With this setting we can use the Cartan-Kähler Theorem to prove existence of local Calabi-Yau structures, but in order to do so effectively we also need to understand the Cartan Test in this contest. We may regard the ideal $\mathcal{ℐ}$  as living on $F$  as well as on $S$  . Let $\tau :F\to S$  and $\overline{\pi }:S\to M$  be the projections and denote the bundle $ker{\tau }_{*}$  by $\mathbf{h}$  . The fibres of $\mathbf{h}$  are canonically isomorphic to $\mathfrak{s}\mathfrak{u}\left(n\right)$  . We are interested in integral elements $E\in {Gr}_k\left(T_{u}F\right)$  (or $E\in {Gr}_k\left(T_{\tau \left(u\right)}S\right)$  ) which are mapped isomorphically onto $T_{\pi \left(u\right)}M$  by ${\pi }_{*}$  (or by ${\overline{\pi }}_{*}$  ). Let us denote the set of such elements by $V_k(\mathcal{ℐ},\pi )$  (or $V_k(\mathcal{ℐ},\overline{\pi })$  respectively). Given $E\in {Gr}_k\left(TS\right)$  we denote by $\tilde{E}$  a lift of $E$  , i.e. an element in ${Gr}_k\left(TF\right)$  such that ${\tau }_{*}\left(\tilde{E}\right)=E$  . Clearly ${\tilde{E}}_1$  , ${\tilde{E}}_2$  are lifts of the same element $E$  if and only if ${\tilde{E}}_1+\mathbf{h}={\tilde{E}}_2+\mathbf{h}$  . One can verify that the following holds $$H\left(\tilde{E}\right)={\tau }_{*}^{-1}\left(H\right(E\left)\right).$$  We deduce that

$$\begin{array}{c}c\left(\tilde{E}\right)={codim}_{T_{u}F}\left(H\right(\tilde{E}\left)\right)={codim}_{T_{\tau \left(u\right)}S}\left(H\right(E\left)\right)=c\left(E\right),\end{array}$$

(4)

which is useful because it is often simpler to compute these numbers on lifts.

Now suppose $E\in V_{2n}(\mathcal{ℐ},\overline{\pi })$  with a lift $\tilde{E}\in V_{2n}(\mathcal{ℐ},\pi )$  . There is a natural integral flag $\tilde{B}=({\tilde{E}}_0,\dots ,{\tilde{E}}_{2n})$  with ${\tilde{E}}_{2n}=\tilde{E}$  . In fact let ${\epsilon }_1,\dots ,{\epsilon }_{2n}$  be the dual standard basis in ${\mathbb{R}}^{2n}$  , then we can define $${\tilde{E}}_k=\{v\in \tilde{E}|{\epsilon }_{k+1}\left(\theta \right(v\left)\right)=\dots ={\epsilon }_{2n}\left(\theta \right(v\left)\right)=0\}.$$  Clearly ${\tau }_{*}$  maps $\tilde{B}$  to a well defined flag $B=(E_0,\dots ,E_{2n})$  with $E_{2n}=E$  .

We want to apply the Cartan Test to prove that $B$  is regular. First of all one can show that near $E$  and $\tilde{E}$  , $V_{2n}(\mathcal{ℐ},\pi )$  and $V_{2n}(\mathcal{ℐ},\overline{\pi })$  are submanifolds of ${Gr}_{2n}\left(TF\right)$  and ${Gr}_{2n}\left(TS\right)$  respectively of the same co-dimension $2nq$  , where $q$  is the co-dimension of $SU\left(n\right)$  in $SO\left(2n\right)$  . For example, when $n=3$  , which is the case we are interested in, they are submanifolds of co-dimension $42$  . To show that $B$  is regular we have to compute the numbers $c_k=codim\left(H\right(E_k\left)\right)$  and show that $C\left(B\right)={\sum }_{k=0}^{2n-1}{c}_k$  is equal to $2nq$  . By ( 4 ) one can do this computation for $\tilde{B}$  , which is easier. Clearly $\tilde{E}\subseteq H\left({\tilde{E}}_k\right)$  .

Therefore $H\left({\tilde{E}}_k\right)=\tilde{E}+{\mathbf{h}}_k$  , where ${\mathbf{h}}_k$  is a subspace of the vertical bundle, i.e.

a suspace of $\mathfrak{s}\mathfrak{o}\left(2n\right)$  . For example suppose $x\in \mathfrak{s}\mathfrak{o}\left(2n\right)$  and $v_1,v_2,\dots ,v_n\in {\tilde{E}}_k$  , then

$$\begin{array}{ccc}d\hat{\Omega }(x,v_1,\dots ,v_n)& =& {\mathcal{ℒ}}_x\hat{\Omega }(v_1,\dots ,v_n)\end{array}$$

$$\begin{array}{ccc}& =& {\mathcal{ℒ}}_x{\Omega }_0\left(\theta \right(v_1),\dots ,\theta (v_n\left)\right),\end{array}$$

where $\mathcal{ℒ}$  denotes the usual Lie derivative. Similar equalities clearly hold for $d\hat{\omega }$  . If we now think of ${\mathbb{R}}^k$  as standardly embedded in ${\mathbb{R}}^{2n}$  , i.e. as the set of vectors whose last $2n-k$  -coordinates are zero, then we see that $${\mathbf{h}}_k=\{x\in \mathfrak{s}\mathfrak{o}(n\left)\right|({\mathcal{ℒ}}_x{\Omega }_0{)}_{|{\mathbb{R}}^k}=({\mathcal{ℒ}}_x{\omega }_0{)}_{|{\mathbb{R}}^k}=0\}.$$  Therefore $$c_k={codim}_{\mathfrak{s}\mathfrak{o}\left(2n\right)}\left({\mathbf{h}}_k\right).$$  So, computing these numbers is a completely algebraic problem. In [2] , Bryant points out that whether $C\left(B\right)=2nq$  holds or not is a property both of $SU\left(n\right)$  and of the choice of its representation. For example he checks that indeed $C\left(F\right)=42$  in the three dimensional case when we identify ${\mathbb{C}}^n$  with ${\mathbb{R}}^{2n}$  via $z\mapsto (x,y)$  , where $z=x+iy$  , i.e. when the complex structure $J$  on ${\mathbb{R}}^{2n}$  is the matrix $$J=\left(\begin{array}{cc}0& I_n\\ -I_n& 0\end{array}\right).$$  In this case one says that $SU\left(n\right)$  is regularly presented. If one represents $SU\left(n\right)$  in a different way then it may not satisfy $C\left(B\right)=2nq$  (see [2] for more details).

We are now ready to sketch Bryant's theorem.

Theorem 4 (Bryant, [2] ) Let $(Y,g)$  be a $3$  -dimensional oriented, connected real-analytic manifold with real-analytic Riemannian metric $g$  . Then there exists a three dimensional Calabi-Yau manifold $X$  and an isometric embedding $\iota :Y\to X$  such that $\iota \left(Y\right)$  is special Lagrangian.

Sketch of proof. It is known that every $3$  -dimensional Riemannian manifold is parallelisable, moreover, when $M$  and $g$  are real-analytic, one can find a real-analytic orthonormal parallelisation.

This means we can assume that we have a globally defined, real-analytic, oriented frame $e=(e_1,e_2,e_3)$  on $Y$  . When $Y$  is embedded in some Calabi-Yau manifold as a special Lagrangian submanifold, $(J{e}_1,J{e}_2,J{e}_3)$  is an orthonormal frame for the normal bundle $\nu \left(Y\right)$  of $Y$  . Via the exponential map we know that a neighbourhood of $Y$  in $X$  is diffeomorphic to a neighbourhood of $Y\times \left\{0\right\}$  in $M=Y\times {\mathbb{R}}^3$  . Moreover, if $({\epsilon }_1,{\epsilon }_2,{\epsilon }_3,{\phi }_1,{\phi }_2,{\phi }_3)$  is the co-frame dual to $(e_1,e_2,e_3,J{e}_1,J{e}_2,J{e}_3)$  , then on ${\iota }^{*}TX$  , where $\iota :Y\to X$  is the embedding, we have $$\Omega =({\epsilon }_1+i{\phi }_1)\wedge ({\epsilon }_2+i{\phi }_2)\wedge ({\epsilon }_3+i{\phi }_3),$$  $$\omega ={\sum }_{k=1}^3{\epsilon }_k\wedge {\phi }_k.$$  Following these observations, it is reasonable to set the problem in the following way. Let $M=Y\times {\mathbb{R}}^3$  and let $({\phi }_1,{\phi }_2,{\phi }_3)$  be the canonical co-frame on ${\mathbb{R}}^3$  , then with the choice of global co-frame $({\epsilon }_1,{\epsilon }_2,{\epsilon }_3,{\phi }_1,{\phi }_2,{\phi }_3)$  on $M$  we let $F=M\times {GL}_6(\mathbb{R})$  and $S=M\times \left({GL}_6\right(\mathbb{R})/SU(3\left)\right)$  . We also define the forms ${\hat{\Omega }}_0$  and ${\hat{\omega }}_0$  on $F$  (and on $S$  ) to be $${\hat{\Omega }}_0=({\epsilon }_1+i{\phi }_1)\wedge ({\epsilon }_2+i{\phi }_2)\wedge ({\epsilon }_3+i{\phi }_3),$$  $${\hat{\omega }}_0={\sum }_{k=1}^3{\epsilon }_k\wedge {\phi }_k.$$  Let ${\sigma }_0:Y\times \left\{0\right\}\to S$  be defined by ${\sigma }_0(y,0)=\tau \left(\right(y,0),I_6)$  , where $I_6$  is the identity matrix. Denote ${\sigma }_0(Y\times \{0\left\}\right)$  by $Y_0$  . The goal is to find an open neighbourhood $X$  of $Y\times \left\{0\right\}$  in $M$  and a section $\sigma :X\to S$  such that $\sigma (y,0)={\sigma }_0(y,0)$  and such that $\sigma \left(X\right)$  is an integral submanifold of the exterior differential system generated by $d{\hat{\Omega }}_0$  and $d{\hat{\omega }}_0$  on $S$  . Clearly ${\sigma }_0^{*}{\hat{\omega }}_0=0$  and ${\sigma }_0^{*}\left({\hat{\Omega }}_0\right)={\epsilon }_1\wedge {\epsilon }_2\wedge {\epsilon }_3$  , and therefore $Y_0$  is an integral manifold.

With a bit of thought one can see that at each $p\in Y_0$  the tangent space $T_p{Y}_0$  is contained in some $E\in V_6(\mathcal{ℐ},\overline{\pi })$  and it is the $E_3$  in the canonical flag ending in $E$  . In particular $Y_0$  is a regular submanifold.

Bryant computes explicitly the (linear)-equations defining ${\mathbf{h}}_3$  and finds that it has dimension $31$  in $M_6(\mathbb{R})$  , i.e. that $c_3=5$  and that $H\left(T_p{Y}_0\right)$  has dimension $29$  (i.e. $dim{\mathbf{h}}_3+dimM-dimSU\left(3\right)$  ). Therefore the extension rank of $Y_0$  is $25$  . We want to apply the Cartan-Kähler Theorem to extend $Y_0$  to a $4$  -dimensional integral manifold whose tangent spaces are in $V_4(\mathcal{ℐ},\overline{\pi })$  .

We need to find a $9$  dimensional submanifold $Z_0$  of $S$  , containing $Y_0$  , such that $T_p{Z}_0\cap H\left(T_p{Y}_0\right)$  is of dimension $4$  and is in $V_4(\mathcal{ℐ},\overline{\pi })$  .

The idea to construct $Z_0$  is the following. Being a linear subspace of $M_6(\mathbb{R})$  , ${\mathbf{h}}_3$  has a complementary subspace $W_0$  , i.e. a subspace of $M_6(\mathbb{R})$  of dimension $5$  such that $W_0\cap {\mathbf{h}}_3=\left\{0\right\}$  . There exists a neighbourhood $U_0$  of $0$  in $W_0$  such that for all $x\in U_0$  , $I_6+x$  is in ${GL}_n(\mathbb{R})$  , therefore we can define the following submanifold of $F$  : $${\tilde{Z}}_0=\left\{\right(p,(y_1,0,0),I_6+x)\in Y\times {\mathbb{R}}^3\times {GL}_6(\mathbb{R}\left)\right|p\in Y,x\in U_0\}.$$  It is a submanifold of dimension $9$  , and since $\mathfrak{s}\mathfrak{u}\left(3\right)\subset {\mathbf{h}}_3$  , $W_0$  is transversal to the $SU\left(3\right)$  orbit, therefore ${\tilde{Z}}_0$  maps down to a well defined $9$  dimensional submanifold $Z_0$  of $S$  , containing $Y_0$  . One can check that indeed $T_p{Z}_0\cap H\left(T_p{Y}_0\right)$  is of dimension $4$  and it is in $V_4(\mathcal{ℐ},\overline{\pi })$  . In fact $T_p{Z}_0\cap H\left(T_p{Y}_0\right)$  is sent via ${\overline{\pi }}_{*}$  isomorphically onto the subspace of $T_{(p,0)}M$  spanned by $T_{p}Y$  and $\partial /\partial y_1$  . We can now apply the Cartan-Kähler Theorem to find an integral, $4$  dimensional submanifold $Y_1$  , such that $Y_0\subset Y_1\subset Z_0$  and whose tangent spaces are all contained in $V_4(\mathcal{ℐ},\overline{\pi })$  . In particular we may assume that $Y_1$  is the image of a section ${\sigma }_1$  of $S$  defined over $\left\{\right(p,(y_1,0,0))\in Y\times {\mathbb{R}}^3|y_1\text{is sufficiently small}\}$  .

We now repeat similar arguments to extend $Y_1$  to a $5$  dimensional integral submanifold containing $Y_1$  which is the image of a section of $S$  defined over a $5$  dimensional submanifold in $M$  . One observes that each tangent space of $Y_1$  is contained in some $E\in V_6(\mathcal{ℐ},\overline{\pi })$  and that it is the $E_4$  of the canonical flag of $E$  , therefore $Y_1$  is a regular submanifold. One computes the equations defining ${\mathbf{h}}_4$  and finds that it is a $24$  dimensional subspace of $M_6(\mathbb{R})$  . Therefore $H\left(T_p{Y}_1\right)$  is $22$  dimensional and $T_p{Y}_1$  has extension rank $17$  . We need to find $Z_1$  , a submanifold of $S$  , of dimension $17$  containing $Y_1$  and with the property that $T_p{Z}_1\cap H\left(T_p{Y}_1\right)$  is of dimension $5$  at every $p$  and it is contained in $V_5(\mathcal{ℐ},\overline{\pi })$  . One finds a subspace $W_1$  of $M_6(\mathbb{R})$  of dimension $12$  which is complementary to ${\mathbf{h}}_4$  and a neighbourhood $U_1$  of $0$  in $W_1$  such that $I_6+x\in {GL}_6(\mathbb{R})$  for all $x\in U_1$  . Then we can define $${\tilde{Z}}_1=\left\{\right((p,y_1,y_2,0),I_6+x)\in Y\times {\mathbb{R}}^3\times {GL}_6(\mathbb{R}\left)\right|p\in Y,x\in U_1\}.$$  We have that ${\tilde{Z}}_1$  maps down to a well defined $17$  dimensional submanifold $Z_1$  in $S$  , containing $Z_0$  and therefore $Y_1$  which has the desired properties.

We apply Cartan-Kähler and we find a five dimensional integral submanifold $Y_2$  containing $Y_1$  which is the image of a section ${\sigma }_2$  defined over the set $\left\{\right(p,y_1,y_2,0)\in Y\times {\mathbb{R}}^3|y_1,y_2\text{are sufficiently small}\}$  .

Finally one can extend one more time to find the desired six dimensional integral submanfold $Y_3$  which is the image of a section $\sigma$  defined over an open neighbourhood $X$  of $Y\times \left\{0\right\}$  in $M$  . This can be done since $h_5$  is $14$  dimensional (i.e. $c_5=22$  ), implying $H\left(T_p{Y}_2\right)$  is $12$  dimensional and the extension rank of $Y_2$  is $6$  . The arguments above carry through. The forms ${\sigma }^{*}{\hat{\Omega }}_0$  and ${\sigma }^{*}{\hat{\omega }}_0$  define a Calabi-Yau structure on $X$  with respect to which $Y\times \left\{0\right\}$  is special Lagrangian and induces the given metric $g$  . $\square$ 

5 SL isometric embedding of $1$  -parameter families

We now show how Bryant's result can be refined to prove an isometric embedding theorem of one parameter families of Riemannian manifolds as one parameter families of special Lagrangian submanifolds in a Calabi-Yau manifold.

First let us recall some important facts. Suppose that $Y$  is a $3$  dimensional manifold and suppose that for some $\kappa >0$  there exists a map $I:Y\times [0,\kappa )\to X$  into some $3$  -dimensional Calabi-Yau manifold $(X,\Omega ,\omega )$  , such that for each $t\in [0,\kappa )$  the map $I_t=I(\cdot ,t)$  is an immersion and $Y_t=I_t\left(Y\right)$  is special Lagrangian. We call $I$  a special Lagrangian variation of Y. The vector field in $X$  along $Y_t$  defined by $$V_t\left(p\right)=\frac{\partial I}{\partial t}(p,t),$$  is called the variational vector field of $I$  . If $V_t$  is normal to $Y_t$  for all $t$  we call $I$  a normal (special Lagrangian) variation. Let $${\theta }_t=I_t^{*}\left({\iota }_{V_t}\omega \right).$$  MacLean has shown that ${\theta }_t$  is a harmonic one-form on $Y$  with respect to the metric $g_t=I_t^{*}g$  where $g$  is the metric on $X$  . We call ${\theta }_t$  the variational harmonic form of the variation $I$  .

We now prove

Theorem 5 Let $Y$  be a real-analytic $3$  -dimensional oriented manifold. Suppose we are given the data of a pair $(g_t,{\theta }_t{)}_{t\in (-\kappa ,\kappa )}$  where $g_t$  is a real-analytic one parameter family of metrics on $Y$  and ${\theta }_t$  a real-analytic one parameter family of $1$  -forms, satisfying

• (i) ${Vol}_{g_t}$  , the volume form with respect to $g_t$  , is independent of $t$  , i.e. $$\frac{\partial }{\partial t}{Vol}_{g_t}=0;$$ 
• (ii) ${\theta }_t$  is harmonic with respect to $g_t$  for all $t\in (-\kappa ,\kappa )$  ;
• (iii) ${\theta }_t\left(p\right)\ne 0$  , for all $t\in (-\kappa ,\kappa )$  and $p\in Y$  .

Then there exists a $3$  -dimensional Calabi-Yau manifold $(X,\Omega ,\omega )$  and a normal special Lagrangian variation $I:Y\times (-\kappa ,\kappa )\to X$  such that for every $t\in (-\kappa ,\kappa )$  , the map $I_t=I(\cdot ,t)$  is an isometry with respect to $g_t$  and ${\theta }_t$  is the variational harmonic one-form with respect to $I$  .

Proof. Let $X_1=Y\times (-\kappa ,\kappa )$  . Define the following one forms on $X_1$  : $${\epsilon }_1(p,t)=\frac{{\theta }_t}{\left|{\theta }_t\right|},$$  $${\phi }_1(p,t)=\left|{\theta }_t\right|dt,$$  where the length of ${\theta }_t$  is computed w.r.t. $g_t$  and it is non-zero because of (iii). We can complete these two forms to a global, real-analytic co-frame $({\epsilon }_1,{\epsilon }_2,{\epsilon }_3,{\phi }_1)$  such that for every $t\in (-\kappa ,\kappa )$  , $\left({\epsilon }_1\right(\cdot ,t),{\epsilon }_2(\cdot ,t),{\epsilon }_3(\cdot ,t\left)\right)$  is an oriented, orthonormal co-frame on $Y$  with respect to $g_t$  .

Now define $M=X_1\times {\mathbb{R}}^2$  and let $(y_1,y_2)$  be coordinates on ${\mathbb{R}}^2$  . Let $({\phi }_2,{\phi }_3)=(d{y}_1,d{y}_2)$  be the standard co-frame on ${\mathbb{R}}^2$  . Let the co-frame bundle $F$  of $M$  be trivialised by $({\epsilon }_1,{\epsilon }_2,{\epsilon }_3,{\phi }_1,{\phi }_2,{\phi }_3)$  , so that we can identify $F$  with $M\times {GL}_6(\mathbb{R})$  and $S$  with $M\times \left({GL}_6\right(\mathbb{R})/SU(n\left)\right)$  . As usual, define on $F$  (and on $S$  ) the forms $${\hat{\Omega }}_0=({\epsilon }_1+i{\phi }_1)\wedge ({\epsilon }_2+i{\phi }_2)\wedge ({\epsilon }_3+i{\phi }_3),$$  $${\hat{\omega }}_0={\sum }_{k=1}^3{\epsilon }_k\wedge {\phi }_k.$$  We let $\mathcal{ℐ}$  be the exterior differential system on $F$  (or $S$  ) generated by $d{\hat{\Omega }}_0$  and $d{\hat{\omega }}_0$  .

Now consider the section of $S$  defined on $X_1\times \left\{0\right\}$  by ${\sigma }_1\left(\right(p,t),0)=\tau \left(\right(p,t),0,I_6)$  . We show that $Y_1={\sigma }_1(X_1\times \{0\left\}\right)$  is an integral submanifold of $(S,\mathcal{ℐ})$  . In fact we have $${\sigma }_1^{*}\left({\hat{\omega }}_0\right)={\epsilon }_1\wedge {\phi }_1={\theta }_t\wedge dt$$  and

$$\begin{array}{ccc}{\sigma }_1^{*}\left({\hat{\Omega }}_0\right)& =& {\epsilon }_1\wedge {\epsilon }_2\wedge {\epsilon }_3+i{\phi }_1\wedge {\epsilon }_2\wedge {\epsilon }_3\end{array}$$

$$\begin{array}{ccc}& =& {\epsilon }_1\wedge {\epsilon }_2\wedge {\epsilon }_3+i\left|{\theta }_t\right|dt\wedge {\epsilon }_2\wedge {\epsilon }_3\end{array}$$

By assumption (ii), ${\theta }_t$  is closed on $Y$  for every $t$  , therefore ${\sigma }_1^{*}\left({\hat{\omega }}_0\right)$  is also closed on $X_1$  . Now observe that ${\epsilon }_1\wedge {\epsilon }_2\wedge {\epsilon }_3$  restricts to the volume form w.r.t. $g_t$  on $Y\times \left\{t\right\}\subset X_1$  , therefore, by assumption (i), it is independent of $t$  , i.e. it is closed on $X_1$  . Moreover, if $◆$  is the Hodge-star operator w.r.t the metric $g_t$  , then $$◆{\theta }_t=\left|{\theta }_t\right|{\epsilon }_2\wedge {\epsilon }_3,$$  therefore, by assumption (ii), $\left|{\theta }_t\right|{\epsilon }_2\wedge {\epsilon }_3$  is closed on $Y$  . It follows that $\left|{\theta }_t\right|dt\wedge {\epsilon }_2\wedge {\epsilon }_3$  , and therefore ${\sigma }_1^{*}\left({\hat{\Omega }}_0\right)$  , is closed in $X_1$  . We conclude that $Y_1$  is an integral submanifold.

To complete the proof we can now follow exactly the steps in Theorem  4 , starting from the second extension. One replaces the $Y_1$  in Theorem  4 , obtained by extending $Y_0$  , with the $Y_1$  we have defined here and follows the argument word by word from then on. $\square$ 

6 Structure preserving torus actions

The action of a Lie group $G$  on an $n$  -dimensional ACY manifold $(X,\Omega ,\omega )$  is called structure preserving if it is Hamiltonian and preserves $\Omega$  . Clearly structure preserving actions induce holomorphic isometries on $X$  , although the converse is not in general true as we may have holomorphic actions which do not preserve $\Omega$  .

We restrict to the case when $G$  is the $m$  -torus $T=U(1{)}^m$  , with $m<n$  , and the action is free. Denote by $\mathfrak{t}=i{\mathbb{R}}^m$  the Lie algebra of $T$  . Let ${\eta }_1,\dots ,{\eta }_m\in \Gamma \left(TX\right)$  be the vector fields induced by the action corresponding to the standard basis $i{e}_1,\dots ,i{e}_m$  of $\mathfrak{t}$  . We denote by ${\mu }_j:X\to \mathbb{R}$  a Hamiltonian of ${\eta }_j$  , i.e. a function such that ${\iota }_{{\eta }_j}\omega =d{\mu }_j$  . The function $\mu :X\to {\mathbb{R}}^m$  given by $\mu =({\mu }_1,\dots ,{\mu }_m)$  is the Hamiltonian of the action.

For any $t=(t_1,\dots ,t_m)\in \mu \left(X\right)$  we can form the manifold $N={\mu }^{-1}\left(t\right)/T^m$  together with its symplectic form ${\omega }_{\text{r}}$  given by symplectic reduction. On $N$  we can also naturally define an $(n-m)$  -form ${\Omega }_{\text{r}}$  as follows. Denote by $\pi :{\mu }^{-1}\left(t\right)\to N$  the projection. Let $v_1,\dots ,v_{n-m}\in T{\mu }^{-1}\left(t\right)$  . Then $${\Omega }_{\text{r}}({\pi }_{*}{v}_1,\dots ,{\pi }_{*}{v}_{n-m})={\iota }_{{\eta }_1}\dots {\iota }_{{\eta }_m}\Omega (v_1,\dots ,v_{n-m}).$$  Goldstein [3] and Gross [6] proved the following theorem:

Theorem 6 (Goldstein, Gross) The manifold $(N,{\Omega }_{\text{r}},{\omega }_{\text{r}})$  is almost Calabi-Yau. Moreover, suppose there exists a $T^m$  -invariant continuous map $g:X\to B$  to an $(n-m)$  -dimensional real manifold $B$  such that the induced maps $g:N\to B$  are sL fibrations for all (or almost all) $t\in \mu \left(X\right)$  . Then $(g,\mu ):X\to B\times {\mathbb{R}}^m$  is a sL fibration on $X$  .

This theorem is our main motivation to study CY manifolds with $T$  action.

In [11] Pedersen and Poon prove an ansatz for Kähler-Einstein metrics with a torus acting through holomorphic isometries. Here we wish to formulate the ansatz for the stronger case of structure preserving actions on CY manifolds. We start by describing the data required in the ansatz.

Let $\pi :E\to N$  be a principal $T$  -bundle over the complex $(n-m)$  -dimensional manifold $N$  . We identify $\mathfrak{t}$  with $i{\mathbb{R}}^m$  . Let $U\subseteq {\mathbb{R}}^m$  be a connected open subset with coordinates $t=(t_1,\dots ,t_m)$  . On $E$  we have the vertical vector fields ${\eta }_1,\dots ,{\eta }_m$  associated to the standard basis of $i{\mathbb{R}}^m$  . Let $W=\left(w_{jk}\right)$  be an $m\times m$  positive definite symmetric matrix of smooth functions on $N\times U$  . Let ${\omega }_{\text{r}}\left(t\right)$  be a smooth family of Kähler forms on $N$  parametrised by $t\in U$  and $\theta \left(t\right)\in {\Omega }^1(E,i{\mathbb{R}}^n)$  a smooth family of connection forms, written in components as $\theta =({\theta }_1,\dots ,{\theta }_m)$  . Assume also that there exists on $N$  a nowhere vanishing holomorphic $(n-m)$  -form ${\Omega }_{\text{r}}$  such that $({\Omega }_{\text{r}},{\omega }_{\text{r}}(t\left)\right)$  is an ACY structure for all $t$  . On the $2n$  -dimensional real manifold $X=E\times U$  we can define the following $T$  -invariant forms

$$\begin{array}{ccc}& \omega ={\pi }^{*}{\omega }_{\text{r}}-i{\theta }_j\wedge d{t}_j,& \end{array}$$

(5)

$$\begin{array}{ccc}& \Omega =(i{)}^m\left({\wedge }_{j=1}^m(w^{jk}d{t}_k-{\theta }_j)\right)\wedge {\pi }^{*}{\Omega }_{\text{r}}.& \end{array}$$

(6)

Proposition 1 (Pedersen-Poon,[11] ) Let $\pi :E\to N$  be a principal $T$  -bundle over the complex $(n-m)$  -dimensional manifold $N$  and let $W=\left(w_{jk}\right)$  , $({\Omega }_{\text{r}},{\omega }_{\text{r}})$  , and $\theta$  be the data defined above. Then the pair of forms $(\Omega ,\omega )$  on $X=E\times U$  written in ( 5 ) and ( 6 ) form an almost Calabi-Yau structure on $X$  if and only if the data satisfy

$$\begin{array}{ccc}& \frac{\partial }{\partial t_j}{\omega }_{\text{r}}=id{\theta }_j,& \end{array}$$

(7)

$$\begin{array}{ccc}& {\left[\frac{\partial {\theta }_i}{\partial t_j}\right]}_{(0,1)}=-\overline{\partial }{w}^{ij}& \end{array}$$

(8)

$$\begin{array}{ccc}& \frac{\partial w^{ij}}{\partial t_k}-\frac{\partial w^{ik}}{\partial t_j}=0,& \end{array}$$

(9)

where in the second equation, $[\cdot {]}_{(0,1)}$  is the projection onto the $(0,1)$  part with respect to the complex structure on $N$  . The operators $d$  and $\overline{\partial }$  are computed on $N$  .

The structure is CY if $({\Omega }_{\text{r}},{\omega }_{\text{r}})$  satisfy

$$\begin{array}{c}{\omega }_{\text{r}}^{n-m}/(n-m)!=(-1{)}^{(n-m)(n-m-1)/2}(i/2{)}^{n-m}det{W}^{-1}{\Omega }_{\text{r}}\wedge {\overline{\Omega }}_{\text{r}}.\end{array}$$

(10)

Moreover any ACY manifold $(X,\Omega ,\omega )$  with a free structure preserving $T$  -action can be described locally by such a construction.

Proof. We use Theorem  1 . Clearly $\Omega$  is locally decomposable since this is true of ${\Omega }_{\text{r}}$  .

Notice that since $\theta$  has values in $i{\mathbb{R}}^m$  $$\overline{\theta }=-\theta .$$  It is not difficult to prove the identity

$$\begin{array}{c}\Omega \wedge \overline{\Omega }=(-2{)}^{m}det{W}^{-1}d{t}_1\wedge \dots \wedge d{t}_m\wedge {\theta }_1\wedge \dots \wedge {\theta }_m\wedge {\pi }^{*}({\Omega }_{\text{r}}\wedge {\overline{\Omega }}_{\text{r}}).\end{array}$$

(11)

Since ${\theta }_1\wedge \dots \wedge {\theta }_m$  restricts to a volume form on the vertical space of $E$  , ( 11 ) implies condition (iii) of Theorem  1 . Condition (iv) easily follows from the identity $${\theta }_j\wedge d{t}_j=-\frac{w_{jl}}{2}(w^{lk}d{t}_k-{\theta }_l)\wedge (w^{jm}d{t}_m+{\theta }_j).$$  A tedious but straight forward computation, which uses ( 11 ), also shows that if ${\omega }_{\text{r}}$  and ${\Omega }_{\text{r}}$  satisfy ( 10 ) then $\omega$  and $\Omega$  satisfy condition $\left(5\right)$  . Finally one can easily check that $\Omega$  and $\omega$  are closed if and only if equations ( 7 ), ( 8 ) and ( 9 ) are satisfied.

We now prove that every ACY manifold with a structure preserving free $T$  action can be described by the ansatz. Let $\mu =({\mu }_1,\dots ,{\mu }_n)$  be the moment map of the action. Then, for $t\in \mu \left(X\right)$  let $E_t={\mu }^{-1}\left(t\right)$  and $N_t={\mu }^{-1}\left(t\right)/T$  with the reduced ACY structure $\left({\Omega }_{\text{r}}\right(t),{\omega }_{\text{r}}(t\left)\right)$  . Let $J$  denote the almost complex structure of $X$  . Define

$$\begin{array}{c}{w}_{jk}=\langle {\eta }_j,{\eta }_k\rangle =\omega ({\eta }_j,J{\eta }_k),\end{array}$$

(12)

then for fixed $t$  , the matrix $W=\left(w_{jk}\right)$  is a symmetric positive definite matrix of functions on $N_t$  . Now denote

$$\begin{array}{c}{\xi }_j=w^{jk}J{\eta }_k,\end{array}$$

(13)

The ${\xi }_j$  's are linearly independent vector fields orthogonal to ${\mu }^{-1}\left(t\right)$  . Using the fact that the ${\eta }_j$  's are Killing holomorphic fields one can show that $$[{\xi }_j,{\xi }_k]=0.$$  Therefore the flows ${\Phi }_{{\xi }_j}^{t_j}$  of the ${\xi }_j$  's, which exist for sufficiently small $t_j$  's, define coordinates $t=(t_1,\dots ,t_m)$  on the leafs of the distribution defined by the ${\xi }_j$  's. Let $${\Phi }^t={\Phi }_{{\xi }_1}^{t_1}\circ \dots \circ {\Phi }_{{\xi }_m}^{t_m}.$$  The fact that

$$\begin{array}{ccc}d{\mu }_i\left({\xi }_j\right)& =& \omega ({\eta }_i,w^{jk}J{\eta }_k)\end{array}$$

$$\begin{array}{ccc}& =& w_{ik}{w}^{jk}={\delta }_{ij}.\end{array}$$

implies that $$\mu \left({\Phi }^t\right(p\left)\right)=t$$  when $p\in {\mu }^{-1}\left(0\right)$  . Therefore ${\Phi }^t$  identifies $E_0={\mu }^{-1}\left(0\right)$  with $E_t={\mu }^{-1}\left(t\right)$  .

Now, since ${\eta }_k$  commutes with ${\xi }_j$  , ${\Phi }^t$  is also equivariant, i.e. $N_0$  and $N_t$  are also naturally identified. Let $E=E_0$  and $N=N_0$  . If we take $U\subseteq {\mathbb{R}}^m$  to be an open neighbourhood of $0$  such that ${\Phi }^t$  is defined for $t\in U$  , then ${\Phi }^t$  identifies $E\times U$  with ${\mu }^{-1}\left(U\right)$  in an equivariant way, moreover $\mu$  is identified with the projection onto $U$  .

The reduced ACY structure on $N_t$  can be considered as a family of ACY structures $\left({\Omega }_{\text{r}}\right(t),{\omega }_{\text{r}}(t\left)\right)$  on $N$  . We now prove

$$\begin{array}{c}\frac{\partial }{\partial t_j}{\Omega }_{\text{r}}=0.\end{array}$$

(14)

Goldstein [3,Lemma1] proved that ${\iota }_{{\eta }_1}\dots {\iota }_{{\eta }_m}\Omega$  is closed. Therefore ( 14 ) follows from $${\mathcal{ℒ}}_{{\xi }_j}({\iota }_{{\eta }_1}\dots {\iota }_{{\eta }_m}\Omega )={\iota }_{{\xi }_j}d{\iota }_{{\eta }_1}\dots {\iota }_{{\eta }_m}\Omega +d{\iota }_{{\xi }_j}{\iota }_{{\eta }_1}\dots {\iota }_{{\eta }_m}\Omega =0.$$  where the second summand vanishes because $\Omega$  is of type $(n,0)$  .

The Riemannian metric on $X$  also induces a family of connections $\theta \left(t\right)=\left({\theta }_1\right(t),\dots ,{\theta }_m(t\left)\right)$  on $E$  . Proving that ${\Phi }^t$  identifies $\Omega$  and $\omega$  on $X$  with the forms in ( 6 ) and ( 5 ) respectively is just a matter of a simple verification.

The fact that $({\Omega }_{\text{r}},{\omega }_{\text{r}})$  always satisfy ( 10 ) whenever $(\Omega ,\omega )$  satisfy property (v) in Theorem  1 is also a rather straight forward computation. $\square$ 

Remark 1. The way we stated the Pedersen-Poon ansatz is different from how it was stated in the original paper. This is partially due to the fact that we are only considering the case of Ricci-flat metrics where the action preserves also the holomorphic form. Our formulation is also more convenient for the applications in the following sections. $\square$ 

7 Isometric embeddings of torus bundles

We now prove, using essentially Proposition  1 and the Cauchy-Kowalesky theorem, the following isometric embedding theorem for principal $T^m$  bundles over an ACY base.

Theorem 7 Let $\pi :E\to N$  be a principal $T^m$  -bundle over an $(n-m)$  -dimensional ACY manifold $(N,{\Omega }_0,{\omega }_0)$  with connection ${\theta }_0$  whose curvature is of type $(1,1)$  . Let $W_0$  be a positive definite $m\times m$  symmetric matrix of functions on $N$  such that

$$\begin{array}{c}{\omega }_0^{n-m}/(n-m)!=(-1{)}^{(n-m)(n-m-1)/2}(i/2{)}^{n-m}det{W}_0^{-1}{\Omega }_0\wedge {\overline{\Omega }}_0.\end{array}$$

(15)

If all the data is real-analytic, then there exists a neighbourhood $U\subset {\mathbb{R}}^m$  of $0$  and a $T^m$  -invariant CY structure $(\Omega ,\omega )$  on $E\times U$  , inducing the connection ${\theta }_0$  and the matrix $W_0$  on $E\times \left\{0\right\}$  . Moreover the projection onto $U$  is the Hamiltonian of the action and the reduced space over $t=0$  is $(N,{\Omega }_0,{\omega }_0)$  .

Proof. By Proposition  1 we can assume ${\Omega }_{\text{r}}={\Omega }_0$  . For given ${\omega }_{\text{r}}$  , denote by $f_{{\omega }_{\text{r}}}$  the unique function such that $${\omega }_{\text{r}}^{n-m}/(n-m)!=(-1{)}^{(n-m)(n-m-1)/2}(i/2{)}^{n-m}{f}_{{\omega }_{\text{r}}}{\Omega }_{\text{r}}\wedge {\overline{\Omega }}_r.$$  For any $m$  -tuple of positive numbers $\epsilon =({\epsilon }_1,\dots ,{\epsilon }_m)$  denote $$U_l=\left\{\right(x_1,\dots ,x_m)\in {\mathbb{R}}^m|x_j<{\epsilon }_j,1\le j\le l;x_j=0,l+1\le j\le m\}.$$  For fixed $1\le l\le m$  consider the following system of equations $$\left(S_l\right)\{\begin{array}{c}\frac{\partial }{\partial t_l}{\omega }_{\text{r}}=id{\theta }_l,\\ {\left[\frac{\partial {\theta }_i}{\partial t_l}\right]}_{(0,1)}=-\overline{\partial }{w}^{il},i=1,\dots ,m\\ \frac{\partial w^{ij}}{\partial t_l}=\frac{\partial w^{il}}{\partial t_j},j\le i,j\le l\\ \frac{\partial w^{ij}}{\partial t_l}=\frac{\partial w^{jl}}{\partial t_i},i\le j\le l\\ det{W}^{-1}=f_{{\omega }_{\text{r}}}\end{array}$$  We regard this system as an evolution equation with time variable $t_l$  . It is a well defined system on $N\times {\mathbb{R}}^l$  . We now show that by subsequently applying the Cauchy-Kowalesky Theorem, we can find an $m$  -tuple $\epsilon =({\epsilon }_1,\dots ,{\epsilon }_m)$  such that for every $l=1,\dots ,m$  , there is a solution $({\omega }_{\text{r}},W,\theta {)}_l$  to $\left(S_l\right)$  on $U_l$  .

Moreover $({\omega }_{\text{r}},W,\theta {)}_l$  coincides with $({\omega }_{\text{r}},W,\theta {)}_{l-1}$  on $U_{l-1}$  .

By induction assume that for some $l\le m$  , we have constructed a real-analytic $({\omega }_{\text{r}},W,\theta {)}_{l-1}$  on $U_{l-1}$  which satisfies $\left(S_k\right)$  for all $k\le l-1$  .

We now solve $S_l$  on some $U_l$  extending $U_{l-1}$  with the initial conditions $$({\omega }_{\text{r}},W,\theta ){|}_{t_l=0}=({\omega }_{\text{r}},W,\theta {)}_{l-1}.$$  Notice that $S_l$  is underdetermined, in fact the $w^{ij}$  's with $j>l$  appear only in the last equation. Moreover, if we ignore the last equation, the unknowns $w^{ij}$  and ${\theta }_i$  with $i>l$  and $j\le l$  , appear only in the second and third set of equations which are independent from the others. To solve the second and third set of equations with $i>l$  we can arbitrarily choose real analytic $w^{il}$  's extending the initial conditions and then obtain $w^{ij}$  and ${\theta }_i$  by integration.

Now also arbitrarily extend $w^{ij}$  with $i,j>l$  and set $w^{ij}$  with $j>l$  and $i\le l$  equal to $w^{ji}$  (this is coherent with requiring $W$  to be symmetric). One can now see that $\left(S_l\right)$  restricted to the cases $i\le l$  is completely determined and a unique solution exists by Cauchy-Kowalesky on some $U_l$  .

Suppose that on $U_{l-1}$  , $w^{ij}=w^{ji}$  for all $i$  and $j$  , we now show that the same continues to be true on $U_l$  for the solution just constructed. By construction this is true for all $i,j$  with $i>l$  or $j>l$  . So assume $i\le j\le l$  .

We have $$\frac{\partial }{\partial t_l}(w^{ij}-w^{ji})=\frac{\partial w^{jl}}{\partial t_i}-\frac{\partial w^{jl}}{\partial t_i}=0,$$  where in the first equality we have used the third and fourth equation of $\left(S_l\right)$  . Now the claim follows from $(w^{ij}-w^{ji}){|}_{t_l=0}=0$  .

We now show that $({\omega }_{\text{r}},W,\theta {)}_l$  continues to satisfy $\left(S_k\right)$  for all $k<l$  on $U_l$  .

First we show that the third equation of $\left(S_k\right)$  holds. Assume first $j\le k\le i$  .

Then $$\frac{\partial }{\partial t_l}\left(\frac{\partial w^{ij}}{\partial t_k}-\frac{\partial w^{ik}}{\partial t_j}\right)=\frac{\partial }{\partial t_k}\left(\frac{\partial w^{ij}}{\partial t_l}-\frac{\partial w^{il}}{\partial t_j}\right)=0,$$  where both equal signs follows from applying the third equation of $\left(S_l\right)$  .

Now, assuming $j\le i\le k$  , we have

$$\begin{array}{ccc}\frac{\partial }{\partial t_l}\left(\frac{\partial w^{ij}}{\partial t_k}-\frac{\partial w^{ik}}{\partial t_j}\right)& =& \frac{{\partial }^2{w}^{ij}}{\partial t_k\partial t_l}-\frac{{\partial }^2{w}^{kl}}{\partial t_j\partial t_i}=\end{array}$$

$$\begin{array}{ccc}& =& \frac{{\partial }^2{w}^{ij}}{\partial t_k\partial t_l}-\frac{{\partial }^2{w}^{lk}}{\partial t_j\partial t_i}=\end{array}$$

$$\begin{array}{ccc}& =& \frac{\partial }{\partial t_k}\left(\frac{\partial w^{ij}}{\partial t_l}-\frac{\partial w^{li}}{\partial t_j}\right)=0,\end{array}$$

where in the first equality we have used the fourth equation, in the second we used symmetry, in the third we used the previous case and in the last the third equation. We now show that the fourth equation of $\left(S_k\right)$  holds.

Assume $i\le j\le k<l$  , then

$$\begin{array}{ccc}\frac{\partial }{\partial t_l}\left(\frac{\partial w^{ij}}{\partial t_k}-\frac{\partial w^{jk}}{\partial t_i}\right)& =& \frac{{\partial }^2{w}^{ij}}{\partial t_k\partial t_l}-\frac{{\partial }^2{w}^{kl}}{\partial t_i\partial t_j}=\end{array}$$

$$\begin{array}{ccc}& =& \frac{{\partial }^2{w}^{ij}}{\partial t_k\partial t_l}-\frac{{\partial }^2{w}^{lk}}{\partial t_i\partial t_j}=\end{array}$$

$$\begin{array}{ccc}& =& \frac{\partial }{\partial t_k}\left(\frac{\partial w^{ij}}{\partial t_l}-\frac{\partial w^{lj}}{\partial t_i}\right)=0,\end{array}$$

where the first equality follows from the fourth equation of $\left(S_l\right)$  , the second from symmetry, the third from the third equation of $\left(S_k\right)$  and the last from symmetry and the fourth equation of $\left(S_l\right)$  . This proves that the fourth equation of $\left(S_k\right)$  continues to hold on $U_l$  . Notice also that if $i\le k<l$  , we have

$$\begin{array}{c}\frac{\partial w^{ik}}{\partial t_l}=\frac{\partial w^{kl}}{\partial t_i}=\frac{\partial w^{lk}}{\partial t_i}=\frac{\partial w^{li}}{\partial t_k}=\frac{\partial w^{il}}{\partial t_k}.\end{array}$$

(16)

We now prove that the second equation of $\left(S_k\right)$  continues to hold. We have

$$\begin{array}{ccc}\frac{\partial }{\partial t_l}\left({\left[\frac{\partial {\theta }_i}{\partial t_k}\right]}_{(0,1)}+\overline{\partial }{w}^{ik}\right)& =& {\left[\frac{{\partial }^2{\theta }_i}{\partial t_k\partial t_l}\right]}_{(0,1)}+\overline{\partial }\left(\frac{\partial w^{ik}}{\partial t_l}\right)=\end{array}$$

$$\begin{array}{ccc}& =& \frac{\partial }{\partial t_k}\left({\left[\frac{\partial {\theta }_i}{\partial t_l}\right]}_{(0,1)}+\overline{\partial }{w}^{il}\right)=0,\end{array}$$

where in the second equality we used the third equation of $\left(S_l\right)$  or ( 16 ) and in the last we used the second equation of $\left(S_l\right)$  . Notice also that by symmetry and by the fact that $\theta$  has imaginary values we have $$\frac{\partial {\theta }_i}{\partial t_k}=\frac{\partial {\theta }_k}{\partial t_i}.$$  Using this fact it is easy to show that also the first equation of $\left(S_k\right)$  continues to hold. We leave it to the reader. This concludes the proof that on some $U=U_m$  there exists a $({\omega }_{\text{r}},W,\theta )=({\omega }_{\text{r}},W,\theta {)}_m$  which satisfies $\left(S_l\right)$  for all $l$  and such that $W$  is symmetric. Together with ( 16 ), this also proves that $({\omega }_{\text{r}},W,\theta )$  is a solution to the system in Proposition  1 and satisfies the required initial conditions. $\square$ 

Remark . It is interesting that in the case of the circle bundle, the Calabi-Yau $X$  constructed by this theorem is unique. This means that the geometry of $X$  is completely determined by the initial data on $\pi :E\to N$  .

$\square$ 

8 From $T$  actions to sL submanifolds

We apply Theorems  7 and we combine it with Theorem  6 to construct new families of compact sL submanifolds. All the data is assumed to be real-analytic.

Example 1. Let $N$  be the complex torus ${\mathbb{C}}^n/\Lambda$  , where $\Lambda$  is a $2n$  -dimensional lattice generated by vectors $e_1,\dots ,e_n,f_1,\dots ,f_n$  , where $e_1,\dots ,e_n$  is the standard basis of ${\mathbb{C}}^n$  and $f_1,\dots ,f_n$  are arbitrary. Choose holomorphic line-bundles $L_1,\dots ,L_m$  with $U\left(1\right)$  -connections ${\theta }_1,\dots ,{\theta }_m$  . Let $U_1,\dots ,U_m$  be the associated $U\left(1\right)$  -bundles respectively. Let $E=U_1\oplus \dots \oplus U_m$  . Suppose $${\omega }_0=\frac{i}{2}\sum h_{jk}d{z}_j\wedge d{\overline{z}}_k$$  is a Kähler form on $N$  whose coefficients $h_{jk}$  are all real valued. As holomorphic $n$  -form we take the standard one $${\Omega }_0=d{z}_1\wedge \dots \wedge d{z}_n.$$  Define $\theta \left(0\right)=({\theta }_1,\dots ,{\theta }_m)$  and choose a matrix $W_0$  satisfying the conditions of Theorem  7 . It then follows that we have a CY structure on $X=E\times U$  for some $U\subseteq {\mathbb{R}}^m$  with the properties of Theorem  7 . Now, if $Im\Lambda \subseteq {\mathbb{R}}^n$  is the lattice spanned by $Im{f}_1,\dots ,Im{f}_n$  , let $B={\mathbb{R}}^n/Im\Lambda$  . From our choice of ${\Omega }_0$  and ${\omega }_0$  it follows that the map $f:N\to B$  given by $$f(z_1,\dots ,z_n)=(Im{z}_1,\dots ,Im{z}_n)$$  is an $n$  -dimensional sL fibration on $(N,\Omega (0),\omega (0\left)\right)$  . Therefore, using Theorem  6 , we can lift it to a family of sL $n+m$  -tori of $X$  .

Take for example $m=1$  and $U_1=N\times S^1$  with ${\theta }_1=ids$  , where $s$  is the coordinate on $S^1$  . In this case, $W_0$  is a function uniquely determined by ${\omega }_0$  and ${\Omega }_0$  via ( 15 ). The data $(U_1,{\Omega }_0,{\omega }_0,{\theta }_0)$  can be interpreted as an $n$  -parameter family of Riemannian $n+1$  -tori $T_y=S^1\times f^{-1}\left(y\right)$  , where $y\in B$  .

The metric on $T_y$  is given by $g_y=det\left(h^{jk}\right)d{s}^2+h_{jk}d{x}_{j}d{x}_k$  , where the $x_k$  's are the coordinates on $f^{-1}\left(y\right)$  . Theorem  7 embeds this family isometrically into $X$  . We obtain different metrics by choosing more general ${\theta }_0$  's. $\square$ 

Example 2. The following example is taken from Bryant [1] . Let $c:{\mathbb{C}}^n\to {\mathbb{C}}^n$  be the standard conjugation. The map $c$  obviously descends to a map of $\mathbb{C}{\mathbb{P}}^{n-1}$  (which we still call $c$  ). The fixed locus of $c$  is $\mathbb{R}{\mathbb{P}}^{n-1}\subseteq \mathbb{C}{\mathbb{P}}^{n-1}$  . Let $N$  be a smooth algebraic hyper-surface of $\mathbb{C}{\mathbb{P}}^{n-1}$  with trivial canonical bundle.

Assume $N$  to be invariant under $c$  , for example $N$  can be originating from a homogeneous polynomial of degree $n$  in ${\mathbb{C}}^n$  with real coefficients. Let $Y$  be the fixed locus of $c$  restricted to $N$  , i.e. $Y=N\cap \mathbb{R}{\mathbb{P}}^{n-1}$  . On $N$  we have a natural Kähler form ${\omega }_0$  , namely the restriction of the Fubini-Study metric, with respect to which $Y$  is Lagrangian. It is also possible to chose a nowhere vanishing holomorphic $(n-2)$  -form ${\Omega }_0$  on $N$  such that $c^{*}{\Omega }_0={\overline{\Omega }}_0$  . Given such a form we have $Im{\Omega }_{0|Y}=0$  . Thus $Y$  is sL with respect to the ACY structure $(N,{\Omega }_0,{\omega }_0)$  . We can now take $E=U_1\oplus \dots \oplus U_m$  , where the $U_j$  's are holomorphic $U\left(1\right)$  -bundles on $N$  with $U\left(1\right)$  -connections ${\theta }_1,\dots ,{\theta }_m$  . We apply the corollary to construct a Calabi-Yau metric on $X=E\times U$  for some $U\to {\mathbb{R}}^m$  . Theorem  6 then allows us to lift $Y$  to an $(n+m-2)$  -dimensional sL submanifold. It was shown by Bryant [1] that in the cases $n=3,4$  and $5$  , $Y$  can also be a real torus. $\square$ 

Example 3 This is a particular case of Example 1. Let $N={\mathbb{C}}^{*}$  and $\pi :E\to N$  be a $T^2$  principal bundle with a connection ${\Theta }_0$  . We let ${\Omega }_{\text{r}}=du$  .

Given a $(2\times 2)$  positive definite matrix of functions $W_0$  on $N$  , define $${\omega }_0=\frac{i}{2}det{W}_0^{-1}{\Omega }_0\wedge {\overline{\Omega }}_0.$$  Then Theorem  7 gives us a $T^2$  -symmetric CY structure on $X=E\times (-\epsilon ,\epsilon {)}^2$  , for some $\epsilon$  .

Now consider $f:N\to \mathbb{R}$  given by $u\mapsto log\left|u\right|$  . Clearly $f$  is sL with respect to the ACY structure $(du,{\omega }_0)$  . Lifting it, we obtain that $E\times \left\{0\right\}\times \left\{0\right\}$  is a one parameter family of sL manifolds inside $X$  . This example therefore could have been constructed also using Theorem  5 . In the case $E=T^2\times N$  , we will study this example in more detail in the next section.

9 SL fibrations with $T^2$  symmetry

Let $B\subseteq {\mathbb{R}}^3$  be open and let $u=x+iy$  be the complex coordinate on $\mathbb{C}$  . Define

$$\begin{array}{c}\tilde{X}=\left\{\right({\phi }_1,{\phi }_2,u,t_1,t_2)\in S^1\times S^1\times \mathbb{C}\times \mathbb{R}\times \mathbb{R}|(t_1,t_2,Imu)\in B\}\end{array}$$

(17)

Let $\tilde{N}={\pi }_{\mathbb{C}}\left(\tilde{X}\right)$  , where ${\pi }_{\mathbb{C}}$  is the projection on the $\mathbb{C}$  component. We assume that on $\tilde{X}$  we have a CY structure such that the obvious $T^2$  action on $({\phi }_1,{\phi }_2)$  is structure preserving, the coordinates $t_j$  are Hamiltonians for ${\eta }_j=\partial /\partial {\phi }_j$  and $\tilde{N}$  is the reduced space such that ${\Omega }_{\text{r}}=du$  . The CY structure is then described by Proposition  1 . We recall the main data involved. We have a $(2\times 2)$  matrix $W$  and a connection $\Theta =({\theta }_1,{\theta }_2)$  , which we write as $${\theta }_j={\alpha }_{j}du-{\overline{\alpha }}_{j}d\overline{u}+id{\phi }_j.$$  We then have the reduced Kähler form on $N$  given by $${\omega }_{\text{r}}=\frac{i}{2}det{W}^{-1}du\wedge d\overline{u}.$$  This data satisfies the equations of Proposition  1 . The forms $\omega$  and $\Omega$  on the total space $\tilde{X}$  are

$$\begin{array}{ccc}\omega & =& {\pi }^{*}{\omega }_{\text{r}}-i{\theta }_j\wedge d{t}_j=\end{array}$$

$$\begin{array}{ccc}& =& \frac{i}{2}\left(det{W}^{-1}du\wedge d\overline{u}+w_{rl}(w^{lk}d{t}_k-{\theta }_l)\wedge (w^{rm}d{t}_m+{\theta }_r)\right),\end{array}$$

(18)

$$\begin{array}{ccc}\Omega & =& -(w^{1k}d{t}_k-{\theta }_1)\wedge (w^{2k}d{t}_k-{\theta }_2)\wedge du.\end{array}$$

(19)

Assume also that $W$  and $\Theta$  are periodic in $x=Reu$  of period $\kappa$  . Then $(\Omega ,\omega )$  defines a CY structure on $X=\tilde{X}/\kappa \mathbb{Z}$  , where $\kappa \mathbb{Z}$  acts by translations on $x$  . Let $N=\tilde{N}/\kappa \mathbb{Z}$  . Observe now that the map

$$\begin{array}{ccc}f:X& ⟶& B\end{array}$$

(20)

$$\begin{array}{ccc}({\phi }_1,{\phi }_2,u,t_1,t_2)& \mapsto & (t_1,t_2,Imu)\end{array}$$

is a sL 3-torus fibration. We will now compute: the period matrix $P$  with respect to $\omega$  , the metric on the fibres, McLean's metric on $B$  , the volume of the fibres and the semi-flat volumes function $\Phi$  . We refer to the Introduction for the definition of these concepts. Denote by $F_b$  a fibre over $b\in B$  .

For every $b\in B$  fix a basis ${\Sigma }_1,{\Sigma }_2,{\Sigma }_3$  for $H_1(F_b,\mathbb{Z})$  to be represented by the $1$  -cycles $\{x={\phi }_2=0\}$  , $\{x={\phi }_1=0\}$  and $\{{\phi }_1={\phi }_2=0\}$  respectively.

Define

$$\begin{array}{ccc}& {\Xi }_k=({\iota }_{\partial /\partial t_k}\omega {)}_{|F_b},\text{when}k=1,2\text{and}& \end{array}$$

$$\begin{array}{ccc}& {\Xi }_3=({\iota }_{\partial /\partial y}\omega {)}_{|F_b}.& \end{array}$$

These are harmonic $1$  -forms, by McLean's theorem.

The period matrix $P=\left(P_{jk}\right)$  of the fibration with respect to the frame $\Xi =({\Xi }_1,{\Xi }_2,{\Xi }_3)$  and basis ${\Sigma }_1,{\Sigma }_2,{\Sigma }_3$  of $H_1(F_b,\mathbb{Z})$  is given by $$P_{jk}=-{\int }_{{\Sigma }_j}{\Xi }_k.$$  We have

Lemma 1 The period matrix with respect to the frame $\Xi =({\Xi }_1,{\Xi }_2,{\Xi }_3)$  and basis ${\Sigma }_1,{\Sigma }_2,{\Sigma }_3$  of $H_1(F_b,\mathbb{Z})$  is

$$\begin{array}{c}P=\left(\begin{array}{ccc}1& 0& 0\\ & & \\ 0& 1& 0\\ & & \\ {\int }_0^{\kappa }2Im{\alpha }_{1}dx& {\int }_0^{\kappa }2Im{\alpha }_{2}dx& {\int }_0^{\kappa }det{W}^{-1}dx\end{array}\right).\end{array}$$

(21)

The computation is straight forward and we leave it to the reader.

Notice that $P$  depends on $b\in B$  . From the period matrix $P$  we obtain the period $1$  -forms on $B$  : $${\lambda }_j=P_{j1}d{t}_1+P_{j2}d{t}_2+P_{j3}dy$$  which uniquely determine the lattice $\Lambda ={\text{span}}_{\mathbb{Z}}({\lambda }_1,{\lambda }_3,{\lambda }_3)$  inside $T^{*}B$  .

Standard theory of Lagrangian torus fibrations tells us that given a Lagrangian section $\tau :B\to X$  of $f$  , we can naturally identify $X$  with $T^{*}B/\Lambda$  . The identification, in our case, goes as follows. We know that ${\eta }_j$  is the Hamiltonian vector field corresponding to $t_j$  . One can check that $$\zeta =detW\left(\frac{\partial }{\partial x}-2Im{\alpha }_k{\eta }_k\right)$$  is the Hamiltonian vector field corresponding to $y$  . Let ${\Phi }_{{\eta }_j}^s$  and ${\Phi }_{\zeta }^s$  denote the flows of ${\eta }_j$  and $\zeta$  respectively. Then, the map $\Psi :T^{*}B/\Lambda \to X$  given by $$\Psi (b,s_{1}d{t}_1+s_{2}d{t}_2+s_{3}dy)={\Phi }_{{\eta }_1}^{s_1}\circ {\Phi }_{{\eta }_2}^{s_2}\circ {\Phi }_{\zeta }^{s_3}\left(\tau \right(b\left)\right)$$  is well defined with respect to the quotient and provides the identification.

We have the following

Theorem 8 With respect to the frame $({\eta }_1,{\eta }_2,\zeta )$  , the metric of the fibre $F_b$  has the form

$$\begin{array}{c}g=\left(\begin{array}{cc}W& 0\\ & \\ 0& detW\end{array}\right).\end{array}$$

(22)

With respect to the frame $\Xi =({\Xi }_1,{\Xi }_2,{\Xi }_3)$  , McLean's metric $G=(\langle {\Xi }_j,{\Xi }_k{\rangle }_{L^2\left(F_b\right)})$  has the form

$$\begin{array}{c}G={\int }_0^{\kappa }{g}^{-1}dx.\end{array}$$

(23)

Proof. We first compute the metric on $F_b$  with respect to the frame $({\eta }_1,{\eta }_2,\partial /\partial x)$  which we denote $\tilde{g}=\left({\tilde{g}}_{jk}\right)$  . Obviously when $j$  and $k$  are $1$  or $2$  we have $${\tilde{g}}_{jk}=w_{jk}.$$  Using the fact that the forms $(w^{jk}d{t}_k-{\theta }_j)$  with $j=1,2$  and $du$  are of type $(1,0)$  on $X$  and formula ( 18 ) we can compute that

$$\begin{array}{ccc}{\tilde{g}}_{j3}=\omega (\partial /\partial x,J{\eta }_j)=2Im{\alpha }_k{w}_{kj},& & \end{array}$$

when $j=1,2$  , and $${\tilde{g}}_{33}={\left|\frac{\partial }{\partial x}\right|}^2=det{W}^{-1}+4Im{\alpha }_{j}Im{\alpha }_k{w}_{jk}.$$  One can verify that, on passing from the frame $({\eta }_1,{\eta }_2,\partial /\partial x)$  to the frame $({\eta }_1,{\eta }_2,\zeta )$  , one obtains that the metric has the form ( 22 ).

We now compute McLean's metric. We easily see that

$$\begin{array}{ccc}& {\Xi }_k=-(2Im{\alpha }_{k}dx+d{\phi }_k)\text{when}k=1,2\text{and}& \end{array}$$

$$\begin{array}{ccc}& {\Xi }_3=-det{W}^{-1}dx.& \end{array}$$

Notice that, as a co-frame along $F_b$  , $-\Xi$  is dual to $({\eta }_1,{\eta }_2,\zeta )$  , therefore we automatically obtain ( 23 ). $\square$ 

Notice that $g$  can be interpreted (via the identification $\Psi$  ) as a metric on the fibres of $T^{*}B/\Lambda$  , the same is true for $G^{-1}$  : it may be interpreted as a flat metric on the fibres of $T^{*}B/\Lambda$  .

To compute the volume of the fibres we observe that the volume form of $F_b$  is ${\Omega }_{|F_b}$  , by the sL property. Thus applying it to the frame $({\eta }_1,{\eta }_2,\partial /\partial x)$  and integrating over $F_b$  we obtain $$Vol\left(F_b\right)=\kappa .$$  We can now compute the semi-flat volumes function $\Phi$  , mentioned in the Introduction (formula ( 3 )) and defined by Hitchin [7] 

Corollary 1 The semi-flat volumes function $\Phi$  for $T^2$  symmetric $3$  -dimensional sL fibrations is

$$\begin{array}{c}\Phi =\frac{det\left({\int }_0^{\kappa }{W}^{-1}dx\right)}{{\int }_0^{\kappa }det{W}^{-1}dx}.\end{array}$$

(24)

Proof. A basis of harmonic forms with respect to which the period matrix is the identity is $${\Xi }^{\prime }=\Xi \cdot P^{-1}.$$  With respect to this new basis, McLean's metric is $$G^{\prime }=(P^{-1}{)}^{t}G{P}^{-1}.$$  By definition $\Phi =det{G}^{\prime }$  . The reader can check that we obtain ( 24 ). $\square$ 

This formula generalizes a similar one obtained in Section 5 of [9] . By applying Theorem  7 we see that one can find CY metrics on $X$  with $W$  chosen arbitrarily along the line $\{t_1=t_2=0\}$  . Therefore one obtains examples where $\Phi$  is not constant.

References

1. R. L. Bryant. Some examples of special lagrangian tori. Adv. Theor. Math. Phys., 3:83–90, 1999.
2. R. L. Bryant. Calibrated embeddings in the special Lagrangian and coassociative cases. Ann. Global Anal. Geom., 18:405–435, 2000. Special issue in memory of Alfred Gray (1939–1998).
3. E. Goldstein. Calibrated fibrations on noncompact manifolds via torus actions. Duke Mathematical Journal, 110:309–343, 2001.
4. M. Gross. Special Lagrangian fibrations I: topology. In Integrable Systems and Algebraic Geometry (Kobe/Kyoto, 1997), pages 156–193, River Edge, NJ, 1998. World Sci. Publishing.
5. M. Gross. Special Lagrangian fibrations II: geometry. In Surveys in Differential Geometry, pages 341–403, Somerville:MA, 1999. International Press.
6. M. Gross. Examples of special Lagrangian fibrations. In Symplectic geometry and mirror symmetry (Seoul, 2000), pages 81–109. World Sci. Publishing, River Edge, NJ, 2001.
7. N. J. Hitchin. The moduli space of special Lagrangian submanifolds. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 25:503–515, 1998. Dedicated to Ennio De Giorgi.
8. N. J. Hitchin. The moduli space of complex Lagrangian submanifolds. Asian J. Math., 3:77–91, 1999.
9. D. Matessi. Some families of special Lagrangian tori. Mathematische Annalen, 325(2):211–228, 2003.
10. R. C. McLean. Deformations of calibrated submanifolds. Comm. Anal. Geom., 6:705–747, 1998.
11. H. Pedersen and Y. Sun Poon. Hamiltonian constructions of Kähler–einstein metrics and constant scalar curvature. Comm. in Math. Phys., 136:309–326, 1991.
12. A. Strominger, S.–T. Yau, and E. Zaslow. Mirror symmetry is T-duality. Nucl. Phys., B479:243–259, 1996.