk}\varphi_i^{jk}-\langle\alpha,\varphi\rangle\delta_i^a, α i a = 2 1 α j k a φ i j k − ⟨ α , φ ⟩ δ i a ,
φ ( α ^ 1 , α ^ 2 , α ^ 3 ) ∧ ψ = φ a b c φ i j k ( α ^ 1 ) i a ( α ^ 2 ) j b ( α ^ 3 ) k c vol . (46) \varphi(\widehat{\alpha}_1,\widehat{\alpha}_2,\widehat{\alpha}_3)\wedge\psi=\varphi_{abc}\varphi^{ijk}(\widehat{\alpha}_1)_i^a(\widehat{\alpha}_2)_j^b(\widehat{\alpha}_3)_k^c\operatorname{vol}. \tag*{(46)} φ ( α 1 , α 2 , α 3 ) ∧ ψ = φ ab c φ ij k ( α 1 ) i a ( α 2 ) j b ( α 3 ) k c vol . (46)
For α = χ h \alpha=\chi_h α = χ h with h ∈ W h\in W h ∈ W , we have ⟨ χ h , φ ⟩ = 0 \langle\chi_h,\varphi\rangle=0 ⟨ χ h , φ ⟩ = 0 by (19) , and the second identity in (22) gives ( χ h ^ ) i a = 2 h i a (\widehat{\chi_h})^a_i=2h^a_i ( χ h ) i a = 2 h i a . Therefore
φ ( χ ^ h 1 , χ ^ h 2 , χ ^ h 3 ) ∧ ψ = 1 8 φ a 1 a 2 a 3 φ b 1 b 2 b 3 ( χ h 1 ) p q a 1 φ b 1 p q ( χ h 2 ) r s a 2 φ b 2 r s ( χ h 3 ) m n a 3 φ b 3 m n vol = 8 φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 vol = 8 C φ ( h 1 , h 2 , h 3 ) vol . (47) \begin{aligned}
\varphi(\widehat{\chi}_{h_1},\widehat{\chi}_{h_2},\widehat{\chi}_{h_3})\wedge\psi=\frac{1}{8}\varphi_{a_1a_2a_3}\varphi^{b_1b_2b_3}(\chi_{h_1})^{a_1}_{pq}\varphi^{b_1pq}(\chi_{h_2})^{a_2}_{rs}\varphi^{b_2rs}(\chi_{h_3})^{a_3}_{mn}\varphi^{b_3mn}\operatorname{vol} \\
&=8\varphi_{a_1a_2a_3}\varphi_{b_1b_2b_3}h^{a_1b_1}_1h^{a_2b_2}_2h^{a_3b_3}_3\operatorname{vol}=8C_\varphi(h_1,h_2,h_3)\operatorname{vol}. \tag*{(47)}
\end{aligned} φ ( χ h 1 , χ h 2 , χ h 3 ) ∧ ψ = 8 1 φ a 1 a 2 a 3 φ b 1 b 2 b 3 ( χ h 1 ) pq a 1 φ b 1 pq ( χ h 2 ) r s a 2 φ b 2 r s ( χ h 3 ) mn a 3 φ b 3 mn vol = 8 φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 vol = 8 C φ ( h 1 , h 2 , h 3 ) vol . (47)
The pointwise Yukawa integrand, pulled back to W W W by I φ \mathcal{I}_\varphi I φ , is therefore 8 C φ 8C_\varphi 8 C φ under this evaluation convention. We also call C φ C_\varphi C φ itself the pointwise Yukawa cubic ; it is the φ \varphi φ -cubic of Definition 2.6 . A different evaluation convention changes the scalar factor but not the line C C φ \mathbb{C}C_\varphi C C φ , the endpoint [ 0 : 1 ] [0:1] [ 0 : 1 ] of the pencil (36) . We return to this comparison in Proposition 4.16 .
We use these product and insertion identities to relate the cubics to volume.
Volume and the two cubics
The two generators of the invariant cubic plane arise, up to constant factors, as third derivatives of volume along two different affine paths. Fix a positive 3-form φ \varphi φ on V V V , with metric g = g φ g=g_\varphi g = g φ , put ψ = ∗ φ φ \psi=*_\varphi\varphi ψ = ∗ φ φ , and write
ν = vol g = 1 7 φ ∧ ψ . (48) \nu=\operatorname{vol}_g=\frac{1}{7}\varphi\wedge\psi. \tag*{(48)} ν = vol g = 7 1 φ ∧ ψ . (48)
The factor 1 7 \frac{1}{7} 7 1 comes from ∣ φ ∣ g 2 = 7 |\varphi|_g^2=7 ∣ φ ∣ g 2 = 7 . The form ν \nu ν is the pointwise density in Hitchin’s volume functional [35 §7.3] . For h ∈ W h\in W h ∈ W , consider the paths
q h ( t ) = g + 2 t h , φ h ( t ) = φ + t χ h , χ h = I φ ( h ) , q_h(t)=g+2th,\qquad\varphi_h(t)=\varphi+t\chi_h,\qquad\chi_h=\mathcal{I}_\varphi(h), q h ( t ) = g + 2 t h , φ h ( t ) = φ + t χ h , χ h = I φ ( h ) ,
g h ( t ) = g φ h ( t ) , ψ h ( t ) = ∗ g h ( t ) φ h ( t ) , vol g h ( t ) = 1 7 φ h ( t ) ∧ ψ h ( t ) . (49) g_h(t)=g_{\varphi_h(t)},\qquad\psi_h(t)=*_{g_h(t)}\varphi_h(t),\qquad\operatorname{vol}_{g_h(t)}=\frac{1}{7}\varphi_h(t)\wedge\psi_h(t). \tag*{(49)} g h ( t ) = g φ h ( t ) , ψ h ( t ) = ∗ g h ( t ) φ h ( t ) , vol g h ( t ) = 7 1 φ h ( t ) ∧ ψ h ( t ) . (49)
For sufficiently small t t t , q h ( t ) q_h(t) q h ( t ) is a metric and φ h ( t ) \varphi_h(t) φ h ( t ) is positive. The paths q h ( t ) q_h(t) q h ( t ) and g h ( t ) g_h(t) g h ( t ) have the same initial metric velocity 2 h 2h 2 h , but g h ( t ) g_h(t) g h ( t ) is generally not affine. All insertions and index identifications in these paths are taken at the basepoint ( φ , g ) (\varphi,g) ( φ , g ) ; in particular, χ h \chi_h χ h is held fixed. We first compute the second derivatives and then the third derivatives.
We use the pointwise deformation formulas of Grigorian–Yau [31 Section 5] . Their averaged insertion map satisfies i φ = I φ / 3 i_\varphi=\mathcal{I}_\varphi/3 i φ = I φ /3 , so the symmetric tensor that they associate with our direction χ h \chi_h χ h is 3 h 3h 3 h . With this substitution, the second-order terms of their (5.19), (5.35), (5.43), and (5.48) give
g h ( t ) = g + 2 t h + t 2 ( 2 h 2 − 1 2 P ( h , h ) − 1 6 tr ( h 2 ) g ) + O ( t 3 ) , (50) g_h(t)=g+2th+t^2\left(2h^2-\frac{1}{2}\mathcal{P}(h,h)-\frac{1}{6}\operatorname{tr}(h^2)g\right)+O(t^3), \tag*{(50)} g h ( t ) = g + 2 t h + t 2 ( 2 h 2 − 2 1 P ( h , h ) − 6 1 tr ( h 2 ) g ) + O ( t 3 ) , (50)
ψ h ( t ) = ψ − t ∗ χ h + t 2 ∗ ( 1 2 I φ ( B ( h , h ) ) − 1 21 tr ( h 2 ) φ ) + O ( t 3 ) . (51) \psi_h(t)=\psi-t*\chi_h+t^2*\left(\frac{1}{2}\mathcal{I}_\varphi(B(h,h))-\frac{1}{21}\operatorname{tr}(h^2)\varphi\right)+O(t^3). \tag*{(51)} ψ h ( t ) = ψ − t ∗ χ h + t 2 ∗ ( 2 1 I φ ( B ( h , h )) − 21 1 tr ( h 2 ) φ ) + O ( t 3 ) . (51)
Here and below, a Hodge star without a subscript is the star of the basepoint, and h 2 h^2 h 2 is the contracted square (12) . The expansions are Taylor expansions in t t t , so each second derivative is twice the coefficient of t 2 t^2 t 2 . Polarizing gives the following.
Proposition 2.11. For h 1 , h 2 , h 3 ∈ W h_1,h_2,h_3 \in W h 1 , h 2 , h 3 ∈ W , with the ambient form directions fixed,
D φ 2 g [ χ h 1 , χ h 2 ] = 2 ( h 1 h 2 + h 2 h 1 ) − P ( h 1 , h 2 ) − 1 3 tr ( h 1 h 2 ) g , (52) D_{\varphi}^{2}g[\chi_{h_1},\chi_{h_2}] = 2(h_1h_2+h_2h_1)-\mathcal{P}(h_1,h_2)-\frac{1}{3}\operatorname{tr}(h_1h_2)g, \tag*{(52)} D φ 2 g [ χ h 1 , χ h 2 ] = 2 ( h 1 h 2 + h 2 h 1 ) − P ( h 1 , h 2 ) − 3 1 tr ( h 1 h 2 ) g , (52) ∗ D φ 2 ( ∗ φ φ ) [ χ h 1 , χ h 2 ] = I φ ( B ( h 1 , h 2 ) ) − 2 21 tr ( h 1 h 2 ) φ , (53) *D_{\varphi}^{2}(*_{\varphi}\varphi)[\chi_{h_1},\chi_{h_2}] = \mathcal{I}_{\varphi}(B(h_1,h_2))-\frac{2}{21}\operatorname{tr}(h_1h_2)\varphi, \tag*{(53)} ∗ D φ 2 ( ∗ φ φ ) [ χ h 1 , χ h 2 ] = I φ ( B ( h 1 , h 2 )) − 21 2 tr ( h 1 h 2 ) φ , (53) D φ ⟨ χ h 1 , χ h 2 ⟩ g φ [ χ h 3 ] = − 2 C φ ( h 1 , h 2 , h 3 ) . (54) D_{\varphi}\langle\chi_{h_1},\chi_{h_2}\rangle_{g_{\varphi}}[\chi_{h_3}] = -2C_{\varphi}(h_1,h_2,h_3). \tag*{(54)} D φ ⟨ χ h 1 , χ h 2 ⟩ g φ [ χ h 3 ] = − 2 C φ ( h 1 , h 2 , h 3 ) . (54)
Proof. The first two identities are the polarized expansions (50) and (51) . For the last identity, D φ g [ χ h 3 ] = 2 h 3 D_{\varphi}g[\chi_{h_3}]=2h_3 D φ g [ χ h 3 ] = 2 h 3 , so differentiating the three inverse metrics in the inner product of the fixed forms gives
D φ ⟨ χ h 1 , χ h 2 ⟩ g φ [ χ h 3 ] = − h 3 i j ( χ h 1 ) i a b ( χ h 2 ) j a b = − 2 C φ ( h 1 , h 2 , h 3 ) , D_{\varphi}\langle\chi_{h_1},\chi_{h_2}\rangle_{g_{\varphi}}[\chi_{h_3}]
=-h_3^{ij}(\chi_{h_1})_{iab}(\chi_{h_2})_{j}^{\ ab}
=-2C_{\varphi}(h_1,h_2,h_3), D φ ⟨ χ h 1 , χ h 2 ⟩ g φ [ χ h 3 ] = − h 3 ij ( χ h 1 ) iab ( χ h 2 ) j ab = − 2 C φ ( h 1 , h 2 , h 3 ) ,
by (43) .
We now differentiate the volume itself. We regard vol \operatorname{vol} vol both as the function q ↦ vol q q\mapsto\operatorname{vol}_q q ↦ vol q on metrics and as the function φ ↦ vol g φ \varphi\mapsto\operatorname{vol}_{g_{\varphi}} φ ↦ vol g φ on positive forms; the subscript on D D D indicates which is differentiated, and ν \nu ν always denotes the value at the basepoint. For a metric velocity 2 h 2h 2 h , the determinant formula gives
log vol g h ( t ) ν = 1 2 tr log ( I + 2 t g − 1 h ) = − t 2 tr ( h 2 ) + 4 3 t 3 tr ( h 3 ) + O ( t 4 ) . (55) \log\frac{\operatorname{vol}_{g_h(t)}}{\nu}
=\frac{1}{2}\operatorname{tr}\log(I+2t\,g^{-1}h)
=-t^2\operatorname{tr}(h^2)+\frac{4}{3}t^3\operatorname{tr}(h^3)+O(t^4). \tag*{(55)} log ν vol g h ( t ) = 2 1 tr log ( I + 2 t g − 1 h ) = − t 2 tr ( h 2 ) + 3 4 t 3 tr ( h 3 ) + O ( t 4 ) . (55)
Thus the first derivative vanishes, and the Hessian of the metric volume is
D g 2 vol [ 2 h , 2 k ] ν = − 2 tr ( h k ) . (56) \frac{D_g^2\operatorname{vol}[2h,2k]}{\nu}=-2\operatorname{tr}(hk). \tag*{(56)} ν D g 2 vol [ 2 h , 2 k ] = − 2 tr ( hk ) . (56)
For positive forms, we call φ ↦ ∗ φ φ \varphi\mapsto*_{\varphi}\varphi φ ↦ ∗ φ φ the duality map. The first-variation identities, valid at every positive form, are as follows; for the derivative of the duality map, see Hitchin [35 Lemma 20] .
D φ vol [ ξ ] = 1 3 ξ ∧ ψ , (57) D_{\varphi}\operatorname{vol}[\xi]=\frac{1}{3}\xi\wedge\psi, \tag*{(57)} D φ vol [ ξ ] = 3 1 ξ ∧ ψ , (57)
D φ ( ∗ φ φ ) [ ξ ] = ∗ ( 4 3 pr 1 ξ + pr 7 ξ − pr 27 ξ ) . (58) D_{\varphi}(*_{\varphi}\varphi)[\xi]
=*\left(\frac{4}{3}\operatorname{pr}_1\xi+\operatorname{pr}_7\xi-\operatorname{pr}_{27}\xi\right). \tag*{(58)} D φ ( ∗ φ φ ) [ ξ ] = ∗ ( 3 4 pr 1 ξ + pr 7 ξ − pr 27 ξ ) . (58)
The projections are those of the form at which the derivative is taken. To see the factor 1 / 3 1/3 1/3 , differentiate vol g φ = φ ∧ ( ∗ φ φ ) / 7 \operatorname{vol}_{g_{\varphi}}=\varphi\wedge(*_{\varphi}\varphi)/7 vol g φ = φ ∧ ( ∗ φ φ ) /7 : the second identity gives φ ∧ D φ ( ∗ φ φ ) [ ξ ] = 4 3 ξ ∧ ψ \varphi\wedge D_{\varphi}(*_{\varphi}\varphi)[\xi]=\frac{4}{3}\xi\wedge\psi φ ∧ D φ ( ∗ φ φ ) [ ξ ] = 3 4 ξ ∧ ψ . Along the affine path in (49) , these identities yield
d d t vol g h ( t ) = 1 3 χ h ∧ ψ h ( t ) , d 2 d t 2 ∣ t = 0 vol g h ( t ) = 1 3 χ h ∧ ψ ˙ h ( 0 ) = − 1 3 ∣ χ h ∣ 2 ν . \frac{d}{dt}\operatorname{vol}_{g_h(t)}=\frac{1}{3}\chi_h\wedge\psi_h(t),\qquad
\left.\frac{d^2}{dt^2}\right|_{t=0}\operatorname{vol}_{g_h(t)}
=\frac{1}{3}\chi_h\wedge\dot{\psi}_h(0)
=-\frac{1}{3}|\chi_h|^2\nu. d t d vol g h ( t ) = 3 1 χ h ∧ ψ h ( t ) , d t 2 d 2 t = 0 vol g h ( t ) = 3 1 χ h ∧ ψ ˙ h ( 0 ) = − 3 1 ∣ χ h ∣ 2 ν .
Since ∣ χ h ∣ 2 = 2 tr ( h 2 ) |\chi_h|^2=2\operatorname{tr}(h^2) ∣ χ h ∣ 2 = 2 tr ( h 2 ) , polarization gives
D φ 2 vol [ χ h , χ k ] ν = − 1 3 ⟨ χ h , χ k ⟩ = − 2 3 tr ( h k ) . (59) \frac{D_{\varphi}^2\operatorname{vol}[\chi_h,\chi_k]}{\nu}
=-\frac{1}{3}\langle\chi_h,\chi_k\rangle
=-\frac{2}{3}\operatorname{tr}(hk). \tag*{(59)} ν D φ 2 vol [ χ h , χ k ] = − 3 1 ⟨ χ h , χ k ⟩ = − 3 2 tr ( hk ) . (59)
In particular, the two paths with the same initial metric velocity have different volume Hessians.
We can now take the third derivatives. For mixed derivatives the paths are replaced by the affine families
q ( s , t , r ) = g + 2 ( s h 1 + t h 2 + r h 3 ) , φ ( s , t , r ) = φ + s χ h 1 + t χ h 2 + r χ h 3 . q(s,t,r)=g+2(sh_1+th_2+rh_3),\qquad\varphi(s,t,r)=\varphi+s\chi h_1+t\chi h_2+r\chi h_3. q ( s , t , r ) = g + 2 ( s h 1 + t h 2 + r h 3 ) , φ ( s , t , r ) = φ + sχ h 1 + t χ h 2 + r χ h 3 .
The direction forms are fixed in Λ 3 V ∗ \Lambda^3V^* Λ 3 V ∗ , rather than transported into the moving Λ 27 3 \Lambda^3_{27} Λ 27 3 subspaces.
Proposition 2.12. With flat affine derivatives on the cones of metrics and positive 3-forms, for h 1 , h 2 , h 3 ∈ W h_1,h_2,h_3\in W h 1 , h 2 , h 3 ∈ W ,
D g 3 vol [ 2 h 1 , 2 h 2 , 2 h 3 ] ν = 8 C 0 ( h 1 , h 2 , h 3 ) , D φ 3 vol [ χ h 1 , χ h 2 , χ h 3 ] ν = 2 3 C φ ( h 1 , h 2 , h 3 ) . (60) \frac{D_g^3\operatorname{vol}[2h_1,2h_2,2h_3]}{\nu}=8C_0(h_1,h_2,h_3),\qquad\frac{D_\varphi^3\operatorname{vol}[\chi h_1,\chi h_2,\chi h_3]}{\nu}=\frac{2}{3}C_\varphi(h_1,h_2,h_3). \tag*{(60)} ν D g 3 vol [ 2 h 1 , 2 h 2 , 2 h 3 ] = 8 C 0 ( h 1 , h 2 , h 3 ) , ν D φ 3 vol [ χ h 1 , χ h 2 , χ h 3 ] = 3 2 C φ ( h 1 , h 2 , h 3 ) . (60)
Proof. For the metric path, the vanishing first derivative means that the third derivatives of the volume ratio and its logarithm agree at zero. The determinant expansion (55) gives 8 tr ( h 3 ) 8\operatorname{tr}(h^3) 8 tr ( h 3 ) , and polarization gives the first identity.
For the form path, differentiate (57) twice in fixed ambient directions and use (53) :
D φ 3 vol [ χ h 1 , χ h 2 , χ h 3 ] = 1 3 ⟨ χ h 3 ∧ D φ 2 ( ∗ φ φ ) [ χ h 1 , χ h 2 ] ⟩ = 1 3 ⟨ χ h 3 , I φ ( B ( h 1 , h 2 ) ) − 2 21 tr ( h 1 h 2 ) φ ⟩ ν = 2 3 tr ( h 3 B ( h 1 , h 2 ) ) ν = 2 3 C φ ( h 1 , h 2 , h 3 ) ν . \begin{aligned}
D_\varphi^3\operatorname{vol}[\chi h_1,\chi h_2,\chi h_3]=\frac{1}{3}\langle\chi h_3\wedge D_\varphi^2(*\varphi\varphi)[\chi h_1,\chi h_2]\rangle\\
&=\frac{1}{3}\left\langle\chi h_3,\mathcal{I}_\varphi(B(h_1,h_2))-\frac{2}{21}\operatorname{tr}(h_1h_2)\varphi\right\rangle\nu\\
&=\frac{2}{3}\operatorname{tr}(h_3B(h_1,h_2))\nu=\frac{2}{3}C_\varphi(h_1,h_2,h_3)\nu.
\end{aligned} D φ 3 vol [ χ h 1 , χ h 2 , χ h 3 ] = 3 1 ⟨ χ h 3 ∧ D φ 2 ( ∗ φφ ) [ χ h 1 , χ h 2 ]⟩ = 3 1 ⟨ χ h 3 , I φ ( B ( h 1 , h 2 )) − 21 2 tr ( h 1 h 2 ) φ ⟩ ν = 3 2 tr ( h 3 B ( h 1 , h 2 )) ν = 3 2 C φ ( h 1 , h 2 , h 3 ) ν .
Here χ h 3 ⊥ φ \chi h_3\perp\varphi χ h 3 ⊥ φ , the insertion map multiplies the W W W -pairing by 2, and h 3 h_3 h 3 is traceless. Equivalently, the one-parameter volume expansions are
vol q h ( t ) ν = 1 − t 2 tr ( h 2 ) + 4 3 t 3 C 0 ( h ) + O ( t 4 ) , \frac{\operatorname{vol}_{q_h(t)}}{\nu}=1-t^2\operatorname{tr}(h^2)+\frac{4}{3}t^3C_0(h)+O(t^4), ν vol q h ( t ) = 1 − t 2 tr ( h 2 ) + 3 4 t 3 C 0 ( h ) + O ( t 4 ) ,
vol φ h ( t ) ν = 1 − 1 3 t 2 tr ( h 2 ) + 1 9 t 3 C φ ( h ) + O ( t 4 ) . \frac{\operatorname{vol}_{\varphi_h(t)}}{\nu}=1-\frac{1}{3}t^2\operatorname{tr}(h^2)+\frac{1}{9}t^3C_\varphi(h)+O(t^4). ν vol φ h ( t ) = 1 − 3 1 t 2 tr ( h 2 ) + 9 1 t 3 C φ ( h ) + O ( t 4 ) .
Thus the second-order term of ψ h ( t ) \psi_h(t) ψ h ( t ) in Grigorian–Yau’s (5.48) already determines the third derivative of the volume density of positive forms.
Thus C φ C_\varphi C φ measures both the third affine derivative of the positive-form volume and the change of the inner product on fixed ambient 3-forms in (54) . For the metric volume, the corresponding cubic is C 0 C_0 C 0 .
An arbitrary nonlinear metric curve with initial velocity 2 h 2h 2 h need not have these volume derivatives. For instance, along a curve in the fixed-volume hypersurface of metrics the volume is constant, so all its derivatives vanish.
The second derivative identities are the polarized second-order terms of Grigorian–Yau’s expansions [31 Section 5] . The contraction formula (43) appears in Karigiannis–Leung [40 Lemma 3.12] , and the third derivative identity appears in integrated form in [40 Theorem 3.10] . A cubic C s C_s C s on W W W , with s ∈ { 0 , φ } s\in\{0,\varphi\} s ∈ { 0 , φ } , is transported to Λ 27 3 V ∗ \Lambda^3_{27}V^* Λ 27 3 V ∗ through I φ − 1 \mathcal{I}_\varphi^{-1} I φ − 1 . Replacing I φ \mathcal{I}_{\varphi} I φ by a nonzero multiple, with the cubics on W W W fixed, multiplies both transported cubics by the inverse cube of the multiple. For the isometry I φ / 2 \mathcal{I}_{\varphi}/\sqrt{2} I φ / 2 , this factor is 2 2 2\sqrt{2} 2 2 . Since both cubics are scaled by the same factor, the pencil parameter is unchanged.
In Section 3, we use the invariant cubic plane, its pencil, the product B B B and the decomposition of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) into irreducible representations to locate the parameters at which the stabilizer jumps.
The exceptional cubic pencil
The question and the theorem
In Section 2 we identified the G 2 G_{2} G 2 -invariant cubics on W C W_{\mathbb{C}} W C as the pencil N a , b = a C 0 + b C φ N_{a,b}=aC_{0}+bC_{\varphi} N a , b = a C 0 + b C φ . Every member therefore has infinitesimal symmetries ρ ( g 2 ) \rho(\mathfrak{g}_{2}) ρ ( g 2 ) . We determine the parameters [ a : b ] ∈ P 1 ( C ) [a:b]\in\mathbb{P}^{1}(\mathbb{C}) [ a : b ] ∈ P 1 ( C ) for which the stabilizer is larger and identify the additional symmetries.
We use the splitting (2.22) (with Λ 14 2 = g 2 \Lambda^{2}_{14}=\mathfrak{g}_{2} Λ 14 2 = g 2 ), the G 2 G_{2} G 2 -contraction identities of Lemma 2.1, and the contraction rules of Lemma 2.9 from Section 2. The contraction data are stated where they enter the proof; their detailed derivations are collected in Appendix A.
We work over C \mathbb{C} C unless a statement explicitly allows F ∈ { R , C } \mathbb{F}\in\{\mathbb{R},\mathbb{C}\} F ∈ { R , C } ; then g 2 F \mathfrak{g}_{2}^{\mathbb{F}} g 2 F denotes g 2 \mathfrak{g}_{2} g 2 for F = R \mathbb{F}=\mathbb{R} F = R and g 2 C \mathfrak{g}_{2}^{\mathbb{C}} g 2 C for F = C \mathbb{F}=\mathbb{C} F = C . As in Section 2, we identify a cubic polynomial with its symmetric trilinear polarization, normalized by N ( h ) = N ( h , h , h ) N(h)=N(h,h,h) N ( h ) = N ( h , h , h ) as in (2.28). All pairings and adjoints on W C W_{\mathbb{C}} W C use the complex-bilinear extension of the W W W -pairing (2.11); in particular X T W X^{T_{W}} X T W is the W W W -adjoint (2.25). In the stabilizer formulas, X X X (and in Section 3.6, L L L ) denotes an endomorphism of W F W_{\mathbb{F}} W F .
Definition 3.1. Let N N N be a cubic on W F W_{\mathbb{F}} W F . Its exact infinitesimal stabilizer is
stab ( N ) = { X ∈ End ( W F ) : N ( X h , h , h ) = 0 for all h ∈ W F } . \operatorname{stab}(N)=\left\{X\in\operatorname{End}(W_{\mathbb{F}}):N(Xh,h,h)=0\text{ for all }h\in W_{\mathbb{F}}\right\}. stab ( N ) = { X ∈ End ( W F ) : N ( X h , h , h ) = 0 for all h ∈ W F } . Its action map is Φ N : End ( W F ) → S 3 W F ∗ \Phi_{N}:\operatorname{End}(W_{\mathbb{F}})\to S^{3}W_{\mathbb{F}}^{*} Φ N : End ( W F ) → S 3 W F ∗ ,
Φ N ( X ) ( h 1 , h 2 , h 3 ) = N ( X h 1 , h 2 , h 3 ) + N ( h 1 , X h 2 , h 3 ) + N ( h 1 , h 2 , X h 3 ) . (61) \Phi_{N}(X)(h_{1},h_{2},h_{3})=N(Xh_{1},h_{2},h_{3})+N(h_{1},Xh_{2},h_{3})+N(h_{1},h_{2},Xh_{3}). \tag*{(61)} Φ N ( X ) ( h 1 , h 2 , h 3 ) = N ( X h 1 , h 2 , h 3 ) + N ( h 1 , X h 2 , h 3 ) + N ( h 1 , h 2 , X h 3 ) . (61) Since N N N is symmetric, Φ N ( X ) \Phi_{N}(X) Φ N ( X ) is a symmetric trilinear form, and Φ N ( X ) ( h , h , h ) = 3 N ( X h , h , h ) \Phi_{N}(X)(h,h,h)=3N(Xh,h,h) Φ N ( X ) ( h , h , h ) = 3 N ( X h , h , h ) . A symmetric trilinear form vanishes when its cubic does, so
stab ( N ) = ker Φ N . \operatorname{stab}(N)=\ker\Phi_{N}. stab ( N ) = ker Φ N .
For the pencil members we write Φ a , b = Φ N a , b = a Φ C 0 + b Φ C φ \Phi_{a,b}=\Phi_{N_{a,b}}=a\Phi_{C_{0}}+b\Phi_{C_{\varphi}} Φ a , b = Φ N a , b = a Φ C 0 + b Φ C φ . The exact stabilizer preserves the cubic itself, whereas preserving its one-dimensional span F N \mathbb{F}N F N requires only Φ N ( X ) ∈ F N \Phi_{N}(X)\in\mathbb{F}N Φ N ( X ) ∈ F N . We return to stabilizers of cubic lines in Sections 3.6 and 4.
The G 2 G_{2} G 2 -invariance of the generators in (2.27) has two immediate consequences for the action map.
Lemma 3.2. For every [ a : b ] ∈ P 1 ( F ) [a:b]\in\mathbb{P}^{1}(\mathbb{F}) [ a : b ] ∈ P 1 ( F ) , with F ∈ { R , C } \mathbb{F}\in\{\mathbb{R},\mathbb{C}\} F ∈ { R , C } ,
ρ ( g 2 F ) ⊆ ker Φ a , b = stab ( N a , b ) . \rho(\mathfrak{g}_{2}^{\mathbb{F}})\subseteq\ker\Phi_{a,b}=\operatorname{stab}(N_{a,b}). ρ ( g 2 F ) ⊆ ker Φ a , b = stab ( N a , b ) . Moreover, Φ a , b \Phi_{a,b} Φ a , b is g 2 F \mathfrak{g}_{2}^{\mathbb{F}} g 2 F -equivariant:
Φ a , b ( [ ρ ( M ) , X ] ) = M ⋅ Φ a , b ( X ) . \Phi_{a,b}\bigl([\rho(M),X]\bigr)=M\cdot\Phi_{a,b}(X). Φ a , b ( [ ρ ( M ) , X ] ) = M ⋅ Φ a , b ( X ) .
Here M ∈ g 2 F M \in\mathfrak{g}_2^{\mathrm{F}} M ∈ g 2 F acts on endomorphisms by commutators with ρ ( M ) \rho(M) ρ ( M ) and on cubic tensors by the induced action. Consequently, every stabilizer is a g 2 F \mathfrak{g}_2^{\mathrm{F}} g 2 F -submodule of End ( W F ) \operatorname{End}(W_{\mathrm{F}}) End ( W F ) .
Proof. Differentiating the invariance of N a , b N_{a,b} N a , b using (26) and (61) gives the containment. By the same invariance, the action map is equivariant, so its kernel is a submodule. Both statements extend complex-linearly.
We show in § 5 that the stabilizers of any two distinct members intersect precisely in ρ ( g 2 F ) \rho(\mathfrak{g}_2^{\mathrm{F}}) ρ ( g 2 F ) .
The main theorem of this section determines the dimension of the stabilizer of every member of the pencil, including the parameters at which the stabilizer is larger than ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) . We use the normalization of C 0 C_0 C 0 and C φ C_\varphi C φ fixed in Definition 2.6; Remark 2.8 describes how the parameters depend on this choice.
Theorem 3.3. For [ a : b ] ∈ P 1 ( C ) [a:b]\in\mathbb{P}^{1}(\mathbb{C}) [ a : b ] ∈ P 1 ( C ) , let N a , b = a C 0 + b C φ N_{a,b}=aC_0+bC_\varphi N a , b = a C 0 + b C φ on W C W_{\mathbb{C}} W C . Then
dim C stab ( N a , b ) = { 21 , [ a : b ] = [ 1 : 0 ] , 78 , a 2 − 2 a b + 8 b 2 = 0 , 14 , otherwise, including [ 0 : 1 ] . (62) \dim_{\mathbb{C}}\operatorname{stab}(N_{a,b}) =
\begin{cases}
21, & [a:b]=[1:0],\\
78, & a^2-2ab+8b^2=0,\\
14, & \text{otherwise, including }[0:1].
\end{cases} \tag*{(62)} dim C stab ( N a , b ) = ⎩ ⎨ ⎧ 21 , 78 , 14 , [ a : b ] = [ 1 : 0 ] , a 2 − 2 ab + 8 b 2 = 0 , otherwise, including [ 0 : 1 ] . (62)
The quadratic a 2 − 2 a b + 8 b 2 a^2-2ab+8b^2 a 2 − 2 ab + 8 b 2 has two distinct non-real zeros in P 1 ( C ) \mathbb{P}^{1}(\mathbb{C}) P 1 ( C ) . In the affine coordinate λ = b / a \lambda=b/a λ = b / a they are the roots of 8 λ 2 − 2 λ + 1 = 0 8\lambda^2-2\lambda+1=0 8 λ 2 − 2 λ + 1 = 0 ,
λ ± = 1 ± i 7 8 . (63) \lambda_{\pm}=\frac{1\pm i\sqrt{7}}{8}. \tag*{(63)} λ ± = 8 1 ± i 7 . (63)
We use the critical cubics N ± N_{\pm} N ± and critical lines C N ± CN_{\pm} C N ± introduced in § 1 , with the parameters λ ± \lambda_{\pm} λ ± of (63) . Since C 0 C_0 C 0 and C φ C_\varphi C φ are real and linearly independent, complex conjugation exchanges N + N_+ N + and N − N_- N − and their stabilizers. Neither critical line contains a nonzero real cubic. Consequently, by (62) , the real pencil has exactly one member with stabilizer dimension greater than 14: the trace cubic C 0 C_0 C 0 , whose stabilizer has dimension 21. The proof of Theorem 3.3 has four stages.
We develop the contraction identities of the two cubic tensors on W W W , the c-contraction identities, including two universal quartic identities. Their symbolic proof in Section A.3 uses the evaluated one-index c-contraction identity, a formal resolvent and a triple-form contraction (Lemma A.1 ).
We characterize the affine members N λ = C 0 + λ C φ N_\lambda=C_0+\lambda C_\varphi N λ = C 0 + λ C φ whose quadratic adjoints satisfy a double-adjoint identity (Proposition 3.7 ). Evaluation on elementary test tensors proves necessity, and the quartic identities prove sufficiency and produce the parameters λ ± \lambda_{\pm} λ ± .
Using the cubic contractions, we exclude five isotypic components from every kernel. We treat the double 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) component with the four-map Gram matrix G \mathcal{G} G , and the multiplicity-one components with four 2 × 2 2\times2 2 × 2 matrices Γ \Gamma Γ . Their contraction proofs are given in Section A.2; the invertibility argument remains in § 4.4 . These exclusions and the Koecher operators constructed from the double adjoints give dim stab ( N ± ) ≥ 78 \dim\operatorname{stab}(N_{\pm})\geq78 dim stab ( N ± ) ≥ 78 (Proposition 3.18 ).
Finally, the trace-cubic stabilizer and the Koecher lower bound determine the two critical kernels inside the double V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) component. A single explicit computation determines their graphs in the parity model. From the resulting relative operator we obtain the exact kernel dimension throughout the projective pencil (Proposition 3.31), including the endpoint C φ C_\varphi C φ , and rule out all other jumps.
An isotypic component is the sum of all irreducible submodules of a fixed type. By equivariance, components of different types do not mix, and by Schur’s lemma, maps between repeated copies reduce to small matrices. This approach has recently been applied to the octonionic structure operator, see [29 , 30 ] . In this paper, the main example is the pair of copies of the 64-dimensional module. We use the decompositions from § 3 , Schur’s lemma and complete reducibility throughout these module arguments. We use the classification of complex simple Lie algebras only in Theorem 3.32 , after the dimensions have been established, to identify the critical stabilizer as e 6 \mathfrak{e}_6 e 6 .
Quadratic adjoints and c-contraction identities
Recall the G 2 G_2 G 2 -equivariant symmetric bilinear products A , B : S 2 W → W A,B:S^2W\to W A , B : S 2 W → W from (37) and (38) . We write A ( h ) = A ( h , h ) A(h)=A(h,h) A ( h ) = A ( h , h ) and B ( h ) = B ( h , h ) B(h)=B(h,h) B ( h ) = B ( h , h ) , and for λ ∈ C \lambda\in\mathbb{C} λ ∈ C set
F λ = A + λ B . F_\lambda=A+\lambda B. F λ = A + λ B .
We use the same symbol F λ F_\lambda F λ for its halved polarization:
F λ ( h 1 , h 2 ) = 1 2 ( F λ ( h 1 + h 2 ) − F λ ( h 1 ) − F λ ( h 2 ) ) = A ( h 1 , h 2 ) + λ B ( h 1 , h 2 ) . (64) \begin{aligned}
F_\lambda(h_1,h_2)&=\frac{1}{2}\left(F_\lambda(h_1+h_2)-F_\lambda(h_1)-F_\lambda(h_2)\right)\\
&=A(h_1,h_2)+\lambda B(h_1,h_2). \tag*{(64)}
\end{aligned} F λ ( h 1 , h 2 ) = 2 1 ( F λ ( h 1 + h 2 ) − F λ ( h 1 ) − F λ ( h 2 ) ) = A ( h 1 , h 2 ) + λ B ( h 1 , h 2 ) . (64)
Thus F λ ( h , h ) = F λ ( h ) F_\lambda(h,h)=F_\lambda(h) F λ ( h , h ) = F λ ( h ) , and (39) gives
⟨ F λ ( h 1 , h 2 ) , h 3 ⟩ W = N λ ( h 1 , h 2 , h 3 ) . (65) \langle F_\lambda(h_1,h_2),h_3\rangle_W=N_\lambda(h_1,h_2,h_3). \tag*{(65)} ⟨ F λ ( h 1 , h 2 ) , h 3 ⟩ W = N λ ( h 1 , h 2 , h 3 ) . (65)
These identities extend complex-bilinearly to W C W_{\mathbb{C}} W C .
Definition 3.4. The map F λ F_\lambda F λ is the quadratic adjoint of N λ N_\lambda N λ relative to the W W W -pairing. It is characterized by
N λ ( h , h , k ) = ⟨ F λ ( h ) , k ⟩ W . (66) N_\lambda(h,h,k)=\langle F_\lambda(h),k\rangle_W. \tag*{(66)} N λ ( h , h , k ) = ⟨ F λ ( h ) , k ⟩ W . (66) In particular, d N λ ∣ h ( k ) = 3 ⟨ F λ ( h ) , k ⟩ W dN_\lambda|_h(k)=3\langle F_\lambda(h),k\rangle_W d N λ ∣ h ( k ) = 3 ⟨ F λ ( h ) , k ⟩ W , so F λ = 1 3 ∇ N λ F_\lambda=\frac{1}{3}\nabla N_\lambda F λ = 3 1 ∇ N λ is one third of the W W W -gradient of N λ N_\lambda N λ , where the W W W -gradient ∇ N \nabla N ∇ N of a cubic N N N is defined by ⟨ ∇ N ( h ) , k ⟩ W = d N ∣ h ( k ) \langle\nabla N(h),k\rangle_W=dN|_h(k) ⟨ ∇ N ( h ) , k ⟩ W = d N ∣ h ( k ) . For the critical parameters we write F ± = F λ ± F_\pm=F_{\lambda_\pm} F ± = F λ ± .
The terminology comes from the adjoint, or sharp, map of a cubic norm structure. However, in this definition we do not assume a Jordan algebra structure.
Fix a real orthonormal basis ( k α ) α = 1 27 (k_\alpha)_{\alpha=1}^{27} ( k α ) α = 1 27 of W W W for the W W W -pairing of Definition 2.3 and use it also as a basis of W C W_{\mathbb{C}} W C , with the pairing extended complex-bilinearly. Greek indices run from 1 to 27 and are raised and lowered with δ α β \delta_{\alpha\beta} δ α β ; Latin indices still run from 1 to 7 on V V V . All repeated indices are contracted with these bilinear metrics, without complex conjugation.
For such a basis we have the projection identity
∑ α = 1 27 ( k α ) i j ( k α ) m n = 1 2 ( δ i m δ j n + δ i n δ j m ) − 1 7 g i j g m n . (67) \sum_{\alpha=1}^{27} (k_\alpha)_{ij}(k_\alpha)^{mn} = \frac{1}{2}\left(\delta_i^m\delta_j^n+\delta_i^n\delta_j^m\right)-\frac{1}{7}g_{ij}g^{mn}. \tag*{(67)} α = 1 ∑ 27 ( k α ) ij ( k α ) mn = 2 1 ( δ i m δ j n + δ i n δ j m ) − 7 1 g ij g mn . (67)
Indeed, adjoining g / 7 g/\sqrt{7} g / 7 gives a basis of S 2 V ∗ S^2V^* S 2 V ∗ that is orthonormal for the W W W -pairing tr ( h 1 h 2 ) \operatorname{tr}(h_1h_2) tr ( h 1 h 2 ) , extended to S 2 V ∗ S^2V^* S 2 V ∗ by the same formula; subtracting the rank-one projection onto g / 7 g/\sqrt{7} g / 7 from the symmetrization gives (67) . Thus (67) represents, on covariant 2-tensors X X X , the projection X ↦ 1 2 ( X + X T ) − 1 7 ( tr X ) g X\mapsto\frac{1}{2}(X+X^\mathsf{T})-\frac{1}{7}(\operatorname{tr}X)g X ↦ 2 1 ( X + X T ) − 7 1 ( tr X ) g onto W W W , and it is independent of the chosen basis.
The superscripts 0 and φ \varphi φ indicate that a quantity is associated with C 0 C_0 C 0 or C φ C_\varphi C φ , respectively. For s , t ∈ { 0 , φ } s,t\in\{0,\varphi\} s , t ∈ { 0 , φ } define
c α β γ s = C s ( k α , k β , k γ ) , T α β γ δ s t = c α β σ s ( c t ) σ γ δ . (68) c^s_{\alpha\beta\gamma}=C_s(k_\alpha,k_\beta,k_\gamma),\qquad T^{st}_{\alpha\beta\gamma\delta}=c^s_{\alpha\beta\sigma}(c^t)^\sigma{}_{\gamma\delta}. \tag*{(68)} c α β γ s = C s ( k α , k β , k γ ) , T α β γ δ s t = c α β σ s ( c t ) σ γ δ . (68)
Each c s c^s c s is fully symmetric. The pairings (39) between the products A A A and B B B and the two cubics give their coordinate description:
A ( h 1 , h 2 ) α = ( c 0 ) α β γ h 1 β h 2 γ , B ( h 1 , h 2 ) α = ( c φ ) α β γ h 1 β h 2 γ , A(h_1,h_2)^\alpha=(c^0)^{\alpha\beta\gamma}h_{1\beta}h_{2\gamma},\qquad B(h_1,h_2)^\alpha=(c^\varphi)^{\alpha\beta\gamma}h_{1\beta}h_{2\gamma}, A ( h 1 , h 2 ) α = ( c 0 ) α β γ h 1 β h 2 γ , B ( h 1 , h 2 ) α = ( c φ ) α β γ h 1 β h 2 γ ,
F λ ( h ) α = ( c 0 + λ c φ ) α β γ h β h γ . F_\lambda(h)^\alpha=(c^0+\lambda c^\varphi)^{\alpha\beta\gamma}h_\beta h_\gamma. F λ ( h ) α = ( c 0 + λ c φ ) α β γ h β h γ .
We call the four-index tensors T s t T^{st} T s t , which contract one index of c s c^s c s with one of c t c^t c t , the one-index c-contractions ; contracting two indices gives the two-index c-contractions below. Formulas that evaluate them are the c-contraction identities . The one-index c-contractions are symmetric within each pair and satisfy
T α β γ δ s t = T γ δ α β t s . T^{st}_{\alpha\beta\gamma\delta}=T^{ts}_{\gamma\delta\alpha\beta}. T α β γ δ s t = T γ δ α β t s .
Thus interchanging the two cubic types also interchanges the two pairs of indices. The two mixed one-index c-contractions T 0 φ T^{0\varphi} T 0 φ and T φ 0 T^{\varphi0} T φ 0 need not agree when all four indices are kept in place.
Contracting two indices produces an invariant bilinear form on W W W , hence a scalar multiple of the W W W -pairing. The resulting 2 × 2 2\times2 2 × 2 matrix a = ( a s t ) \mathbf{a}=(a_{st}) a = ( a s t ) of scalars, indexed by s , t ∈ { 0 , φ } s,t\in\{0,\varphi\} s , t ∈ { 0 , φ } , computes the pairings of the multiplication operators associated with A A A and B B B . Its nonsingularity will distinguish the two copies of W W W in End ( W ) \operatorname{End}(W) End ( W ) . The two-index c-contraction identity is
( c s ) α β γ ( c t ) β γ = a s t δ α δ , a = ( 55 / 28 11 / 14 11 / 14 159 / 7 ) , det a = 44. (69) (c^s)_{\alpha\beta\gamma}(c^t)^{\beta\gamma}=a_{st}\delta_\alpha^\delta,\qquad
\mathbf{a}=
\begin{pmatrix}
55/28 & 11/14\\
11/14 & 159/7
\end{pmatrix},
\qquad\det\mathbf{a}=44. \tag*{(69)} ( c s ) α β γ ( c t ) β γ = a s t δ α δ , a = ( 55/28 11/14 11/14 159/7 ) , det a = 44. (69)
We derive all three entries in Section A.1 from the projection identity (67) and the one-index c-contraction identities. Identity (69) is tensorial and does not depend on the choice of orthonormal basis.
In terms of the products, the one-index c-contractions are
T s t ( h 1 , h 2 ; h 3 , h 4 ) = ⟨ F ( s ) ( h 1 , h 2 ) , F ( t ) ( h 3 , h 4 ) ⟩ W , F ( 0 ) = A , F ( φ ) = B , T^{st}(h_1,h_2;h_3,h_4)=\left\langle F^{(s)}(h_1,h_2),F^{(t)}(h_3,h_4)\right\rangle_W,\qquad F^{(0)}=A,\qquad F^{(\varphi)}=B, T s t ( h 1 , h 2 ; h 3 , h 4 ) = ⟨ F ( s ) ( h 1 , h 2 ) , F ( t ) ( h 3 , h 4 ) ⟩ W , F ( 0 ) = A , F ( φ ) = B ,
where s , t ∈ { 0 , φ } s,t\in\{0,\varphi\} s , t ∈ { 0 , φ } and h 1 , h 2 , h 3 , h 4 ∈ W h_1,h_2,h_3,h_4\in W h 1 , h 2 , h 3 , h 4 ∈ W . Expanding these contractions gives
T 00 ( h 1 , h 2 ; h 3 , h 4 ) = 1 2 tr ( h 1 h 2 h 3 h 4 + h 1 h 2 h 4 h 3 ) − 1 7 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , T 0 φ ( h 1 , h 2 ; h 3 , h 4 ) = C φ ( h 3 , h 4 , 1 2 ( h 1 h 2 + h 2 h 1 ) ) + 1 7 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , T φ 0 ( h 1 , h 2 ; h 3 , h 4 ) = C φ ( h 1 , h 2 , 1 2 ( h 3 h 4 + h 4 h 3 ) ) + 1 7 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W . (70) \begin{aligned}
T^{00}(h_1,h_2;h_3,h_4)&=\frac{1}{2}\operatorname{tr}(h_1h_2h_3h_4+h_1h_2h_4h_3)-\frac{1}{7}\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W,\\
T^{0\varphi}(h_1,h_2;h_3,h_4)&=C_\varphi\left(h_3,h_4,\frac{1}{2}(h_1h_2+h_2h_1)\right)+\frac{1}{7}\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W,\\
T^{\varphi0}(h_1,h_2;h_3,h_4)&=C_\varphi\left(h_1,h_2,\frac{1}{2}(h_3h_4+h_4h_3)\right)+\frac{1}{7}\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W. \tag*{(70)}
\end{aligned} T 00 ( h 1 , h 2 ; h 3 , h 4 ) T 0 φ ( h 1 , h 2 ; h 3 , h 4 ) T φ 0 ( h 1 , h 2 ; h 3 , h 4 ) = 2 1 tr ( h 1 h 2 h 3 h 4 + h 1 h 2 h 4 h 3 ) − 7 1 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , = C φ ( h 3 , h 4 , 2 1 ( h 1 h 2 + h 2 h 1 ) ) + 7 1 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , = C φ ( h 1 , h 2 , 2 1 ( h 3 h 4 + h 4 h 3 ) ) + 7 1 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W . (70)
Here the contraction defining C φ C_\varphi C φ is extended to all symmetric matrices when its third argument is not traceless.
To evaluate the fourth contraction, define the vector obtained by contracting the commutator [ h 1 , h 2 ] [h_1,h_2] [ h 1 , h 2 ] with φ \varphi φ ,
η ( h 1 , h 2 ) i = φ i a b ( h 1 h 2 ) a b = 1 2 φ i a b [ h 1 , h 2 ] a b , \eta(h_1,h_2)_i=\varphi_{iab}(h_1h_2)^{ab}=\frac{1}{2}\varphi_{iab}[h_1,h_2]^{ab}, η ( h 1 , h 2 ) i = φ iab ( h 1 h 2 ) ab = 2 1 φ iab [ h 1 , h 2 ] ab ,
so that η ( h 2 , h 1 ) = − η ( h 1 , h 2 ) \eta(h_2,h_1)=-\eta(h_1,h_2) η ( h 2 , h 1 ) = − η ( h 1 , h 2 ) , and let Ξ \Xi Ξ be the sum of the six mixed cubic terms,
Ξ ( h 1 , h 2 , h 3 , h 4 ) = C φ ( h 1 , h 2 , A ( h 3 , h 4 ) ) + C φ ( h 1 , h 3 , A ( h 2 , h 4 ) ) + C φ ( h 1 , h 4 , A ( h 2 , h 3 ) ) + C φ ( h 2 , h 3 , A ( h 1 , h 4 ) ) + C φ ( h 2 , h 4 , A ( h 1 , h 3 ) ) + C φ ( h 3 , h 4 , A ( h 1 , h 2 ) ) . \begin{aligned}
\Xi(h_1,h_2,h_3,h_4)={}&C_\varphi(h_1,h_2,A(h_3,h_4))+C_\varphi(h_1,h_3,A(h_2,h_4))+C_\varphi(h_1,h_4,A(h_2,h_3))\\
&+C_\varphi(h_2,h_3,A(h_1,h_4))+C_\varphi(h_2,h_4,A(h_1,h_3))+C_\varphi(h_3,h_4,A(h_1,h_2)).
\end{aligned} Ξ ( h 1 , h 2 , h 3 , h 4 ) = C φ ( h 1 , h 2 , A ( h 3 , h 4 )) + C φ ( h 1 , h 3 , A ( h 2 , h 4 )) + C φ ( h 1 , h 4 , A ( h 2 , h 3 )) + C φ ( h 2 , h 3 , A ( h 1 , h 4 )) + C φ ( h 2 , h 4 , A ( h 1 , h 3 )) + C φ ( h 3 , h 4 , A ( h 1 , h 2 )) .
Then
T φ φ ( h 1 , h 2 ; h 3 , h 4 ) = 2 7 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W + 10 7 ⟨ h 1 , h 3 ⟩ W ⟨ h 2 , h 4 ⟩ W + 10 7 ⟨ h 1 , h 4 ⟩ W ⟨ h 2 , h 3 ⟩ W − 8 tr ( h 1 h 3 h 2 h 4 ) + 2 Ξ ( h 1 , h 2 , h 3 , h 4 ) − 2 ( η ( h 1 , h 3 ) ⋅ η ( h 2 , h 4 ) + η ( h 1 , h 4 ) ⋅ η ( h 2 , h 3 ) ) . (71) \begin{aligned}
T^{\varphi\varphi}(h_1,h_2;h_3,h_4)={}&\frac{2}{7}\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W+\frac{10}{7}\langle h_1,h_3\rangle_W\langle h_2,h_4\rangle_W+\frac{10}{7}\langle h_1,h_4\rangle_W\langle h_2,h_3\rangle_W\\
&-8\operatorname{tr}(h_1h_3h_2h_4)+2\Xi(h_1,h_2,h_3,h_4)\\
&-2\bigl(\eta(h_1,h_3)\cdot\eta(h_2,h_4)+\eta(h_1,h_4)\cdot\eta(h_2,h_3)\bigr). \tag*{(71)}
\end{aligned} T φφ ( h 1 , h 2 ; h 3 , h 4 ) = 7 2 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W + 7 10 ⟨ h 1 , h 3 ⟩ W ⟨ h 2 , h 4 ⟩ W + 7 10 ⟨ h 1 , h 4 ⟩ W ⟨ h 2 , h 3 ⟩ W − 8 tr ( h 1 h 3 h 2 h 4 ) + 2Ξ ( h 1 , h 2 , h 3 , h 4 ) − 2 ( η ( h 1 , h 3 ) ⋅ η ( h 2 , h 4 ) + η ( h 1 , h 4 ) ⋅ η ( h 2 , h 3 ) ) . (71)
The dot denotes the complex-bilinear metric pairing on V C V_{\mathbb C} V C . We prove these identities in Section A.1 by contracting the cubic products using the projection identity (67) and evaluating the resulting 4-form contraction.
We obtain a scalar quartic identity by setting h 1 = h 2 = h 3 = h 4 = h h_1=h_2=h_3=h_4=h h 1 = h 2 = h 3 = h 4 = h in (71) . On the left-hand side,
T φ φ ( h , h ; h , h ) = ⟨ B ( h ) , B ( h ) ⟩ W . T^{\varphi\varphi}(h,h;h,h)=\langle B(h),B(h)\rangle_W. T φφ ( h , h ; h , h ) = ⟨ B ( h ) , B ( h ) ⟩ W .
Since η ( h , h ) = 0 \eta(h,h)=0 η ( h , h ) = 0 and
Ξ ( h , h , h , h ) = 6 ⟨ A ( h ) , B ( h ) ⟩ W , \Xi(h,h,h,h)=6\langle A(h),B(h)\rangle_W, Ξ ( h , h , h , h ) = 6 ⟨ A ( h ) , B ( h ) ⟩ W ,
the contraction identity gives
⟨ B ( h ) , B ( h ) ⟩ W = 22 7 ⟨ h , h ⟩ W 2 − 8 tr ( h 4 ) + 12 ⟨ A ( h ) , B ( h ) ⟩ W . \langle B(h),B(h)\rangle_W=\frac{22}{7}\langle h,h\rangle_W^2-8\operatorname{tr}(h^4)+12\langle A(h),B(h)\rangle_W. ⟨ B ( h ) , B ( h ) ⟩ W = 7 22 ⟨ h , h ⟩ W 2 − 8 tr ( h 4 ) + 12 ⟨ A ( h ) , B ( h ) ⟩ W .
Using
tr ( h 4 ) = ⟨ A ( h ) , A ( h ) ⟩ W + 1 7 ⟨ h , h ⟩ W 2 , \operatorname{tr}(h^4)=\langle A(h),A(h)\rangle_W+\frac{1}{7}\langle h,h\rangle_W^2, tr ( h 4 ) = ⟨ A ( h ) , A ( h ) ⟩ W + 7 1 ⟨ h , h ⟩ W 2 ,
we therefore obtain
8 ⟨ A ( h ) , A ( h ) ⟩ W − 12 ⟨ A ( h ) , B ( h ) ⟩ W + ⟨ B ( h ) , B ( h ) ⟩ W = 2 ⟨ h , h ⟩ W 2 . (72) 8\langle A(h),A(h)\rangle_W-12\langle A(h),B(h)\rangle_W+\langle B(h),B(h)\rangle_W=2\langle h,h\rangle_W^2. \tag*{(72)} 8 ⟨ A ( h ) , A ( h ) ⟩ W − 12 ⟨ A ( h ) , B ( h ) ⟩ W + ⟨ B ( h ) , B ( h ) ⟩ W = 2 ⟨ h , h ⟩ W 2 . (72)
Equivalently, polarization gives
Sym α β γ δ ( 8 T α β γ δ 00 − 12 T α β γ δ 0 φ + T α β γ δ φ φ ) = 2 Sym α β γ δ ( δ α β δ γ δ ) . \operatorname{Sym}_{\alpha\beta\gamma\delta}\left(8T^{00}_{\alpha\beta\gamma\delta}-12T^{0\varphi}_{\alpha\beta\gamma\delta}+T^{\varphi\varphi}_{\alpha\beta\gamma\delta}\right)=2\operatorname{Sym}_{\alpha\beta\gamma\delta}\left(\delta_{\alpha\beta}\delta_{\gamma\delta}\right). Sym α β γ δ ( 8 T α β γ δ 00 − 12 T α β γ δ 0 φ + T α β γ δ φφ ) = 2 Sym α β γ δ ( δ α β δ γ δ ) .
Here symmetrization is normalized to average over all permutations of the displayed indices.
Although this scalar identity yields further relations by polarization and differentiation, it does not by itself establish the W W W -valued quartic identities needed below. Their proofs also use the triple-form reduction of Lemma A.1 in Section A.1 and the formal resolvent of Section A.3.
The double-adjoint argument will use the two general identities of Lemma 3.5 below. Their proof is given in Section A.3. Composing two quadratic adjoints produces six quartic expressions formed from A A A and B B B . The following two identities relate those expressions to C 0 ( h ) h C_0(h)h C 0 ( h ) h and C φ ( h ) h C_\varphi(h)h C φ ( h ) h .
Lemma 3.5. For every h ∈ W C h \in W_{\mathbb{C}} h ∈ W C ,
8 A ( A ( h ) ) − A ( B ( h ) ) + 2 B ( A ( h ) ) + 2 B ( A ( h ) , B ( h ) ) = 16 3 C 0 ( h ) h (73) 8A(A(h))-A(B(h))+2B(A(h))+2B(A(h),B(h))=\frac{16}{3}C_0(h)h \tag*{(73)} 8 A ( A ( h )) − A ( B ( h )) + 2 B ( A ( h )) + 2 B ( A ( h ) , B ( h )) = 3 16 C 0 ( h ) h (73) 16 A ( A ( h ) , B ( h ) ) + 2 A ( B ( h ) ) − 8 B ( A ( h ) ) + B ( B ( h ) ) = 16 3 C φ ( h ) h (74) 16A(A(h),B(h))+2A(B(h))-8B(A(h))+B(B(h))=\frac{16}{3}C_\varphi(h)h \tag*{(74)} 16 A ( A ( h ) , B ( h )) + 2 A ( B ( h )) − 8 B ( A ( h )) + B ( B ( h )) = 3 16 C φ ( h ) h (74)
These two identities supply the sufficiency argument in Proposition 3.7: at the critical parameters, the composition of the quadratic adjoints is a linear combination of their left-hand sides.
Double-adjoint characterization of the critical cubics
For a cubic norm structure, the classical adjoint identity ( x ♯ ) ♯ = N ( x ) x (x^\sharp)^\sharp=N(x)x ( x ♯ ) ♯ = N ( x ) x says that applying the adjoint twice returns the algebra element x x x multiplied by its norm. Cubic norm structures and their adjoints are recalled in § 5.2 , see (120) . This identity motivates the following question. Here we allow two different quadratic adjoints F λ F_\lambda F λ and F μ F_\mu F μ (Definition 3.4) and ask when
F μ ( F λ ( h ) ) = κ N λ ( h ) h for every h ∈ W C . (75) F_\mu(F_\lambda(h))=\kappa N_\lambda(h)h \qquad\text{for every } h\in W_{\mathbb{C}}. \tag*{(75)} F μ ( F λ ( h )) = κ N λ ( h ) h for every h ∈ W C . (75)
The parameters λ \lambda λ , μ \mu μ , κ ∈ C \kappa\in\mathbb{C} κ ∈ C are initially unrestricted.
Definition 3.6. A real 3-plane Σ ⊂ V \Sigma\subset V Σ ⊂ V is associative if ∣ φ ( a , b , c ) ∣ = 1 |\varphi(a,b,c)|=1 ∣ φ ( a , b , c ) ∣ = 1 for an orthonormal basis ( a , b , c ) (a,b,c) ( a , b , c ) of Σ \Sigma Σ . The calibration φ \varphi φ then specifies its orientation.
Proposition 3.7. Let λ \lambda λ , μ \mu μ , κ ∈ C \kappa\in\mathbb{C} κ ∈ C . Then, (3.15) holds if and only if ( λ , μ , κ ) = ( λ ± , λ ∓ , 2 3 ) (\lambda,\mu,\kappa)=(\lambda_\pm,\lambda_\mp,\frac{2}{3}) ( λ , μ , κ ) = ( λ ± , λ ∓ , 3 2 ) .
In particular, identically on W C W_{\mathbb{C}} W C ,
F − ( F + ( h ) ) = 2 3 N + ( h ) h , and F + ( F − ( h ) ) = 2 3 N − ( h ) h . F_-(F_+(h))=\frac{2}{3}N_+(h)h,\qquad\text{and}\qquad F_+(F_-(h))=\frac{2}{3}N_-(h)h. F − ( F + ( h )) = 3 2 N + ( h ) h , and F + ( F − ( h )) = 3 2 N − ( h ) h .
Proof. Necessity. Assume (3.15) holds. We will use three test tensors in (3.15) to determine ( λ , μ , κ ) (\lambda,\mu,\kappa) ( λ , μ , κ ) . Recall from (16) that P i = e i ∘ ⊗ e i ∘ ∈ S 2 V ∗ P_i=e_i^\circ\otimes e_i^\circ\in S^2V^* P i = e i ∘ ⊗ e i ∘ ∈ S 2 V ∗ is the projection tensor of R e i \mathbb{R}e_i R e i , so that g − 1 P i = pr R e i g^{-1}P_i=\operatorname{pr}_{\mathbb{R}e_i} g − 1 P i = pr R e i . For the standard G 2 G_2 G 2 3-form, the 3-plane Σ = span ( e 1 , e 2 , e 3 ) \Sigma=\operatorname{span}(e_1,e_2,e_3) Σ = span ( e 1 , e 2 , e 3 ) is associative, with e 1 × e 2 = e 3 e_1\times e_2=e_3 e 1 × e 2 = e 3 and its cyclic variants. Its projection tensor is P Σ = P 1 + P 2 + P 3 P_\Sigma=P_1+P_2+P_3 P Σ = P 1 + P 2 + P 3 . We then use the test tensors (2.14) of the line R e 1 \mathbb{R}e_1 R e 1 and of Σ \Sigma Σ , and also define Δ \Delta Δ :
E 1 = 7 P 1 − g , E Σ = 7 P Σ − 3 g , Δ = P 1 − P 2 . E_1=7P_1-g,\qquad E_\Sigma=7P_\Sigma-3g,\qquad\Delta=P_1-P_2. E 1 = 7 P 1 − g , E Σ = 7 P Σ − 3 g , Δ = P 1 − P 2 .
Clearly, the tensors E 1 E_1 E 1 , E Σ E_\Sigma E Σ , Δ \Delta Δ are traceless. Geometrically, E 1 = 7 ( P 1 ) 0 E_1=7(P_1)_0 E 1 = 7 ( P 1 ) 0 and E Σ = 7 ( P Σ ) 0 E_\Sigma=7(P_\Sigma)_0 E Σ = 7 ( P Σ ) 0 are multiples of the traceless parts of the projection tensors of a line and of the associative 3-plane, respectively, while Δ \Delta Δ has rank 2. We will use the identities (2.34a) and (2.34b):
P ( P u , P v ) = P u × v , P ( g , h ) = ( tr h ) g − h . (76) \mathcal{P}(P_u,P_v)=P_{u\times v},\qquad\mathcal{P}(g,h)=(\operatorname{tr}h)g-h. \tag*{(76)} P ( P u , P v ) = P u × v , P ( g , h ) = ( tr h ) g − h . (76)
Here P u = u ♭ ⊗ u ♭ P_u = u^\flat\otimes u^\flat P u = u ♭ ⊗ u ♭ as in (16) , and h h h is any symmetric 2-tensor. The preceding identities give
P ( P 1 , P 1 ) = 0 , P ( P Σ , P Σ ) = 2 P Σ , P ( g , g ) = 6 g . \mathcal{P}(P_1,P_1)=0,\qquad\mathcal{P}(P_\Sigma,P_\Sigma)=2P_\Sigma,\qquad\mathcal{P}(g,g)=6g. P ( P 1 , P 1 ) = 0 , P ( P Σ , P Σ ) = 2 P Σ , P ( g , g ) = 6 g .
For example, the six ordered pairs of distinct basis vectors in the associative 3-plane give two copies of each P i P_i P i in P ( P Σ , P Σ ) \mathcal{P}(P_\Sigma,P_\Sigma) P ( P Σ , P Σ ) . We substitute h = E 1 h=E_1 h = E 1 and h = E Σ h=E_\Sigma h = E Σ into A ( h ) = ( h 2 ) 0 A(h)=(h^2)_0 A ( h ) = ( h 2 ) 0 and B ( h ) = P ( h , h ) 0 B(h)=\mathcal{P}(h,h)_0 B ( h ) = P ( h , h ) 0 , and then into F λ ( h ) = A ( h ) + λ B ( h ) F_\lambda(h)=A(h)+\lambda B(h) F λ ( h ) = A ( h ) + λ B ( h ) and N λ ( h ) = ⟨ F λ ( h ) , h ⟩ W N_\lambda(h)=\langle F_\lambda(h),h\rangle_W N λ ( h ) = ⟨ F λ ( h ) , h ⟩ W . This gives the following values.
Table 3.16.
We then evaluate (75) at E 1 E_1 E 1 and E Σ E_\Sigma E Σ , respectively. Since F μ F_\mu F μ is quadratic, we obtain the following equations, where we do not yet cancel any parameter-dependent factors:
( 5 + 2 λ ) 2 ( 5 + 2 μ ) = 42 κ ( 5 + 2 λ ) , (77) (5+2\lambda)^2(5+2\mu)=42\kappa(5+2\lambda), \tag*{(77)} ( 5 + 2 λ ) 2 ( 5 + 2 μ ) = 42 κ ( 5 + 2 λ ) , (77)
( 1 + 20 λ ) 2 ( 1 + 20 μ ) = 84 κ ( 1 + 20 λ ) . (78) (1+20\lambda)^2(1+20\mu)=84\kappa(1+20\lambda). \tag*{(78)} ( 1 + 20 λ ) 2 ( 1 + 20 μ ) = 84 κ ( 1 + 20 λ ) . (78)
For Δ = P 1 − P 2 \Delta=P_1-P_2 Δ = P 1 − P 2 , the same identities give
A ( Δ ) = P 1 + P 2 − 2 7 g , B ( Δ ) = − 2 P 3 + 2 7 g . A(\Delta)=P_1+P_2-\frac{2}{7}g,\qquad B(\Delta)=-2P_3+\frac{2}{7}g. A ( Δ ) = P 1 + P 2 − 7 2 g , B ( Δ ) = − 2 P 3 + 7 2 g .
Both are orthogonal to Δ \Delta Δ under the W W W -pairing, so N λ ( Δ ) = 0 N_\lambda(\Delta)=0 N λ ( Δ ) = 0 . Thus (75) forces F μ ( X ) = 0 F_\mu(X)=0 F μ ( X ) = 0 , where
X = F λ ( Δ ) = z g + R − 2 λ P 3 , R = P 1 + P 2 , z = 2 ( λ − 1 ) 7 . X=F_\lambda(\Delta)=zg+R-2\lambda P_3,\qquad R=P_1+P_2,\qquad z=\frac{2(\lambda-1)}{7}. X = F λ ( Δ ) = z g + R − 2 λ P 3 , R = P 1 + P 2 , z = 7 2 ( λ − 1 ) .
Only the nonscalar terms of X 2 + μ P ( X , X ) X^2+\mu\mathcal{P}(X,X) X 2 + μ P ( X , X ) matter, since F μ ( X ) F_\mu(X) F μ ( X ) is its traceless part. Orthogonality of the projection tensors, P i P j = δ i j P i P_iP_j=\delta_{ij}P_i P i P j = δ ij P i for the contracted product (12) , gives
( X 2 ) 0 = ( ( 2 z + 1 ) R + ( 4 λ 2 − 4 λ z ) P 3 ) 0 . (X^2)_0=\bigl((2z+1)R+(4\lambda^2-4\lambda z)P_3\bigr)_0. ( X 2 ) 0 = ( ( 2 z + 1 ) R + ( 4 λ 2 − 4 λ z ) P 3 ) 0 .
Also,
P ( R , R ) = 2 P 3 , P ( R , P 3 ) = R , P ( P 3 , P 3 ) = 0. \mathcal{P}(R,R)=2P_3,\qquad\mathcal{P}(R,P_3)=R,\qquad\mathcal{P}(P_3,P_3)=0. P ( R , R ) = 2 P 3 , P ( R , P 3 ) = R , P ( P 3 , P 3 ) = 0.
Bilinear expansion, using (41) to discard scalar terms, now gives
P ( X , X ) 0 = ( ( − 2 z − 4 λ ) R + ( 2 + 4 λ z ) P 3 ) 0 . \mathcal{P}(X,X)_0=\bigl((-2z-4\lambda)R+(2+4\lambda z)P_3\bigr)_0. P ( X , X ) 0 = ( ( − 2 z − 4 λ ) R + ( 2 + 4 λ z ) P 3 ) 0 .
Since the tensors g g g , R R R , P 3 P_3 P 3 are linearly independent, F μ ( X ) = 0 F_\mu(X)=0 F μ ( X ) = 0 holds exactly when the coefficients of R R R and P 3 P_3 P 3 vanish. Rewriting z z z in terms of λ \lambda λ , we obtain
3 + 4 λ + ( 4 − 32 λ ) μ = 0 , (79) 3+4\lambda+(4-32\lambda)\mu=0, \tag*{(79)} 3 + 4 λ + ( 4 − 32 λ ) μ = 0 , (79)
8 λ + 20 λ 2 + ( 14 − 8 λ + 8 λ 2 ) μ = 0. (80) 8\lambda+20\lambda^2+(14-8\lambda+8\lambda^2)\mu=0. \tag*{(80)} 8 λ + 20 λ 2 + ( 14 − 8 λ + 8 λ 2 ) μ = 0. (80)
These two equations say that the matrix
( 3 + 4 λ 4 − 32 λ 8 λ + 20 λ 2 14 − 8 λ + 8 λ 2 ) \begin{pmatrix}
3+4\lambda& 4-32\lambda\\
8\lambda+20\lambda^2 & 14-8\lambda+8\lambda^2
\end{pmatrix} ( 3 + 4 λ 8 λ + 20 λ 2 4 − 32 λ 14 − 8 λ + 8 λ 2 )
annihilates the nonzero vector ( 1 , μ ) ⊤ (1,\mu)^{\top} ( 1 , μ ) ⊤ . Its determinant must therefore vanish, which gives the following equation
( 2 λ + 1 ) ( 8 λ 2 − 2 λ + 1 ) = 0. (2\lambda+1)(8\lambda^2-2\lambda+1)=0. ( 2 λ + 1 ) ( 8 λ 2 − 2 λ + 1 ) = 0.
If λ = − 1 2 \lambda=-\frac{1}{2} λ = − 2 1 , equation (79) gives μ = − 1 20 \mu=-\frac{1}{20} μ = − 20 1 . The line test (77) then forces κ = 7 15 \kappa=\frac{7}{15} κ = 15 7 , whereas the 3-plane test (78) forces κ = 0 \kappa=0 κ = 0 , giving a contradiction.
Hence 8 λ 2 − 2 λ + 1 = 0 8\lambda^2-2\lambda+1=0 8 λ 2 − 2 λ + 1 = 0 . In particular, 4 − 32 λ ≠ 0 4-32\lambda\ne0 4 − 32 λ = 0 , and thus solving (79) and using the quadratic relation yields
μ = 1 4 − λ . \mu=\frac{1}{4}-\lambda. μ = 4 1 − λ .
This immediately implies that λ \lambda λ and μ \mu μ are the two distinct roots λ ± \lambda_{\pm} λ ± . Finally, 5 + 2 λ ≠ 0 5+2\lambda\ne0 5 + 2 λ = 0 , so we may now solve (77) for κ \kappa κ . Using λ + μ = 1 4 \lambda+\mu=\frac{1}{4} λ + μ = 4 1 and λ μ = 1 8 \lambda\mu=\frac{1}{8} λ μ = 8 1 from the quadratic equation, we obtain
κ = ( 5 + 2 λ ) ( 5 + 2 μ ) 42 = 25 + 10 ( λ + μ ) + 4 λ μ 42 = 2 3 . \kappa=\frac{(5+2\lambda)(5+2\mu)}{42}=\frac{25+10(\lambda+\mu)+4\lambda\mu}{42}=\frac{2}{3}. κ = 42 ( 5 + 2 λ ) ( 5 + 2 μ ) = 42 25 + 10 ( λ + μ ) + 4 λ μ = 3 2 .
Sufficiency. Conversely, let λ , μ \lambda,\mu λ , μ be the two roots of 8 t 2 − 2 t + 1 8t^2-2t+1 8 t 2 − 2 t + 1 , so λ + μ = 1 / 4 \lambda+\mu=1/4 λ + μ = 1/4 and λ μ = 1 / 8 \lambda\mu=1/8 λ μ = 1/8 . Bilinearity of the halved polarizations gives
F μ ( F λ ( h ) ) = A ( A ( h ) ) + 2 λ A ( A ( h ) , B ( h ) ) + λ 2 A ( B ( h ) ) + μ B ( A ( h ) ) + 2 λ μ B ( A ( h ) , B ( h ) ) + λ 2 μ B ( B ( h ) ) . \begin{aligned}
F_{\mu}(F_{\lambda}(h))={}&A(A(h))+2\lambda A(A(h),B(h))+\lambda^2A(B(h))\\
&+\mu B(A(h))+2\lambda\mu B(A(h),B(h))+\lambda^2\mu B(B(h)).
\end{aligned} F μ ( F λ ( h )) = A ( A ( h )) + 2 λ A ( A ( h ) , B ( h )) + λ 2 A ( B ( h )) + μ B ( A ( h )) + 2 λ μ B ( A ( h ) , B ( h )) + λ 2 μ B ( B ( h )) .
Use μ = 1 / 4 − λ \mu=1/4-\lambda μ = 1/4 − λ , λ 2 = λ / 4 − 1 / 8 \lambda^2=\lambda/4-1/8 λ 2 = λ /4 − 1/8 and λ 2 μ = λ / 8 \lambda^2\mu=\lambda/8 λ 2 μ = λ /8 to collect this as
8 F μ ( F λ ( h ) ) = [ 8 A ( A ( h ) ) − A ( B ( h ) ) + 2 B ( A ( h ) ) + 2 B ( A ( h ) , B ( h ) ) ] + λ [ 16 A ( A ( h ) , B ( h ) ) + 2 A ( B ( h ) ) − 8 B ( A ( h ) ) + B ( B ( h ) ) ] . (81) \begin{aligned}
8F_{\mu}(F_{\lambda}(h))={}&[8A(A(h))-A(B(h))+2B(A(h))+2B(A(h),B(h))]\\
&+\lambda[16A(A(h),B(h))+2A(B(h))-8B(A(h))+B(B(h))]. \tag*{(81)}
\end{aligned} 8 F μ ( F λ ( h )) = [ 8 A ( A ( h )) − A ( B ( h )) + 2 B ( A ( h )) + 2 B ( A ( h ) , B ( h ))] + λ [ 16 A ( A ( h ) , B ( h )) + 2 A ( B ( h )) − 8 B ( A ( h )) + B ( B ( h ))] . (81)
Lemma 3.5 identifies the brackets as 16 C 0 ( h ) h / 3 16C_0(h)h/3 16 C 0 ( h ) h /3 and 16 C φ ( h ) h / 3 16C_{\varphi}(h)h/3 16 C φ ( h ) h /3 , respectively. Dividing by 8 8 8 , we obtain F μ ( F λ ( h ) ) = 2 3 N λ ( h ) h F_{\mu}(F_{\lambda}(h))=\frac{2}{3}N_{\lambda}(h)h F μ ( F λ ( h )) = 3 2 N λ ( h ) h on all of W C W_{\mathbb{C}} W C .
Lemma 3.8. The image of F ± F_{\pm} F ± contains { u ∈ W C : N ∓ ( u ) ≠ 0 } \{u\in W_{\mathbb{C}:N_{\mp}(u)\ne0\}} { u ∈ W C : N ∓ ( u ) = 0 } . In particular, F ± F_{\pm} F ± is dominant, and a polynomial identity holding on its image holds on all of W C W_{\mathbb{C}} W C .
Proof. Proposition 3.7 gives F ± ( F ∓ ( u ) ) = 2 3 N ∓ ( u ) u F_{\pm}(F_{\mp}(u))=\frac{2}{3}N_{\mp}(u)u F ± ( F ∓ ( u )) = 3 2 N ∓ ( u ) u . If N ∓ ( u ) ≠ 0 N_{\mp}(u)\ne0 N ∓ ( u ) = 0 , choose t ∈ C t\in\mathbb{C} t ∈ C with t 2 2 3 N ∓ ( u ) = 1 t^2\frac{2}{3}N_{\mp}(u)=1 t 2 3 2 N ∓ ( u ) = 1 . By homogeneity, F ± ( t F ∓ ( u ) ) = u F_{\pm}(tF_{\mp}(u))=u F ± ( t F ∓ ( u )) = u . The nonvanishing set of the nonzero cubic N ∓ N_{\mp} N ∓ is Zariski-dense, which proves the remaining assertions.
The bilinear expansion of F μ ( F λ ( h ) ) F_{\mu}(F_{\lambda}(h)) F μ ( F λ ( h )) in the proof of Proposition 3.7 gives an identity for unrestricted λ , μ ∈ C \lambda,\mu\in\mathbb{C} λ , μ ∈ C . Put
R λ ( 1 ) ( h ) = B ( A ( h ) ) + λ A ( B ( h ) ) , \mathcal{R}^{(1)}_{\lambda}(h)=B(A(h))+\lambda A(B(h)), R λ ( 1 ) ( h ) = B ( A ( h )) + λ A ( B ( h )) ,
R λ ( 2 ) ( h ) = 2 B ( A ( h ) , B ( h ) ) − A ( B ( h ) ) + λ B ( B ( h ) ) . \mathcal{R}^{(2)}_{\lambda}(h)=2B(A(h),B(h))-A(B(h))+\lambda B(B(h)). R λ ( 2 ) ( h ) = 2 B ( A ( h ) , B ( h )) − A ( B ( h )) + λ B ( B ( h )) .
Then, as a polynomial identity on W C W_{\mathbb{C}} W C ,
F μ ( F λ ( h ) ) = 2 3 N λ ( h ) h + ( λ + μ − 1 4 ) R λ ( 1 ) ( h ) + ( λ μ − 1 8 ) R λ ( 2 ) ( h ) . (82) \begin{aligned}
F_{\mu}(F_{\lambda}(h))={}&\frac{2}{3}N_{\lambda}(h)h\\
&+\left(\lambda+\mu-\frac{1}{4}\right)\mathcal{R}^{(1)}_{\lambda}(h)
+\left(\lambda\mu-\frac{1}{8}\right)\mathcal{R}^{(2)}_{\lambda}(h). \tag*{(82)}
\end{aligned} F μ ( F λ ( h )) = 3 2 N λ ( h ) h + ( λ + μ − 4 1 ) R λ ( 1 ) ( h ) + ( λ μ − 8 1 ) R λ ( 2 ) ( h ) . (82)
Indeed, subtracting the two remainder terms from that expansion leaves one eighth of the left-hand side of (73) plus λ / 8 \lambda/8 λ /8 times that of (74) . For example, the coefficient of A ( B ( h ) ) A(B(h)) A ( B ( h )) becomes λ 2 − λ ( λ + μ − 1 4 ) + ( λ μ − 1 8 ) = λ / 4 − 1 / 8 \lambda^2-\lambda(\lambda+\mu-\frac{1}{4})+(\lambda\mu-\frac{1}{8})=\lambda/4-1/8 λ 2 − λ ( λ + μ − 4 1 ) + ( λ μ − 8 1 ) = λ /4 − 1/8 . Thus (82) specializes to the identity F μ ( F λ ( h ) ) = 2 3 N λ ( h ) h F_\mu(F_\lambda(h))=\frac{2}{3}N_\lambda(h)h F μ ( F λ ( h )) = 3 2 N λ ( h ) h precisely when the two remainder coefficients vanish.
Remark 3.9. In Proposition 3.7, the parameters λ ± \lambda_{\pm} λ ± are determined without computing a stabilizer. The later stabilizer calculation shows that these same parameters are exactly the two 78-dimensional cases in Theorem 3.3 . Double-adjoint identities are classical in the theory of cubic Jordan norms [56 ] . Their relation to polynomial multiplicative Legendre transforms and semisimple rank-three Jordan algebras is developed in [18 Theorem 3.10] , [4 Theorem 2 and Corollary 4] , and [53 Corollary 5.9] , under the hypotheses stated there. McCrimmon gives a coordinate proof of x # # = N ( x ) x x^{\mathbin{\#\#}}=N(x)x x ## = N ( x ) x in the Hermitian-matrix model [48 pp. 502–503] . In the proof of Proposition 3.7, we use neither Jordan theory nor representation theory. The parameters are determined directly from the G 2 G_2 G 2 model.
The double-adjoint identity of Proposition 3.7 will produce stabilizer elements through the Koecher operators of § 4.5 . We first exclude the five isotypic blocks that can never contribute, so that the resulting lower bound can be compared with the contribution of the remaining blocks.
Stabilizer exclusion
We will show that five isotypic blocks of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) never enter the infinitesimal stabilizer ker Φ a , b \ker\Phi_{a,b} ker Φ a , b for ( a , b ) ≠ 0 (a,b)\ne0 ( a , b ) = 0 . We prove this exclusion here because it shortens the determination of the stabilizer: after the scalar component is excluded for a nonzero cubic, only V ( 1 , 0 ) V(1,0) V ( 1 , 0 ) , V ( 0 , 1 ) V(0,1) V ( 0 , 1 ) and V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) remain as possible components in Table 1. The two copies of V ( 2 , 0 ) V(2,0) V ( 2 , 0 ) are handled together by the Gram matrix of four maps. On each of the four multiplicity-one blocks, a 2 × 2 2\times2 2 × 2 matrix of the scalars by which the compositions Φ s T Φ t \Phi_s^{\mathsf T}\Phi_t Φ s T Φ t of (88) act gives the exclusion.
With the cubic coordinates already fixed, write Φ s = Φ C s \Phi_s=\Phi_{C_s} Φ s = Φ C s for s ∈ { 0 , φ } s\in\{0,\varphi\} s ∈ { 0 , φ } . The action (61) has components
Φ s ( X ) α β γ = X σ α c σ β γ s + X σ β c α σ γ s + X σ γ c α β σ s . (83) \Phi_s(X)_{\alpha\beta\gamma}=X^\sigma{}_\alpha c^s_{\sigma\beta\gamma}+X^\sigma{}_\beta c^s_{\alpha\sigma\gamma}+X^\sigma{}_\gamma c^s_{\alpha\beta\sigma}. \tag*{(83)} Φ s ( X ) α β γ = X σ α c σ β γ s + X σ β c α σ γ s + X σ γ c α β σ s . (83)
Use the full ordered pairing on cubic tensors and the trace-adjoint pairing on endomorphisms,
⟨ U , U ′ ⟩ 3 = U α β γ U ′ α β γ , ⟨ X , Y ⟩ End = tr W ( X T W Y ) . (84) \langle U,U'\rangle_3=U_{\alpha\beta\gamma}U'^{\alpha\beta\gamma},\qquad\langle X,Y\rangle_{\operatorname{End}}=\operatorname{tr}_W(X^{\mathsf T_W}Y). \tag*{(84)} ⟨ U , U ′ ⟩ 3 = U α β γ U ′ α β γ , ⟨ X , Y ⟩ End = tr W ( X T W Y ) . (84)
The cubic pairing has no factor 1 / 3 ! 1/3! 1/3 ! . The algebraic transpose Φ s T \Phi_s^{\mathsf T} Φ s T is characterized by ⟨ Φ s T U , X ⟩ End = ⟨ U , Φ s ( X ) ⟩ 3 \langle\Phi_s^{\mathsf T}U,X\rangle_{\operatorname{End}}=\langle U,\Phi_s(X)\rangle_3 ⟨ Φ s T U , X ⟩ End = ⟨ U , Φ s ( X ) ⟩ 3 , and therefore
( Φ s T U ) σ α = 3 ( c s ) σ β γ U α β γ . (85) (\Phi_s^{\mathsf T}U)^\sigma{}_\alpha=3(c^s)^{\sigma\beta\gamma}U_{\alpha\beta\gamma}. \tag*{(85)} ( Φ s T U ) σ α = 3 ( c s ) σ β γ U α β γ . (85)
Define the operators of multiplication by h h h in the two symmetric products A A A and B B B ,
m h = A ( h , ⋅ ) , n h = B ( h , ⋅ ) , m h ( 0 ) = m h , m h ( φ ) = n h . (86) m_h=A(h,\mathord{\cdot}),\qquad n_h=B(h,\mathord{\cdot}),\qquad m_h^{(0)}=m_h,\qquad m_h^{(\varphi)}=n_h. \tag*{(86)} m h = A ( h , ⋅ ) , n h = B ( h , ⋅ ) , m h ( 0 ) = m h , m h ( φ ) = n h . (86)
Lemma 3.10. The operators m h ( s ) m_h^{(s)} m h ( s ) are W W W -self-adjoint and satisfy
⟨ m h 1 ( s ) , m h 2 ( t ) ⟩ End = a s t ⟨ h 1 , h 2 ⟩ W . (87) \langle m_{h_1}^{(s)},m_{h_2}^{(t)}\rangle_{\operatorname{End}}=a_{st}\langle h_1,h_2\rangle_W. \tag*{(87)} ⟨ m h 1 ( s ) , m h 2 ( t ) ⟩ End = a s t ⟨ h 1 , h 2 ⟩ W . (87)
The map ( h 1 , h 2 ) ↦ m h 1 + n h 2 (h_1,h_2) \mapsto m h_1+n h_2 ( h 1 , h 2 ) ↦ m h 1 + n h 2 is a G 2 G_2 G 2 -equivariant isomorphism from W C ⊕ W C W_{\mathbb{C}} \oplus W_{\mathbb{C}} W C ⊕ W C onto the full 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) -isotypic component of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) .
Proof. Self-adjointness follows from the full symmetry of c s c^s c s , and (3.25) is exactly the two-index c-contraction identity (3.9). If m h 1 + n h 2 = 0 m h_1+n h_2=0 m h 1 + n h 2 = 0 , pair it with m k m k mk and n k n k nk for arbitrary k ∈ W C k \in W_{\mathbb{C}} k ∈ W C . The resulting system has coefficient matrix a a a , whose determinant is 44, so the nondegenerate W W W -pairing gives h 1 = h 2 = 0 h_1=h_2=0 h 1 = h 2 = 0 . The map is G 2 G_2 G 2 -equivariant, and by Table 1, End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) contains exactly two copies of W = V ( 2 , 0 ) W=V(2,0) W = V ( 2 , 0 ) . The image of the map, of dimension 54, is therefore the whole isotypic component.
To apply the c-contraction identities to endomorphisms, we define Z s t \mathcal{Z}_{st} Z s t by the first line below; the second line then holds:
Z s t ( X ) α β = ( c s ) α γ δ c β σ δ t X σ γ , Φ s T Φ t ( X ) = 3 a s t X + 6 Z s t ( X ) . (88) \begin{aligned}
\mathcal{Z}_{st}(X)^\alpha{}_\beta&= (c^s)^{\alpha\gamma\delta}c^t_{\beta\sigma\delta}X^\sigma{}_\gamma,\\
\Phi_s^{\mathsf{T}}\Phi_t(X) &= 3a_{st}X+6\mathcal{Z}_{st}(X). \tag*{(88)}
\end{aligned} Z s t ( X ) α β Φ s T Φ t ( X ) = ( c s ) α γ δ c β σ δ t X σ γ , = 3 a s t X + 6 Z s t ( X ) . (88)
Indeed, substitute the three terms of (3.21) into (3.23). The first contracts to 3 a s t X 3a_{st}X 3 a s t X by (3.9); the other two coincide by cubic symmetry and give 6 Z s t ( X ) 6\mathcal{Z}_{st}(X) 6 Z s t ( X ) . Equivalently,
Z s t ( X ) = ∑ α m k α ( s ) X T W m k α ( t ) . \mathcal{Z}_{st}(X)=\sum_\alpha m_{k\alpha}^{(s)}X^{\mathsf{T}W}m_{k\alpha}^{(t)}. Z s t ( X ) = α ∑ m k α ( s ) X T W m k α ( t ) .
Applying Φ 0 T \Phi_0^{\mathsf{T}} Φ 0 T and Φ φ T \Phi_\varphi^{\mathsf{T}} Φ φ T to Φ a , b ( X ) = a Φ 0 ( X ) + b Φ φ ( X ) = 0 \Phi_{a,b}(X)=a\Phi_0(X)+b\Phi_\varphi(X)=0 Φ a , b ( X ) = a Φ 0 ( X ) + b Φ φ ( X ) = 0 shows that every X ∈ ker Φ a , b X\in\ker\Phi_{a,b} X ∈ ker Φ a , b satisfies
a Φ 0 T Φ 0 ( X ) + b Φ 0 T Φ φ ( X ) = 0 , (89) a\Phi_0^{\mathsf{T}}\Phi_0(X)+b\Phi_0^{\mathsf{T}}\Phi_\varphi(X)=0, \tag*{(89)} a Φ 0 T Φ 0 ( X ) + b Φ 0 T Φ φ ( X ) = 0 , (89)
a Φ φ T Φ 0 ( X ) + b Φ φ T Φ φ ( X ) = 0. (90) a\Phi_\varphi^{\mathsf{T}}\Phi_0(X)+b\Phi_\varphi^{\mathsf{T}}\Phi_\varphi(X)=0. \tag*{(90)} a Φ φ T Φ 0 ( X ) + b Φ φ T Φ φ ( X ) = 0. (90)
For the 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) block, consider the four maps
B s t : W C ⟶ S 3 W C ∗ , B s t ( h ) = Φ s ( m h ( t ) ) . \mathcal{B}_{st}:W_{\mathbb{C}}\longrightarrow S^3W_{\mathbb{C}}^*,\qquad\mathcal{B}_{st}(h)=\Phi_s(m_h^{(t)}). B s t : W C ⟶ S 3 W C ∗ , B s t ( h ) = Φ s ( m h ( t ) ) .
Enumerate the four maps by
B 1 = B 00 , B 2 = B φ 0 , B 3 = B 0 φ , B 4 = B φ φ . \mathcal{B}_1=\mathcal{B}_{00},\qquad\mathcal{B}_2=\mathcal{B}_{\varphi0},\qquad\mathcal{B}_3=\mathcal{B}_{0\varphi},\qquad\mathcal{B}_4=\mathcal{B}_{\varphi\varphi}. B 1 = B 00 , B 2 = B φ 0 , B 3 = B 0 φ , B 4 = B φφ .
Let B q T : S 3 W C ∗ → W C \mathcal{B}_q^{\mathsf{T}}:S^3W_{\mathbb{C}}^*\to W_{\mathbb{C}} B q T : S 3 W C ∗ → W C denote the transpose of B q \mathcal{B}_q B q with respect to ⟨ ⋅ , ⋅ ⟩ 3 \langle\cdot,\cdot\rangle_3 ⟨ ⋅ , ⋅ ⟩ 3 and the W W W -pairing. Write G \mathcal{G} G for the 4 × 4 4\times4 4 × 4 matrix of the scalars by which the compositions B p T B q \mathcal{B}_p^{\mathsf{T}}\mathcal{B}_q B p T B q act.
Lemma 3.11. The four maps satisfy B p T B q = G p q id W \mathcal{B}_p^{\mathsf{T}}\mathcal{B}_q=\mathcal{G}_{pq}\operatorname{id}_W B p T B q = G pq id W , where
G = 1 196 ( 4455 1782 1782 − 9636 1782 16236 − 9636 27192 1782 − 9636 16236 27192 − 9636 27192 27192 456816 ) , det G = 179397504 7 ≠ 0. (91) \mathcal{G}=\frac{1}{196}
\begin{pmatrix}
4455 & 1782 & 1782 & -9636\\
1782 & 16236 & -9636 & 27192\\
1782 & -9636 & 16236 & 27192\\
-9636 & 27192 & 27192 & 456816
\end{pmatrix},
\qquad
\det\mathcal{G}=\frac{179397504}{7}\ne0. \tag*{(91)} G = 196 1 4455 1782 1782 − 9636 1782 16236 − 9636 27192 1782 − 9636 16236 27192 − 9636 27192 27192 456816 , det G = 7 179397504 = 0. (91)
The contraction proof is given in Section A.2, the proof of Lemma 3.11.
On a multiplicity-one block, write Γ s t \Gamma_{st} Γ s t for the scalar by which Φ s T Φ t \Phi_s^{\mathsf{T}}\Phi_t Φ s T Φ t acts, with s , t ∈ { 0 , φ } s,t\in\{0,\varphi\} s , t ∈ { 0 , φ } , and let Γ = ( Γ s t ) \Gamma=(\Gamma_{st}) Γ = ( Γ s t ) in the order 0 , φ 0,\varphi 0 , φ . The following table records all four matrices.
Lemma 3.12. On the four multiplicity-one blocks, the compositions satisfy Φ s ⊤ Φ t = Γ s t i d \Phi_s^{\top}\Phi_t=\Gamma_{st}\,\mathrm{id} Φ s ⊤ Φ t = Γ s t id , with the following scalars and determinants:
The contraction proof is given in Section A.2, the proof of Lemma 3.12.
Proposition 3.13. For every ( a , b ) ∈ C 2 ∖ { 0 } (a,b)\in\mathbb{C}^{2}\setminus\{0\} ( a , b ) ∈ C 2 ∖ { 0 } , the restriction of the action map Φ a , b = a Φ C 0 + b Φ C φ \Phi_{a,b}=a\Phi_{C_0}+b\Phi_{C_\varphi} Φ a , b = a Φ C 0 + b Φ C φ to each of the following isotypic components is injective:
W W W -self-adjoint part S 2 W C S^{2}W_{\mathbb{C}} S 2 W C | 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) , V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) , V ( 4 , 0 ) V(4,0) V ( 4 , 0 ) W W W -skew-adjoint part Λ 2 W C \Lambda^{2}W_{\mathbb{C}} Λ 2 W C | V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) , V ( 2 , 1 ) V(2,1) V ( 2 , 1 ) .
Here injectivity on 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) means injectivity on the sum of both copies. None of these five types occurs in ker Φ a , b \ker\Phi_{a,b} ker Φ a , b .
Proof. By Lemma 3.10, an element of the 2 V ( 2 , 0 ) 2V(2,0) 2 V ( 2 , 0 ) block has the unique form X = m h 1 + n h 2 X=mh_1+nh_2 X = m h 1 + n h 2 . If Φ a , b ( X ) = 0 \Phi_{a,b}(X)=0 Φ a , b ( X ) = 0 , then
B 1 ( a h 1 ) + B 2 ( b h 1 ) + B 3 ( a h 2 ) + B 4 ( b h 2 ) = 0. \mathcal{B}_1(ah_1)+\mathcal{B}_2(bh_1)+\mathcal{B}_3(ah_2)+\mathcal{B}_4(bh_2)=0. B 1 ( a h 1 ) + B 2 ( b h 1 ) + B 3 ( a h 2 ) + B 4 ( b h 2 ) = 0. Applying the four transposes gives G ( a h 1 , b h 1 , a h 2 , b h 2 ) ⊤ = 0 \mathcal{G}(ah_1,bh_1,ah_2,bh_2)^{\top}=0 G ( a h 1 , b h 1 , a h 2 , b h 2 ) ⊤ = 0 . By Lemma 3.11, G \mathcal{G} G is invertible, so a h 1 = b h 1 = a h 2 = b h 2 = 0 ah_1=bh_1=ah_2=bh_2=0 a h 1 = b h 1 = a h 2 = b h 2 = 0 . Since ( a , b ) ≠ 0 (a,b)\ne0 ( a , b ) = 0 , we have h 1 = h 2 = 0 h_1=h_2=0 h 1 = h 2 = 0 .
On any of the four multiplicity-one blocks, applying Φ 0 ⊤ \Phi_0^{\top} Φ 0 ⊤ and Φ φ ⊤ \Phi_\varphi^{\top} Φ φ ⊤ to Φ a , b ( X ) = 0 \Phi_{a,b}(X)=0 Φ a , b ( X ) = 0 gives Γ ( a X , b X ) ⊤ = 0 \Gamma(aX,bX)^{\top}=0 Γ ( a X , b X ) ⊤ = 0 by (3.27a) and (3.27b). Every determinant in Lemma 3.12 is nonzero. Thus a X = b X = 0 aX=bX=0 a X = b X = 0 , and again X = 0 X=0 X = 0 . □ \square □
Once the scalar endomorphisms are also excluded (see the proof of Proposition 3.31), only the 7, 14 and 64 components can occur in the stabilizer of a nonzero cubic. We next construct enough infinitesimal symmetries at the critical parameters to force the 64 component to occur.
Koecher operators and a conceptual lower bound
Differentiating the double-adjoint identity produces endomorphisms that preserve the cubic. We call these Koecher operators because they are related to the Tits–Kantor–Koecher construction in the theory of Jordan algebras [13 , 23 , 57 ] . Lemma 3.15 applies to any pair of cubics with quadratic adjoints. We first need a precise definition.
Definition 3.14. Let X \mathcal{X} X be a finite-dimensional vector space over a field F \mathbb{F} F of characteristic other than 2 and 3, with a nondegenerate symmetric bilinear form ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ . Let N N N and N ′ N' N ′ be
cubic forms on X \mathcal{X} X with the polarization convention N ( h , h , h ) = N ( h ) N(h,h,h)=N(h) N ( h , h , h ) = N ( h ) and N ′ ( h , h , h ) = N ′ ( h ) N'(h,h,h)=N'(h) N ′ ( h , h , h ) = N ′ ( h ) .
Define symmetric bilinear maps F , F ′ : X × X → X F,F'\colon\mathcal{X}\times\mathcal{X}\to\mathcal{X} F , F ′ : X × X → X by
⟨ F ( h 1 , h 2 ) , h 3 ⟩ = N ( h 1 , h 2 , h 3 ) , ⟨ F ′ ( h 1 , h 2 ) , h 3 ⟩ = N ′ ( h 1 , h 2 , h 3 ) . \langle F(h_1,h_2),h_3\rangle=N(h_1,h_2,h_3),\qquad\langle F'(h_1,h_2),h_3\rangle=N'(h_1,h_2,h_3). ⟨ F ( h 1 , h 2 ) , h 3 ⟩ = N ( h 1 , h 2 , h 3 ) , ⟨ F ′ ( h 1 , h 2 ) , h 3 ⟩ = N ′ ( h 1 , h 2 , h 3 ) .
Write F ( h ) = F ( h , h ) F(h)=F(h,h) F ( h ) = F ( h , h ) and F ′ ( h ) = F ′ ( h , h ) F'(h)=F'(h,h) F ′ ( h ) = F ′ ( h , h ) , and define stab ( N ) \operatorname{stab}(N) stab ( N ) as in Definition 3.1. For 0 ≠ κ ∈ F 0\ne\kappa\in\mathbb{F} 0 = κ ∈ F and h 1 , h 2 ∈ X h_1,h_2\in\mathcal{X} h 1 , h 2 ∈ X define the Koecher operator S ( h 1 , h 2 ) ∈ End ( X ) S(h_1,h_2)\in\operatorname{End}(\mathcal{X}) S ( h 1 , h 2 ) ∈ End ( X ) by
S ( h 1 , h 2 ) h = ⟨ h , h 2 ⟩ h 1 − 4 3 κ F ′ ( F ( h 1 , h ) , h 2 ) + 1 3 ⟨ h 1 , h 2 ⟩ h . (92) S(h_1,h_2)h=\langle h,h_2\rangle h_1-\frac{4}{3\kappa}F'(F(h_1,h),h_2)+\frac{1}{3}\langle h_1,h_2\rangle h. \tag*{(92)} S ( h 1 , h 2 ) h = ⟨ h , h 2 ⟩ h 1 − 3 κ 4 F ′ ( F ( h 1 , h ) , h 2 ) + 3 1 ⟨ h 1 , h 2 ⟩ h . (92)
Lemma 3.15. With notation as in Definition 3.14, the following are equivalent:
(i) F ′ ( F ( h ) ) = κ N ( h ) h F'(F(h))=\kappa N(h)h F ′ ( F ( h )) = κ N ( h ) h for all h ∈ X h\in\mathcal{X} h ∈ X ;
(ii) S ( h 1 , h 2 ) ∈ stab ( N ) S(h_1,h_2)\in\operatorname{stab}(N) S ( h 1 , h 2 ) ∈ stab ( N ) for all h 1 , h 2 ∈ X h_1,h_2\in\mathcal{X} h 1 , h 2 ∈ X .
For the pencil, Lemma 3.15 gives the following corollary immediately.
Corollary 3.16. *Take X = W C \mathcal{X}=W_{\mathbb{C}} X = W C with the W W W -pairing, N = N λ N=N_\lambda N = N λ and N ′ = N μ N'=N_\mu N ′ = N μ , so that F = F λ F=F_\lambda F = F λ and F ′ = F μ F'=F_\mu F ′ = F μ by (3.5). For λ , μ ∈ C \lambda,\mu\in\mathbb{C} λ , μ ∈ C and κ ≠ 0 \kappa\ne0 κ = 0 , the identity (3.15) holds at ( λ , μ , κ ) (\lambda,\mu,\kappa) ( λ , μ , κ ) if and only if every Koecher operator S ( h 1 , h 2 ) = S λ μ κ ( h 1 , h 2 ) S(h_1,h_2)=S_{\lambda\mu\kappa}(h_1,h_2) S ( h 1 , h 2 ) = S λ μ κ ( h 1 , h 2 ) lies in stab ( N λ ) \operatorname{stab}(N_\lambda) stab ( N λ ) .
We now proceed with the proof of Lemma 3.15.
Proof of Lemma 3.15. Regarding (i) as a quartic polynomial equation in h h h , we differentiate it in the direction h 1 h_1 h 1 . Since F ( h + t h 1 ) = F ( h ) + 2 t F ( h , h 1 ) + t 2 F ( h 1 ) F(h+th_1)=F(h)+2tF(h,h_1)+t^2F(h_1) F ( h + t h 1 ) = F ( h ) + 2 tF ( h , h 1 ) + t 2 F ( h 1 ) , the coefficient of t t t in F ′ ( F ( h + t h 1 ) ) F'(F(h+th_1)) F ′ ( F ( h + t h 1 )) is 2 F ′ ( F ( h ) , 2 F ( h , h 1 ) ) 2F'(F(h),2F(h,h_1)) 2 F ′ ( F ( h ) , 2 F ( h , h 1 )) , while that of N ( h + t h 1 ) ( h + t h 1 ) N(h+th_1)(h+th_1) N ( h + t h 1 ) ( h + t h 1 ) is N ( h ) h 1 + 3 N ( h , h , h 1 ) h = N ( h ) h 1 + 3 ⟨ F ( h ) , h 1 ⟩ h N(h)h_1+3N(h,h,h_1)h=N(h)h_1+3\langle F(h),h_1\rangle h N ( h ) h 1 + 3 N ( h , h , h 1 ) h = N ( h ) h 1 + 3 ⟨ F ( h ) , h 1 ⟩ h . Hence (i) gives
4 F ′ ( F ( h ) , F ( h , h 1 ) ) = κ ( N ( h ) h 1 + 3 ⟨ F ( h ) , h 1 ⟩ h ) . (93) 4F'(F(h),F(h,h_1))=\kappa\left(N(h)h_1+3\langle F(h),h_1\rangle h\right). \tag*{(93)} 4 F ′ ( F ( h ) , F ( h , h 1 )) = κ ( N ( h ) h 1 + 3 ⟨ F ( h ) , h 1 ⟩ h ) . (93)
Next pair S ( h 1 , h 2 ) h S(h_1,h_2)h S ( h 1 , h 2 ) h with F ( h ) F(h) F ( h ) . By the full symmetry of ⟨ F ′ ( ⋅ , ⋅ ) , ⋅ ⟩ = N ′ \langle F'(\mathord{\cdot},\mathord{\cdot}),\mathord{\cdot}\rangle=N' ⟨ F ′ ( ⋅ , ⋅ ) , ⋅ ⟩ = N ′ ,
⟨ F ( h ) , F ′ ( F ( h 1 , h ) , h 2 ) ⟩ = ⟨ F ′ ( F ( h ) , F ( h , h 1 ) ) , h 2 ⟩ . \langle F(h),F'(F(h_1,h),h_2)\rangle=\langle F'(F(h),F(h,h_1)),h_2\rangle. ⟨ F ( h ) , F ′ ( F ( h 1 , h ) , h 2 )⟩ = ⟨ F ′ ( F ( h ) , F ( h , h 1 )) , h 2 ⟩ .
Also N ( h 1 , h , h ) = ⟨ F ( h ) , h 1 ⟩ N(h_1,h,h)=\langle F(h),h_1\rangle N ( h 1 , h , h ) = ⟨ F ( h ) , h 1 ⟩ , so
N ( S ( h 1 , h 2 ) h , h , h ) = ⟨ F ( h ) , S ( h 1 , h 2 ) h ⟩ = ⟨ h , h 2 ⟩ ⟨ F ( h ) , h 1 ⟩ − 4 3 κ ⟨ F ′ ( F ( h ) , F ( h , h 1 ) ) , h 2 ⟩ + 1 3 ⟨ h 1 , h 2 ⟩ N ( h ) . (94) \begin{aligned}
N(S(h_1,h_2)h,h,h)
&=\langle F(h),S(h_1,h_2)h\rangle\\
&=\langle h,h_2\rangle\langle F(h),h_1\rangle-\frac{4}{3\kappa}\langle F'(F(h),F(h,h_1)),h_2\rangle+\frac{1}{3}\langle h_1,h_2\rangle N(h). \tag*{(94)}
\end{aligned} N ( S ( h 1 , h 2 ) h , h , h ) = ⟨ F ( h ) , S ( h 1 , h 2 ) h ⟩ = ⟨ h , h 2 ⟩ ⟨ F ( h ) , h 1 ⟩ − 3 κ 4 ⟨ F ′ ( F ( h ) , F ( h , h 1 )) , h 2 ⟩ + 3 1 ⟨ h 1 , h 2 ⟩ N ( h ) . (94)
and (ii) is the vanishing of (94) for all h , h 1 , h 2 h,h_1,h_2 h , h 1 , h 2 . If (i) holds, then by (93) the middle term equals − 1 3 ( N ( h ) ⟨ h 1 , h 2 ⟩ + 3 ⟨ F ( h ) , h 1 ⟩ ⟨ h , h 2 ⟩ ) -\frac{1}{3}\left(N(h)\langle h_1,h_2\rangle+3\langle F(h),h_1\rangle\langle h,h_2\rangle\right) − 3 1 ( N ( h ) ⟨ h 1 , h 2 ⟩ + 3 ⟨ F ( h ) , h 1 ⟩ ⟨ h , h 2 ⟩ ) , and (94) vanishes identically. Conversely, if (94) vanishes for all h 2 h_2 h 2 , nondegeneracy gives (93) , and setting h 1 = h h_1=h h 1 = h in it gives 4 F ′ ( F ( h ) ) = 4 κ N ( h ) h 4F'(F(h))=4\kappa N(h)h 4 F ′ ( F ( h )) = 4 κ N ( h ) h . □
Each O ∈ G 2 O\in G_2 O ∈ G 2 acts on W C W_{\mathbb{C}} W C by ρ ( O ) \rho(O) ρ ( O ) , preserving the W W W -pairing and the maps F λ F_\lambda F λ and F μ F_\mu F μ . Consequently,
S ( ρ ( O ) h 1 , ρ ( O ) h 2 ) = ρ ( O ) S ( h 1 , h 2 ) ρ ( O ) − 1 , (95) S(\rho(O)h_1,\rho(O)h_2)=\rho(O)S(h_1,h_2)\rho(O)^{-1}, \tag*{(95)} S ( ρ ( O ) h 1 , ρ ( O ) h 2 ) = ρ ( O ) S ( h 1 , h 2 ) ρ ( O ) − 1 , (95)
and the span of the Koecher operators is a G 2 G_2 G 2 -submodule of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) . At κ = 2 3 \kappa= \frac{2}{3} κ = 3 2 put
D ( h 1 , h 2 ) = S ( h 1 , h 2 ) + 2 3 ⟨ h 1 , h 2 ⟩ W id W , D(h_1,h_2)=S(h_1,h_2)+\frac{2}{3}\langle h_1,h_2\rangle_W\operatorname{id}_W, D ( h 1 , h 2 ) = S ( h 1 , h 2 ) + 3 2 ⟨ h 1 , h 2 ⟩ W id W ,
D ( h 1 , h 2 ) h = ⟨ h 1 , h 2 ⟩ W h + ⟨ h , h 2 ⟩ W h 1 − 2 F μ ( F λ ( h 1 , h ) , h 2 ) . D(h_1,h_2)h=\langle h_1,h_2\rangle_W h+\langle h,h_2\rangle_W h_1-2F_\mu(F_\lambda(h_1,h),h_2). D ( h 1 , h 2 ) h = ⟨ h 1 , h 2 ⟩ W h + ⟨ h , h 2 ⟩ W h 1 − 2 F μ ( F λ ( h 1 , h ) , h 2 ) .
When the double-adjoint identity holds, so that Lemma 3.15 (ii) holds as well, equation (94) gives 3 ⟨ F λ ( h ) , D ( h 1 , h 2 ) h ⟩ W = 2 ⟨ h 1 , h 2 ⟩ W N λ ( h ) 3\langle F_\lambda(h),D(h_1,h_2)h\rangle_W=2\langle h_1,h_2\rangle_WN_\lambda(h) 3 ⟨ F λ ( h ) , D ( h 1 , h 2 ) h ⟩ W = 2 ⟨ h 1 , h 2 ⟩ W N λ ( h ) , that is, Φ N λ ( D ( h 1 , h 2 ) ) = 2 ⟨ h 1 , h 2 ⟩ W N λ \Phi_{N_\lambda}(D(h_1,h_2))=2\langle h_1,h_2\rangle_WN_\lambda Φ N λ ( D ( h 1 , h 2 )) = 2 ⟨ h 1 , h 2 ⟩ W N λ : each D ( h 1 , h 2 ) D(h_1,h_2) D ( h 1 , h 2 ) is an infinitesimal similarity of N λ N_\lambda N λ with multiplier 2 ⟨ h 1 , h 2 ⟩ W 2\langle h_1,h_2\rangle_W 2 ⟨ h 1 , h 2 ⟩ W .
Remark 3.17. It is classical that, in the Jordan algebra setting, the operators D ( h 1 , h 2 ) D(h_1,h_2) D ( h 1 , h 2 ) act as infinitesimal similarities of the norm with multiplier 2 ⟨ h 1 , h 2 ⟩ W 2\langle h_1,h_2\rangle_W 2 ⟨ h 1 , h 2 ⟩ W [57 §§4.3–4.4 and 14.22] , [23 (33a.30) and Proposition 50.5] , and De Medts–Wiedemann also establish this property in [13 Lemma 3.3] (arXiv v1, February 2026). Lemma 3.15 also establishes the converse (ii)⇒ \Rightarrow ⇒ (i) and requires only the one-sided composition F μ ∘ F λ F_\mu\circ F_\lambda F μ ∘ F λ . We have not found these two features explicitly stated in the literature.
We now use the Koecher operators to obtain a lower bound for the dimension of the critical stabilizers. The argument has two steps. First, evaluation at a suitable element E 1 E_1 E 1 shows that the Koecher operators span a space of dimension at least 26, containing symmetries outside of ρ ( s o ( 7 , C ) ) \rho(\mathfrak{so}(7,\mathbb{C})) ρ ( so ( 7 , C )) . We then use Proposition 3.13 and the G 2 G_2 G 2 -module decomposition of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) to show that their span must contain a copy of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) , giving the lower bound 14 + 64 = 78 14+64=78 14 + 64 = 78 .
Let E 1 = 7 P 1 − g E_1=7P_1-g E 1 = 7 P 1 − g , as in (17) , be the test tensor in W W W used for the line test in the proof of Proposition 3.7 , so that
F λ ( E 1 ) = ( 5 + 2 λ ) E 1 , ⟨ E 1 , E 1 ⟩ W = 42. F_\lambda(E_1)=(5+2\lambda)E_1,\qquad\langle E_1,E_1\rangle_W=42. F λ ( E 1 ) = ( 5 + 2 λ ) E 1 , ⟨ E 1 , E 1 ⟩ W = 42.
For a real unit vector u ∈ V u\in V u ∈ V , define Θ u ∈ s o ( V ) \Theta_u\in\mathfrak{so}(V) Θ u ∈ so ( V ) by Θ u ( v ) = v × u \Theta_u(v)=v\times u Θ u ( v ) = v × u , and put Θ 1 = Θ e 1 \Theta_1=\Theta_{e_1} Θ 1 = Θ e 1 , so that
Θ 1 ( v ) = v × e 1 , ( Θ 1 ) i a = φ i a 1 . \Theta_1(v)=v\times e_1,\qquad(\Theta_1)_{ia}=\varphi_{ia1}. Θ 1 ( v ) = v × e 1 , ( Θ 1 ) ia = φ ia 1 .
In the standard frame of (4) , the terms of φ \varphi φ containing the index 1 are e 123 e^{123} e 123 , e 145 e^{145} e 145 and e 167 e^{167} e 167 , giving
Θ 1 = 0 ⊕ J 0 ⊕ J 0 ⊕ J 0 , J 0 = ( 0 1 − 1 0 ) , \Theta_1=0\oplus J_0\oplus J_0\oplus J_0,\qquad J_0=\begin{pmatrix}0&1\\-1&0\end{pmatrix}, Θ 1 = 0 ⊕ J 0 ⊕ J 0 ⊕ J 0 , J 0 = ( 0 − 1 1 0 ) ,
on the blocks ( e 1 ) (e_1) ( e 1 ) , ( e 2 , e 3 ) (e_2,e_3) ( e 2 , e 3 ) , ( e 4 , e 5 ) (e_4,e_5) ( e 4 , e 5 ) and ( e 6 , e 7 ) (e_6,e_7) ( e 6 , e 7 ) . In particular, Θ 1 e 1 = 0 \Theta_1e_1=0 Θ 1 e 1 = 0 and Θ 1 2 = − pr e 1 ⊥ \Theta_1^2=-\operatorname{pr}_{e_1^\perp} Θ 1 2 = − pr e 1 ⊥ , so its restriction to e 1 ⊥ e_1^\perp e 1 ⊥ is a complex structure. The choice of e 1 e_1 e 1 reduces G 2 G_2 G 2 to its stabilizer S U ( 3 ) SU(3) S U ( 3 ) . The complex structure Θ 1 \Theta_1 Θ 1 on e 1 ⊥ e_1^\perp e 1 ⊥ identifies this space with C 3 \mathbb{C}^3 C 3 . We use the corresponding standard S U ( 3 ) SU(3) S U ( 3 ) -module decomposition
W C ∣ S U ( 3 ) ≅ 1 ⊕ ( 3 ⊕ 3 ‾ ) ⊕ 8 ⊕ ( 6 ⊕ 6 ‾ ) , W_{\mathbb{C}}\big|_{SU(3)}\cong1\oplus(3\oplus\overline{3})\oplus8\oplus(6\oplus\overline{6}), W C S U ( 3 ) ≅ 1 ⊕ ( 3 ⊕ 3 ) ⊕ 8 ⊕ ( 6 ⊕ 6 ) ,
and thus define, for a real unit vector u ∈ V u\in V u ∈ V , the following real subspaces of W W W , which we call the S U ( 3 ) SU(3) S U ( 3 ) sectors:
W 6 ( u ) = { u ♭ ⊗ v ♭ + v ♭ ⊗ u ♭ : v ⊥ u } , W_6(u)=\{u^\flat\otimes v^\flat+v^\flat\otimes u^\flat:v\perp u\}, W 6 ( u ) = { u ♭ ⊗ v ♭ + v ♭ ⊗ u ♭ : v ⊥ u } ,
W 8 ( u ) = { h ∈ W : h ( u , ⋅ ) = 0 , Θ u h = h Θ u } , W_8(u)=\{h\in W:h(u,\cdot)=0,\ \Theta_u h=h\Theta_u\}, W 8 ( u ) = { h ∈ W : h ( u , ⋅ ) = 0 , Θ u h = h Θ u } ,
W 12 ( u ) = { h ∈ W : h ( u , ⋅ ) = 0 , Θ u h = − h Θ u } , W_{12}(u)=\{h\in W:h(u,\cdot)=0,\ \Theta_u h=-h\Theta_u\}, W 12 ( u ) = { h ∈ W : h ( u , ⋅ ) = 0 , Θ u h = − h Θ u } ,
where indices are contracted with the metric, so that ( Θ u h ) a b = ( Θ u ) a i h b j g i j (\Theta_u h)_{ab}=(\Theta_u)_{ai}h_{bj}g^{ij} ( Θ u h ) ab = ( Θ u ) ai h bj g ij . The subscript C denotes complexification: W r ( u ) C = W r ( u ) ⊗ C W_r(u)_{\mathbb{C}}=W_r(u)\otimes\mathbb{C} W r ( u ) C = W r ( u ) ⊗ C for r = 6 , 8 , 12 r=6,8,12 r = 6 , 8 , 12 . At u = e 1 u=e_1 u = e 1 these give the decomposition
W C = C E 1 ⊕ W 6 ( e 1 ) C ⊕ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C , E 1 ⊥ = W 6 ( e 1 ) C ⊕ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C , W_{\mathbb{C}}=\mathbb{C}E_1\oplus W_6(e_1)_{\mathbb{C}}\oplus W_8(e_1)_{\mathbb{C}}\oplus W_{12}(e_1)_{\mathbb{C}},\qquad E_1^\perp=W_6(e_1)_{\mathbb{C}}\oplus W_8(e_1)_{\mathbb{C}}\oplus W_{12}(e_1)_{\mathbb{C}}, W C = C E 1 ⊕ W 6 ( e 1 ) C ⊕ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C , E 1 ⊥ = W 6 ( e 1 ) C ⊕ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C ,
where E 1 ⊥ E_1^\perp E 1 ⊥ is the W-orthogonal complement of E 1 E_1 E 1 . The subscripts 6, 8, 12 indicate the dimensions of the summands. The components W 6 ( e 1 ) C W_6(e_1)_{\mathbb{C}} W 6 ( e 1 ) C and W 12 ( e 1 ) C W_{12}(e_1)_{\mathbb{C}} W 12 ( e 1 ) C are sums of the conjugate irreducible S U ( 3 ) SU(3) S U ( 3 ) -modules 3 ⊕ 3 ‾ 3\oplus\overline{3} 3 ⊕ 3 and 6 ⊕ 6 ‾ 6\oplus\overline{6} 6 ⊕ 6 , respectively. By Proposition 3.7, the double-adjoint identity (75) holds precisely for
( λ , μ , κ ) = ( λ + , λ − , 2 3 ) or ( λ − , λ + , 2 3 ) . (\lambda,\mu,\kappa)=\left(\lambda_+,\lambda_-,\frac{2}{3}\right)\quad\text{or}\quad\left(\lambda_-,\lambda_+,\frac{2}{3}\right). ( λ , μ , κ ) = ( λ + , λ − , 3 2 ) or ( λ − , λ + , 3 2 ) .
For either choice, let S = S λ μ κ S=S_{\lambda\mu\kappa} S = S λ μ κ denote the Koecher operators of Corollary 3.16. By Definition 3.1 and (3.6), every X ∈ stab ( N λ ) X\in\operatorname{stab}(N_\lambda) X ∈ stab ( N λ ) satisfies
0 = N λ ( X h , h , h ) = ⟨ F λ ( h ) , X h ⟩ W ( h ∈ W C ) . 0=N_\lambda(Xh,h,h)=\langle F_\lambda(h),Xh\rangle_W\qquad(h\in W_{\mathbb{C}}). 0 = N λ ( X h , h , h ) = ⟨ F λ ( h ) , X h ⟩ W ( h ∈ W C ) .
Taking h = E 1 h=E_1 h = E 1 and using F λ ( E 1 ) = ( 5 + 2 λ ) E 1 F_\lambda(E_1)=(5+2\lambda)E_1 F λ ( E 1 ) = ( 5 + 2 λ ) E 1 shows that X E 1 ∈ E 1 ⊥ XE_1\in E_1^\perp X E 1 ∈ E 1 ⊥ whenever 5 + 2 λ ≠ 0 5+2\lambda\ne0 5 + 2 λ = 0 , in particular at either critical parameter. We therefore have the evaluation map
ev E 1 : stab ( N λ ) ⟶ E 1 ⊥ , X ⟼ X E 1 . (96) \operatorname{ev}_{E_1}:\operatorname{stab}(N_\lambda)\longrightarrow E_1^\perp,\qquad X\longmapsto XE_1. \tag*{(96)} ev E 1 : stab ( N λ ) ⟶ E 1 ⊥ , X ⟼ X E 1 . (96)
Proposition 3.18. Let λ ∈ { λ + , λ − } \lambda\in\{\lambda_+,\lambda_-\} λ ∈ { λ + , λ − } , let μ \mu μ be the other critical parameter, and let S = S λ μ κ S=S_{\lambda\mu\kappa} S = S λ μ κ with κ = 2 3 \kappa=\frac{2}{3} κ = 3 2 . For the critical cubic N λ N_\lambda N λ , the evaluation map ev E 1 \operatorname{ev}_{E_1} ev E 1 in (3.32) is surjective. Moreover, for every 0 ≠ h ∈ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C 0\ne h\in W_8(e_1)_{\mathbb{C}}\oplus W_{12}(e_1)_{\mathbb{C}} 0 = h ∈ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C , the operator S ( h , E 1 ) S(h,E_1) S ( h , E 1 ) belongs to stab ( N λ ) \operatorname{stab}(N_\lambda) stab ( N λ ) but not to ρ ( s o ( 7 , C ) ) \rho(\mathfrak{so}(7,\mathbb{C})) ρ ( so ( 7 , C )) .
The span of the Koecher operators contains a G 2 G_2 G 2 -submodule τ ≅ V ( 1 , 1 ) \tau\cong V(1,1) τ ≅ V ( 1 , 1 ) , and
ρ ( g 2 C ) ⊕ τ ⊆ stab ( N λ ) , dim stab ( N λ ) ≥ 14 + 64 = 78. \rho(\mathfrak{g}_2^{\mathbb{C}})\oplus\tau\subseteq\operatorname{stab}(N_\lambda),\qquad\dim\operatorname{stab}(N_\lambda)\ge14+64=78. ρ ( g 2 C ) ⊕ τ ⊆ stab ( N λ ) , dim stab ( N λ ) ≥ 14 + 64 = 78. Equality is established in Section 3.7.
Proof. By Corollary 3.16, the Koecher operators S ( h 1 , h 2 ) S(h_1,h_2) S ( h 1 , h 2 ) of each critical cubic N λ N_\lambda N λ belong to stab ( N λ ) \operatorname{stab}(N_\lambda) stab ( N λ ) . Since the evaluation map ev E 1 \operatorname{ev}_{E_1} ev E 1 takes values in E 1 ⊥ E_1^\perp E 1 ⊥ by (3.32), it suffices to show that the elements S ( h , E 1 ) E 1 S(h,E_1)E_1 S ( h , E 1 ) E 1 , for h ∈ W C h\in W_{\mathbb{C}} h ∈ W C , span E 1 ⊥ E_1^\perp E 1 ⊥ .
We begin by computing the action of F λ ( ⋅ , E 1 ) F_\lambda(\,\cdot\,,E_1) F λ ( ⋅ , E 1 ) on the four summands of the S U ( 3 ) SU(3) S U ( 3 ) -decomposition of W C W_{\mathbb{C}} W C . These are preserved by A ( ⋅ , E 1 ) A(\,\cdot\,,E_1) A ( ⋅ , E 1 ) and B ( ⋅ , E 1 ) B(\,\cdot\,,E_1) B ( ⋅ , E 1 ) , which act by the scalars listed in Table 2 . The first row follows from (3.16). For h ∈ E 1 ⊥ h\in E_1^\perp h ∈ E 1 ⊥ , we have h 11 = 0 h_{11}=0 h 11 = 0 , and
Table 2. Eigenvalues of A ( ⋅ , E 1 ) A(\mathord{\cdot},E_1) A ( ⋅ , E 1 ) , B ( ⋅ , E 1 ) B(\mathord{\cdot},E_1) B ( ⋅ , E 1 ) and F λ ( ⋅ , E 1 ) F_\lambda(\mathord{\cdot},E_1) F λ ( ⋅ , E 1 ) on the S U ( 3 ) SU(3) S U ( 3 ) -summands of W C W_{\mathbb C} W C , at E 1 = 7 P 1 − g E_1=7P_1-g E 1 = 7 P 1 − g .
A ( h , E 1 ) = 7 2 ( h P 1 + P 1 h ) − h , B ( h , E 1 ) = 7 P ( h , P 1 ) − P ( h , g ) . A(h,E_1)=\frac{7}{2}(hP_1+P_1h)-h,\qquad B(h,E_1)=7\mathcal{P}(h,P_1)-\mathcal{P}(h,g). A ( h , E 1 ) = 2 7 ( h P 1 + P 1 h ) − h , B ( h , E 1 ) = 7 P ( h , P 1 ) − P ( h , g ) .
The remaining entries follow from P ( h , g ) = − h \mathcal{P}(h,g)=-h P ( h , g ) = − h , by (2.34b), and from P ( h , P 1 ) = h ( Θ 1 ⋅ , Θ 1 ⋅ ) \mathcal{P}(h,P_1)=h(\Theta_1\cdot,\Theta_1\cdot) P ( h , P 1 ) = h ( Θ 1 ⋅ , Θ 1 ⋅ ) , which in the standard orthonormal frame is the matrix − Θ 1 h Θ 1 -\Theta_1h\Theta_1 − Θ 1 h Θ 1 .
On each sector, we thus have F λ ( h , E 1 ) = f ( λ ) h F_\lambda(h,E_1)=f(\lambda)h F λ ( h , E 1 ) = f ( λ ) h , where f ( λ ) f(\lambda) f ( λ ) is the entry of the last column for that sector. Applying F μ F_\mu F μ gives F μ ( F λ ( h , E 1 ) , E 1 ) = f ( μ ) f ( λ ) h F_\mu(F_\lambda(h,E_1),E_1)=f(\mu)f(\lambda)h F μ ( F λ ( h , E 1 ) , E 1 ) = f ( μ ) f ( λ ) h . At a triple satisfying (3.15), we have λ + μ = 1 4 \lambda+\mu=\frac{1}{4} λ + μ = 4 1 and λ μ = 1 8 \lambda\mu=\frac{1}{8} λ μ = 8 1 . Direct computation of the eigenvalues of F μ ( ⋅ , E 1 ) ∘ F λ ( ⋅ , E 1 ) F_\mu(\mathord{\cdot},E_1)\circ F_\lambda(\mathord{\cdot},E_1) F μ ( ⋅ , E 1 ) ∘ F λ ( ⋅ , E 1 ) at these values therefore gives 7 on each of W 6 ( e 1 ) C W_6(e_1)_{\mathbb C} W 6 ( e 1 ) C , W 8 ( e 1 ) C W_8(e_1)_{\mathbb C} W 8 ( e 1 ) C and W 12 ( e 1 ) C W_{12}(e_1)_{\mathbb C} W 12 ( e 1 ) C , and 28 on C E 1 CE_1 C E 1 . Consequently,
F μ ( F λ ( h , E 1 ) , E 1 ) = { 7 h , h ∈ E 1 ⊥ , 28 h , h ∈ C E 1 . F_\mu(F_\lambda(h,E_1),E_1)=
\begin{cases}
7h, & h\in E_1^\perp,\\
28h, & h\in CE_1.
\end{cases} F μ ( F λ ( h , E 1 ) , E 1 ) = { 7 h , 28 h , h ∈ E 1 ⊥ , h ∈ C E 1 .
Substituting this into the definition of the Koecher operators (92) , with κ = 2 3 \kappa=\frac{2}{3} κ = 3 2 and ⟨ E 1 , E 1 ⟩ W = 42 \langle E_1,E_1\rangle_W=42 ⟨ E 1 , E 1 ⟩ W = 42 , gives for h ∈ E 1 ⊥ h\in E_1^\perp h ∈ E 1 ⊥
S ( h , E 1 ) E 1 = 42 h − 2 ( 7 h ) = 28 h S(h,E_1)E_1=42h-2(7h)=28h S ( h , E 1 ) E 1 = 42 h − 2 ( 7 h ) = 28 h
while
S ( E 1 , E 1 ) E 1 = ( 42 − 2 ⋅ 28 + 14 ) E 1 = 0. S(E_1,E_1)E_1=(42-2\cdot28+14)E_1=0. S ( E 1 , E 1 ) E 1 = ( 42 − 2 ⋅ 28 + 14 ) E 1 = 0.
By Corollary 3.16 , every S ( h , E 1 ) S(h,E_1) S ( h , E 1 ) belongs to stab ( N λ ) \operatorname{stab}(N_\lambda) stab ( N λ ) . Hence the evaluation map has image exactly E 1 ⊥ E_1^\perp E 1 ⊥ , of dimension 26.
The same computation also yields symmetries outside ρ ( s o ( 7 , C ) ) \rho(\mathfrak{so}(7,\mathbb{C})) ρ ( so ( 7 , C )) . Indeed, for M ∈ s o ( 7 , C ) M\in\mathfrak{so}(7,\mathbb{C}) M ∈ so ( 7 , C ) ,
ρ ( M ) E 1 = 7 ( ( M e 1 ) ♭ ⊗ e 1 ♭ + e 1 ♭ ⊗ ( M e 1 ) ♭ ) ∈ W 6 ( e 1 ) C . \rho(M)E_1=7\left((Me_1)^\flat\otimes e_1^\flat+e_1^\flat\otimes(Me_1)^\flat\right)\in W_6(e_1)_{\mathbb C}. ρ ( M ) E 1 = 7 ( ( M e 1 ) ♭ ⊗ e 1 ♭ + e 1 ♭ ⊗ ( M e 1 ) ♭ ) ∈ W 6 ( e 1 ) C .
On the other hand, S ( h , E 1 ) E 1 = 28 h S(h,E_1)E_1=28h S ( h , E 1 ) E 1 = 28 h for h ∈ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C h\in W_8(e_1)_{\mathbb C}\oplus W_{12}(e_1)_{\mathbb C} h ∈ W 8 ( e 1 ) C ⊕ W 12 ( e 1 ) C . Thus S ( h , E 1 ) ∉ ρ ( s o ( 7 , C ) ) S(h,E_1)\notin\rho(\mathfrak{so}(7,\mathbb{C})) S ( h , E 1 ) ∈ / ρ ( so ( 7 , C )) whenever h ≠ 0 h\ne0 h = 0 .
The G 2 G_2 G 2 -module count. Let K \mathcal{K} K be the linear span of all Koecher operators S ( h 1 , h 2 ) S(h_1,h_2) S ( h 1 , h 2 ) . The map ( h 1 , h 2 ) ↦ S ( h 1 , h 2 ) (h_1,h_2)\mapsto S(h_1,h_2) ( h 1 , h 2 ) ↦ S ( h 1 , h 2 ) is G 2 G_2 G 2 -equivariant, so K \mathcal{K} K is a G 2 G_2 G 2 -submodule of stab ( N λ ) = ker Φ 1 , λ \operatorname{stab}(N_\lambda)=\ker\Phi_{1,\lambda} stab ( N λ ) = ker Φ 1 , λ . Since evaluation at E 1 E_1 E 1 maps K \mathcal{K} K onto E 1 ⊥ E_1^\perp E 1 ⊥ , we have
dim K ≥ 26. \dim\mathcal{K}\ge26. dim K ≥ 26.
We now use the decomposition of End ( W C ) \operatorname{End}(W_{\mathbb C}) End ( W C ) . By Proposition 3.13, none of the five isotypic types
V ( 2 , 0 ) , V ( 0 , 2 ) , V ( 3 , 0 ) , V ( 4 , 0 ) , V ( 2 , 1 ) V(2,0),\quad V(0,2),\quad V(3,0),\quad V(4,0),\quad V(2,1) V ( 2 , 0 ) , V ( 0 , 2 ) , V ( 3 , 0 ) , V ( 4 , 0 ) , V ( 2 , 1 )
can occur in the stabilizer. Of the remaining types, the three smaller ones are V ( 0 , 0 ) V(0,0) V ( 0 , 0 ) , V ( 1 , 0 ) V(1,0) V ( 1 , 0 ) and V ( 0 , 1 ) V(0,1) V ( 0 , 1 ) . Each occurs with multiplicity one in Table 1, so their isotypic components have total dimension
1 + 7 + 14 = 22 < 26. 1+7+14=22<26. 1 + 7 + 14 = 22 < 26.
They cannot account for all of K \mathcal{K} K . The only remaining isotypic type is V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) , of dimension 64. By complete reducibility, K \mathcal{K} K therefore contains an irreducible submodule τ ≅ V ( 1 , 1 ) \tau\cong V(1,1) τ ≅ V ( 1 , 1 ) .
Finally, ρ ( g 2 C ) ≅ V ( 0 , 1 ) \rho(\mathfrak{g}_{2}^{\mathbb{C}}) \cong V(0,1) ρ ( g 2 C ) ≅ V ( 0 , 1 ) is contained in every stabilizer by Lemma 3.2 . Since V ( 0 , 1 ) V(0,1) V ( 0 , 1 ) and V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) are inequivalent, ρ ( g 2 C ) ∩ τ = 0 \rho(\mathfrak{g}_{2}^{\mathbb{C}}) \cap\tau= 0 ρ ( g 2 C ) ∩ τ = 0 . Hence
ρ ( g 2 C ) ⊕ τ ⊆ stab ( N λ ) and dim stab ( N λ ) ≥ 14 + 64 = 78. \rho(\mathfrak{g}_{2}^{\mathbb{C}}) \oplus\tau\subseteq\operatorname{stab}(N_{\lambda}) \qquad\text{and} \qquad\dim\operatorname{stab}(N_{\lambda}) \geq14 + 64 = 78. ρ ( g 2 C ) ⊕ τ ⊆ stab ( N λ ) and dim stab ( N λ ) ≥ 14 + 64 = 78.
Remark 3.19. For the Albert norm, the Lie algebra of infinitesimal similarities has dimension 78 modulo scalar operators [57 14.24–14.25] , [51 §1.8] . The surjectivity of evaluation at E 1 E_1 E 1 also has a classical counterpart at the unit of the Albert algebra [57 proof of 5.5 and 14.12] . Although the dimension bound 78 is classical for the Albert norm, we obtain the lower bound 78 directly from the double-adjoint identity, without first identifying the critical cubic with the Albert norm and without any classification result. The G 2 G_2 G 2 -isotypic count also identifies a 64-dimensional constituent of type V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) alongside ρ ( g 2 C ) \rho(\mathfrak{g}_{2}^{\mathbb{C}}) ρ ( g 2 C ) .
Generator stabilizers and a model of the 64
The exact count needs two further ingredients: the stabilizers of C 0 C_0 C 0 and C φ C_\varphi C φ , and a concrete realization of the two copies of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) . Although W = S 0 2 V ∗ W = S_0^2 V^* W = S 0 2 V ∗ consists of symmetric tensors on V V V , a general endomorphism L ∈ End ( W ) L \in\operatorname{End}(W) L ∈ End ( W ) need not arise from an endomorphism of V V V . In particular, when determining the full infinitesimal stabilizer of a cubic on W W W , we cannot initially assume that L = ρ ( M ) L = \rho(M) L = ρ ( M ) for some M ∈ End V M \in\operatorname{End} V M ∈ End V . Lemma 3.20 gives a criterion for this reduction. In Propositions 3.21 and 3.23 , we first show that the stabilizer condition implies this criterion, and only then identify the corresponding skew-adjoint endomorphism of V V V .
Let F ∈ { R , C } \mathbb{F} \in\{\mathbb{R}, \mathbb{C}\} F ∈ { R , C } . As in (17) , for u ∈ V u \in V u ∈ V let
E u = 7 P u − g ( u , u ) g ∈ W . E_u = 7P_u - g(u,u)g \in W. E u = 7 P u − g ( u , u ) g ∈ W .
For L ∈ End ( W F ) L \in\operatorname{End}(W_{\mathbb{F}}) L ∈ End ( W F ) , let l ( u ) = g − 1 ( L E u ) ∈ End ( V F ) l(u) = g^{-1}(LE_u) \in\operatorname{End}(V_{\mathbb{F}}) l ( u ) = g − 1 ( L E u ) ∈ End ( V F ) be the endomorphism corresponding to L E u LE_u L E u under the metric identification (13) . Thus
g ( l ( u ) v , w ) = ( L E u ) ( v , w ) , v , w ∈ V F . g(l(u)v,w) = (LE_u)(v,w), \qquad v,w \in V_{\mathbb{F}}. g ( l ( u ) v , w ) = ( L E u ) ( v , w ) , v , w ∈ V F .
For u , v ∈ V u,v \in V u , v ∈ V we define the test form β \beta β by
β ( u , v ) = 1 7 ⟨ L E u , E v ⟩ W = ( L E u ) ( v , v ) = g ( l ( u ) v , v ) , \beta(u,v) = \frac{1}{7}\langle LE_u,E_v\rangle_W = (LE_u)(v,v) = g(l(u)v,v), β ( u , v ) = 7 1 ⟨ L E u , E v ⟩ W = ( L E u ) ( v , v ) = g ( l ( u ) v , v ) ,
where 1 7 ⟨ L E u , E v ⟩ W = ( L E u ) ( v , v ) \frac{1}{7}\langle LE_u,E_v\rangle_W = (LE_u)(v,v) 7 1 ⟨ L E u , E v ⟩ W = ( L E u ) ( v , v ) because L E u LE_u L E u is symmetric and traceless. By Lemma 3.20 , β \beta β vanishes on orthogonal pairs exactly for L ∈ ρ ( s o ( 7 , F ) ) L \in\rho(\mathfrak{so}(7,\mathbb{F})) L ∈ ρ ( so ( 7 , F )) .
The following lemma characterizes the endomorphisms of W F W_{\mathbb{F}} W F that arise from the action of s o ( 7 , F ) \mathfrak{so}(7,\mathbb{F}) so ( 7 , F ) . Recall from (26) that this Lie algebra acts on W F W_{\mathbb{F}} W F by ρ \rho ρ and that the operators ρ ( M ) \rho(M) ρ ( M ) are W W W -skew-adjoint. Although stated for dimension 7, the lemma and its proof apply to S 0 2 ( F n ) S_0^2(\mathbb{F}^n) S 0 2 ( F n ) for every n ≥ 2 n \geq2 n ≥ 2 .
Lemma 3.20 (Rank-one recognition).
(i) The elements E u E_u E u , for u ∈ V u \in V u ∈ V , span W F W_{\mathbb{F}} W F .
(ii) An endomorphism L ∈ End ( W F ) L \in\operatorname{End}(W_{\mathbb{F}}) L ∈ End ( W F ) lies in ρ ( s o ( 7 , F ) ) \rho(\mathfrak{so}(7,\mathbb{F})) ρ ( so ( 7 , F )) if and only if β ( u , v ) = 0 \beta(u,v)=0 β ( u , v ) = 0 for all u , v ∈ V u,v \in V u , v ∈ V with g ( u , v ) = 0 g(u,v)=0 g ( u , v ) = 0 . In that case, L = ρ ( M ) L=\rho(M) L = ρ ( M ) for a unique M ∈ s o ( 7 , F ) M \in\mathfrak{so}(7,\mathbb{F}) M ∈ so ( 7 , F ) .
The test vectors u , v u,v u , v are real even when F = C \mathbb{F}=\mathbb{C} F = C .
Proof. For (i), the identities
E i − E j = 7 ( e i b ⊗ e i b − e j b ⊗ e j b ) , E_i-E_j=7(e_i^b\otimes e_i^b-e_j^b\otimes e_j^b), E i − E j = 7 ( e i b ⊗ e i b − e j b ⊗ e j b ) ,
E e i + e j − E i − E j = 7 ( e i b ⊗ e j b + e j b ⊗ e i b ) E_{e_i+e_j}-E_i-E_j=7(e_i^b\otimes e_j^b+e_j^b\otimes e_i^b) E e i + e j − E i − E j = 7 ( e i b ⊗ e j b + e j b ⊗ e i b )
for i ≠ j i\ne j i = j give a spanning set of W F W_{\mathbb{F}} W F .
For (ii), if L = ρ ( M ) L=\rho(M) L = ρ ( M ) with M ∈ s o ( 7 , F ) M\in\mathfrak{so}(7,\mathbb{F}) M ∈ so ( 7 , F ) , then
L E u = 7 ( u b ⊗ ( M u ) b + ( M u ) b ⊗ u b ) , β ( u , v ) = 14 g ( u , v ) g ( M u , v ) . LE_u=7(u^b\otimes(Mu)^b+(Mu)^b\otimes u^b),\qquad\beta(u,v)=14g(u,v)g(Mu,v). L E u = 7 ( u b ⊗ ( M u ) b + ( M u ) b ⊗ u b ) , β ( u , v ) = 14 g ( u , v ) g ( M u , v ) .
so β \beta β vanishes on orthogonal pairs.
Conversely, suppose that β ( u , v ) = 0 \beta(u,v)=0 β ( u , v ) = 0 whenever g ( u , v ) = 0 g(u,v)=0 g ( u , v ) = 0 . For fixed real u ≠ 0 u\ne0 u = 0 , by polarization, L E u LE_u L E u vanishes as a bilinear form on u ⊥ u^\perp u ⊥ . Since L E u LE_u L E u is traceless, there is a unique w ( u ) ∈ V F w(u)\in V_{\mathbb{F}} w ( u ) ∈ V F , orthogonal to u u u , such that
L E u = 7 ( u b ⊗ w ( u ) b + w ( u ) b ⊗ u b ) . LE_u=7(u^b\otimes w(u)^b+w(u)^b\otimes u^b). L E u = 7 ( u b ⊗ w ( u ) b + w ( u ) b ⊗ u b ) .
Each diagonal entry d i ( u ) = ( L E u ) ( e i , e i ) d_i(u)=(LE_u)(e_i,e_i) d i ( u ) = ( L E u ) ( e i , e i ) is a homogeneous quadratic polynomial vanishing when u i = 0 u_i=0 u i = 0 , since then e i ∈ u ⊥ e_i\in u^\perp e i ∈ u ⊥ . Dividing by u i u_i u i therefore gives a linear polynomial, so there is a matrix M M M , independent of u u u , with d i ( u ) = 14 u i ( M u ) i d_i(u)=14u_i(Mu)_i d i ( u ) = 14 u i ( M u ) i . Comparing with d i ( u ) = 14 u i w i ( u ) d_i(u)=14u_iw_i(u) d i ( u ) = 14 u i w i ( u ) gives w ( u ) = M u w(u)=Mu w ( u ) = M u whenever all coordinates of u u u are nonzero. Both sides of
L E u = 7 ( u b ⊗ ( M u ) b + ( M u ) b ⊗ u b ) LE_u=7(u^b\otimes(Mu)^b+(Mu)^b\otimes u^b) L E u = 7 ( u b ⊗ ( M u ) b + ( M u ) b ⊗ u b )
are polynomial in u u u , so the identity holds for every real u u u . Taking traces gives 14 g ( M u , u ) = 0 14g(Mu,u)=0 14 g ( M u , u ) = 0 for all real u u u , hence M M M is g g g -skew-adjoint. Thus L E u = ρ ( M ) E u LE_u=\rho(M)E_u L E u = ρ ( M ) E u ; since the E u E_u E u span W F W_{\mathbb{F}} W F by (i), L = ρ ( M ) L=\rho(M) L = ρ ( M ) , and the injectivity of ρ \rho ρ gives uniqueness. The polynomial argument applies over either coefficient field with the test vectors kept real.
The stabilizer of the trace cubic is already known. Garibaldi and Guralnick showed that, for n = 3 n=3 n = 3 and n ≥ 5 n\ge5 n ≥ 5 , the identity component of the stabilizer of tr ( h 3 ) \operatorname{tr}(h^3) tr ( h 3 ) on S 0 2 ( C n ) S_0^2(\mathbb{C}^n) S 0 2 ( C n ) is the image of S O ( n , C ) SO(n,\mathbb{C}) S O ( n , C ) [21 ] , proof of Theorem 8.1. Their proof uses Seitz’s classification. We give a direct proof of the corresponding infinitesimal statement for n = 7 n=7 n = 7 .
Both generator stabilizers can be found by the same rank-one test. For L L L annihilating the cubic, evaluation on ( E u , E v , E v ) (E_u,E_v,E_v) ( E u , E v , E v ) first forces β ( v , v ) = 0 \beta(v,v)=0 β ( v , v ) = 0 . Evaluations on orthonormal pairs, supplemented for C φ C_\varphi C φ by their cross products, then force β ( u , v ) = 0 \beta(u,v)=0 β ( u , v ) = 0 whenever u ⊥ v u\perp v u ⊥ v . Lemma 3.20 gives L = ρ ( M ) L=\rho(M) L = ρ ( M ) for a skew endomorphism M M M of V V V . Every such M M M preserves C 0 C_0 C 0 ; for C φ C_\varphi C φ , the evaluation formula of Lemma 3.22 will identify those that also preserve φ \varphi φ . In Propositions 3.21 and 3.23 below, the line stabilizer is obtained from the kernel by adding multiples of id W \operatorname{id}_W id W : for any nonzero cubic N N N ,
Φ N ( id W ) = 3 N , { X : Φ N ( X ) ∈ F N } = ker Φ N ⊕ F id W . (97) \Phi_N(\operatorname{id}_W)=3N,\qquad\{X:\Phi_N(X)\in\mathbb{F}N\}=\ker\Phi_N\oplus\mathbb{F}\operatorname{id}_W. \tag*{(97)} Φ N ( id W ) = 3 N , { X : Φ N ( X ) ∈ F N } = ker Φ N ⊕ F id W . (97)
Indeed, Φ N ( X ) = c N \Phi_N(X)=cN Φ N ( X ) = c N if and only if X − c 3 id W ∈ ker Φ N X-\frac{c}{3}\operatorname{id}_W\in\ker\Phi_N X − 3 c id W ∈ ker Φ N .
Proposition 3.21. For F ∈ { R , C } \mathbb{F} \in\{\mathbb{R},\mathbb{C}\} F ∈ { R , C } , ker Φ C 0 = ρ ( s o ( 7 , F ) ) \ker\Phi_{C_0} = \rho(\mathfrak{so}(7,\mathbb{F})) ker Φ C 0 = ρ ( so ( 7 , F )) , of dimension 21 21 21 over F \mathbb{F} F , and the stabilizer { X ∈ End ( W F ) : Φ C 0 ( X ) ∈ F C 0 } \{X \in\operatorname{End}(W_{\mathbb{F}}) : \Phi_{C_0}(X) \in\mathbb{F}C_0\} { X ∈ End ( W F ) : Φ C 0 ( X ) ∈ F C 0 } of the line of C 0 C_0 C 0 has dimension 22 22 22 .
Proof. For M ∈ s o ( 7 , F ) M \in\mathfrak{so}(7,\mathbb{F}) M ∈ so ( 7 , F ) , the action is the matrix commutator ρ ( M ) h = [ M , h ] \rho(M)h = [M,h] ρ ( M ) h = [ M , h ] by (26) . Cyclicity of the trace gives
1 3 Φ C 0 ( ρ ( M ) ) ( h , h , h ) = tr ( [ M , h ] h 2 ) = tr ( M h 3 ) − tr ( h M h 2 ) = 0. \frac{1}{3}\Phi_{C_0}(\rho(M))(h,h,h) = \operatorname{tr}([M,h]h^2) = \operatorname{tr}(Mh^3) - \operatorname{tr}(hMh^2) = 0. 3 1 Φ C 0 ( ρ ( M )) ( h , h , h ) = tr ([ M , h ] h 2 ) = tr ( M h 3 ) − tr ( h M h 2 ) = 0.
Polarization therefore gives ρ ( s o ( 7 , F ) ) ⊆ ker Φ C 0 \rho(\mathfrak{so}(7,\mathbb{F})) \subseteq\ker\Phi_{C_0} ρ ( so ( 7 , F )) ⊆ ker Φ C 0 .
Conversely, let L ∈ ker Φ C 0 L \in\ker\Phi_{C_0} L ∈ ker Φ C 0 . Using the contracted product (12) , we have
C 0 ( h 1 , h 2 , h 3 ) = ⟨ h 1 , 1 2 ( h 2 h 3 + h 3 h 2 ) 0 ⟩ W , C_0(h_1,h_2,h_3) = \left\langle h_1,\frac{1}{2}(h_2h_3+h_3h_2)_0\right\rangle_{W}, C 0 ( h 1 , h 2 , h 3 ) = ⟨ h 1 , 2 1 ( h 2 h 3 + h 3 h 2 ) 0 ⟩ W ,
and the definition of E u E_u E u gives
E v 2 = 35 g ( v , v ) P v + g ( v , v ) 2 g , E_v^2 = 35g(v,v)P_v + g(v,v)^2g, E v 2 = 35 g ( v , v ) P v + g ( v , v ) 2 g ,
1 2 ( E u E v + E v E u ) = 49 2 g ( u , v ) ( u ♭ ⊗ v ♭ + v ♭ ⊗ u ♭ ) − 7 g ( v , v ) P u \frac{1}{2}(E_uE_v+E_vE_u) = \frac{49}{2}g(u,v)(u^\flat\otimes v^\flat+v^\flat\otimes u^\flat)-7g(v,v)P_u 2 1 ( E u E v + E v E u ) = 2 49 g ( u , v ) ( u ♭ ⊗ v ♭ + v ♭ ⊗ u ♭ ) − 7 g ( v , v ) P u
− 7 g ( u , u ) P v + g ( u , u ) g ( v , v ) g . \qquad-7g(u,u)P_v+g(u,u)g(v,v)g. − 7 g ( u , u ) P v + g ( u , u ) g ( v , v ) g .
Contracting with the traceless tensors L E u LE_u L E u and L E v LE_v L E v yields
0 = Φ C 0 ( L ) ( E u , E v , E v ) = C 0 ( L E u , E v , E v ) + 2 C 0 ( L E v , E u , E v ) = 35 g ( v , v ) β ( u , v ) + 98 g ( u , v ) ( L E v ) ( u , v ) − 14 g ( v , v ) β ( v , u ) − 14 g ( u , u ) β ( v , v ) . (98) \begin{aligned}
0 &= \Phi_{C_0}(L)(E_u,E_v,E_v) \\
&= C_0(LE_u,E_v,E_v)+2C_0(LE_v,E_u,E_v) \\
&= 35g(v,v)\beta(u,v)+98g(u,v)(LE_v)(u,v) \\
&\qquad-14g(v,v)\beta(v,u)-14g(u,u)\beta(v,v). \tag*{(98)}
\end{aligned} 0 = Φ C 0 ( L ) ( E u , E v , E v ) = C 0 ( L E u , E v , E v ) + 2 C 0 ( L E v , E u , E v ) = 35 g ( v , v ) β ( u , v ) + 98 g ( u , v ) ( L E v ) ( u , v ) − 14 g ( v , v ) β ( v , u ) − 14 g ( u , u ) β ( v , v ) . (98)
At u = v u=v u = v , this becomes 105 g ( v , v ) β ( v , v ) = 0 105g(v,v)\beta(v,v)=0 105 g ( v , v ) β ( v , v ) = 0 . Since g g g is positive definite, β ( v , v ) = 0 \beta(v,v)=0 β ( v , v ) = 0 for every real v v v . For orthonormal u , v u,v u , v , (98) and the same identity with u u u and v v v exchanged give
5 β ( u , v ) = 2 β ( v , u ) , 5 β ( v , u ) = 2 β ( u , v ) , 5\beta(u,v)=2\beta(v,u), \qquad5\beta(v,u)=2\beta(u,v), 5 β ( u , v ) = 2 β ( v , u ) , 5 β ( v , u ) = 2 β ( u , v ) ,
so both values vanish. By homogeneity, β \beta β vanishes on all real orthogonal pairs. Lemma 3.20 now gives L ∈ ρ ( s o ( 7 , F ) ) L \in\rho(\mathfrak{so}(7,\mathbb{F})) L ∈ ρ ( so ( 7 , F )) , proving the equality. The kernel has dimension 21 21 21 by the injectivity of ρ \rho ρ , and by (97) , the stabilizer of the line has dimension 22 22 22 . □ \square □
By (27) and the equivariance and injectivity of ρ \rho ρ , ker Φ C 0 = ρ ( Λ 7 2 ) ⊕ ρ ( Λ 14 2 ) ≅ 7 ⊕ 14 \ker\Phi_{C_0} = \rho(\Lambda^2_7) \oplus\rho(\Lambda^2_{14}) \cong7 \oplus14 ker Φ C 0 = ρ ( Λ 7 2 ) ⊕ ρ ( Λ 14 2 ) ≅ 7 ⊕ 14 as a G 2 G_2 G 2 -module.
The φ \varphi φ -cubic has an equally explicit value on the same test tensors E u E_u E u .
Lemma 3.22. Let φ \varphi φ be the standard G 2 G_2 G 2 3-form, or more generally any positive 3-form on V V V compatible with g g g , obtained by transporting the standard form by an isometry of either orientation. Let C φ C_\varphi C φ be defined as in (33) , and let E u = 7 P u − g ( u , u ) g E_u = 7P_u-g(u,u)g E u = 7 P u − g ( u , u ) g be as in (17) .
(i) For all u , v , w ∈ V u,v,w \in V u , v , w ∈ V ,
C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 + 49 ( g ( u , u ) g ( v , w ) 2 + g ( v , v ) g ( u , w ) 2 + g ( w , w ) g ( u , v ) 2 ) − 63 g ( u , u ) g ( v , v ) g ( w , w ) . (99) \begin{aligned}
C_\varphi(E_u,E_v,E_w) &= 343\varphi(u,v,w)^2 \\
&\quad+49\bigl(g(u,u)g(v,w)^2+g(v,v)g(u,w)^2+g(w,w)g(u,v)^2\bigr) \\
&\quad-63g(u,u)g(v,v)g(w,w). \tag*{(99)}
\end{aligned} C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 + 49 ( g ( u , u ) g ( v , w ) 2 + g ( v , v ) g ( u , w ) 2 + g ( w , w ) g ( u , v ) 2 ) − 63 g ( u , u ) g ( v , v ) g ( w , w ) . (99)
(ii) The cubic C φ C_\varphi C φ determines φ \varphi φ up to sign. More precisely, if φ ′ \varphi' φ ′ is another such form with C φ ′ = C φ C_{\varphi'}=C_\varphi C φ ′ = C φ , then φ ′ = ± φ \varphi'=\pm\varphi φ ′ = ± φ . If the two forms induce the same orientation, then φ ′ = φ \varphi'=\varphi φ ′ = φ .
(iii) The subgroup of O ( 7 ) O(7) O ( 7 ) whose induced action on W W W preserves C φ C_\varphi C φ is { ± I } G 2 \{\pm I\}G_2 { ± I } G 2 , where G 2 G_2 G 2 is the stabilizer of φ \varphi φ . Since ρ ( − I ) = id W \rho(-I)=\operatorname{id}_W ρ ( − I ) = id W , its image in G L ( W ) GL(W) G L ( W ) is precisely ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) .
For orthonormal u , v , w u,v,w u , v , w , (3.35) reduces to
C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 − 63 , C_\varphi(E_u,E_v,E_w)=343\varphi(u,v,w)^2-63, C φ ( E u , E v , E w ) = 343 φ ( u , v , w ) 2 − 63 ,
recovering the identity used in the proof of Lemma 2.7.
Proof. (i) Evaluation on test tensors. Both sides of (3.35) are invariant under simultaneous orthogonal transport of φ \varphi φ and u , v , w u,v,w u , v , w , so it suffices to consider the standard form. Recall, from (2.13) and (2.14),
P u = u ♭ ⊗ u ♭ , E u = 7 P u − g ( u , u ) g . P_u=u^\flat\otimes u^\flat,\qquad E_u=7P_u-g(u,u)g. P u = u ♭ ⊗ u ♭ , E u = 7 P u − g ( u , u ) g .
We extend C φ C_\varphi C φ to all symmetric 2-tensors using the contraction formula (2.27). The rank-one and metric contraction rules (2.34a) and (2.34b) give
C φ ( P u , P v , P w ) = φ ( u , v , w ) 2 , C φ ( g , P v , P w ) = g ( v , v ) g ( w , w ) − g ( v , w ) 2 , C φ ( g , g , P w ) = 6 g ( w , w ) , C φ ( g , g , g ) = 42. \begin{aligned}
C_\varphi(P_u,P_v,P_w)&=\varphi(u,v,w)^2,\\
C_\varphi(g,P_v,P_w)&=g(v,v)g(w,w)-g(v,w)^2,\\
C_\varphi(g,g,P_w)&=6g(w,w),\\
C_\varphi(g,g,g)&=42.
\end{aligned} C φ ( P u , P v , P w ) C φ ( g , P v , P w ) C φ ( g , g , P w ) C φ ( g , g , g ) = φ ( u , v , w ) 2 , = g ( v , v ) g ( w , w ) − g ( v , w ) 2 , = 6 g ( w , w ) , = 42.
The first identity follows from g ( u × v , w ) = φ ( u , v , w ) g(u\times v,w)=\varphi(u,v,w) g ( u × v , w ) = φ ( u , v , w ) . Expanding C φ ( 7 P u − g ( u , u ) g , 7 P v − g ( v , v ) g , 7 P w − g ( w , w ) g ) C_\varphi(7P_u-g(u,u)g,7P_v-g(v,v)g,7P_w-g(w,w)g) C φ ( 7 P u − g ( u , u ) g , 7 P v − g ( v , v ) g , 7 P w − g ( w , w ) g ) using these identities gives (3.35).
(ii) Recovery of the 3-form. Suppose C φ ′ = C φ C_{\varphi'}=C_\varphi C φ ′ = C φ . Since both forms are compatible with g g g , the metric terms in (3.35) agree. Hence
φ ′ ( u , v , w ) 2 = φ ( u , v , w ) 2 ( u , v , w ∈ V ) . \varphi'(u,v,w)^2=\varphi(u,v,w)^2\qquad(u,v,w\in V). φ ′ ( u , v , w ) 2 = φ ( u , v , w ) 2 ( u , v , w ∈ V ) .
Equivalently,
[ ( φ ′ − φ ) ( u , v , w ) ] [ ( φ ′ + φ ) ( u , v , w ) ] = 0. [(\varphi'-\varphi)(u,v,w)][(\varphi'+\varphi)(u,v,w)]=0. [( φ ′ − φ ) ( u , v , w )] [( φ ′ + φ ) ( u , v , w )] = 0.
As an identity in the polynomial ring R [ u , v , w ] \mathbb{R}[u,v,w] R [ u , v , w ] , which is an integral domain, this implies φ ′ = φ \varphi'=\varphi φ ′ = φ or φ ′ = − φ \varphi'=-\varphi φ ′ = − φ . By the normalization in Lemma 2.1, φ \varphi φ and − φ -\varphi − φ induce opposite orientations. Thus the orientation determines the sign.
(iii) Orthogonal stabilizer. For O ∈ O ( 7 ) O\in O(7) O ∈ O ( 7 ) , (2.20) gives ρ ( O ) E u = E O u \rho(O)E_u=E_{Ou} ρ ( O ) E u = E O u . By (3.35), ρ ( O ) \rho(O) ρ ( O ) preserves C φ C_\varphi C φ if and only if
( O ∗ φ ) ( u , v , w ) 2 = φ ( u , v , w ) 2 ( u , v , w ∈ V ) . (O^*\varphi)(u,v,w)^2=\varphi(u,v,w)^2\qquad(u,v,w\in V). ( O ∗ φ ) ( u , v , w ) 2 = φ ( u , v , w ) 2 ( u , v , w ∈ V ) .
By (ii), this is equivalent to O ∗ φ = ± φ O^*\varphi=\pm\varphi O ∗ φ = ± φ . The orientation normalization forces the sign to equal det O \det O det O . Therefore the subgroup in question is
G 2 ∪ ( − I ) G 2 = { ± I } G 2 . G_2\cup(-I)G_2=\{\pm I\}G_2. G 2 ∪ ( − I ) G 2 = { ± I } G 2 .
Finally, ρ ( − I ) = id W \rho(-I)=\operatorname{id}_W ρ ( − I ) = id W , so the image of { ± I } G 2 \{\pm I\}G_2 { ± I } G 2 in G L ( W ) GL(W) G L ( W ) is ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) .
Lemma 3.22 concerns the orthogonal group only. In the next proposition, we consider the full infinitesimal stabilizer of C φ C_\varphi C φ in End ( W ) \operatorname{End}(W) End ( W ) .
Proposition 3.23. Let F ∈ { R , C } \mathbb{F} \in\{\mathbb{R},\mathbb{C}\} F ∈ { R , C } . Recall that θ M \theta_M θ M denotes the infinitesimal pullback of φ \varphi φ by M M M from (2.19a), and that
g 2 F = { M ∈ g l ( V F ) : θ M = 0 } \mathfrak{g}_2^{\mathbb{F}} = \{M \in\mathfrak{gl}(V_{\mathbb{F}}) : \theta_M = 0\} g 2 F = { M ∈ gl ( V F ) : θ M = 0 } is the infinitesimal stabilizer of φ \varphi φ , extended complex-linearly when F = C \mathbb{F} = \mathbb{C} F = C . Then
ker Φ C φ = ρ ( g 2 F ) , dim F ker Φ C φ = 14. \ker\Phi_{C_\varphi} = \rho(\mathfrak{g}_2^{\mathbb{F}}), \qquad\dim_{\mathbb{F}} \ker\Phi_{C_\varphi} = 14. ker Φ C φ = ρ ( g 2 F ) , dim F ker Φ C φ = 14. The infinitesimal stabilizer of the line F C φ \mathbb{F}C_\varphi F C φ is
ρ ( g 2 F ) ⊕ F id W , \rho(\mathfrak{g}_2^{\mathbb{F}}) \oplus\mathbb{F}\operatorname{id}_W, ρ ( g 2 F ) ⊕ F id W , of dimension 15.
Proof. Reduction to skew endomorphisms of V V V . Let L ∈ ker Φ C φ L \in\ker\Phi_{C_\varphi} L ∈ ker Φ C φ . For h ∈ W F h \in W_{\mathbb{F}} h ∈ W F , the identities (2.34a), (2.34b) and (2.35) give
C φ ( h , E u , E v ) = 49 h ( u × v , u × v ) + 7 g ( v , v ) h ( u , u ) + 7 g ( u , u ) h ( v , v ) . C_\varphi(h,E_u,E_v) = 49h(u \times v,u \times v) + 7g(v,v)h(u,u) + 7g(u,u)h(v,v). C φ ( h , E u , E v ) = 49 h ( u × v , u × v ) + 7 g ( v , v ) h ( u , u ) + 7 g ( u , u ) h ( v , v ) .
where h ( u , u ) = h i j u i u j h(u,u)=h_{ij}u^iu^j h ( u , u ) = h ij u i u j . Substituting into 0 = C φ ( L E u , E v , E v ) + 2 C φ ( L E v , E u , E v ) 0=C_\varphi(LE_u,E_v,E_v)+2C_\varphi(LE_v,E_u,E_v) 0 = C φ ( L E u , E v , E v ) + 2 C φ ( L E v , E u , E v ) and dividing by 14 gives
0 = g ( v , v ) [ β ( u , v ) + β ( v , u ) ] + 7 β ( v , u × v ) + g ( u , u ) β ( v , v ) . (100) 0 = g(v,v)\left[\beta(u,v)+\beta(v,u)\right] + 7\beta(v,u\times v) + g(u,u)\beta(v,v). \tag*{(100)} 0 = g ( v , v ) [ β ( u , v ) + β ( v , u ) ] + 7 β ( v , u × v ) + g ( u , u ) β ( v , v ) . (100)
Taking u = v u=v u = v yields 3 g ( v , v ) β ( v , v ) = 0 3g(v,v)\beta(v,v)=0 3 g ( v , v ) β ( v , v ) = 0 , hence β ( v , v ) = 0 \beta(v,v)=0 β ( v , v ) = 0 for every real v v v .
For orthonormal u , v u,v u , v , put w = u × v w=u\times v w = u × v . By (2.7), ( u , v , w ) (u,v,w) ( u , v , w ) is orthonormal with v × w = u v\times w=u v × w = u and w × u = v w\times u=v w × u = v . (100) and the same equation with u u u and v v v exchanged give
β ( u , v ) + β ( v , u ) + 7 β ( v , w ) = 0 , \beta(u,v)+\beta(v,u)+7\beta(v,w)=0, β ( u , v ) + β ( v , u ) + 7 β ( v , w ) = 0 ,
β ( u , v ) + β ( v , u ) + 7 β ( u , w ) = 0 , \beta(u,v)+\beta(v,u)+7\beta(u,w)=0, β ( u , v ) + β ( v , u ) + 7 β ( u , w ) = 0 ,
where β ( u , − w ) = β ( u , w ) \beta(u,-w)=\beta(u,w) β ( u , − w ) = β ( u , w ) . Thus β ( u , w ) = β ( v , w ) \beta(u,w)=\beta(v,w) β ( u , w ) = β ( v , w ) . Applying the same argument cyclically, we may write
b u = β ( v , u ) = β ( w , u ) , b_u=\beta(v,u)=\beta(w,u), b u = β ( v , u ) = β ( w , u ) ,
b v = β ( w , v ) = β ( u , v ) , b_v=\beta(w,v)=\beta(u,v), b v = β ( w , v ) = β ( u , v ) ,
b w = β ( u , w ) = β ( v , w ) . b_w=\beta(u,w)=\beta(v,w). b w = β ( u , w ) = β ( v , w ) .
The three cyclic evaluations of (3.36) become
b u + b v + 7 b w = 0 , b v + b w + 7 b u = 0 , b w + b u + 7 b v = 0. b_u+b_v+7b_w=0,\qquad b_v+b_w+7b_u=0,\qquad b_w+b_u+7b_v=0. b u + b v + 7 b w = 0 , b v + b w + 7 b u = 0 , b w + b u + 7 b v = 0.
Subtracting shows b u = b v = b w b_u=b_v=b_w b u = b v = b w , so all three vanish. Homogeneity now gives β ( u , v ) = 0 \beta(u,v)=0 β ( u , v ) = 0 for all real orthogonal pairs, and Lemma 3.20 gives L = ρ ( M ) L=\rho(M) L = ρ ( M ) for some M ∈ s o ( 7 , F ) M\in\mathfrak{so}(7,\mathbb{F}) M ∈ so ( 7 , F ) .
Identification of g 2 F \mathfrak{g}_2^F g 2 F . For real skew M M M , put O t = e t M O_t=e^{tM} O t = e tM . Since ρ ( O t ) E u = E O t u \rho(O_t)E_u=E_{O_tu} ρ ( O t ) E u = E O t u , differentiating (99) gives
Φ C φ ( ρ ( M ) ) ( E u , E v , E w ) = 343 d d t ∣ t = 0 φ ( O t u , O t v , O t w ) 2 = 686 φ ( u , v , w ) θ M ( u , v , w ) . (101) \begin{aligned}
\Phi_{C_\varphi}(\rho(M))(E_u,E_v,E_w)=343\left.\frac{d}{dt}\right|_{t=0}\varphi(O_tu,O_tv,O_tw)^2 \\
&=686\varphi(u,v,w)\theta_M(u,v,w). \tag*{(101)}
\end{aligned} Φ C φ ( ρ ( M )) ( E u , E v , E w ) = 343 d t d t = 0 φ ( O t u , O t v , O t w ) 2 = 686 φ ( u , v , w ) θ M ( u , v , w ) . (101)
The metric terms have zero derivative because O t O_t O t is orthogonal. By linearity in M M M , the identity extends to complex skew endomorphisms.
If ρ ( M ) \rho(M) ρ ( M ) annihilates C φ C_\varphi C φ , then φ ( u , v , w ) θ M ( u , v , w ) = 0 \varphi(u,v,w)\theta_M(u,v,w)=0 φ ( u , v , w ) θ M ( u , v , w ) = 0 . Hence θ M \theta_M θ M vanishes on the set where φ ≠ 0 \varphi\ne0 φ = 0 , which is dense in V 3 V^3 V 3 , and therefore vanishes everywhere by continuity. Conversely, if θ M = 0 \theta_M=0 θ M = 0 , then (101) vanishes on all triples of test tensors ( E u , E v , E w ) (E_u,E_v,E_w) ( E u , E v , E w ) ; since the E u E_u E u span W F W_F W F by Lemma 3.20 (i), ρ ( M ) ∈ ker Φ C φ \rho(M)\in\ker\Phi_{C_\varphi} ρ ( M ) ∈ ker Φ C φ . This proves ker Φ C φ = ρ ( g 2 F ) \ker\Phi_{C_\varphi}=\rho(\mathfrak{g}_2^F) ker Φ C φ = ρ ( g 2 F ) . Its dimension is 14 by (2.22) and injectivity of ρ \rho ρ ; (3.33) gives the line stabilizer and its dimension 15. □
We next construct the two copies of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in End W C \operatorname{End}W_C End W C explicitly. The same 64-dimensional tensor space will map to both copies, one consisting of W W W -self-adjoint endomorphisms and the other of W W W -skew-adjoint endomorphisms. The tensor model Q 64 Q_{64} Q 64 as used below is standard, and can be found in [7 Lemma 2.1] and [15 (4.27)] .
Definition 3.24. Let Q 64 = ker ( Alt : V ∗ ⊗ Λ 14 2 → Λ 3 V ∗ ) Q_{64}=\ker(\operatorname{Alt}:V^*\otimes\Lambda_{14}^2\to\Lambda^3V^*) Q 64 = ker ( Alt : V ∗ ⊗ Λ 14 2 → Λ 3 V ∗ ) ; equivalently, Q 64 Q_{64} Q 64 is the space of real tensors Q a b c Q_{abc} Q ab c with
Q a b c = − Q a c b , (102) Q_{abc}=-Q_{acb}, \tag*{(102)} Q ab c = − Q a c b , (102) Q a b c φ d b c = 0 , (103) Q_{abc}\varphi^{bc}_d=0, \tag*{(103)} Q ab c φ d b c = 0 , (103) Q a b c + Q b c a + Q c a b = 0 , (104) Q_{abc}+Q_{bca}+Q_{cab}=0, \tag*{(104)} Q ab c + Q b c a + Q c ab = 0 , (104) and Q 64 , C = Q 64 ⊗ C Q_{64,\mathbb{C}}=Q_{64}\otimes\mathbb{C} Q 64 , C = Q 64 ⊗ C . For Q ∈ Q 64 Q\in Q_{64} Q ∈ Q 64 define the bilinear form ε Q \varepsilon_Q ε Q on covariant 2-tensors h , k ∈ V ∗ ⊗ V ∗ h,k\in V^*\otimes V^* h , k ∈ V ∗ ⊗ V ∗ by
ε Q ( h , k ) = Q a b c φ a d c h a d k b e . \varepsilon_Q(h,k)=Q_{abc}\varphi^{c}_{\ ad}h^{ad}k^{be}. ε Q ( h , k ) = Q ab c φ a d c h a d k b e . This is the form through which Q Q Q acts on W W W . Its symmetric and alternating parts are
ε Q s ( h , k ) = 1 2 ( ε Q ( h , k ) + ε Q ( k , h ) ) , ε Q a ( h , k ) = 1 2 ( ε Q ( h , k ) − ε Q ( k , h ) ) . \varepsilon_Q^s(h,k)=\frac{1}{2}\left(\varepsilon_Q(h,k)+\varepsilon_Q(k,h)\right),\qquad
\varepsilon_Q^a(h,k)=\frac{1}{2}\left(\varepsilon_Q(h,k)-\varepsilon_Q(k,h)\right). ε Q s ( h , k ) = 2 1 ( ε Q ( h , k ) + ε Q ( k , h ) ) , ε Q a ( h , k ) = 2 1 ( ε Q ( h , k ) − ε Q ( k , h ) ) . On W W W , let J Q s , J Q a ∈ End W J_Q^s,J_Q^a\in\operatorname{End}W J Q s , J Q a ∈ End W be the corresponding endomorphisms under the W W W -pairing:
⟨ J Q s h , k ⟩ W = ε Q s ( h , k ) , ⟨ J Q a h , k ⟩ W = ε Q a ( h , k ) . (105) \langle J_Q^s h,k\rangle_W=\varepsilon_Q^s(h,k),\qquad
\langle J_Q^a h,k\rangle_W=\varepsilon_Q^a(h,k). \tag*{(105)} ⟨ J Q s h , k ⟩ W = ε Q s ( h , k ) , ⟨ J Q a h , k ⟩ W = ε Q a ( h , k ) . (105) We call Q ↦ J Q s Q\mapsto J_Q^s Q ↦ J Q s and Q ↦ J Q a Q\mapsto J_Q^a Q ↦ J Q a the parity embeddings ; the labels s and a refer to symmetry and skew-symmetry under the W W W -adjoint. All these definitions extend complex-linearly in Q Q Q and complex-bilinearly in h , k h,k h , k . Equivalently, for h ∈ W h\in W h ∈ W , the operators act by
( J Q s h ) i j = 1 4 [ ( Q a i c − Q i a c ) φ d j c + ( Q a j c − Q j a c ) φ d i c ] h a d , (J_Q^s h)_{ij}=\frac{1}{4}\left[(Q_{aic}-Q_{iac})\varphi^c_{\ dj}+(Q_{ajc}-Q_{jac})\varphi^c_{\ di}\right]h^{ad}, ( J Q s h ) ij = 4 1 [ ( Q ai c − Q ia c ) φ d j c + ( Q aj c − Q j a c ) φ d i c ] h a d , ( J Q a h ) i j = 1 4 [ ( Q a i c + Q i a c ) φ d j c + ( Q a j c + Q j a c ) φ d i c ] h a d . (106) (J_Q^a h)_{ij}=\frac{1}{4}\left[(Q_{aic}+Q_{iac})\varphi^c_{\ dj}+(Q_{ajc}+Q_{jac})\varphi^c_{\ di}\right]h^{ad}. \tag*{(106)} ( J Q a h ) ij = 4 1 [ ( Q ai c + Q ia c ) φ d j c + ( Q aj c + Q j a c ) φ d i c ] h a d . (106) The right-hand side is symmetric and traceless in i , j i,j i , j , as follows from the defining identities of Q 64 Q_{64} Q 64 .
Conditions (3.38a) and (3.38b) ensure that each slice Q a = Q ( e a , ⋅ , ⋅ ) Q_a = Q(e_a,\cdot,\cdot) Q a = Q ( e a , ⋅ , ⋅ ) belongs to Λ 14 2 \Lambda^{2}_{14} Λ 14 2 . Since Q Q Q is skew in its last two indices,
Alt ( Q ) a b c = 1 3 ( Q a b c + Q b c a + Q c a b ) , \operatorname{Alt}(Q)_{abc} = \frac{1}{3}(Q_{abc} + Q_{bca} + Q_{cab}), Alt ( Q ) ab c = 3 1 ( Q ab c + Q b c a + Q c ab ) ,
so (3.38c) is equivalent to Alt ( Q ) = 0 \operatorname{Alt}(Q) = 0 Alt ( Q ) = 0 . This establishes the equivalence of the two descriptions of Q 64 Q_{64} Q 64 . Finally, the nondegenerate W W W -pairing uniquely determines J Q s J^{s}_{Q} J Q s from ε Q s \varepsilon^{s}_{Q} ε Q s and J Q a J^{a}_{Q} J Q a from ε Q a \varepsilon^{a}_{Q} ε Q a .
The contraction identities of the model and the symmetries of its forms are as follows.
Lemma 3.25. Let Q ∈ Q 64 Q \in Q_{64} Q ∈ Q 64 .
(i) Every α ∈ Λ 14 2 \alpha\in\Lambda^{2}_{14} α ∈ Λ 14 2 satisfies ψ r s b c α b c = − 2 α r s \psi_{rsbc}\alpha^{bc} = -2\alpha_{rs} ψ r s b c α b c = − 2 α r s . In particular,
ψ r s b c Q a b c = − 2 Q a r s . (107) \psi_{rsbc}Q_{a}^{\ bc} = -2Q_{ars}. \tag*{(107)} ψ r s b c Q a b c = − 2 Q a r s . (107) (ii) The following contractions vanish:
Q a b a = 0 , (108) Q^{a}_{\ ab} = 0, \tag*{(108)} Q ab a = 0 , (108) Q a b c φ d a b = 0. (109) Q_{abc}\varphi^{ab}_{\ \ d} = 0. \tag*{(109)} Q ab c φ d ab = 0. (109) (iii) For all symmetric 2-tensors h , k h,k h , k ,
ε Q ( h , g ) = ε Q ( g , k ) = 0. \varepsilon_{Q}(h,g) = \varepsilon_{Q}(g,k) = 0. ε Q ( h , g ) = ε Q ( g , k ) = 0. Thus ε Q \varepsilon_{Q} ε Q , ε Q s \varepsilon^{s}_{Q} ε Q s and ε Q a \varepsilon^{a}_{Q} ε Q a depend only on the traceless parts of their arguments.
(iv) The maps J s , J a : Q 64 → End ( W ) J^{s},J^{a}: Q_{64} \to\operatorname{End}(W) J s , J a : Q 64 → End ( W ) are G 2 G_{2} G 2 -equivariant, with O ∈ G 2 O \in G_{2} O ∈ G 2 acting on End ( W ) \operatorname{End}(W) End ( W ) by X ↦ ρ ( O ) X ρ ( O ) − 1 X \mapsto\rho(O)X\rho(O)^{-1} X ↦ ρ ( O ) X ρ ( O ) − 1 . Moreover, J Q s J^{s}_{Q} J Q s is W W W -self-adjoint and J Q a J^{a}_{Q} J Q a is W W W -skew-adjoint.
Proof. Part (i) follows from (2.22), since each slice Q a Q_a Q a belongs to Λ 14 2 \Lambda^{2}_{14} Λ 14 2 . For (ii), equation (3.41) and the cyclic identity give
Q a s a − 1 2 ψ s a b c Q a b c = 1 6 ψ s a b c ( Q a b c + Q b c a + Q c a b ) = 0. Q^{a}_{\ as} - \frac{1}{2}\psi_{sabc}Q^{abc} = \frac{1}{6}\psi_{sabc}(Q^{abc} + Q^{bca} + Q^{cab}) = 0. Q a s a − 2 1 ψ s ab c Q ab c = 6 1 ψ s ab c ( Q ab c + Q b c a + Q c ab ) = 0.
Similarly, contracting the cyclic identity with φ d a b \varphi^{ab}_{\ \ d} φ d ab gives
Q a b c φ d a b = − Q b c a φ d a b = − Q a b c φ d a b , Q_{abc}\varphi^{ab}_{\ \ d} = -Q_{bca}\varphi^{ab}_{\ \ d} = -Q_{abc}\varphi^{ab}_{\ \ d}, Q ab c φ d ab = − Q b c a φ d ab = − Q ab c φ d ab ,
where the Q c a b Q_{cab} Q c ab term vanishes by (3.38b), and the last equality uses (3.38a). This proves (ii). For (iii), inserting g g g into the second argument of ε Q \varepsilon_{Q} ε Q gives zero by (3.38b); inserting it into the first gives zero by (3.42b) and (3.38a). Finally, Q 64 Q_{64} Q 64 is G 2 G_{2} G 2 -invariant because Λ 14 2 \Lambda^{2}_{14} Λ 14 2 and the alternation map are equivariant. The defining contraction of ε Q \varepsilon_{Q} ε Q and the W W W -pairing are also G 2 G_{2} G 2 -invariant, proving equivariance of J s J^{s} J s and J a J^{a} J a . Their adjointness properties follow directly from the symmetry of ε Q s \varepsilon^{s}_{Q} ε Q s and the skew-symmetry of ε Q a \varepsilon^{a}_{Q} ε Q a .
Differentiating the equivariance in Lemma 3.25 (iv) gives
J M ⋅ Q s = [ ρ ( M ) , J Q s ] , J M ⋅ Q a = [ ρ ( M ) , J Q a ] , M ∈ g 2 , J^{s}_{M\cdot Q} = [\rho(M),J^{s}_{Q}], \qquad J^{a}_{M\cdot Q} = [\rho(M),J^{a}_{Q}], \qquad M \in\mathfrak{g}_{2}, J M ⋅ Q s = [ ρ ( M ) , J Q s ] , J M ⋅ Q a = [ ρ ( M ) , J Q a ] , M ∈ g 2 ,
where M ⋅ Q M\cdot Q M ⋅ Q denotes the natural infinitesimal action on Q 64 Q_{64} Q 64 .
Corollary 3.26. The complexification Q 64 , C Q_{64,\mathbb{C}} Q 64 , C is the irreducible g 2 C \mathfrak{g}_{2}^{\mathbb{C}} g 2 C -module V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) . Consequently, Q 64 Q_{64} Q 64 has real dimension 64 and is irreducible under both G 2 G_{2} G 2 and g 2 \mathfrak{g}_{2} g 2 .
Moreover, under the metric identification Λ 14 2 ≅ g 2 \Lambda^{2}_{14} \cong\mathfrak{g}_{2} Λ 14 2 ≅ g 2 ,
V C ⊗ g 2 C ≅ Q 64 , C ⊕ W C ⊕ V C . V_{\mathbb{C}} \otimes\mathfrak{g}_{2}^{\mathbb{C}} \cong Q_{64,\mathbb{C}} \oplus W_{\mathbb{C}} \oplus V_{\mathbb{C}}. V C ⊗ g 2 C ≅ Q 64 , C ⊕ W C ⊕ V C . In particular, Q 64 , C Q_{64,\mathbb{C}} Q 64 , C is the unique submodule of type V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in this tensor product.
Proof. The decomposition (30) gives
( V ∗ ⊗ Λ 14 2 ) C ≅ V ( 1 , 1 ) ⊕ V ( 2 , 0 ) ⊕ V ( 1 , 0 ) . (V^{*} \otimes\Lambda^{2}_{14})_{\mathbb{C}} \cong V(1,1) \oplus V(2,0) \oplus V(1,0). ( V ∗ ⊗ Λ 14 2 ) C ≅ V ( 1 , 1 ) ⊕ V ( 2 , 0 ) ⊕ V ( 1 , 0 ) .
By [7 Lemma 2.1] , the kernel of alternation is precisely the V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) summand. The other two summands are W C W_{\mathbb{C}} W C and V C V_{\mathbb{C}} V C , respectively. The dimension and uniqueness follow, while irreducibility over R \mathbb{R} R follows by complexification. □ \square □
Example 3.27. In the standard frame of (4) , choose
Q 1 = − e 14 + e 27 , Q 2 = e 17 + e 24 , Q 3 = ⋯ = Q 7 = 0. Q_{1}=-e^{14}+e^{27}, \qquad Q_{2}=e^{17}+e^{24}, \qquad Q_{3}=\cdots=Q_{7}=0. Q 1 = − e 14 + e 27 , Q 2 = e 17 + e 24 , Q 3 = ⋯ = Q 7 = 0. The contractions of the two nonzero slices with φ \varphi φ cancel in the e 5 e_{5} e 5 and e 6 e_{6} e 6 components, respectively. The cyclic identity holds on repeated indices by skew-symmetry, and the only triple of distinct indices on which it is nontrivial is { 1 , 2 , 7 } \{1,2,7\} { 1 , 2 , 7 } , where Q 127 + Q 271 + Q 712 = 1 − 1 + 0 = 0 Q_{127}+Q_{271}+Q_{712}=1-1+0=0 Q 127 + Q 271 + Q 712 = 1 − 1 + 0 = 0 . Thus Q ∈ Q 64 Q\in Q_{64} Q ∈ Q 64 .
In Section 2 we identified the W W W -self-adjoint and W W W -skew-adjoint endomorphisms of W C W_{\mathbb{C}} W C equivariantly with S 2 W C S^{2}W_{\mathbb{C}} S 2 W C and Λ 2 W C \Lambda^{2}W_{\mathbb{C}} Λ 2 W C ; see (32) , (28) and (29) . Since Q 64 , C Q_{64,\mathbb{C}} Q 64 , C is irreducible, each parity map is injective whenever it is nonzero. A single pair of test elements shows that both maps are nonzero.
Lemma 3.28. For the maps Q ↦ J Q s Q\mapsto J^{s}_{Q} Q ↦ J Q s and Q ↦ J Q a Q\mapsto J^{a}_{Q} Q ↦ J Q a defined by (106) , we have the following.
(i) The maps Q ↦ J Q s Q\mapsto J^{s}_{Q} Q ↦ J Q s and Q ↦ J Q a Q\mapsto J^{a}_{Q} Q ↦ J Q a are injective on Q 64 Q_{64} Q 64 and on Q 64 , C Q_{64,\mathbb{C}} Q 64 , C ; equivalently, for every nonzero Q Q Q , real or complex, neither ε Q s \varepsilon^{s}_{Q} ε Q s nor ε Q a \varepsilon^{a}_{Q} ε Q a vanishes identically on W W W , respectively W C W_{\mathbb{C}} W C .
(ii) The maps Q ↦ J Q s Q\mapsto J^{s}_{Q} Q ↦ J Q s and Q ↦ J Q a Q\mapsto J^{a}_{Q} Q ↦ J Q a are G 2 G_{2} G 2 -equivariant isomorphisms of Q 64 , C Q_{64,\mathbb{C}} Q 64 , C onto the unique copies of the 64-dimensional module V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in S 2 W C S^{2}W_{\mathbb{C}} S 2 W C and in Λ 2 W C \Lambda^{2}W_{\mathbb{C}} Λ 2 W C , respectively.
Proof. Use the tensor Q Q Q of Example 3.27 and take
h = e 1 ⊗ e 1 − e 7 ⊗ e 7 , k = e 1 ⊗ e 5 + e 5 ⊗ e 1 . h=e^{1}\otimes e^{1}-e^{7}\otimes e^{7}, \qquad k=e^{1}\otimes e^{5}+e^{5}\otimes e^{1}. h = e 1 ⊗ e 1 − e 7 ⊗ e 7 , k = e 1 ⊗ e 5 + e 5 ⊗ e 1 .
Both tensors belong to W W W . The defining contraction gives
ε Q ( h , k ) = Q 114 φ 415 = 1 , \varepsilon_{Q}(h,k)=Q_{114}\varphi_{415}=1, ε Q ( h , k ) = Q 114 φ 415 = 1 ,
ε Q ( k , h ) = Q 114 φ 451 − Q 172 φ 257 = − 2. \varepsilon_{Q}(k,h)=Q_{114}\varphi_{451}-Q_{172}\varphi_{257}=-2. ε Q ( k , h ) = Q 114 φ 451 − Q 172 φ 257 = − 2.
Hence ε Q s ( h , k ) = − 1 2 \varepsilon^{s}_{Q}(h,k)=-\frac{1}{2} ε Q s ( h , k ) = − 2 1 and ε Q a ( h , k ) = 3 2 \varepsilon^{a}_{Q}(h,k)=\frac{3}{2} ε Q a ( h , k ) = 2 3 , so both parity maps are nonzero. By Lemma 3.25 (iv), their complexifications are equivariant. Their kernels are therefore submodules of the irreducible Q 64 , C Q_{64,\mathbb{C}} Q 64 , C , so both maps are injective, as are their real restrictions.
The images lie in S 2 W C S^{2}W_{\mathbb{C}} S 2 W C and Λ 2 W C \Lambda^{2}W_{\mathbb{C}} Λ 2 W C , respectively, by the adjointness properties in Lemma 3.25 (iv). Each image is a copy of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) , which occurs with multiplicity one in each target by (28) and (29) . □ \square □
:::
The two generator stabilizers and the parity model of the 64 64 64 now reduce the remaining count to the kernel restricted to the isotypic component of 64 64 64 in End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) .
The exact stabilizer count
The remaining component is
H 64 = End ( W C ) V ( 1 , 1 ) = J s ( Q 64 , C ) ⊕ J a ( Q 64 , C ) . \mathcal{H}_{64}=\operatorname{End}(W_{\mathbb{C}})_{V_{(1,1)}}=\mathcal{J}^{s}(Q_{64,\mathbb{C}})\oplus\mathcal{J}^{a}(Q_{64,\mathbb{C}}). H 64 = End ( W C ) V ( 1 , 1 ) = J s ( Q 64 , C ) ⊕ J a ( Q 64 , C ) .
Write Φ 0 ( 64 ) = Φ 0 ∣ H 64 \Phi^{(64)}_0=\Phi_0|_{\mathcal{H}_{64}} Φ 0 ( 64 ) = Φ 0 ∣ H 64 and Φ φ ( 64 ) = Φ φ ∣ H 64 \Phi^{(64)}_{\varphi}=\Phi_{\varphi}|_{\mathcal{H}_{64}} Φ φ ( 64 ) = Φ φ ∣ H 64 . Proposition 3.21 and the decomposition following it identify the trace-cubic kernel as 7 ⊕ 14 7\oplus14 7 ⊕ 14 , so Φ 0 ( 64 ) \Phi^{(64)}_0 Φ 0 ( 64 ) is injective. Using the Koecher lower bound, we will show that Φ φ ( 64 ) ( H 64 ) ⊆ Φ 0 ( 64 ) ( H 64 ) \Phi^{(64)}_{\varphi}(\mathcal{H}_{64})\subseteq\Phi^{(64)}_0(\mathcal{H}_{64}) Φ φ ( 64 ) ( H 64 ) ⊆ Φ 0 ( 64 ) ( H 64 ) and hence define a relative operator. We then compute its matrix with respect to the parity embeddings. Set
m ± = ker ( Φ 0 ( 64 ) + λ ± Φ φ ( 64 ) ) = ker Φ 1 , λ ± ∣ H 64 . \mathfrak{m}_{\pm}=\ker(\Phi^{(64)}_0+\lambda_{\pm}\Phi^{(64)}_{\varphi})=\ker\Phi_{1,\lambda_{\pm}}|_{\mathcal{H}_{64}}. m ± = ker ( Φ 0 ( 64 ) + λ ± Φ φ ( 64 ) ) = ker Φ 1 , λ ± ∣ H 64 .
Lemma 3.29. For every Q ∈ Q 64 , C Q\in Q_{64,\mathbb{C}} Q ∈ Q 64 , C ,
Φ C φ ( J Q s ) = − Φ C 0 ( J Q s ) − Φ C 0 ( J Q a ) , \Phi_{C_{\varphi}}(\mathcal{J}^{s}_{Q})=-\Phi_{C_0}(\mathcal{J}^{s}_{Q})-\Phi_{C_0}(\mathcal{J}^{a}_{Q}), Φ C φ ( J Q s ) = − Φ C 0 ( J Q s ) − Φ C 0 ( J Q a ) , Φ C φ ( J Q a ) = 7 Φ C 0 ( J Q s ) − Φ C 0 ( J Q a ) . (110) \Phi_{C_{\varphi}}(\mathcal{J}^{a}_{Q})=7\Phi_{C_0}(\mathcal{J}^{s}_{Q})-\Phi_{C_0}(\mathcal{J}^{a}_{Q}). \tag*{(110)} Φ C φ ( J Q a ) = 7 Φ C 0 ( J Q s ) − Φ C 0 ( J Q a ) . (110)
Consequently, there is a unique endomorphism T T T of H 64 \mathcal{H}_{64} H 64 such that Φ φ ( 64 ) = Φ 0 ( 64 ) ∘ T \Phi^{(64)}_{\varphi}=\Phi^{(64)}_0\circ T Φ φ ( 64 ) = Φ 0 ( 64 ) ∘ T . In the ordered pair of embeddings ( J s , J a ) (\mathcal{J}^{s},\mathcal{J}^{a}) ( J s , J a ) , with images written as columns, the matrix of T T T is
T = ( − 1 7 − 1 − 1 ) . T=\begin{pmatrix}-1 & 7 \\ -1 & -1\end{pmatrix}. T = ( − 1 − 1 7 − 1 ) .
In particular, Φ a , b = Φ C 0 ∘ ( a id + b T ) \Phi_{a,b}=\Phi_{C_0}\circ(a\operatorname{id}+bT) Φ a , b = Φ C 0 ∘ ( a id + b T ) on this component, where, for the matrix of T T T ,
det ( a id + b T ) = ( a − b ) 2 + 7 b 2 = a 2 − 2 a b + 8 b 2 . (111) \det(a\operatorname{id}+bT)=(a-b)^2+7b^2=a^2-2ab+8b^2. \tag*{(111)} det ( a id + b T ) = ( a − b ) 2 + 7 b 2 = a 2 − 2 ab + 8 b 2 . (111)
For ( a , b ) ≠ 0 (a,b)\ne0 ( a , b ) = 0 , the kernel of Φ a , b \Phi_{a,b} Φ a , b on this component is nonzero exactly when a 2 − 2 a b + 8 b 2 = 0 a^2-2ab+8b^2=0 a 2 − 2 ab + 8 b 2 = 0 , equivalently when λ = b / a = λ ± \lambda=b/a=\lambda_{\pm} λ = b / a = λ ± . At these parameters, we have
ker Φ a , b ∣ H 64 = m ± = { J Q s + ζ ± J Q a : Q ∈ Q 64 , C } , ζ ± = ± i 7 . \ker\Phi_{a,b}|_{\mathcal{H}_{64}}=\mathfrak{m}_{\pm}=\{\mathcal{J}^{s}_{Q}+\zeta_{\pm}\mathcal{J}^{a}_{Q}:Q\in Q_{64,\mathbb{C}}\},\qquad\zeta_{\pm}=\pm\frac{\mathrm{i}}{\sqrt{7}}. ker Φ a , b ∣ H 64 = m ± = { J Q s + ζ ± J Q a : Q ∈ Q 64 , C } , ζ ± = ± 7 i .
Here ζ = b / ( a − b ) = λ / ( 1 − λ ) \zeta=b/(a-b)=\lambda/(1-\lambda) ζ = b / ( a − b ) = λ / ( 1 − λ ) is the ratio of the coefficient of J Q a \mathcal{J}^{a}_{Q} J Q a to that of J Q s \mathcal{J}^{s}_{Q} J Q s .
Proof. By Proposition 3.18, each critical stabilizer contains a copy of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) , which lies in H 64 \mathcal{H}_{64} H 64 by the definition of H 64 \mathcal{H}_{64} H 64 as the V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) -isotypic component. Thus dim m ± ≥ 64 \dim\mathfrak{m}_{\pm}\ge64 dim m ± ≥ 64 . If X X X lies in both kernels, subtraction gives ( λ + − λ − ) Φ φ ( 64 ) ( X ) = 0 (\lambda_{+}-\lambda_{-})\Phi^{(64)}_{\varphi}(X)=0 ( λ + − λ − ) Φ φ ( 64 ) ( X ) = 0 , hence Φ φ ( 64 ) ( X ) = Φ 0 ( 64 ) ( X ) = 0 \Phi^{(64)}_{\varphi}(X)=\Phi^{(64)}_0(X)=0 Φ φ ( 64 ) ( X ) = Φ 0 ( 64 ) ( X ) = 0 . Injectivity of Φ 0 ( 64 ) \Phi^{(64)}_0 Φ 0 ( 64 ) gives X = 0 X=0 X = 0 . Since dim H 64 = 128 \dim\mathcal{H}_{64}=128 dim H 64 = 128 , it follows that
H 64 = m + ⊕ m − , dim m + = dim m − = 64. \mathcal{H}_{64}=\mathfrak{m}_{+}\oplus\mathfrak{m}_{-},\qquad\dim\mathfrak{m}_{+}=\dim\mathfrak{m}_{-}=64. H 64 = m + ⊕ m − , dim m + = dim m − = 64.
In particular, each kernel is the irreducible copy just obtained. For X ± ∈ m ± X_{\pm}\in\mathfrak{m}_{\pm} X ± ∈ m ± we have Φ φ ( 64 ) ( X ± ) = − λ ± − 1 Φ 0 ( 64 ) ( X ± ) \Phi^{(64)}_{\varphi}(X_{\pm})=-\lambda_{\pm}^{-1}\Phi^{(64)}_0(X_{\pm}) Φ φ ( 64 ) ( X ± ) = − λ ± − 1 Φ 0 ( 64 ) ( X ± ) . Decomposing any X ∈ H 64 X\in\mathcal{H}_{64} X ∈ H 64 as X + + X − X_{+}+X_{-} X + + X − therefore proves Φ φ ( 64 ) ( H 64 ) ⊆ Φ 0 ( 64 ) ( H 64 ) \Phi^{(64)}_{\varphi}(\mathcal{H}_{64}) \subseteq\Phi^{(64)}_{0}(\mathcal{H}_{64}) Φ φ ( 64 ) ( H 64 ) ⊆ Φ 0 ( 64 ) ( H 64 ) . There is a unique equivariant operator T = ( Φ 0 ( 64 ) ) − 1 Φ φ ( 64 ) T=(\Phi^{(64)}_{0})^{-1}\Phi^{(64)}_{\varphi} T = ( Φ 0 ( 64 ) ) − 1 Φ φ ( 64 ) on H 64 \mathcal{H}_{64} H 64 , and
T ∣ m ± = − λ ± − 1 id = ( − 1 ± i 7 ) id . T|_{\mathfrak{m}_{\pm}}=-\lambda_{\pm}^{-1}\operatorname{id}=(-1\pm i\sqrt{7})\operatorname{id}. T ∣ m ± = − λ ± − 1 id = ( − 1 ± i 7 ) id .
These two eigenvalues already give det ( a id + b T ) = a 2 − 2 a b + 8 b 2 \det(a\operatorname{id}+bT)=a^2-2ab+8b^2 det ( a id + b T ) = a 2 − 2 ab + 8 b 2 on the two-dimensional multiplicity space of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) in H 64 \mathcal{H}_{64} H 64 . It remains to identify the eigenlines in the chosen parity basis.
Use the standard φ \varphi φ of (4) , and let E i j E_{ij} E ij denote the matrix unit with its sole nonzero entry equal to 1 in row i i i , column j j j . For this computation, put D a = E a a − E 77 D_a=E_{aa}-E_{77} D a = E aa − E 77 and S i j = E i j + E j i S_{ij}=E_{ij}+E_{ji} S ij = E ij + E j i . Use Q Q Q from Example 3.27 and choose h = D 1 + S 37 h=D_1+S_{37} h = D 1 + S 37 . Direct multiplication and the definition of P \mathcal{P} P give
h 2 = E 11 + E 33 + 2 E 77 − S 37 , P ( h , h ) = − 2 E 44 − 2 E 66 + 2 S 26 . h^2=E_{11}+E_{33}+2E_{77}-S_{37},\qquad\mathcal{P}(h,h)=-2E_{44}-2E_{66}+2S_{26}. h 2 = E 11 + E 33 + 2 E 77 − S 37 , P ( h , h ) = − 2 E 44 − 2 E 66 + 2 S 26 .
For any symmetric k k k , the same four nonzero slice components give
ε Q ( h , k ) = k 15 + k 26 + k 37 , ε Q ( k , h ) = k 11 − k 22 − 2 k 15 . \varepsilon_Q(h,k)=k_{15}+k_{26}+k_{37},\qquad\varepsilon_Q(k,h)=k_{11}-k_{22}-2k_{15}. ε Q ( h , k ) = k 15 + k 26 + k 37 , ε Q ( k , h ) = k 11 − k 22 − 2 k 15 .
Consequently
( ε Q ( h , h 2 ) , ε Q ( h 2 , h ) , ε Q ( h , P ( h , h ) ) , ε Q ( P ( h , h ) , h ) ) = ( − 1 , 1 , 2 , 0 ) . (112) \left(\varepsilon_Q(h,h^2),\varepsilon_Q(h^2,h),\varepsilon_Q(h,\mathcal{P}(h,h)),\varepsilon_Q(\mathcal{P}(h,h),h)\right)=(-1,1,2,0). \tag*{(112)} ( ε Q ( h , h 2 ) , ε Q ( h 2 , h ) , ε Q ( h , P ( h , h )) , ε Q ( P ( h , h ) , h ) ) = ( − 1 , 1 , 2 , 0 ) . (112)
By (39) , (43) , Lemma 3.25 (iii), (105) and the half-sum and half-difference convention, these values give
1 3 Φ 1 , λ ( J Q s ) ( h , h , h ) = λ , 1 3 Φ 1 , λ ( J Q a ) ( h , h , h ) = λ − 1. \frac{1}{3}\Phi_{1,\lambda}(J_Q^s)(h,h,h)=\lambda,\qquad\frac{1}{3}\Phi_{1,\lambda}(J_Q^a)(h,h,h)=\lambda-1. 3 1 Φ 1 , λ ( J Q s ) ( h , h , h ) = λ , 3 1 Φ 1 , λ ( J Q a ) ( h , h , h ) = λ − 1.
Since λ ± − 1 ≠ 0 \lambda_{\pm}-1\ne0 λ ± − 1 = 0 , neither critical kernel is the whole skew copy J a ( Q 64 , C ) J^a(\mathcal{Q}_{64,\mathbb{C}}) J a ( Q 64 , C ) . The projection of each m ± \mathfrak{m}_{\pm} m ± onto the symmetric copy is therefore nonzero, hence an isomorphism by irreducibility and Schur’s lemma. Each m ± \mathfrak{m}_{\pm} m ± is consequently a graph { J Q s + ζ ± J Q a : Q ∈ Q 64 , C } \{J_Q^s+\zeta_{\pm}J_Q^a:Q\in\mathcal{Q}_{64,\mathbb{C}}\} { J Q s + ζ ± J Q a : Q ∈ Q 64 , C } . Evaluating Φ 1 , λ ± \Phi_{1,\lambda_{\pm}} Φ 1 , λ ± on J Q s + ζ ± J Q a J_Q^s+\zeta_{\pm}J_Q^a J Q s + ζ ± J Q a at the displayed Q Q Q and h h h determines ζ ± \zeta_{\pm} ζ ± :
λ ± + ζ ± ( λ ± − 1 ) = 0 , ζ ± = λ ± 1 − λ ± = ± i 7 . \lambda_{\pm}+\zeta_{\pm}(\lambda_{\pm}-1)=0,\qquad\zeta_{\pm}=\frac{\lambda_{\pm}}{1-\lambda_{\pm}}=\pm\frac{i}{\sqrt{7}}. λ ± + ζ ± ( λ ± − 1 ) = 0 , ζ ± = 1 − λ ± λ ± = ± 7 i .
In particular the plus critical parameter corresponds to the plus sign of i / 7 i/\sqrt{7} i / 7 .
The two graph lines ( 1 , ζ ± ) (1,\zeta_{\pm}) ( 1 , ζ ± ) and their eigenvalues − 1 ± i 7 -1\pm i\sqrt{7} − 1 ± i 7 determine the matrix of T T T uniquely. It is
( − 1 7 − 1 − 1 ) , \begin{pmatrix}
-1 & 7\\
-1 & -1
\end{pmatrix}, ( − 1 − 1 7 − 1 ) ,
since 7 ζ ± = ± i 7 7\zeta_{\pm}=\pm i\sqrt{7} 7 ζ ± = ± i 7 and 7 ζ ± 2 = − 1 7\zeta_{\pm}^{2}=-1 7 ζ ± 2 = − 1 . Its columns give (110) .
Finally, Φ a , b = Φ 0 ( 64 ) ∘ ( a id + b T ) \Phi_{a,b}=\Phi^{(64)}_{0}\circ(a\operatorname{id}+bT) Φ a , b = Φ 0 ( 64 ) ∘ ( a id + b T ) on H 64 \mathcal{H}_{64} H 64 and Φ 0 ( 64 ) \Phi^{(64)}_{0} Φ 0 ( 64 ) is injective, so the determinant and the graph description give all the asserted kernels, including the endpoints of the projective pencil.
Remark 3.30. The matrix of T T T in Lemma 3.29 gives ( id + T ) 2 = − 7 id (\operatorname{id}+T)^2=-7\operatorname{id} ( id + T ) 2 = − 7 id . In particular,
( id + T ) ( J Q s + ζ ± J Q a ) = 7 ζ ± ( J Q s + ζ ± J Q a ) , (\operatorname{id}+T)(J_Q^s+\zeta_{\pm}J_Q^a)=7\zeta_{\pm}(J_Q^s+\zeta_{\pm}J_Q^a), ( id + T ) ( J Q s + ζ ± J Q a ) = 7 ζ ± ( J Q s + ζ ± J Q a ) ,
so these are the eigenspaces of id + T \operatorname{id}+T id + T with eigenvalues ± i 7 \pm i\sqrt{7} ± i 7 , respectively. We have not found the explicit parity embeddings, the identities (110) , or the resulting relative operator T T T with ( id + T ) 2 = − 7 id (\operatorname{id}+T)^2=-7\operatorname{id} ( id + T ) 2 = − 7 id in the literature consulted.
The appearance of − 7 \sqrt{-7} − 7 in the earlier G 2 ⊂ E 6 G_2 \subset E_6 G 2 ⊂ E 6 constructions is discussed in § 1 . In particular, Aschbacher’s result, reported in [58 p. 300] , states that for a field F \mathbb{F} F with char F ≠ 2 , 3 , 7 \operatorname{char}\mathbb{F} \ne2,3,7 char F = 2 , 3 , 7 , the group G 2 ( F ) G_2(\mathbb{F}) G 2 ( F ) preserves, up to scalar multiple, exactly two forms similar to the E 6 ( F ) E_6(\mathbb{F}) E 6 ( F ) form if and only if − 7 ∈ F \sqrt{-7} \in\mathbb{F} − 7 ∈ F .
We can now assemble the kernels. For the critical kernels m ± \mathfrak{m}_{\pm} m ± of Lemma 3.29 , define
E ± : Q 64 , C ⟶ m ± , E ± ( Q ) = J Q S + ζ ± J Q a . \mathcal{E}_{\pm}: Q_{64,\mathbb{C}} \longrightarrow\mathfrak{m}_{\pm}, \qquad\mathcal{E}_{\pm}(Q)=J^S_Q+\zeta_{\pm}J^a_Q. E ± : Q 64 , C ⟶ m ± , E ± ( Q ) = J Q S + ζ ± J Q a .
By Lemma 3.25 (iv) and its infinitesimal form above, E ± \mathcal{E}_{\pm} E ± is a g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C -equivariant map onto m ± \mathfrak{m}_{\pm} m ± , and it is injective because the W W W -self-adjoint part of E ± ( Q ) \mathcal{E}_{\pm}(Q) E ± ( Q ) is J Q S J^S_Q J Q S (Lemma 3.28 (i)). Hence m ± ≅ Q 64 , C ≅ V ( 1 , 1 ) \mathfrak{m}_{\pm} \cong Q_{64,\mathbb{C}} \cong V(1,1) m ± ≅ Q 64 , C ≅ V ( 1 , 1 ) by Corollary 3.26 .
Proposition 3.31. Let ( a , b ) ∈ C 2 ∖ { 0 } (a,b) \in\mathbb{C}^2 \setminus\{0\} ( a , b ) ∈ C 2 ∖ { 0 } . Then ker Φ a , b \ker\Phi_{a,b} ker Φ a , b is the direct sum of its intersections with the isotypic components End ( W C ) V ( j 1 , j 2 ) \operatorname{End}(W_{\mathbb{C}})_{V(j_1,j_2)} End ( W C ) V ( j 1 , j 2 ) of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) , and
dim ker Φ a , b = { 14 + 7 = 21 , b = 0 , 14 + 64 = 78 , a 2 − 2 a b + 8 b 2 = 0 , 14 , otherwise . (113) \dim\ker\Phi_{a,b}=
\begin{cases}
14+7=21, & b=0,\\
14+64=78, & a^2-2ab+8b^2=0,\\
14, & \text{otherwise}.
\end{cases} \tag*{(113)} dim ker Φ a , b = ⎩ ⎨ ⎧ 14 + 7 = 21 , 14 + 64 = 78 , 14 , b = 0 , a 2 − 2 ab + 8 b 2 = 0 , otherwise . (113) The contributions are
Table 3.
The contributions of V ( 1 , 0 ) V(1,0) V ( 1 , 0 ) and V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) vanish outside the indicated loci. At a critical parameter, the 64-dimensional contribution is a single copy of V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) , and
g ± : = stab ( N ± ) = ker Φ 1 , λ ± = ρ ( g 2 C ) ⊕ m ± . \mathfrak{g}_{\pm}:=\operatorname{stab}(N_{\pm})=\ker\Phi_{1,\lambda_{\pm}}=\rho(\mathfrak{g}_2^{\mathbb{C}})\oplus\mathfrak{m}_{\pm}. g ± := stab ( N ± ) = ker Φ 1 , λ ± = ρ ( g 2 C ) ⊕ m ± .
Proof. By Lemma 3.2 , Φ a , b \Phi_{a,b} Φ a , b is g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C -equivariant, so its kernel decomposes over the isotypic components of End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) listed in Table 1 .
The component V ( 0 , 1 ) = ρ ( g 2 C ) V(0,1)=\rho(\mathfrak{g}_2^{\mathbb{C}}) V ( 0 , 1 ) = ρ ( g 2 C ) lies in every kernel and has dimension 14. The scalar component contributes nothing, since Φ a , b ( id W ) = 3 ( a C 0 + b C ϕ ) ≠ 0 \Phi_{a,b}(\operatorname{id}_{W})=3(aC_0+bC_{\phi})\ne0 Φ a , b ( id W ) = 3 ( a C 0 + b C ϕ ) = 0 by Lemma 2.7 .
The unique V ( 1 , 0 ) V(1,0) V ( 1 , 0 ) component is ρ ( Λ 7 2 ) C \rho(\Lambda^2_7)_{\mathbb{C}} ρ ( Λ 7 2 ) C . It lies in ker Φ C 0 \ker\Phi_{C_0} ker Φ C 0 by Proposition 3.21 , but has trivial intersection with ker Φ C ϕ = ρ ( g 2 C ) \ker\Phi_{C_{\phi}}=\rho(\mathfrak{g}_2^{\mathbb{C}}) ker Φ C ϕ = ρ ( g 2 C ) by Proposition 3.23 . Thus it contributes dimension 7 when b = 0 b=0 b = 0 and zero otherwise.
On the V ( 1 , 1 ) V(1,1) V ( 1 , 1 ) -isotypic component, Lemma 3.29 gives a trivial kernel unless a 2 − 2 a b + 8 b 2 = 0 a^2-2ab+8b^2=0 a 2 − 2 ab + 8 b 2 = 0 . At the two critical parameters λ ± = b / a \lambda_{\pm}=b/a λ ± = b / a , the kernel is m ± ≅ V ( 1 , 1 ) \mathfrak{m}_{\pm}\cong V(1,1) m ± ≅ V ( 1 , 1 ) , of dimension 64.
By Proposition 3.13, all remaining components have trivial kernel. These are the two copies of V ( 2 , 0 ) V(2,0) V ( 2 , 0 ) and the summands of types V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) , V ( 4 , 0 ) V(4,0) V ( 4 , 0 ) , V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) , and V ( 2 , 1 ) V(2,1) V ( 2 , 1 ) .
Adding these contributions gives (113) . In particular, at the critical parameters,
ker Φ 1 , λ ± = ρ ( g 2 C ) ⊕ m ± \ker\Phi_{1,\lambda_{\pm}}=\rho(\mathfrak{g}_2^{\mathbb{C}})\oplus\mathfrak{m}_{\pm} ker Φ 1 , λ ± = ρ ( g 2 C ) ⊕ m ±
has dimension 14 + 64 = 78 14+64=78 14 + 64 = 78 , attaining the bound of Proposition 3.18 . □ \square □
Since stab ( N a , b ) = ker Φ a , b \operatorname{stab}(N_{a,b})=\ker\Phi_{a,b} stab ( N a , b ) = ker Φ a , b , the count (113) proves Theorem 3.3 : at b = 0 b=0 b = 0 the quadratic equals a 2 ≠ 0 a^2\ne0 a 2 = 0 , so the three cases are disjoint, and on the locus a 2 − 2 a b + 8 b 2 = 0 a^2-2ab+8b^2=0 a 2 − 2 ab + 8 b 2 = 0 we have a ≠ 0 a\ne0 a = 0 , and λ = b / a \lambda=b/a λ = b / a is a root of 8 λ 2 − 2 λ + 1 8\lambda^2-2\lambda+1 8 λ 2 − 2 λ + 1 , one of the two non-real parameters λ ± \lambda_{\pm} λ ± of (63) .
Proposition 3.31 identifies the stabilizer itself as well as its dimension: it is ρ ( s o ( 7 , C ) ) \rho(\mathfrak{so}(7,\mathbb{C})) ρ ( so ( 7 , C )) for C 0 C_0 C 0 , ρ ( g 2 C ) ⊕ m ± \rho(\mathfrak{g}_2^{\mathbb{C}})\oplus\mathfrak{m}_{\pm} ρ ( g 2 C ) ⊕ m ± for N ± N_{\pm} N ± , and ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) for every other member, including C φ C_{\varphi} C φ . Thus the trace endpoint has the orthogonal symmetry algebra, whereas the Yukawa endpoint C φ C_{\varphi} C φ retains exactly ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) . Only the two mixed non-real members N ± N_{\pm} N ± acquire the additional 64-dimensional summand. The exceptional symmetry therefore belongs to this conjugate pair, rather than to the pointwise Yukawa cubic itself. It remains to identify the Lie algebra structure at the critical parameters.
The critical stabilizer
In this section we will identify the 78-dimensional algebras g ± = ρ ( g 2 C ) ⊕ m ± \mathfrak{g}_{\pm}=\rho(\mathfrak{g}_2^{\mathbb{C}})\oplus\mathfrak{m}_{\pm} g ± = ρ ( g 2 C ) ⊕ m ± of Proposition 3.31, with the isomorphism E ± : Q 64 , C → m ± \mathcal{E}_{\pm}:\mathcal{Q}_{64,\mathbb{C}}\to\mathfrak{m}_{\pm} E ± : Q 64 , C → m ± fixed above. We will first recall some terminology from Lie theory. An irreducible representation is minuscule if its weights form a single orbit under the Weyl group, in which case every weight has multiplicity one.
Theorem 3.32. For either sign, g ± \mathfrak{g}_{\pm} g ± is a simple complex Lie algebra of type E 6 E_6 E 6 . Its action on W C W_{\mathbb{C}} W C is one of the two 27-dimensional minuscule representations V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) . These representations are dual to one another and are exchanged by the nontrivial automorphism of the E 6 E_6 E 6 Dynkin diagram.
Proof. Simplicity. For a real endomorphism X X X , write ∥ X ∥ E n d 2 = ⟨ X , X ⟩ E n d \lVert X\rVert_{\mathrm{End}}^2=\langle X,X\rangle_{\mathrm{End}} ∥ X ∥ End 2 = ⟨ X , X ⟩ End using (84) . We choose a sign and write
a = ρ ( g 2 C ) , m = m ± , g = g ± = a ⊕ m . \mathfrak{a}=\rho(\mathfrak{g}_2^{\mathbb{C}}),\qquad\mathfrak{m}=\mathfrak{m}_{\pm},\qquad\mathfrak{g}=\mathfrak{g}_{\pm}=\mathfrak{a}\oplus\mathfrak{m}. a = ρ ( g 2 C ) , m = m ± , g = g ± = a ⊕ m .
The summands a ≅ V ( 0 , 1 ) \mathfrak{a}\cong V(0,1) a ≅ V ( 0 , 1 ) and m ≅ V ( 1 , 1 ) \mathfrak{m}\cong V(1,1) m ≅ V ( 1 , 1 ) are nonisomorphic irreducible g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C -modules. Every ideal of g ± \mathfrak{g}_{\pm} g ± is invariant under the adjoint action of a = ρ ( g 2 C ) \mathfrak{a}=\rho(\mathfrak{g}_2^{\mathbb{C}}) a = ρ ( g 2 C ) . Since g = a ⊕ m \mathfrak{g}=\mathfrak{a}\oplus\mathfrak{m} g = a ⊕ m is a sum of two nonisomorphic irreducible g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C -modules, each of multiplicity one, the only possible ideals are 0 0 0 , a \mathfrak{a} a , m \mathfrak{m} m , and g \mathfrak{g} g .
The action of a \mathfrak{a} a on m \mathfrak{m} m is nontrivial and irreducible, so
[ a , m ] = m . [\mathfrak{a},\mathfrak{m}]=\mathfrak{m}. [ a , m ] = m .
In particular, a \mathfrak{a} a is not an ideal.
To exclude m \mathfrak{m} m , consider the invariant symmetric bilinear form
b ( X , Y ) = tr W ( X Y ) b(X,Y)=\operatorname{tr}_{W}(XY) b ( X , Y ) = tr W ( X Y )
on g \mathfrak{g} g . By Schur’s lemma, b ( a , m ) = 0 b(\mathfrak{a},\mathfrak{m})=0 b ( a , m ) = 0 , since the two summands are nonisomorphic irreducible a \mathfrak{a} a -modules. Moreover, b b b is nondegenerate on m \mathfrak{m} m . Indeed, for any nonzero real Q ∈ Q 64 Q\in\mathcal{Q}_{64} Q ∈ Q 64 , put X = E ± ( Q ) = J Q S + ζ ± J Q A X=\mathcal{E}_{\pm}(Q)=J_Q^{S}+\zeta_{\pm}J_Q^{A} X = E ± ( Q ) = J Q S + ζ ± J Q A . Since ζ ± 2 = − 1 / 7 \zeta_{\pm}^2=-1/7 ζ ± 2 = − 1/7 and J Q S J_Q^{S} J Q S and J Q A J_Q^{A} J Q A are respectively W W W -self-adjoint and W W W -skew-adjoint,
b ( X , X ) = tr W ( ( J Q S ) 2 ) − 1 7 tr W ( ( J Q A ) 2 ) = ∥ J Q S ∥ E n d 2 + 1 7 ∥ J Q A ∥ E n d 2 > 0. \begin{aligned}
b(X,X)=\operatorname{tr}_{W}\left((J_Q^{S})^2\right)-\frac{1}{7}\operatorname{tr}_{W}\left((J_Q^{A})^2\right) \\
&=\lVert J_Q^{S}\rVert_{\mathrm{End}}^2+\frac{1}{7}\lVert J_Q^{A}\rVert_{\mathrm{End}}^2>0.
\end{aligned} b ( X , X ) = tr W ( ( J Q S ) 2 ) − 7 1 tr W ( ( J Q A ) 2 ) = ∥ J Q S ∥ End 2 + 7 1 ∥ J Q A ∥ End 2 > 0.
Thus b \mathfrak{b} b is positive definite on the real span of these elements, and its complex-bilinear extension is nondegenerate on m \mathfrak{m} m . If m \mathfrak{m} m were an ideal, then [ m , m ] ⊆ m [\mathfrak{m},\mathfrak{m}] \subseteq\mathfrak{m} [ m , m ] ⊆ m . By invariance of b \mathfrak{b} b , this would give
b ( [ A , U ] , V ) = b ( A , [ U , V ] ) = 0 \mathfrak{b}([A,U],V)=\mathfrak{b}(A,[U,V])=0 b ([ A , U ] , V ) = b ( A , [ U , V ]) = 0
for all A ∈ a A \in\mathfrak{a} A ∈ a and U , V ∈ m U,V \in\mathfrak{m} U , V ∈ m , since b ( a , m ) = 0 \mathfrak{b}(\mathfrak{a},\mathfrak{m})=0 b ( a , m ) = 0 . But [ a , m ] = m [\mathfrak{a},\mathfrak{m}]=\mathfrak{m} [ a , m ] = m and b \mathfrak{b} b is nondegenerate on m \mathfrak{m} m , a contradiction.
Hence g \mathfrak{g} g has no nonzero proper ideals and is simple.
Identification of the type. By the classification of complex simple Lie algebras, the only types of dimension 78 78 78 are B 6 B_6 B 6 , C 6 C_6 C 6 , and E 6 E_6 E 6 . The action of g ± \mathfrak{g}_{\pm} g ± on W C W_{\mathbb{C}} W C is irreducible, since its restriction to g 2 C \mathfrak{g}_{2}^{\mathbb{C}} g 2 C is already irreducible. Neither B 6 B_6 B 6 nor C 6 C_6 C 6 admits an irreducible representation of dimension 27 27 27 [43 proof of Lemma 4, Table 1, p. 4] . Hence g ± \mathfrak{g}_{\pm} g ± is of type E 6 E_6 E 6 .
The 27-dimensional representation. Finally, the only 27-dimensional irreducible representations of E 6 E_6 E 6 are V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) ; they are mutually dual, minuscule, and exchanged by the diagram automorphism [54 §3, Table 1] , [17 §§26.5, 30.4, 32.1] . Since W C W_{\mathbb{C}} W C is irreducible of dimension 27 27 27 under g ± \mathfrak{g}_{\pm} g ± , it is one of them.
Corollary 3.33. For [ a : b ] ∈ P 1 ( C ) [a:b] \in\mathbb{P}^{1}(\mathbb{C}) [ a : b ] ∈ P 1 ( C ) ,
stab ( N a , b ) ≅ { s o ( 7 , C ) , [ a : b ] = [ 1 : 0 ] , e 6 , a 2 − 2 a b + 8 b 2 = 0 , g 2 C , otherwise. \operatorname{stab}(N_{a,b}) \cong
\begin{cases}
\mathfrak{so}(7,\mathbb{C}), & [a:b]=[1:0],\\
\mathfrak{e}_{6}, & a^{2}-2ab+8b^{2}=0,\\
\mathfrak{g}_{2}^{\mathbb{C}}, & \text{otherwise.}
\end{cases} stab ( N a , b ) ≅ ⎩ ⎨ ⎧ so ( 7 , C ) , e 6 , g 2 C , [ a : b ] = [ 1 : 0 ] , a 2 − 2 ab + 8 b 2 = 0 , otherwise.
Proof. This follows from Proposition 3.31 and Theorem 3.32.
The classical embeddings and branching underlying this realization are discussed in Section 1.
For either sign, let S ± ( h 1 , h 2 ) = S λ ± , λ ∓ , 2 / 3 ( h 1 , h 2 ) S_{\pm}(h_{1},h_{2})=S_{\lambda_{\pm},\lambda_{\mp},2/3}(h_{1},h_{2}) S ± ( h 1 , h 2 ) = S λ ± , λ ∓ , 2/3 ( h 1 , h 2 ) be the Koecher operators of Definition 3.14, in the notation of Corollary 3.16, where h 1 , h 2 ∈ W C h_{1},h_{2}\in W_{\mathbb{C}} h 1 , h 2 ∈ W C . Then, for any h ∈ W C h\in W_{\mathbb{C}} h ∈ W C they are given by
S ± ( h 1 , h 2 ) h = ⟨ h , h 2 ⟩ W h 1 − 2 F ∓ ( F ± ( h 1 , h ) , h 2 ) + 1 3 ⟨ h 1 , h 2 ⟩ W h . (114) S_{\pm}(h_{1},h_{2})h=\langle h,h_{2}\rangle_{W}h_{1}-2F_{\mp}(F_{\pm}(h_{1},h),h_{2})+\frac{1}{3}\langle h_{1},h_{2}\rangle_{W}h. \tag*{(114)} S ± ( h 1 , h 2 ) h = ⟨ h , h 2 ⟩ W h 1 − 2 F ∓ ( F ± ( h 1 , h ) , h 2 ) + 3 1 ⟨ h 1 , h 2 ⟩ W h . (114)
By Proposition 3.7 and Corollary 3.16, all these operators lie in stab ( N ± ) . \operatorname{stab}(N_{\pm}). stab ( N ± ) . Proposition 3.18 shows that their span contains a 64-dimensional submodule. In Corollary 3.34 we show that the span is an ideal of g ± \mathfrak{g}_{\pm} g ± , so the simplicity established in Theorem 3.32 makes it all of g ± \mathfrak{g}_{\pm} g ± .
Corollary 3.34. For either sign, let S ± ( h 1 , h 2 ) S_{\pm}(h_{1},h_{2}) S ± ( h 1 , h 2 ) be the Koecher operators at the critical parameters, as in (3.47). Then
g ± = span C { S ± ( h 1 , h 2 ) : h 1 , h 2 ∈ W C } ≅ e 6 . \mathfrak{g}_{\pm}=\operatorname{span}_{\mathbb{C}}\{S_{\pm}(h_{1},h_{2}):h_{1},h_{2}\in W_{\mathbb{C}}\}\cong\mathfrak{e}_{6}. g ± = span C { S ± ( h 1 , h 2 ) : h 1 , h 2 ∈ W C } ≅ e 6 .
Thus the quadratic adjoints F ± F_{\pm} F ± and F ∓ F_{\mp} F ∓ give an explicit realization of e 6 \mathfrak{e}_{6} e 6 on W C W_{\mathbb{C}} W C .
Proof. Let G G G be the identity component of the stabilizer of N ± N_{\pm} N ± in G L ( W C ) GL(W_{\mathbb{C}}) G L ( W C ) , and write γ ∨ = ( γ T W ) − 1 \gamma^{\vee}=(\gamma^{T_W})^{-1} γ ∨ = ( γ T W ) − 1 for γ ∈ G \gamma\in G γ ∈ G . Invariance of N ± N_{\pm} N ± gives
F ± ( γ h ) = γ ∨ F ± ( h ) . F_{\pm}(\gamma h)=\gamma^{\vee}F_{\pm}(h). F ± ( γ h ) = γ ∨ F ± ( h ) .
The double-adjoint identity
F ∓ ( F ± ( h ) ) = 2 3 N ± ( h ) h F_{\mp}(F_{\pm}(h))=\frac{2}{3}N_{\pm}(h)h F ∓ ( F ± ( h )) = 3 2 N ± ( h ) h
then implies
F ∓ ( γ ∨ u ) = γ F ∓ ( u ) . F_{\mp}(\gamma^{\vee}u)=\gamma F_{\mp}(u). F ∓ ( γ ∨ u ) = γ F ∓ ( u ) .
Indeed, this follows first for u = F ± ( h ) u=F_{\pm}(h) u = F ± ( h ) . By Lemma 3.8 , the identity then extends to all of W C W_{\mathbb{C}} W C , since both sides are polynomial in u u u . Polarizing these identities and substituting into the formula for S ± S_{\pm} S ± gives
S ± ( γ h 1 , γ ∨ h 2 ) = γ S ± ( h 1 , h 2 ) γ − 1 . S_{\pm}(\gamma h_1,\gamma^{\vee}h_2)=\gamma S_{\pm}(h_1,h_2)\gamma^{-1}. S ± ( γ h 1 , γ ∨ h 2 ) = γ S ± ( h 1 , h 2 ) γ − 1 .
Consequently, the linear span of the Koecher operators is invariant under the adjoint action of G G G , hence is an ideal of g ± \mathfrak{g}_{\pm} g ± . It is nonzero by Proposition 3.18 . Since g ± \mathfrak{g}_{\pm} g ± is simple, this span equals g ± \mathfrak{g}_{\pm} g ± .
Remark 3.35. The two critical stabilizers realize dual actions under the W W W -pairing:
g ∓ = { − X T W : X ∈ g ± } . \mathfrak{g}_{\mp}=\{-X^{T_W}:X\in\mathfrak{g}_{\pm}\}. g ∓ = { − X T W : X ∈ g ± } .
Indeed, X ↦ − X T W X\mapsto-X^{T_W} X ↦ − X T W is the Lie algebra isomorphism for which ⟨ X h , k ⟩ W + ⟨ h , − X T W k ⟩ W = 0 \langle Xh,k\rangle_W+\langle h,-X^{T_W}k\rangle_W=0 ⟨ X h , k ⟩ W + ⟨ h , − X T W k ⟩ W = 0 . Thus the action of g ∓ \mathfrak{g}_{\mp} g ∓ corresponds to the contragredient of the action of g ± \mathfrak{g}_{\pm} g ± .
The W W W -pairing identifies W C W_{\mathbb{C}} W C with its dual as G 2 G_2 G 2 -modules, but not as e 6 \mathfrak{e}_6 e 6 -modules. Accordingly, although the Koecher operators are expressed as a G 2 G_2 G 2 -equivariant map
W C ⊗ W C ⟶ g ± , W_{\mathbb{C}}\otimes W_{\mathbb{C}}\longrightarrow\mathfrak{g}_{\pm}, W C ⊗ W C ⟶ g ± ,
their natural e 6 \mathfrak{e}_6 e 6 -equivariant interpretation is
W C ⊗ W C ∗ ⟶ e 6 , h 1 ⊗ h 2 b ⟼ S ± ( h 1 , h 2 ) , W_{\mathbb{C}}\otimes W_{\mathbb{C}}^*\longrightarrow\mathfrak{e}_6,\qquad h_1\otimes h_2^b\longmapsto S_{\pm}(h_1,h_2), W C ⊗ W C ∗ ⟶ e 6 , h 1 ⊗ h 2 b ⟼ S ± ( h 1 , h 2 ) ,
where h 2 b = ⟨ h 2 , ⋅ ⟩ W h_2^b=\langle h_2,\cdot\rangle_W h 2 b = ⟨ h 2 , ⋅ ⟩ W . In terms of the minuscule representations identified in Theorem 3.32 , this map realizes, up to a nonzero scalar, the projection onto the adjoint 78-dimensional summand in the decomposition into irreducible e 6 \mathfrak{e}_6 e 6 -modules:
27 ⊗ 27 ∗ = 1 ⊕ 78 ⊕ 650. 27\otimes27^*=1\oplus78\oplus650. 27 ⊗ 2 7 ∗ = 1 ⊕ 78 ⊕ 650.
These operators are related to the Tits–Kantor–Koecher construction in the theory of Jordan algebras [57 2.25–2.26 and 14.31] .
We have determined the stabilizer of every member of the projective pencil and characterized the two critical parameters in the affine family by the double-adjoint identity. In § 5 we identify the full stabilizer groups, prove uniqueness of their invariant cubics, and identify the critical cubics with the complex Albert norm.
E6, the minuscule 27, and the Albert norm
In Section 3, we identified the critical cubics N ± = C 0 + λ ± C φ N_{\pm}=C_0+\lambda_{\pm}C_{\varphi} N ± = C 0 + λ ± C φ and showed that their infinitesimal stabilizers g ± = s t a b ( N ± ) \mathfrak{g}_{\pm}=\mathop{\rm stab}(N_{\pm}) g ± = stab ( N ± ) are simple Lie algebras of type E 6 E_6 E 6 , each acting on W C W_{\mathbb{C}} W C through one of the two minuscule representations V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) (Theorem 3.32 ). We now determine the full stabilizer groups, establish the uniqueness of their invariant cubics, and identify N ± N_{\pm} N ± with complex Albert norms.
The distinction between a cubic norm and a unital algebra is important here. The G 2 G_2 G 2 -structure canonically determines the two critical norms, but a corresponding Jordan product requires a choice of unit of nonzero norm. We develop the resulting Albert algebra structures, describe their F 4 F_4 F 4 stabilizers and residual G 2 G_2 G 2 symmetries, and construct the real forms needed in Section 5.
Our approach combines direct arguments in the G 2 G_2 G 2 model with classical results about the Albert algebra. Section 4.1 determines the full stabilizer groups using connected algebraic groups and outer automorphisms of e 6 \mathfrak{e}_6 e 6 and g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C . Sections 4.2 and 4.3 establish the norm identification and develop the corresponding unital algebras, including transitivity on nonzero norm levels and their F 4 F_4 F 4 stabilizers. Section 4.4 uses the closed-orbit theorem and the classical stabilizers of a unit vector and an associative plane to analyze the residual G 2 G_2 G 2 symmetry. Section 4.5 distinguishes the pointwise Yukawa cubic from the critical Albert norms. Finally, Sections 4.6 and 4.7 construct compact E 6 E_6 E 6 forms using Cartan decomposition and identify the Euclidean Albert algebra associated with an S U ( 3 ) SU(3) S U ( 3 ) -type basepoint using the Faraut–Korányi classification. No further classification of simple Lie algebras is required.
The stabilizer group and the invariant cubic
The Lie algebra determines the identity component of the stabilizer. To determine the full group, we use the minuscule representation to exclude additional components, then describe the common stabilizer of two distinct members of the pencil to establish uniqueness of their invariant cubics.
Let
G 2 C = { O ∈ S O ( V C , g C ) : O ∗ φ C = φ C } G_2^{\mathbb{C}}=\{O\in SO(V_{\mathbb{C}},g_{\mathbb{C}}):O^{*}\varphi_{\mathbb{C}}=\varphi_{\mathbb{C}}\} G 2 C = { O ∈ S O ( V C , g C ) : O ∗ φ C = φ C }
be the connected complex algebraic group of type G 2 G_2 G 2 , with Lie algebra g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C . Regard W C = S 0 2 V C ∗ W_{\mathbb{C}}=S_0^2V_{\mathbb{C}}^{*} W C = S 0 2 V C ∗ as the space of traceless symmetric 2-tensors, with the complex-bilinear extensions of the conventions of Section 2. The natural left action is
ρ ( O ) h = ( O − 1 ) ∗ h , \rho(O)h=(O^{-1})^{*}h, ρ ( O ) h = ( O − 1 ) ∗ h ,
as in (2.20).
Under the metric identification (2.10), this is conjugation, g − 1 ( ρ ( O ) h ) = O ( g − 1 h ) O − 1 g^{-1}(\rho(O)h)=O(g^{-1}h)O^{-1} g − 1 ( ρ ( O ) h ) = O ( g − 1 h ) O − 1 , and its differential is the action ρ ( M ) \rho(M) ρ ( M ) of (2.21). This action preserves the W W W -pairing and both cubics C 0 C_0 C 0 and C φ C_{\varphi} C φ . Since its differential is injective (Section 2) and G 2 C G_2^{\mathbb{C}} G 2 C has trivial center, the representation is faithful. Its image is a closed connected algebraic subgroup of G L ( W C ) GL(W_{\mathbb{C}}) G L ( W C ) with Lie algebra ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) .
As in Section 1, E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) denotes the simply connected complex group of type E 6 E_6 E 6 . The undecorated e 6 \mathfrak{e}_6 e 6 and f 4 \mathfrak{f}_4 f 4 denote the complex Lie algebras, while g 2 \mathfrak{g}_2 g 2 is the compact real Lie algebra and g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C its complexification. To pass from g ± \mathfrak{g}_{\pm} g ± to the full stabilizer group, we use four standard facts over C \mathbb{C} C [22 , 49 , 54 ] :
a connected algebraic subgroup of G L ( W C ) GL(W_{\mathbb{C}}) G L ( W C ) is determined by its Lie algebra;
a connected semisimple group with Lie algebra e 6 \mathfrak{e}_6 e 6 is a central quotient of E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) ;
the center of E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) is the group μ 3 \mu_3 μ 3 of cube roots of unity, and it acts nontrivially, by scalars, on a minuscule 27 ;
Out ( e 6 ) ≅ Z / 2 \operatorname{Out}(\mathfrak{e}_6) \cong\mathbb{Z}/2 Out ( e 6 ) ≅ Z /2 , and its nontrivial element exchanges V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) .
Theorem 4.1. For either sign, the stabilizer of the critical cubic is
G ± : = Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) . G_{\pm} := \operatorname{Stab}_{GL(W_{\mathbb{C}})}(N_{\pm}) \cong E_6(\mathbb{C}). G ± := Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) . This is the faithful minuscule realization of the simply connected group. It is connected, and its center μ 3 \mu_3 μ 3 acts by scalar cube roots of unity. The stabilizer of the cubic line C N ± \mathbb{C}N_{\pm} C N ± is G ^ ± = G ± ⋅ C × \widehat{G}_{\pm}=G_{\pm}\cdot\mathbb{C}^{\times} G ± = G ± ⋅ C × , where C × \mathbb{C}^{\times} C × acts by scalars; its Lie algebra is g ± ⊕ C id W \mathfrak{g}_{\pm}\oplus\mathbb{C}\operatorname{id}_W g ± ⊕ C id W and its dimension is 79.
A classical construction of E 6 E_6 E 6 as the isometry group of a trilinear form on its 27-dimensional module is given by Aschbacher [1 §5] . Under the identification of Theorem 4.7 , G ± G_{\pm} G ± and G ^ ± \widehat{G}_{\pm} G ± become, respectively, the norm-isometry and norm-similarity groups of the complex Albert algebra [57 12.3(iii), 14.20] , [22 §15, p. 29] . The following proof derives their structure directly from Theorem 3.32 .
Proof. Write G ± ∘ G_{\pm}^{\circ} G ± ∘ for the identity component. Since its Lie algebra is g ± ≅ e 6 \mathfrak{g}_{\pm}\cong\mathfrak{e}_6 g ± ≅ e 6 , the group G ± ∘ G_{\pm}^{\circ} G ± ∘ is connected and almost simple of type E 6 E_6 E 6 , hence a central quotient of E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) . The induced representation of E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) on W C W_{\mathbb{C}} W C is the 27-dimensional minuscule representation. Its kernel is contained in the center μ 3 \mu_3 μ 3 , which acts nontrivially and therefore faithfully. Thus G ± ∘ ≅ E 6 ( C ) G_{\pm}^{\circ}\cong E_6(\mathbb{C}) G ± ∘ ≅ E 6 ( C ) , with center μ 3 id W \mu_3\operatorname{id}_W μ 3 id W .
Now consider the full group. Let γ ∈ G ± \gamma\in G_{\pm} γ ∈ G ± . Conjugation by γ \gamma γ preserves g ± \mathfrak{g}_{\pm} g ± and therefore induces an automorphism α \alpha α of g ± ≅ e 6 \mathfrak{g}_{\pm}\cong\mathfrak{e}_6 g ± ≅ e 6 . Since γ \gamma γ acts on W C W_{\mathbb{C}} W C , this automorphism must preserve the isomorphism class of the 27-dimensional e 6 \mathfrak{e}_6 e 6 -module W C W_{\mathbb{C}} W C . However, the nontrivial outer automorphism of e 6 \mathfrak{e}_6 e 6 exchanges the two nonisomorphic minuscule representations V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) . Thus α \alpha α must be inner.
Choose δ ∈ G ± ∘ \delta\in G_{\pm}^{\circ} δ ∈ G ± ∘ inducing the same automorphism. Then δ − 1 γ \delta^{-1}\gamma δ − 1 γ commutes with g ± \mathfrak{g}_{\pm} g ± . Since W C W_{\mathbb{C}} W C is irreducible under g ± \mathfrak{g}_{\pm} g ± (it is already irreducible under ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) ), Schur’s lemma gives δ − 1 γ = c id W \delta^{-1}\gamma=c\operatorname{id}_W δ − 1 γ = c id W for some c ∈ C × c\in\mathbb{C}^{\times} c ∈ C × . Both δ \delta δ and γ \gamma γ preserve the nonzero cubic N ± N_{\pm} N ± , so c 3 = 1 c^3=1 c 3 = 1 . But μ 3 id W \mu_3\operatorname{id}_W μ 3 id W is the center of G ± ∘ G_{\pm}^{\circ} G ± ∘ , as noted above. Hence δ − 1 γ ∈ G ± ∘ \delta^{-1}\gamma\in G_{\pm}^{\circ} δ − 1 γ ∈ G ± ∘ , and consequently γ ∈ G ± ∘ \gamma\in G_{\pm}^{\circ} γ ∈ G ± ∘ . Therefore G ± = G ± ∘ G_{\pm}=G_{\pm}^{\circ} G ± = G ± ∘ .
The cubic line. If L ∈ G L ( W C ) L\in GL(W_{\mathbb{C}}) L ∈ G L ( W C ) satisfies N ± ( L h ) = c N ± ( h ) N_{\pm}(Lh)=cN_{\pm}(h) N ± ( L h ) = c N ± ( h ) for all h h h , choose t ∈ C × t\in\mathbb{C}^{\times} t ∈ C × with t 3 = c t^3=c t 3 = c . Then t − 1 L ∈ G ± t^{-1}L\in G_{\pm} t − 1 L ∈ G ± , proving G ^ ± = G ± ⋅ C × \widehat{G}_{\pm}=G_{\pm}\cdot\mathbb{C}^{\times} G ± = G ± ⋅ C × . At the Lie-algebra level, Φ N ± ( id W ) = 3 N ± \Phi_{N_{\pm}}(\operatorname{id}_W)=3N_{\pm} Φ N ± ( id W ) = 3 N ± gives
Φ N ± ( X ) = c N ± ⟺ X − c 3 id W ∈ g ± . \Phi_{N_{\pm}}(X)=cN_{\pm}\quad\Longleftrightarrow\quad X-\frac{c}{3}\operatorname{id}_W\in\mathfrak{g}_{\pm}. Φ N ± ( X ) = c N ± ⟺ X − 3 c id W ∈ g ± .
The sum g ± + C id W \mathfrak{g}_{\pm}+\mathbb{C}\operatorname{id}_W g ± + C id W is direct because Φ N ± ( id W ) = 3 N ± ≠ 0 \Phi_{N_{\pm}}(\operatorname{id}_W)=3N_{\pm}\ne0 Φ N ± ( id W ) = 3 N ± = 0 , and its dimension is 78 + 1 = 79 78+1=79 78 + 1 = 79 .
The uniqueness of the invariant cubic will follow from a description of the common stabilizer of two distinct members of the pencil. As before, let F ∈ { R , C } \mathbb{F} \in\{\mathbb{R},\mathbb{C}\} F ∈ { R , C } and N a , b = a C 0 + b C φ N_{a,b}=aC_0+bC_\varphi N a , b = a C 0 + b C φ on W F W_{\mathbb{F}} W F . All stabilizers below fix the indicated cubics exactly.
Proposition 4.2.
(1) If [ a 1 : b 1 ] ≠ [ a 2 : b 2 ] [a_1:b_1]\ne[a_2:b_2] [ a 1 : b 1 ] = [ a 2 : b 2 ] in P 1 ( F ) \mathbb{P}^{1}(\mathbb{F}) P 1 ( F ) , that is, a 1 b 2 − a 2 b 1 ≠ 0 a_1b_2-a_2b_1\ne0 a 1 b 2 − a 2 b 1 = 0 , then stab ( N a 1 , b 1 ) ∩ stab ( N a 2 , b 2 ) = ρ ( g 2 F ) \operatorname{stab}(N_{a_1,b_1})\cap\operatorname{stab}(N_{a_2,b_2})=\rho(\mathfrak{g}_2^{\mathbb{F}}) stab ( N a 1 , b 1 ) ∩ stab ( N a 2 , b 2 ) = ρ ( g 2 F ) .
(2) The intersection of stabilizers of the two critical cubics is given by
G + ∩ G − = μ 3 × ρ ( G 2 C ) , (115) G_+\cap G_-=\mu_3\times\rho(G_2^{\mathbb{C}}), \tag*{(115)} G + ∩ G − = μ 3 × ρ ( G 2 C ) , (115) a direct product of algebraic groups, in which μ 3 \mu_3 μ 3 acts by scalar cube roots of unity and meets ρ ( G 2 C ) \rho(G_2^{\mathbb{C}}) ρ ( G 2 C ) trivially. Relative to the original real form W ⊂ W C W\subset W_{\mathbb{C}} W ⊂ W C , for either sign,
G ± ∩ GL ( W ) = ρ ( G 2 ) , g ± ∩ End ( W ) = ρ ( g 2 ) , G_\pm\cap\operatorname{GL}(W)=\rho(G_2),\qquad\mathfrak{g}_\pm\cap\operatorname{End}(W)=\rho(\mathfrak{g}_2), G ± ∩ GL ( W ) = ρ ( G 2 ) , g ± ∩ End ( W ) = ρ ( g 2 ) , where real maps are identified with their complex-linear extensions.
Proof. (1) Lie algebras. The inclusion of ρ ( g 2 F ) \rho(\mathfrak{g}_2^{\mathbb{F}}) ρ ( g 2 F ) is Lemma 3.2. Conversely, linearity of Φ N ( X ) \Phi_N(X) Φ N ( X ) in N N N and invertibility of the coefficient matrix imply that an X X X annihilating both members annihilates C 0 C_0 C 0 and C φ C_\varphi C φ . By Propositions 3.21 and 3.23,
ker Φ C 0 ∩ ker Φ C φ = ρ ( s o ( 7 , F ) ) ∩ ρ ( g 2 F ) = ρ ( g 2 F ) . \ker\Phi_{C_0}\cap\ker\Phi_{C_\varphi}=\rho(\mathfrak{so}(7,\mathbb{F}))\cap\rho(\mathfrak{g}_2^{\mathbb{F}})=\rho(\mathfrak{g}_2^{\mathbb{F}}). ker Φ C 0 ∩ ker Φ C φ = ρ ( so ( 7 , F )) ∩ ρ ( g 2 F ) = ρ ( g 2 F ) . (2) Groups. Let H = G + ∩ G − H=G_+\cap G_- H = G + ∩ G − . By (1), Lie ( H ) = ρ ( g 2 C ) \operatorname{Lie}(H)=\rho(\mathfrak{g}_2^{\mathbb{C}}) Lie ( H ) = ρ ( g 2 C ) . Since ρ ( G 2 C ) ⊂ H \rho(G_2^{\mathbb{C}})\subset H ρ ( G 2 C ) ⊂ H is connected and has the same Lie algebra, it is the identity component H ∘ H^\circ H ∘ .
Every L ∈ H L\in H L ∈ H normalizes H ∘ H^\circ H ∘ and thus induces an automorphism of g 2 C \mathfrak{g}_2^{\mathbb{C}} g 2 C . Since G 2 G_2 G 2 has no nontrivial diagram automorphisms, every such automorphism is inner. Hence there is an O ∈ G 2 C O\in G_2^{\mathbb{C}} O ∈ G 2 C such that ρ ( O ) − 1 L \rho(O)^{-1}L ρ ( O ) − 1 L commutes with ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) . By irreducibility of W C W_{\mathbb{C}} W C and Schur’s lemma,
L = ω ρ ( O ) , ω ∈ C × . L=\omega\rho(O),\qquad\omega\in\mathbb{C}^{\times}. L = ω ρ ( O ) , ω ∈ C × . Since L L L and ρ ( O ) \rho(O) ρ ( O ) preserve N + N_+ N + , we have ω 3 = 1 \omega^3=1 ω 3 = 1 . Conversely, every such product preserves both critical cubics. The product is direct: if ω id W ∈ ρ ( G 2 C ) \omega\operatorname{id}_W\in\rho(G_2^{\mathbb{C}}) ω id W ∈ ρ ( G 2 C ) , preservation of the W W W -pairing gives ω 2 = 1 \omega^2=1 ω 2 = 1 , and hence ω = 1 \omega=1 ω = 1 . Therefore
G + ∩ G − = μ 3 × ρ ( G 2 C ) . G_+\cap G_-=\mu_3\times\rho(G_2^{\mathbb{C}}). G + ∩ G − = μ 3 × ρ ( G 2 C ) . Finally, both factors are preserved by complex conjugation. Since μ 3 ∩ R = { 1 } \mu_3\cap\mathbb{R}=\{1\} μ 3 ∩ R = { 1 } and the real points of ρ ( G 2 C ) \rho(G_2^{\mathbb{C}}) ρ ( G 2 C ) form ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) , the elements preserving W W W are precisely ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) . Since N + = N − N_+=N_- N + = N − , a real element of either critical group lies in their intersection, giving the asserted real points of each group. Likewise, conjugation exchanges g + \mathfrak{g}_+ g + and g − \mathfrak{g}_- g − . A real element of either lies in ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) by part (1), and the real elements of ρ ( g 2 C ) \rho(\mathfrak{g}_2^{\mathbb{C}}) ρ ( g 2 C ) are exactly ρ ( g 2 ) \rho(\mathfrak{g}_2) ρ ( g 2 ) .
Remark 4.3. The group μ 3 × ρ ( G 2 C ) \mu_3\times\rho(G_2^{\mathbb{C}}) μ 3 × ρ ( G 2 C ) is the normalizer of the irreducible subgroup ρ ( G 2 C ) \rho(G_2^{\mathbb{C}}) ρ ( G 2 C ) in either G ± G_\pm G ± ; for the compact form, see Wolf’s classification [59 Theorem 11.1 and its table] . Proposition 4.2 identifies that known group as the common stabilizer of the two critical cubics, using the standard normalizer argument [21 p. 34] .
Part (1) of Proposition 4.2 now gives uniqueness of the invariant cubic.
Proposition 4.4. For either sign, ( S 3 W C ∗ ) g ± = C N ± (S^{3}W_{\mathbb{C}}^{*})^{\mathfrak{g}_{\pm}}=\mathbb{C}N_{\pm} ( S 3 W C ∗ ) g ± = C N ± ; that is, the cubics invariant under the critical e 6 \mathfrak{e}_{6} e 6 form a complex line.
Proof. Every g ± \mathfrak{g}_{\pm} g ± -invariant cubic is also ρ ( g 2 C ) \rho(\mathfrak{g}_{2}^{\mathbb{C}}) ρ ( g 2 C ) -invariant, and hence lies in the pencil by Lemma 2.7. If such a cubic P P P were independent of N ± N_{\pm} N ± , then Proposition 4.2 (1) would give
g ± ⊆ stab ( N ± ) ∩ stab ( P ) = ρ ( g 2 C ) . \mathfrak{g}_{\pm}\subseteq\operatorname{stab}(N_{\pm})\cap\operatorname{stab}(P)=\rho(\mathfrak{g}_{2}^{\mathbb{C}}). g ± ⊆ stab ( N ± ) ∩ stab ( P ) = ρ ( g 2 C ) . This is impossible, since dim g ± = 78 > 14 \dim\mathfrak{g}_{\pm}=78>14 dim g ± = 78 > 14 .
For the classical minuscule 27 of E 6 E_{6} E 6 , uniqueness of the invariant cubic is also known from invariant theory [55, §4, Propositions 3 and 12; §5, Proposition 47], [57 12.4(ii)] . Proposition 4.4 establishes this uniqueness directly in the present G 2 G_{2} G 2 model.
The Albert norm
The complex Albert algebra carries a minuscule 27 of E 6 E_{6} E 6 and an invariant cubic norm. Using the uniqueness just proved, we will identify that norm with N ± N_{\pm} N ± . We first recall how a norm, together with a chosen unit, determines the Jordan algebra.
Recall that a Jordan algebra over R \mathbb{R} R or C \mathbb{C} C has a commutative bilinear product ∘ \circ ∘ satisfying the Jordan identity
x ∘ ( ( x ∘ x ) ∘ y ) = ( x ∘ x ) ∘ ( x ∘ y ) . (116) x\circ((x\circ x)\circ y)=(x\circ x)\circ(x\circ y). \tag*{(116)} x ∘ (( x ∘ x ) ∘ y ) = ( x ∘ x ) ∘ ( x ∘ y ) . (116)
Jordan algebras over these fields are power-associative: the subalgebra generated by one element is associative, so the powers of that element are unambiguous [51 §3.10] . A Jordan algebra is special if it embeds in an associative algebra endowed with the symmetrized product x ∘ y = ( x y + y x ) / 2 x\circ y=(xy+yx)/2 x ∘ y = ( x y + y x ) /2 , and exceptional otherwise [51 §§1.1–1.3] . Let O C = O ⊗ R C \mathbb{O}_{\mathbb{C}}=\mathbb{O}\otimes_{\mathbb{R}}\mathbb{C} O C = O ⊗ R C be the complexified octonion algebra. Octonion conjugation extends complex-linearly, with norm n O ( a ) = a a ˉ n_{\mathbb{O}}(a)=a\bar{a} n O ( a ) = a a ˉ and trace t O ( a ) = a + a ˉ t_{\mathbb{O}}(a)=a+\bar{a} t O ( a ) = a + a ˉ . A particular case of a Jordan algebra is the complex Albert algebra A = H 3 ( O C ) \mathbb{A}=H_{3}(\mathbb{O}_{\mathbb{C}}) A = H 3 ( O C ) that consists of matrices
x = ( α 1 a 3 a ˉ 2 a ˉ 3 α 2 a 1 a 2 a ˉ 1 α 3 ) , α i ∈ C , a i ∈ O C , x=\begin{pmatrix}
\alpha_{1} & a_{3} & \bar{a}_{2}\\
\bar{a}_{3} & \alpha_{2} & a_{1}\\
a_{2} & \bar{a}_{1} & \alpha_{3}
\end{pmatrix},
\qquad\alpha_{i}\in\mathbb{C},\quad a_{i}\in\mathbb{O}_{\mathbb{C}}, x = α 1 a ˉ 3 a 2 a 3 α 2 a ˉ 1 a ˉ 2 a 1 α 3 , α i ∈ C , a i ∈ O C ,
with unit 1 1 1 and product x ∘ y = ( x y + y x ) / 2 x\circ y=(xy+yx)/2 x ∘ y = ( x y + y x ) /2 [51 §1.5, p. 224] ; see also Baez [2 §3] . Here Hermitian symmetry uses octonion conjugation and does not conjugate the complex coefficients. Its cubic norm is
N A ( x ) = α 1 α 2 α 3 − ∑ i = 1 3 α i n O ( a i ) + t O ( ( a 1 a 2 ) a 3 ) N_{\mathbb{A}}(x)=\alpha_{1}\alpha_{2}\alpha_{3}-\sum_{i=1}^{3}\alpha_{i}n_{\mathbb{O}}(a_{i})+t_{\mathbb{O}}((a_{1}a_{2})a_{3}) N A ( x ) = α 1 α 2 α 3 − i = 1 ∑ 3 α i n O ( a i ) + t O (( a 1 a 2 ) a 3 )
[51 Theorem 15, p. 244] . The trace makes the last term independent of its parenthesization. The algebra is simple [51 Theorem 1, p. 224] . The Albert algebra A \mathbb{A} A is exceptional [51 Theorem 1 and Corollary 18(c)] .
The algebra product is recovered from ( N A , 1 ) (N_A,1) ( N A , 1 ) by the classical construction of Springer and McCrimmon [48 Theorem 5, p. 504] . We recall the conventions needed below, following Petersson [51 §5.2, Theorem 13 and §5.3] .
For a cubic form N N N on a finite-dimensional complex vector space and a point e e e with N ( e ) = 1 N(e)=1 N ( e ) = 1 , define the based trace form by
⟨ x , y ⟩ e = − ∂ x ∂ y log N ∣ e = d N ∣ e ( x ) d N ∣ e ( y ) − d 2 N ∣ e ( x , y ) = 9 N ( e , e , x ) N ( e , e , y ) − 6 N ( e , x , y ) . (117) \langle x,y\rangle_e=-\partial_x\partial_y\log N\big|_e
=dN\big|_e(x)\,dN\big|_e(y)-d^2N\big|_e(x,y)
=9N(e,e,x)N(e,e,y)-6N(e,x,y). \tag*{(117)} ⟨ x , y ⟩ e = − ∂ x ∂ y log N e = d N e ( x ) d N e ( y ) − d 2 N e ( x , y ) = 9 N ( e , e , x ) N ( e , e , y ) − 6 N ( e , x , y ) . (117)
The last expression uses the symmetric trilinear polarization of N N N , normalized as in (34) , and makes the definition independent of any choice of logarithm. Substituting y = e y=e y = e in (117) gives the associated linear trace:
tr e ( x ) : = ⟨ x , e ⟩ e = 3 N ( e , e , x ) , tr e ( e ) = 3. (118) \operatorname{tr}_e(x):=\langle x,e\rangle_e=3N(e,e,x),\qquad\operatorname{tr}_e(e)=3. \tag*{(118)} tr e ( x ) := ⟨ x , e ⟩ e = 3 N ( e , e , x ) , tr e ( e ) = 3. (118)
To express the based algebra operations and construct the real structure in § 5.7 , we represent the based trace form using the fixed W W W -pairing. On W C W_{\mathbb{C}} W C , the based trace operator T e T_e T e is defined by
⟨ T e x , y ⟩ W = ⟨ x , y ⟩ e . (119) \langle T_e x,y\rangle_W=\langle x,y\rangle_e. \tag*{(119)} ⟨ T e x , y ⟩ W = ⟨ x , y ⟩ e . (119)
Explicit formulas for T e T_e T e and its inverse at a critical cubic are given in Lemma 4.8.
The pair ( N , e ) (N,e) ( N , e ) is nonsingular when the based trace form is nondegenerate [51 §5.3] . In that case there is a unique quadratic map x ↦ x ♯ x\mapsto x^\sharp x ↦ x ♯ , the based adjoint , such that ⟨ x ♯ , y ⟩ e = d N ∣ x ( y ) \langle x^\sharp,y\rangle_e=dN|_x(y) ⟨ x ♯ , y ⟩ e = d N ∣ x ( y ) . We use the unhalved polarization
x ♯ y = ( x + y ) ♯ − x ♯ − y ♯ , x ♯ x = 2 x ♯ . x\mathbin{\sharp}y=(x+y)^\sharp-x^\sharp-y^\sharp,\qquad x\mathbin{\sharp}x=2x^\sharp. x ♯ y = ( x + y ) ♯ − x ♯ − y ♯ , x ♯ x = 2 x ♯ .
A nonsingular pair is a cubic norm structure when it satisfies the following identities:
N ( e ) = 1 , e ♯ = e , x ♯ ♯ = N ( x ) x , d N ∣ x ( y ) = ⟨ x ♯ , y ⟩ e , e ♯ x = tr e ( x ) e − x . (120) \begin{aligned}
N(e)&=1,\qquad e^\sharp=e,\qquad x^{\sharp\sharp}=N(x)x,\\
dN|_x(y)&=\langle x^\sharp,y\rangle_e,\qquad e\mathbin{\sharp}x=\operatorname{tr}_e(x)e-x. \tag*{(120)}
\end{aligned} N ( e ) d N ∣ x ( y ) = 1 , e ♯ = e , x ♯♯ = N ( x ) x , = ⟨ x ♯ , y ⟩ e , e ♯ x = tr e ( x ) e − x . (120)
The first and fourth identities hold by construction here; the remaining conditions are the basepoint, adjoint and unit identities. These are polynomial identities, and are required to hold after scalar extension as well. Over C \mathbb{C} C , we can also define a bilinear product:
x ∘ e y = 1 2 ( x ♯ y + tr e ( x ) y + tr e ( y ) x − ( tr e ( x ) tr e ( y ) − ⟨ x , y ⟩ e ) e ) . (121) x\circ_e y=\frac{1}{2}\bigl(x\mathbin{\sharp}y+\operatorname{tr}_e(x)y+\operatorname{tr}_e(y)x-(\operatorname{tr}_e(x)\operatorname{tr}_e(y)-\langle x,y\rangle_e)e\bigr). \tag*{(121)} x ∘ e y = 2 1 ( x ♯ y + tr e ( x ) y + tr e ( y ) x − ( tr e ( x ) tr e ( y ) − ⟨ x , y ⟩ e ) e ) . (121)
as in [42 Proposition 3] . By the classical construction theorem, this product satisfies the Jordan identity (116) and has unit e e e [51 Theorem 13] . The resulting algebra J ( N , e ) \mathcal{J}(N,e) J ( N , e ) , equipped with its norm N N N , is a cubic Jordan algebra ; its based trace form is nondegenerate. We write x ∘ 2 = x ∘ e x x^{\circ2}=x\circ_e x x ∘ 2 = x ∘ e x and, more generally, x ∘ n x^{\circ n} x ∘ n for its Jordan powers, with x ∘ 0 = e x^{\circ0}=e x ∘ 0 = e . The same construction theorem gives the characteristic identity
x ∘ 3 − tr e ( x ) x ∘ 2 + tr e ( x ♯ ) x − N ( x ) e = 0. (122) x^{\circ3}-\operatorname{tr}_e(x)x^{\circ2}+\operatorname{tr}_e(x^\sharp)x-N(x)e=0. \tag*{(122)} x ∘ 3 − tr e ( x ) x ∘ 2 + tr e ( x ♯ ) x − N ( x ) e = 0. (122)
Here tr e ( x # ) \operatorname{tr}_{e}(x^\#) tr e ( x # ) is the quadratic trace. An element x x x is invertible exactly when N ( x ) ≠ 0 N(x) \ne0 N ( x ) = 0 , with Jordan inverse x − 1 = N ( x ) − 1 x # x^{-1}=N(x)^{-1}x^\# x − 1 = N ( x ) − 1 x # [51 Theorem 13, p. 241] .
For a finite-dimensional unital Jordan algebra, its rank, also called its degree here, is the largest degree of an element’s minimal polynomial in the Jordan powers. On the open dense set of elements attaining this degree r r r , the minimal polynomial is the generic characteristic polynomial t r − a 1 ( x ) t r − 1 + ⋯ + ( − 1 ) r a r ( x ) t^r-a_1(x)t^{r-1}+\cdots+(-1)^r a_r(x) t r − a 1 ( x ) t r − 1 + ⋯ + ( − 1 ) r a r ( x ) . Its coefficients extend polynomially to the whole algebra; the Jordan determinant is a r ( x ) a_r(x) a r ( x ) [19 §II.2, pp. 27–30] . Thus (122) bounds the rank by three; when the rank is three, the determinant equals N N N by comparison of constant coefficients. In A A A , diagonal matrices with three distinct entries have minimal polynomial of degree three. Thus A A A has rank three, and N A N_A N A is its Jordan determinant.
The nonsingularity and adjoint identity above are also the admissibility conditions in [42 Definition 2] . Albert algebras are nonsingular cubic Jordan algebras [51 Corollary 18(a)] .
Associated with this product is the quadratic operator U x U_x U x , defined by
U x y = 2 x ∘ e ( x ∘ e y ) − x ∘ 2 ∘ e y . (123) U_xy = 2x \mathbin{\circ_e} (x \mathbin{\circ_e} y) - x^{\circ2} \mathbin{\circ_e} y. \tag*{(123)} U x y = 2 x ∘ e ( x ∘ e y ) − x ∘ 2 ∘ e y . (123)
Equivalently, in terms of the cubic norm and its adjoint,
U x y = ⟨ x , y ⟩ e x − x # # y . U_xy = \langle x,y\rangle_e x - x^\# \# y. U x y = ⟨ x , y ⟩ e x − x # # y .
The equivalence of these expressions is part of the classical cubic norm construction [51 Theorem 13] .
Lemma 4.5 (Quadratic operators). Let ( N , e ) (N,e) ( N , e ) be a nonsingular cubic norm structure, with associated Jordan algebra J ( N , e ) J(N,e) J ( N , e ) . For every x , y x,y x , y ,
U x e = x ∘ 2 , N ( U x y ) = N ( x ) 2 N ( y ) . (124) U_xe = x^{\circ2}, \qquad N(U_xy) = N(x)^2N(y). \tag*{(124)} U x e = x ∘ 2 , N ( U x y ) = N ( x ) 2 N ( y ) . (124) If N ( x ) ≠ 0 N(x) \ne0 N ( x ) = 0 , then U x U_x U x is invertible, with inverse U x − 1 U_{x^{-1}} U x − 1 . Consequently, if N ( x ) 2 = 1 N(x)^2=1 N ( x ) 2 = 1 , then U x U_x U x preserves N N N and carries the unit e e e to x ∘ 2 x^{\circ2} x ∘ 2 .
Proof. Since e e e is the unit of the Jordan algebra, (4.9) gives
U x e = 2 x ∘ e ( x ∘ e e ) − x ∘ 2 ∘ e e = 2 x ∘ 2 − x ∘ 2 = x ∘ 2 . U_xe = 2x \mathbin{\circ_e} (x \mathbin{\circ_e} e) - x^{\circ2} \mathbin{\circ_e} e
= 2x^{\circ2} - x^{\circ2} = x^{\circ2}. U x e = 2 x ∘ e ( x ∘ e e ) − x ∘ 2 ∘ e e = 2 x ∘ 2 − x ∘ 2 = x ∘ 2 .
The second identity in (4.10) is the standard norm-composition identity [51 Theorem 13] . For invertible x x x , the inverse identity U x − 1 = U x − 1 U_x^{-1}=U_{x^{-1}} U x − 1 = U x − 1 is the standard identity for quadratic operators [23 33.11] . The final assertion follows immediately.
These constructions are preserved by linear changes of variables that preserve the cubic norm and basepoint. In the next lemma, the operations on the left are defined by ( N ′ , e ′ ) (N',e') ( N ′ , e ′ ) , and those on the right by ( N , e ) (N,e) ( N , e ) .
Lemma 4.6 (Change of variables). Let N N N and N ′ N' N ′ be cubic forms with points e e e and e ′ e' e ′ , where N ( e ) = 1 N(e)=1 N ( e ) = 1 , and let L L L be a linear isomorphism satisfying
N ′ ( L x ) = N ( x ) , L ( e ) = e ′ . N'(Lx)=N(x), \qquad L(e)=e'. N ′ ( Lx ) = N ( x ) , L ( e ) = e ′ .
Then N ′ ( e ′ ) = 1 N'(e') = 1 N ′ ( e ′ ) = 1 and
⟨ L x , L y ⟩ e ′ = ⟨ x , y ⟩ e and tr e ′ ( L x ) = tr e ( x ) , (125) \langle Lx, Ly\rangle_{e'} = \langle x, y\rangle_e \qquad\text{and} \qquad\operatorname{tr}_{e'}(Lx) = \operatorname{tr}_e(x), \tag*{(125)} ⟨ Lx , L y ⟩ e ′ = ⟨ x , y ⟩ e and tr e ′ ( Lx ) = tr e ( x ) , (125)
so ( N ′ , e ′ ) (N', e') ( N ′ , e ′ ) is nonsingular if and only if ( N , e ) (N,e) ( N , e ) is. In that case
( L x ) # ′ = L ( x # ) , (126) (Lx)^{\#'} = L(x^\#), \tag*{(126)} ( Lx ) # ′ = L ( x # ) , (126)
the same identity holds for the polarized adjoints, and ( N ′ , e ′ ) (N',e') ( N ′ , e ′ ) is a cubic norm structure if and only if ( N , e ) (N,e) ( N , e ) is. If it is, then
L ( x ∘ e y ) = L x ∘ e ′ L y , (127) L(x \circ_e y) = Lx \circ_{e'} Ly, \tag*{(127)} L ( x ∘ e y ) = Lx ∘ e ′ L y , (127)
and L L L is an isomorphism of the corresponding Jordan algebras J ( N , e ) \mathcal{J}(N,e) J ( N , e ) and J ( N ′ , e ′ ) \mathcal{J}(N',e') J ( N ′ , e ′ ) .
Proof. First, N ′ ( e ′ ) = N ′ ( L e ) = N ( e ) = 1 N'(e') = N'(Le) = N(e) = 1 N ′ ( e ′ ) = N ′ ( L e ) = N ( e ) = 1 , so both based trace forms are defined. Taking the second derivative of the identity N ′ ( L x ) = N ( x ) N'(Lx) = N(x) N ′ ( Lx ) = N ( x ) at e e e , and using L ( e ) = e ′ L(e) = e' L ( e ) = e ′ , gives (4.11). Since L L L is an isomorphism, one based trace form is nondegenerate exactly when the other is. In that case, the defining property of the based adjoint and nondegeneracy of the based trace form ⟨ ⋅ , ⋅ ⟩ e ′ \langle\cdot,\cdot\rangle_{e'} ⟨ ⋅ , ⋅ ⟩ e ′ give (4.12). The polarized version follows by polarization. The identities (4.6) are polynomial identities in the norm, the basepoint, the based trace, the based trace form and the adjoint, and L L L intertwines each of these by (4.11) and (4.12); this remains true after scalar extension. Hence they hold for ( N ′ , e ′ ) (N',e') ( N ′ , e ′ ) exactly when they hold for ( N , e ) (N,e) ( N , e ) . Substituting these identities into the product formula (4.7) gives (4.13). Since L L L is a linear isomorphism, it is therefore an isomorphism of Jordan algebras [51 Corollary 18(b)] .
Theorem 4.7. For either sign there is a complex-linear isomorphism Υ : A → W C \Upsilon\colon A \to W_{\mathbb{C}} Υ : A → W C with Υ ∗ N ± = N A \Upsilon^*N_{\pm} = N_A Υ ∗ N ± = N A . Thus each critical cubic is linearly equivalent over C \mathbb{C} C to the determinant of the complex Albert algebra, with any nonzero scalar factor absorbed into the linear map.
This equivalence is over C \mathbb{C} C ; the critical members themselves are not real (Theorem 3.3 ). Note that this theorem establishes linear equivalence but does not specify a preferred coordinate map Υ \Upsilon Υ .
Proof. By a classical theorem, the exact linear stabilizer of N A N_A N A is the simply connected complex group of type E 6 E_6 E 6 [22 §15, p. 29] . As a representation of this group, the 27-dimensional vector space A A A is faithful. It is irreducible because E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) has no nontrivial irreducible representation of smaller dimension, and is therefore one of the two nonisomorphic minuscule modules V ( ω 1 E 6 ) V(\omega_1^{E_6}) V ( ω 1 E 6 ) and V ( ω 6 E 6 ) V(\omega_6^{E_6}) V ( ω 6 E 6 ) , as in Theorem 3.32 .
By Theorem 3.32 , W C W_{\mathbb{C}} W C is likewise one of these two modules for g ± ≅ e 6 \mathfrak{g}_{\pm} \cong\mathfrak{e}_6 g ± ≅ e 6 . Since the diagram automorphism of e 6 \mathfrak{e}_6 e 6 exchanges the two modules, we may choose an identification of the stabilizer Lie algebras under which the representations on A A A and W C W_{\mathbb{C}} W C are equivalent. This gives an intertwining linear isomorphism Υ ′ : A → W C \Upsilon'\colon A \to W_{\mathbb{C}} Υ ′ : A → W C .
By Proposition 4.4 , the space of invariant cubics on W C W_{\mathbb{C}} W C is one-dimensional. Hence
Υ ′ ∗ N ± = c N A \Upsilon'^*N_{\pm} = cN_A Υ ′ ∗ N ± = c N A
for some c ∈ C × c \in\mathbb{C}^{\times} c ∈ C × . Choose t ∈ C × t \in\mathbb{C}^{\times} t ∈ C × with t 3 c = 1 t^3c = 1 t 3 c = 1 and set Υ = t Υ ′ \Upsilon= t\Upsilon' Υ = t Υ ′ . Then Υ ∗ N ± = t 3 c N A = N A \Upsilon^*N_{\pm} = t^3cN_A = N_A Υ ∗ N ± = t 3 c N A = N A .
:::
Basepoints and Jordan products
Fix either sign and choose a basepoint , an element e ∈ W C e \in W_{\mathbb{C}} e ∈ W C with n : = N ± ( e ) ≠ 0 n := N_{\pm}(e) \ne0 n := N ± ( e ) = 0 . Write λ = λ ± \lambda= \lambda_{\pm} λ = λ ± and λ ˉ = λ ∓ \bar{\lambda} = \lambda_{\mp} λ ˉ = λ ∓ . Normalize the cubic at e e e by
N e = N ± / n , N e ( e ) = 1. N_e = N_{\pm}/n,\qquad N_e(e) = 1. N e = N ± / n , N e ( e ) = 1.
The following formulas express T e T_e T e and its inverse in terms of the quadratic adjoints of Definition 3.4.
Lemma 4.8. For n = N λ ( e ) ≠ 0 n = N_{\lambda}(e) \ne0 n = N λ ( e ) = 0 , the based trace operator for N λ / n N_{\lambda}/n N λ / n is
T e x = 3 n tr e ( x ) F λ ( e ) − 6 n F λ ( e , x ) . (128) T_e x = \frac{3}{n}\operatorname{tr}_e(x)F_{\lambda}(e)-\frac{6}{n}F_{\lambda}(e,x). \tag*{(128)} T e x = n 3 tr e ( x ) F λ ( e ) − n 6 F λ ( e , x ) . (128) It is invertible, with
T e − 1 y = ⟨ e , y ⟩ W e − F λ ˉ ( F λ ( e ) , y ) . (129) T_e^{-1}y=\langle e,y\rangle_W e-F_{\bar{\lambda}}(F_{\lambda}(e),y). \tag*{(129)} T e − 1 y = ⟨ e , y ⟩ W e − F λ ˉ ( F λ ( e ) , y ) . (129) Proof. Differentiating N λ / n N_{\lambda}/n N λ / n gives
tr e ( x ) = 3 n ⟨ F λ ( e ) , x ⟩ W , \operatorname{tr}_e(x)=\frac{3}{n}\langle F_{\lambda}(e),x\rangle_W, tr e ( x ) = n 3 ⟨ F λ ( e ) , x ⟩ W , d 2 ( N λ / n ) ∣ e ( x , y ) = 6 n ⟨ F λ ( e , x ) , y ⟩ W . d^2(N_{\lambda}/n)|_e(x,y)=\frac{6}{n}\langle F_{\lambda}(e,x),y\rangle_W. d 2 ( N λ / n ) ∣ e ( x , y ) = n 6 ⟨ F λ ( e , x ) , y ⟩ W . Substitute these in (4.3) and use (4.5) to obtain (4.14). Put f = F λ ( e ) f=F_{\lambda}(e) f = F λ ( e ) . Differentiating F λ ˉ ( F λ ( x ) ) = 2 3 N λ ( x ) x F_{\bar{\lambda}}(F_{\lambda}(x))=\frac{2}{3}N_{\lambda}(x)x F λ ˉ ( F λ ( x )) = 3 2 N λ ( x ) x from Proposition 3.7 at x = e x=e x = e gives
4 F λ ˉ ( f , F λ ( e , x ) ) = 2 ⟨ f , x ⟩ W e + 2 3 n x . 4F_{\bar{\lambda}}(f,F_{\lambda}(e,x))=2\langle f,x\rangle_W e+\frac{2}{3}nx. 4 F λ ˉ ( f , F λ ( e , x )) = 2 ⟨ f , x ⟩ W e + 3 2 n x . Since F λ ˉ ( f ) = 2 3 n e F_{\bar{\lambda}}(f)=\frac{2}{3}ne F λ ˉ ( f ) = 3 2 n e and ⟨ e , f ⟩ W = n \langle e,f\rangle_W=n ⟨ e , f ⟩ W = n , substitution of (4.14) yields
F λ ˉ ( f , T e x ) = 3 n ⟨ f , x ⟩ W e − x , ⟨ e , T e x ⟩ W = 3 n ⟨ f , x ⟩ W . F_{\bar{\lambda}}(f,T_e x)=\frac{3}{n}\langle f,x\rangle_W e-x,\qquad
\langle e,T_e x\rangle_W=\frac{3}{n}\langle f,x\rangle_W. F λ ˉ ( f , T e x ) = n 3 ⟨ f , x ⟩ W e − x , ⟨ e , T e x ⟩ W = n 3 ⟨ f , x ⟩ W . Multiplying the second identity by e e e and subtracting the first shows that the right side of (4.15) is a left inverse of T e T_e T e . Since W C W_{\mathbb{C}} W C is finite-dimensional, it is the inverse.
There is no rank or projection condition on e e e . In particular, the geometric basepoints E u E_u E u and E Σ E_{\Sigma} E Σ of (2.14), built from orthogonal projection tensors and studied in § 5.4 , are convenient examples of possible Jordan units. The next theorem proves that ( N e , e ) (N_e,e) ( N e , e ) is a nonsingular cubic norm structure and we denote its Jordan algebra by J e = J ~ ( N e , e ) J_e=\widetilde{J}(N_e,e) J e = J ( N e , e ) . Table 3 collects the pairings and normalizations used in the explicit operations below. The based adjoint uses the normalized cubic and the based trace form, whereas F λ F_{\lambda} F λ uses the fixed W W W -pairing.
Theorem 4.9.
(1) For every basepoint e e e , ( N e , e ) (N_e,e) ( N e , e ) is a nonsingular cubic norm structure and J ~ e ≅ A \widetilde{J}_e \cong A J e ≅ A , so J ~ e \widetilde{J}_e J e is simple, exceptional and of degree exactly 3. Its adjoint and Jordan product are
h # = 3 n ( ⟨ e , F λ ( h ) ⟩ W e − F λ ˉ ( F λ ( e ) , F λ ( h ) ) ) , h^{\#}=\frac{3}{n}\left(\langle e,F_{\lambda}(h)\rangle_W e-F_{\bar{\lambda}}(F_{\lambda}(e),F_{\lambda}(h))\right), h # = n 3 ( ⟨ e , F λ ( h ) ⟩ W e − F λ ˉ ( F λ ( e ) , F λ ( h )) ) , h ∘ e k = − 3 n F λ ˉ ( F λ ( e ) , F λ ( h , k ) ) + 1 2 ( tr e ( h ) k + tr e ( k ) h ) . (130) h\mathbin{\circ_e} k=-\frac{3}{n}F_{\bar{\lambda}}(F_{\lambda}(e),F_{\lambda}(h,k))
+\frac{1}{2}\left(\operatorname{tr}_e(h)k+\operatorname{tr}_e(k)h\right). \tag*{(130)} h ∘ e k = − n 3 F λ ˉ ( F λ ( e ) , F λ ( h , k )) + 2 1 ( tr e ( h ) k + tr e ( k ) h ) . (130)
Table 3 (PDF p. 61). Pairings and adjoints at a basepoint of a critical cubic.
For Υ \Upsilon Υ as in Theorem 4.7 , the point e A = Υ ( 1 ) e_A=\Upsilon(1) e A = Υ ( 1 ) has N ± ( e A ) = 1 N_\pm(e_A)=1 N ± ( e A ) = 1 , and Υ \Upsilon Υ is an isomorphism A → J ( N ± , e A ) \mathbb{A}\to\mathcal{J}(N_\pm,e_A) A → J ( N ± , e A ) .
(2) For any basepoints e , e ′ ∈ W C e,e'\in W_{\mathbb{C}} e , e ′ ∈ W C ,
e ′ ∈ G ± ⋅ e ⟺ N ± ( e ′ ) = N ± ( e ) . e'\in G_\pm\cdot e \quad\Longleftrightarrow\quad N_\pm(e')=N_\pm(e). e ′ ∈ G ± ⋅ e ⟺ N ± ( e ′ ) = N ± ( e ) .
Equivalently, each nonzero norm level { h ∈ W C : N ± ( h ) = c } \{h\in W_{\mathbb{C}}:N_\pm(h)=c\} { h ∈ W C : N ± ( h ) = c } is a single G ± ≅ E 6 ( C ) G_\pm\cong E_6(\mathbb{C}) G ± ≅ E 6 ( C ) -orbit. Allowing scalar multiplication, all basepoints lie in one orbit of G ^ ± = G ± ⋅ C × \widehat{G}_\pm=G_\pm\cdot\mathbb{C}^{\times} G ± = G ± ⋅ C × .
Proof. We first construct the algebra at one basepoint, prove transitivity using that algebra, and then transport the construction to an arbitrary basepoint.
The Albert basepoint. Let e A = Υ ( 1 ) e_A=\Upsilon(1) e A = Υ ( 1 ) , where Υ \Upsilon Υ is the norm equivalence of Theorem 4.7 . Then
N ± ( e A ) = N A ( 1 ) = 1. N_\pm(e_A)=N_A(1)=1. N ± ( e A ) = N A ( 1 ) = 1.
Since Υ \Upsilon Υ preserves the cubic norm and basepoint, (125) identifies the based trace forms, and (126) identifies the based adjoints. Consequently, ( N ± , e A ) (N_\pm,e_A) ( N ± , e A ) is a nonsingular cubic norm structure satisfying (120) . By (127) , Υ \Upsilon Υ also preserves the Jordan products, giving
J ( N ± , e A ) ≅ A . \mathcal{J}(N_\pm,e_A)\cong\mathbb{A}. J ( N ± , e A ) ≅ A .
Transitivity on nonzero norm levels. We first prove transitivity on N ± − 1 ( 1 ) N_\pm^{-1}(1) N ± − 1 ( 1 ) . Let h ∈ W C h\in W_{\mathbb{C}} h ∈ W C satisfy N ± ( h ) = 1 N_\pm(h)=1 N ± ( h ) = 1 , and regard h h h as an element of J ( N ± , e A ) \mathcal{J}(N_\pm,e_A) J ( N ± , e A ) . By the definition of invertibility, h h h is invertible since N ± ( h ) ≠ 0 N_\pm(h)\ne0 N ± ( h ) = 0 .
Since Jordan algebras are power-associative, the Jordan powers h ∘ j h^{\circ j} h ∘ j of h h h span a commutative associative subalgebra C [ h ] \mathbb{C}[h] C [ h ] . Here the powers are taken with respect to the product ∘ e A \circ_{e_A} ∘ e A , with h ∘ 0 = e A h^{\circ0}=e_A h ∘ 0 = e A , and a polynomial p ( t ) = ∑ j a j t j p(t)=\sum_j a_jt^j p ( t ) = ∑ j a j t j is evaluated in C [ h ] \mathbb{C}[h] C [ h ] as p ( h ) = ∑ j a j h ∘ j p(h)=\sum_j a_jh^{\circ j} p ( h ) = ∑ j a j h ∘ j . All polynomial evaluations, powers and minimal polynomials below are taken in this subalgebra, and U k U_k U k is the quadratic operator of J ( N ± , e A ) \mathcal{J}(N_\pm,e_A) J ( N ± , e A ) .
Moreover, h h h admits a square root k ∈ C [ h ] k\in\mathbb{C}[h] k ∈ C [ h ] . Indeed, by (122) , its minimal polynomial divides t 3 − tr e A ( h ) t 2 + tr e A ( h # ) t − 1 t^3-\operatorname{tr}_{e_A}(h)t^2+\operatorname{tr}_{e_A}(h^\#)t-1 t 3 − tr e A ( h ) t 2 + tr e A ( h # ) t − 1 , whose constant term is nonzero. A polynomial square root can therefore be constructed modulo each primary factor by a truncated binomial expansion, and by the Chinese remainder theorem these combine to give k = p ( h ) k=p(h) k = p ( h ) with k ∘ 2 = h k^{\circ2}=h k ∘ 2 = h . This is a standard square-root argument used in the theory of Albert isotopes [23 Corollary 31.25] .
By Lemma 4.5,
U k e A = k ∘ 2 = h . \mathcal{U}_k e_{\mathrm A}=k^{\circ2}=h. U k e A = k ∘ 2 = h .
Applying the norm identity in (4.10) gives
N ± ( k ) 2 = N ± ( h ) = 1. N_{\pm}(k)^2=N_{\pm}(h)=1. N ± ( k ) 2 = N ± ( h ) = 1.
Thus U k \mathcal{U}_k U k preserves N ± N_{\pm} N ± , belongs to G ± G_{\pm} G ± , and takes e A e_{\mathrm A} e A to h h h . This proves transitivity on N ± − 1 ( 1 ) N_{\pm}^{-1}(1) N ± − 1 ( 1 ) , which is the classical transitivity theorem for elements of norm one [36 Theorem 7] .
For any c ∈ C × c\in\mathbb{C}^{\times} c ∈ C × , multiplication by a cube root of c c c identifies N ± − 1 ( 1 ) N_{\pm}^{-1}(1) N ± − 1 ( 1 ) with N ± − 1 ( c ) N_{\pm}^{-1}(c) N ± − 1 ( c ) . Since scalar multiplication commutes with G ± G_{\pm} G ± , the group acts transitively on every nonzero norm level, as in the classical Albert setting [42 Proposition 12(i)] .
Arbitrary basepoints. Let e ∈ W C e\in W_{\mathbb C} e ∈ W C with n = N ± ( e ) ≠ 0 n=N_{\pm}(e)\ne0 n = N ± ( e ) = 0 , and choose t ∈ C × t\in\mathbb{C}^{\times} t ∈ C × with t 3 = n t^3=n t 3 = n . Then N ± ( e / t ) = 1 N_{\pm}(e/t)=1 N ± ( e / t ) = 1 , so by transitivity there is γ ∈ G ± \gamma\in G_{\pm} γ ∈ G ± such that
γ ( e A ) = e / t . \gamma(e_{\mathrm A})=e/t. γ ( e A ) = e / t .
Define L : W C → W C L:W_{\mathbb C}\to W_{\mathbb C} L : W C → W C by L ( x ) = γ − 1 ( x / t ) L(x)=\gamma^{-1}(x/t) L ( x ) = γ − 1 ( x / t ) . Since γ \gamma γ preserves N ± N_{\pm} N ± , we have
L ( e ) = e A , N ± ( L x ) = N ± ( x ) n = N e ( x ) . L(e)=e_{\mathrm A},\qquad N_{\pm}(Lx)=\frac{N_{\pm}(x)}{n}=N_e(x). L ( e ) = e A , N ± ( Lx ) = n N ± ( x ) = N e ( x ) .
Thus L L L preserves the normalized cubic norm and basepoint. Lemma 4.6 shows that ( N e , e ) (N_e,e) ( N e , e ) is a nonsingular cubic norm structure and identifies the Jordan products: by (4.13),
L ( x ∘ e y ) = L x ∘ e A L y . L(x\circ_e y)=Lx\circ_{e_{\mathrm A}}Ly. L ( x ∘ e y ) = Lx ∘ e A L y .
Hence
J e ≅ J ( N ± , e A ) ≅ A . \mathcal{J}_e\cong\mathcal{J}(N_{\pm},e_{\mathrm A})\cong\mathbb{A}. J e ≅ J ( N ± , e A ) ≅ A .
In particular, every J e \mathcal{J}_e J e is simple, exceptional and of degree 3.
The algebra operations. The defining adjoint relation and (4.5) give h # = 3 T e − 1 F λ ( h ) / n h^{\#}=3T_e^{-1}F_{\lambda}(h)/n h # = 3 T e − 1 F λ ( h ) / n . Its unhalved polarization satisfies
1 2 h # k = 3 n T e − 1 F λ ( h , k ) . \tfrac{1}{2}h\# k=\tfrac{3}{n}T_e^{-1}F_{\lambda}(h,k). 2 1 h # k = n 3 T e − 1 F λ ( h , k ) .
The based trace form (4.3) gives
tr e ( h ) tr e ( k ) − ⟨ h , k ⟩ e = 6 n ⟨ F λ ( e , h ) , k ⟩ W . \operatorname{tr}_e(h)\operatorname{tr}_e(k)-\langle h,k\rangle_e=\tfrac{6}{n}\langle F_{\lambda}(e,h),k\rangle_W. tr e ( h ) tr e ( k ) − ⟨ h , k ⟩ e = n 6 ⟨ F λ ( e , h ) , k ⟩ W .
Substituting these in (4.7), we obtain
h ∘ e k = 3 n T e − 1 F λ ( h , k ) + 1 2 ( tr e ( h ) k + tr e ( k ) h ) − 3 n ⟨ F λ ( e , h ) , k ⟩ W e . h\circ_e k=\tfrac{3}{n}T_e^{-1}F_{\lambda}(h,k)+\tfrac{1}{2}\bigl(\operatorname{tr}_e(h)k+\operatorname{tr}_e(k)h\bigr)-\tfrac{3}{n}\langle F_{\lambda}(e,h),k\rangle_W e. h ∘ e k = n 3 T e − 1 F λ ( h , k ) + 2 1 ( tr e ( h ) k + tr e ( k ) h ) − n 3 ⟨ F λ ( e , h ) , k ⟩ W e .
Now use (4.15). Symmetry of the polarized cubic gives ⟨ e , F λ ( h , k ) ⟩ W = ⟨ F λ ( e , h ) , k ⟩ W \langle e,F_{\lambda}(h,k)\rangle_W=\langle F_{\lambda}(e,h),k\rangle_W ⟨ e , F λ ( h , k ) ⟩ W = ⟨ F λ ( e , h ) , k ⟩ W , so the two scalar multiples of e e e cancel in the product. The same inverse formula applied to the adjoint proves both identities in (4.16).
Finally, since G ± G_{\pm} G ± preserves N ± N_{\pm} N ± and acts transitively on each nonzero norm level, the value of N ± N_{\pm} N ± completely determines the G ± G_{\pm} G ± -orbits on { N ± ≠ 0 } \{N_{\pm}\ne0\} { N ± = 0 } . Allowing scalar multiplication, G ^ ± = G ± ⋅ C × \widehat{G}_{\pm}=G_{\pm}\cdot\mathbb{C}^{\times} G ± = G ± ⋅ C × acts transitively on the full set of basepoints.
Fix any basepoint e 0 e_0 e 0 with n 0 = N ± ( e 0 ) ≠ 0 n_0 = N_{\pm}(e_0) \ne0 n 0 = N ± ( e 0 ) = 0 . For another basepoint e e e , put n = N ± ( e ) n = N_{\pm}(e) n = N ± ( e ) and choose t ∈ C × t \in\mathbb{C}^{\times} t ∈ C × with t 3 = n / n 0 t^3 = n/n_0 t 3 = n / n 0 . For γ ∈ G ± \gamma\in G_{\pm} γ ∈ G ± , write L = t − 1 γ L = t^{-1}\gamma L = t − 1 γ .
Corollary 4.10 (Transport to a model basepoint). There is γ ∈ G ± \gamma\in G_{\pm} γ ∈ G ± with γ e = t e 0 \gamma e = t e_0 γ e = t e 0 whose associated norm similarity L L L satisfies
L ( e ) = e 0 , N e 0 ( L x ) = N e ( x ) , L ( x ∘ e y ) = L x ∘ e 0 L y . (131) L(e) = e_0,\qquad N_{e_0}(Lx) = N_e(x),\qquad L(x \mathbin{\circ_e} y) = Lx \mathbin{\circ_{e_0}} Ly. \tag*{(131)} L ( e ) = e 0 , N e 0 ( Lx ) = N e ( x ) , L ( x ∘ e y ) = Lx ∘ e 0 L y . (131)
Proof. Theorem 4.9 (2) applies because N ± ( e ) = N ± ( t e 0 ) N_{\pm}(e) = N_{\pm}(te_0) N ± ( e ) = N ± ( t e 0 ) . Lemma 4.6 gives the product identity. □
Thus any chosen model basepoint yields, after a scalar adjustment, the same abstract unital Albert algebra as every other basepoint. The scalar adjustment is essential unless n = n 0 n = n_0 n = n 0 : the exact group G ± G_{\pm} G ± itself preserves the value of N ± N_{\pm} N ± . Neither t t t nor γ \gamma γ is canonically determined. The transport is pointwise over C \mathbb{C} C ; it need not preserve the original real G 2 G_2 G 2 -module or assemble into a global gauge transformation on a manifold. In particular, γ \gamma γ need not normalize the embedded ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) , so transport need not preserve the residual G 2 G_2 G 2 -stabilizer.
Remark 4.11. In (131) , take e 0 = E u e_0 = E_u e 0 = E u for a real unit vector u u u (a basepoint by Proposition 4.13), so γ e = t E u \gamma e = tE_u γ e = t E u . With v = s u v = su v = s u and s 2 = t s^2 = t s 2 = t , the model is
t E u = E v = 7 ( P v ) 0 = 7 P v − g ( v , v ) g , g ( v , v ) ≠ 0. tE_u = E_v = 7(P_v)_0 = 7P_v - g(v,v)g,\qquad g(v,v) \ne0. t E u = E v = 7 ( P v ) 0 = 7 P v − g ( v , v ) g , g ( v , v ) = 0.
Thus every basepoint is G ± G_{\pm} G ± -equivalent to the traceless part of a rank-one symmetric tensor. The vector v v v need not be real or of unit length. Requiring the target to be E u E_u E u itself forces N ± ( e ) = N ± ( E u ) N_{\pm}(e) = N_{\pm}(E_u) N ± ( e ) = N ± ( E u ) .
Changing the unit: isotopy. For an invertible p ∈ A p \in\mathcal{A} p ∈ A , the p p p -isotope A ( p ) \mathcal{A}^{(p)} A ( p ) has unit p − 1 p^{-1} p − 1 and quadratic operators U x ( p ) = U x U p U_x^{(p)} = U_xU_p U x ( p ) = U x U p [51 §3.11, p. 232] . Over C \mathbb{C} C , its product is recovered from these data: first x ( p ) 2 = U x ( p ) p − 1 x_{(p)}^2 = U_x^{(p)}p^{-1} x ( p ) 2 = U x ( p ) p − 1 , then polarize this square. Its norm is N A ( p ) = N A ( p ) N A N_{\mathcal{A}^{(p)}} = N_{\mathcal{A}}(p)N_{\mathcal{A}} N A ( p ) = N A ( p ) N A , and it is the Jordan algebra (121) of the cubic norm structure ( N A ( p ) N A , p − 1 ) (N_{\mathcal{A}}(p)N_{\mathcal{A}},p^{-1}) ( N A ( p ) N A , p − 1 ) [48 Theorem 2, pp. 500–501] , [23 33.11–33.12, pp. 326–327] . With a = Υ − 1 ( e ) a = \Upsilon^{-1}(e) a = Υ − 1 ( e ) and p = a − 1 p = a^{-1} p = a − 1 , these data are ( N A / N A ( a ) , a ) (N_{\mathcal{A}}/N_{\mathcal{A}}(a),a) ( N A / N A ( a ) , a ) , so Lemma 4.6 identifies A ( a − 1 ) \mathcal{A}^{(a^{-1})} A ( a − 1 ) with J e \mathcal{J}_e J e via Υ \Upsilon Υ . Thus Theorem 4.9 (1) realizes the classical isomorphism of complex Albert isotopes [23 Corollary 31.25] in the geometric cubic model.
Fixing the norm and a basepoint is equivalent to fixing the resulting unital Albert algebra. Its symmetry group is of type F 4 F_4 F 4 . Recall the based linear trace tr e \operatorname{tr}_e tr e of (118) . Since tr e ( e ) = 3 \operatorname{tr}_e(e) = 3 tr e ( e ) = 3 , the projection onto the 26-dimensional summand is h ↦ h − 1 3 tr e ( h ) e h \mapsto h - \frac{1}{3}\operatorname{tr}_e(h)e h ↦ h − 3 1 tr e ( h ) e . For a basepoint e e e , put F 4 , e = Stab G ± ( e ) F_{4,e} = \operatorname{Stab}_{G_{\pm}}(e) F 4 , e = Stab G ± ( e ) and f e = { X ∈ g ± : X e = 0 } \mathfrak{f}_e = \{X \in\mathfrak{g}_{\pm} : Xe = 0\} f e = { X ∈ g ± : X e = 0 } .
Proposition 4.12. For every basepoint e e e , f e = Der ( J e ) ≅ f 4 \mathfrak{f}_e = \operatorname{Der}(\mathcal{J}_e) \cong\mathfrak{f}_4 f e = Der ( J e ) ≅ f 4 , a simple Lie algebra of dimension 52 and rank 4. The f e \mathfrak{f}_e f e -module W C W_{\mathbb{C}} W C splits as W C = C e ⊕ 26 W_{\mathbb{C}} = \mathbb{C}e \oplus26 W C = C e ⊕ 26 , with 26 = { h : tr e ( h ) = 0 } 26 = \{h : \operatorname{tr}_e(h) = 0\} 26 = { h : tr e ( h ) = 0 } irreducible, and F 4 , e F_{4,e} F 4 , e is connected and isomorphic to the complex group F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) of type F 4 F_4 F 4 .
Proof. A derivation D D D of J e \mathcal{J}_e J e kills the unit, and exp ( t D ) \exp(tD) exp ( t D ) is an algebra automorphism for every t t t . Such automorphisms preserve the cubic norm, so D D D infinitesimally annihilates N e N_e N e , and hence N ± N_{\pm} N ± . Thus D ∈ f e D \in\mathfrak{f}_e D ∈ f e . Conversely, let X ∈ g ± X \in\mathfrak{g}_{\pm} X ∈ g ± satisfy X e = 0 Xe=0 X e = 0 . Then exp ( t X ) \exp(tX) exp ( tX ) preserves both N e N_e N e and e e e , and Lemma 4.6 shows that it preserves the Jordan product. Hence X X X is a derivation.
The remaining claims are transported through the isomorphism J e ≅ A \mathcal{J}_e \cong\mathcal{A} J e ≅ A of Theorem 4.9 (1). This isomorphism takes e e e to 1 1 1 and tr e \operatorname{tr}_e tr e to the trace of A \mathcal{A} A . It takes F 4 , e F_{4,e} F 4 , e to the stabilizer of 1 1 1 in the norm-isometry group of A \mathcal{A} A , and Der ( J e ) \operatorname{Der}(\mathcal{J}_e) Der ( J e ) to Der ( A ) \operatorname{Der}(\mathcal{A}) Der ( A ) . For the complex Albert algebra, Der ( A ) ≅ f 4 \operatorname{Der}(\mathcal{A}) \cong\mathfrak{f}_4 Der ( A ) ≅ f 4 [6 Theorem, p. 137] . Its trace-zero part is an irreducible 26 26 26 [61 Proposition 2.4.6(1)] . The stabilizer of 1 1 1 is the connected group F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) [57 4.6 and 14.24] , [22 §15, p. 29] , and at group level also [51 Theorem 5] . □ \square □
The transitivity in Theorem 4.9 (2) and the F 4 F_4 F 4 stabilizer give, for every c ∈ C × c \in\mathbb{C}^{\times} c ∈ C × and every e c e_c e c with N ± ( e c ) = c N_{\pm}(e_c)=c N ± ( e c ) = c ,
N ± − 1 ( c ) = G ± ⋅ e c ≅ E 6 ( C ) / F 4 ( C ) . N_{\pm}^{-1}(c)=G_{\pm}\cdot e_c \cong E_6(\mathbb{C})/F_4(\mathbb{C}). N ± − 1 ( c ) = G ± ⋅ e c ≅ E 6 ( C ) / F 4 ( C ) .
Each nonzero norm level is therefore a 26 26 26 -dimensional homogeneous hypersurface in W C W_{\mathbb{C}} W C , and the values c ∈ C × c \in\mathbb{C}^{\times} c ∈ C × distinguish these orbits of the exact group G ± G_{\pm} G ± .
The G 2 G_2 G 2 symmetry retained by a basepoint
A basepoint is needed to turn a critical norm into a Jordan product. We ask how much of the original G 2 G_2 G 2 symmetry survives this choice. No nonzero tensor is G 2 G_2 G 2 -invariant, and the bound below shows that at most an 8 8 8 -dimensional subgroup survives, attained at the line basepoints E u E_u E u of Proposition 4.13. This motivates the line-field construction of Section 5. The original group ρ ( G 2 ) \rho(G_2) ρ ( G 2 ) preserves both critical norms. After a basepoint e e e is chosen, the subgroup that also preserves the Jordan product is ρ ( G 2 ) ∩ F 4 , e \rho(G_2) \cap F_{4,e} ρ ( G 2 ) ∩ F 4 , e . For h ∈ W C h \in W_{\mathbb{C}} h ∈ W C , set
g 2 , h = { M ∈ g 2 : ρ ( M ) h = 0 } , g 2 , h C = { M ∈ g 2 C : ρ ( M ) h = 0 } . \mathfrak{g}_{2,h}=\{M\in\mathfrak{g}_2:\rho(M)h=0\},\qquad\mathfrak{g}^{\mathbb{C}}_{2,h}=\{M\in\mathfrak{g}^{\mathbb{C}}_2:\rho(M)h=0\}. g 2 , h = { M ∈ g 2 : ρ ( M ) h = 0 } , g 2 , h C = { M ∈ g 2 C : ρ ( M ) h = 0 } .
These are the annihilators , or infinitesimal stabilizers, of h h h in g 2 \mathfrak{g}_2 g 2 and in g 2 C \mathfrak{g}^{\mathbb{C}}_2 g 2 C . For real h h h , g 2 , h C \mathfrak{g}^{\mathbb{C}}_{2,h} g 2 , h C is the complexification of g 2 , h \mathfrak{g}_{2,h} g 2 , h , since ρ \rho ρ is real.
Since F 4 , e = Stab G ± ( e ) F_{4,e}=\operatorname{Stab}_{G_{\pm}}(e) F 4 , e = Stab G ± ( e ) and ρ ( G 2 ) ⊂ G ± \rho(G_2)\subset G_{\pm} ρ ( G 2 ) ⊂ G ± , the first identity below is immediate, and taking Lie algebras gives the second:
ρ ( G 2 ) ∩ F 4 , e = ρ ( Stab G 2 ( e ) ) , ρ ( g 2 ) ∩ f e = ρ ( g 2 , e ) . (132) \rho(G_2)\cap F_{4,e}=\rho(\operatorname{Stab}_{G_2}(e)),\qquad\rho(\mathfrak{g}_2)\cap\mathfrak{f}_e=\rho(\mathfrak{g}_{2,e}). \tag*{(132)} ρ ( G 2 ) ∩ F 4 , e = ρ ( Stab G 2 ( e )) , ρ ( g 2 ) ∩ f e = ρ ( g 2 , e ) . (132)
We compute this intersection for two geometric families of basepoints, constructed from line and associative-plane projections, and bound its dimension for every choice. These families are not claimed to classify all basepoints up to the original G 2 G_2 G 2 -action. The transport (4.17) uses the larger complex group G ± G_{\pm} G ± ; its relation to the original G 2 G_2 G 2 is discussed after Corollary 4.10.
For a real unit vector u u u and a 3 3 3 -plane Σ ⊂ V \Sigma\subset V Σ ⊂ V , recall the projection tensors P u = u ♭ ⊗ u ♭ P_u=u^\flat\otimes u^\flat P u = u ♭ ⊗ u ♭ and P Σ P_\Sigma P Σ of (2.13) and the corresponding test tensors E u E_u E u and E Σ E_\Sigma E Σ of (2.14). Here g − 1 P Σ = pr Σ g^{-1}P_\Sigma=\operatorname{pr}_\Sigma g − 1 P Σ = pr Σ and P Σ ⊥ = g − P Σ P_{\Sigma^\perp}=g-P_\Sigma P Σ ⊥ = g − P Σ . Associative planes are as in Definition 3.6.
Proposition 4.13 (Line and associative-plane basepoints). For a real unit vector u u u and an associative 3 3 3 -plane Σ \Sigma Σ , the test tensors are
E u = 7 P u − g , E Σ = 7 P Σ − 3 g = 4 P Σ − 3 P Σ ⊥ . (133) E_u=7P_u-g,\qquad E_\Sigma=7P_\Sigma-3g=4P_\Sigma-3P_{\Sigma^\perp}. \tag*{(133)} E u = 7 P u − g , E Σ = 7 P Σ − 3 g = 4 P Σ − 3 P Σ ⊥ . (133) Then, for the compact G 2 G_2 G 2 , the stabilizer subgroups are
Stab G 2 ( E u ) ≅ S U ( 3 ) × Z / 2 and Stab G 2 ( E Σ ) ≅ S O ( 4 ) . \operatorname{Stab}_{G_2}(E_u)\cong SU(3)\times\mathbb{Z}/2\qquad\text{and}\qquad\operatorname{Stab}_{G_2}(E_\Sigma)\cong SO(4). Stab G 2 ( E u ) ≅ S U ( 3 ) × Z /2 and Stab G 2 ( E Σ ) ≅ S O ( 4 ) .
The corresponding annihilators g 2 , E u ≅ s u ( 3 ) \mathfrak{g}_{2,E_u} \cong\mathfrak{su}(3) g 2 , E u ≅ su ( 3 ) and g 2 , E Σ ≅ s o ( 4 ) \mathfrak{g}_{2,E_\Sigma} \cong\mathfrak{so}(4) g 2 , E Σ ≅ so ( 4 ) have dimensions 8 and 6, respectively. In each case the fixed subspace of the annihilator acting on W W W is exactly the line spanned by the indicated basepoint. Their cubic values are
( C 0 ( E u ) , C φ ( E u ) ) = ( 210 , 84 ) , (C_0(E_u), C_\varphi(E_u)) = (210,84), ( C 0 ( E u ) , C φ ( E u )) = ( 210 , 84 ) ,
( C 0 ( E Σ ) , C φ ( E Σ ) ) = ( 84 , 1680 ) . (C_0(E_\Sigma), C_\varphi(E_\Sigma)) = (84,1680). ( C 0 ( E Σ ) , C φ ( E Σ )) = ( 84 , 1680 ) .
In particular, both tensors have nonzero critical norm and are basepoints.
Proof. The natural tensor action (25) gives
ρ ( O ) P u = P O u , ρ ( O ) P Σ = P O Σ ( O ∈ G 2 ) . \rho(O)P_u = P_{Ou}, \qquad\rho(O)P_\Sigma= P_{O\Sigma} \qquad(O \in G_2). ρ ( O ) P u = P O u , ρ ( O ) P Σ = P O Σ ( O ∈ G 2 ) .
As G 2 G_2 G 2 preserves g g g , and the projection tensors determine the corresponding unoriented line and plane, respectively,
Stab G 2 ( E u ) = Stab G 2 ( R u ) , Stab G 2 ( E Σ ) = Stab G 2 ( Σ ) . \operatorname{Stab}_{G_2}(E_u) = \operatorname{Stab}_{G_2}(\mathbb{R}u), \qquad\operatorname{Stab}_{G_2}(E_\Sigma) = \operatorname{Stab}_{G_2}(\Sigma). Stab G 2 ( E u ) = Stab G 2 ( R u ) , Stab G 2 ( E Σ ) = Stab G 2 ( Σ ) .
The classical stabilizer results give Stab G 2 ( u ) ≅ S U ( 3 ) \operatorname{Stab}_{G_2}(u) \cong SU(3) Stab G 2 ( u ) ≅ S U ( 3 ) and Stab G 2 ( Σ ) ≅ S O ( 4 ) \operatorname{Stab}_{G_2}(\Sigma) \cong SO(4) Stab G 2 ( Σ ) ≅ S O ( 4 ) [61 Theorems 1.9.1 and 1.10.1] , [33 Theorem IV.1.8] . The line stabilizer has an additional component, interchanging u u u and − u -u − u . The extension by Z / 2 \mathbb{Z}/2 Z /2 splits: for u = e 1 u=e_1 u = e 1 , the involution diag ( − 1 , 1 , − 1 , 1 , − 1 , 1 , − 1 ) ∈ G 2 \operatorname{diag}(-1,1,-1,1,-1,1,-1) \in G_2 diag ( − 1 , 1 , − 1 , 1 , − 1 , 1 , − 1 ) ∈ G 2 sends e 1 e_1 e 1 to − e 1 -e_1 − e 1 , and the general case follows by G 2 G_2 G 2 -transitivity on the unit sphere [61 Theorem 1.9.2] . This proves the group statements and, on differentiation, the assertions about the annihilators.
For the fixed-line claims, the two relevant real representations decompose as
V = R u ⊕ u ⊥ ( S U ( 3 ) ) , V = Σ ⊕ Σ ⊥ ( S O ( 4 ) ) . V = \mathbb{R}u \oplus u^\perp\quad(SU(3)), \qquad V = \Sigma\oplus\Sigma^\perp\quad(SO(4)). V = R u ⊕ u ⊥ ( S U ( 3 )) , V = Σ ⊕ Σ ⊥ ( S O ( 4 )) .
In each case the summands are inequivalent irreducible real modules. On u ⊥ ≅ C 3 u^\perp\cong\mathbb{C}^3 u ⊥ ≅ C 3 , the real commutant of S U ( 3 ) SU(3) S U ( 3 ) is spanned by the identity and the skew-adjoint complex structure, so its self-adjoint elements are real scalars. On the other summands, real Schur’s lemma gives the same conclusion. A self-adjoint endomorphism commuting with the stabilizer is therefore scalar on each summand. Taking its traceless part leaves exactly R E u \mathbb{R}E_u R E u or R E Σ \mathbb{R}E_\Sigma R E Σ , respectively.
Both tensors in (133) have nonzero critical norm. Indeed, C 0 ( h ) C_0(h) C 0 ( h ) and C φ ( h ) C_\varphi(h) C φ ( h ) are real for real h h h , whereas λ ± ∉ R \lambda_\pm\notin\mathbb{R} λ ± ∈ / R . Thus N ± ( h ) = 0 N_\pm(h)=0 N ± ( h ) = 0 would force both real cubics to vanish. The eigenvalues of g − 1 E u g^{-1}E_u g − 1 E u and g − 1 E Σ g^{-1}E_\Sigma g − 1 E Σ give
C 0 ( E u ) = 6 3 − 6 = 210 , C 0 ( E Σ ) = 3 ⋅ 4 3 − 4 ⋅ 3 3 = 84. C_0(E_u) = 6^3 - 6 = 210, \qquad C_0(E_\Sigma) = 3 \cdot4^3 - 4 \cdot3^3 = 84. C 0 ( E u ) = 6 3 − 6 = 210 , C 0 ( E Σ ) = 3 ⋅ 4 3 − 4 ⋅ 3 3 = 84.
For u = e 1 u=e_1 u = e 1 and the associative plane Σ = span ( e 1 , e 2 , e 3 ) \Sigma=\operatorname{span}(e_1,e_2,e_3) Σ = span ( e 1 , e 2 , e 3 ) , these are the test tensors E 1 E_1 E 1 and E Σ E_\Sigma E Σ of the proof of Proposition 3.7 . The values above agree with (3.16), which gives N λ ( E 1 ) = 42 ( 5 + 2 λ ) N_\lambda(E_1)=42(5+2\lambda) N λ ( E 1 ) = 42 ( 5 + 2 λ ) and N λ ( E Σ ) = 84 ( 1 + 20 λ ) N_\lambda(E_\Sigma)=84(1+20\lambda) N λ ( E Σ ) = 84 ( 1 + 20 λ ) . By G 2 G_2 G 2 -equivariance, these cubic values hold for every u u u and $\Sigma.
Remark 4.14. Either tensor can therefore serve as the model basepoint e 0 e_0 e 0 in (131) . For that transport, F 4 , e = γ − 1 F 4 , e 0 γ F_{4,e}=\gamma^{-1}F_{4,e_0}\gamma F 4 , e = γ − 1 F 4 , e 0 γ , since multiplication of e 0 e_0 e 0 by t t t does not change its stabilizer.
The transport by G ± G_{\pm} G ± does not identify the two models under G 2 G_2 G 2 , even projectively: the ratios of the original invariant cubics are
C φ ( E u ) C 0 ( E u ) = 2 5 , C φ ( E Σ ) C 0 ( E Σ ) = 20. \frac{C_{\varphi}(E_u)}{C_0(E_u)}=\frac{2}{5},\qquad\frac{C_{\varphi}(E_{\Sigma})}{C_0(E_{\Sigma})}=20. C 0 ( E u ) C φ ( E u ) = 5 2 , C 0 ( E Σ ) C φ ( E Σ ) = 20.
These ratios are unchanged by G 2 G_2 G 2 and by nonzero scalar multiplication. For example, choose s ± 3 = N ± ( E u ) / N ± ( E Σ ) s_{\pm}^3=N_{\pm}(E_u)/N_{\pm}(E_{\Sigma}) s ± 3 = N ± ( E u ) / N ± ( E Σ ) . Then E u E_u E u and s ± E Σ s_{\pm}E_{\Sigma} s ± E Σ have equal critical norm and are G ± G_{\pm} G ± -equivalent, but they are not projectively G 2 G_2 G 2 -equivalent.
The next proposition bounds the residual symmetry for every nonzero complex tensor, including tensors of zero critical norm.
Proposition 4.15. For every nonzero h ∈ W C h \in\mathrm{W}_{\mathbb{C}} h ∈ W C , dim C g 2 , h C ≤ 8 \dim_{\mathbb{C}}\mathfrak{g}^{\mathbb{C}}_{2,h}\leq8 dim C g 2 , h C ≤ 8 . Equality is attained both by the nonzero real multiples of E u E_u E u , for every real unit vector u u u , and by the rank-one null tensors P v = v ♭ ⊗ v ♭ P_v=v^{\flat}\otimes v^{\flat} P v = v ♭ ⊗ v ♭ , where 0 ≠ v ∈ V C 0\neq v\in\mathrm{V}_{\mathbb{C}} 0 = v ∈ V C and g ( v , v ) = 0 g(v,v)=0 g ( v , v ) = 0 . No G 2 C G^{\mathbb{C}}_2 G 2 C -orbit meets both families. The latter satisfy N ± ( P v ) = 0 N_{\pm}(P_v)=0 N ± ( P v ) = 0 and hence are not basepoints.
Proof. The irreducible G 2 C G^{\mathbb{C}}_2 G 2 C -module W C = V ( 2 , 0 ) \mathrm{W}_{\mathbb{C}}=\mathrm{V}(2,0) W C = V ( 2 , 0 ) has a unique closed projective orbit Y Y Y . It is the image of the null quadric
Q 5 = { [ v ] ∈ P ( V C ) : g ( v , v ) = 0 } Q^5=\{[v]\in\mathbb{P}(\mathrm{V}_{\mathbb{C}}):g(v,v)=0\} Q 5 = {[ v ] ∈ P ( V C ) : g ( v , v ) = 0 }
under the map [ v ] ↦ [ P v ] [v]\mapsto[P_v] [ v ] ↦ [ P v ] , and has dimension 5. It lies in the closure of every projective orbit [63 Theorem III.1.2] . If [ h ] ∉ Y [h]\notin Y [ h ] ∈ / Y , the projective orbit of [ h ] [h] [ h ] has dimension greater than 5, since Y Y Y lies in the boundary of its orbit closure and that boundary has strictly smaller dimension than the orbit. If [ h ] ∈ Y [h]\in Y [ h ] ∈ Y , a one-parameter subgroup of a maximal torus fixing [ h ] [h] [ h ] acts on C h \mathbb{C}h C h by a nontrivial character. It therefore rescales h h h through all of C × \mathbb{C}^{\times} C × , so the affine orbit has dimension 6. Thus in either case
dim C g 2 , h C = 14 − dim ( G 2 C ⋅ h ) ≤ 8. \dim_{\mathbb{C}}\mathfrak{g}^{\mathbb{C}}_{2,h}=14-\dim(G^{\mathbb{C}}_2\cdot h)\leq8. dim C g 2 , h C = 14 − dim ( G 2 C ⋅ h ) ≤ 8.
Equality holds for E u E_u E u by Proposition 4.13 and for every nonzero null P v P_v P v by the affine-orbit calculation. Finally, C 0 ( P v ) = C φ ( P v ) = 0 C_0(P_v)=C_{\varphi}(P_v)=0 C 0 ( P v ) = C φ ( P v ) = 0 , since g ( v , v ) = 0 g(v,v)=0 g ( v , v ) = 0 and P ( P v , P v ) = 0 P(P_v,P_v)=0 P ( P v , P v ) = 0 by (40) , so N ± ( P v ) = 0 N_{\pm}(P_v)=0 N ± ( P v ) = 0 ; these tensors are not basepoints and cannot lie in the orbit of a nonzero multiple of E u E_u E u , whose critical norm is nonzero.
For comparison, Darpö classifies the G 2 G_2 G 2 -orbits of real symmetric 2-tensors, written there as symmetric matrices; the orbit of E u E_u E u is the case treated in [11 Proposition 3.1(2)] . At a general complex tensor the stabilizer is much smaller: it is trivial by [32 Corollary 6] , which applies because dim V ( 2 , 0 ) = 27 > 14 = dim G 2 C \dim\mathrm{V}(2,0)=27>14=\dim G^{\mathbb{C}}_2 dim V ( 2 , 0 ) = 27 > 14 = dim G 2 C . Proposition 4.15 instead gives a uniform bound at every nonzero complex tensor, attained at the geometric basepoints E u E_u E u . We next compare these constructions with the Yukawa cubic.
Comparison with the Yukawa cubic
Lee and Leung describe an exceptional Jordan interpretation of the cubic on Λ 27 3 \Lambda^3_{27} Λ 27 3 , with an E 6 E_6 E 6 symmetry and an F 4 F_4 F 4 stabilizer after an S U ( 3 ) SU(3) S U ( 3 ) reduction [46 §1.2.1, Remark] . This interpretation was given without proof or an explicit comparison with a normalized Albert norm. Proposition 4.12 and Proposition 4.13 establish such a construction, with a normalized norm and a basepoint, for the critical cubics. We now compare it with their cubic. Recall from (47) that the pointwise Yukawa integrand is a nonzero multiple of C φ C_{\varphi} C φ .
An Albert algebra over R \mathbb{R} R is a real form of A A A , that is, a real Jordan algebra J J J with J ⊗ R C ≅ A J \otimes_{\mathbb{R}} \mathbb{C} \cong A J ⊗ R C ≅ A [51 §1.6, p. 225] . Its norm N J N_J N J complexifies to N A N_A N A .
Proposition 4.16. Let F = R \mathbb{F} = \mathbb{R} F = R or C \mathbb{C} C . For every Albert algebra J J J over F \mathbb{F} F , with norm N J N_J N J , there is no F \mathbb{F} F -linear isomorphism L : W F → J L: W_{\mathbb{F}} \to J L : W F → J and no s ∈ F × s \in\mathbb{F}^{\times} s ∈ F × such that C φ ( h , h , h ) = s N J ( L h ) C_{\varphi}(h,h,h) = sN_J(Lh) C φ ( h , h , h ) = s N J ( L h ) for every h ∈ W F h \in W_{\mathbb{F}} h ∈ W F . Thus C φ C_{\varphi} C φ is not linearly equivalent, even up to a nonzero scalar, to any real or complex Albert norm.
Proof. A nonzero rescaling leaves the exact infinitesimal stabilizer of a cubic unchanged, and pullback by a linear isomorphism conjugates it:
stab ( s N J ∘ L ) = L − 1 stab ( N J ) L . \operatorname{stab}(sN_J \circ L) = L^{-1}\operatorname{stab}(N_J)L. stab ( s N J ∘ L ) = L − 1 stab ( N J ) L .
An equivalence would therefore give stabilizers of equal dimensions. By Proposition 3.23 , dim F stab ( C φ ) = 14 \dim_{\mathbb{F}} \operatorname{stab}(C_{\varphi}) = 14 dim F stab ( C φ ) = 14 . The exact linear stabilizer of an Albert norm is an algebraic group of type E 6 E_6 E 6 [22 §15, p. 29] , whose Lie algebra has dimension 6 + 72 = 78 6 + 72 = 78 6 + 72 = 78 . Over R \mathbb{R} R it is a real form of the same complex Lie algebra and has real dimension 78. Since 14 ≠ 78 14 \ne78 14 = 78 , the proposed equivalence is impossible over either field.
Thus the pointwise Yukawa integrand on Λ 27 3 \Lambda^3_{27} Λ 27 3 is not equivalent to an Albert norm. The E 6 E_6 E 6 symmetry and the F 4 F_4 F 4 stabilizer of a basepoint only occur for the two critical members of the same pencil. At E u E_u E u , the compact G 2 G_2 G 2 subgroup preserving the based algebra is S U ( 3 ) ⋊ Z / 2 SU(3) \rtimes\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 , and choosing the unit vector u u u as additional data selects its S U ( 3 ) SU(3) S U ( 3 ) identity component. This discussion only concerns the pointwise cubic, rather than the integrated Yukawa coupling on the moduli space. We return to the comparison with Lee and Leung in § 6 , using parallel pointings.
Definition of a Jordan product requires the additional choice of a basepoint of nonzero norm and every such choice gives an algebra isomorphic to A A A , with symmetry group F 4 , e F_{4,e} F 4 , e . However, we have shown that there is no nonzero basepoint that is invariant under the original G 2 G_2 G 2 : Proposition 4.15 bounds the dimension of the annihilator g 2 , h C \mathfrak{g}^{\mathbb{C}}_{2,h} g 2 , h C of every nonzero h ∈ W C h \in W_{\mathbb{C}} h ∈ W C by 8 < 14 8 < 14 8 < 14 . Consequently no basepoint can be selected equivariantly from ( V , φ ) (V,\varphi) ( V , φ ) alone. Any choice retains a compact subgroup of dimension at most 8, with this maximum attained at the basepoints E u E_u E u and their nonzero real multiples. In § 6 we organize the choice of pointing by allowing real tangent lines to vary.
Real points and compact E 6 E_6 E 6 forms
The critical cubic lines are non-real and are exchanged by the complex conjugation h ↦ h ‾ h \mapsto\overline{h} h ↦ h associated with the original real module W W W , since λ − = λ + ‾ \lambda_- = \overline{\lambda_+} λ − = λ + in (63) and C 0 , C φ C_0,C_{\varphi} C 0 , C φ are real. We distinguish three real objects attached to these cubics. The real endomorphisms in either critical stabilizer form only ρ ( g 2 ) \rho(\mathfrak{g}_2) ρ ( g 2 ) , of real dimension 14; they do not form a real form of g ± \mathfrak{g}_{\pm} g ± . The Hermitian form induced by the metric, on the other hand, selects a compact real form of g ± \mathfrak{g}_{\pm} g ± , of real dimension 78, and no basepoint is needed for this. Finally, a real line in V V V determines a basepoint and a Euclidean real form of the corresponding complex Albert norm, with the corresponding complex Albert algebra, of real dimension 27. Here Euclidean means admitting a positive-definite associative inner product, equivalently being formally real; see § 5.7.3 . We construct the compact forms here and treat the based real algebras separately in § 5.7 .
For a complex-linear or antilinear map X X X of W C W_{\mathbb C} W C , write X ‾ ( h ) = X h ‾ \overline{X}(h)=X\overline{h} X ( h ) = X h . Thus X ‾ = X \overline{X}=X X = X precisely when X X X preserves W W W .
The W W W -pairing on W W W defines the Hermitian form ⟨ h , k ‾ ⟩ W \langle h,\overline{k}\rangle_W ⟨ h , k ⟩ W on W C W_{\mathbb C} W C . For an endomorphism X X X , X H X^{H} X H denotes its adjoint for this form, whereas X T W X^{T_W} X T W is its W W W -adjoint for the bilinear pairing (2.11). The induced pairing on endomorphisms is (3.22). We use the parity embeddings J Q s J^s_Q J Q s and J Q a J^a_Q J Q a from (3.40) to describe the compact real form selected by this pairing. Recall the real tensor space Q 64 Q_{64} Q 64 of Definition 3.24 and the map E ± \mathcal{E}_{\pm} E ± onto m ± \mathfrak{m}_{\pm} m ± introduced before Proposition 3.31. For the compact normalization, define
E ± c = ± i 7 E ± , E ± c ( Q ) = − J Q a ± i 7 J Q s . \mathcal{E}^{c}_{\pm}=\pm i\sqrt{7}\mathcal{E}_{\pm},\qquad\mathcal{E}^{c}_{\pm}(Q)=-J^{a}_{Q}\pm i\sqrt{7}J^{s}_{Q}. E ± c = ± i 7 E ± , E ± c ( Q ) = − J Q a ± i 7 J Q s .
The superscript c c c stands for compact. This rescaling does not change the complex image m ± \mathfrak{m}_{\pm} m ± . On Q 64 Q_{64} Q 64 the operators E ± ( Q ) \mathcal{E}_{\pm}(Q) E ± ( Q ) are Hermitian, so the imaginary factor makes E ± c ( Q ) \mathcal{E}^{c}_{\pm}(Q) E ± c ( Q ) skew-Hermitian, as the next lemma shows.
Lemma 4.17.
(i) The Hermitian form ⟨ h , k ‾ ⟩ W \langle h,\overline{k}\rangle_W ⟨ h , k ⟩ W on W C W_{\mathbb C} W C is positive definite and invariant under the natural action of O ( 7 ) O(7) O ( 7 ) .
(ii) For every real M ∈ s o ( 7 ) M\in\mathfrak{so}(7) M ∈ so ( 7 ) , the endomorphism ρ ( M ) \rho(M) ρ ( M ) of W C W_{\mathbb C} W C is skew-Hermitian.
(iii) For every Q ∈ Q 64 Q\in Q_{64} Q ∈ Q 64 , the operator E ± c ( Q ) \mathcal{E}^{c}_{\pm}(Q) E ± c ( Q ) lies in g ± \mathfrak{g}_{\pm} g ± and is skew-Hermitian.
Proof. The Hermitian form is positive definite because the W W W -pairing is positive definite on W W W . For O ∈ O ( 7 ) O\in O(7) O ∈ O ( 7 ) , the natural action is ρ ( O ) h = O h O − 1 \rho(O)h=OhO^{-1} ρ ( O ) h = O h O − 1 . Since O O O is real, cyclicity of trace gives
⟨ ρ ( O ) h , ρ ( O ) k ‾ ⟩ W = tr ( O h O − 1 O k O − 1 ‾ ) = ⟨ h , k ‾ ⟩ W . \langle\rho(O)h,\overline{\rho(O)k}\rangle_W=\operatorname{tr}(OhO^{-1}\overline{OkO^{-1}})=\langle h,\overline{k}\rangle_W. ⟨ ρ ( O ) h , ρ ( O ) k ⟩ W = tr ( O h O − 1 O k O − 1 ) = ⟨ h , k ⟩ W .
This proves (i). Differentiating along O ( t ) = exp ( t M ) O(t)=\exp(tM) O ( t ) = exp ( tM ) gives
⟨ ρ ( M ) h , k ‾ ⟩ W + ⟨ h , ρ ( M ) k ‾ ⟩ W = 0 , \langle\rho(M)h,\overline{k}\rangle_W+\langle h,\overline{\rho(M)k}\rangle_W=0, ⟨ ρ ( M ) h , k ⟩ W + ⟨ h , ρ ( M ) k ⟩ W = 0 ,
which proves (ii).
For (iii), Proposition 3.31 and the rescaling above give E ± c ( Q ) ∈ m ± ⊂ g ± \mathcal{E}^{c}_{\pm}(Q)\in\mathfrak{m}_{\pm}\subset\mathfrak{g}_{\pm} E ± c ( Q ) ∈ m ± ⊂ g ± . By Lemma 3.25 (iv), the operators J Q s J^s_Q J Q s and J Q a J^a_Q J Q a are real and respectively self-adjoint and skew-adjoint for the W W W -pairing. Their Hermitian adjoints therefore satisfy
( J Q s ) H = J Q s , ( J Q a ) H = − J Q a . (J^s_Q)^H=J^s_Q,\qquad(J^a_Q)^H=-J^a_Q. ( J Q s ) H = J Q s , ( J Q a ) H = − J Q a .
Consequently,
( E ± c ( Q ) ) H = J Q a ∓ i 7 J Q s = − E ± c ( Q ) . (\mathcal{E}^{c}_{\pm}(Q))^H=J^a_Q\mp i\sqrt{7}J^s_Q=-\mathcal{E}^{c}_{\pm}(Q). ( E ± c ( Q ) ) H = J Q a ∓ i 7 J Q s = − E ± c ( Q ) .
Let U ( W C ) U(W_{\mathbb{C}}) U ( W C ) be the unitary group of this Hermitian form, and let u ( W C ) \mathfrak{u}(W_{\mathbb{C}}) u ( W C ) be its real Lie algebra of skew-Hermitian endomorphisms. Define
K ± : = G ± ∩ U ( W C ) (134) K_{\pm}:=G_{\pm}\cap U(W_{\mathbb{C}}) \tag*{(134)} K ± := G ± ∩ U ( W C ) (134)
and
k ± : = ρ ( g 2 ) ⊕ E ± c ( Q 64 ) . (135) \mathfrak{k}_{\pm}:=\rho(\mathfrak{g}_2)\oplus E^{c}_{\pm}(Q_{64}). \tag*{(135)} k ± := ρ ( g 2 ) ⊕ E ± c ( Q 64 ) . (135)
Proposition 4.18 (The compact real forms).
(1) The space k ± \mathfrak{k}_{\pm} k ± is a real Lie algebra and
k ± = g ± ∩ u ( W C ) . \mathfrak{k}_{\pm}=\mathfrak{g}_{\pm}\cap\mathfrak{u}(W_{\mathbb{C}}). k ± = g ± ∩ u ( W C ) . It is a compact real form of g ± ≅ e 6 \mathfrak{g}_{\pm}\cong\mathfrak{e}_6 g ± ≅ e 6 . In particular,
g ± = k ± ⊕ i k ± , dim R k ± = 78. \mathfrak{g}_{\pm}=\mathfrak{k}_{\pm}\oplus i\mathfrak{k}_{\pm},\qquad\dim_{\mathbb{R}}\mathfrak{k}_{\pm}=78. g ± = k ± ⊕ i k ± , dim R k ± = 78. (2) The group K ± K_{\pm} K ± is the compact, simply connected Lie group of type E 6 E_6 E 6 , with Lie algebra k ± \mathfrak{k}_{\pm} k ± . It is a maximal compact subgroup of G ± ≅ E 6 ( C ) G_{\pm}\cong E_6(\mathbb{C}) G ± ≅ E 6 ( C ) , and
G ± = K ± exp ( i k ± ) , ρ ( G 2 ) ⊂ K ± . G_{\pm}=K_{\pm}\exp(i\mathfrak{k}_{\pm}),\qquad\rho(G_2)\subset K_{\pm}. G ± = K ± exp ( i k ± ) , ρ ( G 2 ) ⊂ K ± . (3) For Q ∈ Q 64 Q\in Q_{64} Q ∈ Q 64 ,
E + c ( Q ) ‾ = E − c ( Q ) . \overline{E^{c}_{+}(Q)}=E^{c}_{-}(Q). E + c ( Q ) = E − c ( Q ) . Complex conjugation exchanges k + \mathfrak{k}_{+} k + and k − \mathfrak{k}_{-} k − , and likewise exchanges K + K_{+} K + and K − K_{-} K − .
Proof. By Proposition 3.31 and the rescaling above, the complex Lie algebra g ± \mathfrak{g}_{\pm} g ± is the direct sum of ρ ( g 2 ) \rho(\mathfrak{g}_2) ρ ( g 2 ) and m ± = E ± c ( Q 64 , C ) \mathfrak{m}_{\pm}=E^{c}_{\pm}(Q_{64,\mathbb{C}}) m ± = E ± c ( Q 64 , C ) . Both maps ρ \rho ρ and E ± c E^{c}_{\pm} E ± c are complex-linear extensions of their restrictions to real arguments. Hence every X ∈ g ± X\in\mathfrak{g}_{\pm} X ∈ g ± has a unique expression
X = Y + i Z , Y , Z ∈ k ± . X=Y+iZ,\qquad Y,Z\in\mathfrak{k}_{\pm}. X = Y + i Z , Y , Z ∈ k ± .
By Lemma 4.17 , both Y Y Y and Z Z Z are skew-Hermitian, so
X H = − Y + i Z . X^{\mathrm{H}}=-Y+iZ. X H = − Y + i Z .
Thus X X X is skew-Hermitian precisely when Z = 0 Z=0 Z = 0 . This gives
g ± ∩ u ( W C ) = k ± , g ± = k ± ⊕ i k ± . \mathfrak{g}_{\pm}\cap\mathfrak{u}(W_{\mathbb{C}})=\mathfrak{k}_{\pm},\qquad\mathfrak{g}_{\pm}=\mathfrak{k}_{\pm}\oplus i\mathfrak{k}_{\pm}. g ± ∩ u ( W C ) = k ± , g ± = k ± ⊕ i k ± .
As the intersection of g ± \mathfrak{g}_{\pm} g ± and u ( W C ) \mathfrak{u}(W_{\mathbb{C}}) u ( W C ) , k ± \mathfrak{k}_{\pm} k ± is a real Lie subalgebra. Moreover, − tr W ( X Y ) -\operatorname{tr}_{W}(XY) − tr W ( X Y ) is invariant and positive definite on it, since − tr W ( X 2 ) = tr W ( X H X ) > 0 -\operatorname{tr}_{W}(X^2)=\operatorname{tr}_{W}(X^{\mathrm{H}}X)>0 − tr W ( X 2 ) = tr W ( X H X ) > 0 for nonzero skew-Hermitian X X X . It follows that k ± \mathfrak{k}_{\pm} k ± is compact. Since its complexification is e 6 \mathfrak{e}_6 e 6 , it has real dimension 78 78 78 .
The same decomposition shows that g ± \mathfrak{g}_{\pm} g ± is invariant under X ↦ − X H X\mapsto-X^{\mathrm{H}} X ↦ − X H . Since G ± G_{\pm} G ± is connected (Theorem 4.1 ), it is invariant under the corresponding group involution γ ↦ ( γ H ) − 1 \gamma\mapsto(\gamma^{\mathrm{H}})^{-1} γ ↦ ( γ H ) − 1 . The Cartan decomposition for self-adjoint complex reductive groups therefore gives
G ± = K ± exp ( i k ± ) , (136) G_{\pm}=K_{\pm}\exp(i\mathfrak{k}_{\pm}), \tag*{(136)} G ± = K ± exp ( i k ± ) , (136)
with K ± K_{\pm} K ± maximal compact and a deformation retract of G ± G_{\pm} G ± . By Theorem 4.1 , G ± G_{\pm} G ± is connected and simply connected. The deformation retraction gives the same properties for K ± K_{\pm} K ± . Together with the Lie algebra identification, this proves that K ± K_{\pm} K ± is the compact, simply connected group of type E 6 E_{6} E 6 . The inclusion ρ ( G 2 ) ⊂ K ± \rho(G_{2}) \subset K_{\pm} ρ ( G 2 ) ⊂ K ± follows because ρ ( G 2 ) \rho(G_{2}) ρ ( G 2 ) preserves both the critical cubic and the Hermitian form.
Finally, the operators J Q s J_{Q}^{s} J Q s and J Q a J_{Q}^{a} J Q a are real, so E + c ( Q ) = E − c ( Q ) \mathcal{E}^{c}_{+}(Q)=\mathcal{E}^{c}_{-}(Q) E + c ( Q ) = E − c ( Q ) . Conjugation exchanges the critical stabilizers and preserves u ( W C ) \mathfrak{u}(W_{\mathbb{C}}) u ( W C ) and U ( W C ) U(W_{\mathbb{C}}) U ( W C ) . It therefore exchanges the compact real forms k ± \mathfrak{k}_{\pm} k ± and the compact subgroups K ± K_{\pm} K ± .
The compact E 6 E_{6} E 6 group is classical; see [61 p. 68] and [62 Theorem 3.1.1, p. 80] . The description above realizes its Lie algebra inside End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) through the parity embeddings. This compact form is selected by the critical cubic together with the fixed Hermitian form, not by the abstract complex cubic alone.
The Euclidean Albert algebra at an S U ( 3 ) SU(3) S U ( 3 ) -type basepoint
We now choose a real line in V V V . A unit vector u u u spanning it determines
E u = 7 P u − g , E_{u}=7P_{u}-g, E u = 7 P u − g ,
as in (17) . Since E − u = E u E_{-u}=E_{u} E − u = E u , this basepoint depends only on the unoriented line. Its stabilizer in G 2 G_{2} G 2 is S U ( 3 ) ⋊ Z / 2 SU(3)\rtimes\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 , with identity component the stabilizer S U ( 3 ) SU(3) S U ( 3 ) of u u u (Proposition 4.13 ); this is what we mean by an S U ( 3 ) SU(3) S U ( 3 ) -type basepoint. The complex-bilinear based trace form and the positive Hermitian form are related by an antilinear map, the Euclidean real structure ς u \varsigma_{u} ς u constructed below. We show that this map is an involutive Jordan automorphism. On its fixed space the based trace form is 1 / 14 1/14 1/14 times the Hermitian form, hence positive definite.
Fix either critical parameter. Where convenient, we suppress its sign on N N N , F F F , T e T_{e} T e , T E u T_{E_{u}} T E u , ς u \varsigma_{u} ς u and J u J_{u} J u (defined below), restoring ± \pm ± when displaying the root-dependent coefficients or comparing the two constructions.
The based trace form and S U ( 3 ) SU(3) S U ( 3 ) sectors
The E 1 E_{1} E 1 row of (3.16), transported from e 1 e_{1} e 1 to u u u by G 2 G_{2} G 2 , gives F ± ( E u ) = ( 5 + 2 λ ± ) E u F_{\pm}(E_{u})=(5+2\lambda_{\pm})E_{u} F ± ( E u ) = ( 5 + 2 λ ± ) E u . Thus the norm and based linear trace are
n ± : = N ± ( E u ) = 42 ( 5 + 2 λ ± ) ≠ 0 , tr E u ( h ) = 1 2 h ( u , u ) = 1 14 ⟨ E u , h ⟩ W . (137) n_{\pm}:=N_{\pm}(E_{u})=42(5+2\lambda_{\pm})\ne0,\qquad\operatorname{tr}_{E_{u}}(h)=\frac{1}{2}h(u,u)=\frac{1}{14}\langle E_{u},h\rangle_{W}. \tag*{(137)} n ± := N ± ( E u ) = 42 ( 5 + 2 λ ± ) = 0 , tr E u ( h ) = 2 1 h ( u , u ) = 14 1 ⟨ E u , h ⟩ W . (137)
In particular, Lemma 4.8 gives
T E u h = 1 14 tr E u ( h ) E u − 6 n ± F ± ( E u , h ) . (138) T_{E_{u}}h=\frac{1}{14}\operatorname{tr}_{E_{u}}(h)E_{u}-\frac{6}{n_{\pm}}F_{\pm}(E_{u},h). \tag*{(138)} T E u h = 14 1 tr E u ( h ) E u − n ± 6 F ± ( E u , h ) . (138)
Write n = n ± n=n_{\pm} n = n ± when suppressing the chosen sign. The based trace form and trace operator are those of (117) and (119) for N / n N/n N / n ; their pairing conventions are collected in Table 3.
Recall from § 4.5 the operator Θ u ( v ) = v × u \Theta_{u}(v)=v\times u Θ u ( v ) = v × u . Like Θ 1 \Theta_{1} Θ 1 , it satisfies Θ u 2 = − I \Theta_{u}^{2}=-I Θ u 2 = − I on u ⊥ u^{\perp} u ⊥ , by G 2 G_{2} G 2 -transitivity on unit vectors. Recall also the real sectors W 6 ( u ) W_{6}(u) W 6 ( u ) , W 8 ( u ) W_{8}(u) W 8 ( u ) and W 12 ( u ) W_{12}(u) W 12 ( u ) , of real dimensions 6 6 6 , 8 8 8 and 12 12 12 , with complexifications W r ( u ) C W_{r}(u)_{\mathbb{C}} W r ( u ) C . They give the orthogonal decompositions
W = R E u ⊕ W 6 ( u ) ⊕ W 8 ( u ) ⊕ W 12 ( u ) , W=\mathbb{R}E_{u}\oplus W_{6}(u)\oplus W_{8}(u)\oplus W_{12}(u), W = R E u ⊕ W 6 ( u ) ⊕ W 8 ( u ) ⊕ W 12 ( u ) ,
W C = C E u ⊕ W 6 ( u ) C ⊕ W 8 ( u ) C ⊕ W 12 ( u ) C . (139) W_{\mathbb{C}}=\mathbb{C}E_{u}\oplus W_{6}(u)_{\mathbb{C}}\oplus W_{8}(u)_{\mathbb{C}}\oplus W_{12}(u)_{\mathbb{C}}. \tag*{(139)} W C = C E u ⊕ W 6 ( u ) C ⊕ W 8 ( u ) C ⊕ W 12 ( u ) C . (139)
We use the same notation for an orthogonal projector on W W W and its complex-linear extension. The sectors are those of Table 2 , transported from e 1 e_1 e 1 to u u u . Products below use the metric identification, so g g g acts as I I I . For h ∈ W C h \in W_{\mathbb{C}} h ∈ W C , put a = h ( u , u ) / 6 a = h(u,u)/6 a = h ( u , u ) /6 and H = ( g − P u ) h ( g − P u ) + a ( g − P u ) H = (g-P_u)h(g-P_u)+a(g-P_u) H = ( g − P u ) h ( g − P u ) + a ( g − P u ) , the traceless horizontal block. We compute the trace spectrum because its common absolute value will make the Euclidean real structure involutive.
Lemma 4.19 (Sector projectors and the based trace operator). With the splitting (4.24), the following hold.
(i) The four orthogonal projectors are
Π 1 h = a E u , Π 6 h = P u h ( g − P u ) + ( g − P u ) h P u , Π 8 h = 1 2 ( H − Θ u H Θ u ) , Π 12 h = 1 2 ( H + Θ u H Θ u ) . (140) \begin{aligned}
\Pi_1h &= aE_u, & \Pi_6h &= P_uh(g-P_u)+(g-P_u)hP_u,\\
\Pi_8h &= \frac{1}{2}(H-\Theta_uH\Theta_u), & \Pi_{12}h &= \frac{1}{2}(H+\Theta_uH\Theta_u). \tag*{(140)}
\end{aligned} Π 1 h Π 8 h = a E u , = 2 1 ( H − Θ u H Θ u ) , Π 6 h Π 12 h = P u h ( g − P u ) + ( g − P u ) h P u , = 2 1 ( H + Θ u H Θ u ) . (140) These projectors are polynomial in ( g , φ , P u ) (g,\varphi,P_u) ( g , φ , P u ) and are invariant under u ↦ − u u\mapsto-u u ↦ − u .
(ii) The based trace operator and its inverse are
T E u = 1 14 Π 1 − 1 14 Π 6 − 1 ± 3 i 7 112 Π 8 + 3 ± i 7 56 Π 12 , T_{E_u}=\frac{1}{14}\Pi_1-\frac{1}{14}\Pi_6-\frac{1\pm3i\sqrt{7}}{112}\Pi_8+\frac{3\pm i\sqrt{7}}{56}\Pi_{12}, T E u = 14 1 Π 1 − 14 1 Π 6 − 112 1 ± 3 i 7 Π 8 + 56 3 ± i 7 Π 12 , T E u − 1 = 14 Π 1 − 14 Π 6 − 7 4 ( 1 ∓ 3 i 7 ) Π 8 + 7 2 ( 3 ∓ i 7 ) Π 12 . (141) T_{E_u}^{-1}=14\Pi_1-14\Pi_6-\frac{7}{4}(1\mp3i\sqrt{7})\Pi_8+\frac{7}{2}(3\mp i\sqrt{7})\Pi_{12}. \tag*{(141)} T E u − 1 = 14 Π 1 − 14 Π 6 − 4 7 ( 1 ∓ 3 i 7 ) Π 8 + 2 7 ( 3 ∓ i 7 ) Π 12 . (141) Every eigenvalue of T E u T_{E_u} T E u has absolute value 1 / 14 1/14 1/14 .
Proof. Split a symmetric endomorphism into its mixed block, scalar block and horizontal traceless block. This gives the expressions for Π 1 \Pi_1 Π 1 , Π 6 \Pi_6 Π 6 and H H H in (4.25). On the horizontal traceless block, the map h ↦ − Θ u h Θ u h\mapsto-\Theta_uh\Theta_u h ↦ − Θ u h Θ u is an orthogonal involution. Its + 1 +1 + 1 and − 1 -1 − 1 eigenspaces are W 8 ( u ) W_8(u) W 8 ( u ) and W 12 ( u ) W_{12}(u) W 12 ( u ) , respectively, giving the remaining projectors. The two occurrences of u u u in Θ u H Θ u \Theta_uH\Theta_u Θ u H Θ u enter quadratically, through P u P_u P u . Changing u u u to − u -u − u therefore leaves every projector unchanged.
By G 2 G_2 G 2 -equivariance, the sector values in Table 2 hold at every real unit vector u u u . Thus F ± ( E u , ⋅ ) F_\pm(E_u,\cdot) F ± ( E u , ⋅ ) acts on the four sectors by
5 + 2 λ ± , 5 2 + λ ± , − 1 + 8 λ ± , − 1 − 6 λ ± . 5+2\lambda_\pm,\qquad\frac{5}{2}+\lambda_\pm,\qquad-1+8\lambda_\pm,\qquad-1-6\lambda_\pm. 5 + 2 λ ± , 2 5 + λ ± , − 1 + 8 λ ± , − 1 − 6 λ ± .
Substituting in (4.23), the scalar sector gives
T E u E u = ( 3 14 − 6 ( 5 + 2 λ ± ) 42 ( 5 + 2 λ ± ) ) E u = 1 14 E u . T_{E_u}E_u=\left(\frac{3}{14}-\frac{6(5+2\lambda_\pm)}{42(5+2\lambda_\pm)}\right)E_u=\frac{1}{14}E_u. T E u E u = ( 14 3 − 42 ( 5 + 2 λ ± ) 6 ( 5 + 2 λ ± ) ) E u = 14 1 E u .
The linear trace tr E u \operatorname{tr}_{E_u} tr E u vanishes on the other three sectors. Their eigenvalues are therefore − 6 f r ( λ ± ) / [ 42 ( 5 + 2 λ ± ) ] -6f_r(\lambda_\pm)/[42(5+2\lambda_\pm)] − 6 f r ( λ ± ) / [ 42 ( 5 + 2 λ ± )] , where f r f_r f r is the corresponding response scalar above. For the mixed sector, f 6 = ( 5 + 2 λ ± ) / 2 f_6=(5+2\lambda_\pm)/2 f 6 = ( 5 + 2 λ ± ) /2 gives − 1 / 14 -1/14 − 1/14 . With λ ± = ( 1 ± i 7 ) / 8 \lambda_\pm=(1\pm i\sqrt{7})/8 λ ± = ( 1 ± i 7 ) /8 from (3.3), the other two become
− 1 + 8 λ ± − 7 ( 5 + 2 λ ± ) = − 1 ± 3 i 7 112 , 1 + 6 λ ± 7 ( 5 + 2 λ ± ) = 3 ± i 7 56 . \frac{-1+8\lambda_\pm}{-7(5+2\lambda_\pm)}=-\frac{1\pm3i\sqrt{7}}{112},\qquad
\frac{1+6\lambda_\pm}{7(5+2\lambda_\pm)}=\frac{3\pm i\sqrt{7}}{56}. − 7 ( 5 + 2 λ ± ) − 1 + 8 λ ± = − 112 1 ± 3 i 7 , 7 ( 5 + 2 λ ± ) 1 + 6 λ ± = 56 3 ± i 7 .
These give (4.26). Since
∣ 1 ± 3 i 7 ∣ 2 = 64 , ∣ 3 ± i 7 ∣ 2 = 16 , \lvert1\pm3i\sqrt{7}\rvert^2=64,\qquad\lvert3\pm i\sqrt{7}\rvert^2=16, ∣ 1 ± 3 i 7 ∣ 2 = 64 , ∣ 3 ± i 7 ∣ 2 = 16 ,
all four eigenvalues have absolute value 1 / 14 1/14 1/14 . Taking their reciprocals gives T E u − 1 T_{E_u}^{-1} T E u − 1 . □ \square □
The Euclidean real structure
To show that the Euclidean real structure defined in (4.28) below also preserves the based Jordan product, we first relate the two critical cubics through the based trace operator. The next identity holds at any basepoint of nonzero norm.
Lemma 4.20 (Opposite-cubic identity). Let e ∈ W C e \in W_{\mathbb{C}} e ∈ W C satisfy n = N ± ( e ) ≠ 0 n=N_{\pm}(e) \ne0 n = N ± ( e ) = 0 . The based trace operator for the normalized cubic N ± / n N_{\pm}/n N ± / n satisfies
N ∓ ( T e x ) = 18 n 2 N ± ( x ) for all x ∈ W C . (142) N_{\mp}(T_e x)=\frac{18}{n^2}N_{\pm}(x) \qquad\text{for all } x \in W_{\mathbb{C}}. \tag*{(142)} N ∓ ( T e x ) = n 2 18 N ± ( x ) for all x ∈ W C . (142)
Proof. We prove the identity at λ + \lambda_{+} λ + . The based trace form is symmetric by (117) and nondegenerate by Theorem 4.9 [1 ] , so T e T_e T e is W W W -self-adjoint and invertible. Define
Q ( y ) = n − 2 N + ( T e − 1 y ) . Q(y)=n^{-2}N_{+}(T_e^{-1}y). Q ( y ) = n − 2 N + ( T e − 1 y ) .
By (66) , its W W W -gradient and the based adjoint satisfy
∇ Q ( y ) = 3 n 2 T e − 1 F + ( T e − 1 y ) , x ♯ = 3 n T e − 1 F + ( x ) . \nabla Q(y)=\frac{3}{n^2}T_e^{-1}F_{+}(T_e^{-1}y), \qquad x^{\sharp}=\frac{3}{n}T_e^{-1}F_{+}(x). ∇ Q ( y ) = n 2 3 T e − 1 F + ( T e − 1 y ) , x ♯ = n 3 T e − 1 F + ( x ) .
Quadratic homogeneity and the based adjoint identity in (120) give
∇ Q ( 3 F + ( x ) ) = n ( x ♯ ) ♯ = N + ( x ) x . \nabla Q(3F_{+}(x))=n(x^{\sharp})^{\sharp}=N_{+}(x)x. ∇ Q ( 3 F + ( x )) = n ( x ♯ ) ♯ = N + ( x ) x .
By Proposition 3.7, the last expression is also 1 6 F − ( 3 F + ( x ) ) \frac{1}{6}F_{-}(3F_{+}(x)) 6 1 F − ( 3 F + ( x )) . By Lemma 3.8, F + F_{+} F + has dense image, so the two polynomial maps agree everywhere:
∇ Q = 1 6 F − = ∇ ( N − / 18 ) . \nabla Q=\frac{1}{6}F_{-}=\nabla(N_{-}/18). ∇ Q = 6 1 F − = ∇ ( N − /18 ) .
Both cubics are homogeneous, so Q = N − / 18 Q=N_{-}/18 Q = N − /18 . Substituting y = T e x y=T_e x y = T e x gives (142) . Complex conjugation gives the identity at λ − \lambda_{-} λ − . □
Return to e = E u e=E_u e = E u . We seek an antilinear map satisfying ⟨ x , y ‾ ⟩ W = 14 ⟨ x , ς u y ⟩ E u \langle x,\overline{y}\rangle_W=14\langle x,\varsigma_u y\rangle_{E_u} ⟨ x , y ⟩ W = 14 ⟨ x , ς u y ⟩ E u . The trace-operator definition (119) forces 14 T E u ς u y = y ‾ 14T_{E_u}\varsigma_u y=\overline{y} 14 T E u ς u y = y . Thus we first conjugate and then apply the inverse trace operator, defining the Euclidean real structure by
ς u ( x ) = 1 14 T E u − 1 ( x ‾ ) . (143) \varsigma_u(x)=\frac{1}{14}T_{E_u}^{-1}(\overline{x}). \tag*{(143)} ς u ( x ) = 14 1 T E u − 1 ( x ) . (143)
The factor 1 / 14 1/14 1/14 is fixed by the common absolute value of the eigenvalues of T E u T_{E_u} T E u . Define the fixed space of ς u \varsigma_u ς u by
J u = { x ∈ W C : ς u ( x ) = x } . (144) \mathcal{J}_u=\{x\in W_{\mathbb{C}}:\varsigma_u(x)=x\}. \tag*{(144)} J u = { x ∈ W C : ς u ( x ) = x } . (144)
From (4.22) and 5 + 2 λ ± = ( 21 ± i 7 ) / 4 5+2\lambda_{\pm}=(21\pm i\sqrt{7})/4 5 + 2 λ ± = ( 21 ± i 7 ) /4 , we have ∣ 5 + 2 λ ± ∣ 2 = ( 441 + 7 ) / 16 = 28 \lvert5+2\lambda_{\pm}\rvert^2=(441+7)/16=28 ∣ 5 + 2 λ ± ∣ 2 = ( 441 + 7 ) /16 = 28 . Consequently,
n ± = N ± ( E u ) = 4 2 ( 5 + 2 λ ± ) , n − = n ‾ + , n + n − = 14 3 ⋅ 18. n_{\pm}=N_{\pm}(E_u)=4^2(5+2\lambda_{\pm}), \qquad n_{-}=\overline{n}_{+}, \qquad n_{+}n_{-}=14^3\cdot18. n ± = N ± ( E u ) = 4 2 ( 5 + 2 λ ± ) , n − = n + , n + n − = 1 4 3 ⋅ 18.
Thus N ± / n ± N_{\pm}/n_{\pm} N ± / n ± takes the value 1 at E u E_u E u .
Lemma 4.21 (The Euclidean real structure). For either critical parameter and every real unit vector u u u , the following hold.
(i) The map ς u \varsigma_u ς u is an antilinear involutive automorphism of J ~ E u \widetilde{J}_{E_u} J E u , fixing E u E_u E u . For every x ∈ W C x \in W_{\mathbb{C}} x ∈ W C ,
N ± ( ς u x ) n ± = N ± ( x ) ‾ n ± . \frac{N_{\pm}(\varsigma_u x)}{n_{\pm}}=\frac{\overline{N_{\pm}(x)}}{n_{\pm}}. n ± N ± ( ς u x ) = n ± N ± ( x ) . (ii) For all x , y ∈ W C x,y \in W_{\mathbb{C}} x , y ∈ W C ,
⟨ x , y ‾ ⟩ W = 14 ⟨ x , ς u y ⟩ E u . (145) \langle x,\overline{y}\rangle_W=14\langle x,\varsigma_u y\rangle_{E_u}. \tag*{(145)} ⟨ x , y ⟩ W = 14 ⟨ x , ς u y ⟩ E u . (145) Moreover, ς u \varsigma_u ς u is the unique antilinear involution η \eta η for which
⟨ x , y ‾ ⟩ W = s ⟨ x , η y ⟩ E u for all x , y ∈ W C \langle x,\overline{y}\rangle_W=s\langle x,\eta y\rangle_{E_u}\qquad\text{for all }x,y\in W_{\mathbb{C}} ⟨ x , y ⟩ W = s ⟨ x , η y ⟩ E u for all x , y ∈ W C with some s > 0 s>0 s > 0 . Necessarily s = 14 s=14 s = 14 .
Proof. By Lemma 4.19, the operator T E u T_{E_u} T E u acts by scalars on the complexifications of mutually orthogonal real subspaces. With T ‾ E u ( x ) = T E u x ‾ \overline{T}_{E_u}(x)=T_{E_u}\overline{x} T E u ( x ) = T E u x , the spectral decomposition of T E u T_{E_u} T E u gives
T E u T ‾ E u = 1 14 2 id . T_{E_u}\overline{T}_{E_u}=\frac{1}{14^2}\operatorname{id}. T E u T E u = 1 4 2 1 id .
Consequently,
ς u 2 = 1 14 2 T E u − 1 T ‾ E u − 1 = id . \varsigma_u^2=\frac{1}{14^2}T_{E_u}^{-1}\overline{T}_{E_u}^{-1}=\operatorname{id}. ς u 2 = 1 4 2 1 T E u − 1 T E u − 1 = id .
Since E u E_u E u is real and T E u E u = E u / 14 T_{E_u}E_u=E_u/14 T E u E u = E u /14 , we also have ς u E u = E u \varsigma_u E_u=E_u ς u E u = E u .
We next check the normalized cubic identity at λ + \lambda_+ λ + . By Lemma 4.20,
N − ∘ T E u = 18 n + 2 N + . N_-\circ T_{E_u}=\frac{18}{n_+^2}N_+. N − ∘ T E u = n + 2 18 N + .
Applying this identity to T E u − 1 x ‾ T_{E_u}^{-1}\overline{x} T E u − 1 x and using homogeneity gives
N + ( ς u x ) = 1 14 3 N + ( T E u − 1 x ‾ ) = n + 2 14 3 ⋅ 18 N − ( x ‾ ) = n + n ‾ + N + ( x ‾ ) . \begin{aligned}
N_+(\varsigma_u x)&=\frac{1}{14^3}N_+(T_{E_u}^{-1}\overline{x})\\
&=\frac{n_+^2}{14^3\cdot18}N_-(\overline{x})\\
&=\frac{n_+}{\overline{n}_+}N_+(\overline{x}).
\end{aligned} N + ( ς u x ) = 1 4 3 1 N + ( T E u − 1 x ) = 1 4 3 ⋅ 18 n + 2 N − ( x ) = n + n + N + ( x ) .
Dividing by n + n_+ n + proves the identity. The identity at the other root follows by conjugation.
Since ς u \varsigma_u ς u fixes E u E_u E u , polarization of the normalized cubic identity at E u E_u E u yields
tr E u ( ς u x ) = tr E u ( x ) ‾ \operatorname{tr}_{E_u}(\varsigma_u x)=\overline{\operatorname{tr}_{E_u}(x)} tr E u ( ς u x ) = tr E u ( x )
and
⟨ ς u x , ς u y ⟩ E u = ⟨ x , y ⟩ ‾ E u . \langle\varsigma_u x,\varsigma_u y\rangle_{E_u}=\overline{\langle x,y\rangle}_{E_u}. ⟨ ς u x , ς u y ⟩ E u = ⟨ x , y ⟩ E u .
By the adjoint characterization in (4.6), the normalized cubic and the nondegenerate based trace form determine the based adjoint, as in (4.12). Hence
( ς u x ) # = ς u ( x # ) . (\varsigma_u x)^\#=\varsigma_u(x^\#). ( ς u x ) # = ς u ( x # ) .
Substitution in (4.7) now gives
ς u ( x ∘ E u y ) = ς u x ∘ E u ς u y . \varsigma_u(x\mathbin{\circ_{E_u}} y)=\varsigma_u x\mathbin{\circ_{E_u}}\varsigma_u y. ς u ( x ∘ E u y ) = ς u x ∘ E u ς u y .
This proves (i).
For (ii), the definition of ς u \varsigma_u ς u gives 1 4 T E u ς u y = y ˉ 1_4T_{E_u}\varsigma_u y=\bar{y} 1 4 T E u ς u y = y ˉ . Therefore
1 4 ⟨ x , ς u y ⟩ E u = 1 4 ⟨ x , T E u ς u y ⟩ W = ⟨ x , y ˉ ⟩ W . 1_4\langle x,\varsigma_u y\rangle_{E_u}=1_4\langle x,T_{E_u}\varsigma_u y\rangle_W=\langle x,\bar{y}\rangle_W. 1 4 ⟨ x , ς u y ⟩ E u = 1 4 ⟨ x , T E u ς u y ⟩ W = ⟨ x , y ˉ ⟩ W .
Finally, suppose that another antilinear involution η \eta η satisfies the stated identity with s > 0 s>0 s > 0 . Nondegeneracy of the based trace form gives
η = 1 4 s ς u . \eta=\frac{1_4}{s}\varsigma_u. η = s 1 4 ς u .
Squaring yields ( 1 4 / s ) 2 = 1 (1_4/s)^2=1 ( 1 4 / s ) 2 = 1 . Positivity forces s = 1 4 s=1_4 s = 1 4 , and hence η = ς u \eta=\varsigma_u η = ς u .
The Euclidean real algebra and its compact stabilizer
The fixed space of ς u \varsigma_u ς u can be described separately in each sector. Let S S S be one of the real summands in (4.24), and let ϑ \vartheta ϑ be the eigenvalue of T E u T_{E_u} T E u on S C S_{\mathbb{C}} S C . For nonzero v ∈ S v\in S v ∈ S ,
ς u ( a v ) = a v ⟺ a ˉ = 1 4 ϑ a . (146) \varsigma_u(av)=av\quad\Longleftrightarrow\quad\bar{a}=1_4\vartheta a. \tag*{(146)} ς u ( a v ) = a v ⟺ a ˉ = 1 4 ϑ a . (146)
Since ∣ 1 4 ϑ ∣ = 1 |1_4\vartheta|=1 ∣ 1 4 ϑ ∣ = 1 , this equation defines a real line in C \mathbb{C} C . A nonzero solution a a a determines the real subspace a S aS a S inside S C S_{\mathbb{C}} S C . Replacing a a a by any nonzero real multiple leaves this subspace unchanged. Thus the coefficients below specify real lines in C \mathbb{C} C ; they need not be unit complex numbers. Such a rescaling changes neither the real algebra nor its inherited trace form. For the last two sectors we choose generators
a 8 , ± = − 7 ± 3 i 7 4 , a 12 , ± = 7 ∓ i . a_{8,\pm}=-\frac{7\pm3i\sqrt{7}}{4},\qquad a_{12,\pm}=\sqrt{7}\mp i. a 8 , ± = − 4 7 ± 3 i 7 , a 12 , ± = 7 ∓ i .
A real Jordan algebra J J J is called formally real if
x 1 ∘ x 1 + ⋯ + x k ∘ x k = 0 ⟹ x 1 = ⋯ = x k = 0. x_1\mathbin{\circ}x_1+\cdots+x_k\mathbin{\circ}x_k=0\quad\Longrightarrow\quad x_1=\cdots=x_k=0. x 1 ∘ x 1 + ⋯ + x k ∘ x k = 0 ⟹ x 1 = ⋯ = x k = 0.
For finite-dimensional unital Jordan algebras, formal reality is equivalent to the existence of a positive-definite inner product associative with respect to the Jordan product [19 Proposition VIII.4.2] . Such an algebra is called a Euclidean Jordan algebra. Let H 3 ( O ) H_3(\mathbb{O}) H 3 ( O ) be the real Jordan algebra of 3 × 3 3\times3 3 × 3 Hermitian matrices over the real octonions O \mathbb{O} O , with
x ∘ y = 1 2 ( x y + y x ) . x\mathbin{\circ}y=\frac{1}{2}(xy+yx). x ∘ y = 2 1 ( x y + y x ) .
It is Euclidean of rank 3 [19 Corollary V.2.6] , and H 3 ( O C ) = H 3 ( O ) ⊗ R C H_3(\mathbb{O}_{\mathbb{C}})=H_3(\mathbb{O})\otimes_{\mathbb{R}}\mathbb{C} H 3 ( O C ) = H 3 ( O ) ⊗ R C is the complex Albert algebra introduced in Section 4.2. Recall that the Jordan determinant is the signed constant coefficient of the generic characteristic polynomial (see the discussion of rank and determinant following (4.8)).
Proposition 4.22 (The Euclidean Albert real form). For either critical parameter and every real unit vector u u u , the following hold.
(1) With the real sectors of (139) ,
J u = { x ∈ W C : x ˉ = 14 T E u x } = R E u ⊕ i W 6 ( u ) ⊕ a 8 , ± W 8 ( u ) ⊕ a 12 , ± W 12 ( u ) . (147) \mathcal{J}_u=\{x\in W_{\mathbb{C}}:\bar{x}=14T_{E_u}x\}
=\mathbb{R}E_u\oplus iW_6(u)\oplus a_{8,\pm}W_8(u)\oplus a_{12,\pm}W_{12}(u). \tag*{(147)} J u = { x ∈ W C : x ˉ = 14 T E u x } = R E u ⊕ i W 6 ( u ) ⊕ a 8 , ± W 8 ( u ) ⊕ a 12 , ± W 12 ( u ) . (147)
Each coefficient is a nonzero generator of the corresponding real scalar line and may be replaced independently by a nonzero real multiple.
The real Jordan algebra J u \mathcal{J}_u J u , with unit E u E_u E u , is simple and Euclidean, and is isomorphic to H 3 ( O ) H_3(\mathbb{O}) H 3 ( O ) . Under this isomorphism, N ± / n ± N_{\pm}/n_{\pm} N ± / n ± is the Jordan determinant. Its based trace form is positive definite: for every nonzero x ∈ J u x\in\mathcal{J}_u x ∈ J u ,
⟨ x , x ⟩ E u = 1 14 ⟨ x , x ˉ ⟩ W > 0. \langle x,x\rangle_{E_u}=\frac{1}{14}\langle x,\bar{x}\rangle_W>0. ⟨ x , x ⟩ E u = 14 1 ⟨ x , x ˉ ⟩ W > 0.
(2) The intersection with the original real module is
J u ∩ W = R E u . \mathcal{J}_u\cap W=\mathbb{R}E_u. J u ∩ W = R E u .
Proof. The fixed-space equation follows directly from (143) . Applying (146) to the four eigenvalues in Lemma 4.19 gives the respective real scalar lines
R , i R , a 8 , ± R , a 12 , ± R . \mathbb{R},\qquad i\mathbb{R},\qquad a_{8,\pm}\mathbb{R},\qquad a_{12,\pm}\mathbb{R}. R , i R , a 8 , ± R , a 12 , ± R .
This proves (147) and explains the freedom to rescale its coefficients by nonzero real numbers. By Lemma 4.21, the fixed space is a real Jordan algebra with unit E u E_u E u . The same lemma shows that its based trace form is real and that
⟨ x , x ⟩ E u = 1 14 ⟨ x , x ˉ ⟩ W ( x ∈ J u ) . \langle x,x\rangle_{E_u}=\frac{1}{14}\langle x,\bar{x}\rangle_W\qquad(x\in\mathcal{J}_u). ⟨ x , x ⟩ E u = 14 1 ⟨ x , x ˉ ⟩ W ( x ∈ J u ) .
It is therefore positive definite.
To verify formal reality directly, set e = E u e=E_u e = E u and N e = N / n N_e=N/n N e = N / n , and polarize d N e ∣ x ( y ) = ⟨ x # , y ⟩ e dN_e|_x(y)=\langle x^{\#},y\rangle_e d N e ∣ x ( y ) = ⟨ x # , y ⟩ e in (120) . The resulting symmetry of ⟨ x # y , z ⟩ e \langle x\mathbin{\#}y,z\rangle_e ⟨ x # y , z ⟩ e , together with e # x = tr e ( x ) e − x e\mathbin{\#}x=\operatorname{tr}_e(x)e-x e # x = tr e ( x ) e − x , gives
tr e ( x # y ) = ⟨ x # y , e ⟩ e = ⟨ e # x , y ⟩ e = tr e ( x ) tr e ( y ) − ⟨ x , y ⟩ e . \operatorname{tr}_e(x\mathbin{\#}y)=\langle x\mathbin{\#}y,e\rangle_e=\langle e\mathbin{\#}x,y\rangle_e=\operatorname{tr}_e(x)\operatorname{tr}_e(y)-\langle x,y\rangle_e. tr e ( x # y ) = ⟨ x # y , e ⟩ e = ⟨ e # x , y ⟩ e = tr e ( x ) tr e ( y ) − ⟨ x , y ⟩ e .
Apply tr E u \operatorname{tr}_{E_u} tr E u to (121) , using this identity and tr E u ( E u ) = 3 \operatorname{tr}_{E_u}(E_u)=3 tr E u ( E u ) = 3 . This gives
tr E u ( x ∘ E u y ) = ⟨ x , y ⟩ E u . \operatorname{tr}_{E_u}(x\circ_{E_u}y)=\langle x,y\rangle_{E_u}. tr E u ( x ∘ E u y ) = ⟨ x , y ⟩ E u .
If ∑ j x j ∘ E u x j = 0 \sum_j x_j\circ_{E_u}x_j=0 ∑ j x j ∘ E u x j = 0 in J u \mathcal{J}_u J u , then
0 = ∑ j tr E u ( x j ∘ E u x j ) = ∑ j ⟨ x j , x j ⟩ E u . 0=\sum_j\operatorname{tr}_{E_u}(x_j\circ_{E_u}x_j)=\sum_j\langle x_j,x_j\rangle_{E_u}. 0 = j ∑ tr E u ( x j ∘ E u x j ) = j ∑ ⟨ x j , x j ⟩ E u .
Positivity forces every x j x_j x j to vanish. Thus J u \mathcal{J}_u J u is formally real and hence Euclidean.
Since ς u \varsigma_u ς u is an antilinear involution, the complexification of J u \mathcal{J}_u J u is J E u \mathcal{J}_{E_u} J E u , which is the simple complex Albert algebra of degree three by Theorem 4.9 (1). A proper nonzero real ideal of J u \mathcal{J}_u J u would complexify to a proper nonzero ideal of J E u \mathcal{J}_{E_u} J E u . The latter algebra is simple, so J u \mathcal{J}_u J u is simple as well.
We must still determine its rank: the classification of simple Euclidean Jordan algebras, recalled below, also includes rank-two spin factors, so dimension and simplicity alone do not suffice. The discussion of rank and determinant following (122) bounds the rank by three. Suppose that every element had minimal polynomial of degree at most two. Then E u E_u E u , x x x and x 2 x^2 x 2 would be linearly dependent for every x ∈ J u x \in\mathcal{J}_u x ∈ J u . The corresponding coordinate minors are polynomial functions. If they vanished on the real form J u \mathcal{J}_u J u , they would vanish on its complexification. This contradicts the degree-three assertion of Theorem 4.9 (1). Hence J u \mathcal{J}_u J u has rank three.
By the classification of simple Euclidean Jordan algebras [19 Theorem V.3.7] , a simple algebra of rank three has dimension 3 + 3 d 3+3d 3 + 3 d , where d ∈ { 1 , 2 , 4 , 8 } d \in\{1,2,4,8\} d ∈ { 1 , 2 , 4 , 8 } . Since dim R J u = 27 \dim_{\mathbb{R}}\mathcal{J}_u=27 dim R J u = 27 , we obtain d = 8 d=8 d = 8 and
J u ≅ H 3 ( O ) . \mathcal{J}_u \cong H_3(\mathbb{O}). J u ≅ H 3 ( O ) .
The rank-three determinant argument following (122) , applied to the real characteristic identity (122) , identifies the Jordan determinant with N ± / n ± N_{\pm}/n_{\pm} N ± / n ± . This completes (1). Finally, if x ∈ W x \in W x ∈ W , then x ˉ = x \bar{x}=x x ˉ = x . Thus
x ∈ J u ∩ W ⟺ x ∈ W and T E u x = 1 14 x . x \in\mathcal{J}_u \cap W \Longleftrightarrow x \in W \text{ and } T_{E_u}x=\frac{1}{14}x. x ∈ J u ∩ W ⟺ x ∈ W and T E u x = 14 1 x .
By Lemma 4.19 , the 1 / 14 1/14 1/14 -eigenspace is precisely C E u \mathbb{C}E_u C E u . Its intersection with W W W is R E u \mathbb{R}E_u R E u , proving (2). □
We now compare the based algebra with the compact group of § 5.6 . For this comparison, restore the signs in ς u , ± \varsigma_{u,\pm} ς u , ± and J u , ± \mathcal{J}_{u,\pm} J u , ± . By (145) ,
K ± = { γ ∈ G ± : ⟨ γ x , ς u , ± γ y ⟩ E u = ⟨ x , ς u , ± y ⟩ E u for all x , y ∈ W C } . (148) K_{\pm}=\{\gamma\in G_{\pm}:\langle\gamma x,\varsigma_{u,\pm}\gamma y\rangle_{E_u}=\langle x,\varsigma_{u,\pm}y\rangle_{E_u}
\text{ for all }x,y\in W_{\mathbb{C}}\}. \tag*{(148)} K ± = { γ ∈ G ± : ⟨ γ x , ς u , ± γ y ⟩ E u = ⟨ x , ς u , ± y ⟩ E u for all x , y ∈ W C } . (148)
The based trace form here is taken at the corresponding critical parameter. Although this expression uses E u E_u E u , the group K ± K_{\pm} K ± itself does not depend on u u u . The dependence on the basepoint enters through the stabilizer of E u E_u E u in K ± K_{\pm} K ± . Define K ± , u = Stab K ± ( E u ) K_{\pm,u}=\operatorname{Stab}_{K_{\pm}}(E_u) K ± , u = Stab K ± ( E u ) .
Corollary 4.23 (The compact stabilizer of the basepoint). For either critical parameter,
K ± , u = { γ ∈ G ± : γ E u = E u , γ ς u , ± = ς u , ± γ } . (149) K_{\pm,u}=\{\gamma\in G_{\pm}:\gamma E_u=E_u,\ \gamma\varsigma_{u,\pm}=\varsigma_{u,\pm}\gamma\}. \tag*{(149)} K ± , u = { γ ∈ G ± : γ E u = E u , γ ς u , ± = ς u , ± γ } . (149) Restriction to J u , ± \mathcal{J}_{u,\pm} J u , ± gives an isomorphism
K ± , u ≅ Aut R ( J u , ± ) . K_{\pm,u}\cong\operatorname{Aut}_{\mathbb{R}}(\mathcal{J}_{u,\pm}). K ± , u ≅ Aut R ( J u , ± ) . In particular, K ± , u K_{\pm,u} K ± , u is the compact Lie group of type F 4 F_4 F 4 . Moreover,
ρ ( Stab G 2 ( E u ) ) ⊂ K ± , u ⊂ K ± , Stab G 2 ( E u ) ≅ S U ( 3 ) ⋊ Z / 2. \rho(\operatorname{Stab}_{G_2}(E_u))\subset K_{\pm,u}\subset K_{\pm},\qquad\operatorname{Stab}_{G_2}(E_u)\cong SU(3)\rtimes\mathbb{Z}/2. ρ ( Stab G 2 ( E u )) ⊂ K ± , u ⊂ K ± , Stab G 2 ( E u ) ≅ S U ( 3 ) ⋊ Z /2.
Proof. Fix a sign and suppress it. Suppose that γ ∈ G \gamma\in G γ ∈ G fixes E u E_u E u . Since it preserves the normalized cubic and the basepoint, it preserves the based trace form and the based Jordan product by Lemma 4.6 , equations (125) and (127) . Therefore
⟨ γ x , ς u γ y ⟩ E u = ⟨ x , γ − 1 ς u γ y ⟩ E u . \langle\gamma x,\varsigma_u\gamma y\rangle_{E_u}=\langle x,\gamma^{-1}\varsigma_u\gamma y\rangle_{E_u}. ⟨ γ x , ς u γ y ⟩ E u = ⟨ x , γ − 1 ς u γ y ⟩ E u .
By nondegeneracy, this equals ⟨ x , ς u y ⟩ E u \langle x,\varsigma_u y\rangle_{E_u} ⟨ x , ς u y ⟩ E u for every x , y x,y x , y precisely when γ ς u = ς u γ \gamma\varsigma_u=\varsigma_u\gamma γ ς u = ς u γ . Together with (148) , this proves (149) .
Such a γ \gamma γ restricts to a real Jordan automorphism of J u \mathcal{J}_u J u . Restriction is injective because J u \mathcal{J}_u J u complexifies to W C W_{\mathbb{C}} W C . Conversely, a real Jordan automorphism of J u \mathcal{J}_u J u fixes its unit E u E_u E u and preserves Jordan powers, hence minimal polynomials. It therefore preserves the generic characteristic polynomial and its signed constant coefficient, the Jordan determinant (see the discussion following (122) ). Its complex-linear extension therefore preserves N / n N/n N / n , and hence N N N , and commutes with ς u \varsigma_u ς u . By (149) , this extension belongs to K ± , u K_{\pm,u} K ± , u . Restriction is thus an isomorphism.
By Proposition 4.22 , J u ≅ H 3 ( O ) \mathcal{J}_u\cong H_3(\mathbb{O}) J u ≅ H 3 ( O ) . The automorphism group of this real algebra is compact F 4 F_4 F 4 ; see [61 § 2.2 and Theorem 3.7.1] . Finally, ρ ( G 2 ) ⊂ K ± \rho(G_2)\subset K_{\pm} ρ ( G 2 ) ⊂ K ± by Proposition 4.18 (2), and Proposition 4.13 identifies the stabilizer of E u E_u E u in G 2 G_2 G 2 with S U ( 3 ) ⋊ Z / 2 SU(3)\rtimes\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 . This proves the inclusions in the last display of the corollary.
Thus K ± , u = K ± ∩ F 4 , E u K_{\pm,u}=K_{\pm}\cap F_{4,E_u} K ± , u = K ± ∩ F 4 , E u is a compact form of the complex stabilizer F 4 , E u F_{4,E_u} F 4 , E u of Proposition 4.12 . Indeed, the Lie algebra of K ± , u K_{\pm,u} K ± , u consists of skew-Hermitian operators, so it intersects its imaginary multiple trivially. This real Lie algebra and f E u \mathfrak{f}_{E_u} f E u have dimension 52 over their respective fields; the complexification of the former is therefore f E u \mathfrak{f}_{E_u} f E u .
The based construction depends only on the unoriented line R u \mathbb{R}u R u . Indeed, E − u = E u E_{-u}=E_u E − u = E u , and the sector projectors, trace operator, involution and fixed algebra are all invariant under u ↦ − u u\mapsto-u u ↦ − u . The construction is also G 2 G_2 G 2 -equivariant: for O ∈ G 2 O\in G_2 O ∈ G 2 ,
ς O u , ± ρ ( O ) = ρ ( O ) ς u , ± , J O u , ± = ρ ( O ) J u , ± . \varsigma_{Ou,\pm}\rho(O)=\rho(O)\varsigma_{u,\pm},\qquad\mathcal{J}_{Ou,\pm}=\rho(O)\mathcal{J}_{u,\pm}. ς O u , ± ρ ( O ) = ρ ( O ) ς u , ± , J O u , ± = ρ ( O ) J u , ± .
These identities follow from the equivariance of the normalized cubic and based trace form, together with the reality of ρ ( O ) \rho(O) ρ ( O ) . Conjugation exchanges the two based constructions as well. Since T E u , + ‾ = T E u , − \overline{T_{E_u,+}}=T_{E_u,-} T E u , + = T E u , − ,
ς u , + ‾ = ς u , − , J u , + ‾ = J u , − . \overline{\varsigma_{u,+}}=\varsigma_{u,-},\qquad\overline{\mathcal{J}_{u,+}}=\mathcal{J}_{u,-}. ς u , + = ς u , − , J u , + = J u , − .
Thus the metric-induced Hermitian form determines the compact E 6 E_6 E 6 groups without a basepoint, whereas a tangent line determines the Euclidean Albert algebras and their compact F 4 F_4 F 4 stabilizers. On a manifold, a smooth line subbundle of the tangent bundle gives these based objects fiberwise: local unit sections differ only by sign, so the constructions agree on overlaps. In § 6 , we globalize these pointwise constructions. By themselves, the resulting bundle reductions do not imply parallelism or a holonomy reduction.
The exceptional reductions on a 7-manifold
In the preceding sections we established the exceptional nature of the two critical cubics on a single copy of the deformation representation W C W_{\mathbb{C}} W C . We now return to a 7-manifold ( M , φ ) (\mathcal{M},\varphi) ( M , φ ) and ask what these pointwise cubics define globally.
The exceptional reductions F ± \mathcal{F}_{\pm} F ± and their compact forms F ± c \mathcal{F}^{c}_{\pm} F ± c are canonical on every 7-manifold with a G 2 G_2 G 2 -structure (Sections 5.1 and 5.2). An Albert algebra bundle requires an additional choice of a section with nonzero critical norm, called a pointing . A tangent line field supplies one; such fields exist on every 7-manifold without boundary, and the resulting reductions are homotopic (§ 6.3 ). Finally, among metric connections on T M TM T M , exactly the G 2 G_2 G 2 -connections preserve the critical cubics. Orthogonal projection of the Levi-Civita connection selects the minimal connection, while parallel pointings require reduced holonomy (§ 6.4 ). As in § 1 , no torsion assumption is made, except for the statements on Levi-Civita parallelism and the compact non-flat example in Remark 5.9, which assume torsion-freeness. Compactness is required only for that example.
It is important to distinguish the projective and the exact statements. A line of cubic forms determines a reduction to the group preserving that line; elements of this group may rescale a cubic. To obtain a reduction to the exact stabilizer E 6 ( C ) E_6(\mathbb{C}) E 6 ( C ) , one must select a nowhere-vanishing cubic in the line. In our setting this selection is automatic: the fixed normalizations of C 0 C_0 C 0 and C φ C_\varphi C φ give the global critical fields N ± = C 0 + λ ± C φ N_\pm=C_0+\lambda_\pm C_\varphi N ± = C 0 + λ ± C φ . Thus the exceptional reductions depend only on φ \varphi φ and the labeling of the two conjugate roots.
The exceptional reductions
Recall from Definition 2.2 the real deformation bundle W φ = S 0 2 T ∗ M \mathcal{W}_\varphi=S_0^2T^*\mathcal{M} W φ = S 0 2 T ∗ M , its complexification W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C , and the principal G 2 G_2 G 2 -bundle Fr φ \operatorname{Fr}_\varphi Fr φ of adapted coframes, together with its complexification Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C . Every adapted coframe p p p over x ∈ M x\in\mathcal{M} x ∈ M induces a complex-linear frame
f p : W C ⟶ W φ , C , x , f p O = f p ρ ( O ) ( O ∈ G 2 ) . f_p:\mathcal{W}_{\mathbb{C}}\longrightarrow\mathcal{W}_{\varphi,\mathbb{C},x},\qquad f_pO=f_p\rho(O)\quad(O\in G_2). f p : W C ⟶ W φ , C , x , f p O = f p ρ ( O ) ( O ∈ G 2 ) .
Here ρ \rho ρ is the representation of § 3 , extended to G 2 C G_2^{\mathbb{C}} G 2 C when required. By Theorem 4.1 , the exact stabilizer of the model critical cubic N ± N_\pm N ± is
G ± = Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) , G_\pm=\operatorname{Stab}_{GL(\mathcal{W}_{\mathbb{C}})}(N_\pm)\cong E_6(\mathbb{C}), G ± = Stab G L ( W C ) ( N ± ) ≅ E 6 ( C ) ,
whereas the stabilizer of its line is G ^ ± = Stab G L ( W C ) ( C N ± ) = G ± ⋅ C × \widehat{G}_\pm=\operatorname{Stab}_{GL(\mathcal{W}_{\mathbb{C}})}(\mathbb{C}N_\pm)=G_\pm\cdot\mathbb{C}^{\times} G ± = Stab G L ( W C ) ( C N ± ) = G ± ⋅ C × . Since both C 0 C_0 C 0 and C φ C_\varphi C φ are G 2 G_2 G 2 -invariant, the representation ρ \rho ρ induces embeddings
ȷ ± : G 2 ↪ G ± , ȷ ± ( O ) = ρ ( O ) . \jmath_\pm:G_2\hookrightarrow G_\pm,\qquad\jmath_\pm(O)=\rho(O). ± : G 2 ↪ G ± , ± ( O ) = ρ ( O ) .
We use the same symbol N ± N_\pm N ± for the fixed model cubic on W C \mathcal{W}_{\mathbb{C}} W C and its global realization; we write N ± , x N_{\pm,x} N ± , x for the cubic on the fiber over x x x . Conjugation of frames means f ‾ ( h ) = f ( h ‾ ) \overline{f}(h)=f(\overline{h}) f ( h ) = f ( h ) . Define the corresponding cubic-preserving frame bundles by
F ± = { f : W C → ∼ W φ , C , x | x ∈ M , f ∗ N ± , x = N ± } . (150) \mathcal{F}_\pm=\left\{f:\mathcal{W}_{\mathbb{C}}\xrightarrow{\sim}\mathcal{W}_{\varphi,\mathbb{C},x}\ \middle|\ x\in\mathcal{M},\ f^*N_{\pm,x}=N_\pm\right\}. \tag*{(150)} F ± = { f : W C ∼ W φ , C , x x ∈ M , f ∗ N ± , x = N ± } . (150)
Theorem 5.1 (Global critical cubics and exceptional reductions).
(1) The invariant cubics C 0 C_0 C 0 and C φ C_\varphi C φ frame the trivial rank-two subbundle
C φ = R C 0 ⊕ R C φ ⊂ S 3 W φ ∗ . \mathcal{C}_\varphi=\mathbb{R}C_0\oplus\mathbb{R}C_\varphi\subset S^3\mathcal{W}_\varphi^*. C φ = R C 0 ⊕ R C φ ⊂ S 3 W φ ∗ . The critical equation a 2 − 2 a b + 8 b 2 = 0 a^2-2ab+8b^2=0 a 2 − 2 ab + 8 b 2 = 0 defines two constant sections of P ( C φ , C ) ≅ M × P 1 ( C ) \mathbb{P}(\mathcal{C}_\varphi,\mathbb{C})\cong\mathcal{M}\times\mathbb{P}^1(\mathbb{C}) P ( C φ , C ) ≅ M × P 1 ( C ) , corresponding to the trivial complex line subbundles
N ± = C N ± , N ± = C 0 + λ ± C φ , λ ± = 1 ± i 7 8 . N_\pm=\mathbb{C}N_\pm,\qquad N_\pm=C_0+\lambda_\pm C_\varphi,\qquad\lambda_\pm=\frac{1\pm i\sqrt{7}}{8}. N ± = C N ± , N ± = C 0 + λ ± C φ , λ ± = 8 1 ± i 7 . These lines are complex conjugate, with N − = N + ‾ N_-=\overline{N_+} N − = N + .
(2) Each F ± \mathcal{F}_{\pm} F ± is a canonical reduction of the complex frame bundle of W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C to G ± ≅ E 6 ( C ) G_{\pm}\cong E_{6}(\mathbb{C}) G ± ≅ E 6 ( C ) , with
Fr φ × ȷ ± G ± → ≅ F ± , [ p , L ] ⟼ f p L , \operatorname{Fr}_{\varphi}\mathbin{\times}_{\jmath_{\pm}}G_{\pm}\xrightarrow{\cong}\mathcal{F}_{\pm},\qquad[p,L]\longmapsto f_{p}L, Fr φ × ± G ± ≅ F ± , [ p , L ] ⟼ f p L ,
and W φ , C ≅ F ± × G ± W C \mathcal{W}_{\varphi,\mathbb{C}}\cong\mathcal{F}_{\pm}\mathbin{\times}_{G_{\pm}}\mathcal{W}_{\mathbb{C}} W φ , C ≅ F ± × G ± W C . The lines N ± \mathcal{N}_{\pm} N ± also define reductions to their respective line stabilizers G ^ ± \widehat{G}_{\pm} G ± .
(3) The bundle Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C is a common G 2 C G_{2}^{\mathbb{C}} G 2 C -reduction of F + \mathcal{F}_{+} F + and F − \mathcal{F}_{-} F − , and conjugation of frames exchanges the two exceptional reductions.
(4) The infinitesimal stabilizers form adjoint bundles
ad F ± ≅ Fr φ × G 2 g ± , g ± = ρ ( g 2 C ) ⊕ m ± ≅ 14 ⊕ 64. \operatorname{ad}\mathcal{F}_{\pm}\cong\operatorname{Fr}_{\varphi}\mathbin{\times}_{G_{2}}\mathfrak{g}_{\pm},\qquad\mathfrak{g}_{\pm}=\rho(\mathfrak{g}_{2}^{\mathbb{C}})\oplus\mathfrak{m}_{\pm}\cong14\oplus64. ad F ± ≅ Fr φ × G 2 g ± , g ± = ρ ( g 2 C ) ⊕ m ± ≅ 14 ⊕ 64.
The 64 64 64 summand gives the infinitesimal symmetries of N ± \mathcal{N}_{\pm} N ± beyond those of φ \varphi φ .
Proof. The universal contractions defining C 0 C_{0} C 0 and C φ C_{\varphi} C φ patch under changes of adapted coframe and are pointwise independent by Lemma 2.7 . Thus they frame C φ C_{\varphi} C φ , and Theorem 3.3 gives the constant critical equation and its conjugate solutions.
Let p p p be a local φ \varphi φ -adapted coframe over U ⊂ M U\subset\mathcal{M} U ⊂ M . By invariance, f p ∗ N ± , x = N ± f_{p}^{*}\mathcal{N}_{\pm,x}=\mathcal{N}_{\pm} f p ∗ N ± , x = N ± , so
F ± ∣ U = { f p L : L ∈ G ± } ≅ U × G ± . \mathcal{F}_{\pm}|_{U}=\{f_{p}L:L\in G_{\pm}\}\cong U\times G_{\pm}. F ± ∣ U = { f p L : L ∈ G ± } ≅ U × G ± .
Since f p O = f p ρ ( O ) f_{p}O=f_{p}\rho(O) f p O = f p ρ ( O ) for O ∈ G 2 O\in G_{2} O ∈ G 2 , these local identifications give the stated extension of structure group and associated-bundle isomorphism. The exact stabilizer is identified with E 6 ( C ) E_{6}(\mathbb{C}) E 6 ( C ) in Theorem 4.1 . The line-stabilizer reduction follows by extension of structure group.
Since ρ ( G 2 C ) ⊂ G ± \rho(G_{2}^{\mathbb{C}})\subset G_{\pm} ρ ( G 2 C ) ⊂ G ± , the maps [ p , O ] ↦ f p ρ ( O ) [p,O]\mapsto f_{p}\rho(O) [ p , O ] ↦ f p ρ ( O ) identify Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C with a reduction common to F + \mathcal{F}_{+} F + and F − \mathcal{F}_{-} F − . Finally, the adapted frames f p f_{p} f p are real and N − = N + ‾ \mathcal{N}_{-}=\overline{\mathcal{N}_{+}} N − = N + , so conjugation sends f p L ∈ F + f_{p}L\in\mathcal{F}_{+} f p L ∈ F + to f p L ‾ ∈ F − f_{p}\overline{L}\in\mathcal{F}_{-} f p L ∈ F − . The adjoint-bundle identification follows from the same extension of structure group, with its splitting given by Proposition 3.31.
Together with Proposition 3.7 , Theorems 3.3 and 4.1 , and Theorem 4.9 , this proves Theorem A.
The fixed normalization gives the canonical reductions above. Changing it can change the subbundle of frames.
Remark 5.2 (Normalization and rescaling). Any nowhere-vanishing section s ± = a N ± s_{\pm}=a\mathcal{N}_{\pm} s ± = a N ± , where a : M → C × a:\mathcal{M}\to\mathbb{C}^{\times} a : M → C × is smooth, defines a principal G ± G_{\pm} G ± -bundle
F ± ( s ± ) = { f : W C → ≅ W φ , C , x | x ∈ M , f ∗ s ± , x = N ± } . (151) \mathcal{F}_{\pm}(s_{\pm})=\left\{f:\mathcal{W}_{\mathbb{C}}\xrightarrow{\cong}\mathcal{W}_{\varphi,\mathbb{C},x}\ \middle|\ x\in\mathcal{M},\ f^{*}s_{\pm,x}=\mathcal{N}_{\pm}\right\}. \tag*{(151)} F ± ( s ± ) = { f : W C ≅ W φ , C , x x ∈ M , f ∗ s ± , x = N ± } . (151)
This need not coincide with the canonical reduction F ± \mathcal{F}_{\pm} F ± . Indeed, if we choose t t t locally with t 3 = a − 1 t^{3}=a^{-1} t 3 = a − 1 , the frames f p t f_{p}t f p t trivialize F ± ( s ± ) \mathcal{F}_{\pm}(s_{\pm}) F ± ( s ± ) . Different choices of cube root differ by elements of the center μ 3 ⊂ G ± \mu_{3}\subset G_{\pm} μ 3 ⊂ G ± , so no global cube root is required.
Geometric meaning and common G 2 G_2 G 2 origin. The two exceptional reductions realize the pointwise Albert norms as smooth cubic fields on the same complexified deformation bundle. They arise by extension of structure group from Fr φ \operatorname{Fr}_{\varphi} Fr φ , so their existence imposes no additional topological obstruction beyond the given G 2 G_2 G 2 -structure. Here we use extension of structure group to describe the passage from G 2 G_2 G 2 to the larger group G ± ≅ E 6 ( C ) G_{\pm} \cong E_6(\mathbb{C}) G ± ≅ E 6 ( C ) . Viewed inside the full GL ( 27 , C ) \operatorname{GL}(27,\mathbb{C}) GL ( 27 , C ) frame bundle of W φ , C W_{\varphi,\mathbb{C}} W φ , C , however, the resulting G ± G_{\pm} G ± -structure is a reduction of structure group. Since ρ ( G 2 ) ⊂ G ± ⊂ GL ( W C ) \rho(G_2) \subset G_{\pm} \subset\operatorname{GL}(W_{\mathbb{C}}) ρ ( G 2 ) ⊂ G ± ⊂ GL ( W C ) , this reduction exists automatically. The new content is the identification of the exact stabilizers of the two critical cubics as G ± ≅ E 6 ( C ) G_{\pm} \cong E_6(\mathbb{C}) G ± ≅ E 6 ( C ) , while the passage from invariant tensors to reductions is standard (cf. [3 §3.1.1] for the standard approach in the G 2 G_2 G 2 context). The common G 2 C G_2^{\mathbb{C}} G 2 C -reduction of Theorem 5.1 (3) is not the full intersection of F + \mathcal{F}_+ F + and F − \mathcal{F}_- F − . By Proposition 4.2 ,
G + ∩ G − = μ 3 × ρ ( G 2 C ) , G_+ \cap G_- = \mu_3 \times\rho(G_2^{\mathbb{C}}), G + ∩ G − = μ 3 × ρ ( G 2 C ) ,
so F + ∩ F − \mathcal{F}_+ \cap\mathcal{F}_- F + ∩ F − has this structure group, with Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C as its distinguished identity-component reduction. Thus the central μ 3 \mu_3 μ 3 survives in the intersection of the two exceptional reductions. The real adapted frames also lie in both compact reductions of § 6.2 , which do not contain Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C .
The Hermitian form and the compact reductions
Apply the Hermitian form of Lemma 4.17 fiberwise to W φ , C W_{\varphi,\mathbb{C}} W φ , C . Its value on h , k h,k h , k is ⟨ h , k ‾ ⟩ W \langle h,\overline{k}\rangle_W ⟨ h , k ⟩ W , with conjugation fixing W φ W_\varphi W φ . The unitary frames inside each exceptional reduction give its compact form.
Proposition 5.3 (The compact exceptional reductions). Recall K ± K_{\pm} K ± from (4.20), which contains ȷ ± ( G 2 ) = ρ ( G 2 ) \jmath_{\pm}(G_2)=\rho(G_2) ± ( G 2 ) = ρ ( G 2 ) by Proposition 4.18 (2), and put
F ± c = { f ∈ F ± : ⟨ f h , f k ‾ ⟩ W = ⟨ h , k ‾ ⟩ W for all h , k ∈ W C } . \mathcal{F}^{\mathrm{c}}_{\pm}=\left\{f\in\mathcal{F}_{\pm}:\langle fh,\overline{fk}\rangle_W=\langle h,\overline{k}\rangle_W\ \text{for all }h,k\in W_{\mathbb{C}}\right\}. F ± c = { f ∈ F ± : ⟨ f h , f k ⟩ W = ⟨ h , k ⟩ W for all h , k ∈ W C } . Then F ± c \mathcal{F}^{\mathrm{c}}_{\pm} F ± c is a principal K ± K_{\pm} K ± -subbundle of F ± \mathcal{F}_{\pm} F ± , and the map [ p , L ] ↦ f p L [p,L]\mapsto f_pL [ p , L ] ↦ f p L gives
Fr φ × ȷ ± K ± ≅ F ± c , W φ , C ≅ F ± c × K ± W C . \operatorname{Fr}_{\varphi}\times_{\jmath_{\pm}}K_{\pm}\cong\mathcal{F}^{\mathrm{c}}_{\pm},\qquad W_{\varphi,\mathbb{C}}\cong\mathcal{F}^{\mathrm{c}}_{\pm}\times_{K_{\pm}}W_{\mathbb{C}}. Fr φ × ± K ± ≅ F ± c , W φ , C ≅ F ± c × K ± W C . Each F ± c \mathcal{F}^{\mathrm{c}}_{\pm} F ± c contains Fr φ \operatorname{Fr}_{\varphi} Fr φ but does not contain Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C . Conjugation of frames exchanges F + c \mathcal{F}^{\mathrm{c}}_+ F + c and F − c \mathcal{F}^{\mathrm{c}}_- F − c , covering conjugation K + → K − K_+\to K_- K + → K − .
Proof. Adapted coframes are Hermitian isometries, so a local adapted coframe over an open set U ⊂ M U\subset M U ⊂ M identifies the unitary frames in F ± ∣ U \mathcal{F}_{\pm}|_U F ± ∣ U with U × K ± U\times K_{\pm} U × K ± . Conjugation preserves the unitarity condition and exchanges the critical cubics, while G 2 C G_2^{\mathbb{C}} G 2 C contains nonunitary elements, so F ± c \mathcal{F}^{\mathrm{c}}_{\pm} F ± c does not contain Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C .
The compact real form is classical (cf. [61 Theorem 3.9.2] ); here it is identified inside the deformation module, and the passage to a bundle uses the same standard reduction correspondence as Theorem 5.1 .
Thus every G 2 G_2 G 2 -structure gives two canonical compact-E 6 E_6 E 6 structures on its complexified deformation bundle, exchanged by conjugation.
Pointings by line fields and Albert algebra bundles
A global Albert algebra structure on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C can be obtained from a pointing , namely a smooth section e ∈ Γ ( W φ , C ) e \in\Gamma(\mathcal{W}_{\varphi,\mathbb{C}}) e ∈ Γ ( W φ , C ) satisfying N ± ( e ) ≠ 0 N_{\pm}(e) \ne0 N ± ( e ) = 0 everywhere. By Theorem 4.9 , the pair ( N ± / N ± ( e ) , e ) (N_{\pm}/N_{\pm}(e),e) ( N ± / N ± ( e ) , e ) defines a smooth bundle of complex Albert algebras, and hence an F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reduction of the G L ( 27 , C ) GL(27,\mathbb{C}) G L ( 27 , C ) -frame bundle of W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C preserving the normalized cubic. To obtain local trivializations, choose a smooth cube root of N ± ( e ) N_{\pm}(e) N ± ( e ) locally. Transitivity in Theorem 4.9 (2) and local sections of the orbit projection G ± → G ± / F 4 ( C ) G_{\pm} \to G_{\pm}/F_{4}(\mathbb{C}) G ± → G ± / F 4 ( C ) , with stabilizer taken at a fixed model norm-one basepoint, then give smooth Albert frames. The F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reduction obtained in this way need not be a reduction of the original exceptional bundle F ± \mathcal{F}_{\pm} F ± , whose frames preserve N ± N_{\pm} N ± itself.
More generally, consider a smooth complex line subbundle L ⊂ W φ , C \mathcal{L} \subset\mathcal{W}_{\varphi,\mathbb{C}} L ⊂ W φ , C on which N ± N_{\pm} N ± is nowhere zero. Fix a model basepoint e 0 ∈ W C e_{0} \in W_{\mathbb{C}} e 0 ∈ W C with n 0 = N ± ( e 0 ) ≠ 0 n_{0}=N_{\pm}(e_{0}) \ne0 n 0 = N ± ( e 0 ) = 0 . The stabilizer of its line in G ± G_{\pm} G ± is
Stab G ± ( C e 0 ) ≅ F 4 ( C ) × μ 3 , \operatorname{Stab}_{G_{\pm}}(\mathbb{C}e_{0}) \cong F_{4}(\mathbb{C}) \times\mu_{3}, Stab G ± ( C e 0 ) ≅ F 4 ( C ) × μ 3 ,
where μ 3 \mu_{3} μ 3 is the group of cube roots of unity, acting by scalar multiplication. Thus L \mathcal{L} L determines an F 4 ( C ) × μ 3 F_{4}(\mathbb{C}) \times\mu_{3} F 4 ( C ) × μ 3 -reduction of F ± \mathcal{F}_{\pm} F ± .
The restriction of N ± N_{\pm} N ± to L \mathcal{L} L is nowhere zero, and therefore trivializes ( L ∗ ) ⊗ 3 (\mathcal{L}^{*})^{\otimes3} ( L ∗ ) ⊗ 3 . Consequently,
L ⊗ 3 ≅ C , with 3 c 1 ( L ) = 0. \mathcal{L}^{\otimes3} \cong\mathbb{C}, \qquad\text{with} \qquad3c_{1}(\mathcal{L})=0. L ⊗ 3 ≅ C , with 3 c 1 ( L ) = 0.
The local sections of L \mathcal{L} L satisfying N ± ( e ) = n 0 N_{\pm}(e)=n_{0} N ± ( e ) = n 0 form a principal μ 3 \mu_{3} μ 3 -bundle. Its class
α ± ( L ) ∈ H 1 ( M ; Z / 3 ) \alpha_{\pm}(\mathcal{L}) \in H^{1}(\mathcal{M};\mathbb{Z}/3) α ± ( L ) ∈ H 1 ( M ; Z /3 )
is the obstruction to a compatible F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reduction of F ± \mathcal{F}_{\pm} F ± . This class detects both possible three-torsion in c 1 ( L ) c_{1}(\mathcal{L}) c 1 ( L ) and nontrivial monodromy of the cubic normalization when L \mathcal{L} L is trivial. In the latter case, if L = C e \mathcal{L}=\mathbb{C}e L = C e for a global nowhere-vanishing section e e e , the obstruction is the mod-three winding class of N ± ( e ) N_{\pm}(e) N ± ( e ) . It vanishes precisely when N ± ( e ) / n 0 N_{\pm}(e)/n_{0} N ± ( e ) / n 0 admits a global smooth cube root. Thus a constant critical norm is sufficient, but not necessary, for an F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reduction of the original exceptional bundle compatible with L \mathcal{L} L .
A real pointing e ∈ Γ ( W φ ) e \in\Gamma(\mathcal{W}_{\varphi}) e ∈ Γ ( W φ ) with nonzero critical norm defines conjugate Albert algebra structures for both signs, since N − ( e ) = N + ( e ) N_{-}(e)=N_{+}(e) N − ( e ) = N + ( e ) . The μ 3 \mu_{3} μ 3 ambiguity may persist for real pointings, as the critical norms may have nontrivial winding even on real sections. The corresponding obstruction classes satisfy
α − ( C e ) = − α + ( C e ) . \alpha_{-}(\mathbb{C}e)=-\alpha_{+}(\mathbb{C}e). α − ( C e ) = − α + ( C e ) .
Consequently, a compatible F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reduction exists for one sign if and only if it exists for the other.
Table 4 distinguishes the data fixed in these constructions from the unrestricted reduction problem.
Table 4. Three global pointing questions.
A natural geometric source of real pointings is a line field , by which we mean a smooth real line subbundle ℓ ⊂ T M \ell\subset T\mathcal{M} ℓ ⊂ T M . Let P ℓ P_{\ell} P ℓ denote the orthogonal projection onto ℓ \ell ℓ , viewed as a symmetric covariant tensor using g φ g_{\varphi} g φ , and define
E ℓ = 7 P ℓ − g φ ∈ Γ ( W φ ) . (152) E_{\ell}=7P_{\ell}-g_{\varphi} \in\Gamma(\mathcal{W}_{\varphi}). \tag*{(152)} E ℓ = 7 P ℓ − g φ ∈ Γ ( W φ ) . (152)
Locally, P ℓ = P u = u ♭ ⊗ u ♭ P_{\ell}=P_u=u^{\flat}\otimes u^{\flat} P ℓ = P u = u ♭ ⊗ u ♭ and E ℓ = E u E_{\ell}=E_u E ℓ = E u for a unit section u u u of ℓ \ell ℓ , as in (16) and (17) . At each point, E ℓ E_{\ell} E ℓ is a basepoint of S U ( 3 ) SU(3) S U ( 3 ) type. The stabilizer of an unoriented line in the compact G 2 G_2 G 2 is S U ( 3 ) ⋊ Z / 2 SU(3)\rtimes\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 , with identity component S U ( 3 ) SU(3) S U ( 3 ) (Proposition 4.13). Accordingly, the construction requires no orientation of ℓ \ell ℓ , although a corresponding global S U ( 3 ) SU(3) S U ( 3 ) -reduction of the G 2 G_2 G 2 -bundle would require such an orientation.
Finally, by the transport (131) and Remark 4.11, every basepoint is G ± G_{\pm} G ± -equivalent, up to a nonzero scalar, to one of the form E u = 7 P u − g E_u=7P_u-g E u = 7 P u − g . This transport is pointwise and uses the complex group G ± G_{\pm} G ± , so it need not assemble into a global gauge transformation (see the discussion after Corollary 4.10). Real line fields therefore provide a canonical geometric construction of constant-norm pointings, but do not exhaust all possible pointings.
Fix the model unit vector e 1 ∈ V e_1 \in V e 1 ∈ V and recall E 1 = 7 P 1 − g E_1=7P_1-g E 1 = 7 P 1 − g from (17) . Write H E 1 = Stab G 2 ( E 1 ) ≅ S U ( 3 ) ⋊ Z / 2 H_{E_1}=\operatorname{Stab}_{G_2}(E_1)\cong SU(3)\rtimes\mathbb{Z}/2 H E 1 = Stab G 2 ( E 1 ) ≅ S U ( 3 ) ⋊ Z /2 , as in Proposition 4.13. Define the bundle of coframes adapted to ℓ \ell ℓ by
Fr ℓ = { p ∈ Fr φ ∣ x : x ∈ M , f p ( E 1 ) = E ℓ ( x ) } . (153) \operatorname{Fr}_{\ell}=\{p\in\operatorname{Fr}_{\varphi|x}:x\in M,\ f_p(E_1)=E_{\ell}(x)\}. \tag*{(153)} Fr ℓ = { p ∈ Fr φ ∣ x : x ∈ M , f p ( E 1 ) = E ℓ ( x )} . (153)
Equivalently, the dual frame of p p p sends the model line R e 1 \mathbb{R}e_1 R e 1 to ℓ x \ell_x ℓ x . The transitive action of G 2 G_2 G 2 on unit vectors ensures that such coframes exist locally [61 ] , Theorem 1.9.2, p. 17. If an orientation of ℓ \ell ℓ is fixed by a unit section u u u , requiring instead that the dual frame send e 1 e_1 e 1 to u x u_x u x gives the principal S U ( 3 ) SU(3) S U ( 3 ) -subbundle of Fr ℓ \operatorname{Fr}_{\ell} Fr ℓ .
Lemma 5.4 (The basepoint of a line field).
(i) Every 7-manifold without boundary carries a smooth line field.
(ii) For every line field ℓ \ell ℓ , E ℓ = 7 P ℓ − g φ E_{\ell}=7P_{\ell}-g_{\varphi} E ℓ = 7 P ℓ − g φ is a smooth global section of W φ W_{\varphi} W φ , independent of the orientation of ℓ \ell ℓ , and
N ± ( E ℓ ) = n ± = 42 ( 5 + 2 λ ± ) ≠ 0. N_{\pm}(E_{\ell})=n_{\pm}=42(5+2\lambda_{\pm})\ne0. N ± ( E ℓ ) = n ± = 42 ( 5 + 2 λ ± ) = 0. (iii) For the model basepoint E 1 = 7 P 1 − g E_1=7P_1-g E 1 = 7 P 1 − g , the bundle Fr ℓ \operatorname{Fr}_{\ell} Fr ℓ is a principal H E 1 H_{E_1} H E 1 -subbundle of Fr φ \operatorname{Fr}_{\varphi} Fr φ .
Proof. The first assertion follows componentwise from the classical existence of a nowhere-vanishing section of T M TM T M ; see [41 ] , Corollary 14.7 and the following paragraph, p. 145 for closed odd-dimensional manifolds, including nonorientable ones, and [34 ] , Chapter 5, Section 2, Exercise 6(c), p. 139] for the connected noncompact case. Since u ♭ ⊗ u ♭ u^\flat\otimes u^\flat u ♭ ⊗ u ♭ is invariant under u ↦ − u u \mapsto-u u ↦ − u , E ℓ E_\ell E ℓ is globally defined even when ℓ \ell ℓ is nonorientable. Its constant norm follows from (4.22). For the final assertion, a local adapted coframe identifies the frames taking E 1 E_1 E 1 to E ℓ E_\ell E ℓ with the stabilizer H E 1 H_{E_1} H E 1 .
Thus every real line field determines global Albert algebra structures and F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reductions of both original exceptional bundles F ± \mathcal{F}_\pm F ± . In Table 4 , this is an explicit constant-norm solution for the line L = C E ℓ \mathcal{L}=\mathbb{C}E_\ell L = C E ℓ , with n 0 = n ± n_0=n_\pm n 0 = n ± . Figure 2 shows these reductions together with the pointed reductions constructed below.
Figure 2. Inclusions of principal bundles, with their structure groups. The lower row requires a line subbundle ℓ \ell ℓ of T M TM T M and is constructed in Lemma 5.4 and Theorem 5.5 (Section 5.3).
For each tangent line we write the classical based Albert operations explicitly on the complex deformation bundle. We use the real sector bundles W r ( ℓ ) ⊂ W φ \mathcal{W}_r(\ell)\subset W_\varphi W r ( ℓ ) ⊂ W φ , r = 6 , 8 , 12 r=6,8,12 r = 6 , 8 , 12 , obtained from the splitting (4.24); their complexifications will be denoted by W r ( ℓ ) C \mathcal{W}_r(\ell)_{\mathbb{C}} W r ( ℓ ) C .
The decomposition and trace operators. Fix a line field ℓ ⊂ T M \ell\subset TM ℓ ⊂ T M , and let E ℓ E_\ell E ℓ and n ± n_\pm n ± be as above. We suppress the chosen root on T E ℓ TE_\ell T E ℓ throughout, restoring signs on real structures when both occur. The pointwise decomposition and projectors of Lemma 4.19 apply fiberwise to ℓ \ell ℓ . For a local unit section u u u of ℓ \ell ℓ , use the projectors Π 1 , Π 6 , Π 8 , Π 12 \Pi_1,\Pi_6,\Pi_8,\Pi_{12} Π 1 , Π 6 , Π 8 , Π 12 of (4.25), with P u = P ℓ P_u=P_\ell P u = P ℓ and E u = E ℓ E_u=E_\ell E u = E ℓ . Their formulas depend smoothly on g φ g_\varphi g φ , φ \varphi φ and P ℓ P_\ell P ℓ and are invariant under u ↦ − u u\mapsto-u u ↦ − u , so their images define smooth global sector bundles. The based trace and trace form are given by (4.22) and (4.3); the trace operator T E ℓ T_{E_\ell} T E ℓ and its inverse have the sector expressions (4.26). Define the pointed frame subbundle and based algebra bundle by
F ± , ℓ = { f ∈ F ± ∣ x : x ∈ M , f ( E 1 ) = E ℓ ( x ) } , J E ℓ = ( W φ , C , N ± / n ± , E ℓ , ∘ E ℓ ) . \mathcal{F}_{\pm,\ell}=\{f\in\mathcal{F}_{\pm|x}:x\in M,\ f(E_1)=E_\ell(x)\},\qquad\mathcal{J}_{E_\ell}=(W_{\varphi,\mathbb{C}},N_\pm/n_\pm,E_\ell,\circ_{E_\ell}). F ± , ℓ = { f ∈ F ± ∣ x : x ∈ M , f ( E 1 ) = E ℓ ( x )} , J E ℓ = ( W φ , C , N ± / n ± , E ℓ , ∘ E ℓ ) .
For the explicit operations, write λ = λ ± \lambda=\lambda_\pm λ = λ ± and λ ‾ = λ ∓ \overline{\lambda}=\lambda_\mp λ = λ ∓ . We use F ν = A + ν B F_\nu=A+\nu B F ν = A + ν B also on symmetric tensors with nonzero trace, with A A A and B B B as defined in (2.31)–(2.32).
Theorem 5.5 (The Albert algebra bundle of a line field). Let ℓ ⊂ T M \ell\subset TM ℓ ⊂ T M be a line field. Then
F ± , ℓ ≅ Fr ℓ × H E 1 F 4 , E 1 \mathcal{F}_{\pm,\ell}\cong\operatorname{Fr}_\ell\times_{H_{E_1}}F_{4,E_1} F ± , ℓ ≅ Fr ℓ × H E 1 F 4 , E 1 is a principal F 4 , E 1 ≅ F 4 ( C ) F_{4,E_1}\cong F_4(\mathbb{C}) F 4 , E 1 ≅ F 4 ( C ) -subbundle of F ± \mathcal{F}_\pm F ± . The based algebras J E ℓ \mathcal{J}_{E_\ell} J E ℓ form a smooth bundle of complex Albert algebras, locally trivialized by sections of Fr ℓ \operatorname{Fr}_\ell Fr ℓ .
Their adjoint and Jordan product are given explicitly by
h ♯ = 3 + 4 λ 56 F λ ( h ) − 1 2 F λ ˉ ( P ℓ , F λ ( h ) ) + 3 n ± ⟨ E ℓ , F λ ( h ) ⟩ W E ℓ , h ∘ E ℓ k = 3 + 4 λ 56 F λ ( h , k ) − 1 2 F λ ˉ ( P ℓ , F λ ( h , k ) ) + 1 4 ( h ( u , u ) k + k ( u , u ) h ) . (154) \begin{aligned}
h^\sharp&= \frac{3+4\lambda}{56}F_\lambda(h)-\frac{1}{2}F_{\bar{\lambda}}(P_\ell,F_\lambda(h))+\frac{3}{n_\pm}\langle E_\ell,F_\lambda(h)\rangle_W E_\ell,\\
h\circ_{E_\ell} k &= \frac{3+4\lambda}{56}F_\lambda(h,k)-\frac{1}{2}F_{\bar{\lambda}}(P_\ell,F_\lambda(h,k))+\frac{1}{4}\left(h(u,u)k+k(u,u)h\right). \tag*{(154)}
\end{aligned} h ♯ h ∘ E ℓ k = 56 3 + 4 λ F λ ( h ) − 2 1 F λ ˉ ( P ℓ , F λ ( h )) + n ± 3 ⟨ E ℓ , F λ ( h ) ⟩ W E ℓ , = 56 3 + 4 λ F λ ( h , k ) − 2 1 F λ ˉ ( P ℓ , F λ ( h , k )) + 4 1 ( h ( u , u ) k + k ( u , u ) h ) . (154)
Here u u u is any local unit section of ℓ \ell ℓ . These operations are polynomial contractions of g φ g_\varphi g φ , φ \varphi φ and P ℓ P_\ell P ℓ , and are defined also at noninvertible elements.
Proof. By (132) , H E 1 ⊂ F 4 , E 1 H_{E_1}\subset F_{4,E_1} H E 1 ⊂ F 4 , E 1 , and by Proposition 4.12, we have F 4 , E 1 ≅ F 4 ( C ) F_{4,E_1}\cong F_4(\mathbb{C}) F 4 , E 1 ≅ F 4 ( C ) . A local section p p p of F r ℓ F_r\ell F r ℓ induces a frame f p ∈ F ± f_p\in\mathcal{F}_\pm f p ∈ F ± with f p ( E 1 ) = E ℓ f_p(E_1)=E_\ell f p ( E 1 ) = E ℓ . Every other such frame is of the form f p L f_pL f p L , with L ∈ F 4 , E 1 L\in F_{4,E_1} L ∈ F 4 , E 1 . This proves the principal-bundle isomorphism. Theorem 4.9 then gives the smooth Albert algebra bundle.
For the Jordan operations, recall F λ ( E ℓ ) = ( 5 + 2 λ ) E ℓ F_\lambda(E_\ell)=(5+2\lambda)E_\ell F λ ( E ℓ ) = ( 5 + 2 λ ) E ℓ , and n ± = 42 ( 5 + 2 λ ) n_\pm=42(5+2\lambda) n ± = 42 ( 5 + 2 λ ) from (4.22). For traceless X X X , the definition (37) and the identity (41) give A ( g φ , X ) = X A(g_\varphi,X)=X A ( g φ , X ) = X and B ( g φ , X ) = − X B(g_\varphi,X)=-X B ( g φ , X ) = − X , hence F λ ˉ ( g φ , X ) = ( 1 − λ ˉ ) X F_{\bar{\lambda}}(g_\varphi,X)=(1-\bar{\lambda})X F λ ˉ ( g φ , X ) = ( 1 − λ ˉ ) X . Since E ℓ = 7 P ℓ − g φ E_\ell=7P_\ell-g_\varphi E ℓ = 7 P ℓ − g φ and λ ˉ = 1 4 − λ \bar{\lambda}=\frac{1}{4}-\lambda λ ˉ = 4 1 − λ , it follows that
− 3 n ± F λ ˉ ( F λ ( E ℓ ) , X ) = − 1 14 F λ ˉ ( E ℓ , X ) = 3 + 4 λ 56 X − 1 2 F λ ˉ ( P ℓ , X ) . -\frac{3}{n_\pm}F_{\bar{\lambda}}(F_\lambda(E_\ell),X)=-\frac{1}{14}F_{\bar{\lambda}}(E_\ell,X)=\frac{3+4\lambda}{56}X-\frac{1}{2}F_{\bar{\lambda}}(P_\ell,X). − n ± 3 F λ ˉ ( F λ ( E ℓ ) , X ) = − 14 1 F λ ˉ ( E ℓ , X ) = 56 3 + 4 λ X − 2 1 F λ ˉ ( P ℓ , X ) .
Apply this identity with X = F λ ( h ) X=F_\lambda(h) X = F λ ( h ) and X = F λ ( h , k ) X=F_\lambda(h,k) X = F λ ( h , k ) in (130) , and use tr E ℓ ( h ) = 1 2 h ( u , u ) \operatorname{tr}_{E_\ell}(h)=\frac{1}{2}h(u,u) tr E ℓ ( h ) = 2 1 h ( u , u ) . This proves (5.5) directly. The formulas depend on the local unit section u u u only through P ℓ P_\ell P ℓ and evaluations such as h ( u , u ) h(u,u) h ( u , u ) , so they are globally defined. □
The reconstruction of a Jordan product from a normalized cubic and a basepoint is classical, due to Springer [56 p. 261, equations (9)–(12)] , and was recalled in Section 4.2. Here we realize the classical construction globally, with explicit tensor operations determined by the line field.
On ( M , φ , ℓ ) (\mathcal{M},\varphi,\ell) ( M , φ , ℓ ) the exceptional Jordan product is therefore a natural tensor operation. The Euclidean real forms of Section 4 also transfer to a manifold. We keep the root as a subscript on the real structure ς ℓ , ± \varsigma_{\ell,\pm} ς ℓ , ± and its fixed bundle J ℓ , ± \mathcal{J}_{\ell,\pm} J ℓ , ± , since both signs occur below.
The Euclidean real structure. Write s ↦ s ˉ s\mapsto\bar{s} s ↦ s ˉ for the natural complex conjugation on W φ , C = W φ ⊗ R C W_{\varphi,\mathbb{C}}=W_\varphi\otimes_{\mathbb{R}}\mathbb{C} W φ , C = W φ ⊗ R C . For either critical parameter, define the antilinear map
ς ℓ , ± : Γ ( W φ , C ) ⟶ Γ ( W φ , C ) , s ⟼ 1 14 T E ℓ − 1 ( s ˉ ) . (155) \varsigma_{\ell,\pm}:\Gamma(W_{\varphi,\mathbb{C}})\longrightarrow\Gamma(W_{\varphi,\mathbb{C}}),\qquad
s\longmapsto\frac{1}{14}T_{E_\ell}^{-1}(\bar{s}). \tag*{(155)} ς ℓ , ± : Γ ( W φ , C ) ⟶ Γ ( W φ , C ) , s ⟼ 14 1 T E ℓ − 1 ( s ˉ ) . (155)
Here T E ℓ T_{E_\ell} T E ℓ is the trace operator for the chosen root, and the bar denotes fiberwise conjugation. Thus, for a ∈ C ∞ ( M , C ) a\in C^\infty(\mathcal{M},\mathbb{C}) a ∈ C ∞ ( M , C ) ,
ς ℓ , ± ( a s ) = a ˉ ς ℓ , ± ( s ) . \varsigma_{\ell,\pm}(as)=\bar{a}\,\varsigma_{\ell,\pm}(s). ς ℓ , ± ( a s ) = a ˉ ς ℓ , ± ( s ) .
Equivalently, using the fiberwise projectors of (140) ,
ς ℓ , ± ( s ) = Π 1 s ˉ − Π 6 s ˉ + − 1 ± 3 i 7 8 Π 8 s ˉ + 3 ∓ i 7 4 Π 12 s ˉ . \varsigma_{\ell,\pm}(s)=\Pi_{1\bar{s}}-\Pi_{6\bar{s}}+\frac{-1\pm3i\sqrt{7}}{8}\Pi_{8\bar{s}}+\frac{3\mp i\sqrt{7}}{4}\Pi_{12\bar{s}}. ς ℓ , ± ( s ) = Π 1 s ˉ − Π 6 s ˉ + 8 − 1 ± 3 i 7 Π 8 s ˉ + 4 3 ∓ i 7 Π 12 s ˉ .
The next proposition shows that this map is an involutive automorphism of the based Albert algebra bundle, with a Euclidean Albert algebra as its fixed real form.
Proposition 5.6 (The Euclidean real form). Fix either sign and a line field ℓ \ell ℓ . The map ς ℓ , ± \varsigma_{\ell,\pm} ς ℓ , ± is an antilinear involutive automorphism of J E ℓ \mathcal{J}_{E_\ell} J E ℓ . Its fixed bundle is
J ℓ , ± = R E ℓ ⊕ i W 6 ( ℓ ) ⊕ a 8 , ± W 8 ( ℓ ) ⊕ a 12 , ± W 12 ( ℓ ) , \mathcal{J}_{\ell,\pm}=\mathbb{R}E_\ell\oplus iW_6(\ell)\oplus a_{8,\pm}W_8(\ell)\oplus a_{12,\pm}W_{12}(\ell), J ℓ , ± = R E ℓ ⊕ i W 6 ( ℓ ) ⊕ a 8 , ± W 8 ( ℓ ) ⊕ a 12 , ± W 12 ( ℓ ) , where the coefficients a 8 , ± a_{8,\pm} a 8 , ± and a 12 , ± a_{12,\pm} a 12 , ± are those of Proposition 4.22. In particular,
Γ ( J ℓ , ± ) = { s ∈ Γ ( W φ , C ) : ς ℓ , ± ( s ) = s } . \Gamma(\mathcal{J}_{\ell,\pm})=\{s\in\Gamma(\mathcal{W}_{\varphi,\mathbb{C}}):\varsigma_{\ell,\pm}(s)=s\}. Γ ( J ℓ , ± ) = { s ∈ Γ ( W φ , C ) : ς ℓ , ± ( s ) = s } . Each fiber is a Euclidean Albert algebra with determinant N ± / n ± N_\pm/n_\pm N ± / n ± , and complex conjugation exchanges J ℓ , + \mathcal{J}_{\ell,+} J ℓ , + and J ℓ , − \mathcal{J}_{\ell,-} J ℓ , − .
At λ + \lambda_+ λ + , the compact reduction F + c \mathcal{F}_+^c F + c consists precisely of the frames f ∈ F + f\in\mathcal{F}_+ f ∈ F + satisfying
⟨ f h , ς ℓ , + ( f k ) ⟩ E ℓ = ⟨ h , ς e 1 , + k ⟩ E 1 ( h , k ∈ W C ) , \langle fh,\varsigma_{\ell,+}(f k)\rangle_{E_\ell}=\langle h,\varsigma_{e_1,+}k\rangle_{E_1}\qquad(h,k\in W_{\mathbb{C}}), ⟨ f h , ς ℓ , + ( f k ) ⟩ E ℓ = ⟨ h , ς e 1 , + k ⟩ E 1 ( h , k ∈ W C ) , where ⟨ ⋅ , ⋅ ⟩ E 1 \langle\mathord{\cdot},\mathord{\cdot}\rangle_{E_1} ⟨ ⋅ , ⋅ ⟩ E 1 is the based trace form at the model basepoint E 1 E_1 E 1 .
Proof. A local section of Fr ℓ \operatorname{Fr}_\ell Fr ℓ transfers Lemma 4.21 and Proposition 4.22 to each fiber. These transfers agree because ς A u , ± ρ ( A ) = ρ ( A ) ς u , ± \varsigma_{Au,\pm}\rho(A)=\rho(A)\varsigma_{u,\pm} ς A u , ± ρ ( A ) = ρ ( A ) ς u , ± for A ∈ G 2 A\in G_2 A ∈ G 2 , as recalled at the end of § 5.7 . Since the projectors and trace operator are smooth and invariant under u ↦ − u u\mapsto-u u ↦ − u , the resulting involutions and their fixed spaces are globally defined. Complex conjugation exchanges the trace operators for the two critical parameters, and hence exchanges the fixed real forms J ℓ , + \mathcal{J}_{\ell,+} J ℓ , + and J ℓ , − \mathcal{J}_{\ell,-} J ℓ , − .
Finally, Lemma 4.21 gives the Hermitian identity
14 ⟨ h , ς ℓ , ± k ⟩ E ℓ = ⟨ h , k ‾ ⟩ W . 14\langle h,\varsigma_{\ell,\pm}k\rangle_{E_\ell}=\langle h,\overline{k}\rangle_W. 14 ⟨ h , ς ℓ , ± k ⟩ E ℓ = ⟨ h , k ⟩ W .
Thus the displayed frame condition at λ + \lambda_+ λ + is equivalent to preservation of the Hermitian form induced by the W W W -pairing, which defines F + c \mathcal{F}_+^c F + c . □ \square □
Proposition 5.6 applies the pointwise results of Proposition 4.22, and their classical sources, fiberwise; it involves no new classification of real forms. As in Proposition 4.22, the Euclidean real form meets the real deformation bundle only in R E ℓ \mathbb{R}E_\ell R E ℓ .
Together, Lemma 5.4, Theorem 5.5 and Proposition 5.6 prove Theorem B.
We now consider the unrestricted reduction problem in the last row of Table 4 . The choice of a tangent line field changes the pointing geometrically, but does not change the homotopy class of the resulting F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reduction.
Proposition 5.7 (Homotopy uniqueness). For either sign,
{ N ± = 1 } ≅ E 6 ( C ) / F 4 ( C ) ≃ E 6 / F 4 , \{N_\pm=1\}\cong E_6(\mathbb{C})/F_4(\mathbb{C})\simeq E_6/F_4, { N ± = 1 } ≅ E 6 ( C ) / F 4 ( C ) ≃ E 6 / F 4 , where E 6 / F 4 E_6/F_4 E 6 / F 4 denotes the quotient of the compact simply connected groups of these types, a compact symmetric space that is 8-connected and has π 9 ≅ Z \pi_9\cong\mathbb{Z} π 9 ≅ Z . On a 7-manifold every F 4 ( C ) F_4(\mathbb{C}) F 4 ( C ) -reduction of F ± \mathcal{F}_\pm F ± is homotopic to F ± , ℓ \mathcal{F}_{\pm,\ell} F ± , ℓ , and its principal-bundle isomorphism class is independent of ℓ \ell ℓ .
The homotopy equivalence follows from the polar decompositions [61 Theorems 2.13.2 and 3.14.1] ; the connectivity and π 9 \pi_9 π 9 follow from Conlon’s cell structure [9 Theorem 1.1, p. 411, the following cell-structure remarks on p. 412, and the report on p. 414] . Obstruction theory for sections then gives homotopy uniqueness over a 7-manifold. Thus an unrestricted norm-one pointing exists and is unique up to homotopy; compatibility with a prescribed complex line remains subject to the obstruction described above.
We can also treat all tangent lines at once.
The G 2 G_2 G 2 -structure also supplies a geometric octonion bundle. We compare its entrywise Hermitian construction with the Albert bundle above.
Connections and parallel structures
We now determine which connections preserve these exceptional structures and show that the metric singles out one of them. Write D D D for a connection on T M TM T M and for the induced connections on the deformation bundles and frame bundles. In particular, D L C D^{LC} D L C is the Levi-Civita connection. Fix the intrinsic torsion convention
D X L C φ = ι T ( X ) ψ , D m L C φ a b c = T ( e m ) r ψ r a b c , D^{LC}_X\varphi=\iota_{T(X)}\psi,\qquad D^{LC}_m\varphi_{abc}=T(e_m)^r\psi_{rabc}, D X L C φ = ι T ( X ) ψ , D m L C φ ab c = T ( e m ) r ψ r ab c ,
as in Bryant’s description of intrinsic torsion and its four classes [3 §4] . Extend the notation Θ v ( w ) = w × v \Theta_v(w)=w\times v Θ v ( w ) = w × v to arbitrary tangent vectors v v v . From the decomposition s o ( 7 ) = g 2 ⊕ Λ 7 2 \mathfrak{so}(7)=\mathfrak{g}_2\oplus\Lambda^2_7 so ( 7 ) = g 2 ⊕ Λ 7 2 , write every metric connection on T M TM T M uniquely as
D = D L C + β + Θ w ( ⋅ ) , β ∈ Ω 1 ( M , g 2 ) , w ∈ Ω 1 ( M , T M ) , (156) D=D^{LC}+\beta+\Theta_w(\mathord{\cdot}),\qquad\beta\in\Omega^1(M,\mathfrak{g}_2),\quad w\in\Omega^1(M,TM), \tag*{(156)} D = D L C + β + Θ w ( ⋅ ) , β ∈ Ω 1 ( M , g 2 ) , w ∈ Ω 1 ( M , T M ) , (156)
where the Lie-algebra-valued forms are read in adapted frames. The minimal connection is
D min = D L C − 1 3 Θ T ( ⋅ ) . D^{\min}=D^{LC}-\frac{1}{3}\Theta_{T(\mathord{\cdot})}. D m i n = D L C − 3 1 Θ T ( ⋅ ) .
It is metric and preserves φ \varphi φ and ψ \psi ψ [7 §3.1] .
Among metric connections on T M \mathcal{T}M T M , preservation of an exceptional cubic is precisely the G 2 G_{2} G 2 -compatibility condition. First we need a deformation property of C φ C_{\varphi} C φ . With the metric held fixed, define the infinitesimal variation by
L φ : T M ⟶ S 3 W φ ∗ , L φ ( v ) = d d t ∣ t = 0 C φ + t ι v ψ . L_{\varphi}:\mathcal{T}M\longrightarrow S^{3}W_{\varphi}^{*},\qquad L_{\varphi}(v)=\left.\frac{d}{dt}\right|_{t=0}C_{\varphi+t\iota_{v}\psi}. L φ : T M ⟶ S 3 W φ ∗ , L φ ( v ) = d t d t = 0 C φ + t ι v ψ .
Lemma 5.10 (Variation of the φ \varphi φ -cubic). The map L φ L_{\varphi} L φ is G 2 G_{2} G 2 -equivariant and fiberwise injective, and its image is an irreducible 7 component. In particular,
im L φ ∩ span C { C 0 , C φ } = { 0 } . \operatorname{im}L_{\varphi}\cap\operatorname{span}_{\mathbb{C}}\{C_{0},C_{\varphi}\}=\{0\}. im L φ ∩ span C { C 0 , C φ } = { 0 } .
Proof. We work in a G 2 G_{2} G 2 -adapted orthonormal frame. This deformation keeps the metric fixed to first order, so differentiating the contraction (33) defining C φ C_{\varphi} C φ with a constant metric gives
L φ ( v ) ( h 1 , h 2 , h 3 ) = 2 ( ι v ψ ) a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 . L_{\varphi}(v)(h_{1},h_{2},h_{3})=2(\iota_{v}\psi)_{a_{1}a_{2}a_{3}}\varphi_{b_{1}b_{2}b_{3}}h^{a_{1}b_{1}}_{1}h^{a_{2}b_{2}}_{2}h^{a_{3}b_{3}}_{3}. L φ ( v ) ( h 1 , h 2 , h 3 ) = 2 ( ι v ψ ) a 1 a 2 a 3 φ b 1 b 2 b 3 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 .
This expression is G 2 G_{2} G 2 -equivariant since it involves only the invariant tensors g g g , φ \varphi φ and ψ \psi ψ .
Recall that s o ( 7 ) \mathfrak{so}(7) so ( 7 ) decomposes as g 2 ⊕ Λ 7 2 \mathfrak{g}_{2}\oplus\Lambda^{2}_{7} g 2 ⊕ Λ 7 2 , where Λ 7 2 \Lambda^{2}_{7} Λ 7 2 consists of the maps Θ v \Theta_{v} Θ v . The standard contraction identities give
θ Θ v = − 3 ι v ψ . \theta_{\Theta_{v}}=-3\iota_{v}\psi. θ Θ v = − 3 ι v ψ .
where θ M \theta_{M} θ M is the 3-form (23) . Indeed,
( θ Θ v ) i j k = v b ( φ i b a φ j k a + φ j b a φ i k a + φ k b a φ i j a ) = − 3 v b ψ b i j k . (\theta_{\Theta_{v}})_{ijk}=v^{b}\left(\varphi_{iba}\varphi^{a}_{jk}+\varphi_{jba}\varphi^{a}_{ik}+\varphi_{kba}\varphi^{a}_{ij}\right)=-3v^{b}\psi_{bijk}. ( θ Θ v ) ij k = v b ( φ iba φ j k a + φ j ba φ ik a + φ k ba φ ij a ) = − 3 v b ψ bij k .
Set R t = exp ( t Θ v ) R_{t}=\exp(t\Theta_{v}) R t = exp ( t Θ v ) . By naturality under orthogonal transformations,
C R t ∗ φ ( h 1 , h 2 , h 3 ) = C φ ( ρ ( R t ) h 1 , ρ ( R t ) h 2 , ρ ( R t ) h 3 ) . C_{R_{t}^{*}\varphi}(h_{1},h_{2},h_{3})=C_{\varphi}\bigl(\rho(R_{t})h_{1},\rho(R_{t})h_{2},\rho(R_{t})h_{3}\bigr). C R t ∗ φ ( h 1 , h 2 , h 3 ) = C φ ( ρ ( R t ) h 1 , ρ ( R t ) h 2 , ρ ( R t ) h 3 ) .
Recall the infinitesimal action map Φ N \Phi_{N} Φ N of (61) . Differentiation at t = 0 t=0 t = 0 gives
− 3 L φ ( v ) = Φ C φ ( ρ ( Θ v ) ) . -3L_{\varphi}(v)=\Phi_{C_{\varphi}}\bigl(\rho(\Theta_{v})\bigr). − 3 L φ ( v ) = Φ C φ ( ρ ( Θ v ) ) .
By Proposition 3.23 , ker Φ C φ = ρ ( g 2 ) \ker\Phi_{C_{\varphi}}=\rho(\mathfrak{g}_{2}) ker Φ C φ = ρ ( g 2 ) . Since ρ \rho ρ is injective and Θ v ∈ Λ 7 2 \Theta_{v}\in\Lambda^{2}_{7} Θ v ∈ Λ 7 2 , the right-hand side vanishes only when v = 0 v=0 v = 0 . Hence L φ L_{\varphi} L φ is injective.
Its image is therefore isomorphic to the irreducible 7 representation of G 2 G_{2} G 2 . The invariant plane spanned by C 0 C_{0} C 0 and C φ C_{\varphi} C φ consists of trivial representations, so its intersection with im L φ \operatorname{im}L_{\varphi} im L φ is zero.
The variation formula determines the critical-cubic derivative of any connection induced by a metric connection on T M \mathcal{T}M T M .
Proposition 5.11 (Metric connections and critical cubics). Let D D D be a metric connection on T M \mathcal{T}M T M , written as in (156) , and use the same symbol for its induced connection on W φ , C W_{\varphi,C} W φ , C . Then
D X N ± = λ ± L φ ( T ( X ) + 3 w ( X ) ) . (157) D_{X}N_{\pm}=\lambda_{\pm}L_{\varphi}\bigl(T(X)+3w(X)\bigr). \tag*{(157)} D X N ± = λ ± L φ ( T ( X ) + 3 w ( X ) ) . (157)
Proof. Since β X \beta_X β X annihilates φ \varphi φ and θ Θ v = − 3 ι v ψ \theta_{\Theta_v}=-3\iota_v\psi θ Θ v = − 3 ι v ψ , we have D X φ = ι T ( X ) + 3 w ( X ) ψ D_X\varphi=\iota_{T(X)}+3w(X)\psi D X φ = ι T ( X ) + 3 w ( X ) ψ . Metric compatibility gives D C 0 = 0 DC_0=0 D C 0 = 0 , and since D g = 0 Dg=0 D g = 0 , D X C φ D_XC_\varphi D X C φ is the variation of C φ C_\varphi C φ in the direction D X φ D_X\varphi D X φ with the metric fixed. The definition of L φ L_\varphi L φ therefore gives (5.8).
The injectivity and representation type of L φ L_\varphi L φ turn this derivative formula into preservation criteria.
Corollary 5.12 (Critical-cubic preservation criteria). (1) For a metric connection D D D on T M TM T M , the following conditions are equivalent:
(a) D φ = 0 D\varphi=0 D φ = 0 ;
(b) D N ± = 0 DN_\pm=0 D N ± = 0 ;
(c) D D D preserves the line C N ± \mathbb{C}N_\pm C N ± .
In particular, a metric connection preserves either critical cubic if and only if it is a G 2 G_2 G 2 -connection, in which case it preserves both critical cubics.
(2) For the Levi-Civita connection,
D X L C N ± = λ ± L φ ( T ( X ) ) . D_X^{LC}N_\pm=\lambda_\pm L_\varphi(T(X)). D X L C N ± = λ ± L φ ( T ( X )) . Thus D L C N ± = 0 D^{LC}N_\pm=0 D L C N ± = 0 if and only if T = 0 T=0 T = 0 , that is, if and only if φ \varphi φ is torsion-free.
(3) For an arbitrary complex connection on W φ , C \mathbb{W}_{\varphi,\mathbb{C}} W φ , C preserving the line C N ± \mathbb{C}N_\pm C N ± , its local connection form in a G 2 G_2 G 2 -adapted frame is ω + α i d W \omega+\alpha\,\mathrm{id}_{W} ω + α id W , where ω \omega ω is g ± \mathfrak{g}_\pm g ± -valued and α \alpha α is a complex-valued one-form. Then
D N ± = − 3 α ⊗ N ± . DN_\pm=-3\alpha\otimes N_\pm. D N ± = − 3 α ⊗ N ± . Hence preserving the line C N ± \mathbb{C}N_\pm C N ± is in general weaker than preserving the cubic N ± N_\pm N ± itself.
Proof. By the preceding proof, D φ = 0 D\varphi=0 D φ = 0 if and only if T + 3 w = 0 T+3w=0 T + 3 w = 0 . Since L φ L_\varphi L φ is injective and λ ± ≠ 0 \lambda_\pm\ne0 λ ± = 0 , we have D N ± = 0 DN_\pm=0 D N ± = 0 if and only if T + 3 w = 0 T+3w=0 T + 3 w = 0 , equivalently, D φ = 0 D\varphi=0 D φ = 0 .
Suppose instead that D D D preserves the line C N ± \mathbb{C}N_\pm C N ± . Then D X N ± D_XN_\pm D X N ± is proportional to N ± N_\pm N ± for every X X X . However, by (5.8), D X N ± D_XN_\pm D X N ± lies in im L φ \operatorname{im}L_\varphi im L φ , which has zero intersection with C N ± \mathbb{C}N_\pm C N ± by Lemma 5.10. Hence D X N ± = 0 D_XN_\pm=0 D X N ± = 0 , and the preceding argument gives D φ = 0 D\varphi=0 D φ = 0 . Conversely, a G 2 G_2 G 2 -connection preserves both invariant cubics, and therefore both critical cubics.
Part 2 follows by taking β = w = 0 \beta=w=0 β = w = 0 . The injectivity of L φ L_\varphi L φ then gives D L C N ± = 0 D^{LC}N_\pm=0 D L C N ± = 0 precisely when T = 0 T=0 T = 0 .
For part 3, recall that the infinitesimal stabilizer of the line C N ± \mathbb{C}N_\pm C N ± is g ± ⊕ C i d W \mathfrak{g}_\pm\oplus\mathbb{C}\,\mathrm{id}_W g ± ⊕ C id W . In a G 2 G_2 G 2 -adapted frame, the components of N ± N_\pm N ± are constant, and the induced action of a connection form on covariant cubic tensors is the negative of Φ N ± \Phi_{N_\pm} Φ N ± . Since
Φ N ± ( ω ) = 0 , Φ N ± ( i d W ) = 3 N ± , \Phi_{N_\pm}(\omega)=0,\qquad\Phi_{N_\pm}(\mathrm{id}_W)=3N_\pm, Φ N ± ( ω ) = 0 , Φ N ± ( id W ) = 3 N ± ,
we obtain
D N ± = − 3 α ⊗ N ± . DN_\pm=-3\alpha\otimes N_\pm. D N ± = − 3 α ⊗ N ± .
Thus the exact cubic is parallel precisely when α = 0 \alpha=0 α = 0 .
Beyond connections induced from T M TM T M , compatible exceptional connections have additional freedom in the 64 64 64 directions. Recall from Proposition 3.31 that
g ± = ρ ( g 2 C ) ⊕ m ± , \mathfrak{g}_{\pm}=\rho(\mathfrak{g}_{2}^{\mathbb{C}})\oplus\mathfrak{m}_{\pm}, g ± = ρ ( g 2 C ) ⊕ m ± ,
where m ± \mathfrak{m}_{\pm} m ± is the irreducible 64 64 64 -dimensional G 2 C G_{2}^{\mathbb{C}} G 2 C -module. The minimal G 2 G_{2} G 2 -connection gives a reference connection on each F ± \mathcal{F}_{\pm} F ± . The difference from any other connection is an adjoint-bundle-valued one-form. The displayed equivariant splitting therefore gives a unique expression
D = D min + ρ ( β ) + Ψ , D=D^{\min}+\rho(\beta)+\Psi, D = D m i n + ρ ( β ) + Ψ ,
where β \beta β is now allowed to be g 2 C \mathfrak{g}_{2}^{\mathbb{C}} g 2 C -valued:
β ∈ Ω 1 ( M , Fr φ × G 2 g 2 C ) , \beta\in\Omega^{1}\left(\mathcal{M},\operatorname{Fr}_{\varphi}\times_{G_{2}}\mathfrak{g}_{2}^{\mathbb{C}}\right), β ∈ Ω 1 ( M , Fr φ × G 2 g 2 C ) ,
Ψ ∈ Ω 1 ( M , Fr φ × G 2 m ± ) . \Psi\in\Omega^{1}\left(\mathcal{M},\operatorname{Fr}_{\varphi}\times_{G_{2}}\mathfrak{m}_{\pm}\right). Ψ ∈ Ω 1 ( M , Fr φ × G 2 m ± ) .
Conversely, both summands are g ± \mathfrak{g}_{\pm} g ± -valued, so every such pair defines a connection preserving N ± N_{\pm} N ± .
Proposition 5.13 (Connections on the exceptional reductions). For a connection D D D on F ± \mathcal{F}_{\pm} F ± written as above, the form Ψ \Psi Ψ is the second fundamental form of the intermediate reduction Fr φ , C ⊂ F ± \operatorname{Fr}_{\varphi,\mathbb{C}}\subset\mathcal{F}_{\pm} Fr φ , C ⊂ F ± , with group G 2 C G_{2}^{\mathbb{C}} G 2 C . It vanishes if and only if D D D preserves the intermediate reduction, equivalently, if D D D preserves both critical cubics.
In particular, a connection preserving one critical cubic and the real structure of W φ , C \mathbb{W}_{\varphi,\mathbb{C}} W φ , C preserves both critical cubics and has Ψ = 0 \Psi=0 Ψ = 0 . Such a connection is induced by a real G 2 G_{2} G 2 -connection on T M TM T M .
Here Ψ \Psi Ψ represents the class of the connection form along Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C in g ± / ρ ( g 2 C ) \mathfrak{g}_{\pm}/\rho(\mathfrak{g}_{2}^{\mathbb{C}}) g ± / ρ ( g 2 C ) , identified with m ± \mathfrak{m}_{\pm} m ± by the reductive splitting. This class is the covariant derivative of a principal reduction [52 authors’ PDF, § 2.10, Lemma 2.10.1 and Definition 2.10.4] , also called its second fundamental form [47 arXiv v4, pp. 34–35] .
Proof. The quotient g ± / ρ ( g 2 C ) \mathfrak{g}_{\pm}/\rho(\mathfrak{g}_{2}^{\mathbb{C}}) g ± / ρ ( g 2 C ) is identified with m ± \mathfrak{m}_{\pm} m ± . The projection of the connection form to this quotient is the second fundamental form of Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C in F ± \mathcal{F}_{\pm} F ± . Since D min D^{\min} D m i n and ρ ( β ) \rho(\beta) ρ ( β ) preserve the intermediate reduction, this projection is precisely Ψ \Psi Ψ . Thus D D D preserves Fr φ , C \operatorname{Fr}_{\varphi,\mathbb{C}} Fr φ , C if and only if Ψ = 0 \Psi=0 Ψ = 0 .
By Proposition 4.2 (1),
g + ∩ g − = ρ ( g 2 C ) . \mathfrak{g}_{+}\cap\mathfrak{g}_{-}=\rho(\mathfrak{g}_{2}^{\mathbb{C}}). g + ∩ g − = ρ ( g 2 C ) .
Hence a connection on F ± \mathcal{F}_{\pm} F ± preserves both critical cubics if and only if its connection form is ρ ( g 2 C ) \rho(\mathfrak{g}_{2}^{\mathbb{C}}) ρ ( g 2 C ) -valued, which is equivalent to Ψ = 0 \Psi=0 Ψ = 0 .
Finally, suppose that D D D also preserves the canonical real structure on W φ , C \mathbb{W}_{\varphi,\mathbb{C}} W φ , C , given by complex conjugation with respect to W φ \mathbb{W}_{\varphi} W φ . Since N − = N + ‾ N_{-}=\overline{N_{+}} N − = N + , preservation of one critical cubic implies preservation of the other. We have just shown that this forces Ψ = 0 \Psi=0 Ψ = 0 . In a real G 2 G_{2} G 2 -adapted frame, the remaining connection form is real and lies in ρ ( g 2 C ) \rho(\mathfrak{g}_{2}^{\mathbb{C}}) ρ ( g 2 C ) , hence in ρ ( g 2 ) \rho(\mathfrak{g}_{2}) ρ ( g 2 ) . It therefore comes from a real G 2 G_{2} G 2 -connection on T M TM T M .
An exceptional reduction does not by itself determine a compatible connection. Using the metric g φ g_\varphi g φ , however, we can select a canonical compatible connection by orthogonally projecting the Levi-Civita connection. We first identify the relevant orthogonal complement. By Remark 3.35, we have the decomposition
End ( W C ) ≅ 1 ⊕ 78 ⊕ 650 \operatorname{End}(W_{\mathbb{C}}) \cong\mathbf{1} \oplus\mathbf{78} \oplus\mathbf{650} End ( W C ) ≅ 1 ⊕ 78 ⊕ 650
into irreducible e 6 \mathfrak{e}_6 e 6 -modules, where the summand 78 \mathbf{78} 78 is the adjoint representation on g ± \mathfrak{g}_{\pm} g ± . The invariant form tr W ( X Y ) \operatorname{tr}_{W}(XY) tr W ( X Y ) is nondegenerate and preserved by conjugation. Since these three irreducible modules are pairwise nonisomorphic and have distinct dimensions, they are mutually orthogonal. We therefore have the tr W \operatorname{tr}_{W} tr W -orthogonal decomposition
End ( W C ) = C id W ⊕ g ± ⊕ c ± , \operatorname{End}(W_{\mathbb{C}})=\mathbb{C}\operatorname{id}_{W}\oplus\mathfrak{g}_{\pm}\oplus\mathfrak{c}_{\pm}, End ( W C ) = C id W ⊕ g ± ⊕ c ± ,
so that g ± = C id W ⊕ c ± \mathfrak{g}_{\pm}=\mathbb{C}\operatorname{id}_{W}\oplus\mathfrak{c}_{\pm} g ± = C id W ⊕ c ± , where c ± ≅ 650 \mathfrak{c}_{\pm}\cong\mathbf{650} c ± ≅ 650 .
Let π ± : End ( W C ) → g ± \pi_{\pm}:\operatorname{End}(W_{\mathbb{C}})\to\mathfrak{g}_{\pm} π ± : End ( W C ) → g ± denote the corresponding orthogonal projection. The projected connection of a connection on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C is the connection on F ± \mathcal{F}_{\pm} F ± whose connection form, in a local frame of F ± \mathcal{F}_{\pm} F ± , is the π ± \pi_{\pm} π ± -component of the original one (see Remark 5.16). Write π g 2 \pi_{\mathfrak{g}_2} π g 2 for the projection associated with s o ( 7 ) = g 2 ⊕ { Θ v : v ∈ V } \mathfrak{so}(7)=\mathfrak{g}_2\oplus\{\Theta_v:v\in V\} so ( 7 ) = g 2 ⊕ { Θ v : v ∈ V } .
Lemma 5.14 (Orthogonality of the 7-directions). For either sign and every v ∈ V v\in V v ∈ V ,
ρ ( Θ v ) ∈ c ± . \rho(\Theta_v)\in\mathfrak{c}_{\pm}. ρ ( Θ v ) ∈ c ± . In particular, on s o ( 7 ) \mathfrak{so}(7) so ( 7 ) ,
π ± ∘ ρ = ρ ∘ π g 2 . \pi_{\pm}\circ\rho=\rho\circ\pi_{\mathfrak{g}_2}. π ± ∘ ρ = ρ ∘ π g 2 . Proof. Since ρ ( G 2 ) ⊂ G ± \rho(G_2)\subset G_{\pm} ρ ( G 2 ) ⊂ G ± and the invariant form tr W ( X Y ) \operatorname{tr}_{W}(XY) tr W ( X Y ) is preserved by conjugation, π ± \pi_{\pm} π ± is G 2 G_2 G 2 -equivariant. Hence the complexified composition
V C → v ↦ Θ v s o ( 7 , C ) → ρ End ( W C ) → π ± g ± V_{\mathbb{C}}\xrightarrow{v\mapsto\Theta_v}\mathfrak{so}(7,\mathbb{C})\xrightarrow{\rho}\operatorname{End}(W_{\mathbb{C}})\xrightarrow{\pi_{\pm}}\mathfrak{g}_{\pm} V C v ↦ Θ v so ( 7 , C ) ρ End ( W C ) π ± g ± is G 2 G_2 G 2 -equivariant. The decomposition
g ± = ρ ( g 2 C ) ⊕ m ± ≅ 14 ⊕ 64 \mathfrak{g}_{\pm}=\rho(\mathfrak{g}_{2}^{\mathbb{C}})\oplus\mathfrak{m}_{\pm}\cong\mathbf{14}\oplus\mathbf{64} g ± = ρ ( g 2 C ) ⊕ m ± ≅ 14 ⊕ 64 contains no copy of 7 \mathbf{7} 7 , so this composition vanishes by Schur’s lemma. Moreover, ρ ( Θ v ) \rho(\Theta_v) ρ ( Θ v ) is skew-adjoint on the real space W W W , hence traceless. Thus it belongs to c ± \mathfrak{c}_{\pm} c ± . The projection identity follows because π ± \pi_{\pm} π ± is the identity on ρ ( g 2 ) \rho(\mathfrak{g}_2) ρ ( g 2 ) .
We apply Lemma 5.14 to the connections induced on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C . As before, we use the same notation for a metric connection on T M TM T M and its induced connection on the deformation bundle.
Proposition 5.15 (The minimal connection as projected Levi-Civita connection).
(1) Let D D D be the connection on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C induced by a metric connection on T M TM T M , written as in (5.7). Then
D = ( D min + ρ ( β ) ) + ρ ( Θ w ( ⋅ ) + T ( ⋅ ) / 3 ) . D=\left(D^{\min}+\rho(\beta)\right)+\rho\left(\Theta_{w(\cdot)+T(\cdot)/3}\right). D = ( D m i n + ρ ( β ) ) + ρ ( Θ w ( ⋅ ) + T ( ⋅ ) /3 ) . For either sign, the first term preserves F ± \mathcal{F}_{\pm} F ± and the second is c ± \mathfrak{c}_{\pm} c ± -valued. Thus both exceptional reductions give the same projected connection, D min + ρ ( β ) D^{\min}+\rho(\beta) D m i n + ρ ( β ) .
(2) In particular, both signs give D min D^{\min} D m i n as the projected Levi-Civita connection. For either sign, it is the unique connection D ~ \widetilde{D} D on F ± \mathcal{F}_{\pm} F ± for which D L C − D ~ D^{LC}-\widetilde{D} D L C − D is g ± ⊥ \mathfrak{g}_{\pm}^{\perp} g ± ⊥ -valued. It is also the unique pointwise closest connection to D L C D^{LC} D L C among all connections on F ± \mathcal{F}_{\pm} F ± , in the norm induced by g φ g_{\varphi} g φ and the Hermitian form on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C . Moreover, D min D^{\min} D m i n preserves the compact reduction F ± c \mathcal{F}_{\pm}^{c} F ± c .
Proof. Substituting
D L C = D min + 1 3 ρ ( Θ T ( ⋅ ) ) D^{LC}=D^{\min}+\frac{1}{3}\rho\bigl(\Theta_{T(\cdot)}\bigr) D L C = D m i n + 3 1 ρ ( Θ T ( ⋅ ) )
gives the decomposition in part 1. Its first term is induced by a G 2 G_{2} G 2 -connection, so it preserves both exceptional reductions and their compact reductions. By Lemma 5.14, the second term lies in c ± \mathfrak{c}_{\pm} c ± .
In particular, D L C − D min D^{LC}-D^{\min} D L C − D m i n is g ± ⊥ \mathfrak{g}_{\pm}^{\perp} g ± ⊥ -valued. Every connection on F ± \mathcal{F}_{\pm} F ± has the form D min + ω D^{\min}+\omega D m i n + ω , with ω \omega ω a g ± ⊥ \mathfrak{g}_{\pm}^{\perp} g ± ⊥ -valued one-form, so the tr W \operatorname{tr}_{W} tr W -orthogonal splitting gives the asserted uniqueness. By Proposition 4.18, g ± \mathfrak{g}_{\pm} g ± is the complexification of a compact real form consisting of skew-Hermitian operators. It is therefore stable under Hermitian adjunction, and, writing ⟨ ⋅ , ⋅ ⟩ H \langle\cdot,\cdot\rangle_{H} ⟨ ⋅ , ⋅ ⟩ H for the Hermitian extension of the pairing ⟨ ⋅ , ⋅ ⟩ End \langle\cdot,\cdot\rangle_{\operatorname{End}} ⟨ ⋅ , ⋅ ⟩ End of Section 3,
⟨ X , Y ⟩ H = tr W ( X Y H ) = 0 ( X ∈ g ± ⊥ , Y ∈ g ± ) . \langle X,Y\rangle_{H}=\operatorname{tr}_{W}(XY^{H})=0 \qquad(X\in\mathfrak{g}_{\pm}^{\perp},\ Y\in\mathfrak{g}_{\pm}). ⟨ X , Y ⟩ H = tr W ( X Y H ) = 0 ( X ∈ g ± ⊥ , Y ∈ g ± ) .
Thus the tr W \operatorname{tr}_{W} tr W -orthogonal splitting is also Hermitian-orthogonal. For every such ω \omega ω ,
∣ D L C − ( D min + ω ) ∣ 2 = ∣ D L C − D min ∣ 2 + ∣ ω ∣ 2 \left|D^{LC}-(D^{\min}+\omega)\right|^{2}=\left|D^{LC}-D^{\min}\right|^{2}+|\omega|^{2} D L C − ( D m i n + ω ) 2 = D L C − D m i n 2 + ∣ ω ∣ 2
pointwise, which proves that D min D^{\min} D m i n is the unique pointwise closest connection to D L C D^{LC} D L C among all connections on F ± \mathcal{F}_{\pm} F ± .
Together with Lemma 5.10, Corollary 5.12 and Lemma 5.14, Proposition 5.15 proves Theorem C.
For the compact reduction F ± c \mathcal{F}_{\pm}^{c} F ± c , Proposition 5.15 reduces to the orthogonal-projection characterization of the minimal connection given by Cleyton and Swann [8 Definition 3.1 and Proposition 3.2] . The proposition shows that the same connection is also the unique closest connection among all connections on the complex exceptional reduction F ± \mathcal{F}_{\pm} F ± .
Remark 5.16 (Projection and the defect of a connection). The projection construction applies to any connection D 0 D^{0} D 0 on W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C , although its two projections need not coincide. Fix a sign and choose a local frame in F ± \mathcal{F}_{\pm} F ± . Decompose the connection form of D 0 D^{0} D 0 according to the tr W \operatorname{tr}_{W} tr W -orthogonal splitting
End ( W C ) = g ± ⊕ g ± ⊥ . \operatorname{End}(\mathcal{W}_{\mathbb{C}})=\mathfrak{g}_{\pm}\oplus\mathfrak{g}_{\pm}^{\perp}. End ( W C ) = g ± ⊕ g ± ⊥ .
The g ± \mathfrak{g}_{\pm} g ± -component defines a connection D ± D^{\pm} D ± preserving F ± \mathcal{F}_{\pm} F ± . Since the projection is G ± G_{\pm} G ± -equivariant, this construction is independent of the chosen frame. For this fixed sign, we may therefore write
D 0 = D ± + ϖ , ϖ = ϖ 1 id W + ϖ c , D^{0}=D^{\pm}+\varpi,\qquad\varpi=\varpi_{1}\operatorname{id}_{W}+\varpi_{c}, D 0 = D ± + ϖ , ϖ = ϖ 1 id W + ϖ c ,
using the decomposition
g ± = C id W ⊕ c ± . \mathfrak{g}_{\pm}=\mathbb{C}\operatorname{id}_{W}\oplus\mathfrak{c}_{\pm}. g ± = C id W ⊕ c ± .
Here ϖ 1 \varpi_{1} ϖ 1 is a complex-valued one-form and ϖ c \varpi_{c} ϖ c is a c ± \mathfrak{c}_{\pm} c ± -valued one-form. We call ϖ \varpi ϖ the defect of D 0 D^{0} D 0 .
Since D ± D^{\pm} D ± preserves N ± N_{\pm} N ± , the covariant derivative of the critical cubic is
D 0 N ± = − 3 ϖ 1 ⊗ N ± − Φ N ± ( ϖ c ) . D^{0}N_{\pm}=-3\varpi_{1}\otimes N_{\pm}-\Phi_{N_{\pm}}(\varpi_{c}). D 0 N ± = − 3 ϖ 1 ⊗ N ± − Φ N ± ( ϖ c ) .
Recall that ker Φ N ± = g ± \ker\Phi_{N_{\pm}}=\mathfrak{g}_{\pm} ker Φ N ± = g ± and Φ N ± ( id W ) = 3 N ± \Phi_{N_{\pm}}(\operatorname{id}_{W})=3N_{\pm} Φ N ± ( id W ) = 3 N ± . Hence Φ N ± \Phi_{N_{\pm}} Φ N ± is injective on g ± \mathfrak{g}_{\pm} g ± , and
Φ N ± ( c ± ) ∩ C N ± = 0. \Phi_{N_{\pm}}(\mathfrak{c}_{\pm})\cap\mathbb{C}N_{\pm}=0. Φ N ± ( c ± ) ∩ C N ± = 0.
It follows that D 0 D^{0} D 0 preserves the line C N ± \mathbb{C}N_{\pm} C N ± if and only if ϖ c = 0 \varpi_{c}=0 ϖ c = 0 , and preserves N ± N_{\pm} N ± itself if and only if ϖ = 0 \varpi=0 ϖ = 0 .
We return to the Albert algebra bundle of a line field and ask when its product is parallel under a compatible connection.
Proposition 5.17 (Parallel pointings). Let D D D preserve g g g and φ \varphi φ , and use the same symbol for its induced connections.
The product ∘ E ℓ \circ_{E_{\ell}} ∘ E ℓ is D D D -parallel if and only if D ℓ ⊂ ℓ D\ell\subset\ell D ℓ ⊂ ℓ . When this holds, D D D also preserves N ± / n ± N_{\pm}/n_{\pm} N ± / n ± , the based trace form, the adjoint, and ς ℓ , ± \varsigma_{\ell,\pm} ς ℓ , ± .
If φ \varphi φ is torsion-free, then D L C E ℓ = 0 D^{LC}E_{\ell}=0 D L C E ℓ = 0 if and only if ℓ \ell ℓ is parallel. On each connected component this forces Hol ( g ) ⊂ H E 1 \operatorname{Hol}(g)\subset H_{E_{1}} Hol ( g ) ⊂ H E 1 up to conjugacy, with H E 1 H_{E_{1}} H E 1 as in Proposition 4.13; conversely, such holonomy gives a parallel E ℓ E_{\ell} E ℓ .
If ℓ \ell ℓ is oriented, then D ℓ ⊂ ℓ D\ell\subset\ell D ℓ ⊂ ℓ if and only if its globally defined unit section u u u is D D D -parallel. In this case,
Hol ( D ) ⊂ Stab G 2 ( u ) ≅ S U ( 3 ) . \operatorname{Hol}(D)\subset\operatorname{Stab}_{G_{2}}(u)\cong SU(3). Hol ( D ) ⊂ Stab G 2 ( u ) ≅ S U ( 3 ) . In particular, for torsion-free φ \varphi φ and D = D L C D=D^{LC} D = D L C , an oriented parallel line field forces Hol ( g ) ⊂ S U ( 3 ) \operatorname{Hol}(g)\subset SU(3) Hol ( g ) ⊂ S U ( 3 ) up to conjugacy. Conversely, such holonomy gives an oriented parallel line field.
Proof. If D D D preserves ℓ \ell ℓ , it preserves P ℓ P_{\ell} P ℓ and every tensor contraction defining the operations, which proves one implication of (1) and the last sentence of (1). Conversely, suppose that D D D preserves the product. Differentiating E ℓ ∘ E ℓ h = h E_{\ell}\circ_{E_{\ell}}h=h E ℓ ∘ E ℓ h = h gives
( D X E ℓ ) ∘ E ℓ h + E ℓ ∘ E ℓ D X h = D X h . (D_{X}E_{\ell})\circ_{E_{\ell}}h+E_{\ell}\circ_{E_{\ell}}D_{X}h=D_{X}h. ( D X E ℓ ) ∘ E ℓ h + E ℓ ∘ E ℓ D X h = D X h .
The second term equals D X h D_{X}h D X h , so ( D X E ℓ ) ∘ E ℓ h = 0 (D_{X}E_{\ell})\circ_{E_{\ell}}h=0 ( D X E ℓ ) ∘ E ℓ h = 0 . Taking h = E ℓ h=E_{\ell} h = E ℓ and using the unit identity gives D X E ℓ = 0 D_{X}E_{\ell}=0 D X E ℓ = 0 . Since D g = 0 Dg=0 D g = 0 and P ℓ = ( E ℓ + g ) / 7 P_{\ell}=(E_{\ell}+g)/7 P ℓ = ( E ℓ + g ) /7 , it follows that D P ℓ = 0 DP_{\ell}=0 D P ℓ = 0 , equivalently D ℓ ⊂ ℓ D\ell\subset\ell D ℓ ⊂ ℓ .
In the torsion-free case D L C D^{LC} D L C preserves φ \varphi φ . By the holonomy principle, parallel lines correspond to lines preserved by holonomy, and the stabilizer in G 2 G_{2} G 2 of such a line is H E 1 H_{E_{1}} H E 1 . This proves the first assertion of (2), including its converse.
For the oriented assertion in (2), orientation gives a global unit section u u u of ℓ \ell ℓ . If D D D preserves ℓ \ell ℓ , then D X u = a ( X ) u D_{X}u=a(X)u D X u = a ( X ) u for some one-form a a a . Metric compatibility implies
0 = X ( g ( u , u ) ) = 2 g ( D X u , u ) = 2 a ( X ) , 0=X(g(u,u))=2g(D_{X}u,u)=2a(X), 0 = X ( g ( u , u )) = 2 g ( D X u , u ) = 2 a ( X ) ,
so D u = 0 Du=0 D u = 0 . The converse is immediate. Since the stabilizer of a unit vector in G 2 G_{2} G 2 is S U ( 3 ) SU(3) S U ( 3 ) , the holonomy inclusion follows. Conversely, if Hol ( g ) \operatorname{Hol}(g) Hol ( g ) lies in an S U ( 3 ) SU(3) S U ( 3 ) subgroup of G 2 G_{2} G 2 , it fixes a unit vector, which extends by parallel transport to a global parallel unit vector field.
The holonomy principle and the absence of fixed elements in the nontrivial irreducible 27 immediately give the following.
Corollary 5.18 (Full restricted holonomy). If Hol 0 ( g ) = G 2 \operatorname{Hol}^{0}(g)=G_{2} Hol 0 ( g ) = G 2 up to conjugacy, then W φ , C \mathcal{W}_{\varphi,\mathbb{C}} W φ , C has no nonzero D L C D^{LC} D L C -parallel section.
The case Hol ( g ) = H E 1 ≅ S U ( 3 ) ⋊ Z / 2 \operatorname{Hol}(g)=H_{E_{1}}\cong SU(3)\rtimes\mathbb{Z}/2 Hol ( g ) = H E 1 ≅ S U ( 3 ) ⋊ Z /2 of Proposition 5.17 (2), with disconnected holonomy, occurs in the compact barely G 2 G_{2} G 2 -manifolds constructed as quotients of Y × S 1 \mathcal{Y}\times S^{1} Y × S 1 by ( y , t ) ↦ ( σ ( y ) , − t ) (y,t)\mapsto(\sigma(y),-t) ( y , t ) ↦ ( σ ( y ) , − t ) , where Y \mathcal{Y} Y is a Calabi–Yau threefold and σ \sigma σ is a free antiholomorphic isometric involution. Their holonomy is stated to be S U ( 3 ) ⋊ Z / 2 SU(3)\rtimes\mathbb{Z}/2 S U ( 3 ) ⋊ Z /2 in [26 , 37 ] . The parallel-pointing criterion of Proposition 5.17 can be seen directly: the parallel line spanned by ∂ t \partial_{t} ∂ t on the product descends to the quotient, since the involution sends ∂ t \partial_{t} ∂ t to − ∂ t -\partial_{t} − ∂ t . This gives an unoriented parallel line, explaining the disconnected stabilizer rather than S U ( 3 ) SU(3) S U ( 3 ) .
These results give a geometric realization of the exceptional algebraic structures suggested by Lee and Leung [46 ] . The E 6 E_{6} E 6 -invariant cubics themselves are distinguished complex linear combinations of the two geometrically relevant real cubics C 0 C_{0} C 0 and C φ C_{\varphi} C φ . On any 7-manifold with a G 2 G_{2} G 2 -structure, these critical cubics define canonical conjugate E 6 ( C ) E_{6}(\mathbb{C}) E 6 ( C ) -reductions of the rank-27 deformation bundle. A tangent line field supplies the additional choice of a unit, yielding Albert algebra bundles and F 4 ( C ) F_{4}(\mathbb{C}) F 4 ( C ) -reductions. The reduced-holonomy setting considered by Lee and Leung appears when this choice is parallel.
The differential geometry of these exceptional structures is governed by the intrinsic torsion of φ \varphi φ . The Levi-Civita derivative of either critical cubic determines the full intrinsic torsion, while orthogonal projection onto either exceptional reduction gives the connection induced by the minimal G 2 G_{2} G 2 -connection. Thus the exceptional cubic structures are intrinsic to the G 2 G_{2} G 2 -geometry, whereas parallel Albert products require additional holonomy reduction. The constructions provide a starting point for investigating the curvature and differential invariants of exceptional connections, and their relationship with global G 2 G_{2} G 2 deformation theory.
Proofs of the contraction and quartic identities
This appendix contains the contraction derivations deferred from § 4 . In Section A.1, we derive the one-index c-contraction identities (3.10) and (3.11) and the two-index constants (3.9). Section A.2 uses these identities to evaluate the Gram matrix and normal-operator scalars for stabilizer exclusion. In Section A.3, we prove Lemma 3.5 . Section A.4 collects the notation used across the paper. These derivations use the conventions and G 2 G_{2} G 2 -contraction identities of § 3.1 , Lemma 2.9 , and (3.7). We retain the up/down contraction convention of § 1 . Metric-dual basis indices are written k α k^{\alpha} k α , Θ i \Theta^{i} Θ i and L i L^{i} L i ; in an adapted orthonormal frame their numerical entries agree with those of k α k_{\alpha} k α , Θ i \Theta_{i} Θ i and L i L_{i} L i , respectively.
The c-contraction identities
We use the cubic components C α β γ s C^{s}_{\alpha\beta\gamma} C α β γ s and the ordered one-index c-contractions T s t \mathsf{T}^{st} T s t of § 4.2 . Greek indices refer to W W W and Latin indices to V V V . We first evaluate the one-index c-contractions, then sum them to obtain the two-index constants. To evaluate the one-index contractions without choosing coordinates on W W W , we write the pairings T s t ( h 1 , h 2 ; h 3 , h 4 ) T^{st}(h_1,h_2;h_3,h_4) T s t ( h 1 , h 2 ; h 3 , h 4 ) of Section 3.2 as
T s t ( h 1 , h 2 ; h 3 , h 4 ) = ∑ σ C s ( h 1 , h 2 , k σ ) C t ( h 3 , h 4 , k σ ) . T^{st}(h_1,h_2;h_3,h_4)=\sum_{\sigma} C_s(h_1,h_2,k_\sigma)C_t(h_3,h_4,k^\sigma). T s t ( h 1 , h 2 ; h 3 , h 4 ) = σ ∑ C s ( h 1 , h 2 , k σ ) C t ( h 3 , h 4 , k σ ) .
This sum pairs the corresponding traceless products A A A and B B B , so T α β γ δ s t = T s t ( k α , k β ; k γ , k δ ) T^{st}_{\alpha\beta\gamma\delta}=T^{st}(k_\alpha,k_\beta;k_\gamma,k_\delta) T α β γ δ s t = T s t ( k α , k β ; k γ , k δ ) . The projection identity (67) gives, for arbitrary symmetric matrices h 1 , h 2 h_1,h_2 h 1 , h 2 ,
∑ σ tr ( h 1 k σ ) tr ( h 2 k σ ) = tr ( h 1 h 2 ) − 1 7 tr ( h 1 ) tr ( h 2 ) . \sum_{\sigma}\operatorname{tr}(h_1k_\sigma)\operatorname{tr}(h_2k^\sigma)=\operatorname{tr}(h_1h_2)-\frac{1}{7}\operatorname{tr}(h_1)\operatorname{tr}(h_2). σ ∑ tr ( h 1 k σ ) tr ( h 2 k σ ) = tr ( h 1 h 2 ) − 7 1 tr ( h 1 ) tr ( h 2 ) .
For traceless inputs, the trace corrections are determined by tr 1 2 ( h 1 h 2 + h 2 h 1 ) = ⟨ h 1 , h 2 ⟩ W \operatorname{tr}\frac{1}{2}(h_1h_2+h_2h_1)=\langle h_1,h_2\rangle_W tr 2 1 ( h 1 h 2 + h 2 h 1 ) = ⟨ h 1 , h 2 ⟩ W and tr P ( h 1 , h 2 ) = − ⟨ h 1 , h 2 ⟩ W \operatorname{tr}P(h_1,h_2)=-\langle h_1,h_2\rangle_W tr P ( h 1 , h 2 ) = − ⟨ h 1 , h 2 ⟩ W . Substituting these into the projection identity evaluates the first three one-index c c c -contractions and yields (70) . Exchanging the two cubic types while swapping the two index pairs gives the symmetry
T s t ( h 1 , h 2 ; h 3 , h 4 ) = T t s ( h 3 , h 4 ; h 1 , h 2 ) . T^{st}(h_1,h_2;h_3,h_4)=T^{ts}(h_3,h_4;h_1,h_2). T s t ( h 1 , h 2 ; h 3 , h 4 ) = T t s ( h 3 , h 4 ; h 1 , h 2 ) .
For the remaining one-index c c c -contraction, two applications of (10) give
T φ φ ( h 1 , h 2 ; h 3 , h 4 ) = ( δ a 1 a 3 δ a 2 a 4 − δ a 1 a 4 δ a 2 a 3 + ψ a 1 a 2 a 3 a 4 ) ( δ b 1 b 3 δ b 2 b 4 − δ b 1 b 4 δ b 2 b 3 + ψ b 1 b 2 b 3 b 4 ) ⋅ h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 h 4 a 4 b 4 − 1 7 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W . (158) \begin{aligned}
T^{\varphi\varphi}(h_1,h_2;h_3,h_4)={}&(\delta_{a_1a_3}\delta_{a_2a_4}-\delta_{a_1a_4}\delta_{a_2a_3}+\psi_{a_1a_2a_3a_4})\\
&(\delta_{b_1b_3}\delta_{b_2b_4}-\delta_{b_1b_4}\delta_{b_2b_3}+\psi_{b_1b_2b_3b_4})\\
&\cdot h_1^{a_1b_1}h_2^{a_2b_2}h_3^{a_3b_3}h_4^{a_4b_4}-\frac{1}{7}\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W. \tag*{(158)}
\end{aligned} T φφ ( h 1 , h 2 ; h 3 , h 4 ) = ( δ a 1 a 3 δ a 2 a 4 − δ a 1 a 4 δ a 2 a 3 + ψ a 1 a 2 a 3 a 4 ) ( δ b 1 b 3 δ b 2 b 4 − δ b 1 b 4 δ b 2 b 3 + ψ b 1 b 2 b 3 b 4 ) ⋅ h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 h 4 a 4 b 4 − 7 1 ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W . (158)
To evaluate the 4-form product in this expression, we first derive a local triple-form contraction and take its trace. The same calculation will also be used in the proof of Lemma 3.5, and it does not use any of the later stabilizer arguments.
For h ∈ W h\in W h ∈ W , put Y = P ( h , h ) Y=P(h,h) Y = P ( h , h ) , r = tr ( h 2 ) r=\operatorname{tr}(h^2) r = tr ( h 2 ) and γ φ = C φ ( h ) \gamma_\varphi=C_\varphi(h) γ φ = C φ ( h ) , and write { X , Y } = X Y + Y X \{X,Y\}=XY+YX { X , Y } = X Y + Y X for the anticommutator.
Lemma A.1 (Triple-form reduction). For every h ∈ W h\in W h ∈ W , we have
P ( Y , Y ) = 4 γ φ h + 4 h Y h + 4 r Y − 8 P ( h , h 3 ) − 8 { h 2 , Y } , (159) P(Y,Y)=4\gamma_\varphi h+4hYh+4rY-8P(h,h^3)-8\{h^2,Y\}, \tag*{(159)} P ( Y , Y ) = 4 γ φ h + 4 hY h + 4 r Y − 8 P ( h , h 3 ) − 8 { h 2 , Y } , (159) tr ( Y 2 ) = 5 r 2 − 8 tr ( h 4 ) + 12 C φ ( h , h , h 2 ) . (160) \operatorname{tr}(Y^2)=5r^2-8\operatorname{tr}(h^4)+12C_\varphi(h,h,h^2). \tag*{(160)} tr ( Y 2 ) = 5 r 2 − 8 tr ( h 4 ) + 12 C φ ( h , h , h 2 ) . (160)
Proof. The reduction below uses the following one index φ ψ \varphi\psi φ ψ G 2 G_2 G 2 -contraction identity:
φ i j m ψ a b c m = δ i a φ j b c + δ i b φ j c a + δ i c φ j a b − δ j a φ i b c − δ j b φ i c a − δ j c φ i a b . (161) \varphi_{ijm}\psi^{m}_{\ abc}=\delta_{ia}\varphi_{jbc}+\delta_{ib}\varphi_{jca}+\delta_{ic}\varphi_{jab}
-\delta_{ja}\varphi_{ibc}-\delta_{jb}\varphi_{ica}-\delta_{jc}\varphi_{iab}. \tag*{(161)} φ ij m ψ ab c m = δ ia φ j b c + δ ib φ j c a + δ i c φ j ab − δ j a φ ib c − δ j b φ i c a − δ j c φ iab . (161)
In Bryant’s notation, after relabeling, this is [3 (2.9)] .
We reduce T i ; a b c d = φ i p q φ p a b φ q c d T_{i;abcd}=\varphi_i{}^{pq}\varphi_{pab}\varphi_{qcd} T i ; ab c d = φ i pq φ p ab φ q c d in two steps: contract its first two factors using (10) , then contract the resulting 4-form using (A.4). This gives
T i ; a b c d = − δ i a φ b c d + δ i b φ a c d + δ i c φ a b d − δ i d φ a b c − δ a c φ i b d + δ b c φ i a d + δ a d φ i b c − δ b d φ i a c . (162) T_{i;abcd}=-\delta_{ia}\varphi_{bcd}+\delta_{ib}\varphi_{acd}+\delta_{ic}\varphi_{abd}-\delta_{id}\varphi_{abc}
-\delta_{ac}\varphi_{ibd}+\delta_{bc}\varphi_{iad}+\delta_{ad}\varphi_{ibc}-\delta_{bd}\varphi_{iac}. \tag*{(162)} T i ; ab c d = − δ ia φ b c d + δ ib φ a c d + δ i c φ ab d − δ i d φ ab c − δ a c φ ib d + δ b c φ ia d + δ a d φ ib c − δ b d φ ia c . (162)
In particular, the first contraction is φ i q φ p a b = − δ a i δ q b + δ b i δ q a − ψ i q a b \varphi^i{}_q\varphi_{pab}=-\delta^i_a\delta_{qb}+\delta^i_b\delta_{qa}-\psi_{iqab} φ i q φ p ab = − δ a i δ q b + δ b i δ q a − ψ i q ab . Denote the four signed summands in the first line of (A.5) by T 1 ( 1 ) , … , T 4 ( 1 ) \mathcal{T}^{(1)}_1,\ldots,\mathcal{T}^{(1)}_4 T 1 ( 1 ) , … , T 4 ( 1 ) and those in its second line by T 1 ( 2 ) , … , T 4 ( 2 ) \mathcal{T}^{(2)}_1,\ldots,\mathcal{T}^{(2)}_4 T 1 ( 2 ) , … , T 4 ( 2 ) . Pair two such summands by
[ T u , i ( p ) , T v , j ( q ) ] h = T u , i ; a b c d ( p ) T v , j ; a ′ b ′ c ′ d ′ ( q ) h a a ′ h b b ′ h c c ′ h d d ′ . [\mathcal{T}^{(p)}_{u,i},\mathcal{T}^{(q)}_{v,j}]_h=\mathcal{T}^{(p)}_{u,i;abcd}\mathcal{T}^{(q)}_{v,j;a'b'c'd'}h^{aa'}h^{bb'}h^{cc'}h^{dd'}. [ T u , i ( p ) , T v , j ( q ) ] h = T u , i ; ab c d ( p ) T v , j ; a ′ b ′ c ′ d ′ ( q ) h a a ′ h b b ′ h c c ′ h d d ′ .
Here i , j i,j i , j are the free indices and p , q = 1 , 2 p,q=1,2 p , q = 1 , 2 ; the subscripts 1 , … , 4 1,\ldots,4 1 , … , 4 below enumerate summands. To display the pairing arrays, abbreviate
C = γ φ h , S = h Y h , R = h 2 Y , D = r Y , W = P ( h , h 3 ) . C=\gamma_\varphi h,\qquad S=hYh,\qquad R=h^2Y,\qquad D=rY,\qquad W=\mathcal{P}(h,h^3). C = γ φ h , S = hY h , R = h 2 Y , D = r Y , W = P ( h , h 3 ) .
We pair the summands within each group and between the two groups. Substituting the Kronecker deltas gives the following three arrays, with rows and columns in the order of the summands above:
( [ T u ( 1 ) , T v ( 1 ) ] h ) u , v = ( C − S S S − S C S S S S C − S S S − S C ) , ( [ T u ( 1 ) , T v ( 2 ) ] h ) u , v = − ( R 0 R 0 0 R 0 R R R 0 0 0 0 R R ) , ( [ T u ( 2 ) , T v ( 2 ) ] h ) u , v = ( D − W − W 0 − W D 0 − W − W 0 D − W 0 − W − W D ) . (163) \left([\mathcal{T}^{(1)}_u,\mathcal{T}^{(1)}_v]_h\right)_{u,v}=
\begin{pmatrix}
C & -S & S & S\\
-S & C & S & S\\
S & S & C & -S\\
S & S & -S & C
\end{pmatrix},
\qquad
\left([\mathcal{T}^{(1)}_u,\mathcal{T}^{(2)}_v]_h\right)_{u,v}=
-\begin{pmatrix}
R & 0 & R & 0\\
0 & R & 0 & R\\
R & R & 0 & 0\\
0 & 0 & R & R
\end{pmatrix},
\qquad
\left([\mathcal{T}^{(2)}_u,\mathcal{T}^{(2)}_v]_h\right)_{u,v}=
\begin{pmatrix}
D & -W & -W & 0\\
-W & D & 0 & -W\\
-W & 0 & D & -W\\
0 & -W & -W & D
\end{pmatrix}. \tag*{(163)} ( [ T u ( 1 ) , T v ( 1 ) ] h ) u , v = C − S S S − S C S S S S C − S S S − S C , ( [ T u ( 1 ) , T v ( 2 ) ] h ) u , v = − R 0 R 0 0 R R 0 R 0 0 R 0 R 0 R , ( [ T u ( 2 ) , T v ( 2 ) ] h ) u , v = D − W − W 0 − W D 0 − W − W 0 D − W 0 − W − W D . (163)
The reverse mixed array, pairing group (2) with group (1), is obtained from the displayed (1)–(2) array by transposing both its 4 × 4 4\times4 4 × 4 matrix and the free indices i , j i,j i , j in each entry. Representative entries account for every tensor type and every sign:
[ T 1 , i ( 1 ) , T 1 , j ( 1 ) ] h = h i j φ b c d φ b ′ c ′ d ′ h b b ′ h c c ′ h d d ′ = C i j , [\mathcal{T}^{(1)}_{1,i},\mathcal{T}^{(1)}_{1,j}]_h=h_{ij}\varphi_{bcd}\varphi_{b'c'd'}h^{bb'}h^{cc'}h^{dd'}=C_{ij}, [ T 1 , i ( 1 ) , T 1 , j ( 1 ) ] h = h ij φ b c d φ b ′ c ′ d ′ h b b ′ h c c ′ h d d ′ = C ij ,
[ T 1 , i ( 1 ) , T 2 , j ( 2 ) ] h = − h i a ′ h j b φ b c d φ a ′ c ′ d ′ h c c ′ h d d ′ = − S i j , [\mathcal{T}^{(1)}_{1,i},\mathcal{T}^{(2)}_{2,j}]_h=-h_i{}^{a'}h_j{}^b\varphi_{bcd}\varphi_{a'c'd'}h^{cc'}h^{dd'}=-S_{ij}, [ T 1 , i ( 1 ) , T 2 , j ( 2 ) ] h = − h i a ′ h j b φ b c d φ a ′ c ′ d ′ h c c ′ h d d ′ = − S ij ,
[ T 1 , i ( 1 ) , T 1 , j ( 2 ) ] h = ( h 2 ) i c φ b c d φ j b ′ d ′ h b b ′ h d d ′ = − R i j , [\mathcal{T}^{(1)}_{1,i},\mathcal{T}^{(2)}_{1,j}]_h=(h^2)^i{}_c\varphi_{bcd}\varphi_{jb'd'}h^{bb'}h^{dd'}=-R_{ij}, [ T 1 , i ( 1 ) , T 1 , j ( 2 ) ] h = ( h 2 ) i c φ b c d φ j b ′ d ′ h b b ′ h d d ′ = − R ij ,
[ T 1 , i ( 1 ) , T 2 , j ( 2 ) ] h = − h i a ′ ( h 2 ) b c φ b c d φ j a ′ d ′ h d d ′ = 0 , [\mathcal{T}^{(1)}_{1,i},\mathcal{T}^{(2)}_{2,j}]_h=-h_i{}^{a'}(h^2)^{bc}\varphi_{bcd}\varphi_{ja'd'}h^{dd'}=0, [ T 1 , i ( 1 ) , T 2 , j ( 2 ) ] h = − h i a ′ ( h 2 ) b c φ b c d φ j a ′ d ′ h d d ′ = 0 ,
[ T 1 , i ( 2 ) , T 1 , j ( 2 ) ] h = r φ i b d φ j b ′ d ′ h b b ′ h d d ′ = D i j , [\mathcal{T}^{(2)}_{1,i},\mathcal{T}^{(2)}_{1,j}]_h=r\varphi_{ibd}\varphi_{jb'd'}h^{bb'}h^{dd'}=D_{ij}, [ T 1 , i ( 2 ) , T 1 , j ( 2 ) ] h = r φ ib d φ j b ′ d ′ h b b ′ h d d ′ = D ij ,
[ T 1 , i ( 2 ) , T 2 , j ( 2 ) ] h = − φ i b d φ j a ′ d ′ ( h 3 ) b a ′ h d d ′ = − W i j . [\mathcal{T}^{(2)}_{1,i},\mathcal{T}^{(2)}_{2,j}]_h=-\varphi_{ibd}\varphi_{ja'd'}(h^3)^{ba'}h^{dd'}=-W_{ij}. [ T 1 , i ( 2 ) , T 2 , j ( 2 ) ] h = − φ ib d φ j a ′ d ′ ( h 3 ) b a ′ h d d ′ = − W ij .
In the fourth entry, the symmetric tensor ( h 2 ) b c (h^2)^{bc} ( h 2 ) b c is contracted against alternating slots, so the entry vanishes. The other entries follow by permuting the four paired dummy slots. To recover the full contraction, we sum the entries of each array: the first, mixed and last give 4 C + 4 S 4C+4S 4 C + 4 S , − 8 R -8R − 8 R and 4 D − 8 W 4D-8W 4 D − 8 W , respectively. The reverse mixed array contributes − 8 R ⊤ -8R^\top − 8 R ⊤ . Since [ T i , T j ] h = P ( Y , Y ) i j [\mathcal{T}_i,\mathcal{T}_j]_h=\mathcal{P}(Y,Y)_{ij} [ T i , T j ] h = P ( Y , Y ) ij , the sum is (A.2). Taking its trace, using (2.34c) and tr Y = − r \operatorname{tr}Y=-r tr Y = − r , gives (A.3).
The derivation of (A.2) and (A.3) does not use the quartic identities of Lemma 3.5. □
Proposition A.2. The one-index c c c -contractions satisfy (3.10) and (3.11).
Proof. The first three contractions were evaluated above. For the remaining contraction, we use Lemma A.1 . We use the alternating vector η ( h 1 , h 2 ) \eta(h_1,h_2) η ( h 1 , h 2 ) of § 4.2 , extended to arbitrary symmetric inputs by the same formula. Extend C φ C_\varphi C φ to arbitrary symmetric inputs by the same ordered contraction as in its definition, and set
C ( h 1 , h 2 , h 3 , h 4 ) = C φ ( h 1 , h 2 , 1 2 ( h 3 h 4 + h 4 h 3 ) ) + C φ ( h 1 , h 3 , 1 2 ( h 2 h 4 + h 4 h 2 ) ) + C φ ( h 1 , h 4 , 1 2 ( h 2 h 3 + h 3 h 2 ) ) + C φ ( h 2 , h 3 , 1 2 ( h 1 h 4 + h 4 h 1 ) ) + C φ ( h 2 , h 4 , 1 2 ( h 1 h 3 + h 3 h 1 ) ) + C φ ( h 3 , h 4 , 1 2 ( h 1 h 2 + h 2 h 1 ) ) . \begin{aligned}
C(h_1,h_2,h_3,h_4)={}&C_\varphi\left(h_1,h_2,\frac{1}{2}(h_3h_4+h_4h_3)\right)+C_\varphi\left(h_1,h_3,\frac{1}{2}(h_2h_4+h_4h_2)\right)\\
&+C_\varphi\left(h_1,h_4,\frac{1}{2}(h_2h_3+h_3h_2)\right)+C_\varphi\left(h_2,h_3,\frac{1}{2}(h_1h_4+h_4h_1)\right)\\
&+C_\varphi\left(h_2,h_4,\frac{1}{2}(h_1h_3+h_3h_1)\right)+C_\varphi\left(h_3,h_4,\frac{1}{2}(h_1h_2+h_2h_1)\right).
\end{aligned} C ( h 1 , h 2 , h 3 , h 4 ) = C φ ( h 1 , h 2 , 2 1 ( h 3 h 4 + h 4 h 3 ) ) + C φ ( h 1 , h 3 , 2 1 ( h 2 h 4 + h 4 h 2 ) ) + C φ ( h 1 , h 4 , 2 1 ( h 2 h 3 + h 3 h 2 ) ) + C φ ( h 2 , h 3 , 2 1 ( h 1 h 4 + h 4 h 1 ) ) + C φ ( h 2 , h 4 , 2 1 ( h 1 h 3 + h 3 h 1 ) ) + C φ ( h 3 , h 4 , 2 1 ( h 1 h 2 + h 2 h 1 ) ) .
It differs from the sum Ξ \Xi Ξ of § 4.2 by trace terms. We derive (3.11) by expanding (A.1) into the delta, mixed and 4-form contributions. Write each parenthesis in (A.1) as D + ψ D+\psi D + ψ , with D a b e f = δ a e δ b f − δ a f δ b e D_{abef}=\delta_{ae}\delta_{bf}-\delta_{af}\delta_{be} D ab e f = δ a e δ b f − δ a f δ b e , and abbreviate
p 0 = ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , p 1 = ⟨ h 1 , h 3 ⟩ W ⟨ h 2 , h 4 ⟩ W , p 2 = ⟨ h 1 , h 4 ⟩ W ⟨ h 2 , h 3 ⟩ W , p_0=\langle h_1,h_2\rangle_W\langle h_3,h_4\rangle_W,\quad p_1=\langle h_1,h_3\rangle_W\langle h_2,h_4\rangle_W,\quad p_2=\langle h_1,h_4\rangle_W\langle h_2,h_3\rangle_W, p 0 = ⟨ h 1 , h 2 ⟩ W ⟨ h 3 , h 4 ⟩ W , p 1 = ⟨ h 1 , h 3 ⟩ W ⟨ h 2 , h 4 ⟩ W , p 2 = ⟨ h 1 , h 4 ⟩ W ⟨ h 2 , h 3 ⟩ W ,
t 1 = tr ( h 1 h 2 h 3 h 4 ) , t 2 = tr ( h 1 h 2 h 4 h 3 ) , t 3 = tr ( h 1 h 3 h 2 h 4 ) , t_1=\operatorname{tr}(h_1h_2h_3h_4),\quad t_2=\operatorname{tr}(h_1h_2h_4h_3),\quad t_3=\operatorname{tr}(h_1h_3h_2h_4), t 1 = tr ( h 1 h 2 h 3 h 4 ) , t 2 = tr ( h 1 h 2 h 4 h 3 ) , t 3 = tr ( h 1 h 3 h 2 h 4 ) ,
e = η ( h 1 , h 3 ) ⋅ η ( h 2 , h 4 ) + η ( h 1 , h 4 ) ⋅ η ( h 2 , h 3 ) . e=\eta(h_1,h_3)\cdot\eta(h_2,h_4)+\eta(h_1,h_4)\cdot\eta(h_2,h_3). e = η ( h 1 , h 3 ) ⋅ η ( h 2 , h 4 ) + η ( h 1 , h 4 ) ⋅ η ( h 2 , h 3 ) .
The two delta factors contribute p 1 + p 2 − 2 t 3 p_1+p_2-2t_3 p 1 + p 2 − 2 t 3 . For the mixed factors, we rearrange (2.7) as
ψ a b e f = δ a b δ e f − δ a f δ e b − φ a e i φ i b f . \psi_{abef}=\delta_{ab}\delta_{ef}-\delta_{af}\delta_{eb}-\varphi^{i}_{ae}\varphi_{ibf}. ψ ab e f = δ ab δ e f − δ a f δ e b − φ a e i φ ib f .
Write X ⌟ φ X\mathbin{\lrcorner}\varphi X ┘ φ for the vector with components ( X ⌟ φ ) k = X i j φ i j k (X\mathbin{\lrcorner}\varphi)_k=X^{ij}\varphi_{ijk} ( X ┘ φ ) k = X ij φ ij k . Thus for arbitrary matrices X , Y X,Y X , Y ,
ψ a b e f X a e Y b f = tr ( X T Y ) − tr ( X Y ) − ( X ⌟ φ ) ⋅ ( Y ⌟ φ ) . \psi_{abef}X^{ae}Y^{bf}=\operatorname{tr}(X^{\mathsf T}Y)-\operatorname{tr}(XY)-(X\mathbin{\lrcorner}\varphi)\cdot(Y\mathbin{\lrcorner}\varphi). ψ ab e f X a e Y b f = tr ( X T Y ) − tr ( X Y ) − ( X ┘ φ ) ⋅ ( Y ┘ φ ) .
We apply this identity to ( X , Y ) = ( h 1 h 3 , h 2 h 4 ) (X,Y)=(h_1h_3,h_2h_4) ( X , Y ) = ( h 1 h 3 , h 2 h 4 ) and ( h 1 h 4 , h 2 h 3 ) (h_1h_4,h_2h_3) ( h 1 h 4 , h 2 h 3 ) , then collect the traces using cyclicity and reversal; by the definition of η \eta η , ( h 1 h 3 ) ⌟ φ = η ( h 1 , h 3 ) (h_1h_3)\mathbin{\lrcorner}\varphi=\eta(h_1,h_3) ( h 1 h 3 ) ┘ φ = η ( h 1 , h 3 ) , and similarly for the other three products. The sum of the two mixed contributions is 2 t 1 + 2 t 2 − 4 t 3 − 2 e 2t_1+2t_2-4t_3-2e 2 t 1 + 2 t 2 − 4 t 3 − 2 e .
For the product of the 4-forms, contracting the two φ \varphi φ factors sharing the middle index of Y 2 Y^2 Y 2 gives
Y 2 − 2 P ( h 2 2 , h 2 2 ) = ( φ i a 1 a 2 φ j a 3 a 4 ψ b 1 b 2 b 3 b 4 h a 1 b 1 h a 2 b 2 h a 3 b 3 h a 4 b 4 ) i j . Y^2-2P(h_2^2,h_2^2)=\left(\varphi_{ia_1a_2}\varphi_{ja_3a_4}\psi_{b_1b_2b_3b_4}h^{a_1b_1}h^{a_2b_2}h^{a_3b_3}h^{a_4b_4}\right)_{ij}. Y 2 − 2 P ( h 2 2 , h 2 2 ) = ( φ i a 1 a 2 φ j a 3 a 4 ψ b 1 b 2 b 3 b 4 h a 1 b 1 h a 2 b 2 h a 3 b 3 h a 4 b 4 ) ij .
Taking the trace gives the left-hand side of the scalar identity below, because the two metric terms in the i , j i,j i , j contraction vanish against ψ \psi ψ . We can therefore use the trace calculation (A.3) to obtain the scalar identity
ψ a 1 a 2 a 3 a 4 ψ b 1 b 2 b 3 b 4 h a 1 b 1 h a 2 b 2 h a 3 b 3 h a 4 b 4 = 3 r 2 − 6 tr ( h 4 ) + 12 C φ ( h , h , h 2 ) . \psi_{a_1a_2a_3a_4}\psi_{b_1b_2b_3b_4}h^{a_1b_1}h^{a_2b_2}h^{a_3b_3}h^{a_4b_4}=3r^2-6\operatorname{tr}(h^4)+12C_\varphi(h,h,h^2). ψ a 1 a 2 a 3 a 4 ψ b 1 b 2 b 3 b 4 h a 1 b 1 h a 2 b 2 h a 3 b 3 h a 4 b 4 = 3 r 2 − 6 tr ( h 4 ) + 12 C φ ( h , h , h 2 ) .
We recover the four-input expression by normalized polarization. The contraction of two copies of ψ \psi ψ with four matrix inputs is fully symmetric in those inputs, because interchanging two inputs interchanges two slots in each alternating form. Polarizing the scalar identity in the ordered inputs h 1 , h 2 , h 3 , h 4 h_1,h_2,h_3,h_4 h 1 , h 2 , h 3 , h 4 therefore gives the explicit contraction
ψ a 1 a 2 a 3 a 4 ψ b 1 b 2 b 3 b 4 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 h 4 a 4 b 4 = p 0 + p 1 + p 2 − 2 ( t 1 + t 2 + t 3 ) + 2 C . \psi_{a_1a_2a_3a_4}\psi_{b_1b_2b_3b_4}h_1^{a_1b_1}h_2^{a_2b_2}h_3^{a_3b_3}h_4^{a_4b_4}=p_0+p_1+p_2-2(t_1+t_2+t_3)+2C. ψ a 1 a 2 a 3 a 4 ψ b 1 b 2 b 3 b 4 h 1 a 1 b 1 h 2 a 2 b 2 h 3 a 3 b 3 h 4 a 4 b 4 = p 0 + p 1 + p 2 − 2 ( t 1 + t 2 + t 3 ) + 2 C .
The trace term splits into three classes under cyclicity and reversal. For the last term, there are six choices of the two linear inputs, with the Jordan product taken in the other pair. We now add the delta, mixed and 4-form contributions, and subtract the trace correction p 0 / 7 p_0/7 p 0 /7 :
T φ φ = ( p 1 + p 2 − 2 t 3 ) + ( 2 t 1 + 2 t 2 − 4 t 3 − 2 e ) + p 0 + p 1 + p 2 − 2 ( t 1 + t 2 + t 3 ) + 2 C − 1 7 p 0 . (164) T^{\varphi\varphi}=(p_1+p_2-2t_3)+(2t_1+2t_2-4t_3-2e)
+p_0+p_1+p_2-2(t_1+t_2+t_3)+2C-\frac{1}{7}p_0. \tag*{(164)} T φφ = ( p 1 + p 2 − 2 t 3 ) + ( 2 t 1 + 2 t 2 − 4 t 3 − 2 e ) + p 0 + p 1 + p 2 − 2 ( t 1 + t 2 + t 3 ) + 2 C − 7 1 p 0 . (164)
This is the intermediate expression with symmetrized products. If every symmetrized product in C C C is replaced by its traceless part A ( ⋅ , ⋅ ) A(\cdot,\cdot) A ( ⋅ , ⋅ ) , the new sum Ξ \Xi Ξ satisfies C = Ξ − 2 ( p 0 + p 1 + p 2 ) / 7 C=\Xi-2(p_0+p_1+p_2)/7 C = Ξ − 2 ( p 0 + p 1 + p 2 ) /7 , because C φ ( h 1 , h 2 , g ) = − ⟨ h 1 , h 2 ⟩ W C_{\varphi}(h_1,h_2,g)=-\langle h_1,h_2\rangle_W C φ ( h 1 , h 2 , g ) = − ⟨ h 1 , h 2 ⟩ W . The coefficients of p 0 , p 1 , p 2 p_0,p_1,p_2 p 0 , p 1 , p 2 in (A.7) then become 2 / 7 , 10 / 7 , 10 / 7 2/7,10/7,10/7 2/7 , 10/7 , 10/7 , respectively, which gives (3.11). Both versions keep the alternating-vector terms and the order of the inputs. □
The two-index c-contraction constants. To derive the entries of a a a in (3.9), we use the projection sums, a mixed cubic contraction, and the alternating-vector contraction. We evaluate these before summing the one-index c-contraction identities. For h ∈ W h\in W h ∈ W , the projection identity gives
∑ α k α k α = 27 7 g , ∑ α k α h k α = 5 14 h , ∑ α P ( k α , k α ) = − 27 7 g . (165) \sum_{\alpha}k_{\alpha}k^{\alpha}=\frac{27}{7}g,\qquad
\sum_{\alpha}k_{\alpha}hk^{\alpha}=\frac{5}{14}h,\qquad
\sum_{\alpha}\mathcal{P}(k_{\alpha},k^{\alpha})=-\frac{27}{7}g. \tag*{(165)} α ∑ k α k α = 7 27 g , α ∑ k α h k α = 14 5 h , α ∑ P ( k α , k α ) = − 7 27 g . (165)
For the last identity, inserting the projection identity in φ i a b φ j c d ∑ α ( k α ) a c ( k α ) b d \varphi_{iab}\varphi_{jcd}\sum_{\alpha}(k_{\alpha})^{ac}(k^{\alpha})^{bd} φ iab φ j c d ∑ α ( k α ) a c ( k α ) b d gives 0 − 3 δ i j − 6 δ i j / 7 0-3\delta_{ij}-6\delta_{ij}/7 0 − 3 δ ij − 6 δ ij /7 . The mixed cubic contraction needed below is
u ( h 1 , h 2 ) : = ∑ α C φ ( h 1 , k α , 1 2 ( h 2 k α + k α h 2 ) ) = 9 14 ⟨ h 1 , h 2 ⟩ W . u(h_1,h_2):=\sum_{\alpha}C_{\varphi}\left(h_1,k_{\alpha},\frac{1}{2}(h_2k^{\alpha}+k^{\alpha}h_2)\right)=\frac{9}{14}\langle h_1,h_2\rangle_W. u ( h 1 , h 2 ) := α ∑ C φ ( h 1 , k α , 2 1 ( h 2 k α + k α h 2 ) ) = 14 9 ⟨ h 1 , h 2 ⟩ W .
The two terms of the symmetrized product agree by transposition. After inserting the projection identity, their sum reduces to φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 2 a 1 b 1 h 2 a 2 b 2 Π a 2 b 2 b 3 a 2 b 2 \varphi_{a_1a_2a_3}\varphi_{b_1b_2b_3}h_2^{a_1b_1}h_2^{a_2b_2}\Pi^{a_2b_2}_{\phantom{a_2b_2}b_3} φ a 1 a 2 a 3 φ b 1 b 2 b 3 h 2 a 1 b 1 h 2 a 2 b 2 Π a 2 b 2 b 3 a 2 b 2 , where Π \Pi Π denotes the right-hand side of (3.7). The three summands contribute 0 0 0 , ⟨ h 1 , h 2 ⟩ W / 2 \langle h_1,h_2\rangle_W/2 ⟨ h 1 , h 2 ⟩ W /2 and ⟨ h 1 , h 2 ⟩ W / 7 \langle h_1,h_2\rangle_W/7 ⟨ h 1 , h 2 ⟩ W /7 , respectively. To evaluate the latter two, we apply (2.7) to the alternating tensors. The 4-form part vanishes on contraction with a symmetric input, and the pair-trace term vanishes by tracelessness.
For the vector terms of (A.7), put Θ i : = Θ e i \Theta_i:=\Theta_{e_i} Θ i := Θ e i , with ( Θ i ) j k = φ i j k (\Theta_i)_{jk}=\varphi_{ijk} ( Θ i ) j k = φ ij k , as in Section 3.5. The G 2 G_2 G 2 -contraction identities imply ∑ i Θ i Θ i = − 6 g \sum_i\Theta_i\Theta^i=-6g ∑ i Θ i Θ i = − 6 g and ∑ i Θ i h Θ i = h \sum_i\Theta_i h\Theta^i=h ∑ i Θ i h Θ i = h on W W W . Also η ( h 1 , h 2 ) i = ⟨ h 1 , [ Θ i , h 2 ] / 2 ⟩ W \eta(h_1,h_2)_i=\langle h_1,[\Theta_i,h_2]/2\rangle_W η ( h 1 , h 2 ) i = ⟨ h 1 , [ Θ i , h 2 ] /2 ⟩ W . The projection identity and cyclicity of trace therefore give
∑ α η ( h 1 , k α ) ⋅ η ( h 2 , k α ) = 1 4 ∑ i tr ( [ Θ i , h 1 ] [ Θ i , h 2 ] ) = 1 2 ∑ i tr ( Θ i h 1 Θ i h 2 − Θ i Θ i h 1 h 2 ) = 7 2 ⟨ h 1 , h 2 ⟩ W . \begin{aligned}
\sum_{\alpha}\eta(h_1,k_{\alpha})\cdot\eta(h_2,k^{\alpha})
=\frac{1}{4}\sum_i\operatorname{tr}([\Theta_i,h_1][\Theta^i,h_2]) \\
&=\frac{1}{2}\sum_i\operatorname{tr}(\Theta_i h_1\Theta^i h_2-\Theta_i\Theta^i h_1h_2)
=\frac{7}{2}\langle h_1,h_2\rangle_W.
\end{aligned} α ∑ η ( h 1 , k α ) ⋅ η ( h 2 , k α ) = 4 1 i ∑ tr ([ Θ i , h 1 ] [ Θ i , h 2 ]) = 2 1 i ∑ tr ( Θ i h 1 Θ i h 2 − Θ i Θ i h 1 h 2 ) = 2 7 ⟨ h 1 , h 2 ⟩ W .
We can now sum the one-index contractions using a s t ⟨ h 1 , h 2 ⟩ W = ∑ α T s t ( h 1 , k α ; h 2 , k α ) a_{st}\langle h_1,h_2\rangle_W=\sum_{\alpha}T^{st}(h_1,k_{\alpha};h_2,k^{\alpha}) a s t ⟨ h 1 , h 2 ⟩ W = ∑ α T s t ( h 1 , k α ; h 2 , k α ) . For the first two entries, substituting the sums above into (3.10) gives
a 00 = 1 2 ( 5 14 + 27 7 ) − 1 7 = 55 28 , a 0 φ = 9 14 + 1 7 = 11 14 . a_{00}=\frac{1}{2}\left(\frac{5}{14}+\frac{27}{7}\right)-\frac{1}{7}=\frac{55}{28},
\qquad
a_{0\varphi}=\frac{9}{14}+\frac{1}{7}=\frac{11}{14}. a 00 = 2 1 ( 14 5 + 7 27 ) − 7 1 = 28 55 , a 0 φ = 14 9 + 7 1 = 14 11 .
For the remaining entry a φ φ a_{\varphi\varphi} a φφ , we substitute ( h 1 , k α ; h 2 , k α ) (h_1,k_{\alpha};h_2,k^{\alpha}) ( h 1 , k α ; h 2 , k α ) in (A.7) and sum each type of term separately. The three metric terms contribute ( 6 / 7 + 54 + 2 ) ⟨ h 1 , h 2 ⟩ W (6/7+54+2)\langle h_1,h_2\rangle_W ( 6/7 + 54 + 2 ) ⟨ h 1 , h 2 ⟩ W , and the crossed trace contributes − 8 ( 27 / 7 ) ⟨ h 1 , h 2 ⟩ W -8(27/7)\langle h_1,h_2\rangle_W − 8 ( 27/7 ) ⟨ h 1 , h 2 ⟩ W . The six terms of C C C sum to 4 u ( h 1 , h 2 ) − 54 ⟨ h 1 , h 2 ⟩ W / 7 = − 36 ⟨ h 1 , h 2 ⟩ W / 7 4u(h_1,h_2)-54\langle h_1,h_2\rangle_W/7=-36\langle h_1,h_2\rangle_W/7 4 u ( h 1 , h 2 ) − 54 ⟨ h 1 , h 2 ⟩ W /7 = − 36 ⟨ h 1 , h 2 ⟩ W /7 : four are copies of u ( h 1 , h 2 ) u(h_1,h_2) u ( h 1 , h 2 ) , while the other two use ∑ α k α k α = 27 g / 7 \sum_\alpha k_\alpha k^\alpha=27g/7 ∑ α k α k α = 27 g /7 and ∑ α P ( k α , k α ) = − 27 g / 7 \sum_\alpha\mathcal{P}(k_\alpha,k^\alpha)=-27g/7 ∑ α P ( k α , k α ) = − 27 g /7 . Of the two vector terms, one vanishes since η ( k α , k α ) = 0 \eta(k_\alpha,k^\alpha)=0 η ( k α , k α ) = 0 . The other sums to − 7 ⟨ h 1 , h 2 ⟩ W / 2 -7\langle h_1,h_2\rangle_W/2 − 7 ⟨ h 1 , h 2 ⟩ W /2 before multiplication by − 2 -2 − 2 . Combining these contributions gives
a φ φ = ( 6 7 + 54 + 2 ) − 216 7 − 72 7 + 7 = 159 7 . (166) a_{\varphi\varphi}=\left(\frac{6}{7}+54+2\right)-\frac{216}{7}-\frac{72}{7}+7=\frac{159}{7}. \tag*{(166)} a φφ = ( 7 6 + 54 + 2 ) − 7 216 − 7 72 + 7 = 7 159 . (166)
These entries give the matrix and determinant in (69) . All sums use the complete ordered cubic contraction; introducing a factor 1 / 3 ! 1/3! 1/3 ! would change these constants. The projection identity represents the identity operator on W W W , so this derivation and the resulting contraction are independent of the choice of orthonormal basis. These contractions extend complex-bilinearly.
Contractions for stabilizer exclusion
We prove the matrix evaluations of Lemmas 3.11 and 3.12 . We use the action maps Φ s \Phi_s Φ s , their algebraic transposes and pairings from (3.21)–(3.23), the multiplication operators m h ( s ) m_h^{(s)} m h ( s ) of (86) , and the operators Z s t Z_{st} Z s t of (88) . The coefficients a s t a_{st} a s t and one-index c c c -contractions T s t T_{st} T s t have been evaluated above. We first describe the four multiplicity-one blocks as tensors and evaluate the seven-generator contraction used in their normal operators.
Represent X ∈ End W C X\in\operatorname{End}W_{\mathbb{C}} X ∈ End W C by its four-tensor X i j , k l X_{ij,kl} X ij , k l , symmetric and traceless in each pair, with ( X h ) i j = X i j , k l h k l (Xh)_{ij}=X_{ij,kl}h^{kl} ( X h ) ij = X ij , k l h k l . Here the comma is only used to separate groups of indices, and does not denote differentiation. Then ( X T W ) i j , k l = X k l , i j (X^{T_W})_{ij,kl}=X_{kl,ij} ( X T W ) ij , k l = X k l , ij . Define the partial trace r ( X ) r(X) r ( X ) and the φ \varphi φ -contraction q ( X ) q(X) q ( X ) of the four-tensor of X X X ,
r ( X ) i k = X i j , k j , q ( X ) i k m = X i j , k l φ j l m . r(X)_{ik}=X_{ij,k}{}^{j},\qquad q(X)_{ikm}=X_{ij,kl}\varphi^{jl}{}_{m}. r ( X ) ik = X ij , k j , q ( X ) ik m = X ij , k l φ j l m .
From (10) , we obtain
q ( X ) i k m φ k n m = X i j , k l φ j l m φ k n m = X i j , k l ( δ j k δ l n − δ j n δ l k + ψ j l k n ) = r ( X ) i n . (167) \begin{aligned}
q(X)_{ikm}\varphi^k{}_{n}{}^m
&=X_{ij,kl}\varphi^{jl}{}_{m}\varphi^k{}_{n}{}^m\\
&=X_{ij,kl}\left(\delta^{jk}\delta^l{}_n-\delta^j{}_n\delta^{lk}+\psi^{jlk}{}_{n}\right)\\
&=r(X)_{in}. \tag*{(167)}
\end{aligned} q ( X ) ik m φ k n m = X ij , k l φ j l m φ k n m = X ij , k l ( δ j k δ l n − δ j n δ l k + ψ j l k n ) = r ( X ) in . (167)
since each pair of indices of X i j , k l X_{ij,kl} X ij , k l is traceless and symmetric. Therefore, q ( X ) = 0 q(X)=0 q ( X ) = 0 forces r ( X ) = 0 r(X)=0 r ( X ) = 0 . Moreover, if X T W = ϵ X X^{T_W}=\epsilon X X T W = ϵ X with ϵ = ± 1 \epsilon=\pm1 ϵ = ± 1 , then q ( X ) k i m = − ϵ q ( X ) i k m q(X)_{kim}=-\epsilon q(X)_{ikm} q ( X ) k im = − ϵ q ( X ) ik m . In particular, for W W W -self-adjoint X X X , q ( X ) q(X) q ( X ) is skew in the indices i , k i,k i , k , and for W W W -skew-adjoint X X X , q ( X ) q(X) q ( X ) is instead symmetric on these indices. Let S ( X ) S(X) S ( X ) denote the orthogonal projection of the full symmetrization of X i j , k l X_{ij,kl} X ij , k l in its four V V V -indices onto S 0 4 V ∗ S^4_0V^* S 0 4 V ∗ , the space of fully symmetric traceless four-tensors, regarded again as an endomorphism through its two pairs of indices. For W W W -skew-adjoint X X X , the full symmetrization vanishes, so S ( X ) = 0 S(X)=0 S ( X ) = 0 . For W W W -self-adjoint X X X the full symmetrization is the average of the three pairings below, and its trace over one pair is 2 3 r ( X ) \frac{2}{3}r(X) 3 2 r ( X ) . Hence on a W W W -self-adjoint X X X with r ( X ) = 0 r(X)=0 r ( X ) = 0 , no trace correction is needed, and
S ( X ) i j , k l = 1 3 ( X i j , k l + X i k , j l + X i l , j k ) . (168) S(X)_{ij,kl}=\frac{1}{3}\left(X_{ij,kl}+X_{ik,jl}+X_{il,jk}\right). \tag*{(168)} S ( X ) ij , k l = 3 1 ( X ij , k l + X ik , j l + X i l , j k ) . (168)
Here S ( X ) S(X) S ( X ) is a fully symmetric traceless four-tensor, S 2 ( X ) = S ( X ) S^2(X)=S(X) S 2 ( X ) = S ( X ) , and q ( S ( X ) ) = 0 q(S(X))=0 q ( S ( X )) = 0 .
We will also use the equivariant map ȷ 3 : S 0 3 V ∗ → Λ 2 W C \jmath_3:S_0^3V^* \to\Lambda^2W_{\mathbb{C}} 3 : S 0 3 V ∗ → Λ 2 W C defined by
ȷ 3 ( t ) i j , k l = t i k m φ j l m + t i l m φ j k m + t j k m φ i l m + t j l m φ i k m . \jmath_3(t)_{ij,kl}=t_{ikm}\varphi_{jl}^{m}+t_{ilm}\varphi_{jk}^{m}+t_{jkm}\varphi_{il}^{m}+t_{jlm}\varphi_{ik}^{m}. 3 ( t ) ij , k l = t ik m φ j l m + t i l m φ j k m + t j k m φ i l m + t j l m φ ik m .
The transpose q T q^{\mathsf{T}} q T is taken with respect to the full ordered pairing ⟨ u , v ⟩ 3 = u i k m v i k m \langle u,v\rangle_3=u_{ikm}v^{ikm} ⟨ u , v ⟩ 3 = u ik m v ik m on three-tensors and the pairing ⟨ X , Y ⟩ E n d = tr W ( X T W Y ) \langle X,Y\rangle_{\mathrm{End}}=\operatorname{tr}_{W}(X^{\mathsf{T}W}Y) ⟨ X , Y ⟩ End = tr W ( X T W Y ) on End ( W C ) \operatorname{End}(W_{\mathbb{C}}) End ( W C ) in (84) .
Lemma A.3. The four higher-dimensional multiplicity-one blocks in End W C \operatorname{End}W_{\mathbb{C}} End W C (from Table 1) have the following properties:
Table 1 (PDF p. 99).
In the third row, X = ȷ 3 ( t ) X=\jmath_3(t) X = 3 ( t ) for a unique t ∈ S 0 3 V ∗ t\in S_0^3V^* t ∈ S 0 3 V ∗ ; moreover, q T q ( X ) = 2 X q^{\mathsf{T}}q(X)=2X q T q ( X ) = 2 X on this block.
Proof. From (A.10), q ( X ) = 0 q(X)=0 q ( X ) = 0 implies r ( X ) = 0 r(X)=0 r ( X ) = 0 , hence to prove r ( X ) = 0 r(X)=0 r ( X ) = 0 , it is sufficient to show q ( X ) = 0 q(X)=0 q ( X ) = 0 . The identification V ( 4 , 0 ) = S 0 4 V ∗ V(4,0)=S_0^4V^* V ( 4 , 0 ) = S 0 4 V ∗ gives the first row: full symmetry implies q ( X ) = 0 q(X)=0 q ( X ) = 0 , and S ( X ) = X S(X)=X S ( X ) = X .
For V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) , the vanishing of q q q and S S S follows by equivariance. The map q q q sends W W W -self-adjoint X X X into
Λ 2 V ∗ ⊗ V ∗ ≅ 1 ⊕ 2 ⋅ 7 ⊕ 14 ⊕ 2 ⋅ 27 ⊕ 64 , \Lambda^2V^*\otimes V^*\cong\mathbf{1}\oplus2\cdot\mathbf{7}\oplus\mathbf{14}\oplus2\cdot\mathbf{27}\oplus\mathbf{64}, Λ 2 V ∗ ⊗ V ∗ ≅ 1 ⊕ 2 ⋅ 7 ⊕ 14 ⊕ 2 ⋅ 27 ⊕ 64 ,
where the decomposition follows from Λ 2 V = Λ 7 2 ⊕ Λ 14 2 \Lambda^2V=\Lambda^2_7\oplus\Lambda^2_{14} Λ 2 V = Λ 7 2 ⊕ Λ 14 2 , Λ 7 2 ≅ V \Lambda^2_7\cong V Λ 7 2 ≅ V , V ⊗ V ≅ 1 ⊕ 7 ⊕ 14 ⊕ 27 V\otimes V\cong\mathbf{1}\oplus\mathbf{7}\oplus\mathbf{14}\oplus\mathbf{27} V ⊗ V ≅ 1 ⊕ 7 ⊕ 14 ⊕ 27 , and (30) . On the other hand, S S S maps into
S 4 V ∗ = S 0 4 V ∗ ⊕ ( g ⋅ S 0 2 V ∗ ) ⊕ C g ⋅ g , S^4V^*=S_0^4V^*\oplus(g\cdot S_0^2V^*)\oplus\mathbb{C}g\cdot g, S 4 V ∗ = S 0 4 V ∗ ⊕ ( g ⋅ S 0 2 V ∗ ) ⊕ C g ⋅ g ,
where ⋅ \cdot ⋅ denotes the product in the symmetric algebra S ∙ V ∗ S^\bullet V^* S ∙ V ∗ , and neither target contains V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) . Therefore the images must be 0 by Schur’s lemma.
To compute q ( ȷ 3 ( t ) ) q(\jmath_3(t)) q ( 3 ( t )) , we apply (10) . The four terms then contribute 6 t , t , t , 0 6t,t,t,0 6 t , t , t , 0 to q ( ȷ 3 ( t ) ) = 8 t q(\jmath_3(t))=8t q ( 3 ( t )) = 8 t . Hence, ȷ 3 \jmath_3 3 is injective, and since S 0 3 V ∗ = V ( 3 , 0 ) S_0^3V^*=V(3,0) S 0 3 V ∗ = V ( 3 , 0 ) occurs once in Λ 2 W C \Lambda^2W_{\mathbb{C}} Λ 2 W C , its image is precisely that component. Note that since t t t is symmetric, we have r ( ȷ 3 ( t ) ) = 0 r(\jmath_3(t))=0 r ( 3 ( t )) = 0 .
The 189-dimensional component V ( 2 , 1 ) V(2,1) V ( 2 , 1 ) is W W W -skew-adjoint and satisfies q ( X ) = 0 q(X)=0 q ( X ) = 0 , and hence r ( X ) = 0 r(X)=0 r ( X ) = 0 . Indeed, q q q maps it into the invariant direct sum
S 2 V ∗ ⊗ V ∗ = S 3 V ∗ ⊕ ker ( Sym ) , S^2V^*\otimes V^*=S^3V^*\oplus\ker(\operatorname{Sym}), S 2 V ∗ ⊗ V ∗ = S 3 V ∗ ⊕ ker ( Sym ) ,
whose summands have dimensions 84 and 112. Each projection of q q q to these summands must vanish: a nonzero equivariant map from the irreducible 189-dimensional module would be injective, which is impossible for either target.
For the two skew-adjoint components, S ( X ) = 0 S(X)=0 S ( X ) = 0 by antisymmetry under exchange of the two index pairs. Full ordered pairings also give q T ( t ) = ȷ 3 ( t ) / 4 q^{\mathsf{T}}(t)=\jmath_3(t)/4 q T ( t ) = 3 ( t ) /4 , hence q T q ( X ) = 2 X q^{\mathsf{T}}q(X)=2X q T q ( X ) = 2 X on V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) . □
To evaluate Z φ φ \mathcal{Z}_{\varphi\varphi} Z φφ on these blocks, we must contract the two alternating-vector terms in (71) . One of them gives ∑ i L i X L i \sum_iL_iXL_i ∑ i L i X L i . Here Θ i \Theta_i Θ i is the generator of Λ 7 2 \Lambda^2_7 Λ 7 2 dual to e i e_i e i , and L i = 1 2 ρ ( Θ i ) L_i=\frac{1}{2}\rho(\Theta_i) L i = 2 1 ρ ( Θ i ) is one half of its action on W W W , given by
( L i ) α β = η ( k α , k β ) i , L i ( h ) = 1 2 [ Θ i , h ] . (L_i)_{\alpha\beta}=\eta(k_\alpha,k_\beta)_i,\qquad L_i(h)=\frac{1}{2}[\Theta_i,h]. ( L i ) α β = η ( k α , k β ) i , L i ( h ) = 2 1 [ Θ i , h ] .
The next lemma expresses this sum through q q q and S S S , whose values on the four blocks are known from Lemma A.3 . This lets us compute the required contraction using only the two-φ \varphi φ contraction identity (10) .
Lemma A.4. For a W W W -self-adjoint or W W W -skew-adjoint endomorphism X ∈ End ( W C ) X \in\operatorname{End}(W_{\mathbb{C}}) X ∈ End ( W C ) with r ( X ) = 0 r(X)=0 r ( X ) = 0 , the following identity holds:
∑ i L i X L i = { ⊤ q ( X ) + 3 S ( X ) − 2 X , X T W = X , r ( X ) = 0 , ⊤ q ( X ) − X , X T W = − X , r ( X ) = 0. (169) \sum_i L_i X L^i =
\begin{cases}
\top q(X)+3S(X)-2X, & X^{\mathsf{T}W}=X,\ r(X)=0,\\
\top q(X)-X, & X^{\mathsf{T}W}=-X,\ r(X)=0.
\end{cases} \tag*{(169)} i ∑ L i X L i = { ⊤ q ( X ) + 3 S ( X ) − 2 X , ⊤ q ( X ) − X , X T W = X , r ( X ) = 0 , X T W = − X , r ( X ) = 0. (169)
Proof. For r ( X ) = 0 r(X)=0 r ( X ) = 0 , put
D ( X ) a b , c d = 1 4 ( X c b , a d + X d b , c a + X a c , b d + X a d , c b ) . D(X)_{ab,cd}=\frac{1}{4}\left(X_{cb,ad}+X_{db,ca}+X_{ac,bd}+X_{ad,cb}\right). D ( X ) ab , c d = 4 1 ( X c b , a d + X d b , c a + X a c , b d + X a d , c b ) .
Expanding L i ( h ) = 1 2 [ Θ i , h ] L_i(h)=\frac{1}{2}[\Theta_i,h] L i ( h ) = 2 1 [ Θ i , h ] on both sides of X X X gives
( ∑ i L i X L i ) a b , c d = − 1 4 ( φ i a j φ c i l X b j , d l + φ i a j φ d i l X b j , c l + φ i b j φ c i l X a j , d l + φ i b j φ d i l X a j , c l ) . \left(\sum_i L_i X L^i\right)_{ab,cd}
=-\frac{1}{4}\left(\varphi_{ia}{}^j\varphi_c{}^{il}X_{bj,dl}+\varphi_{ia}{}^j\varphi_d{}^{il}X_{bj,cl}
+\varphi_{ib}{}^j\varphi_c{}^{il}X_{aj,dl}+\varphi_{ib}{}^j\varphi_d{}^{il}X_{aj,cl}\right). ( i ∑ L i X L i ) ab , c d = − 4 1 ( φ ia j φ c i l X bj , d l + φ ia j φ d i l X bj , c l + φ ib j φ c i l X aj , d l + φ ib j φ d i l X aj , c l ) .
For example, the first product contracts by (10) as
φ i a j φ c l i = δ a c δ j l − δ a l δ j c + ψ a j c l . \varphi_{iaj}\varphi_{cl}{}^i=\delta_{ac}\delta_{jl}-\delta_{al}\delta_{jc}+\psi_{ajcl}. φ iaj φ c l i = δ a c δ j l − δ a l δ j c + ψ aj c l .
The first delta product contributes − 1 4 δ a c r ( X ) b d = 0 -\frac{1}{4}\delta_{ac}r(X)_{bd}=0 − 4 1 δ a c r ( X ) b d = 0 , and the second contributes 1 4 X b c , d a \frac{1}{4}X_{bc,da} 4 1 X b c , d a . Applying the same identity to the other three products gives
( ∑ i L i X L i ) a b , c d = D ( X ) a b , c d − 1 4 ( ψ a c j l X b j , d l + ψ a d j l X b j , c l + ψ b c j l X a j , d l + ψ b d j l X a j , c l ) . \left(\sum_i L_i X L^i\right)_{ab,cd}
=D(X)_{ab,cd}-\frac{1}{4}\left(\psi_{a c}{}^j{}_lX_{bj,dl}+\psi_{a d}{}^j{}_lX_{bj,cl}
+\psi_{b c}{}^j{}_lX_{aj,dl}+\psi_{b d}{}^j{}_lX_{aj,cl}\right). ( i ∑ L i X L i ) ab , c d = D ( X ) ab , c d − 4 1 ( ψ a c j l X bj , d l + ψ a d j l X bj , c l + ψ b c j l X aj , d l + ψ b d j l X aj , c l ) .
Thus the surviving delta terms sum to D ( X ) D(X) D ( X ) . On the other hand, pair symmetrization in the definition of ⊤ q \top q ⊤ q gives
( ⊤ q ( X ) ) a b , c d = 1 4 ( φ b d m q ( X ) a c m + φ b c m q ( X ) a d m + φ a d m q ( X ) b c m + φ a c m q ( X ) b d m ) . (\top q(X))_{ab,cd}
=\frac{1}{4}\left(\varphi_{bd}{}^m q(X)_{acm}+\varphi_{bc}{}^m q(X)_{adm}
+\varphi_{ad}{}^m q(X)_{bcm}+\varphi_{ac}{}^m q(X)_{bdm}\right). ( ⊤ q ( X ) ) ab , c d = 4 1 ( φ b d m q ( X ) a c m + φ b c m q ( X ) a d m + φ a d m q ( X ) b c m + φ a c m q ( X ) b d m ) .
Its pair traces vanish because q ( X ) i k m φ n m = r ( X ) i n = 0 q(X)_{ikm}\varphi_n{}^{m}=r(X)_{in}=0 q ( X ) ik m φ n m = r ( X ) in = 0 . After substituting q ( X ) i k m = X i j , k l φ j l m q(X)_{ikm}=X_{ij,kl}\varphi^{jl}{}_{m} q ( X ) ik m = X ij , k l φ j l m , the two-φ \varphi φ products appear explicitly:
( ⊤ q ( X ) ) a b , c d = 1 4 ( φ b d m φ j l m X a j , c l + φ b c m φ j l m X a j , d l + φ a d m φ j l m X b j , c l + φ a c m φ j l m X b j , d l ) . (\top q(X))_{ab,cd}
=\frac{1}{4}\left(\varphi_{bd}{}^m\varphi^{jl}{}_{m}X_{aj,cl}+\varphi_{bc}{}^m\varphi^{jl}{}_{m}X_{aj,dl}
+\varphi_{ad}{}^m\varphi^{jl}{}_{m}X_{bj,cl}+\varphi_{ac}{}^m\varphi^{jl}{}_{m}X_{bj,dl}\right). ( ⊤ q ( X ) ) ab , c d = 4 1 ( φ b d m φ j l m X aj , c l + φ b c m φ j l m X aj , d l + φ a d m φ j l m X bj , c l + φ a c m φ j l m X bj , d l ) .
For the first product, (10) gives
φ b d m φ j l m X a j , c l = X a b , c d − X a d , c b + ψ b d j l X a j , c l . \varphi_{bd}{}^m\varphi^{jl}{}_{m}X_{aj,cl}=X_{ab,cd}-X_{ad,cb}+\psi_{bd}{}^{jl}X_{aj,cl}. φ b d m φ j l m X aj , c l = X ab , c d − X a d , c b + ψ b d j l X aj , c l .
The four delta contributions therefore sum to X − D ( X ) X-\mathcal{D}(X) X − D ( X ) . The four ψ \psi ψ contributions agree with those in ∑ i L i X L i \sum_i L_iXL^i ∑ i L i X L i : for example, ψ b d j l X a j , c l = − ψ b j d l X a j , c l \psi_{bd}^{jl}X_{aj,cl}=-\psi_{bj}^{dl}X_{aj,cl} ψ b d j l X aj , c l = − ψ bj d l X aj , c l , and the other three terms agree by the same interchange of indices. Consequently,
∑ i L i X L i = q ⊤ q ( X ) − X + 2 D ( X ) . \sum_i L_iXL^i=q^\top q(X)-X+2\mathcal{D}(X). i ∑ L i X L i = q ⊤ q ( X ) − X + 2 D ( X ) .
For self-adjoint X X X , D ( X ) = ( 3 S ( X ) − X ) / 2 \mathcal{D}(X)=(3\mathcal{S}(X)-X)/2 D ( X ) = ( 3 S ( X ) − X ) /2 ; for skew-adjoint X X X , D ( X ) = 0 \mathcal{D}(X)=0 D ( X ) = 0 . This proves (A.12). □ \square □
The four maps B p B_p B p and their ordering are fixed before Lemma 3.11. We now evaluate their Gram matrix.
Proof of Lemma 3.11. For s , t , u , v ∈ { 0 , φ } s,t,u,v\in\{0,\varphi\} s , t , u , v ∈ { 0 , φ } , the definition of Φ s ⊤ \Phi_s^\top Φ s ⊤ , (3.26) and (3.25) give
⟨ B s t ( h 1 ) , B u v ( h 2 ) ⟩ 3 = ⟨ m h 1 ( t ) , Φ s ⊤ Φ u ( m h 2 ( v ) ) ⟩ E n d = 3 a s u a t v ⟨ h 1 , h 2 ⟩ W + 6 tr W ( m h 1 ( t ) Z s u ( m h 2 ( v ) ) ) . \langle B_{st}(h_1),B_{uv}(h_2)\rangle_3
=\langle m_{h_1}^{(t)},\Phi_s^\top\Phi_u(m_{h_2}^{(v)})\rangle_{\mathrm{End}}
=3a_{su}a_{tv}\langle h_1,h_2\rangle_W+6\operatorname{tr}_W\left(m_{h_1}^{(t)}\mathcal{Z}_{su}(m_{h_2}^{(v)})\right). ⟨ B s t ( h 1 ) , B uv ( h 2 ) ⟩ 3 = ⟨ m h 1 ( t ) , Φ s ⊤ Φ u ( m h 2 ( v ) ) ⟩ End = 3 a s u a t v ⟨ h 1 , h 2 ⟩ W + 6 tr W ( m h 1 ( t ) Z s u ( m h 2 ( v ) ) ) .
By equivariance and Schur’s lemma, B p ⊤ B q = G p q id W B_p^\top B_q=G_{pq}\operatorname{id}_W B p ⊤ B q = G pq id W . Taking the trace over W W W in the preceding pairing formula gives
G ( s , t ) , ( u , v ) = 3 a s u a t v + 2 9 T t s δ α β γ ( T u v ) α γ δ β . (170) \mathcal{G}_{(s,t),(u,v)}=3a_{su}a_{tv}+\frac{2}{9}T_{ts}^{\delta\alpha\beta\gamma}(T^{uv})_{\alpha\gamma\delta\beta}. \tag*{(170)} G ( s , t ) , ( u , v ) = 3 a s u a t v + 9 2 T t s δ α β γ ( T uv ) α γ δ β . (170)
To see the index order, note that the second trace is the sum
∑ δ tr W ( m k δ ( t ) Z s u ( m k δ ( v ) ) ) = c δ α β t ( c s ) β γ ϵ ( c u ) α σ ϵ ( c v ) σ γ δ . \sum_\delta\operatorname{tr}_W\left(m_{k_\delta}^{(t)}\mathcal{Z}_{su}(m_{k_\delta}^{(v)})\right)
=c_{\delta\alpha\beta}^{t}(c^{s})^{\beta\gamma\epsilon}(c^{u})^\alpha{}_{\sigma\epsilon}(c^{v})^\sigma{}_{\gamma}{}^\delta. δ ∑ tr W ( m k δ ( t ) Z s u ( m k δ ( v ) ) ) = c δ α β t ( c s ) β γ ϵ ( c u ) α σ ϵ ( c v ) σ γ δ .
Renaming the contracted slots pairs the first two cubics into T t s T^{ts} T t s and the last two into T u v T^{uv} T uv as in (A.13). Division by dim W = 27 \dim W=27 dim W = 27 changes the coefficient 6 6 6 into 2 / 9 2/9 2/9 .
For a representative entry, expand the two T 00 T^{00} T 00 factors using (3.10). Successive contractions with the projection identity and the sums (A.8) give
T δ α β γ 00 ( T 00 ) α γ δ β = 1 4 ( 3 4995 392 + 11043 49 ) − 1 7 ( 135 14 + 729 7 ) + 27 49 = 78705 1568 . T^{00}_{\delta\alpha\beta\gamma}(T^{00})^{\alpha\gamma\delta\beta}
=\frac{1}{4}\left(3\frac{4995}{392}+\frac{11043}{49}\right)-\frac{1}{7}\left(\frac{135}{14}+\frac{729}{7}\right)+\frac{27}{49}
=\frac{78705}{1568}. T δ α β γ 00 ( T 00 ) α γ δ β = 4 1 ( 3 392 4995 + 49 11043 ) − 7 1 ( 14 135 + 7 729 ) + 49 27 = 1568 78705 .
The first bracket consists of the four trace-times-trace terms; three agree by cyclicity and reversal of trace. The next bracket contains the two trace corrections, and the last term comes from contracting the two metric products. Thus G 11 = 3 ( 55 / 28 ) 2 + ( 2 / 9 ) ( 78705 / 1568 ) = 4455 / 196 \mathcal{G}_{11}=3(55/28)^2+(2/9)(78705/1568)=4455/196 G 11 = 3 ( 55/28 ) 2 + ( 2/9 ) ( 78705/1568 ) = 4455/196 . The full table of contractions in (A.13), in the ordered pairs ( 0 , 0 ) , ( φ , 0 ) , ( 0 , φ ) , ( φ , φ ) (0,0),(\varphi,0),(0,\varphi),(\varphi,\varphi) ( 0 , 0 ) , ( φ , 0 ) , ( 0 , φ ) , ( φ , φ ) , is
( T δ α β γ t s ( T u v ) α γ δ β ) ( s , t ) , ( u , v ) = 1 1568 ( 78705 31482 31482 − 359964 31482 − 359964 − 359964 601128 31482 − 359964 − 359964 601128 − 359964 601128 601128 5523984 ) . \left(T^{ts}_{\delta\alpha\beta\gamma}(T^{uv})^{\alpha\gamma\delta\beta}\right)_{(s,t),(u,v)}
=\frac{1}{1568}
\begin{pmatrix}
78705 & 31482 & 31482 & -359964 \\
31482 & -359964 & -359964 & 601128 \\
31482 & -359964 & -359964 & 601128 \\
-359964 & 601128 & 601128 & 5523984
\end{pmatrix}. ( T δ α β γ t s ( T uv ) α γ δ β ) ( s , t ) , ( u , v ) = 1568 1 78705 31482 31482 − 359964 31482 − 359964 − 359964 601128 31482 − 359964 − 359964 601128 − 359964 601128 601128 5523984 .
For every other entry we use the same contraction order in (A.13). We first expand the two T T T factors, using (70) for the 00 and mixed factors and (71) for a φ φ \varphi\varphi φφ factor. We then sum each repeated basis index with the projection identity and contract the resulting V V V -indices with (10) . In particular, the two alternating-vector terms of the φ φ \varphi\varphi φφ identity retain the displayed index order throughout this expansion. This gives the contraction table above, including the off-diagonal entries. Substitution in (A.13) gives the matrix in (91) . Its determinant, 179397504 / 7 179397504/7 179397504/7 , is nonzero.
On each multiplicity-one block, Γ s t \Gamma_{st} Γ s t is the scalar of Φ s T Φ t \Phi_s^{\mathsf T}\Phi_t Φ s T Φ t as in Lemma 3.12 .
Proof of Lemma 3.12. By Lemma A.3 , r ( X ) = 0 r(X)=0 r ( X ) = 0 on each of these components, and the same lemma gives the values of q q q and S S S there. First contract the three identities in (70) with X X X in the index order specified by (88) . A single insertion of the projection identity gives
Z 00 ( X ) = { − 11 28 X + 3 4 S ( X ) , X T W = X , r ( X ) = 0 , 1 7 X , X T W = − X , r ( X ) = 0. (171) Z_{00}(X)=
\begin{cases}
-\frac{11}{28}X+\frac{3}{4}S(X), & X^{\mathsf T}W=X,\ r(X)=0,\\
\frac{1}{7}X, & X^{\mathsf T}W=-X,\ r(X)=0.
\end{cases} \tag*{(171)} Z 00 ( X ) = { − 28 11 X + 4 3 S ( X ) , 7 1 X , X T W = X , r ( X ) = 0 , X T W = − X , r ( X ) = 0. (171)
More explicitly, before the normalized trace tensor g / 7 g/\sqrt{7} g / 7 is removed, the four-tensor of Z 00 ( X ) Z_{00}(X) Z 00 ( X ) is
1 8 ( ( X T W ) l j , k i + ( X T W ) k j , l i + ( X T W ) l i , k j + ( X T W ) k i , l j ) , \frac{1}{8}\left((X^{\mathsf T}W)_{lj,ki}+(X^{\mathsf T}W)_{kj,li}+(X^{\mathsf T}W)_{li,kj}+(X^{\mathsf T}W)_{ki,lj}\right), 8 1 ( ( X T W ) l j , k i + ( X T W ) k j , l i + ( X T W ) l i , k j + ( X T W ) k i , l j ) ,
plus terms involving r ( X ) r(X) r ( X ) . Removing g / 7 g/\sqrt{7} g / 7 subtracts X T W / 7 X^{\mathsf T}W/7 X T W /7 . The bracket equals 2 ( 3 S ( X ) − X ) 2(3S(X)-X) 2 ( 3 S ( X ) − X ) in the self-adjoint case and zero in the skew-adjoint case, giving (A.14).
For X T W = ϵ X X^{\mathsf T}W=\epsilon X X T W = ϵ X with r ( X ) = 0 r(X)=0 r ( X ) = 0 , the same contraction of the mixed identity gives
Z 0 φ ( X ) i j , k l = − ϵ 4 ( φ i k c q ( X ) j c l + φ i l c q ( X ) j c k + φ j k c q ( X ) i c l + φ j l c q ( X ) i c k ) + ϵ 7 X i j , k l . Z_{0\varphi}(X)_{ij,kl}=-\frac{\epsilon}{4}\left(\varphi^{c}_{ik}q(X)_{jcl}+\varphi^{c}_{il}q(X)_{jck}+\varphi^{c}_{jk}q(X)_{icl}+\varphi^{c}_{jl}q(X)_{ick}\right)+\frac{\epsilon}{7}X_{ij,kl}. Z 0 φ ( X ) ij , k l = − 4 ϵ ( φ ik c q ( X ) j c l + φ i l c q ( X ) j c k + φ j k c q ( X ) i c l + φ j l c q ( X ) i c k ) + 7 ϵ X ij , k l .
The last term is the correction from the normalized trace tensor g / 7 g/\sqrt{7} g / 7 , since P ( g , h ) = − h \mathcal{P}(g,h)=-h P ( g , h ) = − h on W W W . The reversed-order operator is obtained by W W W -adjoint transposition: Z φ 0 ( X ) = Z 0 φ ( X T W ) T W Z_{\varphi0}(X)=Z_{0\varphi}(X^{\mathsf T}W)^{\mathsf T}W Z φ 0 ( X ) = Z 0 φ ( X T W ) T W . Consequently both mixed operators equal ϵ X / 7 \epsilon X/7 ϵ X /7 when q ( X ) = 0 q(X)=0 q ( X ) = 0 . For V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) , Lemma A.3 gives q ( ȷ 3 ( t ) ) = 8 t q(\jmath_3(t))=8t q ( 3 ( t )) = 8 t , so the bracket is 8 ȷ 3 ( t ) 8\jmath_3(t) 8 3 ( t ) . Thus both mixed operators equal 13 X / 7 13X/7 13 X /7 .
Finally evaluate (71) at ( h 1 , h 2 , h 3 , h 4 ) = ( k α , k γ , k β , k σ ) (h_1,h_2,h_3,h_4)=(k_\alpha,k^\gamma,k_\beta,k_\sigma) ( h 1 , h 2 , h 3 , h 4 ) = ( k α , k γ , k β , k σ ) and contract with X σ γ X^\sigma{}_\gamma X σ γ . The crossed-trace term − 8 tr ( h 1 h 3 h 2 h 4 ) -8\operatorname{tr}(h_1h_3h_2h_4) − 8 tr ( h 1 h 3 h 2 h 4 ) vanishes because its contraction contains the transpose of r ( X ) r(X) r ( X ) . Write m T , n T m^{\mathsf T},n^{\mathsf T} m T , n T for the algebraic transposes of h ↦ m h , n h h\mapsto m_h,n_h h ↦ m h , n h . Then the six mixed cubic terms contract to
Z φ 0 ( X + X T W ) + Z 0 φ ( X + X T W ) + n m T X + m n T X . Z_{\varphi0}(X+X^{\mathsf T}W)+Z_{0\varphi}(X+X^{\mathsf T}W)+nm^{\mathsf T}X+mn^{\mathsf T}X. Z φ 0 ( X + X T W ) + Z 0 φ ( X + X T W ) + n m T X + m n T X .
The last two terms vanish for W W W -skew-adjoint X X X . For self-adjoint X X X with q ( X ) = 0 q(X)=0 q ( X ) = 0 they vanish as well: m T X = r ( X ) 0 m^{\mathsf T}X=r(X)_0 m T X = r ( X ) 0 , and n T X n^{\mathsf T}X n T X is the traceless part of φ i a φ j d X i j , b d \varphi^i{}_a\varphi^j{}_{d}X_{ij,bd} φ i a φ j d X ij , b d , a contraction of q ( X ) q(X) q ( X ) . Of the two η \eta η -terms, the first contracts to zero by r ( X ) = 0 r(X)=0 r ( X ) = 0 ; the second is ∑ i L i X L i \sum_i L_iXL^i ∑ i L i X L i . The metric contribution is $12X/7$ for W W W -self-adjoint X X X and 8 X / 7 8X/7 8 X /7 for W W W -skew-adjoint $X.
For the former, the mixed contribution is 2 ( 4 X / 7 ) 2(4X/7) 2 ( 4 X /7 ) when q ( X ) = 0 q(X)=0 q ( X ) = 0 ; for the latter it is zero. Lemma A.4 therefore gives
Z φ φ ( X ) = 48 7 X − 6 S ( X ) ( X T W = X , q ( X ) = 0 ) , Z φ φ ( X ) = 22 7 X − 2 q T q ( X ) ( X T W = − X , r ( X ) = 0 ) . (172) \begin{aligned}
Z_{\varphi\varphi}(X)&=\frac{48}{7}X-6S(X) &&(X^{\mathsf{T}}W=X,\ q(X)=0),\\
Z_{\varphi\varphi}(X)&=\frac{22}{7}X-2q^{\mathsf{T}}q(X) &&(X^{\mathsf{T}}W=-X,\ r(X)=0). \tag*{(172)}
\end{aligned} Z φφ ( X ) Z φφ ( X ) = 7 48 X − 6 S ( X ) = 7 22 X − 2 q T q ( X ) ( X T W = X , q ( X ) = 0 ) , ( X T W = − X , r ( X ) = 0 ) . (172)
By Lemma A.3 , S S S acts as 1 1 1 on V ( 4 , 0 ) V(4,0) V ( 4 , 0 ) and as 0 0 0 on V ( 0 , 2 ) V(0,2) V ( 0 , 2 ) , and q T q q^{\mathsf{T}}q q T q acts as 2 2 2 on V ( 3 , 0 ) V(3,0) V ( 3 , 0 ) and as 0 0 0 on V ( 2 , 1 ) V(2,1) V ( 2 , 1 ) . Substituting these values gives:
Table 1 (PDF p. 103).
Finally, by (88) , Γ s t \Gamma_{st} Γ s t is 3 a s t 3a_{st} 3 a s t plus six times the scalar of Z s t \mathcal{Z}_{st} Z s t in the table above, which yields the stated normal-operator scalars and determinants.
Quartic identities
In this section, we prove Lemma 3.5 of § 4 . The proof uses one contraction of φ \varphi φ with the Hodge dual of a 3-form, which we record first.
Lemma A.5. For u , v ∈ V u,v\in V u , v ∈ V ,
∗ ( ( ι u φ ) ∧ ( ι v φ ) ) = u ♭ ∧ ι v φ + v ♭ ∧ ι u φ . (173) *((\iota_u\varphi)\wedge(\iota_v\varphi))=u^{\flat}\wedge\iota_v\varphi+v^{\flat}\wedge\iota_u\varphi. \tag*{(173)} ∗ (( ι u φ ) ∧ ( ι v φ )) = u ♭ ∧ ι v φ + v ♭ ∧ ι u φ . (173) For any 3-form ξ \xi ξ ,
φ i a b φ j c d ( ∗ ξ ) a b c d = 2 ( ξ i a b φ j a b + ξ j a b φ i a b ) . (174) \varphi_{iab}\varphi_{jcd}(*\xi)^{abcd}=2\left(\xi_{iab}\varphi_j^{ab}+\xi_{jab}\varphi_i^{ab}\right). \tag*{(174)} φ iab φ j c d ( ∗ ξ ) ab c d = 2 ( ξ iab φ j ab + ξ j ab φ i ab ) . (174)
Proof. Since ψ = ∗ φ \psi=* \varphi ψ = ∗ φ , the mixed contraction (11) implies
[ ∗ ( ( ι u φ ) ∧ φ ) ] i j = 1 2 ψ i j a b ( ι u φ ) a b = 2 ( ι u φ ) i j . \left[*((\iota_u\varphi)\wedge\varphi)\right]_{ij}=\frac{1}{2}\psi_{ijab}(\iota_u\varphi)^{ab}=2(\iota_u\varphi)_{ij}. [ ∗ (( ι u φ ) ∧ φ ) ] ij = 2 1 ψ ij ab ( ι u φ ) ab = 2 ( ι u φ ) ij .
Together with ∗ ( ι u φ ) = u ♭ ∧ ψ *(\iota_u\varphi)=u^{\flat}\wedge\psi ∗ ( ι u φ ) = u ♭ ∧ ψ , this gives ( ι u φ ) ∧ φ = 2 u ♭ ∧ ψ (\iota_u\varphi)\wedge\varphi=2u^{\flat}\wedge\psi ( ι u φ ) ∧ φ = 2 u ♭ ∧ ψ . Write w = u × v w=u\times v w = u × v for the cross product (8) . Contracting that identity with v v v gives
( ι u φ ) ∧ ( ι v φ ) = 2 g ( u , v ) ψ − 2 u ♭ ∧ ι v ψ + φ ∧ w ♭ . (\iota_u\varphi)\wedge(\iota_v\varphi)=2g(u,v)\psi-2u^{\flat}\wedge\iota_v\psi+\varphi\wedge w^{\flat}. ( ι u φ ) ∧ ( ι v φ ) = 2 g ( u , v ) ψ − 2 u ♭ ∧ ι v ψ + φ ∧ w ♭ .
We interchange u , v u,v u , v in this last equation and average the two equations:
( ι u φ ) ∧ ( ι v φ ) = 2 g ( u , v ) ψ − u ♭ ∧ ι v ψ − v ♭ ∧ ι u ψ . (\iota_u\varphi)\wedge(\iota_v\varphi)=2g(u,v)\psi-u^{\flat}\wedge\iota_v\psi-v^{\flat}\wedge\iota_u\psi. ( ι u φ ) ∧ ( ι v φ ) = 2 g ( u , v ) ψ − u ♭ ∧ ι v ψ − v ♭ ∧ ι u ψ .
The standard Hodge identity for these degrees in dimension seven is
∗ ( u ♭ ∧ ι v ψ ) = g ( u , v ) φ − v ♭ ∧ ι u φ . *(u^{\flat}\wedge\iota_v\psi)=g(u,v)\varphi-v^{\flat}\wedge\iota_u\varphi. ∗ ( u ♭ ∧ ι v ψ ) = g ( u , v ) φ − v ♭ ∧ ι u φ .
Here ∗ * ∗ takes 4-forms to 3-forms. Applying this identity to the two insertion terms of the averaged equation, with ∗ ψ = φ *\psi=\varphi ∗ ψ = φ , proves (A.16). Finally, (A.16), paired with an arbitrary 3-form ξ \xi ξ , gives (A.17) in components. Converting the normalized pairing of forms to a sum over all ordered indices gives the factor 2.
For h ∈ W h \in W h ∈ W , recall Y = P ( h , h ) Y = \mathcal{P}(h,h) Y = P ( h , h ) and r = tr ( h 2 ) r = \operatorname{tr}(h^2) r = tr ( h 2 ) , and put γ 0 = C 0 ( h ) \gamma_0 = C_0(h) γ 0 = C 0 ( h ) . Also recall that { X , Y } = X Y + Y X \{X,Y\} = XY + YX { X , Y } = X Y + Y X denotes the anticommutator. Introduce a formal variable t t t and set L ( t ) j i = δ j i + t h j i L(t)^i_j = \delta^i_j + th^i_j L ( t ) j i = δ j i + t h j i , using the metric identification (13) . Its inverse M ( t ) = L ( t ) − 1 M(t) = L(t)^{-1} M ( t ) = L ( t ) − 1 is given by the Neumann series
M ( t ) j i = ∑ m ≥ 0 ( − t ) m ( h m ) j i = δ j i − t h j i + t 2 ( h 2 ) j i − t 3 ( h 3 ) j i + t 4 ( h 4 ) j i + O ( t 5 ) . (175) M(t)^i_j = \sum_{m \ge0} (-t)^m (h^m)^i_j
= \delta^i_j - th^i_j + t^2(h^2)^i_j - t^3(h^3)^i_j + t^4(h^4)^i_j + O(t^5). \tag*{(175)} M ( t ) j i = m ≥ 0 ∑ ( − t ) m ( h m ) j i = δ j i − t h j i + t 2 ( h 2 ) j i − t 3 ( h 3 ) j i + t 4 ( h 4 ) j i + O ( t 5 ) . (175)
Since tr ( h ) = 0 \operatorname{tr}(h) = 0 tr ( h ) = 0 , the identity det L = exp ( tr log L ) \det L = \exp(\operatorname{tr}\log L) det L = exp ( tr log L ) gives the determinant expansion
det L ( t ) = 1 − r 2 t 2 + γ 0 3 t 3 + ( r 2 8 − tr ( h 4 ) 4 ) t 4 + O ( t 5 ) . (176) \det L(t) = 1 - \frac{r}{2}t^2 + \frac{\gamma_0}{3}t^3 + \left(\frac{r^2}{8} - \frac{\operatorname{tr}(h^4)}{4}\right)t^4 + O(t^5). \tag*{(176)} det L ( t ) = 1 − 2 r t 2 + 3 γ 0 t 3 + ( 8 r 2 − 4 tr ( h 4 ) ) t 4 + O ( t 5 ) . (176)
For the inverse series, the logarithm series and the determinant–exponential identity, see [50 §3.4, eq. (187), and §10.1.5, eqs. (497), (502)] . Here we use these identities in End ( V ) [ [ t ] ] \operatorname{End}(V)[[t]] End ( V ) [[ t ]] and R [ [ t ] ] \mathbb{R}[[t]] R [[ t ]] ; the constant term of L L L is I I I , so no invertibility assumption on h h h is needed.
Both L L L and M M M are self-adjoint with respect to g g g . In tensor products such as P ( M , M ) \mathcal{P}(M,M) P ( M , M ) we write M M M for the symmetric covariant 2-tensor M i j = g i a M j a M_{ij} = g_{ia}M^a_j M ij = g ia M j a , as in (13) . Pullbacks insert the endomorphism in each covariant slot; for example, ( L ∗ ψ ) a b c d = L a p L b q L c r L d s ψ p q r s (L^*\psi)_{abcd} = L^p_aL^q_bL^r_cL^s_d\psi_{pqrs} ( L ∗ ψ ) ab c d = L a p L b q L c r L d s ψ pq r s . Define the symmetric polynomial tensor
Z ( t ) i j = φ i a b φ j c d ( L ∗ ψ ) a b c d . (177) Z(t)_{ij} = \varphi_i^{ab}\varphi_j^{cd}(L^*\psi)_{abcd}. \tag*{(177)} Z ( t ) ij = φ i ab φ j c d ( L ∗ ψ ) ab c d . (177)
Lemma A.6 (Formal resolvent). The pullback of ψ \psi ψ satisfies
∗ ( L ∗ ψ ) = det ( L ) M ∗ φ . (178) *(L^*\psi) = \det(L)M^*\varphi. \tag*{(178)} ∗ ( L ∗ ψ ) = det ( L ) M ∗ φ . (178) Consequently,
Z ( t ) = 2 det ( L ) { M , P ( M , M ) } . (179) Z(t) = 2\det(L)\{M,\mathcal{P}(M,M)\}. \tag*{(179)} Z ( t ) = 2 det ( L ) { M , P ( M , M )} . (179) Proof. Let ε \varepsilon ε be the oriented volume tensor. The cofactor identity is
ε b 1 ⋯ b 7 L a 1 b 1 ⋯ L a 7 b 7 = det ( L ) ε a 1 ⋯ a 7 . \varepsilon_{b_1\cdots b_7}L^{b_1}_{a_1}\cdots L^{b_7}_{a_7} = \det(L)\varepsilon_{a_1\cdots a_7}. ε b 1 ⋯ b 7 L a 1 b 1 ⋯ L a 7 b 7 = det ( L ) ε a 1 ⋯ a 7 . Contracting three slots with M M M and raising indices with g g g , using self-adjointness, gives
ε i j k L a p L b q L c r L d s = det ( L ) M i a M j b M k c ε a b c p q r s . \varepsilon_{ijk}L^p_aL^q_bL^r_cL^s_d = \det(L)M^a_iM^b_jM^c_k\varepsilon_{abc}^{\ \ \ \ pqrs}. ε ij k L a p L b q L c r L d s = det ( L ) M i a M j b M k c ε ab c pq r s . Contracting with ψ p q r s / 4 ! \psi_{pqrs}/4! ψ pq r s /4 ! and using ∗ ψ = φ *\psi= \varphi ∗ ψ = φ proves (A.21).
Take ξ = det ( L ) M ∗ φ \xi= \det(L)M^*\varphi ξ = det ( L ) M ∗ φ in (A.17). By (A.21), ∗ ξ = L ∗ ψ *\xi= L^*\psi ∗ ξ = L ∗ ψ , so the left-hand side is Z ( t ) i j Z(t)_{ij} Z ( t ) ij . On the right-hand side,
ξ i a b φ j a b = det ( L ) M i p M a q M b r φ p q r φ j a b = det ( L ) M p i P ( M , M ) p j . \xi_{iab}\varphi_j^{ab} = \det(L)M^p_iM^q_aM^r_b\varphi_{pqr}\varphi_j^{ab} = \det(L)M^i_p\mathcal{P}(M,M)_{pj}. ξ iab φ j ab = det ( L ) M i p M a q M b r φ pq r φ j ab = det ( L ) M p i P ( M , M ) p j . The other term is its transpose. Their sum gives (A.22), with the contracted products of (12) .
We now compare fourth-order coefficients in Lemma A.6 to compute Y 2 Y^2 Y 2 .
Lemma A.7 (Formal-resolvent quartic identity). For every h ∈ W h \in W h ∈ W , we have
Y 2 = 6 P ( h 2 , h 2 ) + 8 P ( h , h 3 ) + 4 { h , P ( h , h 2 ) } + 2 { h 2 , Y } − 2 r Y − 8 h 4 + 4 r h 2 + 8 3 γ 0 h + ( 2 tr ( h 4 ) − r 2 ) g . (180) Y^2 = 6\mathcal{P}(h^2,h^2) + 8\mathcal{P}(h,h^3) + 4\{h,\mathcal{P}(h,h^2)\} + 2\{h^2,Y\} - 2rY
- 8h^4 + 4rh^2 + \frac{8}{3}\gamma_0h + (2\operatorname{tr}(h^4)-r^2)g. \tag*{(180)} Y 2 = 6 P ( h 2 , h 2 ) + 8 P ( h , h 3 ) + 4 { h , P ( h , h 2 )} + 2 { h 2 , Y } − 2 r Y − 8 h 4 + 4 r h 2 + 3 8 γ 0 h + ( 2 tr ( h 4 ) − r 2 ) g . (180)
Proof. We compute the fourth-order coefficient in (A.22), then compute it directly in (A.20). Write P 1 = P ( h , h 2 ) P_1=\mathcal{P}(h,h^2) P 1 = P ( h , h 2 ) and P 2 = P ( h 2 , h 2 ) P_2=\mathcal{P}(h^2,h^2) P 2 = P ( h 2 , h 2 ) , and recall that W = P ( h , h 3 ) W=\mathcal{P}(h,h^3) W = P ( h , h 3 ) is as in Section A.1. The inverse series (A.18) and the elementary identity P ( g , h 1 ) = tr ( h 1 ) g − h 1 \mathcal{P}(g,h_1)=\operatorname{tr}(h_1)g-h_1 P ( g , h 1 ) = tr ( h 1 ) g − h 1 (41) give each coefficient of the quadratic expression:
P ( M , M ) = 6 g + 2 t h + t 2 ( 2 r g − 2 h 2 + Y ) + t 3 ( − 2 γ 0 g + 2 h 3 − 2 P 1 ) + t 4 ( 2 tr ( h 4 ) g − 2 h 4 + 2 W + P 2 ) + O ( t 5 ) . (181) \begin{aligned}
\mathcal{P}(M,M)={}&6g+2th+t^2(2rg-2h^2+Y)\\
&+t^3(-2\gamma_0g+2h^3-2P_1)\\
&+t^4(2\operatorname{tr}(h^4)g-2h^4+2W+P_2)+O(t^5). \tag*{(181)}
\end{aligned} P ( M , M ) = 6 g + 2 t h + t 2 ( 2 r g − 2 h 2 + Y ) + t 3 ( − 2 γ 0 g + 2 h 3 − 2 P 1 ) + t 4 ( 2 tr ( h 4 ) g − 2 h 4 + 2 W + P 2 ) + O ( t 5 ) . (181)
Before multiplying by the determinant, we multiply by M M M using (A.18). The coefficients of orders 0 , 1 , 2 0,1,2 0 , 1 , 2 are 6 g 6g 6 g , − 4 h -4h − 4 h , 2 r g + 2 h 2 + Y 2rg+2h^2+Y 2 r g + 2 h 2 + Y , respectively, and the coefficient of degree four is
2 tr ( h 4 ) g + 2 γ 0 h + 2 r h 2 − 2 h 4 + 2 W + P 2 + 2 h P 1 + h 2 Y . 2\operatorname{tr}(h^4)g+2\gamma_0h+2rh^2-2h^4+2W+P_2+2hP_1+h^2Y. 2 tr ( h 4 ) g + 2 γ 0 h + 2 r h 2 − 2 h 4 + 2 W + P 2 + 2 h P 1 + h 2 Y .
The determinant series (A.19) contributes to degree four through its terms of degrees 0 , 2 , 3 , 4 0,2,3,4 0 , 2 , 3 , 4 . Thus the required coefficient is
[ t 4 ] det ( L ) M P ( M , M ) = ( tr ( h 4 ) 2 − r 2 4 ) g + 2 γ 0 3 h + r h 2 − 2 h 4 + 2 W + P 2 + 2 h P 1 + h 2 Y − r 2 Y . (182) [t^4]\det(L)M\mathcal{P}(M,M)
=\left(\frac{\operatorname{tr}(h^4)}{2}-\frac{r^2}{4}\right)g+\frac{2\gamma_0}{3}h+rh^2-2h^4
+2W+P_2+2hP_1+h^2Y-\frac{r}{2}Y. \tag*{(182)} [ t 4 ] det ( L ) M P ( M , M ) = ( 2 tr ( h 4 ) − 4 r 2 ) g + 3 2 γ 0 h + r h 2 − 2 h 4 + 2 W + P 2 + 2 h P 1 + h 2 Y − 2 r Y . (182)
For the direct calculation, we return to the defining four-slot contraction. The coefficient [ t 4 ] L ∗ ψ [t^4]L^*\psi [ t 4 ] L ∗ ψ is ψ \psi ψ with h h h inserted in each of its four slots. We identify this term in Y 2 Y^2 Y 2 by contracting the two φ \varphi φ factors that share the middle index using (10) . The two metric terms each equal P ( h 2 , h 2 ) \mathcal{P}(h^2,h^2) P ( h 2 , h 2 ) , and the 4-form term is exactly the coefficient being computed. Thus
[ t 4 ] Z ( t ) = Y 2 − 2 P 2 . (183) [t^4]Z(t)=Y^2-2P_2. \tag*{(183)} [ t 4 ] Z ( t ) = Y 2 − 2 P 2 . (183)
Equating twice the sum of (A.25) and its transpose to (A.26) gives (A.23). □
Proof of Lemma 3.5. We work over R \mathbb{R} R . We combine the differentiated scalar c-contraction with Lemma A.7 to prove the first identity, then use Lemma A.1 for the second. Throughout the proof put
a = A ( h ) , b = B ( h ) , r = tr ( h 2 ) , γ 0 = C 0 ( h ) , γ φ = C φ ( h ) , Y = P ( h , h ) . a=A(h),\qquad b=B(h),\qquad r=\operatorname{tr}(h^2),\qquad\gamma_0=C_0(h),\qquad\gamma_\varphi=C_\varphi(h),\qquad Y=\mathcal{P}(h,h). a = A ( h ) , b = B ( h ) , r = tr ( h 2 ) , γ 0 = C 0 ( h ) , γ φ = C φ ( h ) , Y = P ( h , h ) .
The operators m h = A ( h , ⋅ ) m_h=A(h,\cdot) m h = A ( h , ⋅ ) and n h = B ( h , ⋅ ) n_h=B(h,\cdot) n h = B ( h , ⋅ ) are those defined in (86) ; in this proof we write m , n m,n m , n for them. The six quartic products and eight iterated products that occur below will be denoted by
Table A.27.
These are elements of W W W , with subscripts enumerating products rather than tensor components. We keep the mixed iterated products distinct throughout the calculation, so that their cancellations in the two eliminations can be followed explicitly.
Differentiated scalar c-contraction. We use the scalar identity (3.12), 8 ⟨ a , a ⟩ W − 12 ⟨ a , b ⟩ W + ⟨ b , b ⟩ W = 2 r 2 8\langle a,a\rangle_W-12\langle a,b\rangle_W+\langle b,b\rangle_W=2r^2 8 ⟨ a , a ⟩ W − 12 ⟨ a , b ⟩ W + ⟨ b , b ⟩ W = 2 r 2 . The factors of two in the differentiation come from the quadratic maps and their halved polarizations: D h a [ k ] = 2 m h k D h_a[k]=2m h_k D h a [ k ] = 2 m h k and D h b [ k ] = 2 n h k D h_b[k]=2n h_k D h b [ k ] = 2 n h k . We differentiate the scalar identity, use self-adjointness of m , n m,n m , n , and divide by four to obtain
8 m a − 6 m b − 6 n a + n b = 2 r h . (184) 8ma-6mb-6na+nb=2rh. \tag*{(184)} 8 ma − 6 mb − 6 na + nb = 2 r h . (184)
Applying m m m and n n n to this equation gives two relations among the iterated products:
R m : = 8 V 1 − 6 V 2 − 6 V 3 + V 4 − 2 r a = 0 , R_m:=8\mathcal{V}_1-6\mathcal{V}_2-6\mathcal{V}_3+\mathcal{V}_4-2ra=0, R m := 8 V 1 − 6 V 2 − 6 V 3 + V 4 − 2 r a = 0 ,
R n : = 8 V 5 − 6 V 6 − 6 V 7 + V 8 − 2 r b = 0. (185) R_n:=8\mathcal{V}_5-6\mathcal{V}_6-6\mathcal{V}_7+\mathcal{V}_8-2rb=0. \tag*{(185)} R n := 8 V 5 − 6 V 6 − 6 V 7 + V 8 − 2 r b = 0. (185)
We also differentiate (A.28) in a direction k ∈ W k\in W k ∈ W , held fixed during differentiation and set equal to a a a or b b b only afterwards. On the left we obtain
8 A ( k , a ) − 6 A ( k , b ) − 6 B ( k , a ) + B ( k , b ) + 16 m 2 k − 12 ( m n + n m ) k + 2 n 2 k . 8A(k,a)-6A(k,b)-6B(k,a)+B(k,b)+16m^2k-12(mn+nm)k+2n^2k. 8 A ( k , a ) − 6 A ( k , b ) − 6 B ( k , a ) + B ( k , b ) + 16 m 2 k − 12 ( mn + nm ) k + 2 n 2 k .
On the right, the derivative of 2 r h 2rh 2 r h is 4 ⟨ h , k ⟩ W h + 2 r k 4\langle h,k\rangle_W h+2rk 4 ⟨ h , k ⟩ W h + 2 r k . We now take k = a , b k=a,b k = a , b and use ⟨ h , a ⟩ W = γ 0 \langle h,a\rangle_W=\gamma_0 ⟨ h , a ⟩ W = γ 0 and ⟨ h , b ⟩ W = γ φ \langle h,b\rangle_W=\gamma_\varphi ⟨ h , b ⟩ W = γ φ to obtain
H a : = 8 A 1 − 6 A 2 − 6 A 4 + A 5 + 16 V 1 − 12 ( V 3 + V 5 ) + 2 V 7 − 4 γ 0 h − 2 r a = 0 , H_a:=8\mathcal{A}_1-6\mathcal{A}_2-6\mathcal{A}_4+\mathcal{A}_5+16\mathcal{V}_1-12(\mathcal{V}_3+\mathcal{V}_5)+2\mathcal{V}_7-4\gamma_0h-2ra=0, H a := 8 A 1 − 6 A 2 − 6 A 4 + A 5 + 16 V 1 − 12 ( V 3 + V 5 ) + 2 V 7 − 4 γ 0 h − 2 r a = 0 ,
H b : = 8 A 2 − 6 A 3 − 6 A 5 + A 6 + 16 V 2 − 12 ( V 4 + V 6 ) + 2 V 8 − 4 γ φ h − 2 r b = 0. (186) H_b:=8\mathcal{A}_2-6\mathcal{A}_3-6\mathcal{A}_5+\mathcal{A}_6+16\mathcal{V}_2-12(\mathcal{V}_4+\mathcal{V}_6)+2\mathcal{V}_8-4\gamma_\varphi h-2rb=0. \tag*{(186)} H b := 8 A 2 − 6 A 3 − 6 A 5 + A 6 + 16 V 2 − 12 ( V 4 + V 6 ) + 2 V 8 − 4 γ φ h − 2 r b = 0. (186)
To reduce these relations, we express two of the iterated terms in the product notation:
V 1 = A 1 + 1 7 r a − 1 7 γ 0 h , V 7 = A 5 + 8 7 r a − 8 7 γ 0 h . (187) \mathcal{V}_1=\mathcal{A}_1+\frac{1}{7}ra-\frac{1}{7}\gamma_0h,\qquad
\mathcal{V}_7=\mathcal{A}_5+\frac{8}{7}ra-\frac{8}{7}\gamma_0h. \tag*{(187)} V 1 = A 1 + 7 1 r a − 7 1 γ 0 h , V 7 = A 5 + 7 8 r a − 7 8 γ 0 h . (187)
The first identity is an elementary Jordan-associator calculation starting from m a = h 3 − r h / 7 − γ 0 g / 7 ma=h^3-rh/7-\gamma_0g/7 ma = h 3 − r h /7 − γ 0 g /7 . Applying m m m once more and removing the trace gives the claimed expression. For the second, we subtract (A.7) evaluated at ( h , a ; h , k ) (h,a;h,k) ( h , a ; h , k ) from its value at ( h , h ; a , k ) (h,h;a,k) ( h , h ; a , k ) . Full symmetry of C φ C_\varphi C φ and self-adjointness of n n n identify the two contractions as
T φ φ ( h , h ; a , k ) = ⟨ b , B ( a , k ) ⟩ W = ⟨ A 5 , k ⟩ W , T^{\varphi\varphi}(h,h;a,k)=\langle b,B(a,k)\rangle_W=\langle\mathcal{A}_5,k\rangle_W, T φφ ( h , h ; a , k ) = ⟨ b , B ( a , k ) ⟩ W = ⟨ A 5 , k ⟩ W ,
T φ φ ( h , a ; h , k ) = ⟨ n a , n k ⟩ W = ⟨ V 7 , k ⟩ W . T^{\varphi\varphi}(h,a;h,k)=\langle na,nk\rangle_W=\langle\mathcal{V}_7,k\rangle_W. T φφ ( h , a ; h , k ) = ⟨ na , nk ⟩ W = ⟨ V 7 , k ⟩ W .
The six cubic terms cancel by their full symmetry. The crossed traces cancel because h a = a h ha=ah ha = ah , and the alternating-vector terms vanish because η ( h , a ) = 0 \eta(h,a)=0 η ( h , a ) = 0 . Only the metric terms remain, so ⟨ A 5 − V 7 , k ⟩ W = − 8 ( r ⟨ a , k ⟩ W − γ 0 ⟨ h , k ⟩ W ) / 7 \langle\mathcal{A}_5-\mathcal{V}_7,k\rangle_W=-8(r\langle a,k\rangle_W-\gamma_0\langle h,k\rangle_W)/7 ⟨ A 5 − V 7 , k ⟩ W = − 8 ( r ⟨ a , k ⟩ W − γ 0 ⟨ h , k ⟩ W ) /7 . The second identity in (A.31) follows because k k k is arbitrary. Substituting (A.31) into H a = 0 H_a=0 H a = 0 yields
8 ( V 3 + V 5 ) = 16 A 1 − 4 A 2 − 4 A 4 + 2 A 5 + 12 7 r a − 40 7 γ 0 h . (188) 8(\mathcal{V}_3+\mathcal{V}_5)=16\mathcal{A}_1-4\mathcal{A}_2-4\mathcal{A}_4+2\mathcal{A}_5+\frac{12}{7}ra-\frac{40}{7}\gamma_0h. \tag*{(188)} 8 ( V 3 + V 5 ) = 16 A 1 − 4 A 2 − 4 A 4 + 2 A 5 + 7 12 r a − 7 40 γ 0 h . (188)
Trace removal and the first quartic identity. First we take traceless parts, expressing the matrix terms in the notation of (A.27):
( Y 2 ) 0 = A 3 − 2 7 r b , ( P ( h 2 , h 2 ) ) 0 = A 4 − 2 7 r a , (Y^2)_0=\mathcal{A}_3-\frac{2}{7}rb,\qquad
(\mathcal{P}(h^2,h^2))_0=\mathcal{A}_4-\frac{2}{7}ra, ( Y 2 ) 0 = A 3 − 7 2 r b , ( P ( h 2 , h 2 ) ) 0 = A 4 − 7 2 r a ,
( P ( h , h 3 ) ) 0 = V 5 − 1 7 γ 0 h + 1 7 r b , { h , P ( h , h 2 ) } 0 = 2 V 3 − 2 7 γ 0 h − 2 7 r a , (\mathcal{P}(h,h^3))_0=\mathcal{V}_5-\frac{1}{7}\gamma_0h+\frac{1}{7}rb,\qquad
\{h,\mathcal{P}(h,h^2)\}_0=2\mathcal{V}_3-\frac{2}{7}\gamma_0h-\frac{2}{7}ra, ( P ( h , h 3 ) ) 0 = V 5 − 7 1 γ 0 h + 7 1 r b , { h , P ( h , h 2 ) } 0 = 2 V 3 − 7 2 γ 0 h − 7 2 r a ,
{ h 2 , Y } 0 = 2 A 2 + 2 7 r b − 2 7 r a , ( h 4 ) 0 = A 1 + 2 7 r a . (189) \{h^2,Y\}_0=2\mathcal{A}_2+\frac{2}{7}rb-\frac{2}{7}ra,\qquad
(h^4)_0=\mathcal{A}_1+\frac{2}{7}ra. \tag*{(189)} { h 2 , Y } 0 = 2 A 2 + 7 2 r b − 7 2 r a , ( h 4 ) 0 = A 1 + 7 2 r a . (189)
For example, P ( h , h 2 ) = n a − γ 0 g / 7 − r h / 7 \mathcal{P}(h,h^{2})=na-\gamma_{0}g/7-rh/7 P ( h , h 2 ) = na − γ 0 g /7 − r h /7 and h 3 = m a + γ 0 g / 7 + r h / 7 h^{3}=ma+\gamma_{0}g/7+rh/7 h 3 = ma + γ 0 g /7 + r h /7 ; these follow from a = h 2 − r g / 7 a=h^{2}-rg/7 a = h 2 − r g /7 , tr P ( h , a ) = − tr ( h a ) = − γ 0 \operatorname{tr}\mathcal{P}(h,a)=-\operatorname{tr}(ha)=-\gamma_{0} tr P ( h , a ) = − tr ( ha ) = − γ 0 and P ( h , g ) = − h \mathcal{P}(h,g)=-h P ( h , g ) = − h . The other entries follow by expanding Y = b − r g / 7 Y=b-rg/7 Y = b − r g /7 and applying the trace projection. In particular the factors 1 / 7 1/7 1/7 enter before any elimination. Using the table to take the traceless part of (A.23), we obtain
A 3 = 6 A 4 + 8 ( V 5 + V 3 ) + 4 A 2 − 8 A 1 − 12 7 r a + 8 21 γ 0 h . (190) \mathcal{A}_{3}=6\mathcal{A}_{4}+8(\mathcal{V}_{5}+\mathcal{V}_{3})+4\mathcal{A}_{2}-8\mathcal{A}_{1}-\frac{12}{7}ra+\frac{8}{21}\gamma_{0}h. \tag*{(190)} A 3 = 6 A 4 + 8 ( V 5 + V 3 ) + 4 A 2 − 8 A 1 − 7 12 r a + 21 8 γ 0 h . (190)
Collecting the trace corrections cancels the r b rb r b terms and gives the displayed coefficients of r a ra r a and γ 0 h \gamma_{0}h γ 0 h . We eliminate the mixed iterated products by substituting (A.32). The mixed product A 2 \mathcal{A}_{2} A 2 and the r a ra r a terms cancel as follows:
A 3 = 6 A 4 + 16 A 1 − 4 A 2 − 4 A 4 + 2 A 5 + 12 7 r a − 40 7 γ 0 h + 4 A 2 − 8 A 1 − 12 7 r a + 8 21 γ 0 h = 8 A 1 + 2 A 4 + 2 A 5 − 16 3 γ 0 h . \begin{aligned}
\mathcal{A}_{3}&=6\mathcal{A}_{4}+16\mathcal{A}_{1}-4\mathcal{A}_{2}-4\mathcal{A}_{4}+2\mathcal{A}_{5}+\frac{12}{7}ra-\frac{40}{7}\gamma_{0}h\\
&\quad+4\mathcal{A}_{2}-8\mathcal{A}_{1}-\frac{12}{7}ra+\frac{8}{21}\gamma_{0}h\\
&=8\mathcal{A}_{1}+2\mathcal{A}_{4}+2\mathcal{A}_{5}-\frac{16}{3}\gamma_{0}h.
\end{aligned} A 3 = 6 A 4 + 16 A 1 − 4 A 2 − 4 A 4 + 2 A 5 + 7 12 r a − 7 40 γ 0 h + 4 A 2 − 8 A 1 − 7 12 r a + 21 8 γ 0 h = 8 A 1 + 2 A 4 + 2 A 5 − 3 16 γ 0 h .
Thus 8 A 1 − A 3 + 2 A 4 + 2 A 5 = 16 γ 0 h / 3 8\mathcal{A}_{1}-\mathcal{A}_{3}+2\mathcal{A}_{4}+2\mathcal{A}_{5}=16\gamma_{0}h/3 8 A 1 − A 3 + 2 A 4 + 2 A 5 = 16 γ 0 h /3 , the first identity of the lemma. The second quartic identity. For the second identity, we return to (A.2), proved above using the eight-term contraction (A.5). We first express that relation in traceless form, using
P ( Y , Y ) 0 = A 6 + 2 7 r b , \mathcal{P}(Y,Y)_{0}=\mathcal{A}_{6}+\frac{2}{7}rb, P ( Y , Y ) 0 = A 6 + 7 2 r b ,
( h Y h ) 0 = 2 V 2 − A 2 − 1 7 r b + 2 7 γ φ h − 1 7 r a , (hYh)_{0}=2\mathcal{V}_{2}-\mathcal{A}_{2}-\frac{1}{7}rb+\frac{2}{7}\gamma_{\varphi}h-\frac{1}{7}ra, ( hY h ) 0 = 2 V 2 − A 2 − 7 1 r b + 7 2 γ φ h − 7 1 r a ,
P ( h , h 3 ) 0 = V 5 − 1 7 γ 0 h + 1 7 r b , { h 2 , Y } 0 = 2 A 2 + 2 7 r b − 2 7 r a . \mathcal{P}(h,h^{3})_{0}=\mathcal{V}_{5}-\frac{1}{7}\gamma_{0}h+\frac{1}{7}rb,\qquad\{h^{2},Y\}_{0}=2\mathcal{A}_{2}+\frac{2}{7}rb-\frac{2}{7}ra. P ( h , h 3 ) 0 = V 5 − 7 1 γ 0 h + 7 1 r b , { h 2 , Y } 0 = 2 A 2 + 7 2 r b − 7 2 r a .
For the second equality, write J ( h 1 , h 2 ) = ( h 1 h 2 + h 2 h 1 ) / 2 J(h_{1},h_{2})=(h_{1}h_{2}+h_{2}h_{1})/2 J ( h 1 , h 2 ) = ( h 1 h 2 + h 2 h 1 ) /2 . The matrix identity and its traceless conversion are
h b h = 2 J ( h , J ( h , b ) ) − J ( h 2 , b ) , hbh=2J(h,J(h,b))-J(h^{2},b), hbh = 2 J ( h , J ( h , b )) − J ( h 2 , b ) ,
( h b h ) 0 = 2 m 2 b − A ( a , b ) − 1 7 r b + 2 7 γ φ h . (hbh)_{0}=2m^{2}b-A(a,b)-\frac{1}{7}rb+\frac{2}{7}\gamma_{\varphi}h. ( hbh ) 0 = 2 m 2 b − A ( a , b ) − 7 1 r b + 7 2 γ φ h .
Indeed, J ( h , b ) = m b + γ φ g / 7 J(h,b)=mb+\gamma_{\varphi}g/7 J ( h , b ) = mb + γ φ g /7 and J ( h 2 , b ) = A ( a , b ) + r b / 7 + ⟨ a , b ⟩ W g / 7 J(h^{2},b)=A(a,b)+rb/7+\langle a,b\rangle_{W}g/7 J ( h 2 , b ) = A ( a , b ) + r b /7 + ⟨ a , b ⟩ W g /7 . Substituting Y = b − r g / 7 Y=b-rg/7 Y = b − r g /7 then gives the displayed expression for ( h Y h ) 0 (hYh)_{0} ( hY h ) 0 . Substitution in (A.2) gives
R 3 : = A 6 − 8 V 2 + 20 A 2 + 8 V 5 − 36 7 γ φ h − 8 7 γ 0 h − 12 7 r a + 2 7 r b = 0. (191) R_{3}:=\mathcal{A}_{6}-8\mathcal{V}_{2}+20\mathcal{A}_{2}+8\mathcal{V}_{5}-\frac{36}{7}\gamma_{\varphi}h-\frac{8}{7}\gamma_{0}h-\frac{12}{7}ra+\frac{2}{7}rb=0. \tag*{(191)} R 3 := A 6 − 8 V 2 + 20 A 2 + 8 V 5 − 7 36 γ φ h − 7 8 γ 0 h − 7 12 r a + 7 2 r b = 0. (191)
We complete the elimination by writing the second quartic expression as a combination of the relations already proved. Let Q 1 = 8 A 1 − A 3 + 2 A 4 + 2 A 5 − 16 γ 0 h / 3 Q_{1}=8\mathcal{A}_{1}-\mathcal{A}_{3}+2\mathcal{A}_{4}+2\mathcal{A}_{5}-16\gamma_{0}h/3 Q 1 = 8 A 1 − A 3 + 2 A 4 + 2 A 5 − 16 γ 0 h /3 , which vanishes by the first identity. Using R m R_{m} R m , R n R_{n} R n , H a H_{a} H a , H b H_{b} H b from (A.29) and (A.30), we collect terms to obtain
16 A 2 + 2 A 3 − 8 A 4 + A 6 − 16 3 γ φ h = − 2 R m + 1 3 R n + H a − 1 6 H b − Q 1 + 7 6 R 3 . (192) 16\mathcal{A}_{2}+2\mathcal{A}_{3}-8\mathcal{A}_{4}+\mathcal{A}_{6}-\frac{16}{3}\gamma_{\varphi}h=-2R_{m}+\frac{1}{3}R_{n}+H_{a}-\frac{1}{6}H_{b}-Q_{1}+\frac{7}{6}R_{3}. \tag*{(192)} 16 A 2 + 2 A 3 − 8 A 4 + A 6 − 3 16 γ φ h = − 2 R m + 3 1 R n + H a − 6 1 H b − Q 1 + 6 7 R 3 . (192)
Collecting coefficients on the right cancels every iterated product and every r a ra r a , r b rb r b , γ 0 h \gamma_{0}h γ 0 h term, leaving the displayed quartic expression. All residuals on its right-hand side vanish, proving the second identity.
All identities are polynomial in h h h , so they extend to W C W_{\mathbb{C}} W C by complexification. Equivalently, their normalized polarizations give the symmetric four-input tensor identities among the components Sym β γ δ ϵ ( c σ τ α ( c t ) β γ σ ( c u ) δ ϵ τ ) \operatorname{Sym}_{\beta\gamma\delta\epsilon}(c^{\alpha}_{\sigma\tau}(c^{t})^{\sigma}_{\beta\gamma}(c^{u})^{\tau}_{\delta\epsilon}) Sym β γ δ ϵ ( c σ τ α ( c t ) β γ σ ( c u ) δ ϵ τ ) and Sym β γ δ ϵ ( c β γ δ α c ϵ ) \operatorname{Sym}_{\beta\gamma\delta\epsilon}(c_{\beta\gamma\delta}{}^{\alpha}c_{\epsilon}) Sym β γ δ ϵ ( c β γ δ α c ϵ ) . The proof has used only the G 2 G_{2} G 2 -contraction identities, the evaluated one-index c c c -contraction identity, the formal inverse of I + t h I+th I + t h , and the local triple-form reduction; no critical parameter relation or stabilizer result enters. ∙்கள்
Notation
The following table collects the notation used across sections. Its final column points to the definition or convention. Pairings are complex bilinear unless a conjugate or a Hermitian adjoint is written explicitly.
Table 5. Notation and its defining locations.
Table 5 (continued).
Table 5 (PDF p. 110).