Einstein four-manifolds: curvature rigidity and an L² topological gap

Source and verification

This article explains family 348 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.

🏷️ Curvature, Einstein metrics, and the defect energy

For a Riemannian manifold $(M,g)$, the sectional curvature $K(\sigma )$ measures curvature along a tangent two-plane $\sigma$. The full curvature tensor is denoted $\mathrm{Rm}$. The Ricci tensor sums sectional curvatures through a tangent direction; its trace is the scalar curvature $R$. In dimension four an Einstein metric satisfies

$$\mathrm{Ric}=\lambda g,R=4\lambda ,$$

where $\lambda$ is constant on a connected manifold. The Einstein equation fixes these sums without making all sectional curvatures equal. A closed manifold is compact without boundary.

Positive Einstein constant permits a normalization to $\mathrm{Ric}=3g$. The standard models in this normalization are the unit round four-sphere, the Fubini–Study complex projective plane with the appropriate scale, and the product of two round two-spheres of radius $1/\sqrt{3}$. The product has zero sectional curvature on planes spanned by one direction from each factor, despite its positive Ricci tensor.

For a metric that need not be Einstein, define its trace-free Ricci tensor and defect energy by

$$\begin{array}{rl}{E}_g& ={\mathrm{Ric}}_g-\frac{R_g}{4}g,\\ \mathcal{D}(g)& ={\int }_M|E_g{|}_g^{2}d{\mu }_g.\end{array}$$

On a connected four-manifold, $E_g=0$ forces constant scalar curvature by the contracted Bianchi identity and therefore makes $g$ Einstein. The energy is scale invariant: replacing $g$ by $cg$ multiplies the squared tensor norm by $c^{-2}$ and the volume form by $c^2$.

Prerequisites. The post uses differential forms, covariant derivatives, integration by parts, elementary spectral theory, and the basic heat-equation viewpoint on Ricci flow. The four-dimensional curvature decomposition, normalized integral estimates, and compactness argument are introduced below. Representation-theoretic tensor calculations are treated as source inputs rather than developed as an algebra course.

Three related conclusions

The positive-curvature manuscript reports a metric classification. The zero-plane companion identifies the boundary case and obtains a classification of universal covers under nonnegative sectional curvature. The energy-gap companion transfers that exact classification to the diffeomorphism type of metrics with sufficiently small trace-free Ricci energy. Its conclusion does not assert that the original metric is a standard metric.

🏷️ The three statements and their dependencies

Manuscript	Hypotheses	Reported conclusion
Positive sectional curvature (OpenAI, 2026)	Connected, closed, Einstein four-manifold with $K>0$	Up to scaling and isometry: round $S^4$, Fubini–Study ${\mathbb{CP}}^2$, or round ${\mathbb{RP}}^4$
Zero-plane rigidity (OpenAI, 2026b)	Connected, closed, $\mathrm{Ric}=3g$, $K\ge 0$, and one zero-curvature plane	Universal Riemannian cover is $S^2(1/\sqrt{3})\times S^2(1/\sqrt{3})$
Small Einstein defect (OpenAI, 2026a)	Connected, simply connected, closed, $K\ge 0$, and $\mathcal{D}(g)<{\epsilon }_0$	Diffeomorphic to standard $S^4$, ${\mathbb{CP}}^2$, or $S^2\times S^2$

The first two statements require no orientability assumption. In the first, the projective four-space is the nonorientable possibility; the oriented conclusion has only the sphere and complex projective plane. The second specifies a universal cover, not a classification of all possible quotients. For $\mathrm{Ric}=\lambda g$ with $\lambda >0$, its sphere factors have radius $1/\sqrt{\lambda }$.

The zero-plane paper also reports the full nonnegative-sectional classification: the universal cover of a closed Einstein four-manifold with $\mathrm{Ric}=3g$ and $K\ge 0$ is one of the sphere, complex projective plane, and product models. The gap paper states this exact classification as a premise and invokes it only after constructing an exactly Einstein limit. (OpenAI, 2026c) (OpenAI, 2026a)

Reported topological energy gap

Assuming the companion’s exact nonnegative-sectional Einstein classification, there is a universal ${\epsilon }_0>0$ such that the third row holds. No auxiliary volume, diameter, injectivity-radius, Sobolev, curvature-maximum, or curvature-derivative bounds are assumed. The paper proves existence of a positive threshold, without giving an explicit numerical value. (OpenAI, 2026a)

The zero-plane assertion concerns one plane at one point. It does not assume an everywhere sharp curvature inequality or a pre-existing product structure. Conversely, the positive Einstein constant is essential to its formulation: the theorem is not a classification of all Ricci-flat nonnegatively curved manifolds.

🏷️ The two Weyl blocks in dimension four

On an oriented four-manifold, the Hodge star satisfies ${\ast }^2=1$ on two-forms. Its $+1$ and $-1$ eigenspaces, denoted ${\Lambda }^{+}$ and ${\Lambda }^{-}$, each have rank three. The curvature tensor acts as a symmetric operator on two-forms. When $\mathrm{Ric}=3g$, its block form is

$$\mathcal{R}=\left(\begin{array}{cc}I+T& 0\\ 0& I+U\end{array}\right),$$

where $T,U$ are symmetric trace-free operators, the two Weyl blocks. Vanishing of one block is called half-conformal flatness. The Einstein equation removes the off-diagonal block, but does not force either diagonal Weyl block to vanish.

Using Hilbert–Schmidt operator norms, put

$$\begin{array}{rlrl}v& =|T|/\sqrt{6},& b& =|U|/\sqrt{6},\\ x& =-{\lambda }_{\min }(T),& y& =-{\lambda }_{\min }(U).\end{array}$$

A unit simple two-form representing a plane has self-dual and anti-self-dual components of equal squared norm $1/2$. Conversely the two unit directions can be varied independently to describe oriented planes. Minimizing their curvature contributions gives

$$K_{\min }=1-\frac{x+y}{2}.$$

Elementary trace-free three-by-three spectral inequalities give

$$v\le x\le 2v,b\le y\le 2b.$$

Thus nonnegative sectional curvature is exactly $x+y\le 2$, and implies $v+b\le 2$. Strict sectional positivity gives strict inequalities. The converse from the norm bound alone is unavailable: the proof must retain the least eigenvalues as well as the norms. (OpenAI, 2026c) (OpenAI, 2026b)

For the nonnegative case, pass to the universal Riemannian cover. It is complete, and Bonnet–Myers makes it compact because $\mathrm{Ric}=3g$. Simple connectivity supplies an orientation. Reversing it merely interchanges $T$ and $U$.

🏷️ Separate norm gaps and balanced moments

Write $\mathbf{E}f=\mathrm{Vol}(M{)}^{-1}{\int }_{M}fd\mu$. The papers use two scalar estimates:

$$\begin{array}{c}T\not\equiv 0⟹\mathbf{E}{v}^2\ge 1,\\ \mathrm{Var}(a)\le \mathbf{E}\frac{|\nabla a{|}^2}{4+v^2+b^2}.\end{array}$$

The corresponding gap holds for $U$. Equality in the nonzero-block gap makes that Weyl block parallel, with normalized norm one and eigenvalues $(2,-1,-1)$. (OpenAI, 2026c)

The norm gap combines the Weyl equation, a refined gradient estimate, and a sharp normalized Sobolev inequality. To display the mechanism, put

$$\begin{array}{c}e=|\nabla T{|}^2/6,\\ \sigma =\frac{\mathrm{tr}(T^3)}{6v^3}(v>0).\end{array}$$

The spectral shape parameter satisfies $|\sigma |\le 1$. The differential identities are

$$\begin{array}{rl}\Delta (v^2)& =12(v^2-\sigma v^3)+2e,\\ e& \ge \frac{5}{3}|\nabla v{|}^2.\end{array}$$

Test the regularized equation using $f_{\eta }=(v^2+{\eta }^2{)}^{1/6}$ and apply

$$(\mathbf{E}|f{|}^4{)}^{1/2}\le \mathbf{E}\left(f^2+\frac{1}{2}|\nabla f{|}^2\right).$$

Passing to $f=v^{1/3}$ gives

$$\begin{array}{rl}(\mathbf{E}{f}^4{)}^{1/2}& \le \mathbf{E}(f^{2}v)\\ & \le (\mathbf{E}{f}^4{)}^{1/2}(\mathbf{E}{v}^2{)}^{1/2}.\end{array}$$

For a nonzero block the common factor is positive, proving the gap. Equality in Cauchy–Schwarz and connectedness make $v$ constant; the norm equation then forces $\nabla T=0$. Regularization is needed because a nonzero block can still vanish at individual points. (OpenAI, 2026c)

The weighted variance bound is stronger than ordinary Poincaré control. A curvature contraction with the trace-free Hessian, combined with Bochner’s identity, gives

$$\mathbf{E}(\Delta \varphi {)}^2\ge \mathbf{E}(4+v^2+b^2)|\nabla \varphi {|}^2.$$

Solve $\Delta \varphi =a-\mathbf{E}a$, integrate by parts, and apply weighted Cauchy–Schwarz. This yields the displayed variance inequality. Its tensor estimate uses the full spectral restriction $x+y\le 2$.

The separate gaps do not yet settle the problem: nonconstant $v,b$ can have large second moments despite a pointwise bound on their sum. A global argument first balances the moments. Let

$$s_0=\mathbf{E}(v^2+b^2),V=\mathrm{Vol}(M).$$

If $n_{\pm }$ are the dimensions of the spaces of harmonic self-dual and anti-self-dual two-forms, the Euler characteristic and signature satisfy

$$\begin{array}{rl}\chi & =2+n_{+}+n_{-},\\ \tau & =n_{+}-n_{-},\\ 8{\pi }^2\chi & =6V(1+s_0),\\ 12{\pi }^2\tau & =6V\mathbf{E}(v^2-b^2).\end{array}$$

Here the topology formulas use simple connectivity. The paper derives an upper volume deficit

$$\frac{V}{8{\pi }^2/3}\le 1-\frac{3}{40}{s}_0.$$

When neither block vanishes identically, the two gaps and the pointwise constraint give $2\le s_0<4$. The endpoint four would force the pair $(v,b)$ to be constantly $(2,0)$ or $(0,2)$, contradicting the assumption. The volume and characteristic identities then imply $\chi <7$ and $|\tau |/\chi <4/15$. Since $\chi$ and $\tau$ have the same parity, the only possible nonzero-signature case is $(\chi ,|\tau |)=(5,1)$. (OpenAI, 2026c)

To exclude it, label $b$ so that $P=\mathbf{E}{b}^2$ is the smaller moment. The identities and $v\le 2-b$ imply

$$P\ge 1,28\mathbf{E}b+6P\le 25.$$

For

$$\begin{array}{rl}H(z)& =\sqrt{z}\left(1+\frac{z^2}{40}-\frac{z^4}{1152}\right),\\ \Phi (z)& =16H(z{)}^2+9z(1-z),\end{array}$$

the weighted variance estimate yields $16(\mathbf{E}H(b){)}^2-\mathbf{E}\Phi (b)\ge 0$. The paper proves a constrained-moment lemma: every probability measure on $[0,2]$ obeying those two moment constraints makes the same expression strictly negative. Its proof reduces a hypothetical maximizing measure to two-point support and then uses explicit polynomial signs. Applied to the distribution of $b$, it gives a contradiction. Consequently

$$\tau =0,\mathbf{E}{v}^2=\mathbf{E}{b}^2.$$

This balance is a conclusion of the volume, topology, and moment arguments, not a symmetry assumption. (OpenAI, 2026c)

🏷️ Radial volume comparison without commuting curvature matrices

Along a unit-speed geodesic, let $\mathcal{R}(t)$ be the three-by-three normal curvature matrix in a parallel frame. The normal Jacobi matrix satisfies

$$\begin{array}{c}{J}^{\prime \prime }+\mathcal{R}J=0,\\ J(0)=0,J^{\prime }(0)=I.\end{array}$$

Its determinant is the polar volume density. With $A=\mathcal{R}-I$, nonnegative sectional curvature gives $0\le \mathcal{R}\le 3I$. Geodesic-flow averaging gives

$$\mathcal{E}|A{|}^2=\frac{3}{2}{s}_0,\mathcal{E}|A^{\prime }{|}^2\le \frac{3}{2}\mathcal{E}|A{|}^2,$$

where $\mathcal{E}$ averages over the unit tangent bundle. These relate radial anisotropy to the global Weyl moments.

The comparison does not diagonalize curvature simultaneously along the geodesic. Divide $[0,t]$ into $N$ intervals of length $a=t/N$ and form the block tridiagonal Dirichlet matrix with diagonal blocks $2I-a^2\mathcal{R}(ia)$ and adjacent blocks $-I$. Its determinant satisfies

$$a^3\det {\mathcal{L}}_N⟶\det J(t).$$

Before conjugacy, these matrices are eventually positive definite. Comparing them to the round matrices and isolating the first sine mode gives a determinant deficit controlled by the sine-weighted average of $A$. Schur complements preserve the order of matrix products, and a variance estimate controls loss from averaging. Exact rational integral bounds supply the coefficient $3/40$ above. (OpenAI, 2026c)

For the global upper bound, integrate over all radial vectors before their first conjugate point and of length at most $\pi$. Bonnet–Myers gives the required distance bound. The exponential map may cover some points more than once, which only increases this integral. Nonconjugate cut points are allowed; no injectivity-radius lower bound is used in this upper comparison. This distinction permits the argument to reach the nonnegative-curvature boundary.

🏷️ A weighted tensor inequality and its equality case

Balanced moments allow the two Weyl equations to interact. The difficult tensor input is a weighted Hessian identity. The Bianchi constraint and curvature commutator decompose the full Hessian of a Weyl block into a prescribed trace component and a remaining component. The Hessian is not assumed symmetric in its derivative slots. A curvature-dependent operator $\mathcal{P}$ on the remaining component satisfies

$$\mathcal{P}\ge (4-2x-2y)I\ge 0.$$

Completing a weighted square and integrating by parts gives a first-derivative inequality. Both derivatives of the weight and derivatives of $\mathcal{P}$ must be retained. The identity is valid for positive Lipschitz weights, including

$$\begin{array}{rl}{m}_{+}& ={\left(1-\frac{3}{8}(v-b)\right)}^2,\\ m_{-}& ={\left(1+\frac{3}{8}(v-b)\right)}^2.\end{array}$$

Each weight is at least $1/16$ on $v+b\le 2$. The square requires nonnegativity of $\mathcal{P}$; a kernel at $x+y=2$ causes no failure. (OpenAI, 2026c)

The paper adds three expressions to the two weighted inequalities: a nonnegative integral from the Laplacian of $2-v-b$, the balanced variance defect, and zero-integral tests of the norm equations with polynomial multipliers. The resulting integrand $\mathcal{I}$ has $\mathbf{E}\mathcal{I}\ge 0$. Its quartic, quadratic, and zeroth-order derivative terms obey

$$\begin{array}{c}\mathcal{I}\le -cj+2\Theta r-\frac{{\Theta }^2}{c},\\ c=\frac{2}{5}-\frac{3}{10}vb\ge \frac{1}{10}.\end{array}$$

Here $\Theta$ is an explicitly chosen polynomial comparison function in $v,b$, and

$$\begin{array}{rl}p& =\nabla v,q=\nabla b,\\ R_v& =\frac{4(1+v)}{1+6v+v^2},\\ r& =R_v|p{|}^2+R_b|q{|}^2,\\ j& =(|p{|}^2+|q{|}^2)\left(\frac{|p{|}^2}{v}+\frac{|q{|}^2}{b}\right),\end{array}$$

with $R_b$ obtained by replacing $v$ by $b$. These are specific estimates derived from the tensor calculation and finite polynomial certificates, not a generic inequality for arbitrary curvature data. Their consequence has a short proof. The identity

$$\begin{array}{c}1+6v+v^2-4\sqrt{v}(1+v)\\ =(\sqrt{v}-1{)}^4\end{array}$$

gives $R_v\le 1/\sqrt{v}$ for $v>0$, with equality only at $v=1$. Cauchy–Schwarz gives $r^2\le j$, so

$$\mathcal{I}\le -\frac{(\Theta -cr{)}^2}{c}-c(j-r^2)\le 0.$$

A Lipschitz function has zero gradient almost everywhere on any one of its level sets. Thus, almost everywhere where $p\ne 0$, one has both $v>0$ and $v\ne 1$; there the corresponding comparison is strict. The same holds for $q$. The upper bound for $\mathcal{I}$ is therefore strictly negative wherever either norm gradient is nonzero, outside a null set. Since its integral is nonnegative, both gradients vanish almost everywhere. Connectedness makes $v,b$ constant.

If both blocks are nonzero, their gaps imply $v,b\ge 1$. Under strict positive sectional curvature this contradicts $v+b<2$, proving that one block vanishes. Under nonnegative sectional curvature it instead forces

$$v=b=1.$$

The equality argument therefore detects precisely the product boundary case. Its strictness comes from the level-set argument, not from strict positivity of the Hessian operator. (OpenAI, 2026c) (OpenAI, 2026b)

Two zero-set issues are part of the proof. First, regularizing $v$ by $\sqrt{v^2+{\eta }^2}$ proves that $e/v$ is integrable and that the weak norm equation has an $L^1$ density, with no residual singular measure. Terms whose numerator and denominator vanish on the zero set are assigned zero almost everywhere. Second, the pointwise polynomial comparison at $v=0$ uses a small nonzero auxiliary Weyl block while slightly shrinking the opposite block, keeping its derivative tensor fixed. This preserves the spectral cone and passes the estimates to the boundary. These are pointwise tensor data, not a deformation through Einstein metrics. (OpenAI, 2026c)

🏷️ From equality to the model geometries

In the two-block equality case the norm equations force both blocks to be parallel with eigenvalues $(2,-1,-1)$. Their simple eigenlines yield parallel unit self-dual and anti-self-dual two-forms on the simply connected cover. Viewed as skew endomorphisms, the appropriately normalized forms give commuting orthogonal complex structures $I,J$. Their product satisfies

$$\begin{array}{c}(IJ{)}^{\ast }=IJ,(IJ{)}^2=I_{TM},\\ \mathrm{tr}(IJ)=0.\end{array}$$

Its positive and negative eigenspaces are orthogonal parallel distributions of rank two. The de Rham theorem turns this into a global product of complete simply connected surfaces. Each factor has Ricci curvature $3g$, hence Gaussian curvature three, and is the round sphere of radius $1/\sqrt{3}$. If a zero plane exists, the half-conformally-flat sphere and complex-projective alternatives, which have strictly positive sectional curvature, are excluded. This proves the reported zero-plane conclusion. (OpenAI, 2026c)

The strictly positive manuscript also supplies its own last step after one block vanishes, in addition to citing classical half-conformally-flat classification. If exactly one block survives, characteristic identities leave two possible harmonic counts. The exceptional one forces

$$s_0=3,V=\frac{4{\pi }^2}{3}.$$

A conditional lower polar-volume estimate under $s_0=3$ gives instead $V>4{\pi }^2/3$. Unlike the earlier upper bound, this lower bound uses an injectivity domain of radius $\pi /\sqrt{3}$ supplied by the strict-positive topology argument. The remaining harmonic count has Weyl moment one, whose equality case makes the curvature parallel. Matching its tensor with the sphere or Fubini–Study tensor then extends to a global isometry by simple connectivity. (OpenAI, 2026b)

For a nonorientable positively curved Einstein manifold, take its orientation double cover. The complex projective plane has no orientation-reversing diffeomorphism: a diffeomorphism preserves the square of a generator of its second integral cohomology and hence its top-degree orientation class. On the sphere, a free isometric involution must be antipodal, because any positive eigendirection of the corresponding orthogonal involution would give a fixed point. The quotient is round ${\mathbb{RP}}^4$. The nonnegative theorem only identifies the universal cover and does not make this same quotient assertion. (OpenAI, 2026b)

🏷️ Small energy supplies the initial geometry

The gap proof argues by contradiction. Suppose a sequence of simply connected closed four-manifolds has $K\ge 0$, $\Vert E{\Vert }_2\to 0$, and lies outside the three proposed diffeomorphism types. Norms here use the actual volume measure, not normalized expectation.

Nonnegative sectional curvature implies a dimensional estimate $|\mathrm{Rm}|\le C_{0}R$. Simple connectivity gives Euler characteristic at least two, so Chern–Gauss–Bonnet excludes $R\equiv 0$. Normalize each metric to have average scalar curvature twelve. An almost-Schur estimate then gives

$$\begin{array}{c}\Vert R-12{\Vert }_2\le 4\Vert E{\Vert }_2,\\ \Vert \mathrm{Ric}-3g{\Vert }_2⟶0.\end{array}$$

For the estimate, solve $\Delta f=R-12$. The identity $\mathrm{div}E=\frac{1}{4}dR$ gives

$$\int (R-12{)}^2=4\int ⟨E,{\nabla }^{2}f⟩.$$

Integrated Bochner under $\mathrm{Ric}\ge 0$ bounds $\Vert {\nabla }^{2}f{\Vert }_2$ by $\Vert \Delta f{\Vert }_2$; Cauchy–Schwarz finishes the argument. Orthogonality of the trace and trace-free parts controls the full Ricci error. (OpenAI, 2026a)

The topology bound, $|\mathrm{Rm}|\le C_{0}R$, and

$$\int R^2=144V+\Vert R-12{\Vert }_2^2$$

give a positive volume lower bound. A segment-averaging argument bounds diameter without assuming a Ricci lower bound close to three pointwise. Take balls near the ends of a diameter and average the integrated Ricci error over minimizing geodesics between them. One such geodesic of length $\ell$ comparable to the diameter has small integrated error. The index form tested with sine-weighted parallel normal fields gives

$$\frac{3{\pi }^2}{2\ell }\ge \frac{3\ell }{2}-\text{integrated error},$$

forcing a uniform upper bound for $\ell$ and hence for diameter. Bishop comparison then bounds volume above. These bounds yield a uniform initial Sobolev inequality

$$\Vert f{\Vert }_4^2\le S\int (|\nabla f{|}^2+f^2).$$

The paper derives this through a distance-kernel estimate, control for functions of small support, and dyadic truncation. No initial curvature maximum is obtained or assumed. (OpenAI, 2026a)

🏷️ Ricci-flow smoothing at the critical exponent

Run unnormalized Ricci flow ${\partial }_{t}g=-2\mathrm{Ric}$. Initially each metric has its own existence interval. The proof first obtains estimates only while each individual flow exists, and then derives a common interval.

Perelman’s entropy propagates a scalar-weighted Sobolev inequality. To remove the scalar weight, the paper controls just the portion of curvature above a moving bounded threshold. Put

$$\begin{array}{rl}u& =|\mathrm{Rm}|,\\ w& =(u-K(t){)}_{+},\\ K(0)& =12C_0,K^{\prime }=C_q{K}^2,\end{array}$$

where $C_q$ is a dimensional constant in $({\partial }_t-\Delta )u\le C_q{u}^2$. Initially $w\le C_0(R-12{)}_{+}$, so $\Vert w(0){\Vert }_2\to 0$. The threshold equation cancels the bounded-background reaction and gives

$$\begin{array}{rl}\frac{d}{dt}\int w^2& \le -2\int |\nabla w{|}^2\\ & +C\int w^3+C\int w^2.\end{array}$$

On an interval where $\Vert w{\Vert }_2\le \eta$, the scalar-weighted Sobolev estimate becomes unweighted by absorbing $\int w{f}^2\le \eta \Vert f{\Vert }_4^2$. Next,

$$\begin{array}{rl}\int w^3& \le \eta \Vert w{\Vert }_4^2\\ & \le C\eta \int (|\nabla w{|}^2+w^2).\end{array}$$

Choose $\eta$ small, absorb the gradient term, and use Gronwall. Since the initial tail tends to zero, a first crossing of the threshold $\eta$ is impossible. The Sobolev constant remains uniform, the total curvature energy stays bounded, and ${\sup }_t\Vert w(t){\Vert }_2\to 0$ throughout the existing short intervals. (OpenAI, 2026a)

Distance cutoffs in the Sobolev inequality yield a volume bound $\mathrm{Vol}B(x,r)\ge c{r}^4$ for $r\le 1$. Iteration starts at arbitrarily small scales on each smooth metric, so no common initial injectivity scale is needed.

Suppose $t\max u(t)$ first reaches a fixed large constant $A$, and set $Q=A/t$. Curvature on a preceding interval of length $Q^{-1}$ is at most $2Q$. Shi’s derivative estimate then gives $|\nabla \mathrm{Rm}|\le C{Q}^{3/2}$ at the final time. Around a curvature maximum, a ball of radius $c{Q}^{-1/2}$ has $w\ge Q/2$. The volume lower bound forces

$$\int w^2\ge c{Q}^2(Q^{-1/2}{)}^4=c>0,$$

contradicting the vanishing tail. Thus $|\mathrm{Rm}(t)|\le A/t$. The curvature continuation criterion supplies a common lifespan. The cancellation of the powers of $Q$ explains the role of the scale-invariant exponent $L^2$ in dimension four. (OpenAI, 2026a)

🏷️ Identifying the Einstein limit and recovering curvature sign

Set $\lambda (t)=3/(1-6t)$, the Einstein constant along a normalized shrinking Einstein flow, and let $L=\mathrm{Ric}-\lambda (t)g$. In moving orthonormal frames its evolution is linear in $L$:

$${\partial }_{t}L=\Delta L+2\mathrm{Rm}(L)+2\lambda L.$$

The same small curvature tail absorbs the potentially large curvature coefficient in the $L^2$ energy estimate. Thus ${\sup }_t\Vert L(t){\Vert }_2\to 0$ on the common short interval.

Nonnegative sectional curvature is not assumed to persist under these flows. Let $F(\mathrm{Rm})$ be the magnitude of the most negative sectional curvature, or zero if all planes have nonnegative curvature. As a function on curvature tensors, $F$ is convex, Lipschitz, and invariant under orthonormal changes of frame. Smooth convex approximation and the curvature evolution give

$$({\partial }_t-\Delta )F\le C|\mathrm{Rm}{|}^2$$

in the corresponding weak sense. Since $R\ge 0$, the evolving volume contributes a favorable sign. The uniform total curvature energy and $F(0)=0$ imply

$${\int }_{M}F(\mathrm{Rm}(t))d{\mu }_{g(t)}\le Ct.$$

At any fixed positive time, the curvature bound, local volume lower bound, and total volume upper bound give bounded diameter and positive injectivity radius. Derivative estimates and Hamilton compactness produce smooth convergence on positive-time intervals. The diameter bound makes the limit manifold $N$ compact; convergence embeddings of $N$ are eventually diffeomorphisms onto the connected manifolds in the sequence.

The vanishing Einstein defect identifies the limiting flow as

$$h(t)=(1-6t)h,{\mathrm{Ric}}_h=3h.$$

Under constant metric scaling by $c$, the negative sectional-curvature magnitude scales by $c^{-1}$ and four-dimensional volume by $c^2$. Passing the integral defect estimate to this explicit limit gives

$$(1-6t)\int F({\mathrm{Rm}}_h)d{\mu }_h\le Ct.$$

Letting $t\downarrow 0$ forces $F({\mathrm{Rm}}_h)=0$, so $h$ has nonnegative sectional curvature. This uses neither smooth convergence of the original metrics at time zero nor curvature-cone preservation under Ricci flow. The exact classification now applies to $(N,h)$, and eventual diffeomorphism contradicts the initial choice of manifolds. (OpenAI, 2026a)

🏷️ Finite certificates and verification boundaries

The classification arguments reduce specific inequalities to rational polynomial identities and positivity of Bernstein coefficients on intervals and triangles. The Bernstein basis is a nonnegative partition of unity; nonnegative scalar coefficients imply polynomial nonnegativity, and positive semidefinite matrix coefficients imply the corresponding matrix inequality. These are exact finite certificates for the stated polynomials and domains.

For this post, the pinned arithmetic package was downloaded, its core file hashes checked against its manifest, and python -B verification/verify.py --check executed successfully. It recomputed and matched the bundled results: 53 identities, 21 interval rows, 12 triangle rows, and 24 exact-value assertions, with all_passed: true. Fifteen interval rows are reported main rows and six are supplementary. The positive-curvature and zero-plane packages have byte-identical checker, polynomial engine, and saved results, so one run checks their shared arithmetic.

This verifies the finite calculations, including rational volume margins. It does not verify the differential identities, geometric reductions, moment optimization, compactness theorem, or equivalence with a Lean statement. The pinned checker coverage states that distinction explicitly.

🏷️ Related geometric mechanisms

• The metric Blaschke theorem also combines matrix Jacobi-field comparison with a separate global rigidity argument; neither calculation permits arbitrary simultaneous diagonalization along geodesics.
• Taming and compatibility on four-manifolds provides another four-dimensional use of the Hodge decomposition and positivity of two-forms.
• The collection guide records the reading order and graduate prerequisites, while the source-scope discussion below separates the three manuscripts from the narrower documented formalization.

🏷️ Key steps and proof outlines

The improved variance estimate comes from a curvature contraction

For a smooth function $\varphi$, let $S={\mathrm{Hess}}_0\varphi$. Under $\mathrm{Ric}=3g$, the integrated Bochner identity reads

$$\mathbf{E}(\Delta \varphi {)}^2=\mathbf{E}\left(4|\nabla \varphi {|}^2+\frac{4}{3}|S{|}^2\right).$$

The two Weyl blocks induce an operator on trace-free symmetric two-tensors. Its eigenvalues are pairwise sums of Weyl eigenvalues, so the sectional-curvature restriction bounds it above by $4I$. On the other hand, integration by parts, the divergence-free Weyl identity, and the curvature commutator give the exact contraction $3\mathbf{E}[(v^2+b^2)|\nabla \varphi {|}^2]$. Comparing the upper operator bound with this identity inserts the extra weight $v^2+b^2$ into Bochner’s estimate.

Solve $\Delta \varphi =a-\mathbf{E}a$. Weighted Cauchy–Schwarz then gives

$$\mathrm{Var}(a)\le {\left(\mathbf{E}\frac{|\nabla a{|}^2}{4+v^2+b^2}\right)}^{1/2}\mathrm{Var}(a{)}^{1/2}.$$

Division, with the zero-variance case treated separately, proves the stronger variance bound. Thus the weight is a consequence of the two curvature blocks acting together, not a freely chosen improvement of Poincaré’s inequality. (OpenAI, 2026c)

Integral balance precedes the pointwise coupling argument

Separate nonzero-block gaps give lower bounds on two second moments. The volume deficit and characteristic identities then restrict the possible Euler characteristic and signature. The constrained-moment lemma removes the last exceptional signature case, producing equal Weyl moments. Only after this balance is available can the coupled weighted inequalities be added with the required signs.

The final square completion retains a strict remainder wherever a norm gradient is nonzero. The level-set fact that a Lipschitz function has zero gradient almost everywhere on a fixed level handles the apparent equality values $v=1$ or $b=1$. At zeros, regularized norm equations eliminate a possible singular measure before the division terms are used. This is why the nonnegative-curvature boundary yields parallel blocks with $v=b=1$, rather than merely a limiting statement from strict positive curvature. (OpenAI, 2026b)

Critical curvature concentration contradicts the small tail

In the $L^2$ gap companion, the moving cutoff $w=(|\mathrm{Rm}|-K(t){)}_{+}$ separates the uniformly bounded background from the small high-curvature part. Entropy supplies a weighted Sobolev estimate; small $\Vert w{\Vert }_2$ absorbs its scalar-curvature weight. The same small factor absorbs the cubic reaction in the energy inequality for $w$.

At a first high-curvature event of size $Q$, derivative bounds keep curvature comparable to $Q$ on a ball of radius $c{Q}^{-1/2}$. Noncollapse gives that ball volume at least $c{Q}^{-2}$. Its contribution to $\int w^2$ is therefore bounded below independently of $Q$, contradicting the vanishing tail. This produces a common smoothing interval before compactness is invoked. The Einstein limit can then be classified by the exact rigidity theorem, and the later stability argument returns the geometric conclusion to the original metrics. (OpenAI, 2026a)

🏷️ Manuscripts and verification scope

• Zero-Plane Rigidity for Einstein Four-Manifolds — pinned manuscript and companion source.
• An L² Einstein Gap for Nonnegatively Curved Four-Manifolds — pinned manuscript and companion source.
• Positively curved Einstein four-manifolds — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. Despite its heading mentioning nonnegative curvature and an L² gap, the scope page selects only the strictly positive sectional-curvature classification from the September manuscript, including its nonorientable round projective-space case. It does not document coverage of zero-plane rigidity, the nonnegative-sectional universal-cover classification, or the L² topological-gap companion. 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 2026a. An L² Einstein Gap for Nonnegatively Curved Four-Manifolds.

🐻  OpenAI 2026b. Positively curved Einstein four-manifolds.

🐻  OpenAI 2026c. Zero-Plane Rigidity for Einstein Four-Manifolds.