Hyperholomorphic sheaves and new examples of hyperkaehler manifolds

This is ?. width 4pt height 4pt depth 0pt

This is Proposition 1.1 of [_Verbitsky:Symplectic_II_]. width 4pt height 4pt depth 0pt

This is [_Verbitsky:Hyperholo_bundles_], Proposition 1.2. width 4pt height 4pt depth 0pt

This is Theorem 4.1 of [_Verbitsky:Symplectic_II_]. width 4pt height 4pt depth 0pt

This is Proposition 2.2 from [_Verbitsky:Symplectic_II_] width 4pt height 4pt depth 0pt

This is Proposition 4.17 of [_Kobayashi_], Chapter I. \hbox{\vrule width 4pt height 4pt depth 0pt}

[_Uhle_Yau_]. width 4pt height 4pt depth 0pt

We use the definition of a hyperholomorphic connection as one with SU(2)-invariant curvature. Then ? follows from the

Lemma .lemma: Let \Theta\in \Lambda^2(M) be a SU(2)-invariant differential 2-form on M. Then \Lambda_L(\Theta)=0 for each induced complex structure L.(By \Lambda_L we understand the Hodge operator \Lambda associated with the Kähler complex structure L.)

This is Lemma 2.1 of [_Verbitsky:Hyperholo_bundles_]. width 4pt height 4pt depth 0pt

This is Theorem 2.5 of [_Verbitsky:Hyperholo_bundles_]. width 4pt height 4pt depth 0pt

[_NHYM_], Proposition 4.5. width 4pt height 4pt depth 0pt

This is [_OSS_], Ch. II, Lemma 1.2.4. width 4pt height 4pt depth 0pt

In [_Bando_Siu_], ? is proved for Kähler M ([_Bando_Siu_], Theorem 3). It is easy to adapt this proof for Hermitian manifolds satisfying conditions of Li–Yau. width 4pt height 4pt depth 0pt

This is [_Bando_Siu_], Theorem 2. width 4pt height 4pt depth 0pt

Consider an exact sequence

Let [F / F^{**}]\in H^4(M) be the fundamental class of the union of all codimension-2 components of support of the sheaf F / F^{**}, taken with appropriate multiplicities. Then, c_2(F^{**}/ F) =- [F / F^{**}]. From the product formula for Chern classes, it follows that

Clearly, if all X_i are trianalytic then the class [F / F^{**}] is SU(2)-invariant. Thus, if the sheaf F^{**} is hyperholomorphic and all X_i are trianalytic, then the second Chern class of F is SU(2)-invariant, and F is hyperholomorphic. Conversely, assume that F is hyperholomorphic. We need to show that all X_i are trianalytic. By definition,

where [X_i] denotes the fundamental class of X_i, and \lambda_i is the multiplicity of F / F^{**} at X_i. By (Proof), (F hyperholomorphic) implies that the class [F / F^{**}] is SU(2)-invariant. Since [F / F^{**}] is SU(2)-invariant, we have

By Wirtinger’s inequality (?),

and the equality is reached only if X_i is trianalytic. By definition, all the numbers \lambda_i are positive. Therefore,

and the equality is reached only if all the subvarieties X_i are trianalytic. This finishes the proof of ?. width 4pt height 4pt depth 0pt

Let F be a torsion-free coherent sheaf over (M, I). Clearly from the definition of induced complex structure of general type, the sheaves F and F^{**} have SU(2)-invariant Chern classes. Now, all SU(2)-invariant 2-forms have degree zero (Proof), and thus, F is semistable. width 4pt height 4pt depth 0pt

Let J be an induced complex structure on M, and U the set of all points of (M,I) where F is non-singular. Clearly, (F, \nabla) is a bundle with admissible connection on (U,J). Therefore, the holomorphic structure on F{\left|_{{\phantom{|}\!\!}_{(U,J)}}\right.} can be extended to (M,J). Thus, the singular set of F is holomorphic with respect to J. This proves ?. width 4pt height 4pt depth 0pt

By ? and ?, F is polystable. We need only to show that the Chern classes c_1(F) and c_2(F) are SU(2)-invariant. Let U\subset M be the maximal open subset of M such that F{\left|_{{\phantom{|}\!\!}_{U}}\right.} is locally trivial. By ?, the metric h and the connection \nabla can be extended to U. Let \operatorname{Tw} U\subset \operatorname{Tw} M be the corresponding twistor space, and \sigma:\; \operatorname{Tw} U {\:\longrightarrow\:} U the standard map. Consider the bundle \sigma^* F{\left|_{{\phantom{|}\!\!}_{U}}\right.}, equipped with a connection \sigma^* \nabla. It is well known (See for instance the section “Direct and inverse twistor transform” in [_NHYM_].) that \sigma^* F{\left|_{{\phantom{|}\!\!}_{U}}\right.} is a bundle with an admissible Yang-Mills metric (we use Yang-Mills in the sense of Li-Yau, see ?). By ?, \sigma^* F{\left|_{{\phantom{|}\!\!}_{U}}\right.} can be extended to a reflexive sheaf {\cal F} on \operatorname{Tw} M. Clearly, this extension coincides with the push-forward of \sigma^* F{\left|_{{\phantom{|}\!\!}_{U}}\right.}. The singular set \widetilde S of {\cal F} is a pull-back of the singular set S of F. Thus, S is trianalytic. By desingularization theorem (?), S can be desingularized to a hyperkähler manifold in such a way that its twistors form a desingularization of {\cal S}. From the exact description of the singularities of {\cal S}, provided by the desingularization theorem, we obtain that the standard projection \pi:\; {\cal S} {\:\longrightarrow\:} {{\mathbb C}} P^1 is flat. By the following lemma, the restriction of {\cal F} to the fiber (M,I) = \pi^{-1}(\{I\}) of \pi coincides with F.

Lemma .lemma: Let \pi:\; X {\:\longrightarrow\:} Y be a map of complex varieties, and S\hookrightarrow X a subvariety of X of codimension at least 2, which is flat over Y. Denote by U\stackrel j \hookrightarrow X the complement U = (X\backslash S). Let F be a vector bundle over U, and j_* F its push-forward. Then the restriction of j_* F to the fibers of \pi is reflexive.

Let Z= \pi^{-1}(\{y\}) be a fiber of \pi. Since S is flat over Y and of codimension at least 2, we have j_*({\cal O}_{Z \cap U}) = {\cal O}_{Z}. Clearly, for an open embedding \gamma:\; T_1 {\:\longrightarrow\:} T_2 and coherent sheaves A, B on T_1, we have \gamma_*(A\otimes B) = \gamma_* A \otimes \gamma_* B. Thus, for all coherent sheaves A on U, we have

This implies that j_*(F {\left|_{{\phantom{|}\!\!}_{Z}}\right.}) = j_* F{\left|_{{\phantom{|}\!\!}_{Z}}\right.}. It is well known ([_OSS_], Ch. II, 1.1.12; see also ?) that a push-forward of a reflexive sheaf under an open embedding \gamma is reflexive, provided that the complement of the image of \gamma has codimension at least 2. Therefore, j_* F{\left|_{{\phantom{|}\!\!}_{Z}}\right.} is a reflexive sheaf over Z. This proves ?. width 4pt height 4pt depth 0pt

The following general argument is used.

Sublemma .sublemma: Let M be a Kähler manifold, B a Hermitian vector bundle over M, and \Xi a square-integrable 2-form on M with coefficients in \frak{su}(B). Then:

• (i) For

  \Lambda_I \Xi =0, \ \ \ \Xi\in \Lambda^{1,1}_I(M, \operatorname{End} B)

  we have

  \deg_I \Xi^2 = C \int_M|\Xi|^2 \operatorname{Vol} M ,

  where C= (4\pi^2 n (n-1))^{-1} M.

• (ii) For

  \Xi\in \Lambda^{2,0}_I(M, \operatorname{End} B)\oplus \Lambda^{0,2}_I(M, \operatorname{End} B)

  we have

  \deg_I \Xi^2 = -C \int_M |\Xi|^2\operatorname{Vol} M,

  where C is the same constant as appeared in (i).

The proof is based on a linear-algebraic computation (so-called Lübcke-type argument). The same computation is used to prove Hodge-Riemann bilinear relations. width 4pt height 4pt depth 0pt

Clearly, \Lambda_L \Theta^{1,1}_L= \Lambda_L\Theta. Consider the natural Hermitian structure on the space of 2-forms. Since \Theta is of type (1,1) with respect to I, \Theta is fiberwise orthogonal to the holomorphic symplectic form \Omega= \omega_J +\1 \Omega_K\in \Lambda^{2,0}_I(M). By the same reason, \Theta is orthogonal to \overline\Omega. Therefore, \Theta is orthogonal to \omega_J and \omega_K. Since \Lambda_I \Theta =0, \Theta is also orthogonal to \omega_I. The map \Lambda_L is a projection to the form \omega_L which is a linear combination of \omega_I, \omega_J and \omega_K. Since \Theta is fiberwise orthogonal to \omega_L, we have \Lambda_L \Theta =0. width 4pt height 4pt depth 0pt

Let \varphi:\; F_1 {\:\longrightarrow\:} F_2 be a morphism of hyperholomorphi sheaves. In ?, ? , we introduced a slope of a coherent sheaf. Clearly, sl(F_1)\leqslant sl(\operatorname{im} \varphi)\leqslant sl(F_2). All hyperholomorphic sheaves have slope 0 by ?. Thus, sl(\operatorname{im} \varphi)=0 and the subsheaf \operatorname{im}\varphi\subset F_2 is destabilizing. Since F_2 is polystable, this sheaf is decomposed:

A similar argument proves that F_1 = \ker\varphi \oplus \operatorname{coim} \varphi, with all summands hyperholomorphic. This proves that {\cal F}_{st}(M,I) is abelian. The Tannakian properties are clear. width 4pt height 4pt depth 0pt

Let F\in {\cal F}_{st}(M,I_1) be a reflexive polystable hyperholomorphic sheaf and \nabla the canonical admissible Yang-Mills connection. Consider the sheaf {\cal F} on the twistor space \operatorname{Tw}(M) constructed as in the proof of ?. Restricting {\cal F} to \pi^{-1}(I_2)\subset \operatorname{Tw}(M), we obtain a coherent sheaf F' on (M,I_2). As we have shown in the proof of ?, the sheaf (F')^{**} is polystable hyperholomorphic. Let \Phi_{I_1, I_2}(F):= (F')^{**}. It is easy to check that thus constructed map of objects gives a functor

and moreover, \Phi_{I_1, I_2}\circ \Phi_{I_2, I_1} = Id. This shows that \Phi_{I_1, I_2} is an equivalence. ? is proven. width 4pt height 4pt depth 0pt

This is [_Verb:alge_], Theorem 2.2. width 4pt height 4pt depth 0pt

This statement is a form of [_Verb:alge_], Lemma 2.3. width 4pt height 4pt depth 0pt

The space H^*_o(M) is generated from \pmb 1\in H^0(M) by operators L_R, \Lambda_R. To prove that H^*_o(M) is closed under multiplication, we have to show that H^*_o(M) is generated (as a linear space) by expressions of type L_{r_1} \circ L_{R_2} \circ ... \circ \pmb 1. By (?), the commutators of L_R, \Lambda_R map such expressions to linear combinations of such expressions. On the other hand, the operators \Lambda_R map \pmb 1 to zero. Thus, the operators \Lambda_R map expressions of type L_{r_1} \circ L_{R_2} \circ ... \circ \pmb 1 to linear combinations of such expressions. This proves that H^*_o(M) is closed under multiplication. The second statement of ? is clear (see, e. g. [_Verbitsky:Symplectic_II_], proof of Proposition 4.5). It remains to show that H^*_o(M)\subset H^*(M) is an SU(2)-invariant subspace and that p:\; H^*(M) {\:\longrightarrow\:} H^*_o(M) is compatible with the SU(2)-action. From (?), we obtain that the Lie group G_A associated with \frak a_{{\cal H}}\cong \frak{so}(1,4) contains SU(2) acting in a standard way on H^*(M). Since the map p:\; H^*(M) {\:\longrightarrow\:} H^*_o(M) commutes with G_A-action, p also commutes with SU(2)-action. We proved ?. width 4pt height 4pt depth 0pt

[_Verbitsky:cohomo_] (see also [_coho_announce_]). width 4pt height 4pt depth 0pt

[_Verbitsky:cohomo_] (see also [_coho_announce_]). width 4pt height 4pt depth 0pt

By definition, the classes \omega_1, \omega_2 satisfy \omega_i\in \ker p. By ?, we have \omega_1 \omega_2\in \ker p. By ?, \deg_I\omega_1 \omega_2 =0. width 4pt height 4pt depth 0pt

? (i) is clear, because the action of {\frak g} on H^*(M) is compatible with the rational structure on H^*(M). To prove ? (ii), we notice that the space H^*_r(M) is generated from H^0(M) by the action of {\frak g}, and H^*_o(M) is generated from H^0(M) by the action of {\frak a}_{{\cal H}}. Since {\frak a}_{{\cal H}} is by definition a subalgebra in {\frak g}, we have H^*_o(M) \subset H^*_r(M). The isotypic projection r:\; H^*(M) {\:\longrightarrow\:} H^*_r(M) is by definition compatible with the {\frak g}-action. Since {\frak a}_{{\cal H}}\subset {\frak g}, the map r is also compatible with the {\frak a}_{{\cal H}}-action. Therefore, \ker r\subset \ker p. ? (iii) is implied by ? (ii) and ?. width 4pt height 4pt depth 0pt

Clearly, if an integer class \alpha is SU(2)-invariant for a generic hyperkähler structure, then \alpha is G-invariant, where G is the group defined in ?. On the other hand, H^4_r(M)\cong S^2(H^2(M)), as follows from ?. Clearly, the vector \theta\in H^4_r(M)\cong S^2(H^2(M)) is \frak{so}(H^2(M))-invariant. Moreover, the space of \frak{so}(H^2(M))-invariant vectors in S^2(H^2(M)) is one-dimensional. Finally, from an explicit computation of G it follows that G acts on H^4(M) as SO(H^2(M)), and thus, the Lie algebra invariants coincide with invariants of G. We found that the space of G-invariants in H^4_r(M) is one-dimensional and generated by \theta. This proves ?. width 4pt height 4pt depth 0pt

? (i) is a direct consequence of Wirtinger’s inequality (?). It remains to prove ? (ii).

We assume, temporarily, that F is reflexive. By ?, we have

and the equality happens only if F is hyperholomorphic. Since c_1(F) is SU(2)-invariant, we have \deg_I(c_1(F)^2) = \deg_J (c_1(F)^2) =0 (?). Thus, (?, ?) implies that

and the inequality is strict unless F is hyperholomorphic, in which case, the class c_2(F) is SU(2)-invariant by definition. We have proven ? (ii) for the case of reflexive F.

For F not necessary reflexive sheaf, we have shown in the proof of ? that

where n_i are positive integers, and [X_i] are the fundamental classes of irreducible components of support of the sheaf F^{**}/F. Therefore, the class c_2(F) is a sum of classes of CA-type. Clearly, a sum of cohomology classes of CA-type is again a class of CA-type. This proves ?. width 4pt height 4pt depth 0pt

The proof of ? takes the rest of this section.

Denote by {\cal R} the set {\cal R}\cong {{\mathbb C}} P^1 of all induced complex structures on M. Consider the set {\cal R}/\{\pm 1\} of induced complex structures up to a sign (?). Let \alpha \in H^2(M) be a cohomology class which is not SU(2)-invariant. According to ?, there exists a unique element c(\alpha)\in {\cal R}/\{\pm 1\} such that \alpha \in H^{1,1}_{c(\alpha)}(M). This defines a map

where H^2_{inv}(M)\subset H^2(M) is the set of all SU(2)-invariant cohomology classes. For induced complex structures I and -I, and \eta\in H^{2p}(M), the degree maps satisfy

Thus, the number |\deg_I(\eta)| is independent from the sign of I.

Let \eta\in H^*(M, {{\mathbb Z}}) be a cohomology class. The largest divisor of \eta is the biggest positive integer number k such that the cohomology class \frak \eta k is also integer.

Let \alpha\in H^2(M, {{\mathbb Z}}) be an integer cohomology class, which is not SU(2)-invariant, k its largest divisor and \widetilde \alpha:= \frak \alpha k the corresponding integer class. Denote by \widetilde\deg(\alpha) the number

The induced complex structure c(\alpha) is defined up to a sign, but from (?) it is clear that \widetilde\deg(\alpha) is independent from the choice of a sign.

Lemma .lemma: Let M be a compact hyperkähler manifold, and I be an algebraic induced complex structure, such that the group H^{1,1}_I(M)\cap H^2(M, {{\mathbb Z}}) has rank one, and the group H^{2,2}_I(M)\cap H^4(M, {{\mathbb Z}})/(\ker r) has rank 2. Denote by \alpha the generator of H^{1,1}_I(M)\cap H^2(M, {{\mathbb Z}}). Since the class \alpha is proportional to a Kähler form, \alpha is not SU(2)-invariant (?, (i)). Let d:= \widetilde \deg\alpha. Then, there exists a positive real constant A depending on volume of M, its topology and its dimension, such that I is d\cdot A-restricted.

This lemma is a trivial calculation based on results of [_Verbitsky:cohomo_] (see also [_coho_announce_] and Subsection ?).

Since H^{1,1}_I(M)\cap H^2(M, {{\mathbb Z}}) has rank one, for all \eta \in H^{1,1}_I(M)\cap H^2(M, {{\mathbb Z}}), \eta\neq 0, we have | \deg_I \eta| \geqslant d. This proves the first condition of ?.

Let \theta be the SU(2)-invariant integer cycle \theta\in H^4(M) defined in ?. By ?, ?, \theta\in H^{2,2}_I(M). Consider \alpha^2\in H^{2,2}_I(M), where \alpha is the generator of H^{1,1}_I(M)\cap H^2(M, {{\mathbb Z}}).

Sublemma .sublemma: Let J be an induced complex structure, J\circ I=-J\circ I, and \deg_I, \deg_J the degree maps associated with I, J. Then

Since \alpha is a Kähler class with respect to I, we have \deg_I \alpha^2 >0. Since the cohomology class \theta is SU(2)-invariant, and SU(2) acts transitively on the set of induced complex structures, we have \deg_I\theta=\deg_J\theta. It remains to show that \deg_J\alpha^2 =0 and \deg_J\theta >0. The manifold M is by our assumptions simple; thus, \dim H^{2,0}(M) =1 ([_Besse:Einst_Manifo_]). Therefore, in the natural SU(2)-invariant decomposition

we have \dim H^2_+(M) = 3. In particular, the intersection H^2_+(M)\cap H^{1,1}_I(M) is 1-dimensional. Consider the decomposition of \alpha, associated with (Proof): \alpha = \alpha_+ + \alpha_{inv}. Since \alpha is of type (1,1) with respect to I, the class \alpha_+ is proportional to the Kähler class \omega_I, with positive coefficient. A similar argument leads to the following decomposition for \theta:

where K= I\circ J is an induced complex structure, and the classes x_i belong to H_{inv}^2(M). From ?, we obtain that the classes x_i^2 satisfy \deg_I(x_i^2)=0 (here we use \dim_{{\mathbb H}}(M)>1). Thus,

Similarly one checks that

This proves ?. width 4pt height 4pt depth 0pt

The map \widetilde \deg can be expressed in the following wey. We call an integer cohomology class \alpha\in H^2(M, {{\mathbb Z}}) indivisible if its largest divisor is 1, that is, there are no integer classes \alpha', and numbers k\in {{\mathbb Z}}, k>1, such that \alpha = k \alpha'.

Sublemma .sublemma: Let \alpha \in H^2(M) be an non-SU(2)-invariant cohomology class and \alpha= \alpha_{inv} + \alpha_+ be a decomposition associated with (Proof). Assume that \alpha is indivisible. Then

where (\cdot,\cdot)_{{\cal B}} is the Bogomolov-Beauville pairing on H^2(M) ([_coho_announce_]; see also ?), and C a constant depending on \dim M, \operatorname{Vol} M.

By ?,

(clearly, p(\alpha) = \alpha_+). By definition of (\cdot,\cdot)_{{\cal B}}, we have

On the other hand, \alpha_+ is collinear with \omega_{c(\alpha)} by definition of the map c. Now (?) follows trivially from routine properties of bilinear forms. width 4pt height 4pt depth 0pt

For every plane L\in Gr_0, consider the set s(L) of all isotropic lines l\in L, that is, lines satisfying (l, l)_{{\cal B}}=0. Since (\cdot,\cdot)_{{\cal B}}{\left|_{{\phantom{|}\!\!}_{L}}\right.} is non-degenerate, the set s(L) is a rational curve in {{\mathbb P}}L. Clearly, this curve has degree 2. Therefore, s(L) belongs to X(L). The map L{\:\longrightarrow\:} s(L) is inverse to \kappa. width 4pt height 4pt depth 0pt

The first claim is an immediate consequence of Calabi-Yau Theorem (?). Now, ? follows from the Bogomolov’s theorem (?) and dimension count. width 4pt height 4pt depth 0pt

Consider a natural involution i of Comp which is compatible with the involution \iota:\; X {\:\longrightarrow\:} X inder the period map P_c:\; Comp {\:\longrightarrow\:} X. This involution maps the complex structure I to -I.

Let L_U\subset L(Comp) be the set of all lines which do not intersect V. Since V is a union of subvarieties of codimension at least 2, a general rational line l \in L(Comp) does not intersect V. Therefore, L_U is dense in L(Comp). Thus, the set of real points of L_U is dense L(X)_\iota. Using the identification (Proof), we obtain the statement of ?. width 4pt height 4pt depth 0pt

Let F_1, ..., F_n be the Jordan-Hölder series for the sheaf F'. Since {\cal F} is hyperholomorphic, we have \text{slope}({\cal F})=0 (?). Therefore, \operatorname{slope}(F_i)=0, and \deg_I(c_1(F_i))=0. By ? (i), then, the class c_1(F_i) is SU(2) invariant for all i. To prove that F' is hyperholomorphic it remains to show that the classes c_2(F_i), c_2(F_i^{**}) are SU(2)-invariant for all i.

Consider an exact sequence

Let [F_i / F_i^{**}]\in H^4(M) be the fundamental class of the union of all components of Sup(F_i / F_i^{**}) of complex codimension 2, taken with appropriate multiplicities. Clearly, c_2(F_i)= c_2(F^{**}_i) + [F_i / F_i^{**}]. Since [F_i / F_i^{**}] is an effective cycle, \deg_I([F_i / F_i^{**}])\geqslant0. By the Bogomolov-Miyaoka-Yau inequality (see ?), we have \deg_I (c_2(F^{**}_i)\geqslant 0. Therefore,

Using the product formula for Chern classes, we obtain

By ?, \deg_I(\sum_{i, j} c_2(F_i)\wedge c_2(F_j)) =0. Since the numbers \deg_I c_2(F_i) are non-negative, we have \deg_I c_2(F_i)\leqslant \deg_I c_2(F) =C. By ?, the classes c_2(F_i), c_2(F_i^{**}) are of CA-type. By ? (ii), then, the inequality \deg_I c_2(F_i)\leqslant C implies that the class c_2(F_i) is SU(2)-invariant. By (?), \deg_I c_2(F^{**}_i)\leqslant\deg_I c_2(F_i), so the class c_2(F_i^{**}) is also SU(2)-invariant. ? is proven. width 4pt height 4pt depth 0pt

[_HKLR_], [_Verbitsky:hypercomple_]. width 4pt height 4pt depth 0pt

The map \Psi_{I, J} was defined in [_Verbitsky:DesinguII_], [_Verbitsky:hypercomple_] using the identifications between the real analytic varieties underlying (M,I) and (M, J). We proved that \Psi_{I, J} defined this way acts on T_{x_0}B_0 as a multiplication by the scalar \lambda\in {{\mathbb C}}, 0<|\lambda| <1. It remains to show that the map \Psi_{I, J} defined in [_Verbitsky:DesinguII_], [_Verbitsky:hypercomple_] coincides with \Psi_{I, J} defined above.

Consider the natural identification

which is defined in a neighbourhood B_{{\mathbb C}} of (x_0, x_0). There is a natural projection a_I:\; B_{{\mathbb C}} {\:\longrightarrow\:} (M,I). Consider the embedding b_I:\; (B, I) {\:\longrightarrow\:} B_{{\mathbb C}}, x{\:\longrightarrow\:} (x, x_0), defined in a neighbourhood of x_0\in (B,I). In a similar way we define a_J, b_J. In [_Verbitsky:DesinguII_], [_Verbitsky:hypercomple_] we defined \Psi_{I, J} as a composition b_I\circ a_J \circ b_J \circ a_I. Earlier in this Subsection, we described a local identification of (B, I)\times (B,-I) and \operatorname{Lin}(B). Clearly, under this identification, the maps a_I, b_I correspond to p_I, e_I. Therefore, the definition of \Psi_{I,J} given in this paper is equivalent to the definition given in [_Verbitsky:DesinguII_], [_Verbitsky:hypercomple_]. width 4pt height 4pt depth 0pt

The complex codimension of the singular set S in (M,I) is at least 3, because F is reflexive ([_OSS_], Ch. II, 1.1.10). Since S is trianalytic (?), this codimension is even. Thus, \operatorname{codim}_{{\mathbb C}} (S, (M,I)) \geqslant 4. Therefore,

Consider the set L_S of all twistor lines l\in \operatorname{Lin}(W) passing through \operatorname{Tw}(S). For generic points x,y\in \operatorname{Tw}(W), there exists a unique line l\in \operatorname{Lin}(W) passing through x, y. Therefore,

By definition, L_0:= \operatorname{Lin}(W)\backslash L_S. Since F_1, F_2 are direct images of bundles \underline F_1, \underline F_2 over a subvariety L_S of codimension 3, these sheaves are coherent and reflexive ([_OSS_], Ch. II, 1.1.12; see also ?). To show that they are naturally isomorphic it remains to construct an isomorphism between \underline F_1 and \underline F_2.

Let {\cal F} be a coherent sheaf on \operatorname{Tw}(W) obtained from F{\left|_{{\phantom{|}\!\!}_{W}}\right.} as in the proof of ?. The singular set of {\cal F} is \operatorname{Tw}(S)\subset \operatorname{Tw}(W). Therefore, the restriction {\cal F}{\left|_{{\phantom{|}\!\!}_{\operatorname{Tw}(W)\backslash \operatorname{Tw}(S)}}\right.} is a holomorphic vector bundle. For all horizontal twistor lines l_x\subset \operatorname{Tw}(W)\backslash \operatorname{Tw}(S), the restriction {\cal F}{\left|_{{\phantom{|}\!\!}_{l_x}}\right.} is clearly a trivial vector bundle over l_x\cong {{\mathbb C}} P^1. A small deformation of a trivial vector bundle is again trivial. Shrinking W if necessary, we may assume that for all lines l\in L_0, the restriction of {\cal F} to l\cong {{\mathbb C}} P^1 is a trivial vector bundle.

The isomorphism \underline \Psi_{1,2}:\;\underline F_1{\:\longrightarrow\:}\underline F_2 is constructed as follows. Let l\in L_0 be a twistor line. The restriction {\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.} is trivial. Consider l as a map l:\; {{\mathbb C}} P^1 {\:\longrightarrow\:} \operatorname{Tw}(M). We identify {{\mathbb C}} P^1 with the set of induced complex structures on M. By definition, the fiber of F_1 in l is naturally identified with the space {\cal F}{\left|_{{\phantom{|}\!\!}_{l(I)}}\right.}, and the fiber of F_2 in l is identifies with {\cal F}{\left|_{{\phantom{|}\!\!}_{l(J)}}\right.}. Since {\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.} is trivial, the fibers of the bundle {\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.} are naturally identified. This provides a vector bundle isomorphism \underline \Psi_{1,2}:\;\underline F_1{\:\longrightarrow\:}\underline F_2 mapping \underline F_1 {\left|_{{\phantom{|}\!\!}_{l}}\right.} = {\cal F}{\left|_{{\phantom{|}\!\!}_{l(I)}}\right.} to \underline F_2 {\left|_{{\phantom{|}\!\!}_{l}}\right.} = {\cal F}{\left|_{{\phantom{|}\!\!}_{l(J)}}\right.}. It remains to show that this isomorphism is compatible with the holomorphic structure.

Since the bundle {\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.} is trivial, we have an identification

where \Gamma({\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.}) is the space of global sections of {\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.}. Thus, F_i {\left|_{{\phantom{|}\!\!}_{l}}\right.} = \Gamma({\cal F}{\left|_{{\phantom{|}\!\!}_{l}}\right.}), and this identification is clearly holomorphic. This proves ?. width 4pt height 4pt depth 0pt

By (?), it suffices to prove ? for i=0. In other words, we have to show that \Psi^F_{I,J} acts as identity on F_x /\frak m F_x. We reduced ? to the following claim.

Claim .claim: In the above assumptions, the automorphism \Psi^F_{I,J} acts as identity on F_x /\frak m F_x.

In the course of defining the map \Psi^F_{I,J}, we identified the space \operatorname{Lin}(M) with a complexification of (M,I), and defined the maps

(these maps are smooth, in a neighbourhood of \operatorname{Hor}\subset \operatorname{Lin}(M), by ?), and

(these maps are locally closed embeddings). Consider (M,J) as a subvariety of \operatorname{Tw}(M), (M, J) = \pi^{-1}(J). Let {\cal F} be the lift of F to \operatorname{Tw}(M) (see the proof of ? for details). Denote the completion of {\cal O}_x(M, J) by {\cal O}_{x, J}. Let F_J denote the {\cal O}_{x, J}-module \left({\cal F}{\left|_{{\phantom{|}\!\!}_{(M,J)}}\right.} \right)\otimes_{{\cal O}_{(M,J)}} \hat {\cal O}_{x, J}. Consider the horizontal twistor line l_x\in \operatorname{Lin}(M). Let \operatorname{Lin}_x(M) be the spectre of the completion {\cal O}_{x,\operatorname{Lin}} of the local ring of holomorphic functions on \operatorname{Lin}(M) in l_x. The maps p_I, p_J, e_I, e_J can be considered as maps of corresponding formal manifolds:

As in Subsection ?, we consider the {\cal O}_{x,\operatorname{Lin}}-modules F_1:= p_I^* F_x and F_2:= p_J^* F_J. By ?, there exists a natural isomorphism \Psi_{1,2}:\; F_1{\:\longrightarrow\:} F_2.

Let \frak m_{l_x} be the maximal ideal of {\cal O}_{x,\operatorname{Lin}}. Since the morphism p_I is smooth, the space F_1/ \frak m_{l_x} F_1 is naturally isomorphic to F_x/\frak m F_x. Similarly, the space F_2/ \frak m_{l_x} F_2 is isomorphic to F_J / \frak m_J F_J, where \frak m_J is the maximal ideal of {\cal O}_{x, J}. We have a chain of isomorphisms

By definition, for any f\in F_x/\frak m F_x, the value of \Psi^F_{I,J}(f) is given by the composide map of (?) applied to f. The composition

is identity, because the spaces F_2/ \frak m_{l_x} F_2 and F_J / \frak m_J F_J are canonically identified, and this identification can be performed via e_J^* or p_J^*. Thus, the map (?) is a composition

This map is clearly equivalent to a composition

which is identity according to the same reasoning which proved that (?) is identity. We proved Proof and ?. width 4pt height 4pt depth 0pt

Let C:= \pi^{-1}(x) be the singular locus of \pi. The sheaf F is locally trivial outside of x. Let d be the rank of F{\left|_{{\phantom{|}\!\!}_{B\backslash x}}\right.}. To prove that \widetilde F is locally trivial, we need to show that for all points y\in \widetilde B, the fiber \pi^* F{\left|_{{\phantom{|}\!\!}_{y}}\right.} is d-dimensional. Therefore, to prove ? it suffices to show that \pi^* F{\left|_{{\phantom{|}\!\!}_{C}}\right.} is a vector bundle of dimension d.

The variety C is naturally identified with the projectivization {{\mathbb P}} T_x B of the tangent space T_xB. The total space of T_x B is equipped with a natural action of {{\mathbb C}}^*, acting by dilatations. Clearly, coherent sheaves on {{\mathbb P}} T_x B are in one-to-one correspondence with {{\mathbb C}}^*-equivariant coherent sheaves on T_x B. Consider a local isomorphism \varphi:\; T_x B {\:\longrightarrow\:} B which is compatible with {{\mathbb C}}^*-action, maps 0\in T_x B to x and acts as identity on the tangent space T_0 (T_x B)=T_x B. The sheaf \varphi^* F is {{\mathbb C}}^*-equivariant. Clearly, the corresponding sheaf on {{\mathbb P}} T_x B is canonically isomorphic with \pi^*F {\left|_{{\phantom{|}\!\!}_{C}}\right.}. Let l\in T_x B be a line passing through 0, and l\backslash 0 its complement to 0. Denote the corresponding point of {{\mathbb P}} T_x B by y. The restriction \varphi^* F{\left|_{{\phantom{|}\!\!}_{l\backslash 0}}\right.} is a {{\mathbb C}}^*-equivariant vector bundle. The {{\mathbb C}}^*-equivariant structure identifies all the fibers of the bundle \varphi^* F{\left|_{{\phantom{|}\!\!}_{l\backslash 0}}\right.}. Let F_l be one of these fibers. Clearly, the fiber of \pi^*F {\left|_{{\phantom{|}\!\!}_{C}}\right.} in y is canonically isomorphic to F_l. Therefore, the fiber of \pi^*F {\left|_{{\phantom{|}\!\!}_{C}}\right.} in y is d-dimensional. We proved that \pi^*F is a bundle. width 4pt height 4pt depth 0pt

Follows from ?, ?. width 4pt height 4pt depth 0pt

[_NHYM_], Theorem 5.12. width 4pt height 4pt depth 0pt

Given a pair {\cal F}, {\frak h}:\; {\cal F} {\:\longrightarrow\:} (\iota^* \overline{{\cal F}})^*, we need to construct a non-degenerate semilinear pairing h on {\cal F}{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.}, compatible with a connection. Since {\cal F} is compatible with twistor transform, it is a pullback of a bundle (F, \nabla) on M. This identifies the real analytic bundles {\cal F}{\left|_{{\phantom{|}\!\!}_{(M,I')}}\right.}, for all induced complex structures I'. Taking I' = \pm I, we obtain an identification of the C^{\infty}-bundles {\cal F}{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.}, {\cal F}{\left|_{{\phantom{|}\!\!}_{(M,-I)}}\right.}. Thus, \frak h can be considered as an isomorphism of F= {\cal F}{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.} and (\overline {{\cal F}})^*{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.}. This allows one to consider \frak h as a semilinear form h on F. We need only to show that h is compatible with the connection \nabla. Since \nabla_{{\cal F}} is an invariant of holomorphic structure, the map {\frak h}:\; {\cal F} {\:\longrightarrow\:} (\iota^* \overline{{\cal F}})^* is compatible with the connection \nabla_{{\cal F}}. Thus, the obtained above form h is compatible with the connection \nabla_{{\cal F}}{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.}=\nabla. This proves ?. width 4pt height 4pt depth 0pt

It is easy to modify the proof of the direct and inverse twistor transform theorem from [_NHYM_] to work in quaternionic-Kähler situation. width 4pt height 4pt depth 0pt

[_Nitta:Y-M_]. width 4pt height 4pt depth 0pt

Clearly, the manifold {\cal U}(Q) is equipped with a {{\mathbb H}}^*-action, which is related with the almost hypercomplex structure as prescribed by ? (ii)-(iii). Pick an induced complex structure I\in {{\mathbb H}}. This gives an algebra embedding {{\mathbb C}} {\:\longrightarrow\:} {{\mathbb H}}. Consider the corresponding {{\mathbb C}}^*-action \rho_I on an almost complex manifold ({\cal U}(Q), I). This {{\mathbb C}}^*-action is compatible with the almost complex structure. The quotient {\cal U}(Q)/\rho(I) is an almost complex manifold, which is naturally isomorphic to the twistor space \operatorname{Tw}(Q). Let L^* be a complex vector bundle of all (1,0)-vectors v\in T (\operatorname{Tw}(Q)) tangent to the fibers of the standard projection \sigma:\; \operatorname{Tw}(Q){\:\longrightarrow\:} Q, and L be the dual vector bundle. Denote by Tot_{\neq 0}(L) the complement \operatorname{Tot}(L)\backslash N, where N=\operatorname{Tw}(Q)\subset \operatorname{Tot}(L) is the zero section of L. Using the natural connection in L, we obtain an almost complex structure on \operatorname{Tot}(L).

Consider the natural projection \varphi:\; Tot_{\neq 0}(L){\:\longrightarrow\:} Q. The fibers \varphi^{-1}(q) of \varphi are identified with the space of non-zero vectors in the total space of the cotangent bundle T^* \sigma^{-1}(q)\cong T^*({{\mathbb C}} P^1). This space is naturally isomorphic to

This gives a canonical isomorphism of almost complex manifolds

Therefore, to prove that ({\cal U}(Q), I) is a complex manifold, it suffices to show that the natural almost complex structure on Tot_{\neq 0}(L)\subset \operatorname{Tot}(L) is integrable. Consider the natural connection \nabla_L on L. To prove that \operatorname{Tot}(L) is a complex manifold, it suffices to show that \nabla_L is a holomorphic connection. The bunlde L is known under the name of holomorphic contact bundle, and it is known to be holomorphic ([_Salamon_], [_Besse:Einst_Manifo_]). width 4pt height 4pt depth 0pt

[_Swann_]. width 4pt height 4pt depth 0pt

[_Swann_]. width 4pt height 4pt depth 0pt

Let I be an induced complex structure on M. As we have shown in the proof of ?, the complex manifold (M, I) is identified with an open subset of the total space \operatorname{Tot}(L) of a holomorphic line bundle L over \operatorname{Tw}(Q). The map \varphi is represented as a composition of the projections h:\; \operatorname{Tot}(L){\:\longrightarrow\:} \operatorname{Tw}(Q) and \sigma_Q:\; \operatorname{Tw}(Q) {\:\longrightarrow\:} Q. Since the map h is smooth and holomorphic, the form \varphi^* \omega is of Hodge type (1,1) if and only if \sigma_Q^*\omega is of type (1,1). By ? (i), this happens if and only if \omega is a B_2-form. This proves an equivalence (i) \Leftrightarrow (ii). Since the choice of I is arbitrary, the pullback \varphi^* \omega of a B_2-form is of Hodge type (1,1) with respect to all induced complex structures. By ?, ?, this proves the implication (i) \Rightarrow (iii). The implication (iii) \Rightarrow (ii) is clear. width 4pt height 4pt depth 0pt

Follows from ?, ? applied to \omega = \nabla^2. width 4pt height 4pt depth 0pt

It is obvious from construction that \rho_{{\mathbb C}} is compatible with the complex structure on \operatorname{Tw}(M). width 4pt height 4pt depth 0pt

First of all, let us recall the construction of the natural connection \nabla_{{\cal F}}. Let {\cal F} be an arbitrary bundle compatible with the twistor transform. We construct \nabla_{{\cal F}} in terms of the isomorphism \Psi_{1,2} defined in ?.

Consider an induced complex structure I. Let F_I be the restriction of {\cal F} to (M, I)= \pi^{-1}(I)\subset \operatorname{Tw}(M). Consider the evaluation map

(Subsection ?). In a similar way we define the holomorphic vector bundle F_{-I} on (M, -I) and the map p_{-I}:\; \operatorname{Lin}(M) {\:\longrightarrow\:} (M, -I). Denote by F_1, F_{-1} the sheaves p_I^*(F_I), p_{-I}^*(F_{-I}). In ?, we constructed an isomorphism \Psi_{1,-1}:\; F_1 {\:\longrightarrow\:} F_{-1}.

Let us identify \operatorname{Lin}(M) with (M, I)\times (M, I) (this identification is naturally defined in a neighbourhood of \operatorname{Hor}\subset \operatorname{Lin}(M) – see ?). Then the maps p_I, p_{-I} became projections to the relevant components. Let

be the sheaf maps obtained as pullbacks of de Rham differentials (the tensor product is taken in the category of coherent sheaves over \operatorname{Lin}(M)). Twisting \6 by an isomorphism \Psi_{1,-1}:\; F_1 {\:\longrightarrow\:} F_{-1}, we obtain a map

Adding \overline \6 and \6^\Psi, we obtain

Clearly, \nabla satisfies the Leibniz rule. Moreover, the sheaf p_I^* \Omega^1(M, I) \oplus p_I^* \Omega^1(M, -I) is naturally isomorphic to the sheaf of differentials over

Therefore, \nabla can be considered as a connection in F_1, or as a real analytic connection in a real analytic complex vector bundle underlying F_I. From the definition of \nabla_{{\cal F}} ([_NHYM_]), it is clear that \nabla_{{\cal F}}{\left|_{{\phantom{|}\!\!}_{(M, I)}}\right.} equals \nabla.

Return to the proof of ?. Consider a {{\mathbb C}}^*-action \rho_I(z) on (M, I), (M, -I) induced from the natural embeddings (M,I)\hookrightarrow \operatorname{Tw}(M), (M,-I)\hookrightarrow \operatorname{Tw}(M). Then F_I is a {{\mathbb C}}^*-equivariant bundle. Since \iota\circ \rho(z) = \rho(\overline z) \circ \iota, the identification \operatorname{Lin}(M)=(M, I)\times (M, I) is compatible with {{\mathbb C}}^*-action. Let {\mathbf r}= \frac{d}{dr} be the holomorphic vector field on (M, I) corresponding to the {{\mathbb C}}^*-action. To prove ?, we have to show that the operator

vanishes.

Consider the equivariant structure operator

Let U be a {{\mathbb C}}^*-invariant Stein subset of (M, I). Consider \rho(z)^F an an endomorphism of the space of global holomorphic sections \Gamma_U(F_I). Let

for f\in \Gamma_U(F_I). Clearly, D_r is a well defined sheaf endomorphism of F_I, satisfying

for all \alpha\in {\cal O}_{(M,I)}. We say that a holomorphic section f of F_I is {{\mathbb C}}^*-invariant if D_r(f)=0. Clearly, the {\cal O}_{(M,I)}-sheaf F_I is generated by {{\mathbb C}}^*-invariant sections. Therefore, it suffices to check the equality

for holomorphic {{\mathbb C}}^*-invariant f\in F_I.

Since f is holomorphic, we have \nabla_{\overline {\mathbf r}} f =0. Thus,

We obtain that ? is implied by the following lemma.

Lemma .lemma: In the above assumptions, let f be a {{\mathbb C}}^*-invariant section of F_I. Then \nabla_{\mathbf r}(f)=0.

Return to the notation we used in the beginning of the proof of ?. Then, \nabla(f) = \overline \6(f) +\6^\Psi(f). Since f is holomorphic, \overline \6(f)=0, so we need to show that \6^\Psi(f)({\mathbf r})=0. By definition of \6^\Psi, this is equivalent to proving that

Consider the {{\mathbb C}}^*-action on \operatorname{Lin}(M) which is induced by the {{\mathbb C}}^*-action on \operatorname{Tw}(M). Since the maps p_I, p_{-I} are compatible with the {{\mathbb C}}^*-action, the sheaves F_1, F_{-1} are {{\mathbb C}}^*-equivariant. We can repeat the construction of the operator D_r for the sheaf F_{-I}. This allows one to speak of holomorphic {{\mathbb C}}^*-invariant sections of F_{-I}. Pick a {{\mathbb C}}^*-invariant Stein subset U\subset (M, -I). Since the statement of ? is local, we may assume that M=U. Let g_1, ... , g_n be a set of {{\mathbb C}}^*-invariant sections of F_I which generated F_I. Then, the sections p_{-I}^*(g_1), ..., p_{-I}^*(g_n) generate F_{-1}. Consider the section \Psi_{1,-1}(f) of F_{-1}. Clearly, \Psi_{1,-1} commutes with the natural {{\mathbb C}}^*-action. Therefore, the section \Psi_{1,-1}(f) is {{\mathbb C}}^*-invariant, and can be written as

where the functions \alpha_i are {{\mathbb C}}^*-invariant. By definition of \6 we have

On the other hand, g_i is a holomorphic section of F_{-I}, so \6 p_{-I}^* (g_i)=0. We obtain

Thus,

but since the functions \alpha_i are {{\mathbb C}}^*-invariant, their derivatives along {\mathbf r} vanish. We obtain \6\Psi_{1,-1}(f)({\mathbf r})=0. This proves ?. ? is proven. width 4pt height 4pt depth 0pt

The leaves of {{\mathbb H}}^*-action are parametrized by the points of q\in Q. Consider such a leaf M_q:=\varphi^{-1}(q)\subset M. Clearly, M_q is a hyperkähler submanifold in M, equipped with a special action of {{\mathbb H}}^*. Moreover, the restriction of {\cal F} to \operatorname{Tw}(M_q)\subset \operatorname{Tw}(M) satisfies assumptions of ?. To prove that \nabla_{{\cal F}} is flat along the leaves of {{\mathbb H}}^*-action, we have to show that {{\cal F}}{\left|_{{\phantom{|}\!\!}_{\operatorname{Tw}(M_q)}}\right.} is flat, for all q. Therefore, it suffices to prove ? for \dim_{{\mathbb H}} M=1.

Lemma .lemma: We work in notation and assumptions of ?. Assume that \dim_{{\mathbb H}} M=1. Then the connection \nabla_{{\cal F}} is flat.

Let I be an induced complex structure, and F_I:= F{\left|_{{\phantom{|}\!\!}_{(M,I)}}\right.} the corresponding holomorphic bundle on (M, I). Denote by z_I the vector field corresponding to the {{\mathbb C}}^*-action \rho_I on (M, I). By definition, the connection \nabla{\left|_{{\phantom{|}\!\!}_{F_I}}\right.} has SU(2)-invariant curvature \Theta_I. On the other hand, \Theta_I(z_I, \overline z_I)=0 by ?. Since \nabla_{{\cal F}}=\sigma^*\nabla is a pullback of an autodual connection \nabla on M, its curvature is a pullback of \Theta_I. In particular, \Theta= \Theta_I is independent from the choice of induced complex structure I. We obtain that \Theta(z_I, \overline z_I)=0 for all induced complex structures I on M.

Now ? is implied by the following linear-algebraic claim.

Claim .claim: Let M be a hyperkähler manifold equipped with a special {{\mathbb H}}^*-action, \dim_{{\mathbb H}}M=1. Consider the vectors z_I, \overline z_I defined above. Let \Theta be a smooth SU(2)-invariant 2-form, such that for all induced complex structures, I, we have \Theta(z_I, \overline z_I)=0. Then \Theta=0.

The proof of ? is an elementary calculation. Fix a point m_0\in M. Consider the flat hyperkähler manifold {{\mathbb H}}\backslash 0, equipped with a natural special action of {{\mathbb H}}^*. From the definition of a special action, it is clear that the map \rho defines a covering {{\mathbb H}}\backslash 0{\:\longrightarrow\:} M, h{\:\longrightarrow\:} \rho(h)m_0 of hyperkähler manifolds, and this covering is compatible with the special action. Therefore, the hyperkähler manifold M is flat, and the {{\mathbb H}}^*-action is linear in the flat coordinates.

Let

be the standard decomposition of \Lambda^2(M) according to the eigenvalues of the Hodge * operator. Consider the natural Hermitian metric on \Lambda^2(M). Then \Lambda^-(M) is the bundle of SU(2)-invariant 2-forms (see, e. g., [_Verbitsky:Hyperholo_bundles_]), and \Lambda^+(M) is its orthogonal complement. Consider the corresponding orthogonal projection \Pi:\; \Lambda^2(M) {\:\longrightarrow\:} \Lambda^-(M). Denote by dz_I\wedge d \overline z_I the differential form which is dual to the bivector z_I\wedge \overline z_I. Let R\subset \Lambda^-(M) be the C^\infty(M)-subsheaf of \Lambda^-(M) generated by \Pi(dz_I\wedge d \overline z_I), for all induced complex structures I on M. Clearly, \Theta\in \Lambda^-(M) and \Theta is orthogonal to R\subset \Lambda^-(M). Therefore, to prove that \Theta=0 it suffices to show that R= \Lambda^-(M). Since M is covered by {{\mathbb H}}\backslash 0, we may prove R= \Lambda^-(M) in assumption M={{\mathbb H}}\backslash 0.

Let \gamma be the real vector field corresponding to dilatations of M={{\mathbb H}}\backslash 0, and d\gamma the dual 1-form. Clearly,

Averaging d\gamma\wedge I(d\gamma) by SU(2), we obtain

where I, J, K is the standard triple of generators for quaternion algebra. Similarly,

and

Thus, \Pi(R) is a 3-dimensional sub-bundle of \Lambda^-(M). Since \dim \Lambda^-(M) =3, we have \Pi(R)= \Lambda^-(M). This proves ?. ? and ? is proven. width 4pt height 4pt depth 0pt

The argument is clear, but cumbersome, and essentially hinges on taking associate graded quotients everywhere and checking that all equations remain true. We give a simplified version of the proof, which omits some details and notation.

Consider the bundle {\cal F}{\left|_{{\phantom{|}\!\!}_{M\backslash s_x}}\right.}. This bundle is compatible with twistor transform, and therefore, is equipped with a natural connection \nabla_{{\cal F}}. This connection is constructed using the isomorphism \Psi_{1,-1}:\; F_1{\:\longrightarrow\:} F_{-1} (see the proof of ?). We apply the same consideration to {\cal F}^{gr}{\left|_{{\phantom{|}\!\!}_{(T_x M, I)}}\right.}, and show that the resulting connection \nabla_{{\cal F}^{gr}} is hyperholomorphic. This implies that {\cal F}^{gr} admits a (1,1)-connection which is a pullback of some connection on {\cal F}^{gr}{\left|_{{\phantom{|}\!\!}_{(T_x M, I)}}\right.}. This argument is used to prove that {\cal F}^{gr} is compatible with the twistor transform.

We use the notation introduced in the proof of ?. Let \operatorname{Lin}^{gr} be the space of twistor maps in \operatorname{Tw}^{gr}. Consider the maps p_{\pm I}^{gr}:\; \operatorname{Lin}^{gr}{\:\longrightarrow\:} (T_x M, \pm I) and the sheaves F^{gr}_{\pm1}:=(p_{\pm I}^{gr})^*{\cal F}^{gr}_{\pm I} obtained in the same way as the maps p_{\pm I} and the sheaves F_{\pm1} from the corresponding associated graded objects. Taking the associated graded of \Psi_{1,-1} gives an isomorphism \Psi^{gr}_{1,-1}:\; F^{gr}_1{\:\longrightarrow\:} F^{gr}_{-1}. Using the same construction as in the proof of ?, we obtain a connection operator

Since (\overline \6^{gr})^2 = (\6^{\Psi^{gr}})^2=0, the curvature of \nabla^{gr}_I has Hodge type (1,1) with respect to I. To prove that \nabla^{gr}_I is hyperholomorphic, we need to show that the curvature of \nabla^{gr}_I has type (1,1) with respect to every induced complex structure. Starting from another induced complex structure J, we obtain a connection \nabla^{gr}_J, with the curvature of type (1,1) with respect to J. To prove that \nabla^{gr}_J is hyperholomorphic it remains to show that \nabla^{gr}_J=\nabla^{gr}_I.

Let \nabla_I, \nabla_J be the corresponding operators on F_1. From the construction, it is clear that \nabla^{gr}_I, \nabla^{gr}_J are obtained from \nabla_I, \nabla_J by taking the associated graded quotients. On the other hand, \nabla_I = \nabla_J. Therefore, the connections \nabla^{gr}_I and \nabla^{gr}_J are equal. We proved that the bundle {\cal F}^{gr}{\left|_{{\phantom{|}\!\!}_{\operatorname{Tw}^{gr}\backslash s_0}}\right.} is compatible with the twistor transform. To prove ?, it remains to show that the natural connection on {\cal F}^{gr} is Hermitian.

The bundle {{\cal F}} {\left|_{{\phantom{|}\!\!}_{\operatorname{Tw}(M\backslash{x_0})}}\right.} is by definition Hermitian. Consider the corresponding isomorphism {\cal F} {\:\longrightarrow\:} (\iota^*\overline{{\cal F}})^* (?). Taking an associate graded map, we obtain an isomorphism

This gives a non-degenerate semilinear form h^{gr} on {\cal F}^{gr}. It remains only to show that h^{gr} is pseudo-Hermitian (i. e. satisfies h(x, y) = \overline{h(y,x)}) and positive definite.

Let M^{gr}_{{\mathbb C}} be a complexification of M^{gr}=T_x M, M^{gr}_{{\mathbb C}} = \operatorname{Lin}(M^{gr}). Consider the corresponding complex vector bundle {\cal F}^{gr}_{{\mathbb C}} over M^{gr}_{{\mathbb C}} underlying {\cal F}^{gr}. The metric h^{gr} can be considered as a semilinear form {\cal F}^{gr}_{{\mathbb C}}\times {\cal F}^{gr}_{{\mathbb C}}{\:\longrightarrow\:} {\cal O}_{M^{gr}_{{\mathbb C}}}. This semilinear form is obtained from the corresponding form h on {\cal F} by taking the associate graded quotients. Since h is Hermitian, the form h^{gr} is pseudo-Hermitian. To prove that h^{gr} is positive semidefinite, we need to show that for all f\in {\cal F}^{gr}_{{\mathbb C}}, the function h^{gr}(f, \overline f) belongs to {\cal O}_{M^{gr}_{{\mathbb C}}}\cdot\overline{\cal O}_{M^{gr}_{{\mathbb C}}}, where {\cal O}_{M^{gr}_{{\mathbb C}}}\cdot\overline{\cal O}_{M^{gr}_{{\mathbb C}}} denotes the {{\mathbb R}}^{>0}-semigroup of {\cal O}_{M^{gr}_{{\mathbb C}}} generated by x\cdot\overline x, for all x\in {\cal O}_{M^{gr}_{{\mathbb C}}}. A similar property for h holds, because h is positive definite. Clearly, taking associated graded quotient of the semigroup {\cal O}_{M_{{\mathbb C}}}\cdot\overline{\cal O}_{M_{{\mathbb C}}}, we obtain {\cal O}_{M^{gr}_{{\mathbb C}}}\cdot\overline{\cal O}_{M^{gr}_{{\mathbb C}}}. Thus,

This proves that h^{gr} is positive semidefinite. Since h^{gr} is non-degenerate, this form in positive definite. ? is proven. width 4pt height 4pt depth 0pt

By definition, coherent sheaves on C= {{\mathbb P}}T_xM correspond bijectively to {{\mathbb C}}^*-equivariant sheaves on T_xM\backslash 0. Let F^{gr} be the associated graded sheaf of F (Subsection ?). Consider F^{gr} as a bundle on T_xM\backslash 0. In the notation of ?, F^{gr}= {\cal F}^{gr}{\left|_{{\phantom{|}\!\!}_{(M, I)}}\right.}. By ?, the sheaf {\cal F}^{gr} is {{\mathbb C}}^*-equivariant.

The bundle F^{gr} is obtained as a pullback of \pi^*F{\left|_{{\phantom{|}\!\!}_{C}}\right.}, where

denotes the standard projection. The vector bundles on {{\mathbb C}} P T_xM is the same thing as {{\mathbb C}}^*-equivariant vector bundles on T_xM\backslash 0. We obtained two {{\mathbb C}}^*-equivariant structures on F^{gr}: one is given by ?, another is given from the identification F^{gr}=\pi^*F{\left|_{{\phantom{|}\!\!}_{C}}\right.}.

The quotient of these two equivariant structures gives a {{\mathbb C}}^*-action \rho on F^{gr}. This gives a decomposition F^{gr}=\bigoplus F^{gr}_i(i), with \rho(t){\left|_{{\phantom{|}\!\!}_{F^{gr}_i}}\right.}=t^i.

Consider now the quaternionic projective space {{\mathbb H}}P T_xM. For each p\in{{\mathbb H}}P T_xM, denote by H_p the fiber of the natural projection

isomorphic to {{\mathbb H}}\backslash 0. The bundles F^{gr}_i are trivial on each of the fibers H_p. Therefore, these bundles are compatible with the twistor transform. From ? it follows that each of F^{gr}_i is Yang-Mills and polystable. width 4pt height 4pt depth 0pt

[_OSS_], Lemma 1.1.12. width 4pt height 4pt depth 0pt

Clearly, it suffices to prove ? for {\frak F} reflexive. Let \widetilde X be the blow-up of (M, I)\times S in \{x_0\} \times S, and \widetilde {\frak F} the pullback of \frak F to \widetilde X. Clearly, \widetilde X= \widetilde M\times S, where \widetilde M is a blow-up of (M,I) in x_0. Denote by C\subset \widetilde M the blow-up divisor of \widetilde M. Taking S sufficiently small, we may assume that the bundle \frak F{\left|_{{\phantom{|}\!\!}_{\{x_0\}\times(S\backslash\{s_0\})}}\right.} is trivial. Thus, the bundle \widetilde{\frak F}{\left|_{{\phantom{|}\!\!}_{(C\times S)\backslash (C\times \{s_0\})}}\right.}, which is a pullback of \frak F{\left|_{{\phantom{|}\!\!}_{\{x_0\}\times(S\backslash\{s_0\})}}\right.} under the natural projection (C\times S)\backslash (C\times \{s_0\}) {\:\longrightarrow\:}\{x_0\}\times(S\backslash\{s_0\} is trivial. To prove that \frak F is locally trivial, we have to show that \widetilde {\frak F} is locally trivial, and that the restriction of \widetilde {\frak F} to C\times S is trivial along the fibers of the natural projection C \times S{\:\longrightarrow\:} S. Clearly, to show that \widetilde {\frak F} is locally trivial we need only to prove that the fiber \widetilde {\frak F}{\left|_{{\phantom{|}\!\!}_{z}}\right.} has constant dimension for all z\in C \times S. Thus, \widetilde {\frak F} is locally trivial if and only if \widetilde {\frak F}{\left|_{{\phantom{|}\!\!}_{C\times S}}\right.} is locally trivial. This sheaf is reflexive, since it corresponds to an associate graded sheaf of a reflexive sheaf, in the sense of Footnote to ?. It is non-singular in codimension 2, because all reflexive sheaves are non-singular in codimension 2 ([_OSS_], Ch. II, Lemma 1.1.10). By ?, the sheaf \widetilde {\frak F}{\left|_{{\phantom{|}\!\!}_{C\times\{s\}}}\right.} is a direct sum of stable bundles, with the slopes depending on the equivariant action. Since the equivariant action is deformed continuously with the sheaf, and the corresponding numbers are all 0 when the sheaf is smooth, they are all 0 in the singular points too. Then ? implies that it is polystable. ? is implied by the following lemma, applied to the sheaf \widetilde {\frak F}{\left|_{{\phantom{|}\!\!}_{C\times S}}\right.}.

Lemma .lemma: Let C be a complex projective space, S a complex variety and \frak F a torsion-free sheaf over C\times S. Consider an open set U\stackrel j\hookrightarrow C\times S, which is a complement of C\times\{s_0\}\subset C\times S. Assume that the sheaf {\frak F}{\left|_{{\phantom{|}\!\!}_{U}}\right.} is trivial: {\frak F}{\left|_{{\phantom{|}\!\!}_{U}}\right.}\cong {\cal O}_U^n. Assume, moreover, that \frak F is non-singular in codimension 2, and the sheaf \left({\frak F}{\left|_{{\phantom{|}\!\!}_{C \times\{s_0\}}}\right.}\right)^{**} is polystable.

Then the reflexization {\frak F}^{**} of \frak F is a trivial bundle.

Using induction, it suffices to prove ? assuming that it is proven for all \frak F' with \operatorname{rk} {\frak F}'<\operatorname{rk} {\frak F}. We may also assume that S is Stein, smooth and 1-dimensional.

Step 1: We construct an exact sequence

of sheaves of positive rank, which, as we prove in Step 3, satisfy assumptions of ?.

Consider the pushforward sheaf j_* {\cal O}_U^n. From the definition of j_*, we obtain a canonical map

and the kernel of this map is a torsion subsheaf in {\frak F}.

Let f be a coordinate function on S, which vanishes in s_0\in S. Clearly,

On the other hand, the sheaf {\cal O}_{C\times S}\left[\frac{1}{f}\right] is a direct limit of the following diagram:

where \cdot f is the injection given by the multiplication by f. Thus, the map (?) gives an embedding

which is identity outside of (x_0, s_0). Multiplying p by \frac{1}{f} if necessary, we may assume that the restriction p{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.} is non-trivial. Thus, p gives a map

with image of positive rank. Since both sides of (?) are semistable of slope zero, and {\cal O}^n_{C\times \{s_0\}} is polystable, the map (?) satisfies the following conditions. (see [_OSS_], Ch. II, Lemma 1.2.8 for details).

Let O_1 = {\cal O}^k_{C\times S}\subset {\cal O}^n_{C\times S} be the corresponding free subsheaf of {\cal O}^n_{C\times S}. Consider the natural projection \pi_{O_1} of {\cal O}^n_{C\times S} to O_1. Let p_{O_1} be the composition of p and \pi_{O_1}, \frak F_1 the image of p_{O_1}, and \frak F_2 the kernel of p_{O_1}.

Step 2: We show that the sheaves \frak F_2 and \frak F_1 and non-singular in codimension 2.

Consider the exact sequence

obtained by tensoring the sequence

with {\cal O}_{C\times \{s_0\}}. From this sequence, we obtain an isomorphism {\frak F}_1{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}\cong F_1.

A torsion-free coherent sheaf over a smooth manifold is non-singular in codimension 1 ([_OSS_], Ch. II, Corollary 1.1.8).

Since \frak F is non-singular in codimension 2, the restriction \frak F{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.} is non-singular in codimension 1. Therefore, the torsion of \frak F{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.} has support of codimension at least 2 in C\times \{s_0\}. Since the sheaf F_2 is a subsheaf of \frak F{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}, its torsion has support of codimension at least 2. Therefore, the singular set of F_2 has codimension at least 2 in C\times \{s_0\}. The rank of F_2 is by definition equal to n-k.

Since F_1 has rank k, the singular set of \frak F_1 coincides with the singular set of F_1. Since the restriction {\frak F}_1{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}=F_1, is a subsheaf of a trivial bundle of dimension k on C\times \{s_0\}, it is torsion-free. Therefore, the singularities of {\frak F}_1 have codimension at least 2 in C\times \{s_0\}.

We obtain that the support of Tor^1({\cal O}_{C\times \{s_0\}}, {\frak F}_1) has codimension at least 2 in C\times \{s_0\}. Since the quotient sheaf

is isomorphic to the sheaf F_2, this quotient is non-singular in codimension 1. Since we proved that F_2 is non-singular in codimension 1, the sheaf \frak F_2 {\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.} is also non-singular in codimension 1, and its rank is equal to the rank of F_2.

Let R be the union of singular sets of the sheaves \frak F_2, \frak F, {\frak F}_1. Clearly, R is contained in C\times \{s_0\}, and R coincides with the set of all x\in C\times \{s_0\} where the dimension of the fiber of the sheaves \frak F_2, \frak F, {\frak F}_1 is not equal to n-k, n, k. We have seen that the restrictions of \frak F_2, {\frak F}_1 to {C\times \{s_0\}} have ranks n-k, k. Therefore, the singular sets of \frak F_2, {\frak F}_1 coincide with the singular sets of \frak F_2{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}, {\frak F}_1{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}. We have shown that these singular sets have codimension at least 2 in C\times \{s_0\}. On the other hand, \frak F is non-singular in codimension 2, by the conditions of ?. Therefore, R has codimension at least 3 in C\times S.

Step 3: We check the assumptions of ? applied to the sheaves \frak F_2, {\frak F}_1.

Since the singular set of {\frak F}_1 has codimension 2 in C\times \{s_0\}, the {\cal O}_{C\times \{s_0\}}-sheaf Tor^1({\cal O}_{C\times \{s_0\}},{\frak F}_1) is a torsion sheaf with support of codimension 2 in C\times \{s_0\}. By (?), the reflexization of \frak F_2{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.} coincides with the reflexization of F_2. Thus, the sheaf \left(\frak F_2{\left|_{{\phantom{|}\!\!}_{C\times \{s_0\}}}\right.}\right)^{**} is semistable. On the other hand, outside of {C\times \{s_0\}}, the sheaf \frak F_2 is a trivial bundle. Thus, \frak F_2 satisfies assumptions of ?. Similarly, the sheaf \frak F_1 is non-singular in codimension 2, its restriction to C\times \{s_0\} has trivial reflexization, and it is free outside of C\times \{s_0\}.

Step 4: We apply induction and prove ?.

By induction assumption, the reflexization of \frak F_2 is isomorphic to a trivial bundle {\cal O}^{n-k}_{C\times S}. and reflexization of \frak F_1 is {\cal O}^k_{C\times S}. We obtain an exact sequence

where the sheaves \frak F_2 and {\frak F}_1 have trivial reflexizations.

Let V:= C\times S\backslash R. Restricting the exact sequence (?) to V, we obtain an exact sequence

Since V is a complement of a codimension-3 complex subvariety in a smooth Stein domain, the first cohomology of a trivial sheaf on V vanish. Therefore, the sequence (?) splits, and the sheaf {\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.} is a trivial bundle. Consider the pushforward \zeta_*{\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.}, where \zeta:\; V{\:\longrightarrow\:} C\times S is the standard map. Then \zeta_*{\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.} is a reflexization of {\frak F} (a pushforward of a reflexive sheaf over a subvariety of codimension 2 or more is reflexive – see ?). On the other hand, since the sheaf {\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.} is a trivial bundle, its push-forward over a subvariety of codimension at least 2 is also a trivial bundle over C\times S. We proved that the sheaf {\frak F}^{**}=\zeta_*{\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.} is a trivial bundle over C\times S. The push-forward \zeta_*{\frak F}{\left|_{{\phantom{|}\!\!}_{V}}\right.} coincides with reflexization of \frak F, by ?. This proves ? and ?. width 4pt height 4pt depth 0pt

See, e. g., [_Maruyama:Si_]. width 4pt height 4pt depth 0pt

The complex structure I is by definition algebraic, with unique polarization. This makes it possible to speak of Gieseker–Maruyama stability on (M, I). Denote by {\cal M}_{mar} the Maruyama moduli of deformations of F. Then {\cal M} is naturally an open subset of {\cal M}_{mar}. Let s\in {\cal M}_{mar} be an arbitrary point and F_s the corresponding coherent sheaf on (M, I), defined up to S-equivalence. According to ?, the Chern classes of F and F_s are equal. Thus, by ?, the sheaf F_s is hyperholomorphic. Therefore, F_s admits a filtration with hyperholomorphic stable quotient sheaves F_i, i= 1, ..., m. By ?, the singular set S of F_s is trianalytic. Since M has no proper trianalytic subvarieties, S is a collection of points. We obtain that F_s has isolated singularities. Let \frak F be a family of deformations of F, parametrized by Y. The points y\in Y correspond to deformations F_y of F_s. Assume that for all y\in Y, y\neq s, the sheaf F_y is a bundle. Since {\cal M} is open in {\cal M}_{mar}, such a deformation always exists.

The sheaf F_s has isolated singularities and admits a filtration with hyperholomorphic stable quotient sheaves. This implies that the family \frak F satisfies the conditions of ?. By ?, the reflexization \frak F^{**} is locally trivial. To prove that {\cal M}= {\cal M}_{mar}, we have to show that for all s\in {\cal M}_{mar}, the corresponding coherent sheaf F_s is locally trivial. Therefore, to finish the proof of ?, it remains to prove the following algebro-geometric claim.

Claim .claim: Let X be a compact complex manifold, \dim_{{\mathbb C}} X>2, and \frak F a torsion-free coherent sheaf over X\times Y which is flat over Y. Assume that the reflexization of \frak F is locally trivial, \frak F has isolated singularities, and for some point s\in Y, the restriction of \frak F to the complement (X\times Y)\backslash (X\times \{s\}) is locally trivial. Then the reflexization \left(\frak F{\left|_{{\phantom{|}\!\!}_{X}}\right.}\times\{s\}\right)^{**} is locally trivial.

Remark .remark: We say that a kernel of a map from a bundle to an Artinian sheaf is a bundle with holes. In slightly more intuitive terms, ? states that a flat deformation of a bundle with holes is again a bundle with holes, and cannot be smooth, assuming that \dim_{{\mathbb C}} X>2.

? is well known. Here we give a sketch of a proof. Consider a coherent sheaf F_s= \frak F {\left|_{{\phantom{|}\!\!}_{X\times \{s\}}}\right.}, and an exact sequence

where k is an Artinian sheaf. By definition, the sheaf F_s^{**} is locally trivial. The flat deformations of F_s are infinitesimally classified by Ext^1(F_s, F_s). Replacing F_s by a quasi-isomorphic complex of sheaves F_s^{**} {\:\longrightarrow\:} k, we obtain a spectral sequence converging to Ext^\bullet(F_s, F_s). In the E_2-term of this sequence, we observe the group

which is responsible for Ext^1(F_s, F_s). The term Ext^1(F_s^{**}, F_s^{**}) is responsible for deformations of the bundle F_s^{**}, the term Ext^0(F_s^{**}, k) for the deformations of the map F_s^{**} {\:\longrightarrow\:} k, and the term Ext^1(k, k) for the deformations of the Artinian sheaf k. Thus, the term Ext^2(k, F_s^{**}) is responsible for the deformations of F_s which change the dimension of the cokernel of the embedding F_s{\:\longrightarrow\:} F_s^{**}. We obtain that whenever Ext^2(k, F_s^{**})=0, all deformations of F_s are singular. On the other hand, Ext^2(k, F_s^{**})=0, because the i-th Ext from the skyscraper to a free sheaf on a manifold of dimension more than i vanishes (this is a basic result of Grothendieck’s duality, [_Hartshorne:Grothendieck's_]). width 4pt height 4pt depth 0pt

By Proof, for all \omega \in Pic(M, I), we have \deg_I(\omega)=0. Therefore, every subsheaf of B is destabilising. width 4pt height 4pt depth 0pt

? (i) is a consequence of [_coho_announce_], Corollary 5.1. Here we give a sketch of its proof.

Let I be an induced complex structure of general type. By ?, a bundle B over (M, I) is stable if and only if it is simple. Thus, the category Bun_I({\cal H}) is independent from the choice of {\cal H} (?).

In ?, we constructed the equivalence of categories \Phi_{I, J}, which gives the functor \Phi_\gamma for twistor path which consists of a single twistor curve. This proves ? (i) for n=1. A composition of isomorphisms \Phi_{I, J}\circ \Phi_{J, J'} is well defined, because the category Bun_I({\cal H}) is independent from the choice of {\cal H}. Taking successive compositions of the maps \Phi_{I, J}, we obtain an isomorphism \Phi_\gamma. This proves ? (i).

The variety {\cal M}_{I, {\cal H}}(B) is singular hyperkähler ([_Verbitsky:Hyperholo_bundles_]), and the variety {\cal M}_{J, {\cal H}}(B) is the same singular hyperkähler variety, taken with another induced complex structure. By definition of singular hyperkähler varieties, this implies that {\cal M}_{I, {\cal H}}(B), {\cal M}_{J, {\cal H}}(B) are real analytic equivalent, with equivalence provided by \Phi_{I, J}. This proves ? (ii). width 4pt height 4pt depth 0pt

This is [_coho_announce_], Theorem 5.2. $ width 4pt height 4pt depth 0pt

This is [_coho_announce_], Corollary 5.2. \hbox{\vrule width 4pt height 4pt depth 0pt}

Clearly, the map \deg_\omega:\; H^2(M) {\:\longrightarrow\:} {{\mathbb R}} is equal (up to a scalar multiplier) to the orthogonal projection onto the line {{\mathbb R}}\cdot\omega. Then, ? is equivalent to ?, (i). width 4pt height 4pt depth 0pt

Consider the twistor curves P_0, P_n corresponding to {\cal H}_0, {\cal H}_n. Since a, b are generic, the curves P_0, P_n intersect with Comp_0. Applying ?, we connect the curves P_0 and P_n by an admissible path. width 4pt height 4pt depth 0pt

Clear. width 4pt height 4pt depth 0pt

Let \gamma be the twistor path of ?. By ?, \Phi_\gamma(T(M, I)) = T(M, J). Applying ?, we obtain a real analytic equivalence from {\cal M}_{I, a}(T(M, I)) to {\cal M}_{J, b}(T(M, J)). width 4pt height 4pt depth 0pt

The variety {\cal M}_F^I is compact by ?. In [_Verbitsky:Hyperholo_bundles_], it was proven that the space {\cal M}_F^I of stable deformations of F is a singular hyperkähler variety (see also [_NHYM_] for an explicit construction of the twistor space of {\cal M}_F^I). Then ? is a consequence of the Desingularization Theorem for singular hyperkähler varietiess (?). width 4pt height 4pt depth 0pt

Clearly, F' satisfies assumptions of ?, and the moduli space of its stable deformations is compact. Since \Phi_\gamma induces a homeomorphism of moduli spaces (?), the space of stable deformations of F is also compact. width 4pt height 4pt depth 0pt

Follows from ? and ?. width 4pt height 4pt depth 0pt

The proof of ? is based on the same ideas as the proof of ?. width 4pt height 4pt depth 0pt

Follows from ? and ?. width 4pt height 4pt depth 0pt

Consider the Jordan–Hölder serie for B'. Let Q_1 \oplus Q_2 \oplus ... be the associated graded sheaf. By ?, the stable bundles Q_i are hyperholomorphic. Since c_{\operatorname{rk} B}(B)\neq 0, we have c_{\operatorname{rk} Q_i}(Q_i)\neq 0. Therefore, the bundles Q_i are non-trivial. By ?, \operatorname{rk} Q_i >2. Since all the Chern classes of the bundles Q_i are SU(2)-invariant, the odd Chern classes of Q_i vanish (?, ?). Therefore, \operatorname{rk} Q_i\geqslant 4 for all i. Since \operatorname{rk} B\leqslant 6, we have i=1 and the bundle B' is stable. width 4pt height 4pt depth 0pt

By ?, all semistable bundle deformations of TX are stable. Then ? is implied by ?. width 4pt height 4pt depth 0pt

An elementary argument using the period maps, in the spirit of Subsection ?. width 4pt height 4pt depth 0pt

Given a decomposition \operatorname{Def}(M) = M_1\times M_2, we obtain a parallel 2-form on \Omega_1 on \operatorname{Def}(B), which is a pullback of the holomorphic symplectic form on M_1. Consider the space {\cal A} of connections on B, which is an infinitely-dimensional complex analytic Banach manifold. Then \Omega_1 corresponds to a holomorphic 2-form \widetilde \Omega_1 on {\cal A}. Since \Omega_1 is parallel with respect to the natural connection on \operatorname{Def}(B), the form \widetilde \Omega_1 is also a parallel 2-form on the tangent space to {\cal A}, which is identified with \Omega^1(X, \operatorname{End}(B)). It is possible to prove that this 2-form is obtained as

where

is a certain holomorphic pairing on the bundle \Omega^1(Y, \operatorname{End}(B)), and Y is a trianalytic subvariety of X. Since X has no trianalytic subvarieties, \widetilde \Omega_1 is obtained from a {\cal O}_X-linear pairing

Using stability of B, it is possible to show that such a pairing is unique, and thus, \Omega_1 coincides with the holomorphic symplectic form on \operatorname{Def}(B). Therefore, \operatorname{Def}(B) = M_1, and this manifold is simple. width 4pt height 4pt depth 0pt