Balls, tubes and slabs: the least-area shapes in a periodic cube

← All digests

Family 354 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

In ordinary three-dimensional space, a ball encloses a prescribed volume using the least surface area. Now imagine a unit cube whose opposite faces are identified: leaving through one face brings you back through the opposite face. Which shape encloses a prescribed fraction of this periodic space with the least boundary area?

For a small amount of material, a ball still looks best. With more material, a circular tube can wrap once around the space and avoid paying for end caps. With still more, a slab bounded by two flat sheets may be cheaper. These simple competitors suggest a ball–tube–slab sequence.

The hard question is whether a more intricate surface, perhaps with several handles and passages through the cube, can ever do better. Family 354 claims a complete answer for the cubic flat three-torus: the three expected types and their complements are all the minimizers, including at the two volumes where the preferred type changes.

This is a geometric phase transition. The quantity being optimized is boundary area, the parameter is enclosed volume, and the minimizing shape changes its topology as that parameter grows.

2. Background and essential definitions

A periodic space with ordinary local geometry

The cubic flat three-torus is

$$\mathbb T^3=\mathbb R^3/\mathbb Z^3.$$

Points differing by an integer vector represent the same point. Small neighborhoods have the ordinary Euclidean metric, so the space is flat. Globally, however, straight lines in coordinate directions close into loops of length one.

A region $E\subset\mathbb T^3$ has volume $|E|$ between zero and one. Its perimeter $P(E)$ means the area of its boundary. The theorem uses finite-perimeter sets, a generalized class that permits rough competitors as well as smooth regions. One can define perimeter as the total variation of the region’s indicator function; for a smooth boundary this agrees with usual surface area.

The isoperimetric profile is

$$I(V)=\inf\{P(E):|E|=V\}.$$

It records the least boundary area needed to enclose volume $V$. A region and its complement have the same separating boundary, so

$$I(V)=I(1-V).$$

It therefore suffices to understand $0<V\le1/2$.

The three candidates

A sufficiently small embedded ball of radius $r$ has volume $4\pi r^3/3$ and area $4\pi r^2$. Expressed in terms of its volume, its area is

$$P_{\mathrm{ball}}(V)=(36\pi)^{1/3}V^{2/3}.$$

A solid circular tube around a shortest closed coordinate geodesic has length one and no end caps. Its volume is $\pi r^2$, and its boundary area is $2\pi r$, giving

$$P_{\mathrm{tube}}(V)=2\sqrt{\pi V}.$$

A coordinate slab of width $V$ has two boundary tori, each of area one. Its perimeter is consequently

$$P_{\mathrm{slab}}(V)=2.$$

Here a boundary torus is a flat two-dimensional periodic sheet, not a doughnut-shaped surface drawn in ordinary space. The periodic identification removes the slab’s apparent side faces.

These formulas give upper bounds for $I(V)$. They become the exact profile only if every other possible region has at least as much perimeter.

3. Development history and previous partial results

The ball–cylinder–slab prediction appears in Hauswirth, Pérez, Romon and Ros’s The periodic isoperimetric problem and Ros’s later account of isoperimetry. Reflection also links the periodic problem with relative perimeter in a cube, where boundary lying on the cube’s walls is not counted. The canonical pieces there are an eighth of a ball at a corner, a quarter cylinder along an edge, and a coordinate cut.

Important parts of the prediction were known well before the full classification. Hadwiger established the optimality of a coordinate cut at half volume in the cube. Morgan and Johnson’s small-volume comparison gave balls as minimizers for sufficiently small volumes on compact flat manifolds. Acerbi, Fusco and Morini established the uniqueness of coordinate slabs near half volume in the cubic torus. These results controlled neighborhoods of the ends of the relevant volume range, but did not reach the predicted transition points with matching sharp bounds.

The possible geometry of an unknown minimizer was also progressively constrained. Reflection arguments supplied symmetry; volume-constrained stability supplied restrictions on topology. Ritoré and Ros developed results for stable constant-mean-curvature surfaces in flat three-manifolds. Hauswirth and collaborators obtained the connectedness, genus and curvature–area estimates used in this manuscript.

More recently, Emanuel Milman obtained effective ranges, including the cubic spherical regime through volume fraction approximately $0.120582$. He also formulated the reduction to the two proposed transition volumes, using the differential comparison method of Hauswirth and collaborators. This is a substantial precursor: knowing the right endpoint information can force the whole profile between those endpoints.

The remaining difficulty was not simply locating where the three elementary formulas cross. It was ruling out nonstandard minimizers at those crossing volumes, including extra shapes tied with the familiar ones. Family 354 claims to complete those endpoint exclusions.

4. What the new manuscript claims

For every $V\in(0,1)$, set $v=\min(V,1-V)$. The manuscript states

$$I(V)=\min\left\{(36\pi)^{1/3}v^{2/3},\;2\sqrt{\pi v},\;2\right\}.$$

For $V\le1/2$, it gives the following exhaustive classification, up to ambient isometry and modification on a set of volume zero.

Volume range	Minimizing region
$0<V<4\pi/81$	An embedded round ball
$V=4\pi/81$	A ball or a circular tube around a shortest closed geodesic
$4\pi/81<V<1/\pi$	Such a circular tube
$V=1/\pi$	A circular tube or a coordinate slab
$1/\pi<V\le1/2$	A coordinate slab

At the two transitions there are exactly the listed alternatives, with no additional equality cases. For volumes above one half, take the complements of the corresponding small-volume regions.

The transition fractions are approximately $0.1551$ and $0.3183$. The ball has radius $(3V/(4\pi))^{1/3}$, the tube has radius $\sqrt{V/\pi}$, and the slab has width $V$. These formulas use the unit cubic torus throughout.

“Up to ambient isometry” allows moving a shape around the torus and applying the symmetries of the cubic lattice. It does not allow an arbitrary rotation that fails to preserve that lattice. In particular, the optimal tubes follow shortest coordinate loops, and the optimal slabs have coordinate directions.

The classification applies to unrestricted finite-perimeter competitors. It does not assume that the region is convex, symmetric, connected or given by one of the three models in advance.

5. Why the advance matters

Importance: 5/5, conditional on the claim. This would settle a classical, concrete isoperimetric conjecture by giving both the optimal value and every optimizer. Exact profiles are much rarer than nonsharp inequalities or classifications restricted to very small volumes.

The example exposes a basic competition between local geometry and global topology. At small scale, a ball behaves as it does in Euclidean space. A tube benefits from closing around a periodic direction. A slab benefits from extending through two periodic directions. The boundary can become simpler in area while the region becomes less localized.

The transition cases are mathematically significant. A formula for the infimum alone would leave open whether unexpected surfaces share the optimum. The manuscript’s strict exclusions claim to remove that ambiguity as well.

Its proposed method also offers a general lesson for slicing arguments. Optimizing each cross-section independently can underestimate the cost of a whole object, because the optimal sections may not fit together. The additional information here is how neighboring sections are nested.

6. The main idea

Reduce a continuum of volumes to two difficult ones

Classical regularity says an isoperimetric boundary in this three-dimensional setting is smooth and has constant mean curvature. The mean curvature describes its first-order area response when the enclosed volume changes. Volume-constrained stability says that small perturbations preserving volume cannot decrease area to second order.

The existing structural theory severely restricts such minimizers. Planar, spherical and circular-cylinder boundaries give the standard candidates. A nonstandard minimizing boundary must be connected and have genus between two and four. The genus counts handles: a sphere has genus zero, a torus genus one, and more elaborate candidates have at least two.

Curvature–area estimates and differential inequalities for the isoperimetric profile then provide the two-volume reduction. If every minimizer at

$$V_1=\frac{4\pi}{81},\qquad V_2=\frac1\pi$$

is standard, the comparison argument gives the complete profile and the classification in the intervening open ranges. The new work can therefore concentrate on excluding higher-genus competitors at those two values.

Use symmetry to study nested planar slices

Reflection and stability put a hypothetical nonstandard minimizer into a coordinate-monotone form in a cube. Coordinate monotonicity means that its sections shrink as one moves in the appropriate coordinate direction. The remaining configurations have two possible patterns of occupied vertices.

The manuscript relates the area of the three-dimensional boundary to areas, boundary lengths and fluxes of its planar sections. A flux identity integrates a component of the normal vector and links it to the section geometry. The coarea formula expresses an integral over a surface in terms of its level slices, with the appropriate slope factor.

These tools and scalar inequalities exclude one vertex pattern at both transitions and the other pattern at the ball–tube transition. The delicate case is the remaining configuration at the tube–slab transition.

Why nesting improves the bound

A planar isoperimetric inequality supplies a lower bound for the perimeter of each section from its area. But those separate bounds are not always strong enough after integration. Near the difficult curvature range, they leave a small deficit.

The proposed improvement starts with the bottom section. Its complementary corner has two edge extents, which need not be equal. A calibration—here an auxiliary unit vector field paired with the boundary tangent—controls both extents using the corner’s area and boundary data. This step uses monotonicity, without assuming that the corner is circular or convex.

Every intermediate section lies inside the bottom section. Thus either orientation of a strip-like section has an endpoint constrained by one of those extents. The strip cannot simultaneously have the area it needs and the least conceivable planar boundary length. Nesting forces a definite excess.

A useful analogy is packing a sequence of shelves into a tapered cabinet. Each shelf might be optimal when designed alone, but its permitted width is constrained by the cabinet and the shelves around it. Here the constraint is exact set inclusion, and the cost is perimeter.

The manuscript converts this extra section length into a positive gain in total surface area using the flux identity and Jensen’s inequality. The latter controls averages of nonlinear quantities by exploiting convexity. That gain supplies the missing strict margin against a nonstandard minimizer.

Finish with a one-variable estimate

After the geometric reduction, a residual inequality depends on a single curvature parameter. The manuscript supplies outward arithmetic bounds and interpolation estimates between finitely many parameter values, together with exact calculation files. Those are stated components of its argument. This digest does not execute or independently certify them.

The paper performs much of the analysis in a torus of side length two with normalized area. Its final theorem restores the unit torus. Keeping that normalization separate is essential: doubling lengths multiplies physical areas by four and volumes by eight. The profile displayed above is the original unit-torus perimeter, not the intermediate normalized quantity.

7. Scope, limitations, and open questions

The claim concerns the cubic flat three-torus. Changing the lattice, taking unequal side lengths, or moving to higher dimensions changes the available loops, sheet areas and possible competing shapes. The same numerical thresholds cannot be carried over without new analysis.

It is a global minimization and equality theorem, not a classification of all constant-mean-curvature surfaces or all stable surfaces. Complicated periodic surfaces can exist even when none minimizes perimeter at its enclosed volume.

The result also does not by itself provide a quantitative stability estimate: it does not state that a region with nearly optimal perimeter must be close to a canonical optimizer by a particular measured amount. Such a statement becomes especially subtle near a transition, where two distinct types tie.

Further questions include the profiles of rectangular and general flat tori, rates of closeness for almost minimizers, and how the nesting method extends beyond this three-dimensional setting. The manuscript’s full theorem and equality claims are reported here as stated, without proof verification.

8. Sources and further reading

The manuscript’s introduction and Theorem 1.1 give the profile and full equality classification. Sections 2–6 explain the two-volume reduction and the nesting improvement; its appendices describe the scalar estimates.

Primary background includes Hauswirth–Pérez–Romon–Ros, Acerbi–Fusco–Morini, and Milman’s effective isoperimetric comparisons. These distinguish the earlier structural and partial results from the endpoint exclusions claimed here.

Source manuscripts

• M1. OpenAI, The Isoperimetric Conjecture for the Cubic Flat Three-Torus (September 24, 2026).

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