Borel rigidity in dimension four: cyclic covers and realized symmetries

Source and verification

This article explains family 320 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.

The Borel rigidity question asks whether a closed aspherical manifold is determined, up to homeomorphism, by its homotopy type. This manuscript claims a negative answer in dimension four. Its proof first obstructs a specified self-homotopy equivalence from being represented by a homeomorphism, then adds a symmetry argument to distinguish two manifolds regardless of the homotopy class of a possible homeomorphism. Those are separate conclusions. (OpenAI, 2026)

🏷️ Preliminaries: homotopy type and manifold structure

A space is aspherical when its universal cover is contractible. An aspherical connected space with CW homotopy type is a $K(\pi ,1)$: its fundamental group determines its homotopy type. A closed topological four-manifold is compact, has no boundary, and is locally homeomorphic to ${\mathbb{R}}^4$. A homotopy equivalence need not preserve that local structure.

For a group $\Pi$, an outer automorphism is an automorphism considered modulo conjugation by elements of $\Pi$. Self-homotopy equivalences of an aspherical CW-type space correspond to outer automorphisms of its fundamental group. The subgroup realized by homeomorphisms may be smaller.

A Poincaré complex has the homological duality of a manifold, including local coefficients, but need not itself be a manifold. A closed manifold model of such a complex means a closed manifold equipped with a homotopy equivalence to it. A Poincaré pair is the corresponding relative notion for a space with boundary.

Manuscript's four-dimensional pair

There are closed connected aspherical topological four-manifolds $M,N$ that are homotopy equivalent but not homeomorphic. Their common fundamental group is word-hyperbolic. The same $M$ also admits a self-homotopy equivalence that is not homotopic to any homeomorphism. (OpenAI, 2026)

Here word-hyperbolicity means that triangles in a Cayley graph are uniformly thin. The conclusion concerns the topological category, rather than a pair that is homeomorphic but not diffeomorphic. The manuscript explains why known higher-dimensional rigidity theorems do not directly give four-dimensional homeomorphism rigidity: their four-dimensional applications need additional geometric hypotheses, such as the relevant good-group assumptions.

🏷️ The geometric input: a chamber with a parity obstruction

The construction starts with a collared four-dimensional Poincaré pair $(P,S)$ whose interior homotopy type is a finite graph. An epimorphism ${\pi }_1(P)\to \mathbb{Z}$ defines connected cyclic covers $P_d$ and their boundaries $S_d$.

A marked filling of $S_d$ is a compact oriented four-manifold with this identified boundary and a homotopy equivalence to $P_d$ compatible with the prescribed boundary map. The marking is essential: an arbitrary manifold bounding the same three-manifold is not enough.

The source constructs these two different behaviors:

• $S_2$ has a marked filling, with a relative homotopy equivalence equal to the identity on its boundary.
• $S_d$ has no marked filling for any odd $d\ge 3$.

The positive construction uses plumbing caps of two colors. Each grope body has four terminal caps arranged in two red–blue pairs. In the double cover, select red caps at one vertex and blue caps at the other. Each plumbing edge has exactly one selected endpoint, giving disjoint chosen cap sheets. The source compresses the resulting surfaces and replaces disjoint punctured $S^2\times S^2$ neighborhoods by four-balls, retaining the specified boundary and graph homotopy type. (OpenAI, 2026)

The odd-cover obstruction has a simple combinatorial endpoint. Replace each edge of an odd cycle by five red and five blue parallel edges. A hypothetical marked filling supplies formal data to an endpoint-assignment theorem: each edge must be assigned to one endpoint, with no vertex receiving both two red and two blue edges.

At every vertex, therefore, at least one color is deficient: the vertex receives at most one edge of that color. Adjacent vertices cannot have the same deficient color, since their combined capacity would be at most two while five edges of that color join them. The deficient colors must alternate, which is impossible around an odd cycle.

The hard geometric and analytic input is the implication from a marked filling to these formal data. The endpoint theorem requires compatible edge potentials, injective vertex tangent maps, an invertible tangent-difference map, exact vanishing of each vertex potential on its formal parametrization, and exact gradient-ideal relations for every two-red/two-blue choice. It is not a theorem about arbitrary edge decorations. The source obtains those identities using marked holonomy and companion tensor-obstruction results. In its nilpotent parameter rings each individual parameter squares to zero, while products of distinct parameters remain present; discarding those products would lose the four-slot relations. This post treats that formal construction and the companion marked algebraicity theorem as inputs. (OpenAI, 2026)

🏷️ Reflection converts fillings into closed models

The Davis reflection construction glues copies of a chamber along panels in its boundary. A flag triangulation of $S$ determines the panels: “flag” means that any pairwise adjacent set of vertices spans a simplex. Initially the chamber labels form a vector space over ${\mathbb{F}}_2$, with one coordinate per panel. Crossing a panel changes its coordinate.

For the source’s verified chamber hypotheses, the reflected space is a finite aspherical Poincaré complex. A controlled manifold-to-filling theorem says that a closed manifold model for it would produce the marked filling of the original chamber boundary. This implication uses the full framed hyperbolic middle-homology data and the prescribed markings; asphericity alone would not imply it.

The manuscript then quotients the labels by a binary relation space that acts freely and eliminates unwanted simplicial symmetries. Denote the resulting finite aspherical complex by $Q$, and its folding epimorphism by $w:G={\pi }_1(Q)\to \mathbb{Z}$. Put

$$G_d=w^{-1}(d\mathbb{Z}),{\pi }_1(Q_d)=G_d.$$

Replacing the doubled chambers by their marked fillings produces a closed manifold model of $Q_2$. No $Q_d$ with odd $d\ge 3$ can have such a model. Indeed, a hypothetical model would lift along the finite cover from the full reflection; the imported manifold-to-filling implication would then contradict odd nonfilling. (OpenAI, 2026)

🏷️ A cyclic-cover criterion with no finite-order assumption

The next step is a useful general lemma. Suppose a finite aspherical $Q$ has a manifold model $M$ for $Q_2$, but no such models for all sufficiently large odd covers. Choose $\gamma \in G$ with $w(\gamma )=1$. Conjugation by $\gamma$ defines an outer automorphism $\delta$ of $G_2$, and hence a self-homotopy equivalence of $M$.

Assume a homeomorphism $h$ realizes $\delta$. Write $K=\ker w$ and let $X=K\backslash \tilde{M}$ be the infinite cyclic cover. Its deck generator $T$ is represented by ${\gamma }^2$. Adjusting a lift of $h$ by a deck transformation makes its action on the fundamental group exactly conjugation by $\gamma$. It descends to a homeomorphism $h^{\prime }$ of $X$ commuting with $T$.

There is a proper continuous height function $\tau :X\to \mathbb{R}$ with $\tau (Tx)=\tau (x)+1$. Because $h^{\prime }$ commutes with $T$, its height displacement is bounded, say by $c$. For an integer $r>c+1$, set $F_r=T^r{h}^{\prime }$. Then

$$\begin{array}{c}{\Delta }_r(x)=\tau (F_{r}x)-\tau (x),\\ 0<r-c\le {\Delta }_r(x)\le r+c.\end{array}$$

The lower bound makes the action of $⟨F_r⟩$ free and properly discontinuous. The upper bound shows that every orbit meets the compact band $0\le \tau \le r+c$, so the quotient is compact. Thus $X/⟨F_r⟩$ is a closed aspherical manifold.

Its fundamental group is

$$K{⋊}_{{\mathrm{Ad}}_{{\gamma }^{2r+1}}}\mathbb{Z}\cong G_{2r+1}.$$

It would therefore be a manifold model of the forbidden odd cover. This contradiction proves that no homeomorphism realizes $\delta$. Although ${\delta }^2=1$ as an outer class, the argument did not assume that the hypothetical homeomorphism has finite order. Excluding only involutory representatives would be weaker. (OpenAI, 2026)

🏷️ Two models with different realized symmetries

To prove the stronger nonhomeomorphic-pair statement, let $\mathsf{A}$ be the quotient label vector space and choose a nonzero linear functional $\ell :\mathsf{A}\to {\mathbb{F}}_2$. The doubled-chamber filling can be glued with uniform boundary identifications, giving $M=M_0$, or with identifications twisted by the double-cover involution according to $\ell$, giving $N=M_{\ell }$.

Both maps $M_{ϵ}\to Q_2$ are homotopy equivalences. They also realize different finite subgroups of $\mathrm{Out}(\Pi )$, where $\Pi ={\pi }_1(Q_2)$:

$$\begin{array}{rl}\mathsf{A}& =\{(a,0):a\in \mathsf{A}\},\\ {\mathsf{B}}_{\ell }& =\{(a,\ell (a)):a\in \mathsf{A}\}.\end{array}$$

These lie in $\mathsf{A}\times ⟨\delta ⟩$. The first is realized on $M$, the second on $N$, and the cyclic-cover argument forbids $\delta$ on $M$.

The two subgroups are not conjugate in the full outer automorphism group. The source detects this by counting elements that have finite-order lifts to actual automorphisms of $\Pi$. This property is invariant under conjugation. Here the source first embeds the relevant group extension faithfully into $\mathrm{Aut}(\Pi )$ and identifies it with the full preimage of $\mathsf{A}\times ⟨\delta ⟩$. This ensures the count controls every possible automorphism lift. Every element with second coordinate one lacks such a lift, because its lift has nonzero odd value under the cyclic character. The two subgroups agree above $\ker \ell$; outside that kernel, the graph subgroup has no such lifts, whereas at least one generating label in the horizontal subgroup lifts to a Coxeter involution. Their counts differ. (OpenAI, 2026)

🏷️ Controlling all possible homeomorphisms

Nonconjugacy alone is not enough: a homeomorphism between $M$ and $N$ could act through an unanticipated outer automorphism. Two further inputs close this gap:

$$\begin{array}{c}\mathrm{Out}(\Pi )\text{ is finite},\\ N_{\mathrm{Out}(\Pi )}(\mathsf{A})=\mathsf{A}\times ⟨\delta ⟩.\end{array}$$

For the first, the source constructs a proper cocompact cubical action and bounds square grids uniformly, implying hyperbolicity. Poincaré duality excludes splittings over virtually cyclic subgroups. Concretely, a nontrivial splitting with edge group $U$ gives a tree-path cocycle with coefficients in the finite-support module ${\mathbb{F}}_2[\Pi /U]$. Its support grows unboundedly along an unbounded orbit, whereas a coboundary has uniformly bounded support, so its cohomology class is nonzero. But duality and homological Shapiro give

$$\begin{array}{rl}& H^1(\Pi ;{\mathbb{F}}_2[\Pi /U])\\ & \cong H_3(\Pi ;{\mathbb{F}}_2[\Pi /U])\\ & \cong H_3(U;{\mathbb{F}}_2)=0.\end{array}$$

The last equality uses torsion-freeness of $\Pi$: a virtually cyclic subgroup is trivial or infinite cyclic. The source then invokes the hyperbolic-group theorem that an infinite outer automorphism group would force such a splitting.

The normalizer calculation uses rigidity of the Coxeter generators in the lifted triangulation and the chosen binary relations. An automorphism normalizing the label subgroup must preserve the fibers of the lifted vertex labels. It descends to a symmetry preserving the binary relations, which was arranged to be trivial; the remaining lift is a deck transformation. This is a substantial rigidity input, not a consequence of the cyclic-cover criterion alone.

Finally let $R_M$ be the outer automorphisms realized by homeomorphisms of $M$. Since it contains $\mathsf{A}$ but not $\delta$, the normalizer of $\mathsf{A}$ inside $R_M$ is just $\mathsf{A}$. A proper subgroup of a finite $2$-group is strictly contained in its normalizer, so $\mathsf{A}$ must be a Sylow $2$-subgroup of $R_M$. A homeomorphism $M\to N$ would carry a conjugate of ${\mathsf{B}}_{\ell }$ into $R_M$. Its equal order would make it another Sylow subgroup, forcing the conjugacy already excluded. This proves the claimed distinction between the two models. (OpenAI, 2026)

The selected-source review covers the statement, parity interface, cyclic-cover proof, realized-symmetry argument and duality mechanism. The full chamber construction, tensor identities, support cubulation, generator rigidity and imported companion results have not all been independently verified here.

🏷️ Related notes

Marked disk embedding and manifold realization introduces the companion tensor and reflection inputs used by this construction. The four-dimensional Singer manuscript uses a related duality-and-Shapiro reduction for a different invariant of aspherical four-manifolds.

🏷️ Key steps and proof outlines

The even filling removes a complete geometric basis

The two-color choice in the double cover does more than remove a few intersections. Select red ports at vertex zero and blue ports at vertex one. Every plumbing edge has the same color at both ends and joins the two vertices, so exactly one endpoint is selected. The corresponding cap parallels are mutually disjoint.

Compress each upper torus along its selected cap. The source’s framed-dual construction then gives embedded pairs of two-spheres, with one transverse intersection within each pair and none between distinct pairs. Their homology classes form the entire free middle basis in the universal cover. That completeness is essential: eliminating only part of the middle homology would not produce an aspherical filling.

A regular neighborhood of each pair is a punctured $S^2\times S^2$, with boundary $S^3$. Remove these disjoint neighborhoods and glue in four-balls. Van Kampen preserves the marked fundamental group because the removed neighborhoods, their boundaries, and the replacement balls are simply connected.

To check the universal cover, let $\tilde{A}$ be the lifted complement. Mayer–Vietoris gives

$$H_2(\tilde{A};\mathbb{Z})\oplus \underset{j}{⨁}{H}_2({\tilde{N}}_j;\mathbb{Z})\cong H_2({\tilde{D}}_2;\mathbb{Z}).$$

The second summand maps isomorphically onto the right side by the basis property, so $H_2(\tilde{A})=0$. In the neighboring degrees, the boundary three-spheres account for all third homology of the complement. Filling them by balls kills exactly those classes and produces no fourth homology. The resulting cover is simply connected and acyclic, hence contractible by the homological Whitehead theorem. All changes were interior, so the boundary marking survives. This proves the positive filling claim using the manuscript’s earlier cap and basis lemmas. (OpenAI, 2026)

The odd-cover contradiction has an exact capacity count

Once the endpoint-assignment theorem has been obtained from the marked formal data, its combinatorial consequence is short. At each vertex choose a deficient color, one for which that vertex receives at most one edge. If adjacent vertices both chose red, they could receive at most two of the five red edges joining them. Yet all five edges must be assigned to one of those two endpoints. The same argument applies to blue. Thus any choice of deficient colors must alternate along the cycle, contradicting odd length.

This separates the elementary obstruction from its difficult premise. It is the exact four-slot identities and nondegeneracy conditions of the formal construction that force an endpoint assignment. The parity argument is not an obstruction for arbitrary decorated graphs or arbitrary fillings without the specified marking.

A bounded displacement produces a genuine manifold quotient

The cyclic-cover criterion avoids assuming that a homeomorphism realizing a homotopy involution is itself involutory. Suppose such a homeomorphism exists and lift it to $h^{\prime }$ on the infinite cyclic cover, commuting with deck translation $T$. An equivariant height function $\tau$ is proper: a bounded height band meets only finitely many translates of a compact fundamental set for the $T$-action.

The difference $\tau (h^{\prime }x)-\tau (x)$ is $T$-invariant, so compactness downstairs bounds it by $c$ in absolute value. For $r>c+1$, the composite $F_r=T^r{h}^{\prime }$ has increments between $r-c$ and $r+c$.

The lower bound proves freeness and proper discontinuity: a nonzero iterate cannot fix a point, and only finitely many iterates can return a compact set to itself. The upper bound supplies compactness of the quotient: following an orbit from negative height to its first nonnegative height places a representative in the fixed compact band $[0,r+c]$. Both bounds are needed.

The lifted action on the kernel of the cyclic character is conjugation by ${\gamma }^{2r+1}$, so the quotient manifold has the fundamental group of the forbidden odd cover. Asphericity then identifies its homotopy type with that cover. This contradiction rules out every homeomorphism representing the chosen outer class, regardless of its order.

Distinguishing unmarked manifolds needs one more argument

A forbidden self-equivalence alone does not distinguish two manifolds. The last criterion studies the full subgroup $R_M$ of outer automorphisms realized by homeomorphisms. The source’s separate finiteness and normalizer calculations imply

$$N_{R_M}(\mathsf{A})=\mathsf{A}.$$

In a finite two-group, every proper subgroup is strictly smaller than its normalizer. Therefore $\mathsf{A}$ must already be a Sylow two-subgroup of $R_M$. If a homeomorphism $M\to N$ existed, it would transport the equally large realized subgroup $\mathsf{B}$ of $N$ into another Sylow subgroup of $R_M$. Sylow conjugacy would make $\mathsf{A}$ and $\mathsf{B}$ conjugate in the full outer automorphism group, contrary to the construction. This explains the logical role of the substantial symmetry computations: they control all potential homeomorphisms, including ones in unexpected homotopy classes.

🏷️ Manuscripts and verification scope

• Nonhomeomorphic closed aspherical four-manifolds with the same homotopy type — pinned manuscript and companion source.

No family scope page was present at lean/docs/320.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. Nonhomeomorphic closed aspherical four-manifolds with the same homotopy type.