Infinitely many distinct closed geodesics on every sphere

← All digests

Family 345 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

A geodesic is the path followed by a freely moving particle on a curved space. A closed geodesic returns to its starting point with its velocity aligned so that the motion repeats forever. On a round sphere, great circles provide infinitely many examples.

Now deform the metric arbitrarily. The sphere may no longer be round, have positive curvature, or possess any visible symmetry. Must infinitely many different closed geodesics survive?

Family 345 claims that they do for every smooth Riemannian metric on the standard sphere $S^n$, in every dimension $n\geq2$. The two-dimensional case is classical; the claimed advance includes all higher-dimensional metrics, even those with degeneracies that obstruct generic Morse-theoretic arguments.

The counting convention is essential. Running around the same path twice, starting at another point, or reversing direction does not create a new geometric image. The claim supplies infinitely many genuinely different paths in the manifold, rather than infinitely many parametrizations of a finite list.

2. Background and essential definitions

Prime geodesics and geometric distinction

A closed geodesic is prime if its period is the least positive period of the trajectory. Its iterates traverse that same trajectory repeatedly. A prime closed geodesic can have self-intersections; prime does not mean a simple embedded circle.

Two geodesics are geometrically distinct here if their images as subsets of the manifold differ. The theorem identifies changes of starting point and orientation, as well as repeated traversal.

The metric is arbitrary but smooth and Riemannian. Riemannian lengths come from a positive-definite quadratic form on each tangent space. More general Finsler metrics need not have this quadratic structure, and some are not symmetric under reversing velocity. The main theorem does not cover all Finsler metrics.

Geodesics as critical points on a space of loops

Let $\Lambda M$ be the space of loops $\gamma:\mathbb R/\mathbb Z\to M$ with square-integrable first derivative. This is the Sobolev space usually denoted $H^1$. It supports the energy functional

$$E(\gamma)=\int_0^1|\dot\gamma(t)|_g^2\,dt.$$

Its nonconstant critical points are closed geodesics parametrized at constant speed. Constant loops are also critical and must be distinguished from the desired trajectories.

The paper filters loop space using square-root energy

$$F(\gamma)=\sqrt{E(\gamma)}.$$

The usual length satisfies $\ell(\gamma)\leq F(\gamma)$, with equality at constant speed. An energy threshold therefore gives a way to organize the topology of loops by their geometric cost.

Topology that persists, and topology that disappears

Homology measures cycles and holes in a space. A cycle in a low-energy part of loop space may remain nontrivial globally, or it may become the boundary of a higher-dimensional family after a larger energy threshold is allowed.

These temporary features matter. Simply counting local critical points or the final homology of loop space can miss information about when classes appear and disappear. The manuscript’s argument is designed to control that filtered behavior.

A geodesic is degenerate when its second variation has additional zero directions beyond the inevitable symmetry of changing the starting time. Such degeneracies can make families of critical loops interact. The theorem does not assume they are absent.

3. Development history and previous partial results

Birkhoff developed minimax methods for finding closed geodesics on surfaces of genus zero. Lyusternik–Fet later proved that every positive-dimensional closed Riemannian manifold has a nonconstant closed geodesic. Existence of one is much weaker than infinitude of distinct images.

Bott’s iteration theory and Gromoll–Meyer’s local Morse theory explain how the topology contributed by repeated traversals is constrained. For a simply connected closed manifold, unbounded Betti numbers of its free loop space over a field force infinitely many geometrically distinct geodesics. Vigué-Poirrier–Sullivan showed that sufficiently rich rational cohomology of the manifold supplies that growth. Spheres are a central exception: their cohomology has one algebra generator, and their degreewise loop Betti numbers are bounded. Gromoll–Meyer, Vigué-Poirrier–Sullivan.

For the two-sphere, Bangert’s work together with Franks’s or Hingston’s established infinitude. Later work extended this to reversible Finsler two-spheres, and Albach obtained a quantitative growth result in that setting. These powerful surface theorems do not automatically extend to higher-dimensional spheres. De Philippis–Marini–Mazzucchelli–Suhr, Albach.

In higher dimensions, Rademacher proved infinitude for strongly bumpy metrics on compact simply connected manifolds and showed that these metrics are generic. Finite multiplicity results also hold without genericity, including at least two geometrically distinct closed geodesics on every Riemannian three-sphere. Passing to a limit of generic metrics is insufficient for infinitude: different geodesics can merge or lose the distinctions used in the counting argument.

The new manuscript builds on string topology, particularly loop cutting and joining, and on Hingston–Rademacher’s resonance theorem for spheres. Block–Manin–Weinberger’s work provides a direct predecessor for its fixed-degree filling estimates. Resonance theorem, persistent homology of function spaces.

The source also discusses earlier unrestricted infinitude claims and reports objections to their proofs. Those objections are the manuscript’s assessment, not conclusions independently reached here. The present digest explains its claimed new argument without treating those earlier claims as established solutions.

4. What the new manuscript claims

For every integer $n\geq2$ and every smooth Riemannian metric on the standard smooth sphere $S^n$, there are infinitely many prime closed geodesics with pairwise distinct images.

No curvature bound, nondegeneracy, genericity, or symmetry hypothesis is imposed. “Standard smooth sphere” specifies the smooth manifold; the theorem does not automatically include every exotic smooth sphere in high dimension.

The manuscript gives a finite-cover corollary. If a closed manifold admits a finite smooth cover diffeomorphic to a standard $S^n$, every smooth Riemannian metric on that manifold also has infinitely many geodesic images. This includes real projective spaces and smooth spherical space forms. The metric downstairs need not be the standard quotient metric.

It then obtains the same conclusion for every nonempty closed smooth three-manifold, whether orientable or not. For infinite fundamental group, this uses Rademacher–Taimanov’s existing theorem. For finite fundamental group, Perelman’s geometrization gives a finite spherical cover, and the new sphere theorem applies. Disconnected manifolds are covered by working on one component. Rademacher–Taimanov.

These are statements about the number of images, not the number of distinct lengths. On a round sphere, infinitely many prime geodesics all have the same length. The manuscript does not claim a quantitative length-counting rate for its new finite-fundamental-group cases.

5. Why the advance matters

Importance: 5/5. If correct, the theorem resolves a central outstanding class in the closed-geodesic infinitude problem and completes the problem for closed three-manifolds.

Spheres are especially important because ordinary loop-space Betti-number growth does not supply the standard infinitude argument. They require finer use of topology and iteration, even though they are among the most familiar manifolds.

The inclusion of degenerate metrics is equally significant. Generic theorems describe a large and well-behaved collection of metrics, but the universal statement must also handle exceptional metrics where critical trajectories interact in more complicated ways.

The proposed degree-independent filling bound and filtered comparison of loop operations could also be useful beyond the final infinitude theorem. The rating concerns these stated advances, not proof verification.

6. The main idea

Assume there are only finitely many prime images

The argument proceeds by contradiction. If only finitely many prime images exist, every nonconstant critical loop is an iterate of one of finitely many oriented trajectories. Iteration theory constrains its Morse index, which counts independent directions in which energy decreases.

The local homology ranks of these iterates are uniformly bounded and occur near finitely many index-versus-length lines. Degenerate trajectories require additional local analysis, but the manuscript constructs the relevant comparison maps using one common flow, so it keeps track of the actual maps rather than only abstract group dimensions.

This restricts what can happen at each energy crossing. It does not yet control temporary topology accumulated between crossings.

Bound the lifetime of ordinary temporary cycles

The manuscript proves a uniform filling statement: there is a constant $D$, depending on the metric, such that any cycle below square-root energy $a$ which bounds in the full loop space already bounds below

$$a+D.$$

The same $D$ works in every homological degree. The statement concerns cycles that are globally boundaries; it does not say that all loop-space topology vanishes after a fixed threshold.

First, shortening and replacement of loop families give an estimate depending on degree. Then a filtered cut operation lowers a high-degree class into a bounded range of degrees. A loop product joins loops to restore the removed degree. Hingston–Rademacher resonance controls the difference between the lengths removed and added, yielding a bounded total increment.

The paper supplies a cut–join identity with explicit filtration bounds. Its unit is the fundamental class of constant loops in the loop product, so keeping the distinction between absolute and relative constructions is part of the argument.

Add reflection and half-period translation

Loop space has natural symmetries. Reflection reverses the parameter, while translation by half a period identifies loops that repeat a half-length root. The paper uses equivariant cohomology, which records topology together with these group actions, with coefficients in the two-element field $\mathbb F_2$.

Let $B_a$ be the reflection-equivariant cohomology below threshold $a$, relative to constant loops, and let $B_\infty$ be its full-loop-space counterpart. Define

$$T(a)=\dim\operatorname{coker}(B_\infty\longrightarrow B_a),$$

summing over degrees. This measures extra sublevel classes that are not restrictions of global ones.

Under the finite-prime-image assumption, the manuscript proves that $T(a)$ is uniformly bounded. It compares the actual equivariant degree on the half-translation fixed locus with the degree of the half-length root. The total discrepancy at a finite cutoff is at most linear in $a$. A long chain of extra classes would instead force a quadratic discrepancy. The ordinary filling estimate handles the remaining degree ranges.

Doubling forces an impossible first appearance

The full symmetry of loops gives

$$T(2a)\geq T(a).$$

This is monotonicity under doubling, not a claim that $T$ is monotone between arbitrary thresholds. Since $T$ is a bounded integer, an interval where it reaches its maximum propagates to arbitrarily large doubled intervals.

The equality case of the doubling comparison requires certain global classes, computed using the round-metric filtration, to appear with independent partners obtained by cup product with an evaluation class. On a sufficiently large maximal interval, one of these classes must become visible for the first time at a critical crossing.

But the local crossing calculation makes the relevant cup action vanish there. The new class has appeared while its required partner has not. This contradicts the equality condition and rules out the finite-prime assumption.

Finally, the finite-cover argument controls multiplicity: a covering of degree $r$ has at most $r$ upstairs prime images projecting to a given downstairs image. Infinitely many upstairs images therefore cannot collapse to finitely many downstairs ones, including when geodesics self-intersect.

7. Scope, limitations, and open questions

The main theorem concerns standard spheres with smooth Riemannian metrics. The three-manifold conclusion uses additional established theorems, and the finite-cover conclusion requires an actual finite smooth spherical cover.

The general infinitude problem for every closed manifold in every dimension is not claimed here. Nor does the result imply that the geodesics are embedded, mutually disjoint, or have unbounded distinct lengths.

Quantitative questions remain separate: how many images occur below a length threshold, how their lengths are distributed, and what uniform estimates depend only on geometric bounds. The proof’s qualitative contradiction does not supply those conclusions automatically.

This digest reports the manuscript’s filtered-topology mechanism and its stated dependencies. It does not audit the cut–join identity, the equivariant calculations, or the treatment of degenerate local crossings.

8. Sources and further reading

The linked primary sources describe loop-space growth, surface results, resonance, quantitative fillings, and the three-manifold input. Optional definitions appear under closed geodesic, Morse theory, and equivariant cohomology.

Source manuscripts

• M1. OpenAI, Infinitely many closed geodesic images on every Riemannian sphere (September 24, 2026).

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