Deforming positive triples to hyperkähler geometry

Source and verification

This article explains family 341 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

🏷️ Preliminaries: positive triples in four dimensions

On an oriented four-manifold, the wedge product of two two-forms is a four-form. After choosing a positive volume form, this gives a symmetric bilinear form on the six-dimensional space of two-forms. Its signature is $(3,3)$.

A hypersymplectic triple is an ordered triple of smooth closed real two-forms ${\omega }_1,{\omega }_2,{\omega }_3$ spanning a positive definite three-plane at every point. Equivalently, every nonzero constant linear combination is symplectic and induces the given orientation.

The triple determines a conformal metric: its span is the self-dual subspace of two-forms. Choosing a volume fixes the metric within that conformal class. The forms are closed and self-dual, hence harmonic, but they need not be parallel.

A hyperkähler triple is stronger. Its three forms are parallel and correspond to parallel complex structures satisfying the quaternionic identities. In dimension four, closed forms with the pointwise orthonormal wedge identities below give such a structure.

The global intersection matrix is

$$G_{ij}={\int }_X{\omega }_i\wedge {\omega }_j.$$

It is positive definite. A constant linear change of basis normalizes $G$ to the identity. A cohomology class $[{\omega }_i]$ records the closed form up to an exact change, so preserving it preserves all periods over closed two-cycles.

Manuscript result

On a closed connected oriented smooth four-manifold, every hypersymplectic triple normalized by $G_{ij}={\delta }_{ij}$ admits a smooth path of hypersymplectic triples with

$$\begin{array}{c}\omega (0)=\omega ,[{\omega }_i(t)]=[{\omega }_i],\\ {\omega }_i(1)\wedge {\omega }_j(1)=2{\delta }_{ij}{\mu }_1\end{array}$$

for a positive volume form ${\mu }_1$. The endpoint is a parallel hyperkähler triple. (OpenAI, 2026)

The normalization implies ${\int }_X{\mu }_1=1/2$. Without normalization, undoing the constant basis change gives parallel endpoint forms with their original Gram matrix $G$.

This is a deformation theorem. It does not assert global existence, convergence, or uniqueness for the particular PDE called hypersymplectic flow.

The prerequisites are differential forms, Hodge decomposition, symplectic geometry, and elliptic estimates. The topological and curve-existence results used to construct a fibration are stated as interfaces; their classification theory is not developed here.

🏷️ Definite pairs and the third-form problem

Write the triple as $(A,C,B)$. A definite pair $(A,C)$ has a positive definite pointwise wedge matrix. Ratios of four-forms define smooth functions

$$\begin{array}{c}x=\frac{A\wedge C}{A^2},\\ y={\left(\frac{C^2}{A^2}-x^2\right)}^{1/2}>0.\end{array}$$

Here $A^2=A\wedge A$. Put $T=(C-xA)/y$. Then

$$A\wedge T=0,T^2=A^2.$$

The complex two-form $\Omega =A+iT$ satisfies ${\Omega }^2=0$ and $\Omega \wedge \overline{\Omega }>0$. Its kernel in the complexified tangent bundle determines an almost complex structure $J$, up to sign. The forms $A,C$ span the real anti-invariant two-forms for that structure.

A third form $B$ completes a positive triple precisely when it tames one of the two signs:

$$B(v,Jv)>0(v\ne 0).$$

On the connected manifold, $B$ chooses the sign globally. Thus deforming a pair reduces the triple problem to retaining a closed tamer in the original third class.

Existence of some compatible form for each $J$ would not suffice. Its class could vary, while the theorem requires $[B]$ to remain fixed.

🏷️ A prescribed-class criterion

The initial triple and a topological bound imply $b^{+}(X)=3$, where $b^{+}$ is the positive index of the intersection form on real degree-two cohomology. After global normalization,

$$V=[A{]}^{\perp }\cap [C{]}^{\perp }$$

has signature $(1,b^{-})$. Its positive-square vectors have two components. The class $[B]$ chooses one.

The source proves a criterion for this setting. Suppose a positive-square class $U\in V$ contains a tamer for the almost complex structure determined by the closed definite pair. A class $H\in V$ contains a tamer if it has positive square, lies in the same positive component as $U$, and

$$H\cdot d>0$$

for every irreducible closed $J$-holomorphic curve of class $d$.

A $J$-holomorphic curve is a surface whose differential intertwines its complex structure with $J$. The pairing $H\cdot d$ is its area as measured by a closed representative of $H$, so positivity is necessary. The theorem supplies sufficiency under the definite-pair and cohomological hypotheses.

The proof starts from positive-current separation. Mass and angular estimates establish closedness of suitable density weights. The diffuse part is handled using the intersection form, while normalized thresholds of the positive-density part produce integral curve cycles.

Compatibility is established before applying the integral-cycle regularity theorem, so the local symplectic hypothesis required by that theorem is available. This prescribed-class result is an additional step beyond the unrestricted compatibility theorem explained in family 342.

🏷️ From the coefficient sphere to a torus fibration

The initial triple gives a sphere of almost complex structures. For $v\in S^2$, take the anti-invariant plane consisting of forms $a\cdot \omega$ with $a\perp v$, and choose the sign tamed by $v\cdot \omega$.

At each point, this coefficient sphere maps with degree of absolute value one to the sphere of compatible pointwise complex structures in the determined conformal metric. Compatible forms can be chosen smoothly over the parameter sphere by an elliptic correction.

The manuscript combines this degree with a marked family curve theorem to produce curves through every point in a suitable nonzero integral class $F$ with

$$F^2=0.$$

The marked-point condition has a parameter-dependent contribution: the relevant line in the family of equations varies over $S^2$. Its degree must be retained even though the underlying topological structure is fixed.

A topological choice of $F$ prevents these curve configurations from splitting. Adjunction and positivity of curve intersections restrict the possible fibers to tori and rational curves with one singularity. Exact perturbations of the pair $(A,C)$, keeping $B$ fixed, give the needed transversality. A relative local replacement arranges ordinary nodes.

The output is a proper fibration over a closed oriented surface, with torus regular fibers and rational singular fibers having one ordinary node. The forms are then prepared by further exact deformations so that their regular-fiber behavior is invariant under torus translations.

These are substantial construction theorems. The argument does not assume an initial smooth identification with a K3 manifold or a torus, or rationality of the original positive period plane.

🏷️ An auxiliary equal-square form

The fibration is used to build a closed auxiliary form $D=D_{\epsilon }$ with

$$\begin{array}{c}A\wedge D=C\wedge D=0,\\ D^2=A^2>0.\end{array}$$

The parameter $\epsilon$ controls the small positive area of a regular torus fiber. On regular regions a translation-invariant model is available. Near nodes, explicit local models are glued to it across fixed annuli.

Matching fiber areas alone is insufficient to make the discrepancy exact. The construction also matches a period over a surface swept out by a monodromy-invariant fiber cycle. With these periods matched, the overlap error has a primitive.

The gluing error is exponentially small in ${\epsilon }^{-1}$. Correcting it requires estimates whose losses are only powers of ${\epsilon }^{-1}$. Thus the error can dominate the analytic loss as $\epsilon \downarrow 0$.

A useful exact-form identity explains the correction space. If $h=d\alpha$ is an exact two-form on a closed oriented four-manifold, then Stokes gives

$$0={\int }_{X}h\wedge h=\Vert h^{+}{\Vert }_2^2-\Vert h^{-}{\Vert }_2^2.$$

Consequently

$$\Vert h{\Vert }_2^2=2\Vert h^{+}{\Vert }_2^2.$$

Controlling the self-dual part controls the full exact correction. The source combines this identity with finite-order Sobolev estimates as the fibers collapse. A scalar complex Monge–Ampère argument cannot simply be used at this stage: a global integrable complex structure has not yet been constructed.

The class of $D_{\epsilon }$ is auxiliary and generally differs from $[B]$. Its role is to help keep the required class available, not to replace it permanently.

🏷️ One auxiliary choice for every later curve

The source obtains a class expansion

$$\begin{array}{c}[D_{\epsilon }]={\lambda }_{\epsilon }F+\epsilon u_{\epsilon },\\ {\lambda }_{\epsilon }\asymp {\epsilon }^{-1},\Vert u_{\epsilon }\Vert \le K.\end{array}$$

The auxiliary and original classes lie in the same positive component of $V$. More importantly, one sufficiently small choice of $\epsilon$ works for all later definite pairs in the fixed classes $[A],[C]$ and all their irreducible curve classes:

$$[D_{\epsilon }]\cdot d>0⟹[B]\cdot d>0.$$

This uniformity is essential. Choosing $\epsilon$ separately for each curve would not give a single auxiliary form supporting an entire deformation.

Once $D$ tames a moving pair’s structure, its positivity on those curves transfers to $[B]$. The prescribed-class criterion then supplies a tamer in $[B]$ at every parameter.

🏷️ Closed equal-square pairs produce integrability

If $A,D$ are closed with $A\wedge D=0$ and $A^2=D^2>0$, then $\Omega =A+iD$ is closed and decomposable. If vector fields $U,V$ lie in its kernel, Cartan’s formula gives

$$\begin{array}{rl}{\iota }_{[U,V]}\Omega & ={\mathcal{L}}_U({\iota }_V\Omega )\\ & -{\iota }_V({\mathcal{L}}_U\Omega )=0.\end{array}$$

The kernel is therefore involutive. The smooth integrability theorem supplies holomorphic coordinates, and $\Omega$ becomes a nowhere-zero holomorphic two-form.

If a third closed form $C$ is orthogonal to both and has the same square, it is a positive $(1,1)$ form for the appropriate sign, hence Kähler. The holomorphic volume form has constant norm, making it parallel. The resulting three forms are parallel and give the hyperkähler structure.

This explains why the auxiliary equal-square equation is a useful intermediate target. It provides integrability before invoking classical Kähler volume theorems.

🏷️ Two volume equations in the original classes

First use $A\pm iD$ to obtain an integrable complex structure for which the prepared $C$ is a tamer. A compact-surface criterion and the numerical positivity conditions put $[C]$ in its Kähler cone. The prescribed-volume theorem gives

$$\begin{array}{c}{C}_1\in [C],C_1^2=A^2,\\ A\wedge C_1=D\wedge C_1=0.\end{array}$$

Interpolate by $C_t=(1-t)C+t{C}_1$. Both endpoints tame this integrable structure, so their convex combinations do as well. The pairs $(A,C_t)$ remain definite, and $D$ tames their associated almost complex structures $J_t$.

The uniform class implication above and the prescribed-class criterion show that $[B]$ contains a tamer for every $J_t$. The sign at the starting point agrees with the original $B$: both $B$ and $D$ are positive on the oriented regular fibers, and the sign cannot jump on connected $X$.

At $t=1$, the pair $(A,C_1)$ is itself closed, orthogonal, and equal-square. It therefore defines a second integrable structure. The same class and volume criteria now give

$$\begin{array}{c}{B}_1\in [B],B_1^2=A^2,\\ A\wedge B_1=C_1\wedge B_1=0.\end{array}$$

Finally select closed tamers $B_t\in [B]$ smoothly, with the prescribed actual endpoints $B_0=B$ and $B_1$. The selection is elementary once existence at every parameter is known. A fixed tamer remains positive for nearby parameters; choose finitely many such neighborhoods and representatives, then combine them using a partition of unity in the parameter only. Closedness and the cohomology class are preserved, and positivity follows from convexity.

Thus $(A,C_t,B_t)$ is a path in the original three classes ending at $(A,C_1,B_1)$. The preparatory paths are concatenated after flattening their parameterizations at endpoints. Undoing the initial orthogonal coefficient change restores the ordered original classes.

🏷️ Individual forms and the scope of the conclusion

For any fixed nonzero coefficient vector $c$, the path

$${\alpha }_t=\sum _i{c}_i{\omega }_i(t)$$

is symplectic and stays in one class. Write ${\dot{\alpha }}_t=d{\sigma }_t$ and solve ${\iota }_{V_t}{\alpha }_t=-{\sigma }_t$. Cartan’s formula shows that the associated isotopy satisfies ${\varphi }_t^{\ast }{\alpha }_t={\alpha }_0$.

At the endpoint, ${\alpha }_1$ is a Kähler form for a suitable parallel complex structure and rescaling of the endpoint metric. Pulling these back makes the original ${\alpha }_0$ Kähler for some hyperkähler metric. This is Moser’s argument applied to one chosen combination.

The isotopy depends on $c$. The manuscript does not claim that one diffeomorphism simultaneously transports all three original forms to a parallel triple.

The global endpoint also gives the claimed smooth-type consequence: $X$ is diffeomorphic to the standard K3 manifold or $T^4$, using the classification of compact hyperkähler surfaces. That classification is a consequence applied after constructing the endpoint; it is not an assumption used to identify the original smooth manifold.

🏷️ Key steps and proof outlines

Three moment conditions are the entire linear obstruction

Let $(A,C,D_{\epsilon }^0)$ be the approximate positive triple from the glued collapsing models. To correct it by an exact form $h$, linearize the three wedge equations:

$$A\wedge h=f_1,C\wedge h=f_2,D_{\epsilon }^0\wedge h=f_3.$$

The pointwise positive Gram matrix determines the self-dual part $\varphi =h^{+}$ uniquely. The three conditions ${\int }_X{f}_i=0$ make $\varphi$ orthogonal to $A,C,D_{\epsilon }^0$. Since these closed self-dual forms span the three-dimensional harmonic self-dual space, these are precisely the cokernel conditions for solving $d^{+}a=\varphi$.

For a fixed metric, the Hodge Green operator gives an exact solution $h=2d{d}^{\ast }G\varphi$. Quantitative estimates do not require bounding that Green operator on one-forms: Stokes gives $\Vert h{\Vert }_2=\sqrt{2}\Vert \varphi {\Vert }_2$. Moreover, $dh=0$ and $d^{\ast }h=-2\ast d\varphi$. The Weitzenböck identity and its differentiated versions therefore bound higher derivatives using only curvature derivatives and lower derivatives of $h$. Polynomial geometry bounds in ${\epsilon }^{-1}$ give a polynomial inverse bound with no derivative loss. (OpenAI, 2026)

Exactness keeps every nonlinear iterate solvable

Write $D_{\epsilon }=D_{\epsilon }^0+h$. The nonlinear fixed-point map sends $h$ to the linear inverse applied to

$$\left(-A\wedge D_{\epsilon }^0,-C\wedge D_{\epsilon }^0,\frac{A^2-(D_{\epsilon }^0{)}^2-h\wedge h}{2}\right).$$

The approximate triple was arranged to satisfy the three integral moment equations exactly. For every exact iterate, ${\int }_{X}h\wedge h=0$, so the quadratic term preserves those equations. No new obstruction appears during iteration.

If $P_{\epsilon }$ bounds the inverse, $Q_{\epsilon }$ bounds multiplication, and ${\delta }_{\epsilon }$ bounds the gluing residual, use a ball of radius $2P_{\epsilon }{\delta }_{\epsilon }$. Its contraction constant is at most $2P_{\epsilon }^2{Q}_{\epsilon }{\delta }_{\epsilon }$. The first two factors grow polynomially, while ${\delta }_{\epsilon }\le C{\epsilon }^{-s}{e}^{-c/\epsilon }$. Their product tends to zero, proving contraction and an exponentially small correction at one fixed sufficiently high Sobolev order.

Smoothness follows from a square elliptic system

After choosing the small parameter, write $h=da$ in Coulomb gauge $d^{\ast }a=0$. The three wedge equations plus this gauge form four first-order equations for the four components of $a$. At a nonzero covector $\xi$, a kernel vector $b$ would make $\xi \wedge b$ orthogonal to the positive three-plane spanned by the corrected triple. That complement is negative definite, whereas the decomposable form $\xi \wedge b$ has square zero. It must vanish. Then $b$ is proportional to $\xi$, and the gauge equation forces $b=0$.

The square principal symbol is therefore invertible. Elliptic bootstrapping gives smoothness for the fixed parameter, without demanding smallness simultaneously in infinitely many derivative orders. The later uniform chamber estimate keeps the original third class available through the path, while the two integrable volume equations provide the parallel endpoint. The auxiliary class solves the analytic problem and preserves the class-selection condition; it does not replace the prescribed endpoint classes.

See Also

Taming implies compatibility develops the positive-current argument shared with this paper. Its compatible form has unrestricted class; the curve criterion, auxiliary form, and uniform positivity transfer above supply the additional class control required for deformation.

🏷️ Manuscripts and verification scope

• Deforming hypersymplectic four-manifolds to hyperkähler triples — pinned manuscript and companion source.

No family scope page was present at lean/docs/341.md in the pinned snapshot. This does not establish that all related formalization code is absent. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026. Deforming hypersymplectic four-manifolds to hyperkähler triples.