Positive spectral curvature and a controlled loss of two dimensions

← All digests

Family 336 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

Positive scalar curvature need not make a manifold small. A round two-sphere multiplied by a line has positive scalar curvature but extends indefinitely along the line. Its two spherical directions remain bounded while one direction stays large.

Gromov’s width question asks whether a general manifold with a uniformly positive scalar-curvature lower bound has a comparable description at a controlled scale: can it be mapped to a space with two fewer dimensions so that every fiber stays uniformly small?

Family 336 claims a quantitative answer and then weakens the curvature assumption from pointwise positivity to spectral positivity. The latter combines curvature with the gradient energy of test functions, allowing scalar curvature to be negative in some regions.

The three manuscripts have complementary roles. M3 develops the higher-dimensional pointwise theorem and a partition construction. M1 uses that construction to obtain the spectral theorem in dimensions at least four. M2 gives a separate weighted argument in dimension three, with an explicit convenient constant. Throughout, the final fiber diameters are measured in the original metric.

2. Background and essential definitions

For a metric space $X$, its Urysohn $d$-width is

$$\operatorname{UW}_d(X) =\inf_{f:X\to K}\sup_{y\in K} \operatorname{diam}_X f^{-1}(y),$$

where $f$ is continuous and $K$ is a simplicial complex of dimension at most $d$. A simplicial complex is built from vertices, edges, triangles, and higher-dimensional simplices glued along faces. For $d=1$, it is a graph.

A fiber $f^{-1}(y)$ consists of all points sent to the same target point. Its diameter is the supremum of distances between any two of its points, including points in different connected components. Bounding each component separately is weaker: many small components could be arbitrarily far apart while sharing one target value.

The width measures how much uncertainty remains after replacing $X$ by a lower-dimensional description. Small fibers do not imply that the map preserves distances between different fibers, is injective, or is a homotopy equivalence.

A Riemannian manifold is complete when its metric has no missing finite-distance limit points. Let $\Delta_g=\operatorname{div}_g\nabla_g$ be the Laplacian, with this sign convention. The spectral condition in the family is

$$\int_M\bigl(4|\nabla\varphi|_g^2+ \mathrm{Scal}_g\varphi^2\bigr)\,dV_g \geq\lambda\int_M\varphi^2\,dV_g$$

for every smooth compactly supported real test function $\varphi$, where $\lambda>0$. It is abbreviated

$$-4\Delta_g+\mathrm{Scal}_g\geq\lambda.$$

This is a quadratic-form inequality, not a pointwise inequality between functions. A pointwise bound $\mathrm{Scal}_g\geq\lambda$ implies it, since gradient energy is nonnegative. The converse need not hold. On a noncompact manifold the definition does not require a lowest eigenvalue to be attained by an eigenfunction.

3. Development history and previous partial results

Gromov’s macroscopic-dimension program relates positive scalar curvature to a loss of large-scale dimensions. Its quantitative form asks for one fiber bound depending only on dimension after normalizing the curvature scale. This is stronger than asking for some finite bound that can depend on each manifold and metric.

Three-dimensional geometry supplied important precedents. Stable minimal-surface arguments of Schoen and Yau, and distance-level estimates of Gromov and Lawson, connect curvature to the size of separating surfaces. Some distance-level arguments require additional control of first homology; they are not diameter bounds for every distance sphere in every manifold.

Liokumovich and Maximo constructed singular foliations of closed three-manifolds with bounds on the area, diameter, and genus of connected level components under pointwise positive scalar curvature. Liokumovich and Wang extended area and diameter control to complete noncompact three-manifolds. Passing from levels to a Reeb graph separates their connected components, explaining how such results yield a graph-valued description.

In higher dimensions, several macroscopic-dimension theorems concerned universal covers of closed manifolds and used spin, fundamental-group, homotopy, or related assumptions. These results gave important topological cases, but differed from a dimension-only quantitative bound for arbitrary complete manifolds.

The spectral viewpoint also has established origins. Perelman’s scalar-curvature functional naturally involves $-4\Delta+\mathrm{Scal}$. Gromov’s torus stabilization studies warped products with circles while keeping the base metric as the reference for width. His three-dimensional theorem in that framework is a close predecessor to M2. The new manuscript describes its contribution as a direct weighted proof covering nonorientable and noncompact cases without global control of the auxiliary weight.

A terminology warning matters here. “Macroscopic scalar curvature” can also mean a fixed-radius ball-volume comparison. Kumar and Sen give counterexamples to the analogous uniform width implication under that different condition. Neither ordinary scalar curvature nor the spectral condition should be replaced by it.

4. What the new manuscripts claim

The spectral theorem in dimensions at least three

Combining M1 and M2, the family states that every connected complete smooth boundaryless Riemannian $n$-manifold, $n\geq3$, satisfying the spectral inequality admits a continuous map

$$f:M\longrightarrow K,\qquad \dim K\leq n-2,$$

whose entire fibers satisfy

$$\operatorname{diam}_g f^{-1}(y) \leq\frac{C_n}{\sqrt\lambda}.$$

The constant depends only on $n$. M2 gives $C_3=500$, explicitly without claiming optimality. For higher dimensions, M1 asserts a finite dimension-dependent constant.

There is no orientability, spin, compactness, or bounded-geometry assumption. Bounded geometry would impose additional uniform local controls, such as curvature and injectivity-radius bounds; those are not part of these statements. The target complex has its ordinary polyhedral topology.

The pointwise theorem and closed-cover consequences

M3 states the corresponding theorem under $\mathrm{Scal}_g\geq1$ in dimensions $n\geq4$. Its stabilized partition theorem is an imported geometric input in M1’s spectral extension, so these two proofs should not be described as independent alternatives for the full spectral conclusion.

M3 also obtains the continuous macroscopic-dimension statement for universal covers of closed positive-scalar-curvature manifolds in every dimension $n\geq2$:

$$\dim_{\mathrm{mc}}\widetilde M\leq n-2.$$

Here the target has the uniform simplex metric, in which simplices have standard Euclidean shape and edge length, and the map is continuous with uniformly bounded entire fibers. This is not automatically the stronger variant requiring a Lipschitz map. The two- and three-dimensional cases use established low-dimensional inputs; M1’s general spectral theorem itself begins in dimension four, with M2 supplying dimension three.

For closed manifolds of dimension at least four, M3 derives a filling-radius bound from its width theorem. Filling radius measures how large a neighborhood of the manifold is needed before its fundamental cycle becomes a boundary in a canonical ambient function space. Under $\mathrm{Scal}_g\geq\sigma^2>0$, the claimed bound is

$$\operatorname{FillRad}(M,g)\leq\frac{C_n}{2\sigma}.$$

The fundamental class uses integer coefficients for orientable manifolds and mod-two coefficients otherwise. M3 further derives the closed aspherical positive-scalar-curvature obstruction in dimensions at least four, while pointing to the stronger independent rational-inessentiality claim in family 335.

5. Why the advance matters

Importance: 5/5. The family claims a central quantitative dimension-drop theorem under very weak global hypotheses, and extends it to a spectral condition permitting local negative scalar curvature.

Uniformity is the key. A compact manifold can always map to a point with fiber diameter equal to its own diameter, but that says nothing useful across a class of metrics. Here one constant is supposed to work for every manifold of the same dimension once the curvature scale is fixed.

The original metric is equally important. A conformal change might improve scalar curvature while stretching some distances uncontrollably. A width bound in the changed metric would not necessarily answer the original question. These manuscripts explicitly maintain the original metric for the final estimate.

The result would also distinguish geometric dimension at a curvature scale from topological dimension. A smooth $n$-manifold remains locally $n$-dimensional. The theorem only compresses it into an $(n-2)$-dimensional description with bounded ambiguity, as the sphere-times-line example illustrates.

6. The main idea

Convert spectral information into a weight

The spectral inequality supplies a positive smooth supersolution $v$ of

$$(-4\Delta_g+\mathrm{Scal}_g)v\geq v$$

after scaling to $\lambda=1$. With $t=-2\log v$, it yields a weighted curvature quantity

$$D(g,t)=\mathrm{Scal}_g+2\Delta_gt-|dt|_g^2\geq1.$$

The weight can be unbounded and can oscillate substantially. The argument therefore cannot assume that replacing the metric by a weighted or conformal version preserves all distances up to a controlled factor.

In dimension three, cut into pieces controlled by surfaces

M2 uses the positive density associated with $t$ in a weighted area-minus-volume minimization problem. Its critical surfaces have prescribed weighted mean curvature. Stability, together with the ambient weighted curvature inequality, gives diameter control on the resulting two-dimensional surfaces in their original induced metric.

The construction first cuts along a locally finite family of stable weighted minimal surfaces to simplify separation. It then doubles pieces, smooths the metric and weight compatibly, and inserts further separating surfaces. Distance slabs acquire controlled diameter, producing a cover in which at most two sets meet at a point. A partition of unity turns such a cover into a graph-valued map.

Gluing the pieces back together needs more than componentwise estimates. M2 keeps collar homotopies inside graph regions tied to a common cover set. This is what controls the distance between disconnected pieces of the same final fiber.

In higher dimensions, stretch first and partition compatibly

M1 first constructs a larger graph metric and a new signed weight with an additional quadratic reserve. This reserve absorbs the cost of later local stretching even when the starting weight has no global bound.

Choose a separated collection of centers in the original metric. Their inner balls cover the manifold, and the corresponding larger balls form a locally finite family. Local graph deformations make each inner ball far from the exterior of its outer ball in an auxiliary metric. A telescoping inverse-metric estimate controls the cost without requiring a uniform bound on how many balls overlap.

Each location is changed only finitely often. This local stabilization allows the construction to remain smooth on a noncompact manifold with uncontrolled geometry at infinity. Locally finite circle warps convert the weighted inequality into the stabilized curvature condition required by M3’s partition theorem.

M3 supplies the geometric dimension drop through compatible cuts and walls. A wall lowers the dimension of one part of the construction, but both adjacent bulk pieces must also be retained. In high dimensions, singular minimizing regions require shielding and regularization while preserving the curvature information where later cuts need it.

The resulting walls have a controlled intersection pattern: $n-1$ distinct walls cannot meet. Continuous sign coordinates associated with the walls therefore have at most $n-2$ simultaneously varying entries. They define a map to a complex of that dimension. Equality of all coordinates places two original points near the same original covering ball, yielding a bound on the entire fiber.

7. Scope, limitations, and open questions

The spectral assertion is for dimensions at least three and a strictly positive lower spectral bound. It is not a theorem under mere pointwise positivity without a uniform lower bound on a noncompact manifold. Compact positive-scalar-curvature manifolds do have a positive minimum, which is why their universal covers enter the stated corollary.

The map is continuous. No global Lipschitz constant, equivariance under covering transformations, smooth fibration structure, or homotopy equivalence is claimed. The target may be an infinite complex.

The constants are not optimized, including the explicit three-dimensional value 500. Optimal width constants and stronger control on the maps or fibers remain separate questions. Fixed-scale macroscopic scalar curvature is a different hypothesis and cannot be substituted.

Finally, M1 depends on M3’s stabilized partition theorem, while M2 uses its own weighted surface construction. The digest preserves these dependencies and explains the statements without checking their proofs.

8. Sources and further reading

• Gromov, Four lectures on scalar curvature, for width, stabilization, and macroscopic dimension.
• Liokumovich and Maximo, Waist inequality for three-manifolds with positive scalar curvature.
• Liokumovich and Wang, width and waist inequalities in complete three-manifolds.
• Kumar and Sen, Urysohn width and macroscopic scalar curvature, for the distinction from fixed-scale macroscopic curvature.

Source manuscripts

• M1. OpenAI, Spectral scalar curvature and uniform Urysohn width (October 5, 2026).
• M2. OpenAI, Spectral scalar curvature and Urysohn width in dimension three (October 5, 2026).
• M3. OpenAI, Positive scalar curvature and uniform codimension-two width (September 23, 2026).

Sources consulted on 26 October 2026 at repository revision 1581d4c. Download the manuscript citations.