Sharp isoperimetry and fillings in nonpositive curvature

← All digests

Family 337 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

Among planar regions with a fixed area, a disk has the shortest boundary. In Euclidean space of any dimension, the analogous statement says that balls enclose the most volume for a given boundary area.

What happens when the ambient space is curved? Nonpositive curvature tends to spread geodesics apart. The Cartan–Hadamard isoperimetric conjecture predicts that this geometry cannot make enclosing volume cheaper than in Euclidean space. If curvature is bounded above by a negative constant, the proposed comparison is stronger: the benchmark becomes a ball in the corresponding hyperbolic space.

Family 337 claims the full smooth-manifold comparison in every dimension, including nonsmooth regions with finite perimeter. It also classifies bounded equality regions in the Euclidean comparison.

A companion manuscript asks a broader filling question. Given a closed, oriented geometric boundary, possibly singular and carrying integer multiplicities, can it be filled with no more volume than the Euclidean sharp formula allows? It claims this for proper CAT(0) spaces, which can have singularities and need not be manifolds at all.

2. Background and essential definitions

A Cartan–Hadamard manifold is a complete simply connected Riemannian manifold with nonpositive sectional curvature. Completeness rules out missing finite-distance endpoints. Simple connectedness means every loop can be contracted. Sectional curvature measures curvature in individual tangent two-planes; it is a stronger directional condition than a scalar-curvature bound.

Let $\omega_n$ be the volume of the unit ball in $\mathbb R^n$. A Euclidean ball of radius $r$ has volume $\omega_nr^n$ and boundary area $n\omega_nr^{n-1}$. Eliminating $r$ gives the sharp benchmark

$$P\geq n\omega_n^{1/n}V^{(n-1)/n}.$$

For a smooth bounded region, $P$ is boundary area and $V$ is enclosed volume. For a rough measurable region, perimeter is defined through the total variation of its indicator function. This extends boundary area to sets with corners or less regular boundaries. Sets differing on a volume-zero subset are identified.

The perimeter in M1 is full ambient perimeter. If an auxiliary ball confines a region during the proof, boundary lying on that ball still counts. It is not a relative perimeter that ignores contact with the confining boundary.

A CAT(0) space is a geodesic metric space whose triangles are no thicker than Euclidean comparison triangles: corresponding points on their sides are no farther apart. This is a metric version of nonpositive curvature. A space is proper if closed bounded balls are compact.

An integral current is a generalized oriented geometric object with integer multiplicities, an area or volume called its mass, and a boundary operation. It allows singular supports and cancellation between oppositely oriented pieces. A $k$-cycle $T$ has zero boundary, $\partial T=0$. A filling is a $(k+1)$-current $S$ with $\partial S=T$. This describes a geometric boundary equation, not necessarily a parametrized disk or ball.

3. Development history and previous partial results

The surface isoperimetric comparison goes back to Weil and to Beckenbach and Radó; Bol developed nonzero-curvature model comparisons. Kleiner proved the generalized comparison in dimension three. Croke proved the sharp Euclidean comparison in dimension four.

Further work developed different ways to compare domains. Kloeckner and Kuperberg used optical and chord-based methods. Ghomi and Spruck showed that an appropriate total-curvature inequality for convex hypersurfaces implies the Euclidean isoperimetric inequality and equality rigidity. Their analysis of confined minimizers also supplies a regularity input used by M1.

The manuscripts describe several recent overlapping advances. Chen, Ghomi, and Wang’s cited September 2026 preprint proves Euclidean comparison and rigidity in dimensions three through nine, with additional announced or computationally supported extensions. Other cited work gives small-volume comparisons, reductions to rigidity in certain Alexandrov spaces, or all-dimensional conclusions under extra control of negative curvature. M1 claims to remove those remaining dimensional and curvature restrictions in the smooth Cartan–Hadamard setting.

The current-filling problem has a parallel history. Almgren proved the sharp inequality for integral currents in Euclidean space in arbitrary dimension and codimension. Ambrosio and Kirchheim developed integral currents in metric spaces, making it possible to formulate the question without smooth ambient coordinates.

Wenger proved Euclidean-exponent filling inequalities in broad metric settings including CAT(0) spaces, with a dimension-dependent constant. Schulze obtained the sharp constant-curvature comparison for two-dimensional cycles in smooth Cartan–Hadamard manifolds, in any codimension. For Lipschitz loops, Reshetnyak’s majorization theorem already gives the sharp Euclidean disk-area bound.

M2’s claimed contribution is the universal sharp coefficient for every cycle dimension at least two in arbitrary proper CAT(0) spaces. It does not replace the earlier loop theorem, and it does not convert a current filling into every possible prescribed-topology extension problem.

4. What the new manuscripts claim

Generalized isoperimetry and equality

M1 assumes a complete simply connected smooth $n$-manifold, $n\geq2$, with sectional curvature at most $\kappa\leq0$. Every measurable set $E$ of finite volume and finite ambient perimeter is claimed to have perimeter at least that of the equal-volume ball in the simply connected constant-curvature $\kappa$ model.

For $\kappa=0$, this is

$$P_g(E)\geq n\omega_n^{1/n} \operatorname{Vol}_g(E)^{(n-1)/n}.$$

For $\kappa=-b^2<0$, define

$$V_b(r)=n\omega_n\int_0^r \left(\frac{\sinh(bt)}b\right)^{n-1}dt,$$

$$A_b(r)=n\omega_n \left(\frac{\sinh(br)}b\right)^{n-1}.$$

If $V_b(r)=\operatorname{Vol}_g(E)$, the conclusion is $P_g(E)\geq A_b(r)$. Both bounded regions and unbounded sets of finite volume are included in the inequality.

The equality theorem has narrower scope. For a bounded positive-volume set in a Cartan–Hadamard manifold, equality in the Euclidean formula holds precisely when the set agrees up to volume zero with an open region isometric, with its induced metric, to a round Euclidean ball. The conclusion determines the entire equality region, not just its boundary. It does not require the ambient manifold outside that region to be flat.

Sharp integral fillings

M2 uses cycle dimension rather than ambient dimension. To keep the distinction clear, denote it by $k$. For every compactly supported integral $k$-cycle in a proper CAT(0) space, with $k\geq2$, it claims a compactly supported integral filling satisfying

$$\partial S=T,\qquad \mathbf M(S)\leq \frac{\mathbf M(T)^{(k+1)/k}} {(k+1)^{(k+1)/k}\omega_{k+1}^{1/k}}.$$

Here $\mathbf M$ denotes mass. The coefficient is sharp, as Euclidean spheres and their ball fillings show. There is no restriction on integer multiplicities or ambient dimension. For a two-dimensional cycle, the coefficient is $1/(6\sqrt\pi)$.

Applying this theorem to the boundary of a region yields the Euclidean Cartan–Hadamard inequality in ambient dimensions at least three. Classical surface results supply ambient dimension two. This companion route gives the Euclidean comparison, while M1 also treats negative constant-curvature benchmarks and its stated equality classification.

Spectral and torsional consequences

M1 additionally derives model-ball comparisons for smooth relatively compact domains and every $1<p<\infty$. Radial rearrangement preserves the relevant integrals and decreases the $p$-Dirichlet energy $\int|\nabla u|^p$.

Consequently, the first variational Dirichlet $p$-eigenvalue is at least that of the equal-volume model ball, while the corresponding quotient-normalized $p$-torsional rigidity is at most the ball’s value. At $p=2$, these are the familiar lowest-frequency and torsional-response comparisons. They are consequences of the isoperimetric theorem through an established rearrangement principle.

5. Why the advance matters

Importance: 5/5. The family claims sharp all-dimensional resolutions of central isoperimetric and filling problems, retaining the Euclidean optimal constant rather than only the correct power law.

The exponent alone expresses how area and volume scale. A nonoptimal constant already gives useful control, but the optimal coefficient identifies the ball as the universal benchmark. Equality rigidity adds a structural statement: achieving that efficiency forces a Euclidean region throughout the enclosed interior.

The two manuscripts also broaden the admissible objects in different directions. M1 permits rough and unbounded finite-volume regions in smooth negatively curved geometry. M2 permits singular ambient spaces, arbitrary codimension, and integer multiplicities. These are related strengths, not interchangeable formulations.

The comparison has analytical consequences because level sets connect a function’s gradient energy to their boundary areas. Thus a geometric inequality becomes a source of spectral and variational estimates, explaining why the result reaches beyond the geometry of individual domains.

6. The main idea

Compare boundaries through an isoperimetric profile

M1 first claims a sharp critical Sobolev inequality on immersed hypersurfaces. A Sobolev inequality controls an integral norm of a function by its gradient energy and curvature terms. “Critical” refers to the exponent compatible with the dimension and scaling.

The source builds radial comparison maps into a sphere, with weights satisfying compatible curvature inequalities. Their center and scale can be balanced against a chosen measure. If the sharp threshold failed, a variational minimizer and these balanced coordinate variations would yield an incompatible first-order identity. This is the analytical engine of the smooth-manifold argument.

An arbitrary region’s boundary has uncontrolled mean curvature, so the manuscript does not apply that engine directly to every boundary. It first minimizes full perimeter among sets of fixed volume confined in a large geodesic ball. The resulting confined isoperimetric profile records the least perimeter as volume varies.

For a minimizer, first variation relates boundary mean curvature to the derivative of the profile. Contact with the confining sphere requires separate one-sided control. Cutoffs near singular sets make the constant function an admissible limiting test in the hypersurface inequality. The result is a differential inequality for the profile whose integrated form is the model-ball comparison.

Approximation and cutoffs then pass from confined minimizers to general finite-volume finite-perimeter sets. For equality, the manuscript combines boundary regularity, spectral or low-dimensional substitutes, radial flux identities, and equality in metric comparison. These steps recover a round boundary and then the flat metric throughout the occupied region.

Fill cycles without an ambient normal bundle

M2 begins inside a compact convex ball containing the cycle. Properness makes the ball compact, while CAT(0) geometry makes it convex. Compactness and filling continuity supply an extremizing cycle if the sharp inequality fails.

Radial deformation toward a fixed center produces a curvature-like inequality directly from triangle comparison. The calculation retains the exact transverse factor in the swept mass. This is important because a singular metric space need not have an ambient tangent bundle or normal vectors available for a classical hypersurface proof.

For two-dimensional cycles, the radial inequality and Euclidean lower density provide the base case. Higher cycle dimensions use induction: fill boundaries of small restrictions using the lower-dimensional sharp theorem. This gives an almost sharp local perimeter estimate, then a critical Sobolev inequality on the extremizing current.

A variational minimizer supplies a weighted measure and weighted chord lengths. The manuscript obtains an upper bound for their averaged squared size from radial variation and the minimizer equation. CAT(0) variance gives an incompatible lower bound for the same quantity. This closes the induction and yields an attained compactly supported filling with the sharp coefficient.

7. Scope, limitations, and open questions

The curvature assumption is sectional, not merely Ricci or scalar curvature. Simple connectedness and completeness are part of M1’s hypothesis. Its inequality has the full finite-volume scope, but the Euclidean equality theorem requires boundedness and positive volume. No corresponding general negative-curvature equality classification is asserted in its headline theorem.

M2 requires proper CAT(0) spaces and compact support of the input cycle. It does not state the same result for every nonproper CAT(0) space. Ambient dimension is unrestricted, but the cycle has a finite integer dimension at least two.

A current filling need not be a disk, preserve a prescribed topology, or extend a specified parametrization. Those requirements belong to stronger map-extension formulations. Likewise, the sharp universal estimate does not rule out smaller large-mass growth rates for special spaces under additional rank or dimension hypotheses.

The two proof routes and their claimed consequences are reported as stated. Questions about broader equality classifications, prescribed-topology fillings, and removing properness remain distinct from the results presented here.

8. Sources and further reading

• Chen, Ghomi, and Wang, The isoperimetric inequality and CMC hypersurfaces in Cartan–Hadamard manifolds.
• Ghomi and Spruck, Total curvature and the isoperimetric inequality in Cartan–Hadamard manifolds.
• Wenger, Isoperimetric inequalities of Euclidean type in metric spaces.
• Schulze, Optimal isoperimetric inequalities for surfaces in any codimension.
• Optional background: CAT(0) spaces.

Source manuscripts

• M1. OpenAI, Generalized Cartan–Hadamard isoperimetry and Euclidean equality rigidity (September 23, 2026).
• M2. OpenAI, Sharp integral fillings in CAT(0) spaces (September 23, 2026).

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