Taming and compatibility on four-manifolds

Source and verification

This article explains family 342 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: taming and compatibility

An almost complex structure on a real four-manifold is a smooth bundle map $J:TX\to TX$ with $J^2=-1$. It supplies complex directions pointwise, but need not arise from holomorphic coordinate charts.

A real two-form $\omega$ is symplectic if it is closed and nondegenerate. It tames $J$ when

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

It is compatible when it also satisfies

$$\omega (Ju,Jv)=\omega (u,v).$$

In that case $g(u,v)=\omega (u,Jv)$ is a positive definite symmetric metric. A compatible symplectic form gives an almost Kähler structure; integrability of $J$ is an additional condition.

The elementary projection onto invariant forms is

$$\begin{array}{rl}{\omega }^{\mathrm{inv}}(u,v)& =\frac{1}{2}\omega (u,v)\\ & +\frac{1}{2}\omega (Ju,Jv).\end{array}$$

It preserves positivity on complex lines because ${\omega }^{\mathrm{inv}}(v,Jv)=\omega (v,Jv)$. But differentiating it involves derivatives of $J$, so closedness of $\omega$ does not generally imply closedness of ${\omega }^{\mathrm{inv}}$.

Manuscript result

If a smooth almost complex structure on a closed connected smooth real four-manifold is tamed by a symplectic form, it admits a smooth compatible symplectic form. The almost complex structure is kept fixed. (OpenAI, 2026)

The theorem does not prescribe the cohomology class of the new form. It is an existence statement, distinct from a prescribed-volume equation or a general same-class conclusion.

The prerequisites are differential forms, functional analysis, elliptic regularization, and measure theory. The proof uses a positive current as a hypothetical obstruction, then shows that it contradicts the given tamer.

🏷️ Currents as averaged complex lines

A two-current is a continuous linear functional on smooth two-forms. It is closed if it vanishes on exact forms. Integration over an oriented closed surface is an example, but a general current need not be concentrated on a surface.

Choose a $J$-Hermitian metric $g$ and let

$$F(u,v)=g(Ju,v)$$

be its fundamental form. It need not be closed. A positively oriented unit complex-line bivector has the form $\ell =v\wedge Jv$, with $|v|=1$, and satisfies $F(\ell )=1$.

A positive current can be written

$$P(\alpha )={\int }_{\mathcal{C}}{\alpha }_x(\ell )d\lambda (x,\ell ),$$

where $\mathcal{C}$ is the bundle of these lines and $\lambda$ is a positive measure. Normalize its mass to one. Its projection $\mu$ to $X$ is the trace measure, characterized by $P(bF)=\int bd\mu$.

Suppose there is no compatible form. The open convex cone of positive invariant forms is then disjoint from the subspace of closed invariant forms. Hahn–Banach separation gives a nonzero positive $P$ vanishing on that subspace.

This does not yet contradict a tamer: the tamer need not be invariant. The proof must enlarge the annihilation property without losing control of positivity.

🏷️ An elliptic lift and an anti-invariant correction

Let $R$ be the orthogonal projection onto anti-invariant two-forms:

$$\begin{array}{rl}(R\alpha )(u,v)& =\frac{1}{2}\alpha (u,v)\\ & -\frac{1}{2}\alpha (Ju,Jv).\end{array}$$

In real dimension four these forms are self-dual for any $J$-Hermitian metric. This dimension-specific fact is central to the argument.

On anti-invariant forms, the operator

$$\mathcal{L}=Rd{d}^{\ast }$$

has principal symbol $|\xi {|}^2/2$ times the identity. It is elliptic, nonnegative, and self-adjoint. Let $G$ be its inverse off its kernel and let $H$ be projection onto the kernel. Then

$$B=d{d}^{\ast }G+H$$

is an order-zero operator satisfying $dB=0$ and $RB=\mathrm{Id}$. Thus every anti-invariant component has a controlled closed lift.

Define

$$\begin{array}{rl}T(\alpha )& =P(\alpha -BR\alpha ),\\ Q(\alpha )& =-P(BR\alpha ).\end{array}$$

Then $T=P+Q$ annihilates every closed two-form: if $\alpha$ is closed, $\alpha -BR\alpha$ is closed and invariant. The correction $Q$ is anti-invariant and hence self-dual, but initially only a distribution.

For currents identified with distributional two-forms using the $L^2$ pairing, current closedness means $d^{\ast }T=0$, equivalently $d(\ast T)=0$. This convention matters in the later Hodge-star pairing.

🏷️ Quadratic mass growth and angular energy

Corrected radial test functions give

$$\mu (B_r(x))\le C{r}^2$$

uniformly for small $r$. The relevant concentration scale is two-dimensional, even though the ambient manifold has dimension four.

Regularize with a fixed power of the Hodge resolvent:

$$\begin{array}{c}{S}_r=(1+r^2\Delta {)}^{-3},\\ \Delta =d{d}^{\ast }+d^{\ast }d.\end{array}$$

For fixed $r>0$, this puts the currents in $L^2$. It commutes with the differential and, in middle degree, with the Hodge star. Since $T$ annihilates closed forms, every closed current $T^{\prime }$ with the required regularity satisfies

$$⟨S_{r}T,\ast S_r{T}^{\prime }{⟩}_{L^2}=0.$$

The kernel of $S_r^2$ has a nonnegative scalar leading term times parallel transport, plus a smaller remainder. Near the diagonal the leading scale is $r^{-4}$ and the error scale is $r^{-2}$.

Let ${\tau }_{yx}$ transport bivectors from $y$ to $x$. For unit complex lines $\ell ,m$ at these points, the exact identity is

$$\begin{array}{rl}⟨\ell ,\ast {\tau }_{yx}m⟩& =\frac{1}{2}|\ell -{\tau }_{yx}m{|}^2\\ & -\frac{1}{4}|F_x-{\tau }_{yx}{F}_y{|}^2.\end{array}$$

The last term is order $d(x,y{)}^2$. It reflects variation of the almost Hermitian geometry and cannot be discarded as nonnegative.

Combining kernel bounds, mass growth, and a relative projection estimate yields

$$c{r}^{-4}{A}_r+\Vert S_{r}Q{\Vert }_2^2\le C+C\Vert S_{r}Q{\Vert }_2,$$

where

$$\begin{array}{rl}{A}_r& ={\iint }_{d(x,y)<r}|\ell -{\tau }_{yx}m{|}^2\\ & d\lambda (x,\ell )d\lambda (y,m).\end{array}$$

Consequently $Q\in L^2$ and $A_r=O(r^4)$. Smooth approximation within the anti-invariant subspace also makes the mixed regularized pairings of $Q$ with restrictions of $P$ tend to zero.

This first estimate does not force $Q=0$. The negative geometric error is only uniformly bounded at this stage. A second argument is needed to make it vanish.

🏷️ Positive density and closed restrictions

The source proves that the two-dimensional density exists at every point:

$$\theta (x)=\underset{r\downarrow 0}{\lim }{r}^{-2}\mu (B_r(x)).$$

Let $E=\{\theta >0\}$. Its density-splitting theorem says that

$$P_E=1_{E}P$$

is a closed current, under the mass and angular bounds above, assuming only that $P+Q$ is closed and $Q$ is anti-invariant and in $L^2$.

The theorem is not an automatic restriction property of currents. Multiplying a closed current by a function generally creates a boundary, and $P$ itself is not initially known to be closed.

To prove it, smooth the density at scale $r$:

$$\begin{array}{rl}& f_r(x)\\ & =r^{-2}{\int }_{X}h(d(x,y{)}^2/r^2)d\mu (y),\end{array}$$

using a smooth compactly supported radial kernel. Smooth functions of $f_r$ approximate restrictions according to density.

At almost every positive-density center, tangent measures are planar. This gives qualitative smallness of transverse displacements, without a quantitative convergence rate. For nearby pairs, the squared angular moment is $O(r^4)$, while the squared normalized transverse moment is $o(r^2)$.

Closedness of $P+Q$ trades tangential derivatives of the cutoff for a product of these two errors. The cutoff derivative and normalization cost $r^{-3}$. Cauchy–Schwarz gives

$$r^{-3}O(r^2)o(r)=o(1).$$

Keeping the angular and transverse factors together is essential: either estimate alone is insufficient.

The $Q$ contribution is controlled separately. The density is zero at almost every point for four-dimensional volume, and $r^{-2}\mu (B_r(x))$ is uniformly bounded. Dominated convergence against $|Q{|}^2$, together with Fubini for the remaining factor, makes the correction’s cutoff boundary tend to zero. Passing through smooth density weights proves closedness of $P_E$.

🏷️ Why the final pairing uses the residual current

Set

$$\begin{array}{c}{P}^{\prime }=1_{X\setminus E}P,\\ T^{\prime }=P^{\prime }+Q=T-P_E.\end{array}$$

The splitting theorem makes $T^{\prime }$ closed. Pair $T$ with this residual current rather than with itself.

At every zero-density center $y$,

$$\begin{array}{c}{r}^{-2}{\int }_X\frac{d\mu (x)}{(1+d(x,y)/r{)}^6}⟶0.\end{array}$$

On each fixed dyadic shell this follows from vanishing density. Quadratic mass growth bounds the shell tails by a summable geometric sequence. Dominated convergence against the trace measure of $P^{\prime }$ therefore eliminates the negative error in the positive-positive pairing.

Expanding the zero pairing gives

$$\begin{array}{rl}0=& ⟨S_{r}P,\ast S_r{P}^{\prime }⟩\\ & +⟨S_{r}P,\ast S_{r}Q⟩\\ & +⟨S_{r}Q,\ast S_r{P}^{\prime }⟩+\Vert S_{r}Q{\Vert }_2^2.\end{array}$$

The first term has nonnegative lower limit. Both mixed terms tend to zero, and $S_{r}Q\to Q$ strongly in $L^2$. Hence $0\ge \Vert Q{\Vert }_2^2$ and $Q=0$.

Now $P=T$ annihilates every closed form, including the given tamer $\omega$. But $\omega (\ell )$ has a positive minimum on the compact bundle of unit complex lines. Since $\lambda$ has mass one,

$$P(\omega )>0,$$

contradicting annihilation. The separating obstruction cannot exist, so the cone contains a closed invariant positive form.

The decisive improvement is the passage from a bounded error for the full current to a vanishing error on zero-density centers. The positive-density part first has to be removed as a closed current.

🏷️ Cohomology classes and the dimension restriction

A de Rham class records a closed form up to addition of an exact form. Let ${\mathcal{K}}_J^t$ and ${\mathcal{K}}_J^c$ be the classes containing taming and compatible symplectic forms. Let $H_J^{-}$ consist of classes represented by closed anti-invariant forms.

Once the compatible cone is known to be nonempty, the source invokes the four-dimensional cone comparison

$${\mathcal{K}}_J^t={\mathcal{K}}_J^c+H_J^{-}.$$

The right side consists of sums of classes from the two sets. It does not say that every individual taming class is compatible.

If the intersection form has positive index $b_2^{+}=1$, the same comparison gives equality of the two cones. In that case each taming class does contain a compatible representative. The extra hypothesis is needed for this stated same-class corollary.

The proof is specifically four-dimensional: anti-invariant two-forms are self-dual, the current has two-dimensional mass growth, and the kernel and cutoff estimates use ambient dimension four. No higher-dimensional extension or integrability conclusion is asserted.

🏷️ Key steps and proof outlines

The closed lift repairs invariance without preserving positivity

Separation gives a positive invariant current $P$ annihilating closed invariant forms. The elliptic operator $Rd{d}^{\ast }$ on anti-invariant forms has scalar positive principal symbol, so its Green operator and kernel projection produce a closed lift $B$ satisfying $RB=I$. Hence $T(\alpha )=P(\alpha -BR\alpha )$ annihilates every closed form.

The price is the correction $Q=T-P$, which is anti-invariant and self-dual but initially distributional. Positivity belongs to $P$, not to $T$. The proof therefore cannot immediately test $T$ against the tamer and call the result positive. Its task is to show that this correction vanishes. (OpenAI, 2026)

A commutator estimate controls regularization across the two subbundles

The Hodge resolvent $S_r=(1+r^2\Delta {)}^{-3}$ does not commute exactly with a varying pointwise projection $\Pi$. Suppose $\Pi W=0$, and write $U=S_{r}W$ and $A_r=(1+r^2\Delta {)}^3$. Then

$$\Pi U=S_r[A_r,\Pi ]U.$$

Because the principal symbol of ${\Delta }^j$ is scalar, the commutator $[{\Delta }^j,\Pi ]$ has order at most $2j-1$. Resolvent smoothing gives $\Vert S_r{\Vert }_{L^2\to H^{2j-1}}\le C{r}^{-(2j-1)}$. Each commutator term carries a factor $r^{2j}$, leaving one positive power of $r$. Taking the bounded adjoint yields

$$\Vert \Pi S_{r}W{\Vert }_2\le Cr\Vert S_{r}W{\Vert }_2.$$

This is a relative estimate: it remains useful even when $W$ is a measure and its regularized norm diverges. Quadratic mass growth gives $\Vert S_r{P}^{\prime }{\Vert }_2\le C/r$ for restrictions $P^{\prime }\le P$, exactly canceling the extra $r$ in mixed pairings. (OpenAI, 2026)

First obtain finite energy, then make the mixed terms vanish

Pairing the regularized annihilator with itself gives the positive $\Vert S_{r}Q{\Vert }_2^2$ term, an angular term, and mixed terms controlled linearly by $\Vert S_{r}Q{\Vert }_2$. Thus the inequality $X^2\le C+CX$ bounds $X=\Vert S_{r}Q{\Vert }_2$ uniformly. Weak compactness and distributional convergence identify an $L^2$ limit with $Q$.

For smooth anti-invariant $V$, one has $P^{\prime }(S_r^{2}V)\to P^{\prime }(V)=0$. Smooth anti-invariant forms are dense in their $L^2$ subspace: smooth an approximation and project it by $R$. The uniform relative estimate allows this approximation error to be controlled independently of $r$. It follows that the mixed pairings with $Q$ tend to zero for every restriction $P^{\prime }\le P$.

Density splitting is needed to remove the remaining geometric error

The positive-positive kernel pairing has a nonnegative angular leading term but a geometric error of size $d(x,y{)}^2$. Quadratic mass growth only bounds that error; it does not make it vanish. The density-splitting theorem first proves that the positive-density restriction is closed, using coupled angular and transverse errors in the cutoff boundary.

Subtract that closed part. On the residual trace measure every center has zero two-dimensional density. Fixed dyadic shells then tend to zero, and the quadratic growth bound supplies a summable majorant for the shell tails. The geometric error now vanishes. The final zero pairing has nonnegative lower limit except for terms already tending to zero, so it forces $\Vert Q{\Vert }_2^2=0$. Only at this point does positivity of $P=T$ contradict annihilation of the original closed tamer.

See Also

Hypersymplectic deformation uses the same current estimates but adds a prescribed-class criterion and a geometric construction. Those additional ingredients are what allow all three original form classes to remain fixed along its path.

🏷️ Manuscripts and verification scope

• Taming implies compatibility on four-manifolds — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. The documented selection keeps the given almost complex structure fixed and imposes no cohomology class on the compatible form. It does not explicitly list the additional cone identity or the same-class corollary under the positive-index-one hypothesis. 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. Taming implies compatibility on four-manifolds.