This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Some observations on deformed Donaldson–Thomas connections

Kotaro Kawai Yanqi Lake Beijing Institute of Mathematical Sciences and Applications, No. 544, Hefangkou Village, Huaibei Town, Huairou District, Beijing, 101408, China Department of Mathematics, Osaka Metropolitan University, 3-3-138, Sugimoto, Sumiyoshi-ku, Osaka, 558-8585, Japan [email protected]
Abstract.

A deformed Donaldson–Thomas (dDT) connection is a Hermitian connection of a Hermitian line bundle over a G2G_{2}-manifold XX satisfying a certain nonlinear PDE. This is considered to be the mirror of a (co)associative cycle in the context of mirror symmetry. It can also be considered as an analogue of a G2G_{2}-instanton. In this paper, we see that some important observations that appear in other geometric problems are also found in the dDT case as follows.

(1) A dDT connection exists if a 7-manifold has full holonomy G2G_{2} and the G2G_{2}-structure is “sufficiently large”. (2) The dDT equation is described as the zero of a certain multi-moment map. (3) The gradient flow equation of a Chern-Simons type functional of Karigiannis and Leung, whose critical points are dDT connections, agrees with the Spin(7){\rm Spin}(7) version of the dDT equation on a cylinder with respect to a certain metric on a certain space. This can be considered as an analogue of the observation in instanton Floer homology for 3-manifolds.

Key words and phrases:
mirror symmetry, gauge theory, G2G_{2}-manifold, deformed Donaldson–Thomas connections
2020 Mathematics Subject Classification:
53C07, 58E15, 53D37
This work was supported by JSPS KAKENHI Grant Number JP21K03231 and MEXT Promotion of Distinctive Joint Research Center Program JPMXP0723833165.

1. Introduction

Let X7X^{7} be a 77-manifold with a G2\rm G_{2}-structure φΩ3(X)\varphi\in\Omega^{3}(X). For the definition of G2G_{2}-structures, see for example [kawai2021mirror, Section 2.2]. We use the same sign convention there. Denote by gg, vol\mathrm{vol} and \ast the induced Riemannian metric, volume form and Hodge star operator, respectively. Let (L,h)X(L,h)\to X be a smooth complex Hermitian line bundle over XX. We denote by 𝒜0\mathcal{A}_{0} the affine space of Hermitian connections on (L,h)(L,h). Given 𝒜0\nabla\in\mathcal{A}_{0}, we regard its curvature FF_{\nabla} as a 1\sqrt{-1}\mathbb{R}-valued closed 22-form on XX.

Definition 1.1.

A Hermitian connection 𝒜0\nabla\in\mathcal{A}_{0} satisfying

(1.1) 16F3+Fφ=0\frac{1}{6}F_{\nabla}^{3}+F_{\nabla}\wedge\ast\varphi=0

is called a deformed Donaldson-Thomas (dDT) connection.

DDT connections appeared in the context of mirror symmetry. They were introduced in [lee2009geometric] as “mirrors” of calibrated (associative) submanifolds. Historically, deformed Hermitian Yang–Mills (dHYM) connections were introduced first in [lyz2000FM] as “mirrors” of special Lagrangian submanifolds. There is also a similar notion of dDT connections for a manifold with a Spin(7){\rm Spin}(7)-structure ([lee2009geometric, kawai2021FM]). As the names indicate, dDT connections can also be considered as analogues of Donaldson–Thomas connections (G2G_{2}-instantons).

Thus it is natural to expect that dDT connections would have similar properties to associative submanifolds and G2G_{2}-instantons. We show that it is indeed the case in [kawai2020deformation, kawai2021mirror]. For example, the moduli space of dDT connections is b1b^{1}-dimensional and canonically orientable if we perturb the G2G_{2}-structure. Any dDT connection on a compact G2G_{2}-manifold is a global minimizer of the “mirror volume” and its value is topological by the “mirror” of associator equality. We could also prove similar statements in the Spin(7){\rm Spin}(7) case in [kawai2021deformationSpin(7), kawai2021mirror]. Moreover, dDT connections are given by critical points of the Chern-Simons type functional in [karigiannis2009hodge, Theorem 5.13]. The variational characterization is known only for the G2G_{2} case, and no such characterization is known for the Spin(7){\rm Spin}(7) case.

This paper is organized as follows. In Section 2, we study the existence of a dDT connection. Known examples of dDT connections are either trivial or constructed in [lotay2020examples, fowdar2022examples], and are very few in number. So it would be important to consider the existence problem. We first see that the formal “large radius limit” of the defining equation of dDT connections is that of G2G_{2}-instantons. Thus it is natural to expect that dDT connections for a “sufficiently large” G2G_{2}-structure will behave like G2G_{2}-instantons. Moreover, it is known that any complex Hermitian line bundle admits a G2G_{2}-instanton on a compact holonomy G2G_{2}-manifold. Then we show the following from these facts.

Theorem 1.2 (Theorem 2.2).

Suppose that (X,φ)(X,\varphi) is a compact holonomy G2G_{2}-manifold. Let (L,h)X(L,h)\to X be a smooth complex Hermitian line bundle over XX. If the G2G_{2}-structure is “sufficiently large”, there exists a dDT connection.

In Section 3, we formulate the dDT equation in terms of a multi-moment map. The multi-moment map is a generalization of the moment map introduced in [Madsen2012multi, Madsen2013closed]. The dHYM equation is described as the zero of a certain moment map on a infinite dimensional symplectic manifold ([collins2021moment, Section 2], [collins2021survey, Section 2.1]). Analogously, we show that the dDT equation is described as the zero of a certain multi-moment map.

Theorem 1.3 (Theorem 3.4).

The dDT equation is described as the zero of a certain multi-moment map.

In Section 4, we study the gradient flow of the Karigiannis-Leung functional introduced in [karigiannis2009hodge] whose critical points are dDT connections. It is known that the gradient flow equation of the Chern-Simons functional on an oriented 3-manifold X3X^{3} agrees with the ASD equation on ×X3\mathbb{R}\times X^{3}. This is an important observation in instanton Floer homology for 3-manifolds. We show that there is an analogous relation between dDT equations for G2G_{2}- and Spin(7){\rm Spin}(7)-manifolds using the Karigiannis-Leung functional. This will establish a new link between 3, 4-manifold theory and G2G_{2}-, Spin(7){\rm Spin}(7)-geometry, and we might define analogues of instanton Floer homology using dDT connections.

Theorem 1.4 (Theorem 4.3).

The gradient flow equation of a Chern-Simons type functional of Karigiannis and Leung, whose critical points are dDT connections, agrees with the Spin(7){\rm Spin}(7) version of the dDT equation on a cylinder with respect to a certain metric on a certain space.

2. Large radius limit

In this section, we show the existence of a dDT connection if a 7-manifold has full holonomy G2G_{2} and the G2G_{2}-structure is “sufficiently large”.

Suppose that (X,φ)(X,\varphi) is a compact holonomy G2G_{2}-manifold. Let (L,h)X(L,h)\to X be a smooth complex Hermitian line bundle over XX. Set

𝒜0={Hermitian connections of (L,h)}=0+1Ω1idL,\displaystyle\mathcal{A}_{0}=\{\,\mbox{Hermitian connections of }(L,h)\,\}=\nabla_{0}+{\sqrt{-1}}\Omega^{1}\cdot\mathrm{id}_{L},

where 0𝒜0\nabla_{0}\in\mathcal{A}_{0} is any fixed connection and Ω1\Omega^{1} is the space of 1-forms on XX. Denote by 𝒢U\mathcal{G}_{U} the group of unitary gauge transformations of (L,h)(L,h), which acts on 𝒜0\mathcal{A}_{0}. Explicitly,

𝒢U={fidLfΩ0,|f|=1}C(X,S1),\mathcal{G}_{U}=\{\,f\cdot\mathrm{id}_{L}\mid f\in\Omega^{0}_{\mathbb{C}},\ |f|=1\,\}\cong C^{\infty}(X,S^{1}),

where Ω0\Omega^{0}_{\mathbb{C}} is the space of \mathbb{C}-valued smooth functions, and the action 𝒢U×𝒜0𝒜0\mathcal{G}_{U}\times\mathcal{A}_{0}\rightarrow\mathcal{A}_{0} is defined by (λ,)λ:=λ1λ(\lambda,\nabla)\mapsto\lambda^{*}\nabla:=\lambda^{-1}\circ\nabla\circ\lambda. When λ=fidL\lambda=f\cdot\mathrm{id}_{L} for fC(X,S1)f\in C^{\infty}(X,S^{1}), we have

(2.1) λ=λ1λ=+f1dfidL.\displaystyle\lambda^{*}\nabla=\lambda^{-1}\circ\nabla\circ\lambda=\nabla+f^{-1}df\cdot\mathrm{id}_{L}.

Thus the 𝒢U\mathcal{G}_{U}-orbit through 𝒜0{\nabla}\in\mathcal{A}_{0} is given by +𝒦UidL{\nabla}+\mathcal{K}_{U}\cdot\mathrm{id}_{L}, where

(2.2) 𝒦U:={f1df1Ω1|fΩ0,|f|=1}.\mathcal{K}_{U}:=\mathopen{}\mathclose{{}\left\{f^{-1}df\in{\sqrt{-1}}\Omega^{1}\;\middle|\;f\in\Omega^{0}_{\mathbb{C}},\ |f|=1}\right\}.

Note that the curvature 2-form FF_{\nabla} is invariant under the action of 𝒢U\mathcal{G}_{U}.

Consider the family of G2G_{2}-structures

{φr:=r3φ}r>0,\{\varphi_{r}:=r^{3}\varphi\}_{r>0},

all of which induce holonomy G2G_{2} metrics. The defining equation of dDT connections with respect to φr\varphi_{r} is given by

0=r():=16F3+r4Fφ.0=\mathcal{F}_{r}({\nabla}):=\frac{1}{6}F_{\nabla}^{3}+r^{4}F_{\nabla}\wedge*\varphi.

Thus, formally taking the ”large radius limit”, which means the leading behaviour of r()\mathcal{F}_{r}({\nabla}) as rr\to\infty, we obtain

Fφ=0.F_{\nabla}\wedge*\varphi=0.

This is exactly the defining equation of G2G_{2}-instantons. Thus it is natural to expect that dDT connections for a sufficiently large G2G_{2}-structure will behave like G2G_{2}-instantons. The following would be well-known for G2G_{2}-instantons on a smooth complex Hermitian line bundle, but we give the proof for completeness.

Lemma 2.1.

On a compact holonomy G2G_{2}-manifold (X7,φ)(X^{7},\varphi), there is a unique G2G_{2}-instanton on a smooth complex Hermitian line bundle LXL\to X up to the action of 𝒢U\mathcal{G}_{U}.

Proof.

For any 𝒜0{\nabla}\in\mathcal{A}_{0}, we have dF=0dF_{\nabla}=0. So it defines a cohomology class [F]1H2(X,)[F_{\nabla}]\in{\sqrt{-1}}H^{2}(X,\mathbb{R}), which is known to be equal to 2π1c1(L)-2\pi{\sqrt{-1}}c_{1}(L). Then there exists a 1-form α1Ω1\alpha\in{\sqrt{-1}}\Omega^{1} such that F+dαF_{\nabla}+d\alpha is harmonic by Hodge theory.

Denote by ΩkΩk\Omega^{k}_{\ell}\subset\Omega^{k} the subspace of the space of kk-forms corresponding to the \ell-dimensional irreducible representation of G2G_{2}. For more details, see for example [kawai2021mirror, Section 2.2]. Denote by k\mathcal{H}^{k} the space of harmonic kk-forms on XX and set k=kΩk\mathcal{H}^{k}_{\ell}=\mathcal{H}^{k}\cap\Omega^{k}_{\ell}. Then by [joyce2000compact, Theorem 10.2.4], we have 7271=1={0}\mathcal{H}^{2}_{7}\cong\mathcal{H}^{1}_{7}=\mathcal{H}^{1}=\{0\}. Thus we have

F+αidL=F+dα12=172142=1142,F_{{\nabla}+\alpha\cdot\mathrm{id}_{L}}=F_{\nabla}+d\alpha\in{\sqrt{-1}}\mathcal{H}^{2}={\sqrt{-1}}\mathcal{H}^{2}_{7}\oplus\mathcal{H}^{2}_{14}={\sqrt{-1}}\mathcal{H}^{2}_{14},

which implies that F+αidLφ=0F_{{\nabla}+\alpha\cdot\mathrm{id}_{L}}\wedge*\varphi=0.

If =+(α+α)idL{\nabla}^{\prime}={\nabla}+(\alpha+\alpha^{\prime})\cdot\mathrm{id}_{L} for α1Ω1\alpha^{\prime}\in{\sqrt{-1}}\Omega^{1} is also a G2G_{2}-instanton, we have 0=Fφ=dαφ0=F_{{\nabla}^{\prime}}\wedge*\varphi=d\alpha^{\prime}\wedge*\varphi, which is equivalent to

(2.3) dα=(dαφ)=d(αφ).\displaystyle-d\alpha^{\prime}=*(d\alpha^{\prime}\wedge\varphi)=*d(\alpha^{\prime}\wedge\varphi).

Since dΩ1dΩ3={0}d\Omega^{1}\cap d^{*}\Omega^{3}=\{0\}, we have dα=0d\alpha^{\prime}=0. Since H1(X,)={0}H^{1}(X,\mathbb{R})=\{0\} by [joyce2000compact, Theorem 10.2.4] again and 1{\sqrt{-1}}\mathbb{R}-valued exact 1-forms are contained in 𝒦U\mathcal{K}_{U}, the G2G_{2}-instanton is unique up to the action of 𝒢U\mathcal{G}_{U}. ∎

Using this, we can show the following.

Theorem 2.2.

Suppose that (X,φ)(X,\varphi) is a compact holonomy G2G_{2}-manifold. Let (L,h)X(L,h)\to X be a smooth complex Hermitian line bundle over XX. Then for sufficiently large r>0r>0, there exists a dDT connection with respect to φr\varphi_{r}.

Proof.

Define a map :[0,1]×𝒜01dΩ5\mathcal{F}:[0,1]\times\mathcal{A}_{0}\to{\sqrt{-1}}d\Omega^{5} by

(s,)=s46F3+Fφ.\mathcal{F}(s,{\nabla})=\frac{s^{4}}{6}F_{\nabla}^{3}+F_{\nabla}\wedge*\varphi.

Then (0,)1(0)/𝒢U\mathcal{F}(0,\cdot)^{-1}(0)/\mathcal{G}_{U}, which is a point by Lemma 2.1, is the moduli space of G2G_{2}-instantons with respect to φ\varphi and (s,)1(0)/𝒢U\mathcal{F}(s,\cdot)^{-1}(0)/\mathcal{G}_{U} for s0s\neq 0 is the moduli space of dDT connections with respect to φ1/s\varphi_{1/s}.

We want to apply the implicit function theorem to show the statement. Fix a G2G_{2}-instanton 0(0,)1(0){\nabla}_{0}\in\mathcal{F}(0,\cdot)^{-1}(0), whose existence is guaranteed by Lemma 2.1. Denote by the linearization (d)(0,0):1Ω11dΩ5(d\mathcal{F})_{(0,{\nabla}_{0})}:\mathbb{R}\oplus{\sqrt{-1}}\Omega^{1}\to{\sqrt{-1}}d\Omega^{5} of \mathcal{F} at (0,0)(0,{\nabla}_{0}). Then we have

(d)(0,0)(0,1b)=1dbφ.(d\mathcal{F})_{(0,{\nabla}_{0})}(0,{\sqrt{-1}}b)={\sqrt{-1}}db\wedge*\varphi.
Lemma 2.3.

We have

ker(d)(0,0)=1dΩ0,Im(d)(0,0)=1dΩ5.\ker(d\mathcal{F})_{(0,{\nabla}_{0})}=\mathbb{R}\oplus{\sqrt{-1}}d\Omega^{0},\qquad\mathrm{Im}{(d\mathcal{F})_{(0,{\nabla}_{0})}}={\sqrt{-1}}d\Omega^{5}.
Proof.

The first equation is proved as in \tagform@2.3. For the second equation, the Hodge decomposition implies that dΩ2=ddΩ1d^{*}\Omega^{2}=d^{*}d\Omega^{1}. For any bΩ1b\in\Omega^{1}, we have

ddb=d(db+(φdb))dΩ72,d^{*}db=d^{*}(db+*(\varphi\wedge db))\in d^{*}\Omega^{2}_{7},

where we use the fact that φ\varphi is closed. This implies that dΩ2=dΩ72d^{*}\Omega^{2}=d^{*}\Omega^{2}_{7}. Then

dΩ5=dΩ2=dΩ72=dΩ75.d\Omega^{5}=*d^{*}\Omega^{2}=*d^{*}\Omega^{2}_{7}=d\Omega^{5}_{7}.

Since Ω75\Omega^{5}_{7} is spanned by bφb\wedge*\varphi for bΩ1b\in\Omega^{1}, the proof is completed. ∎

By the Hodge decomposition and H1(X,)={0}H^{1}(X,\mathbb{R})=\{0\}, we have Ω1=dΩ0dΩ2\Omega^{1}=d\Omega^{0}\oplus d^{*}\Omega^{2}. By this and Lemma 2.3, we see that (d)(0,0)|1dΩ2:1dΩ21dΩ5(d\mathcal{F})_{(0,{\nabla}_{0})}|_{{\sqrt{-1}}d^{*}\Omega^{2}}:{\sqrt{-1}}d^{*}\Omega^{2}\to{\sqrt{-1}}d\Omega^{5} is an isomorphism. Hence, we can apply the implicit function theorem (after the Banach completion) and we see that (s,)1(0)\mathcal{F}(s,\cdot)^{-1}(0)\neq\emptyset for sufficiently small ss.

Finally, we explain how to recover the regularity of elements in (s,)1(0)\mathcal{F}(s,\cdot)^{-1}(0) after the Banach completion. Since the curvature is invariant under the addition of closed 1-forms, there exists asΩ1a_{s}\in\Omega^{1} such that

(s*_{s}) (s,0+1asidL)=0,das=0\displaystyle\mathcal{F}(s,{\nabla}_{0}+{\sqrt{-1}}a_{s}\cdot\mathrm{id}_{L})=0,\qquad d^{*}a_{s}=0

for sufficiently small ss. In particular, (0)(*_{0}) is given by da0φ=da0=0da_{0}\wedge*\varphi=d^{*}a_{0}=0, which is an overdetermined elliptic equation. To be overdetermined elliptic is an open condition, so we see that (s)(*_{s}) is also overdetermined elliptic for sufficiently small ss. Hence we can find a smooth element in (s,)1(0)\mathcal{F}(s,\cdot)^{-1}(0) around (0,0)(0,{\nabla}_{0}) and the proof is completed. ∎

3. The multi-moment map

It is known that there is a moment map picture in the dHYM case. In particular, the dHYM equation is described as the zero of a certain moment map on a infinite dimensional symplectic manifold. See for example [collins2021moment, Section 2] or the survey article [collins2021survey, Section 2.1]. Analogously, we show that the dDT equation is described as the zero of a certain multi-moment map. First, recall the definition of the multi-moment map in [Madsen2012multi, Madsen2013closed].

Definition 3.1.

Let XX be a smooth manifold and cΩ3c\in\Omega^{3} be a closed 3-form on XX. Suppose that a Lie group GG acts on XX preserving cc. Denote by 𝔤{\mathfrak{g}} the Lie algebra of GG and set

𝒫𝔤=ker(L:Λ2𝔤𝔤)Λ2𝔤,\mathcal{P}_{\mathfrak{g}}=\ker(L:\Lambda^{2}{\mathfrak{g}}\to{\mathfrak{g}})\subset\Lambda^{2}{\mathfrak{g}},

where LL is the linear map induced by the Lie bracket. (Note that 𝒫𝔤=Λ2𝔤\mathcal{P}_{\mathfrak{g}}=\Lambda^{2}{\mathfrak{g}} if GG is abelian.) Denote by uu^{*} the vector field on XX generated by u𝔤u\in{\mathfrak{g}}. For a two vector p=jujvjΛ2𝔤p=\sum_{j}u_{j}\wedge v_{j}\in\Lambda^{2}{\mathfrak{g}}, set

p=jujvj,i(p)c=jc(uj,vj,).p^{*}=\sum_{j}u_{j}^{*}\wedge v_{j}^{*},\qquad i(p^{*})c=\sum_{j}c(u_{j}^{*},v_{j}^{*},\cdot).

Denote by ,:Λ2𝔤×Λ2𝔤\langle\cdot,\cdot\rangle:\Lambda^{2}{\mathfrak{g}}^{*}\times\Lambda^{2}{\mathfrak{g}}\to\mathbb{R} the canonical pairing.

Then a map ν:M𝒫𝔤\nu:M\to\mathcal{P}_{\mathfrak{g}}^{*} is called a multi-moment map if it is GG-equivariant and satisfies

dν,p=i(p)cd\langle\nu,p\rangle=i(p^{*})c

for any p𝒫𝔤p\in\mathcal{P}_{\mathfrak{g}}.

Let XX be a compact 7-manifold with a coclosed G2G_{2}-structure φ\varphi (dφ=0d*\varphi=0) and (L,h)X(L,h)\to X be a smooth complex Hermitian line bundle over XX. Let 𝒜0\mathcal{A}_{0} be the space of Hermitian connections of (L,h)(L,h). Define a map G2:𝒜01Ω6\mathcal{F}_{G_{2}}:\mathcal{A}_{0}\to{\sqrt{-1}}\Omega^{6} by

G2()=16F3+Fφ.\mathcal{F}_{G_{2}}({\nabla})=\frac{1}{6}F_{\nabla}^{3}+F_{\nabla}\wedge*\varphi.

Then the space of dDT connections is given by G21(0)\mathcal{F}_{G_{2}}^{-1}(0). Denote by 𝒢U\mathcal{G}_{U} the group of unitary gauge transformations of (L,h)(L,h) acting 𝒜0\mathcal{A}_{0} canonically as in \tagform@2.1. Since 𝒢U=C(X,S1)\mathcal{G}_{U}=C^{\infty}(X,S^{1}), the Lie algebra 𝔤U{\mathfrak{g}}_{U} of 𝒢U\mathcal{G}_{U} is identified with the space 1Ω0{\sqrt{-1}}\Omega^{0} of 1{\sqrt{-1}}\mathbb{R}-valued functions on XX. Note that 𝒫𝔤=Λ2𝔤\mathcal{P}_{\mathfrak{g}}=\Lambda^{2}{\mathfrak{g}} since 𝒢U\mathcal{G}_{U} is abelian. Define a 3-form ΘΩ3(𝒜0)\Theta\in\Omega^{3}(\mathcal{A}_{0}) on 𝒜0\mathcal{A}_{0} by

Θ(α1,α2,α3)=1Xα1α2α3(12F2+φ),\Theta_{\nabla}(\alpha_{1},\alpha_{2},\alpha_{3})={\sqrt{-1}}\int_{X}\alpha_{1}\wedge\alpha_{2}\wedge\alpha_{3}\wedge\mathopen{}\mathclose{{}\left(\frac{1}{2}F_{\nabla}^{2}+*\varphi}\right),

where 𝒜0{\nabla}\in\mathcal{A}_{0} and α1,α2,α31Ω1=T𝒜0\alpha_{1},\alpha_{2},\alpha_{3}\in{\sqrt{-1}}\Omega^{1}=T_{\nabla}\mathcal{A}_{0}. We first show the following as required in Definition 3.1.

Lemma 3.2.

The 3-form Θ\Theta is 𝒢U\mathcal{G}_{U}-invariant and closed.

Proof.

Take any 𝒜0{\nabla}\in\mathcal{A}_{0}, λ=fidL𝒢U\lambda=f\cdot\mathrm{id}_{L}\in\mathcal{G}_{U}, where fC(X,S1)f\in C^{\infty}(X,S^{1}), and α1,α2,α3,α41Ω1T𝒜0\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{4}\in{\sqrt{-1}}\Omega^{1}\cong T_{\nabla}\mathcal{A}_{0}. Identify αj\alpha_{j} with a vector field on 𝒜0\mathcal{A}_{0} by

(αj)~=ddt(~+tαjidL)|t=0(\alpha_{j})_{\widetilde{\nabla}}=\mathopen{}\mathclose{{}\left.\frac{d}{dt}\mathopen{}\mathclose{{}\left(\widetilde{\nabla}+t\alpha_{j}\cdot\mathrm{id}_{L}}\right)}\right|_{t=0}

for ~𝒜0\widetilde{\nabla}\in\mathcal{A}_{0}. We first show the 𝒢U\mathcal{G}_{U}-invariance of Θ\Theta. That is,

(3.1) Θλ(λ(α1),λ(α2),λ(α3))=Θ(α1,α2,α3).\displaystyle\Theta_{\lambda^{*}{\nabla}}\mathopen{}\mathclose{{}\left(\lambda_{*}(\alpha_{1}),\lambda_{*}(\alpha_{2}),\lambda_{*}(\alpha_{3})}\right)=\Theta_{{\nabla}}\mathopen{}\mathclose{{}\left(\alpha_{1},\alpha_{2},\alpha_{3}}\right).

By \tagform@2.1, we compute

(3.2) λ(αj)=λddt(+tαjidL)|t=0=ddt(+(tαj+f1df)idL)|t=0=(αj)λ.\displaystyle\lambda_{*}(\alpha_{j})_{\nabla}=\lambda_{*}\mathopen{}\mathclose{{}\left.\frac{d}{dt}\mathopen{}\mathclose{{}\left({\nabla}+t\alpha_{j}\cdot\mathrm{id}_{L}}\right)}\right|_{t=0}=\mathopen{}\mathclose{{}\left.\frac{d}{dt}\mathopen{}\mathclose{{}\left({\nabla}+(t\alpha_{j}+f^{-1}df)\cdot\mathrm{id}_{L}}\right)}\right|_{t=0}=(\alpha_{j})_{\lambda^{*}{\nabla}}.

Since Fλ=FF_{\lambda^{*}{\nabla}}=F_{\nabla}, we obtain \tagform@3.1.

Next, we show the closedness of Θ\Theta. Note that [αi,αj]=0[\alpha_{i},\alpha_{j}]=0. Then it follows that

(3.3) dΘ(α1,α2,α3,α4)=\displaystyle d\Theta(\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{4})= α1(Θ(α2,α3,α4))α2(Θ(α1,α3,α4))\displaystyle\alpha_{1}\mathopen{}\mathclose{{}\left(\Theta(\alpha_{2},\alpha_{3},\alpha_{4})}\right)-\alpha_{2}\mathopen{}\mathclose{{}\left(\Theta(\alpha_{1},\alpha_{3},\alpha_{4})}\right)
(3.4) +α3(Θ(α1,α2,α4))α4(Θ(α1,α2,α3)).\displaystyle+\alpha_{3}\mathopen{}\mathclose{{}\left(\Theta(\alpha_{1},\alpha_{2},\alpha_{4})}\right)-\alpha_{4}\mathopen{}\mathclose{{}\left(\Theta(\alpha_{1},\alpha_{2},\alpha_{3})}\right).

Since

(3.5) αi(Θ(αj,αk,α))=\displaystyle\alpha_{i}\mathopen{}\mathclose{{}\left(\Theta(\alpha_{j},\alpha_{k},\alpha_{\ell})}\right)_{\nabla}= 1ddtXαjαkα(12F+tαiidL2+φ)|t=0\displaystyle{\sqrt{-1}}\mathopen{}\mathclose{{}\left.\frac{d}{dt}\int_{X}\alpha_{j}\wedge\alpha_{k}\wedge\alpha_{\ell}\wedge\mathopen{}\mathclose{{}\left(\frac{1}{2}F_{{\nabla}+t\alpha_{i}\cdot\mathrm{id}_{L}}^{2}+*\varphi}\right)}\right|_{t=0}
(3.6) =\displaystyle= 1XαjαkαdαiF,\displaystyle{\sqrt{-1}}\int_{X}\alpha_{j}\wedge\alpha_{k}\wedge\alpha_{\ell}\wedge d\alpha_{i}\wedge F_{\nabla},

we have

(dΘ)(α1,α2,α3,α4)=1Xd(α1α2α3α4F)=0,(d\Theta)_{\nabla}(\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{4})={\sqrt{-1}}\int_{X}d(\alpha_{1}\wedge\alpha_{2}\wedge\alpha_{3}\wedge\alpha_{4}\wedge F_{\nabla})=0,

which implies that dΘ=0d\Theta=0. ∎

We also need the following lemma.

Lemma 3.3.

We have

Ω1={j=1Nf1jdf2j|N,f1j,f2jΩ0}.\Omega^{1}=\mathopen{}\mathclose{{}\left\{\sum_{j=1}^{N}f^{j}_{1}df^{j}_{2}\;\middle|\;N\in\mathbb{N},f^{j}_{1},f^{j}_{2}\in\Omega^{0}}\right\}.
Proof.

Take any 1-form αΩ1\alpha\in\Omega^{1}. We first show that for any xXx\in X, there exists an open neighborhood UxU_{x} of xx and smooth functions {f~x,j1,f~x,j2}j=17\{\widetilde{f}_{x,j}^{1},\widetilde{f}_{x,j}^{2}\}_{j=1}^{7} on XX such that

(3.7) α|Ux=j=17f~x,j1df~x,j2|Ux.\displaystyle\alpha|_{U_{x}}=\sum^{7}_{j=1}\widetilde{f}_{x,j}^{1}\ d\widetilde{f}_{x,j}^{2}|_{U_{x}}.

Indeed, take any local coordinates (V,(x1,,x7))(V,(x^{1},\cdots,x^{7})) of xx and set

α|V=j=17αjdxj.\alpha|_{V}=\sum^{7}_{j=1}\alpha_{j}dx^{j}.

We can take a cutoff function hh such that hh has compact support in VV and h=1h=1 on an open neighborhood UxU_{x} of xx. Then setting f~x,j1=hαj\widetilde{f}_{x,j}^{1}=h\alpha_{j} and f~x,j2=hxj\widetilde{f}_{x,j}^{2}=hx_{j}, which are smooth functions on XX, we obtain \tagform@3.7.

Since {Ux}xX\{U_{x}\}_{x\in X} is an open cover of XX and XX is compact, there exists x1,,xNXx_{1},\cdots,x_{N}\in X such that {Uxp}p=1N\{U_{x_{p}}\}_{p=1}^{N} covers XX. Denote by {hp}p=1N\{h_{p}\}_{p=1}^{N} the partition of unity subordinate to {Uxp}p=1N\{U_{x_{p}}\}_{p=1}^{N}. Set

fp,j1=hpf~xp,j1fp,j2=f~xp,j2.f_{p,j}^{1}=h_{p}\widetilde{f}_{x_{p},j}^{1}\qquad f_{p,j}^{2}=\widetilde{f}_{x_{p},j}^{2}.

Then we have α=p=1Nj=17fp,j1dfp,j2\alpha=\sum^{N}_{p=1}\sum^{7}_{j=1}f_{p,j}^{1}df_{p,j}^{2}. Indeed, take any xXx\in X. We may assume that xUx1Uxkx\in U_{x_{1}}\cap\cdots\cap U_{x_{k}} and xUpx\not\in U_{p} for p=k+1,,Np=k+1,\cdots,N. Then j=17(f~xp,j1df~xp,j2)x=αx\sum^{7}_{j=1}\mathopen{}\mathclose{{}\left(\widetilde{f}_{x_{p},j}^{1}\ d\widetilde{f}_{x_{p},j}^{2}}\right)_{x}=\alpha_{x} for p=1,,kp=1,\cdots,k by \tagform@3.7 and hp(x)=0h_{p}(x)=0 for p=k+1,,Np=k+1,\cdots,N. Hence

p=1Nj=17(fp,j1dfp,j2)x=p=1khp(x)j=17(f~xp,j1df~xp,j2)x=p=1khp(x)αx=p=1Nhp(x)αx=αx.\sum^{N}_{p=1}\sum^{7}_{j=1}\mathopen{}\mathclose{{}\left(f_{p,j}^{1}df_{p,j}^{2}}\right)_{x}=\sum^{k}_{p=1}h_{p}(x)\sum^{7}_{j=1}\mathopen{}\mathclose{{}\left(\widetilde{f}_{x_{p},j}^{1}\ d\widetilde{f}_{x_{p},j}^{2}}\right)_{x}=\sum^{k}_{p=1}h_{p}(x)\alpha_{x}=\sum^{N}_{p=1}h_{p}(x)\alpha_{x}=\alpha_{x}.

Denote by Z6Z^{6} the space of closed 6-forms on XX. Define a map ιZ6:Z6Λ2𝔤U\iota_{Z^{6}}:Z^{6}\to\Lambda^{2}{\mathfrak{g}}_{U}^{*} by

ιZ6(ξ)(f1,f2)=Xξ12(f1df2f2df1)=Xξf1df2\iota_{Z^{6}}(\xi)(f_{1},f_{2})=\int_{X}\xi\wedge\frac{1}{2}(f_{1}df_{2}-f_{2}df_{1})=\int_{X}\xi\wedge f_{1}df_{2}

for ξZ6\xi\in Z^{6} and f1,f21Ω0=𝔤Uf_{1},f_{2}\in{\sqrt{-1}}\Omega^{0}={\mathfrak{g}}_{U}.

Theorem 3.4.

Define a 𝒢U\mathcal{G}_{U}-invariant map ν:𝒜0Λ2𝔤U\nu:\mathcal{A}_{0}\to\Lambda^{2}{\mathfrak{g}}_{U}^{*} by

ν()=ιZ6(1G2()).\nu({\nabla})=\iota_{Z^{6}}({\sqrt{-1}}\mathcal{F}_{G_{2}}({\nabla})).

Then we have dν,p=i(p)Θd\langle\nu,p\rangle=i(p^{*})\Theta for any pΛ2𝔤Up\in\Lambda^{2}{\mathfrak{g}}_{U}.

Since we assume that dφ=0d*\varphi=0, we see that 1G2()Z6{\sqrt{-1}}\mathcal{F}_{G_{2}}({\nabla})\in Z^{6} for any 𝒜0{\nabla}\in\mathcal{A}_{0}. By Lemma 3.3, ιZ6\iota_{Z^{6}} is injective. Hence we have ν1(0)=G21(0)\nu^{-1}(0)=\mathcal{F}_{G_{2}}^{-1}(0). In this sense, we can regard the dDT equation as the zero of a multi-moment map.

Proof.

First note that the vector field ff^{*} generated by f1Ω0=𝔤Uf\in{\sqrt{-1}}\Omega^{0}={\mathfrak{g}}_{U} is given by

f=ddt(etf)|t=0=ddt(+etfdetfidL)|t=0=dff^{*}_{\nabla}=\mathopen{}\mathclose{{}\left.\frac{d}{dt}(e^{tf})^{*}{\nabla}}\right|_{t=0}=\mathopen{}\mathclose{{}\left.\frac{d}{dt}\mathopen{}\mathclose{{}\left({\nabla}+e^{-tf}de^{tf}\cdot\mathrm{id}_{L}}\right)}\right|_{t=0}=df

at 𝒜0{\nabla}\in\mathcal{A}_{0}. Hence for any f1,f21Ω0=𝔤Uf_{1},f_{2}\in{\sqrt{-1}}\Omega^{0}={\mathfrak{g}}_{U} and α1Ω1=T𝒜0\alpha\in{\sqrt{-1}}\Omega^{1}=T_{\nabla}\mathcal{A}_{0}, we have

(3.8) Θ((f1),(f2),α)\displaystyle\Theta_{\nabla}((f_{1}^{*})_{\nabla},(f_{2}^{*})_{\nabla},\alpha)
(3.9) =\displaystyle= 1Xdf1df2α(12F2+φ)\displaystyle{\sqrt{-1}}\int_{X}df_{1}\wedge df_{2}\wedge\alpha\wedge\mathopen{}\mathclose{{}\left(\frac{1}{2}F_{\nabla}^{2}+*\varphi}\right)
(3.10) =\displaystyle= 1Xf1df2dα(12F2+φ)=1Xf1df2(dG2)(α),\displaystyle{\sqrt{-1}}\int_{X}f_{1}df_{2}\wedge d\alpha\wedge\mathopen{}\mathclose{{}\left(\frac{1}{2}F_{\nabla}^{2}+*\varphi}\right)={\sqrt{-1}}\int_{X}f_{1}df_{2}\wedge(d\mathcal{F}_{G_{2}})_{\nabla}(\alpha),

where (dG2):1Ω11Ω6(d\mathcal{F}_{G_{2}})_{\nabla}:{\sqrt{-1}}\Omega^{1}\to{\sqrt{-1}}\Omega^{6} is the linearization of G2\mathcal{F}_{G_{2}} at 𝒜0{\nabla}\in\mathcal{A}_{0}. Hence we obtain

(3.11) Θ((f1),(f2),α)=\displaystyle\Theta_{\nabla}((f_{1}^{*})_{\nabla},(f_{2}^{*})_{\nabla},\alpha)= ιZ6(1(dG2)(α))(f1,f2)\displaystyle\iota_{Z^{6}}({\sqrt{-1}}(d\mathcal{F}_{G_{2}})_{\nabla}(\alpha))(f_{1},f_{2})
(3.12) =\displaystyle= ddtιZ6(1G2(+tαidL))|t=0(f1,f2)=(dν,f1f2)(α).\displaystyle\mathopen{}\mathclose{{}\left.\frac{d}{dt}\iota_{Z^{6}}({\sqrt{-1}}\mathcal{F}_{G_{2}}({\nabla}+t\alpha\cdot\mathrm{id}_{L}))}\right|_{t=0}(f_{1},f_{2})=(d\langle\nu,f_{1}\wedge f_{2}\rangle)_{\nabla}(\alpha).

In the dHYM case, the “𝒥\mathcal{J} functional” defined in [collins2021moment, Remark 2.15] or [collins2021survey, Lemma 2.6 (ii)] is convex along geodesics and the critical points are solutions of the dHYM equation. Hence it plays an important role in the existence problem.

In the dDT case, there is a functional whose critical points are dDT connections. See Section 4.2. However, no metric has yet been found that makes the functional convex along geodesics. Since no such results have been found for associative submanifolds, it might be difficult to relate the functional to the existence problem.

However, as we see in the next section, we have an observation as in the case of instanton Floer homology for 3-manifolds by using the functional in Section 4.2. We might develop the theory like instanton Floer homology using dDT connections.

4. Gradient flow of the Karigiannis-Leung functional

It is known that the gradient flow equation of the Chern-Simons functional on an oriented 3-manifold X3X^{3} agrees with the ASD equation on ×X3\mathbb{R}\times X^{3}. See for example [Donaldson2002, Section 2.5.3]. This is an important observation in instanton Floer homology for 3-manifolds. We show that there is an analogous relation between dDT equations for G2G_{2}- and Spin(7){\rm Spin}(7)-manifolds.

Let X7X^{7} be a 7-manifold with a G2G_{2}-structure φ\varphi and (L,h)X7(L,h)\to X^{7} be a smooth complex Hermitian line bundle over X7X^{7}. Let {t}t\{{\nabla}_{t}\}_{t\in\mathbb{R}} be a family of Hermitian connections of (L,h)X7(L,h)\to X^{7}. We identify this with a connection ~\widetilde{\nabla} of πL×X7\pi^{*}L\to\mathbb{R}\times X^{7}, where π:×X7X7\pi:\mathbb{R}\times X^{7}\to X^{7} is the projection. If we set

t=0+1atidL,{\nabla}_{t}={\nabla}_{0}+{\sqrt{-1}}a_{t}\cdot\mathrm{id}_{L},

where atΩ1(X7)a_{t}\in\Omega^{1}(X^{7}), we have ~=π0+1πatidπL\widetilde{\nabla}=\pi^{*}{\nabla}_{0}+{\sqrt{-1}}\pi^{*}a_{t}\cdot\mathrm{id}_{\pi^{*}L} and the curvature F~F_{\widetilde{\nabla}} of ~\widetilde{\nabla} is given by

F~=1dtπatt+πFt.F_{\widetilde{\nabla}}=\sqrt{-1}dt\wedge\frac{\partial\pi^{*}a_{t}}{\partial t}+\pi^{*}F_{{\nabla}_{t}}.

4.1. The Spin(7){\rm Spin}(7)-dDT condition on ×X7\mathbb{R}\times X^{7}

The product ×X7\mathbb{R}\times X^{7} admits a canonical Spin(7){\rm Spin}(7)-structure. We write down the condition that F~F_{\widetilde{\nabla}} is a Spin(7){\rm Spin}(7)-dDT connection, a dDT connection for a manifold with a Spin(7){\rm Spin}(7)-structure. For simplicity, set

a˙t:=πatt,Et:=1πFt.\dot{a}_{t}:=\frac{\partial\pi^{*}a_{t}}{\partial t},\qquad E_{t}:=-\sqrt{-1}\pi^{*}F_{{\nabla}_{t}}.
Lemma 4.1.

The connection ~\widetilde{\nabla} is a Spin(7){\rm Spin}(7)-dDT connection if and only if

(4.1) φEt+16Et3(112(φEt2))a˙t+(a˙tEtφ)Et\displaystyle-*\varphi\wedge E_{t}+\frac{1}{6}E_{t}^{3}-\mathopen{}\mathclose{{}\left(1-\frac{1}{2}*(\varphi\wedge E_{t}^{2})}\right)*\dot{a}_{t}+*(\dot{a}_{t}\wedge E_{t}\wedge\varphi)\wedge*E_{t} =0\displaystyle=0
(4.2) 12φEt2a˙tEtφ\displaystyle\frac{1}{2}\varphi\wedge*E_{t}^{2}-\dot{a}_{t}\wedge E_{t}\wedge\varphi =0.\displaystyle=0.
Proof.

Denote by 8*_{8} and =7*=*_{7} the Hodge star operators on ×X7\mathbb{R}\times X^{7} and X7X^{7}, respectively. Then, ~\widetilde{\nabla} is a Spin(7){\rm Spin}(7)-dDT connection (in the sense of [kawai2021FM, Definition 1.3]) if and only if

(4.3) F~+168F~3,dtb+i(b)φ=0,F~2,dti(b)φbφ=0\displaystyle\mathopen{}\mathclose{{}\left\langle F_{\widetilde{\nabla}}+\frac{1}{6}*_{8}F_{\widetilde{\nabla}}^{3},\ dt\wedge b+i(b^{\sharp})\varphi}\right\rangle=0,\qquad\mathopen{}\mathclose{{}\left\langle F_{\widetilde{\nabla}}^{2},\ dt\wedge i(b^{\sharp})*\varphi-b\wedge\varphi}\right\rangle=0

for any bΩ1(X7)b\in\Omega^{1}(X^{7}) by [kawai2021FM, Lemma 3.4]. Since

168F~3=168(3dta˙tEt2+Et3)=1(12(a˙tEt2)16dtEt3),\frac{1}{6}*_{8}F_{\widetilde{\nabla}}^{3}=-\frac{\sqrt{-1}}{6}*_{8}(3dt\wedge\dot{a}_{t}\wedge E_{t}^{2}+E_{t}^{3})=\sqrt{-1}\mathopen{}\mathclose{{}\left(-\frac{1}{2}*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}^{2}}\right)-\frac{1}{6}dt\wedge*E_{t}^{3}}\right),

\tagform@4.3 is equivalent to

(4.4) a˙t16Et3,b+Et12(a˙tEt2),i(b)φ=0,\displaystyle\mathopen{}\mathclose{{}\left\langle\dot{a}_{t}-\frac{1}{6}*E_{t}^{3},b}\right\rangle+\mathopen{}\mathclose{{}\left\langle E_{t}-\frac{1}{2}*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}^{2}}\right),i(b^{\sharp})\varphi}\right\rangle=0,
(4.5) 2a˙tEt,i(b)φEt2,bφ=0.\displaystyle\mathopen{}\mathclose{{}\left\langle 2\dot{a}_{t}\wedge E_{t},i(b^{\sharp})*\varphi}\right\rangle-\mathopen{}\mathclose{{}\left\langle E_{t}^{2},b\wedge\varphi}\right\rangle=0.

We compute

Et,i(b)φ=(Et(i(b)φ))=(Etbφ)=φEt,b\mathopen{}\mathclose{{}\left\langle E_{t},i(b^{\sharp})\varphi}\right\rangle=*(E_{t}\wedge*(i(b^{\sharp})\varphi))=*(E_{t}\wedge b\wedge*\varphi)=\langle*\varphi\wedge E_{t},*b\rangle

and

(4.6) 12(a˙tEt2),i(b)φ=\displaystyle\mathopen{}\mathclose{{}\left\langle-\frac{1}{2}*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}^{2}}\right),i(b^{\sharp})\varphi}\right\rangle= 12(a˙tEt2i(b)φ)\displaystyle-\frac{1}{2}*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}^{2}\wedge i(b^{\sharp})\varphi}\right)
(4.7) =\displaystyle= 12(i(b)(a˙tEt2)φ)\displaystyle-\frac{1}{2}*\mathopen{}\mathclose{{}\left(i(b^{\sharp})(\dot{a}_{t}\wedge E_{t}^{2})\wedge\varphi}\right)
(4.8) =\displaystyle= 12(Et2φ)a˙t,b+(a˙tEt(i(b)Et)φ).\displaystyle-\frac{1}{2}*\mathopen{}\mathclose{{}\left(E_{t}^{2}\wedge\varphi}\right)\cdot\langle\dot{a}_{t},b\rangle+*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}\wedge(i(b^{\sharp})E_{t})\wedge\varphi}\right).

Since i(b)Et=(bEt)i(b^{\sharp})E_{t}=-*(b\wedge*E_{t}), we have

(4.9) (a˙tEt(i(b)Et)φ)=a˙tEtφ,bEt=(a˙tEtφ)Et,b.\displaystyle*\mathopen{}\mathclose{{}\left(\dot{a}_{t}\wedge E_{t}\wedge(i(b^{\sharp})E_{t})\wedge\varphi}\right)=\langle\dot{a}_{t}\wedge E_{t}\wedge\varphi,b\wedge*E_{t}\rangle=-\langle*(\dot{a}_{t}\wedge E_{t}\wedge\varphi)\wedge*E_{t},*b\rangle.

Then, we see that \tagform@4.4 is equivalent to \tagform@4.1. Similarly, since

(4.10) 2a˙tEt,i(b)φ\displaystyle\mathopen{}\mathclose{{}\left\langle 2\dot{a}_{t}\wedge E_{t},i(b^{\sharp})*\varphi}\right\rangle =2(a˙tEtbφ)=2a˙tEtφ,b,\displaystyle=-2*(\dot{a}_{t}\wedge E_{t}\wedge b\wedge\varphi)=2\langle\dot{a}_{t}\wedge E_{t}\wedge\varphi,*b\rangle,
(4.11) Et2,bφ\displaystyle-\mathopen{}\mathclose{{}\left\langle E_{t}^{2},b\wedge\varphi}\right\rangle =(bφEt2)=φEt2,b,\displaystyle=-*(b\wedge\varphi\wedge*E_{t}^{2})=-\langle\varphi\wedge*E_{t}^{2},*b\rangle,

we see that \tagform@4.5 is equivalent to \tagform@4.2. ∎

Hence, eliminating (a˙tEtφ)*(\dot{a}_{t}\wedge E_{t}\wedge\varphi) from \tagform@4.1 by \tagform@4.2, we obtain

(4.12) φEt+16Et3+12(φEt2)Et=(112(φEt2))a˙t.\displaystyle-*\varphi\wedge E_{t}+\frac{1}{6}E_{t}^{3}+\frac{1}{2}*(\varphi\wedge*E_{t}^{2})\wedge*E_{t}=\mathopen{}\mathclose{{}\left(1-\frac{1}{2}*(\varphi\wedge E_{t}^{2})}\right)*\dot{a}_{t}.
Remark 4.2.

If 1(φEt2)/201-*(\varphi\wedge E_{t}^{2})/2\neq 0, \tagform@4.1 and \tagform@4.2 are equivalent to \tagform@4.12 by Proposition A.3.

4.2. The Karigiannis-Leung functional

Karigiannis and Leung [karigiannis2009hodge] introduced the functional whose critical points are dDT connections. We first review it.

Let X7X^{7} be a compact 7-manifold with a coclosed G2G_{2}-structure φ\varphi (dφ=0d*\varphi=0) and let (L,h)X7(L,h)\to X^{7} be a smooth complex Hermitian line bundle. Denote by 𝒜0\mathcal{A}_{0} the space of Hermitian connections of (L,h)(L,h). Define a 1-form Θ\Theta on 𝒜0\mathcal{A}_{0} by

Θ(1b)=X1b(16F3+Fφ)\Theta_{\nabla}({\sqrt{-1}}b)=\int_{X}{\sqrt{-1}}b\wedge\mathopen{}\mathclose{{}\left(\frac{1}{6}F_{\nabla}^{3}+F_{\nabla}\wedge*\varphi}\right)

for 𝒜0{\nabla}\in\mathcal{A}_{0} and 1b1Ω1=T𝒜0{\sqrt{-1}}b\in{\sqrt{-1}}\Omega^{1}=T_{\nabla}\mathcal{A}_{0}. Then we see that Θ=0\Theta_{\nabla}=0 if and only if {\nabla} is a dDT connection. We can show that Θ\Theta is closed as in the proof of Lemma 3.2. Since 𝒜0\mathcal{A}_{0} is contractible, there exists :𝒜0\mathcal{F}:\mathcal{A}_{0}\to\mathbb{R} such that d=Θd\mathcal{F}=\Theta. Hence we see that dDT connections are critical points of \mathcal{F}.

Now, we study the relation between Spin(7){\rm Spin}(7)-dDT connections on ×X7\mathbb{R}\times X^{7} and the Karigiannis-Leung functional \mathcal{F}. Set

𝒜ac:={𝒜0| 1+12(φF2)>0}.\mathcal{A}_{ac}:=\mathopen{}\mathclose{{}\left\{{\nabla}\in\mathcal{A}_{0}\;\middle|\;1+\frac{1}{2}*(\varphi\wedge F_{\nabla}^{2})>0}\right\}.

This type of the subset is also considered in the dHYM case. For example, see the survey article [collins2021survey, Definition 2.1]. By the mirror of the associator equality in [kawai2021mirror, Theorem 5.1], it will be natural to call a Hermitian connection {\nabla} satisfying 1+(φF2)/2>01+*(\varphi\wedge F_{\nabla}^{2})/2>0 almost calibrated as in the dHYM case.

Define a metric 𝒢\mathcal{G} on 𝒜ac\mathcal{A}_{ac} by

𝒢(1a,1b)=Xa,b(1+12(φF2))vol\mathcal{G}_{\nabla}(\sqrt{-1}a,\sqrt{-1}b)=\int_{X}\langle a,b\rangle_{\nabla}\mathopen{}\mathclose{{}\left(1+\frac{1}{2}*(\varphi\wedge F_{\nabla}^{2})}\right)\mathrm{vol}

where 𝒜ac,1a,1b1Ω1=T𝒜ac{\nabla}\in\mathcal{A}_{ac},\sqrt{-1}a,\sqrt{-1}b\in{\sqrt{-1}}\Omega^{1}=T_{\nabla}\mathcal{A}_{ac}, vol\mathrm{vol} is the induced volume form from φ\varphi, and ,\langle\cdot,\cdot\rangle_{\nabla} is the induced metric on the space of differential forms from (idTX+(1F))φ(\mathrm{id}_{TX}+(-{\sqrt{-1}}F_{\nabla})^{\sharp})^{*}\varphi. Here, (1F)(-{\sqrt{-1}}F_{\nabla})^{\sharp} is an endomorphism of TXTX defined by g((1F)(u),v)=1F(u,v)g((-{\sqrt{-1}}F_{\nabla})^{\sharp}(u),v)=-{\sqrt{-1}}F_{\nabla}(u,v) for u,vTXu,v\in TX, where gg is the induced metric (on TXTX) from φ\varphi. Note that (1F)(-{\sqrt{-1}}F_{\nabla})^{\sharp} is skew-symmetric with respect to gg. Explicitly, if we denote by gg_{\nabla} the induced metric (on TXTX) from (idTX+(1F))φ(\mathrm{id}_{TX}+(-{\sqrt{-1}}F_{\nabla})^{\sharp})^{*}\varphi, we have g=(idTX+(1F))gg_{\nabla}=(\mathrm{id}_{TX}+(-{\sqrt{-1}}F_{\nabla})^{\sharp})^{*}g and ,\langle\cdot,\cdot\rangle_{\nabla} is the induced metric from gg_{\nabla}.

The following is the main theorem of this paper.

Theorem 4.3.

The gradient flow equation of \mathcal{F} with respect to 𝒢\mathcal{G} on 𝒜ac\mathcal{A}_{ac} agrees with the Spin(7){\rm Spin}(7)-dDT equation on ×X7\mathbb{R}\times X^{7}.

Proof.

We first deduce the gradient flow equation and compare it with the computation in Section 4.1. Take any 𝒜ac{\nabla}\in\mathcal{A}_{ac} and bΩ1b\in\Omega^{1}. Set

E=1FΩ2.E_{\nabla}=-{\sqrt{-1}}F_{\nabla}\in\Omega^{2}.

Denote by ,\langle\cdot,\cdot\rangle the induced metric on the space of differential forms from φ\varphi. Then we compute

(4.13) (d)(1b)=X1b(16F3+Fφ)=Xb,(16E3Eφ)vol.\displaystyle(d\mathcal{F})_{\nabla}({\sqrt{-1}}b)=\int_{X}{\sqrt{-1}}b\wedge\mathopen{}\mathclose{{}\left(\frac{1}{6}F_{\nabla}^{3}+F_{\nabla}\wedge*\varphi}\right)=\int_{X}\mathopen{}\mathclose{{}\left\langle b,*\mathopen{}\mathclose{{}\left(\frac{1}{6}E_{\nabla}^{3}-E_{\nabla}\wedge*\varphi}\right)}\right\rangle\mathrm{vol}.

By Proposition A.1, we have

(16E3Eφ)=((idTX(E)2)1)η,*\mathopen{}\mathclose{{}\left(\frac{1}{6}E_{\nabla}^{3}-E_{\nabla}\wedge*\varphi}\right)=\mathopen{}\mathclose{{}\left(\mathopen{}\mathclose{{}\left(\mathrm{id}_{TX}-(E_{\nabla}^{\sharp})^{2}}\right)^{-1}}\right)^{*}\eta_{\nabla},

where

(4.14) η=(φE+16E3+12(φE2)E)Ω1.\displaystyle\eta_{\nabla}=*\mathopen{}\mathclose{{}\left(-*\varphi\wedge E_{\nabla}+\frac{1}{6}E_{\nabla}^{3}+\frac{1}{2}*(\varphi\wedge*E_{\nabla}^{2})\wedge*E_{\nabla}}\right)\in\Omega^{1}.

Since

idTX(E)2=(idTXE)(idTX+E)=(idTX+E)t(idTX+E),\mathrm{id}_{TX}-(E_{\nabla}^{\sharp})^{2}=(\mathrm{id}_{TX}-E_{\nabla}^{\sharp})(\mathrm{id}_{TX}+E_{\nabla}^{\sharp})={}^{t}(\mathrm{id}_{TX}+E_{\nabla}^{\sharp})(\mathrm{id}_{TX}+E_{\nabla}^{\sharp}),

where (idTX+E)t{}^{t}(\mathrm{id}_{TX}+E_{\nabla}^{\sharp}) is the transpose of idTX+E\mathrm{id}_{TX}+E_{\nabla}^{\sharp} with respect to gg, we have

(4.15) b,(16E3Eφ)=b,((idTX(E)2)1)η=((idTX+E)1)b,((idTX+E)1)η=b,η.\begin{split}\mathopen{}\mathclose{{}\left\langle b,*\mathopen{}\mathclose{{}\left(\frac{1}{6}E_{\nabla}^{3}-E_{\nabla}\wedge*\varphi}\right)}\right\rangle=&\mathopen{}\mathclose{{}\left\langle b,\mathopen{}\mathclose{{}\left(\mathopen{}\mathclose{{}\left(\mathrm{id}_{TX}-(E_{\nabla}^{\sharp})^{2}}\right)^{-1}}\right)^{*}\eta_{\nabla}}\right\rangle\\ =&\mathopen{}\mathclose{{}\left\langle\mathopen{}\mathclose{{}\left(\mathopen{}\mathclose{{}\left(\mathrm{id}_{TX}+E_{\nabla}^{\sharp}}\right)^{-1}}\right)^{*}b,\mathopen{}\mathclose{{}\left(\mathopen{}\mathclose{{}\left(\mathrm{id}_{TX}+E_{\nabla}^{\sharp}}\right)^{-1}}\right)^{*}\eta_{\nabla}}\right\rangle=\mathopen{}\mathclose{{}\left\langle b,\eta_{\nabla}}\right\rangle_{\nabla}.\end{split}

Then by \tagform@4.13 and \tagform@4.15, the gradient vector field of \mathcal{F} with respect to 𝒢\mathcal{G} is given by

𝒜ac1η112(φE2)1Ω1.\mathcal{A}_{ac}\ni{\nabla}\mapsto\frac{{\sqrt{-1}}\eta_{\nabla}}{1-\frac{1}{2}*(\varphi\wedge E_{\nabla}^{2})}\in{\sqrt{-1}}\Omega^{1}.

Thus a family {t}t𝒜ac\{{\nabla}_{t}\}_{t\in\mathbb{R}}\subset\mathcal{A}_{ac} satisfies the gradient flow of \mathcal{F} with respect to 𝒢\mathcal{G} if and only if

(4.16) a˙t=ηt112(φEt2)=(φEt+16Et3+12(φEt2)Et)112(φEt2),\displaystyle\dot{a}_{t}=\frac{\eta_{{\nabla}_{t}}}{1-\frac{1}{2}*(\varphi\wedge E_{{\nabla}_{t}}^{2})}=\frac{*\mathopen{}\mathclose{{}\left(-*\varphi\wedge E_{t}+\frac{1}{6}E_{t}^{3}+\frac{1}{2}*(\varphi\wedge*E_{t}^{2})\wedge*E_{t}}\right)}{1-\frac{1}{2}*(\varphi\wedge E_{t}^{2})},

where t=0+1atidL{\nabla}_{t}={\nabla}_{0}+{\sqrt{-1}}a_{t}\cdot\mathrm{id}_{L}, atΩ1a_{t}\in\Omega^{1}, a˙t=at/t\dot{a}_{t}=\partial a_{t}/\partial t and Et=Et=1FtE_{t}=E_{{\nabla}_{t}}=-{\sqrt{-1}}F_{{\nabla}_{t}}. Then we see that \tagform@4.16 is equivalent to \tagform@4.12. By Remark 4.2, this is equivalent to the Spin(7){\rm Spin}(7)-dDT equation on ×X7\mathbb{R}\times X^{7}. ∎

By Theorem 4.3, we will have to consider the deformation theory of the Spin(7){\rm Spin}(7)-dDT connections on ×X7\mathbb{R}\times X^{7} next for the analogue of instanton Floer homology for 3-manifolds. Deformations of Spin(7){\rm Spin}(7)-dDT connections on a compact manifold with a Spin(7){\rm Spin}(7)-structure are studied in [kawai2021deformationSpin(7), Theorem 1.2], but there are some technical assumptions. We will have to deal with more technical issues, including these, to develop the deformation theory on a cylinder.

Appendix A Algebraic Computations

In this appendix, we give some algebraic computations needed in the proof of Theorem 4.3.

Set V=7V=\mathbb{R}^{7} and let gg be the standard inner product on VV. Denote by * the standard Hodge star operator on VV. For a 2-form FΛ2VF\in\Lambda^{2}V^{*}, define FEnd(V)F^{\sharp}\in{\rm End}(V) by

g(F(u),v)=F(u,v)g(F^{\sharp}(u),v)=F(u,v)

for u,vVu,v\in V. Then, FF^{\sharp} is skew-symmetric, and hence, missingdet(I+F)>0\mathop{\mathrm{missing}}{det}\nolimits(I+F^{\sharp})>0, where II is the identity matrix. We also have

missingdet(I(F)2)=missingdet(I+F)missingdet(IF)=missingdet(I+F)missingdet(I+Ft)=(missingdet(I+F))2>0,\mathop{\mathrm{missing}}{det}\nolimits(I-(F^{\sharp})^{2})=\mathop{\mathrm{missing}}{det}\nolimits(I+F^{\sharp})\mathop{\mathrm{missing}}{det}\nolimits(I-F^{\sharp})=\mathop{\mathrm{missing}}{det}\nolimits(I+F^{\sharp})\mathop{\mathrm{missing}}{det}\nolimits(I+{}^{t}F^{\sharp})=(\mathop{\mathrm{missing}}{det}\nolimits(I+F^{\sharp}))^{2}>0,

where we Ft{}^{t}F^{\sharp} is the transpose of FF^{\sharp} with respect to gg. Define a 3-form φΛ3V\varphi\in\Lambda^{3}V^{*} by

φ=e123+e145+e167+e246e257e347e356,\varphi=e^{123}+e^{145}+e^{167}+e^{246}-e^{257}-e^{347}-e^{356},

where {ei}i=17\{e_{i}\}_{i=1}^{7} is a standard oriented basis of VV with the dual basis {ei}i=17\{e^{i}\}_{i=1}^{7} of VV^{\ast} and ei1ike^{i_{1}\dots i_{k}} is short for ei1eike^{i_{1}}\wedge\cdots\wedge e^{i_{k}}. The stabilizer of φ\varphi is known to be the exceptional 1414-dimensional simple Lie group G2G_{2}. The elements of G2G_{2} preserve the standard inner product gg and volume form vol\mathrm{vol}. The group G2G_{2} acts canonically on ΛkV\Lambda^{k}V^{*}, and Λ2V\Lambda^{2}V^{*} is decomposed as Λ2V=Λ72VΛ142V\Lambda^{2}V^{*}=\Lambda^{2}_{7}V^{*}\oplus\Lambda^{2}_{14}V^{*}, where Λ2V\Lambda^{2}_{\ell}V^{*} is a \ell-dimensional irreducible subrepresentation of G2G_{2} in Λ2V\Lambda^{2}V^{*}. For more details, see for example [kawai2021mirror, Section 2.2]. Set

F=F7+F14=i(u)φ+F14Λ72VΛ142VF=F_{7}+F_{14}=i(u)\varphi+F_{14}\in\Lambda^{2}_{7}V^{*}\oplus\Lambda^{2}_{14}V^{*}

for uVu\in V.

Proposition A.1.

For a 2-form FΛ2VF\in\Lambda^{2}V^{*}, set ξ=φF+F3/6Λ6V\xi=-*\varphi\wedge F+F^{3}/6\in\Lambda^{6}V^{*}. Then we have

(A.1) (I(F)2)ξ=(ξ+12(φF2)F).\displaystyle(I-(F^{\sharp})^{2})^{*}*\xi=*\mathopen{}\mathclose{{}\left(\xi+\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F}\right).
Proof.

Since (I(F)2)ξ=ξ((F)2)ξ(I-(F^{\sharp})^{2})^{*}*\xi=*\xi-((F^{\sharp})^{2})^{*}*\xi, we only have to compute ((F)2)ξ((F^{\sharp})^{2})^{*}*\xi. Set

Fij=F(ei,ej).F_{ij}=F(e_{i},e_{j}).

We have F=i,jFijeiejF^{\sharp}=\sum_{i,j}F_{ij}e^{i}\otimes e_{j}, which implies that (F)2=i,j,kFijFjkeiek(F^{\sharp})^{2}=\sum_{i,j,k}F_{ij}F_{jk}e^{i}\otimes e_{k}. Then we compute

(A.2) ((F)2)ξ=i,j,kFijFjkξ(ek)ei=ji(ej)F,ξi(ej)F.\displaystyle((F^{\sharp})^{2})^{*}*\xi=\sum_{i,j,k}F_{ij}F_{jk}*\xi(e_{k})\cdot e^{i}=-\sum_{j}\langle i(e_{j})F,*\xi\rangle\cdot i(e_{j})F.

Since

(A.3) i(ej)F,ξ=\displaystyle\langle i(e_{j})F,*\xi\rangle= (ξ(i(ej)F))=(ξejF)=ej,(ξF),\displaystyle*\mathopen{}\mathclose{{}\left(*\xi\wedge*(i(e_{j})F)}\right)=-*\mathopen{}\mathclose{{}\left(*\xi\wedge e^{j}\wedge*F}\right)=\langle e^{j},*(*\xi\wedge*F)\rangle,
(A.4) i(ej)F=\displaystyle i(e_{j})F= (ejF),\displaystyle-*(e^{j}\wedge*F),

we have

(A.5) ((F)2)ξ=((ξF)F)=(((φF+F36)F)F).\displaystyle((F^{\sharp})^{2})^{*}*\xi=*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(*\xi\wedge*F}\right)\wedge*F}\right)=*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(-*\varphi\wedge F+\frac{F^{3}}{6}}\right)\wedge*F}\right)\wedge*F}\right).
Lemma A.2.

We have

(A.6) (F3)F\displaystyle(*F^{3})\wedge*F =0,\displaystyle=0,
(A.7) (φF2)\displaystyle*(\varphi\wedge*F^{2}) =6i(u)F.\displaystyle=-6i(u)F.
Proof.

We can prove the first equation as in [kawai2020deformation, Lemma C.2]. For any vVv\in V, set

v=g(v,)V.v^{\flat}=g(v,\cdot)\in V^{*}.

We compute

(A.8) v(F3)F=(F3)(i(v)F)=F3i(v)F=i(v)(F4/4)=0,\displaystyle v^{\flat}\wedge(*F^{3})\wedge*F=(*F^{3})\wedge*(i(v)F)=F^{3}\wedge i(v)F=i(v)(F^{4}/4)=0,

which implies the first equation. Similarly, for any vVv\in V, we have

vφF2=(vφ)F2=i(v)φF2=φi(v)F2=2i(v)FFφ.v^{\flat}\wedge\varphi\wedge*F^{2}=*(v^{\flat}\wedge\varphi)\wedge F^{2}=-i(v)*\varphi\wedge F^{2}=*\varphi\wedge i(v)F^{2}=2i(v)F\wedge F\wedge*\varphi.

Since Fφ=i(u)φφ=3uF\wedge*\varphi=i(u)\varphi\wedge*\varphi=3*u^{\flat} by for example [kawai2020deformation, Lemma B.1], we obtain

vφF2=6u,i(v)Fvol=6vu,Fvol=6v,i(u)Fvol,v^{\flat}\wedge\varphi\wedge*F^{2}=6\langle u^{\flat},i(v)F\rangle\mathrm{vol}=6\langle v^{\flat}\wedge u^{\flat},F\rangle\mathrm{vol}=-6\langle v^{\flat},i(u)F\rangle\mathrm{vol},

which implies the second equation. ∎

Then by \tagform@A.5, Lemma A.2 and the equation Fφ=i(u)φφ=3uF\wedge*\varphi=i(u)\varphi\wedge*\varphi=3*u^{\flat}, we obtain

(A.9) ((F)2)ξ=\displaystyle((F^{\sharp})^{2})^{*}*\xi= (((φF)F)F)\displaystyle*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(-*\varphi\wedge F}\right)\wedge*F}\right)\wedge*F}\right)
(A.10) =\displaystyle= ((3uF)F)\displaystyle*\mathopen{}\mathclose{{}\left(*\mathopen{}\mathclose{{}\left(-3u^{\flat}\wedge*F}\right)\wedge*F}\right)
(A.11) =\displaystyle= 3((i(u)F)F)=12((φF2)F)\displaystyle 3*\mathopen{}\mathclose{{}\left((i(u)F)\wedge*F}\right)=-\frac{1}{2}*\mathopen{}\mathclose{{}\left(*(\varphi\wedge*F^{2})\wedge*F}\right)

and the proof is completed. ∎

Proposition A.3.

For a 1-form aVa\in V^{*} and a 2-form FΛ2VF\in\Lambda^{2}V^{*} such that 1(φF2)/201-*(\varphi\wedge F^{2})/2\neq 0,

(A.12) φF+16F3(112(φF2))a+(aFφ)F\displaystyle-*\varphi\wedge F+\frac{1}{6}F^{3}-\mathopen{}\mathclose{{}\left(1-\frac{1}{2}*(\varphi\wedge F^{2})}\right)*a+*(a\wedge F\wedge\varphi)\wedge*F =0,\displaystyle=0,
(A.13) 12φF2aFφ\displaystyle\frac{1}{2}\varphi\wedge*F^{2}-a\wedge F\wedge\varphi =0\displaystyle=0

if and only if

(A.14) φF+16F3+12(φF2)F=(112(φF2))a.\displaystyle-*\varphi\wedge F+\frac{1}{6}F^{3}+\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F=\mathopen{}\mathclose{{}\left(1-\frac{1}{2}*(\varphi\wedge F^{2})}\right)*a.
Proof.

Eliminating aFφa\wedge F\wedge\varphi from \tagform@A.12 by \tagform@A.13, we obtain \tagform@A.14. Conversely, \tagform@A.14 implies \tagform@A.13 by the following Lemma A.4. By \tagform@A.14, the left hand side of \tagform@A.12 is computed as

(A.15) 12(φF2)F+(aFφ)F=(12φF2+aFφ)F,\displaystyle-\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F+*(a\wedge F\wedge\varphi)\wedge*F=*\mathopen{}\mathclose{{}\left(-\frac{1}{2}\varphi\wedge*F^{2}+a\wedge F\wedge\varphi}\right)\wedge*F,

which vanishes by \tagform@A.13. ∎

Lemma A.4.

For any 2-form FΛ2VF\in\Lambda^{2}V^{*}, we have

(φF+16F3+12(φF2)F)Fφ=12(112(φF2))φF2.*\mathopen{}\mathclose{{}\left(-*\varphi\wedge F+\frac{1}{6}F^{3}+\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F}\right)\wedge F\wedge\varphi=\frac{1}{2}\mathopen{}\mathclose{{}\left(1-\frac{1}{2}*(\varphi\wedge F^{2})}\right)\varphi\wedge*F^{2}.
Proof.

Fix any vVv\in V and set

J1=v(φF+16F3)Fφ,J2=v(12(φF2)F)Fφ.J_{1}=v^{\flat}\wedge*\mathopen{}\mathclose{{}\left(-*\varphi\wedge F+\frac{1}{6}F^{3}}\right)\wedge F\wedge\varphi,\qquad J_{2}=v^{\flat}\wedge*\mathopen{}\mathclose{{}\left(\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F}\right)\wedge F\wedge\varphi.

We compute J1J_{1} and J2J_{2}. We have

(A.16) J1=\displaystyle J_{1}= (i(v)(φF16F3))(2F7F14)\displaystyle*\mathopen{}\mathclose{{}\left(i(v)\mathopen{}\mathclose{{}\left(*\varphi\wedge F-\frac{1}{6}F^{3}}\right)}\right)\wedge*(2F_{7}-F_{14})
(A.17) =\displaystyle= i(v)(φF16F3)(2F7F14)=(3(vu)12i(v)FF2)(2F7F14),\displaystyle i(v)\mathopen{}\mathclose{{}\left(*\varphi\wedge F-\frac{1}{6}F^{3}}\right)\wedge(2F_{7}-F_{14})=\mathopen{}\mathclose{{}\left(-3*(v^{\flat}\wedge u^{\flat})-\frac{1}{2}i(v)F\wedge F^{2}}\right)\wedge(2F_{7}-F_{14}),

where we use φF=3u*\varphi\wedge F=3*u^{\flat}. We also have

3(vu)(2F7F14)=3vu,2F7F14vol=3v,i(u)Fvol-3*(v^{\flat}\wedge u^{\flat})\wedge(2F_{7}-F_{14})=-3\langle v^{\flat}\wedge u^{\flat},2F_{7}-F_{14}\rangle\mathrm{vol}=-3\langle v^{\flat},i(u)F\rangle\mathrm{vol}

as i(u)F7=i(u)i(u)φ=0i(u)F_{7}=i(u)i(u)\varphi=0, and

(A.18) (12i(v)FF2)(2F7F14)\displaystyle\mathopen{}\mathclose{{}\left(-\frac{1}{2}i(v)F\wedge F^{2}}\right)\wedge(2F_{7}-F_{14})
(A.19) =\displaystyle= 12i(v)F(F72+2F7F14+F142)(2F7F14)\displaystyle-\frac{1}{2}i(v)F\wedge(F_{7}^{2}+2F_{7}\wedge F_{14}+F_{14}^{2})\wedge(2F_{7}-F_{14})
(A.20) =\displaystyle= 12(i(v)F7+i(v)F14)(2F73+3F72F14F143)\displaystyle-\frac{1}{2}\mathopen{}\mathclose{{}\left(i(v)F_{7}+i(v)F_{14}}\right)\wedge(2F_{7}^{3}+3F_{7}^{2}\wedge F_{14}-F_{14}^{3})
(A.21) =\displaystyle= 12{i(v)F7(3F72F14F143)+i(v)F14(2F73+3F72F14)},\displaystyle-\frac{1}{2}\mathopen{}\mathclose{{}\left\{i(v)F_{7}\wedge(3F_{7}^{2}\wedge F_{14}-F_{14}^{3})+i(v)F_{14}\wedge(2F_{7}^{3}+3F_{7}^{2}\wedge F_{14})}\right\},

where we use i(v)F7F73=i(v)(F74/4)=0i(v)F_{7}\wedge F_{7}^{3}=i(v)(F_{7}^{4}/4)=0 and i(v)F14F143=i(v)(F144/4)=0i(v)F_{14}\wedge F_{14}^{3}=i(v)(F_{14}^{4}/4)=0. By [kawai2020deformation, (B.7)], we have

(A.22) F73=6|u|2u.\displaystyle F_{7}^{3}=6|u|^{2}*u^{\flat}.

Then

(A.23) 3i(v)F7F72F14=\displaystyle 3i(v)F_{7}\wedge F_{7}^{2}\wedge F_{14}= i(v)F73F14=6|u|2(vu)F14=6|u|2v,i(u)Fvol,\displaystyle i(v)F_{7}^{3}\wedge F_{14}=-6|u|^{2}*(v^{\flat}\wedge u^{\flat})\wedge F_{14}=6|u|^{2}\langle v^{\flat},i(u)F\rangle\mathrm{vol},
(A.24) 2i(v)F14F73=\displaystyle 2i(v)F_{14}\wedge F_{7}^{3}= 12|u|2i(v)F14u=12|u|2F14,vuvol=12|u|2v,i(u)Fvol.\displaystyle 12|u|^{2}i(v)F_{14}\wedge*u^{\flat}=12|u|^{2}\langle F_{14},v^{\flat}\wedge u^{\flat}\rangle\mathrm{vol}=-12|u|^{2}\langle v^{\flat},i(u)F\rangle\mathrm{vol}.

Hence we obtain

(A.25) J1=(3+3|u|2)v,i(u)Fvol+12(i(v)F7F1433i(v)F14F72F14).\displaystyle J_{1}=(-3+3|u|^{2})\langle v^{\flat},i(u)F\rangle\mathrm{vol}+\frac{1}{2}\mathopen{}\mathclose{{}\left(i(v)F_{7}\wedge F_{14}^{3}-3i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}}\right).

Next, we compute J2J_{2}. By Lemma A.2, we have

(A.27) J2=\displaystyle J_{2}= v(3i(u)FF)(2F7F14)\displaystyle v^{\flat}\wedge*\mathopen{}\mathclose{{}\left(-3i(u)F\wedge*F}\right)\wedge*(2F_{7}-F_{14})
(A.28) =\displaystyle= 3(i(u)FF)(i(v)(2F7+F14))=3i(u)FFi(v)(2F7+F14).\displaystyle 3*\mathopen{}\mathclose{{}\left(i(u)F\wedge*F}\right)\wedge*\mathopen{}\mathclose{{}\left(i(v)(-2F_{7}+F_{14})}\right)=3i(u)F\wedge*F\wedge i(v)(-2F_{7}+F_{14}).

Since

(A.29) i(u)FF=\displaystyle i(u)F\wedge*F= i(u)F14(12F7φF14φ)\displaystyle i(u)F_{14}\wedge\mathopen{}\mathclose{{}\left(\frac{1}{2}F_{7}\wedge\varphi-F_{14}\wedge\varphi}\right)
(A.30) =\displaystyle= 12(i(u)(F14F7φ)F14F7i(u)φ)12i(u)F142φ\displaystyle\frac{1}{2}\mathopen{}\mathclose{{}\left(i(u)(F_{14}\wedge F_{7}\wedge\varphi)-F_{14}\wedge F_{7}\wedge i(u)\varphi}\right)-\frac{1}{2}i(u)F_{14}^{2}\wedge\varphi
(A.31) =\displaystyle= 12F72F1412(i(u)(F142φ)F142i(u)φ)\displaystyle-\frac{1}{2}F_{7}^{2}\wedge F_{14}-\frac{1}{2}\mathopen{}\mathclose{{}\left(i(u)(F_{14}^{2}\wedge\varphi)-F_{14}^{2}\wedge i(u)\varphi}\right)
(A.32) =\displaystyle= 12(|F14|2uF72F14+F7F142),\displaystyle\frac{1}{2}\mathopen{}\mathclose{{}\left(|F_{14}|^{2}*u^{\flat}-F_{7}^{2}\wedge F_{14}+F_{7}\wedge F_{14}^{2}}\right),

we have

(A.33) J2=32(F72F14+F7F142)i(v)(2F7+F14)+32|F14|2ui(v)(2F7+F14).\displaystyle J_{2}=\frac{3}{2}\mathopen{}\mathclose{{}\left(-F_{7}^{2}\wedge F_{14}+F_{7}\wedge F_{14}^{2}}\right)\wedge i(v)(-2F_{7}+F_{14})+\frac{3}{2}|F_{14}|^{2}*u^{\flat}\wedge i(v)(-2F_{7}+F_{14}).

We compute

(A.34) (F72F14+F7F142)i(v)(2F7+F14)\displaystyle\mathopen{}\mathclose{{}\left(-F_{7}^{2}\wedge F_{14}+F_{7}\wedge F_{14}^{2}}\right)\wedge i(v)(-2F_{7}+F_{14})
(A.35) =\displaystyle= 2i(v)F7F72F142i(v)F7F7F142i(v)F14F72F14+i(v)F14F7F142.\displaystyle 2i(v)F_{7}\wedge F_{7}^{2}\wedge F_{14}-2i(v)F_{7}\wedge F_{7}\wedge F_{14}^{2}-i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}+i(v)F_{14}\wedge F_{7}\wedge F_{14}^{2}.

By \tagform@A.22, it follows that

2i(v)F7F72F14=23i(v)F73F14=4|u|2v,i(u)Fvol.2i(v)F_{7}\wedge F_{7}^{2}\wedge F_{14}=\frac{2}{3}i(v)F_{7}^{3}\wedge F_{14}=4|u|^{2}\langle v^{\flat},i(u)F\rangle\mathrm{vol}.

Since 2i(v)F7F7F142=i(v)F72F142=F72i(v)F142=2i(v)F14F72F14-2i(v)F_{7}\wedge F_{7}\wedge F_{14}^{2}=-i(v)F_{7}^{2}\wedge F_{14}^{2}=F_{7}^{2}\wedge i(v)F_{14}^{2}=2i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}, we have

2i(v)F7F7F142i(v)F14F72F14=i(v)F14F72F14.-2i(v)F_{7}\wedge F_{7}\wedge F_{14}^{2}-i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}=i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}.

We also have

i(v)F14F7F142=13i(v)F143F7=13F143i(v)F7i(v)F_{14}\wedge F_{7}\wedge F_{14}^{2}=\frac{1}{3}i(v)F_{14}^{3}\wedge F_{7}=-\frac{1}{3}F_{14}^{3}\wedge i(v)F_{7}

and

32|F14|2ui(v)(2F7+F14)=32|F14|22F7+F14,vuvol=32|F14|2v,i(u)Fvol.\frac{3}{2}|F_{14}|^{2}*u^{\flat}\wedge i(v)(-2F_{7}+F_{14})=\frac{3}{2}|F_{14}|^{2}\langle-2F_{7}+F_{14},v^{\flat}\wedge u^{\flat}\rangle\mathrm{vol}=-\frac{3}{2}|F_{14}|^{2}\langle v^{\flat},i(u)F\rangle\mathrm{vol}.

Hence we obtain

(A.36) J2=(6|u|232|F14|2)v,i(u)Fvol+32i(v)F14F72F1412i(v)F7F143.\displaystyle J_{2}=\mathopen{}\mathclose{{}\left(6|u|^{2}-\frac{3}{2}|F_{14}|^{2}}\right)\langle v^{\flat},i(u)F\rangle\mathrm{vol}+\frac{3}{2}i(v)F_{14}\wedge F_{7}^{2}\wedge F_{14}-\frac{1}{2}i(v)F_{7}\wedge F_{14}^{3}.

Then by \tagform@A.25 and \tagform@A.36, we obtain

(A.38) J1+J2=3(1+3|u|212|F14|2)v,i(u)Fvol=3(1+12(φF2))v,i(u)Fvol,\displaystyle J_{1}+J_{2}=3\mathopen{}\mathclose{{}\left(-1+3|u|^{2}-\frac{1}{2}|F_{14}|^{2}}\right)\langle v^{\flat},i(u)F\rangle\mathrm{vol}=3\mathopen{}\mathclose{{}\left(-1+\frac{1}{2}*(\varphi\wedge F^{2})}\right)\langle v^{\flat},i(u)F\rangle\mathrm{vol},

where we use (φF2)=(F(2F7F14))=2|F7|2|F14|2=6|u|2|F14|2*(\varphi\wedge F^{2})=*\mathopen{}\mathclose{{}\left(F\wedge*(2F_{7}-F_{14})}\right)=2|F_{7}|^{2}-|F_{14}|^{2}=6|u|^{2}-|F_{14}|^{2} by [kawai2020deformation, Lemma B.1]. Then it follows that

(φF+16F3+12(φF2)F)Fφ=3(1+12(φF2))(i(u)F).*\mathopen{}\mathclose{{}\left(-*\varphi\wedge F+\frac{1}{6}F^{3}+\frac{1}{2}*(\varphi\wedge*F^{2})\wedge*F}\right)\wedge F\wedge\varphi=3\mathopen{}\mathclose{{}\left(-1+\frac{1}{2}*(\varphi\wedge F^{2})}\right)*(i(u)F).

Since φF2=6(i(u)F)\varphi\wedge*F^{2}=-6*(i(u)F) by Lemma A.2, the proof is completed. ∎

References