Extending Donaldson’s HyperKähler Structure via Minimal Surfaces into \mathrm{PSL}(2,\mathbb{C})

Any local complex coordinate z for (S,J) satisfies

or, equivalently,

Now, a smooth path of complex structures on S corresponds to a smooth 1-parameter family of endomorphisms (J_{t})_t and, by the Generalized Riemann mapping theorem, we can find a corresponding family of local coordinates z_t for (S,J_t) varying regularly in t. Differentiating Equation (Proof) at t=0, we obtain

and, evaluating in \partial_{\overline z} e \partial_z, we deduce

By construction, the tensor

is the first-order variation at t = 0 of the Beltrami differential of the identity maps id : (S,J_0) \to (S,J) and, in particular, it is an element of \mathrm{Belt}_{J_0}(S). From Equation (Proof), we deduce that \dot{J} can be written in local coordinates as

Similarly,

Therefore, combining the two expressions above with Equation (Proof), we have

This proves the formula (2.4). The formula (2.4) then follows from the immediate check that multiplication of \dot\mu by i corresponds to pre-composition of \dot J by J.

The proof relies on ideas from [emam2024bers, rungi2024complex]. Let z and \overline w be local holomorphic coordinates for the pull-back complex structures of J_1 and -J_2 on \Sigma, respectively. Clearly, \mathrm{d}z and \mathrm{d}\overline w are linearly independent at every point of \Sigma, as they have different kernels, corresponding to the two isotropic directions of \mathtt I. In particular, \mathrm{d}z and \mathrm{d}\overline w span \operatorname{\mathbb{C}} T^*\Sigma wherever they are simultaneously defined.

Being \mathtt {II} a (2,0)–tensor it can locally be written as

for smooth local functions \alpha,\beta, \gamma. To prove the statement, we have to show that \tr_{\mathtt I}\mathtt {II}=0 and \nabla^{\mathtt I}\mathtt {II} is tri-symmetric if and only if \alpha is J_1-holomorphic, \gamma is (-J_2)-holomorphic, and \beta=0. Let \mathtt I=\varrho \, \mathrm{d}z\mathrm{d}\overline w be the positive metric written in the isotropic local frame \{\partial_{\overline z},\partial_w\}, where \varrho=\varrho(z,\overline w) is a smooth, nowhere-vanishing function. Let us define s^2:=\mathtt I(\partial_{\overline z},\partial_w)=\frac{1}{2}\varrho(\partial_wz)(\partial_{\overline z}\overline w) and consider the \mathtt I–orthonormal local frame

for a choice of a square root of \varrho, possibly in a smaller contractible open subset. Using this frame we can compute

hence \trace_{\mathtt I}\mathtt {II}=0 if and only if \beta=0. By differentiating \mathtt {II} we get

We show that \nabla^{\mathtt I}\mathrm{d}z is symmetric. By metric compatibility, we have \mathtt I(\nabla^{\mathtt I} \partial_{\overline z}, \partial_{\overline z})=\frac 12 \mathrm{d}(\mathtt I(\partial_{\overline z}, \partial_{\overline z}))=0, hence \nabla^{\mathtt I} \partial_{\overline z}\in \Gamma(\mathbb{C}T^*S \otimes \mathrm{Span}_{\operatorname{\mathbb{C}}}(\partial_{\overline z})) since \mathtt I is nondegenerate. Moreover, the identities

and

imply that \nabla^\mathtt I \mathrm{d}z \in\mathrm{Span}_{\operatorname{\mathbb{C}}}(\mathrm{d}z^2). Similarly we observe that \nabla^\mathtt I \mathrm{d}\overline w\in \mathrm{Span}_{\operatorname{\mathbb{C}}}(\mathrm{d}\overline w^2). By Equation (?), we directly obtain that \nabla^\mathtt I \mathtt {II} is tri-symmetric if and only if \tau:= \mathrm{d}\alpha \otimes \mathrm{d}z^2 + \mathrm{d}\gamma \otimes \mathrm{d}\overline w^2 is tri-symmetric. In order to conclude the proof, notice that if \alpha and \gamma are holomorphic for J_1 and -J_2 respectively, then \tau= \partial_z\alpha\, \mathrm{d}z^3 + \partial_{\overline w} \gamma \, \mathrm{d}\overline w^3, which is clearly tri-symmetric. Conversely, if \tau is tri-symmetric, imposing \tau( \partial_{\overline z}, \partial_z, \partial_w)= \tau( \partial_z, \partial_{\overline z},\partial_w) and \tau(\partial_{w}, \partial_{\overline w}, \partial_{\overline z})= \tau(\partial_{\overline w},\partial_{w}, \partial_{\overline z}), we get respectively that \partial_{\overline z} \alpha=0 and \partial_w \gamma=0, hence \alpha is J_1–holomorphic and \gamma is (-J_2)–holomorphic.

First, a direct consequence of Theorem 3.3 and Remark 3.4 is that equivariant immersions of \widetilde{S} are uniquely determined (up to post-composition with elements of \mathrm{PSL}(2,\operatorname{\mathbb{C}})\times \mathrm{PSL}(2,\operatorname{\mathbb{C}})) by pairs (\mathtt I,\mathtt {II}) on S, solving Equations (?), (?) and (?), (?). By Theorem 3.8 and Corollary 3.9, this is equivalent to pairs of the form (\mathtt I=\varrho \, \vb*h{(J_1, -J_2)},\mathtt {II}=(q_1+\overline{q_2})/2), together with a solution of Equation (3.9) for h=\vb*h{(J_1, -J_2)}.

We need to check that, given a 5-tuple of the form \gamma=(J, q, -J, \pm\overline q, u) for (J,q,u) in \mathrm{AF}(S) (resp. \mathrm{MGH}(S)), the linearized operator \partial_u\mathscr G is injective. Observe that, at \gamma, \partial_u\mathscr G restricts to a linear operator L\colon C^{\infty}(S,\mathbb R)\to C^{\infty}(S,\mathbb R), and, by \operatorname{\mathbb{C}}–linearity, \partial_u\mathscr G is injective if and only if L is injective.

By Equation (?), we have

for v\in C^{\infty}(S,\mathbb R), where h_J=\vb*h(J,-J). The sign is negative for the case of \mathrm{AF}(S) and positive for the case of \mathrm{MGH}(S). In either case (recall that in the former case we assume e^{-4u}|q|^2/h_J^2<1), we deduce:

with equality if and only if v is identically zero. This concludes the argument for the injectivity of L.

Fix F a fiber of the projection \mathcal C(S)\times \mathcal C(\overline S)\to \mathcal T(S)\times \mathcal T(\overline S). By Proposition 3.14, the intersection F\cap \mathcal B_{reg} is non-empty and connected, hence

is connected. By the construction of \mathbb M(S), the image of this subset via the projection \mathrm M^*(S)\to \mathbb M(S) is just the point \{ [ (J_1, 0, -{J_2}, 0, 0) ]\}, which is therefore the same for all (J_1, -J_2)\in F\cap \mathcal B_{reg}. This proves that the map (3.20) is well-defined.

To see that it is holomorphic, observe that the map in Equation (3.20) is trivially holomorphic, hence Equation (3.20) is holomorphic as it can be locally seen as the composition of holomorphic local inverses of \mathcal B_{reg}\to \mathcal T(S)\times \mathcal T(\overline S) with the map (3.20), and the projection \mathrm M^*(S)\to \mathbb M(S).

Finally, the map (3.20) is an inverse of the submersion \mathbb M(S)\to \mathcal T(S)\times \mathcal T(\overline S) on the first and third factor, hence it is a proper embedding.

We show it for \mathcal{AF}(S), and the proof for \mathcal{MGH}(S) follows in the same fashion. As we discussed in Proposition 3.15, the space of almost-Fuchsian minimal immersions corresponds to the subset of \mathrm M^*(S) given by \mathrm{AF}(S). On this subspace, the fibers of the projection [\pi_0] to \mathbb{QD}(S)\times\mathbb{QD}(\overline S) correspond to the orbits of the (diagonal) action of \mathrm{Diff}_0(S), hence they are connected. As a result, this inclusion descends to a continuous injective map \mathcal{AF}(S)\hookrightarrow\mathbb M_0(S). To check that it is a smooth totally real immersion (see Definition A.1), it is sufficient to observe that, on local charts on which \pi\colon \mathbb M_0(S)\to \mathbb{QD}(S)\times\mathbb{QD}(\overline S) restricts to a biholomorphism, the inclusion of \mathcal{AF}(S) into \mathbb M_0(S) corresponds to the inclusion of \mathbb{QD}(S) given by [J,q]\mapsto ([J,q],[-J, \overline q]), whose image is easily checked to be totally real. Finally, to see that this inclusion is an embedding, observe that the map [J,q]\mapsto [(J,q,-J, \overline q, u)] allows to see \mathcal{AF}(S) as the image of a section, defined on an open subset of \mathbb{QD}(S), of the submersion

where pr_1 is the projection on the first factor: as a consequence, \mathcal{AF}(S) is an embedded submanifold of \mathbb M_0(S).

The original result from Bonsante-El Emam [BEE] is stated only in the holomorphic category. However, the proof argument from [BEE] can also be applied to deduce the result in the smooth setting. Indeed, if we assume that the family {(\mathtt I_{\lambda}, \mathtt {II}_{\lambda})}_{\lambda \in \Lambda} depends smoothly on \lambda, Proposition 5.5 in [BEE] implies that there exists a smooth map \Phi\colon \Lambda\times \widetilde S\to \mathrm{PSL}(2,\mathbb{C})^2 such that the evaluation map of \Phi at any point of \mathrm{PSL}(2,\operatorname{\mathbb{C}}) gives a smooth map \sigma\colon \Lambda\times \widetilde S\to \mathrm{PSL}(2,\operatorname{\mathbb{C}}) as in Proposition 4.2. To see the part of the statement involving the holonomy, observe first of all that \sigma_{\lambda} is invariant for at most one homomorphism \rho_{\lambda}\colon \pi_1(S)\to \mathrm{PSL}(2,\operatorname{\mathbb{C}})^2 by Remark 3.4. More precisely, for all \gamma\in \pi_1(S), \rho_{\lambda} is characterized by the conditions \sigma_{\lambda}\circ \gamma= \rho_{\lambda}(\gamma)\circ \sigma_{\lambda} and N_{\sigma_{\lambda}\circ \lambda} = d( \rho_{\lambda}(\gamma)) (N_{\sigma_{\lambda}}), where N_{\sigma_{\lambda}} is the unit normal of \sigma_{\lambda}. These conditions allow to see that the map \lambda\mapsto \rho_{\lambda}(\gamma) is smooth and compute its derivatives explicitly.

By Proposition 2.10, the map (J_1, -J_2) \mapsto \vb*h{(J_1, -J_2)} is holomorphic. Therefore, the function \gamma\mapsto (\mathtt I_{\gamma}, \mathtt {II}_{\gamma}) is holomorphic as it is the restriction of the following holomorphic map:

The composition of F with the map from Equation (4.4) is a smooth/holomorphic map \Lambda\to (\Gamma(\mathrm{Sym}_2(\operatorname{\mathbb{C}} TS)))^2, by Lemma 4.4. By Remark 4.3, the composed map is smooth/holomorphic pointwise, hence it gives a family \lambda\mapsto (\mathtt I_{\lambda}, \mathtt {II}_{\lambda}) as in the hypothesis of Proposition 4.2. The lemma then follows by Proposition 4.2.

By Lemma 4.5, \mathrm{hol} maps C^{\infty}–curves into C^{\infty}–curves and, hence, it is smooth by Theorem 4.1.

To see that the map is holomorphic, we fix \sigma\in \mathrm M^*(S) and \dot \sigma \in T_{\sigma}\mathrm M^*(S). Then, by taking a local chart, we can find a (small) complex submanifold L of \mathrm M^*(S) of complex dimension 1 passing through \sigma and such that T_{\sigma}L=\mathrm{Span}_{\operatorname{\mathbb{C}}}(\dot \sigma). By Lemma 4.5, the restriction of \mathrm{hol} to L is holomorphic and, hence, \mathrm{d}(\mathrm{hol})_{\sigma}(i \, \dot \sigma) = i \, \mathrm{d}(\mathrm{hol})_{\sigma}(\dot \sigma). We conclude that \mathrm{d}(\mathrm{hol})_{\sigma} is \operatorname{\mathbb{C}}–linear for all \sigma\in \mathrm M^*(S). Thus, \mathrm{hol} is holomorphic.

The map hol is holomorphic by Theorem 4.6, and it is clearly \mathrm{Diff}_0(S)-invariant. The statement now follows from Lemma 4.9.

We already know that \mathrm{hol} is a holomorphic map by Theorem 4.10. Therefore, \mathrm{hol} is a local biholomorphism if and only its differential is injective (and, hence, bijective, as domain and codomain have the same dimension) at every point in \mathbb{M}(S). By Theorem 3.19 and [ESrank2, Proposition 6.6], this is equivalent to proving that, for all \gamma\in \mathrm M^*(S), the differential of \mathrm{hol}\colon \mathrm M^*(S)\to \chi^{*}\big(\pi_1(S),\mathrm{PSL}(2,\operatorname{\mathbb{C}})^2\big) at \gamma has kernel given uniquely by elements of the form \mathscr L_X\gamma for X\in \Gamma(\operatorname{\mathbb{C}} TS).

Consider \dot \gamma\in T_{\gamma}\mathrm M^*(S) and assume that \mathrm{d} \mathrm{hol}_{\gamma}(\dot\gamma)=0. For simplicity, fix a path \gamma_t\in \mathrm M^*(S) starting at \gamma with tangent vector \dot \gamma. Let \sigma_t\colon \widetilde S\to \mathrm{PSL}(2,\operatorname{\mathbb{C}}) be a smooth path of regular minimal immersions with immersion data \gamma_t (constructed, for instance, using Proposition 4.2). Denoting by \rho_t\in \mathrm{Hom}\big(\pi_1(S), \mathrm{PSL}(2,\operatorname{\mathbb{C}})^2\big) the holonomy of \sigma_t, and recalling the description of the tangent space to the character variety from ?, the fact that \mathrm{d} \mathrm{hol}_{\gamma}(\dot \gamma)=0 implies that there exists V\in \mathfrak g:=\mathrm{Lie}\big(\mathrm{PSL}(2,\operatorname{\mathbb{C}})^2\big) such that

Hence, by replacing \sigma_t with \sigma'_t= \exp(-tV)\cdot \sigma_t, we get that \sigma_t\in \mathscr U has holonomy \rho_t'= \exp(-tV) \rho_t \exp(tV) and

Since, by assumption, \sigma'_t is a path of minimal immersions and \gamma is infinitesimally rigid, Proposition 4.17 implies that \dot \sigma'=\mathrm{d}\sigma(\widetilde X) for some X\in \Gamma(\operatorname{\mathbb{C}} TS). By Proposition 4.13, we conclude that \dot\gamma=\mathscr L_X\gamma.

By Lemma 4.15, the fundamental forms vary infinitesimally according to \dot \mathtt I=\mathscr L_X \mathtt I and \dot \mathtt {II}=\mathscr L_X\mathtt {II}. Since the differential map of \gamma\mapsto (\mathtt I_{\gamma}, \mathtt {II}_{\gamma}) is injective and sends \dot \gamma into (\mathscr L_X \mathtt I, \mathscr L_X\mathtt {II}), we conclude that \dot \gamma=\mathscr L_X\gamma thanks to Lemma 4.16.

Denote by \mathtt b the Killing form on \mathfrak{sl}(2,\operatorname{\mathbb{C}}). For \sigma\in \mathscr U, denote by N_{\sigma} a unit normal vector field. Let v\in \operatorname{\mathbb{C}} T_xS be any vector such that \mathtt I_{\sigma}(v,v)\ne 0, so that the triple (v, \mathcal Jv, N_{\sigma}) is an orthogonal basis of T_{\sigma(x)}\mathrm{PSL}(2,\operatorname{\mathbb{C}}). Let \dot \sigma\in T_{\sigma}\mathscr U and denote by \dot \mathtt I_\sigma and \dot \mathtt {II}_\sigma the infinitesimal variations of the first and the second fundamental form, respectively, as defined in the statement.

By definition of the metric on \mathrm{PSL}(2,\operatorname{\mathbb{C}}), we have that, for all w\in \operatorname{\mathbb{C}} T\widetilde S

It follows that a smooth curve \sigma_t in \mathscr U determines a family of tensors \mathtt I_{\sigma_t} that depend smoothly on t. Similarly, \mathcal J_{\sigma}(v) is characterized by the following equations:

We observe that, since v and \mathcal J_{\sigma}(v) form a basis of \operatorname{\mathbb{C}} TS whenever v\neq 0, if equations (?) hold for some v\neq 0, then they hold for every v. As before, a smooth curve \sigma_t in \mathscr U determines smoothly varying tensors \mathcal J_{\sigma_t}. We may also repeat the same argument to show that N_{\sigma_t} and, finally, \mathtt {II}_{\sigma_t} depend smoothly on t if \sigma_t does, by observing that N_{\sigma_t} is uniquely determined by:

while \mathtt {II}_{\sigma_t} is defined as:

So, by the Generalized Boman’s Theorem (see Theorem 4.1), the map (?), (?) is C^{\infty}.

Moreover, a direct computation by differentiating the equations (?), (?), (?) and (?), an infinitesimal variation in the direction i\dot \sigma produces the infinitesimal variations i\dot \mathtt I, i\dot{\mathcal J}_{\sigma}, i \dot N_{\sigma}, i\dot \mathtt {II}. Hence the differential of the map (?), (?) is \operatorname{\mathbb{C}}–linear, and we therefore conclude that it is also holomorphic.

Let \mathrm X\in \Gamma(TS) and let \phi_{t\widetilde{\mathrm X}} denote the lift to the universal cover of the corresponding flow. Then, \frac{\mathrm{d}}{\mathrm{d}t}|_{t=0} ( \sigma\circ \phi_{t\widetilde{\mathrm X}}) = \mathrm{d}\sigma(\widetilde{\mathrm X})\in \Gamma(\sigma^* T\mathrm{PSL}(2,\operatorname{\mathbb{C}})). In particular, we can think of \mathrm{d}\sigma(\widetilde{\mathrm X}) as an element of T_{\sigma}\mathscr U. By the properties of the (real) Lie derivative, this infinitesimal deformation determines an infinitesimal variation of \mathtt I_\sigma and \mathtt {II}_\sigma according to the Lie derivative through \mathrm X. By Lemma 4.14, for all X\in \Gamma(\operatorname{\mathbb{C}} TS)

as claimed.

We start by proving that the differential map is injective. Consider a path \gamma^t=(J_1^t, q_1^t, -J_2^t, \overline q_2^t, u^t)\in \mathrm M^*(S), let \mathtt I_t and \mathtt {II}_t be the image path and assume that both derivatives \dot \mathtt I, \dot \mathtt {II} vanish at t = 0. Let z_t and \overline {w_t} be a path of local coordinates for J_1^t and -J_2^t, respectively. Also denote by \mathcal J_t the path of bi-complex structures induced by (J_1^t,-J_2^t).

We show that the derivative \dot{\mathcal J} of \mathcal J_t at time zero is such that \dot{\mathcal J}=0. The condition \mathcal J_t^2=- id implies \mathcal J\dot{\mathcal J}+\dot{\mathcal J}\mathcal J=0, from which we see that \dot {\mathcal J} maps the i–eigenspace to the (-i) eigenspace and vice versa: equivalently, \dot{\mathcal J}(\partial_{\overline z}) \in \mathrm{Span}(\partial_{w}) and \dot{\mathcal J}(\partial_{w})\in \mathrm{Span}(\partial_{ \overline z}). On the other hand, from \mathtt I_t= \varrho_t \mathrm{d}z_t\mathrm{d}\overline {w_t}, we get \mathtt I_t(\mathcal J_t\cdot, \cdot)+\mathtt I_t(\cdot, \mathcal J_t\cdot)=0, hence \dot \mathtt I=0 implies:

By evaluating the left-hand side in (\partial_{\overline z}, \partial_{\overline z}), we have that \mathtt I(\dot {\mathcal J}(\partial_{\overline z}), \partial_{\overline z})=0: using that \dot {\mathcal J}(\partial_{\overline z})\in Span(\partial_ w), \mathtt I(\partial_{\overline z}, \partial_{\overline z})=0, and that \mathtt I is non-degenerate, we conclude that \dot{\mathcal J}(\partial_{\overline z})=0. In the same fashion, \dot{\mathcal J}(\partial_{ w})=0, hence \dot{\mathcal J}=0.

Now, observe that, for any function v we can write

as one can easily see by evaluating both sides in the basis \{\partial_{\overline z}, \partial_w\}. Hence, \mathrm{d}v\circ \mathcal J= i \mathrm{d}v (resp. \mathrm{d}v\circ \mathcal J= -i \mathrm{d}v) if and only if v is holomorphic for J_1 (resp. -J_2).

By differentiating the expressions \mathrm{d}z_t\circ \mathcal J_t= i \mathrm{d}z_t and \mathrm{d}\overline w_t\circ \mathcal J_t= -i\mathrm{d}\overline w_t at t=0, we get that

We therefore conclude that \dot z is J_1–holomorphic, hence the corresponding Beltrami differential is zero, implying that \dot {c}_1=0. In the same spirit, we get that \dot{\overline w} is (-J_2)–holomorphic and \dot{\overline {J_2}}=0.

Since \dot z and \dot{\overline w} are holomorphic for J_1 and -J_2 respectively, \dot q_1 and \dot{ \overline{q_2}} are multiples of \mathrm{d}z^2 and \mathrm{d}\overline w^2 respectively, hence 0=\dot \mathtt {II}=\dot q_1+\dot{ \overline{q_2}} implies that \dot q_1=0=\dot{ \overline{q_2}}.

Finally, since the projection map \mathrm M^*(S)\to \mathrm{QD}(S)\times\mathrm{QD}(\overline S) has injective differential, we conclude that \dot \gamma=0.

To prove the second part of the statement, observe that the map defined in Equation (4.4) is clearly equivariant for the action of \mathrm{Diff}_0(S), hence its derivative maps \mathscr L_{\mathsf X}\gamma to (\mathscr L_{\mathsf X} \mathtt I_{\gamma}, \mathscr L_{\mathsf X} \mathtt {II}_{\gamma} ) for all \mathsf X\in \Gamma(TS). For X\in \Gamma(\operatorname{\mathbb{C}} TS), we get the analogue result by writing \mathscr L_X= \mathscr L_{\Re(X)}+ i\mathscr L_{\Im(X)} and using that the differential map is \operatorname{\mathbb{C}}–linear.

Let \sigma: \widetilde{S} \to \mathrm{PSL}(2,\operatorname{\mathbb{C}}) be an equivariant, regular minimal immersion. We first prove that \mathrm{d}\sigma(\widetilde X) is tangent to \mathscr U and that \mathrm{d}\mathcal R(\mathrm{d}\sigma(\widetilde X))=(0,0) for every X\in \Gamma(\operatorname{\mathbb{C}} TS). Observe that \mathrm{d}\sigma(\widetilde X) is the derivative at time t = 0 of the path of immersions \sigma_t\in \mathscr U given by \sigma_t (p)= \exp_{\sigma(p)} (t\,\mathrm{d}\sigma(\widetilde X)) (for sufficiently small values of |t|). Since \mathrm{d}\sigma(\widetilde X) is \rho–invariant, the maps \sigma_t : \widetilde{S} \to \mathrm{PSL}(2,\operatorname{\mathbb{C}}) are equivariant with respect to the same representation \rho, independently of t. Moreover, by Lemma 4.15 and the properties of the Lie derivative, the infinitesimal deformation of the mean curvature along \dot \sigma is given by

which equals zero since \sigma is minimal. Hence, we conclude that \mathrm{d}\mathcal R_{\sigma} \big( \mathrm{d}\sigma(\widetilde X)\big)=(0,0) for all X\in \Gamma(\operatorname{\mathbb{C}} TS).

We now suppose that \sigma is infinitesimally rigid and show that the kernel of \mathrm{d}\mathcal R_{\sigma} consists only of elements of the form \mathrm{d}\sigma(X). So, assume \dot \sigma\in T_{\sigma}\mathscr U is such that \mathrm{d}\mathcal R_{\sigma}(\dot\sigma)=(0,0). By Equation (?), we see that \dot \sigma is a \rho–invariant section of \sigma^*T\mathrm{PSL}(2,\operatorname{\mathbb{C}}), or equivalently a \rho–invariant vector field on \mathrm{PSL}(2,\operatorname{\mathbb{C}}) along \sigma. Denoting by N the normal vector field along \sigma (which is also \rho–equivariant), we can write \dot \sigma= \mathrm{d}\sigma(\widetilde X)+ \widetilde fN for some X\in \Gamma(\operatorname{\mathbb{C}} TS) and some complex function f\colon S\to \operatorname{\mathbb{C}}.

Clearly \widetilde fN\in T_{\sigma}\mathscr U, since it is the derivative of the path of immersions \sigma_t=\exp_{\sigma(p)} (t \widetilde fN) for |t| small enough. Moreover, by linearity, we get that

where the last equality holds by the first part of the argument. It remains to show that \mathrm{d}\mathcal{R}_{\sigma} (\widetilde fN)=(0,0) if and only if f=0.

In Appendix ? (see in particular Equation (?)), we show that the t-derivative of H_{\sigma_t} at t=0 has the following expression, that holds on the closed surface S since all quantities of interest are \pi_1(S)-invariant:

where we recall that the expression for \partial_u \mathscr G has been given in (?). By definition of infinitesimal rigidity, \partial_u\mathscr G is bijective as a linear map from C^{\infty}(S,\operatorname{\mathbb{C}}) to itself, and \dot H_\sigma=0 by assumption, so we conclude that f=0. This concludes the proof.

We shall show that the Nijenhuis N_{\mathbb L_{\mathbb R}} vanishes identically, for \mathbb L=\mathbb I,\mathbb J,\mathbb K.

For this purpose, just as for almost complex structures defined on the real tangent bundle, let us introduce a Nijenhuis tensor associated with the endomorphism \mathbb L as follows:

Indeed, since the almost complex structure \mathbf{i} is integrable, the distribution T^{1,0}\mathbb X is involutive, and therefore N_{\mathbb L} defines a well-defined (1,2)-tensor on the holomorphic tangent bundle. As in the classical setting, this condition follows from the fact that the endomorphisms \mathbb L are parallel with respect to the holomorphic Levi–Civita connection (Definition 5.1).

Now, as a consequence of the identity

where \mathsf X=X+\overline{X} and \mathsf Y=Y+\overline{Y}, we deduce that N_{\mathbb L_{\mathbb R}}=0. The relation (?) between the Nijenhuis tensors follows directly from the definition of \mathbb L_{\mathbb R} and from the fact that [\overline{X},Y]=0 for X,Y\in T^{1,0}\mathbb X, after choosing local holomorphic extensions.

We first observe that it is enough to prove that N_{\mathbb L}\equiv 0 for every \mathbb L\in\{\mathbb I,\mathbb J,\mathbb K\}. Indeed, once this is established, the classical formula in [kobayashi1996foundations, Chapter IX, Proposition 4.2], extended to the present holomorphic setting, implies directly that if the symplectic forms \omega_{\mathbb L} are closed, then D\mathbb L=0. In the classical setting of almost complex structures defined on the smooth tangent bundle of a manifold, the vanishing of the Nijenhuis tensor is equivalent to the involutivity of the holomorphic, or equivalently anti-holomorphic, tangent distribution. In our setting, as we shall explain shortly, in order to prove that N_{\mathbb L}\equiv 0, we need to show that both eigendistributions \operatorname{Eig}_{\pm i}(\mathbb L) are involutive. Indeed, imposing the involutivity conditions

and expanding the brackets, one finds that the holomorphic vector field N_{\mathbb L}(X,Y) belongs to both \operatorname{Eig}_i(\mathbb L) and \operatorname{Eig}_{-i}(\mathbb L). Since these two eigenspaces have trivial intersection, this is possible if and only if N_{\mathbb L}(X,Y)=0.

In the following, we will only show that \operatorname{Eig}_i(\mathbb K)\subset T^{1,0}\mathbb X is involutive, since the computations for the (-i)-eigenspace and the other endomorphisms are analogous. Given Y,Z\in\Gamma(T^{1,0}\mathbb X), we have

hence \iota_{Y}\omega_{\mathbb{I}}=\iota_{\mathbb{K}(Y)}\omega_{\mathbb{J}}. In particular, \mathbb{K}(Y)=i \, Y if and only if \iota_Y\omega_{\mathbb{I}}=i\, \iota_{Y}\omega_{\mathbb{J}}. In what follows, we denote by \mathscr L the complex Lie derivative introduced in Section ?, which acts naturally on any space of tensors defined on \mathbb X with complex values, and differentiates them along a complex vector field in \mathbb{C}T \mathbb X. Suppose that \mathbb{K}(Y)=i\, Y and \mathbb{K}(Z)=i\, Z, then

and this is equivalent to \mathbb{K}\big([Y,Z]\big)=i\,[Y,Z].

Set \mathbb{U}:=\mathbb{QD}(S)\times \mathbb{QD}(\overline S). We show that \mathbb I_0 is a holomorphic section of \operatorname{End}_{\operatorname{\mathbb{C}}}(T^{1,0}\mathbb U). Since holomorphicity is a local property, let (z_1,\dots,z_n,w_1,\dots,w_n) be local holomorphic coordinates on \mathbb U, where (z_1,\dots,z_n) are holomorphic coordinates on \mathbb{QD}(S), and (w_1,\dots,w_n) are holomorphic coordinates on \mathbb{QD}(\overline S). Then \frac{\partial}{\partial z_1},\dots,\frac{\partial}{\partial z_n}, \frac{\partial}{\partial w_1},\dots,\frac{\partial}{\partial w_n} is a local holomorphic frame of T^{1,0}\mathbb{U}. By definition of \mathbb I_0, one has

Therefore, with respect to the above holomorphic frame, the matrix of \mathbb I_0 is

Its coefficients are constant, hence holomorphic. Thus \mathbb I_0 is a holomorphic section of \operatorname{End}_{\operatorname{\mathbb{C}}}\big(T^{1,0}\mathbb U\big), as claimed.

We will carry out the proof for the tensor \mathbf{I}, since the argument for \mathbf{I}' is analogous. The complex structure \mathbf{I} is equal to the pull-back, via the map \pi_{\mathcal{AF}} from Section ?, of the endomorphism [\dot\mu, \dot q]\mapsto [i\dot\mu, i\dot q] on the real tangent bundle T\mathbb{QD}(S) (see Proposition 5.8 and Remark 5.10). In accordance with the description of the almost-Fuchsian locus in (?), any v \in T\mathcal{AF}(S)\subset T^{1,0}\mathbb M_0(S) can be written as v=[\dot\mu,\dot q,\dot{\overline \mu},\dot{\overline q},\dot u]. Since \mathbb{I} was constructed as the pull-back of the complex structure \mathbb{I}_0 defined in (?), it follows that applying \mathbb{I} to a vector v \in T\mathcal{AF}(S) described as above yields exactly the action of \mathbf{I} on the model T\mathcal{AF}(S) inside T^{1,0}\mathbb M_0(S).

Regarding the restriction to the almost-Fuchsian locus, it suffices to work at the level of smooth character varieties by diagram (?). In fact,

by Proposition 5.11 point (i) it is enough to prove that the action of the endomorphism \mathbb J_0 on the subspace (\Delta^\dagger)_*\big(H^1(S,\mathfrak{sl}_2(\mathbb{C})_{\mathrm{Ad}\rho})\big) is equal to the multiplication by -i. This is a direct computation by using Equation \eqref{eq: definition J_0}

The last term in the equation belongs to the image of H^1(S,\mathfrak{sl}_2(\mathbb{C})_{\mathrm{Ad}\rho}) via \mathrm{d}\Delta^\dagger; indeed,

On the other hand, regarding the restriction to \mathcal{MGH}(S), we observe that

Last chains of equalities directly implies the claim by diagram (?) and Theorem 5.19.

Since \mathbf{IJ}=\mathbf{K} and \mathbf I'\mathbf{J'}=\mathbf K', we easily get that \mathbb K restricts to \mathbf K on \mathcal{AF}(S) and \mathbf i\circ\mathbb K restricts to \mathbf K' on \mathcal{MGH}(S). Now, the tensors \mathbb K^2 and -\mathrm{Id} are \mathbf i–holomorphic endomorphisms of T^{1,0}\mathbb M_0(S) and they coincide on \mathcal{AF}(S), which is a totally real submanifold of \mathbb M_0(S) of full dimension. By Lemma A.2 and Remark A.3, we conclude that \mathbb K^2=-\mathrm{Id} everywhere on \mathbb{M}_0(S).

The tensors \omega_\mathbb I and \mathbb I are both \mathbf i–holomorphic and \mathbb I restricts to \mathbf I on T\mathcal{AF}(S)\subset T^{1,0}\mathbb M_0(S) (see Proposition 5.22). If we show that \omega_{\mathbb{I}}\vert_{\mathcal{AF}(S)} = \omega_{\mathbf{I}}, then we can conclude the proof by Lemma A.2, since \omega_{\mathbf{I}} is compatible with \mathbf{I} on \mathcal{AF}(S) by Theorem 5.7. In fact, it suffices to prove that

which in turn coincides with \omega_{\mathbf{I}} by Proposition 5.11 point (ii). The differential of the map

is given by \dot\rho\mapsto (\dot\rho,-\overline{{}^t\dot{\rho}'}). Therefore, to understand the restriction \omega_{\mathbb{I}}|_{\mathcal{AF}(S)}, according to diagram (?), we need to compute the element

under the conditions \dot{\rho}_2 = -{\overline{{}^t\dot{\rho}_1}} and \dot{\rho}_2' = -{\overline{{}^t\dot{\rho}_1'}}. Hence, by using the description of Goldman’s symplectic form from Equation (?), the term in (Proof) is equal to

where we selected representatives \sum_h \sigma_{1,h}\otimes \phi_{1,h} and \sum_k \sigma_{1,k}'\otimes \phi_{1,k}' of \dot{\rho}_1 and \dot{\rho}_1', respectively. This concludes the proof.

We begin by observing that \omega_\mathbb{K}=\omega_\mathbb{I}(\mathbb{J}\cdot,\cdot), as the following computation shows:

In particular, by Proposition 5.24 and Lemma 5.26, this implies \omega_{\mathbb K}|_{\mathcal{AF}(S)}=\omega_{\mathbf I}(\mathbf J\cdot,\cdot)=\omega_\mathbf K=2(\mathrm{hol}_{\mathcal{AF}})^*\big(\Im\omega_G\big) (see Proposition 5.11 point (ii)). At this point, it is sufficient to prove the equality between i\big(\pi_r^* \omega_G- \pi_l^*\omega_G\big) and 2\Im\omega_G on the image of the \mathrm{PSL}(2,\operatorname{\mathbb{C}})–character variety via the map \Delta^\dagger (see diagram (?)). Indeed, once the equality is established on the almost-Fuchsian locus, the claim follows by applying Lemma A.2. The computation is similar to the one performed in the proof of Lemma 5.26, namely

where we selected representatives \sum_h \sigma_{1,h}\otimes \phi_{1,h} and \sum_k \sigma_{1,k}'\otimes \phi_{1,k}' of \dot{\rho}_1 and \dot{\rho}_1', respectively.

From Proposition \ref{prop: restriction of I} and Lemma \ref{lem: compatibility_omega_I} we know that the holomorphic metric g=\omega_{\mathbb I}(\cdot,\mathbb I\cdot) restricts to the Riemannian metric \mathbf g on \mathcal{AF}(S). Moreover, by Proposition \ref{prop: restriction of J} we have \mathbb J|_{\mathcal{AF}(S)}=\mathbf J. We can therefore deduce that

and we conclude by Lemma A.2.

Let \dot{\gamma},\dot{\gamma}' be tangent vectors to \mathbb{QD}(S)\times\mathbb{QD}(\bar S) in the image of \mathrm{d}\pi|_{T\mathcal{AF}(S)}. Then

By the discussion preceding the statement, \pi^*\Omega is an \mathbf{i}-holomorphic symplectic form on \mathbb{M}_0(S), and therefore Lemma A.2, together with the identity 2\pi_{\mathcal{AF}}^*\mathrm{Im}(\Omega^{\mathbb{C}})|_{\mathcal{AF}(S)}=\omega_{\mathbf{J}} from Proposition 5.9, implies that \omega_{\mathbb{J}}=-\pi^*\Omega(\mathbb{I}\cdot, \cdot) (see also Remark 3.22).

Thanks to Propositions 5.22, 5.24 and 5.25, we already know that the three complex structures \mathbb{I}, \mathbb{J}, \mathbb{K} restrict to the Donaldson’s structures on \mathcal{AF}(S), while \mathbb{I}, \mathbf i{\circ}\mathbb{J}, \mathbf{i}\circ\mathbb{K} restrict to \mathbf{I}', \mathbf{J}', \mathbf{K}' from [MazSepTamb]. Moreover, in the proof of Lemma 5.26 we showed that \omega_{\mathbb{I}} restricts to \omega_{\mathbf{I}} on the almost-Fuchsian locus, and this is sufficient to deduce that the metric g and the other two symplectic forms \omega_{\mathbb{J}} and \omega_{\mathbb{K}} restrict to \omega_{\mathbf{J}} and \omega_{\mathbf{K}}, thanks to formulas (?), (?), (?). This concludes the first item. Analogously, for the second item it suffices to prove that \omega_{\mathbb{I}} restricts to the symplectic form \omega_{\mathbf{I}'} on the locus \mathcal{MGH}(S). To carry out the computation, similarly to Lemma 5.26 we use the inclusion \iota_{\mathbb{R}} introduced in diagram (?). It suffices to evaluate

where \dot{\rho}_i and \dot{\rho}_i' are variations of representations into \mathrm{PSL}(2,\mathbb{R}) for i=1,2. Therefore, we obtain that

As a consequence of Theorem 5.19 we deduce \mathrm{hol}^*\big(\pi_r^*\omega_G+\pi_l^*\omega_G\big)|_{\mathcal{MGH}(S)}=\omega_{\mathbf I'}.

By what was explained at the beginning of the section, the space of quasi-Fuchsian representations \mathcal{QF}(S) is diffeomorphic to a submanifold of \mathbb{M}_0(S), realized as

By the definition of the complex structure \mathbf{i} on \mathbb M(S), when restricted to T^{1,0}\mathcal{T}(S)\times T^{1,0}\mathcal{T}(\overline{S}), it acts by multiplication by i on both the Beltrami differential \dot{\mu}_1 and the conjugate Beltrami differential \dot{\overline{\mu}}_2. In other words, it coincides with the product complex structure \mathbf{i}_{\mathcal{T}(S)}\times\mathbf{i}_{\mathcal{T}(\overline{S})}. By point (1) of Theorem 5.33, this latter complex structure is biholomorphic to \mathbf{i}_{\mathcal{QF}} on \mathcal{QF}(S). On the other hand, the endomorphism \mathbb{I} defined in (?) and (?) restricts to T^{1,0}\mathcal{T}(S)\times T^{1,0}\mathcal{T}(\overline{S}) and it acts as the multiplication by i on the Beltrami differential \dot{\mu}_1 and by -i on the conjugate Beltrami differential \dot{\overline{\mu}}_2. According to Theorem 5.33 point (2), its restriction coincides with \mathbb{I}_{\mathcal{QF}} on \mathcal{QF}(S). The final statement concerning the symplectic form \omega_{\mathbb{I}} follows from its definition in (?), by a computation entirely analogous to that carried out in the proof of Lemma 5.26, where in this case the inclusion at the level of representations is the diagonal embedding.

The complex bi-Kähler structure on \mathcal{QF}(S) (Theorem 5.33) is completely determined by the triple (\mathbf{i}_{\mathcal{QF}},\mathbb{I}_{\mathcal{QF}},\omega_G), since \mathcal{P}=\mathbf{i}_{\mathcal{QF}}\circ \mathbb{I}_{\mathcal{QF}} and g_{\mathcal{QF}}=\omega_G(\mathcal{P}\cdot,\cdot). In particular,

Recall the commutative diagram in Equation (?) and that \pi and \pi_0 are local biholomorphisms. The \mathbb C^* action on \mathrm M(S) commutes through the projection \mathrm M(S)\to \mathrm{QD}(S)\times \mathrm{QD}(\overline S) with the \operatorname{\mathbb{C}}^*-action on \mathrm{QD}(S)\times \mathrm{QD}(\overline S) given by \zeta\cdot (c_1, \overline{c_2}, q_1, \overline{q_2})= (c_1, \overline{c_2}, \zeta q_1, \zeta^{-1}\overline{q_2}); this action is trivially holomorphic and descends to a well-defined holomorphic action of \mathbb C^* on \mathbb{QD}(S)\times\mathbb{QD}(\overline S).

Now, since topologically we have \mathrm{QD}(S)\times\mathrm{QD}(\overline S)\cong\mathbb{QD}(S)\times \mathbb{QD}(\overline S)\times \operatorname{\mathrm{Diff}}_0(S)\times \operatorname{\mathrm{Diff}}_0(S), every element in \mathrm{QD}(S)\times\mathrm{QD}(\overline S) admits a complete system of neighborhoods such that the intersection with every \operatorname{\mathrm{Diff}}_0(S)\times \operatorname{\mathrm{Diff}}_0(S)-orbit is connected. Using this remark, together with the facts that \pi and \pi_0 are local homeomorphisms and that the \mathbb C^*-action on \mathrm M(S) is continuous, we can say that, for all element \gamma_0\in \mathrm M^*(S), there exist U and U_0 neighborhoods of \gamma and V\subset \mathbb C^* neighborhood of 1 such that:

• for all \zeta\in V and \gamma\in U, \zeta\cdot \gamma\in U_0;

• the restrictions of \pi_0 on U and U_0 and the restrictions of \pi on [U] and [U_0] are all injective, hence biholomorphisms with their respective images;

• the intersection of U with every fiber of the projection to \mathrm M^*(S)\to \mathbb M(S) is connected.

Now, let F be the intersection of U with a fiber of \mathrm M^*(S)\to \mathbb M(S). By construction of \mathbb M(S), [\pi_0(F)]\in \mathbb{QD}(S)\times\mathbb{QD}(\overline S) is a point. The assumption on U implies that, for all \zeta\in V, \zeta\cdot F\subset U_0 is connected, and it is mapped through [\pi_0] to the element \zeta\cdot [\pi_0(F)]. We therefore conclude that [\zeta\cdot F] is one point in \mathbb M(S).

From the commutativity of the diagram

we get that the top horizontal map is holomorphic as a consequence of the holomorphicity of the bottom one and of the fact that the vertical maps are biholomorphisms. Similarly for V\times[U]\to [U_0].

By Lemma 6.1, for every element \gamma_0\in \mathrm M_0(S) there exists a neigbourhood U such that, for all \gamma\in U with [\gamma]=[\gamma_0], the maps z\mapsto [e^{iz}\cdot \gamma] and z\mapsto [e^{iz}\cdot \gamma_0], defined for sufficiently small z\in \operatorname{\mathbb{C}}, coincide and are holomorphic. Hence, they determine the same vector

This defines a locally constant map from the fiber over [\gamma] of \mathrm M^*(S)\to \mathbb M(S) to T_{[\gamma]}^{1,0}\mathbb M(S). Since the fibers are connected, we conclude that the map is well defined.

Finally, the holomorphicity in Lemma 6.1 implies the holomorphicity of the vector field \xi.

Let \gamma \in \mathrm{MGH}(S) (see Equation (?)). By Theorem 6.4 and the third item of Theorem 1.1, we know that, at [\gamma] we have the identities

and similar for \mathbb I and \omega_{\mathbb I}. Since they are holomorphic tensors that vanish along \mathcal{MGH}(S), a totally real submanifold of maximal dimension, they vanish on the whole \mathbb M_0(S) (Lemma A.2 and Remark A.3).

We will prove the stronger statement that the map

is holomorphic.

We first consider finite-dimensional families. Let \Lambda be a finite-dimensional smooth manifold and let \lambda \mapsto \gamma_\lambda \in \mathrm{M}^*(S) be a smooth map. Set

By Lemma 4.4, the tensor \mathtt{I}_\lambda depends smoothly on (\lambda,p)\in\Lambda\times S.

The same is true of \mathcal J_\lambda and \omega_\lambda. Indeed, the two isotropic lines of \mathtt{I}_\lambda are distinct roots of the quadratic equation \mathtt{I}_\lambda(v,v)=0 in the projective tangent space, and therefore vary smoothly with (\lambda,p). Since \mathtt{I}_\lambda is positive, these lines lie in distinct components of \mathbb P_{\mathbb C}(\mathbb C T_pS) \setminus \mathbb P_{\mathbb R}(T_pS), and the orientation of S consistently distinguishes them. Since these lines depend smoothly on \lambda, also the tensor \mathcal J_{\lambda}, defined as i on one line and -i on the other one, depends smoothly on \lambda.

Suppose now that \Lambda is a complex manifold and that \lambda\mapsto\gamma_\lambda is holomorphic. Fix a complex structure J_0 compatible with the orientation of S, and let z be a local J_0-holomorphic coordinate on an open subset U\subset S. Keeping this coordinate fixed as \lambda varies, write

By Lemma 4.4, the coefficients E_\lambda,F_\lambda,G_\lambda are holomorphic in \lambda at every point of U, since the frame (\partial_z,\partial_{\bar z}) is independent of \lambda. By a standard computation, we have a_\lambda^2=E_\lambda G_\lambda-F_\lambda^2, thus the area form \omega_\lambda depends holomorphically on \lambda at each point of S, as well. Together with smoothness in (\lambda,p), this shows that

is holomorphic.

We now pass to the Fréchet setting. The argument above, applied to smooth curves, shows that the map \gamma\mapsto \mathrm{d}A_{\mathtt{I}_\gamma} maps smooth curves to smooth curves in \Omega^2(S,\mathbb C). It is therefore smooth by Theorem 4.1. Moreover, its restriction to every holomorphic disk is holomorphic by the preceding computation, thus its differential is \mathbb C-linear. Hence \gamma\mapsto \mathrm{d}A_{\mathtt{I}_\gamma} is holomorphic.

Finally, integration over S defines a continuous \mathbb C-linear map \Omega^2(S,\mathbb C)\rightarrow\mathbb C, therefore

is holomorphic.

First, observe that \hat{\mathcal A} is invariant under the action of \operatorname{\mathrm{Diff}}_0(S). Indeed, for every \varphi\in\operatorname{\mathrm{Diff}}_0(S), one has \mathtt I_{\varphi^*\gamma}=\varphi^*\mathtt I_\gamma and consequently \mathrm{d}A_{\mathtt I_{\varphi^*\gamma}} = \varphi^*\mathrm{d}A_{\mathtt I_\gamma}. Given that \varphi is orientation-preserving, it follows that \hat{\mathcal A}(\varphi^*\gamma)= \hat{\mathcal A}(\gamma). As a consequence of the \mathrm{Diff}_0(S)-invariance and Lemma 4.9, the functional \hat{\mathcal A} descends to a well-defined holomorphic function \mathcal A:\mathbb M(S)\longrightarrow\mathbb C.

Both terms in the equality are holomorphic 1-forms, so it suffices to show that they coincide along the totally real submanifold \mathcal{MGH}(S). This is indeed the case: along \mathcal{MGH}(S), \mathcal A restricts to the usual area functional, \xi corresponds to \frac{\mathrm{d}}{\mathrm{d}\theta} e^{i\theta}, and \omega_{\mathbb I} restricts to \omega_{\mathbf I'}.

Denote h=g|_{TN}. Let D denote the holomorphic Levi-Civita connection of g. Recall that, according to Definition 2.2, D is the \mathbb{C}-linear extension of an affine connection of \mathbb{X}. In particular, D_V keeps the real tangent bundle T \mathbb{X} invariant whenever V \in \Gamma(T \mathbb{X}). As in the classical Riemannian setting, the submanifold N is totally geodesic if and only if, for any pair of tangent vector fields X, Y \in \Gamma(T N), D_X Y is tangent to N. Since N is maximal totally real, for every p\in N one has a direct sum decomposition T_p\mathbb{X}=T_pN\oplus \mathbf{i}(T_pN) as real vector spaces (see Definition A.1). Let X,Y,Z be smooth vector fields tangent to N. The Koszul formula gives

Every term on the right-hand side is real along N. Hence g(D_XY,Z)\in\mathbb R for every vector field Z tangent to N. Fix p\in N and write D_XY=A+\mathbf{i}B, for some A,B\in T_pN. Since g is complex bilinear,

for every Z\in T_pN. The left-hand side is real, so h(B,Z)=0 for every Z\in T_pN. Since by assumption h is non-degenerate, we conclude that B=0 and, hence, D_XY\in T_pN. Since p \in N was arbitrary, we conclude that D_X Y is tangent to N, as desired.

By Corollary 5.31, the holomorphic metric g restricts to \mathbf g on \mathcal{AF}(S) and to \mathbf g' on \mathcal{MGH}(S), and both restricted metrics are real and non-degenerate. Moreover, Proposition 3.21 shows that both spaces are totally real submanifolds of maximal dimension. The first part of the statement therefore follows from Proposition 6.10. As for the second part, Theorem 5.7 implies, for instance, that \mathcal{T}(S) is totally geodesic in \mathcal{AF}(S). The desired conclusion then follows.

Consider any point [\gamma_0] in \mathcal{QF}(S) and take a neighborhood W_0 inside \mathbb M_0(S) over which the projection \pi to \mathbb{QD}(S)\times\mathbb{QD}(\overline S) gives a homeomorphism with the image. Up to shrinking it, we can assume that an element [\gamma]\in W_0 lies in \mathcal {QF}(S) if and only if \pi([\gamma]) lies in the zero section \mathcal T(S)\times \mathcal T(\overline S)\subset\mathbb{QD}(S)\times\mathbb{QD}(\overline S).

By Lemma 6.1, there exists \operatorname{\varepsilon}>0 and a smaller neighborhood W\subset W_0 of [\gamma] such that, for all |z|<\operatorname{\varepsilon}, the action times e^{iz} can be seen as a map from W to W_0. By Theorem 6.5 and by Remark 6.3, we have that (e^{-iz})^*g=g on W, hence every e^{-iz} determines locally an isometry with the image. On the other hand, since \pi conjugates the \operatorname{\mathbb{C}}^*-actions, an element [\gamma] in W is fixed by e^{-iz} for all |z|<\operatorname{\varepsilon} if and only if \pi([\gamma])\in \mathcal T(S)\times \mathcal T(\overline S), hence if and only if [\gamma]\in W_0\cap \mathcal{QF}(S). By a standard argument (e.g. see [kobayashi2012transformation, Chapter II, Theorem 5.1]), every geodesic tangent to \mathcal{QF}(S) starting at [\gamma_0] is then preserved by the action times each e^{iz}, hence it is contained inside \mathcal{QF}(S), proving that \mathcal{QF}(S) is totally geodesic.

Differentiating the expression for \sigma_{\pm}, we have that

For the second part of the statement, just observe that, by the Gauss equation (?), (?), \mathtt I'_{\pm} is a complex metric if and only if

Since id\pm iB is Codazzi for \mathtt I, \det(id\pm iB)\ne 0 implies that the Levi-Civita connection for \mathtt I'_{\pm} is given by

Hence, by a standard computation (see for instance the proof of [ElE, Proposition 3.9]), one gets that

Let \gamma=(J,q,-J, \overline q, u)\in \mathrm{AF}(S), let z be a local holomorphic coordinate for J and write q=\operatorname{\varphi} \mathrm{d}z^2 and h=\lambda \mathrm{d}z\mathrm{d}\overline z for the hyperbolic metric conformal to J. Then, the isotropic directions of \mathtt I'_{\pm} are given by the complex lines generated by

A simple computation shows that a vector \alpha \partial_z+\beta\partial_{\overline z}\ne 0 determines a point in a connected component of \mathbb P(\operatorname{\mathbb{C}} TS)\setminus \mathbb P(TS) according to the sign of Re(\frac{\alpha-\beta}{\alpha+\beta}) or equivalently of Re(\frac{\alpha+\beta}{\alpha-\beta}) (or just one of the two when the other is not defined). In our setting, it is straightforward to check that

implies that X_{\pm} and Y_{\pm} lie in opposite connected components, implying that \mathtt I'_{\pm} are both positive complex metrics. Finally, \mathrm{AF}(S) is connected, \mathtt I'_{\pm} is clearly a Bers metric when q=0, and Bers metrics are connected components of the space of positive complex metrics of curvature -1: we conclude that 2\mathtt I'_\pm are both Bers metrics.

Denote Z= \mathrm{AF}(S)\cup \mathrm{MGH}(S) and \mathcal Z=\mathcal{AF}(S)\cup \mathcal{MGH}(S)\subset \mathbb M(S).

Since \widehat \Theta is \mathrm{Diff}_0(S)–invariant, it is constant within each fiber of the projection {Z}\to \mathcal Z, defining continuously the map \Theta on Z. Moreover, since \widehat \Theta is also holomorphic, we get that, for all \gamma\in \mathrm W(S) and X\in \Gamma(\operatorname{\mathbb{C}} TS),

Hence, by Proposition 4.8, \mathrm{d}\widehat \Theta is constantly zero on the kernel of the differential map of the projection \mathrm W(S)\to \mathbb M(S).

Since the map \pi\colon \mathrm M^*(S)\to \mathrm{QD}(S)\times\mathrm{QD}(S) is a local biholomorphism, for all \gamma in \mathrm Z we can find a neighborhood U\subset \mathrm W(S) such that the fibers of the projection p\colon \mathrm U\to \mathbb M(S) are connected. The map \widehat \Theta|_{U} is therefore constant among the fibers of U\to \mathbb M(S), hence \widehat \Theta|_U descends to a well-defined map \Theta in a neighborhood of [\gamma].

It is simple to check that this local definition of \Theta is holomorphic. Indeed, using the local biholomorphisms \mathrm M^*(S)\to \mathrm{QD}(S)\times \mathrm{QD}(\overline S) and \mathbb M(S)\to \mathbb{QD}(S)\times\mathbb{QD}(\overline S), we can construct local charts for \mathrm M^*(S) of the form by U_1\times U_2 with U_1\subset \mathbb{M}(S) and U_2\subset \mathrm{Diff}_0(S)\times \mathrm{Diff}_0(S), over which \widehat \Theta(u_1,u_2)= \Theta(u_1): the holomorphicity of \Theta then follows by the holomorphicity of \widehat \Theta.

We are left to prove that these local descriptions of \Theta along Z match. This follows from a variation of standard topological argument (see the proof of Theorem 6.20 in [ESrank2], or the suggested proof for the Generalized inverse function theorem in Exercise 1.8.14 in [GuillPoll] and [GIFT]).

Let m be a Riemannian metric on a neighborhood of \mathcal Z inside \mathbb M(S) such that \mathcal Z_1:=\mathcal{AF}(S), \mathcal Z_2:=\mathcal{MGH}(S) and \mathcal Z_3:=\mathcal{AF}(S)\cap \mathcal{MGH}(S) are all totally geodesic. This can be done for instance by restricting \pi a global biholomorphism from a neighborhood of \mathcal Z inside \mathbb M(S) to an open subset of \mathbb{QD}(S)\times\mathbb{QD}(\overline S) as in Theorem 6.20 in [ESrank2], and by pulling back a product metric coming from a Riemannian metric on \mathbb{QD}(S) for which [c,q]\mapsto [c,-q] is an isometry.

For each \mathcal Z_i, one can construct a locally finite cover \{V_i^j\} of open subsets of \mathbb M(S) such that:

• the construction above defines, on each V_i^j, a local holomorphic extension \Theta_i^j coinciding with \Theta|_{\mathcal Z_i} on V_i^j\cap \mathcal Z_i,

• each \overline {V_i^j} is a geodesically convex ball,

• every point of V_i^j has a unique closest point projection to \mathcal Z_i, with the segment realizing the distance being contained in V_i^j.

With these assumptions, it is clear that V_i^j and V_i^k intersect if and only if V_i^j\cap V_i^k\cap \mathcal Z_i is connected and nonempty, therefore \Theta_i^j and \Theta_i^k coincide on V_i^j\cap V_i^k by uniqueness of the holomorphic continuation of \Theta|_{\mathcal Z_i}. As a result, we obtain a globally defined \Theta_i on the open subset \mathbb W_i=\cup_j V_i^j.

Finally, up to shrinking the \mathbb W_i’s, we can assume that \mathbb W_1\cap \mathbb W_2= \mathbb W_3. Clearly, \Theta_1 and \Theta_3 must coincide on \mathbb W_3 by uniqueness of the complex analytic extension on \mathbb W_3\cap \mathcal Z_1. By applying the same argument for \mathbb W_2\cap \mathbb W_3, we get that on \mathbb W(S):=\mathbb W_1\cup \mathbb W_2 we can define \Theta=\Theta_1\cup \Theta_2.

By construction and by Proposition 5.25 the tensor \mathbf i\circ\mathbb K is holomorphic on \mathbb{W}_0(S) and its restriction to \mathcal{MGH}(S) is equal to \mathbf{K}'. Since the tensor \mathbf i\circ\Theta^*(-J_\chi,J_\chi) is also holomorphic on \mathbb{W}_0(S), if we show that its restriction to the anti-de Sitter locus coincides with \mathbf{K}', then we conclude by the analytic continuation argument (Lemma A.2). This follows from the following chain of equalities:

Equation (8.4) follows immediately combining Equation (?) with our definition of \omega_{\mathbb{I}} in (?) and Proposition 5.27. The right-hand side is bi-holomorphic with respect to (\mathbf{i}, \mathbf{i}\circ\mathbb{J}).
Equation (8.4) holds on the maximal totally real submanifold \mathcal{MGH}(S) by [MazSepTamb, Theorem D], hence it is true everywhere on the open subset \mathbb W_0(S)\subset\mathbb M_0(S) which contains \mathcal{MGH}(S). Again, the right-hand side is manifestly bi-holomorphic with respect to (\mathbf{i}, -\mathbf{i}\circ\mathbb{K}).

By complex analiticity, it is sufficient to prove that if f_1 and f_2 coincide on an open subset of \mathbb M, then they coincide everywhere because \mathbb M is connected. Fix a point x\in Z and, up to composing with local charts around x, see f_1, f_2 as holomorphic maps between open subsets \mathbb C^m\supset U\to V\subset \mathbb C^n, and see Z\cap U as a totally real submanifold of U.

It is now simple to check that f_1 and f_2 have the same power series in the point x\in U. Let X_1, \dots X_m be a local frame of linearly independent vector fields on Z in a neighbrohood of x that, up to shrinking the chart, we can assume to be the whole U\cap Z. On U\cap Z, we clearly have that \partial_{X_k} f_1 =\partial_{X_k}f_2, hence, by holomorphicity of f_1 and f_2, we also have that \partial_{\mathbf J X_k} f_1 =i\partial_{X_k} f_1 =i \partial_{X_k}f_2 =\partial_{\mathbf JX_k}f_2: as a consequence, in every point of U\cap Z, the maps f_1 and f_2 have the same 1-jet as functions on U. In particular the first order partial derivatives \partial_{z_k} f_j induced by the holomorphic coordinates (z_1, \dots, z_n) of \operatorname{\mathbb{C}}^n coincide on U'\cap Z for j=1 and j=2. Applying the same argument by induction on k to the holomorphic functions \partial^k_{z_{i_1}, \dots, z_{i_k}} f_1 and \partial^k_{z_{i_1}, \dots, z_{i_k}} f_2, we conclude that f_1 and f_2 share the same partial derivatives of any order in x. As a consequence, their power series at x have the same coefficients, hence f_1 and f_2 coincide in an open neighborhood of x inside \operatorname{\mathbb{C}}^n.

Decomposing d_X Y |_A \in T_A \mathrm{Mat}(2,\operatorname{\mathbb{C}}) with respect to the direct sum decomposition T_A \mathrm{Mat}(2,\operatorname{\mathbb{C}}) \cong T_A \operatorname{\mathrm{SL}}(2,\operatorname{\mathbb{C}}) \oplus \mathrm{Span}_{\operatorname{\mathbb{C}}}(A), the component along \mathrm{Span}_{\operatorname{\mathbb{C}}}(A) is characterized by the following identity:

Moreover, the tangent component to \operatorname{\mathrm{SL}}(2,\operatorname{\mathbb{C}}) is equal to D_X Y|_A, where D is the holomorphic Levi-Civita connection of \mathrm{SL(2,\operatorname{\mathbb{C}})}.

It suffices to apply Lemma B.1, taking into account that \Delta_{\mathtt I}=\varrho^{-1}\Delta_h and, by the Cayley-Hamilton theorem B^2=-\det(B)\mathrm{id} so that \tr(B^2)=-2\det(B)=-2\det_{\mathtt I}\mathtt {II}=2(q_1\overline{q_2})/(\varrho^2 h^2), where the last identity has been verified in the proof of Corollary 3.9.