Ergodicity of the geodesic flow on surfaces of nonpositive curvature

Abstract: We give a full proof that the geodesic flow of a closed surface of genus at least two with a (C^{3+\mathrm{Lip}}) metric of nonpositive curvature is ergodic with respect to Liouville measure. In a forthcoming paper, we will present the corresponding result for smooth rank-one manifolds in every dimension.

• The paper proves the ergodicity of the geodesic flow on closed, connected, oriented surfaces of genus at least two with nonpositive Gaussian curvature and $C^{3+ ext{Lip}}$ metrics.
• it verifies the Liouville probability measure is ergodic on surfaces by proving the vanishing of positive measure singular families of vectors whose geodesic trajectories lie entirely within the zero-curvature locus.
• The proof derives an incompatible lower bound on fundamental frequency modes, leveraging Guillemin-Kazhdan operators, connection Laplacians, and heat smoothing techniques.

Main result and setting

The paper proves that the geodesic flow on a closed, connected, oriented surface of genus at least two is ergodic with respect to Liouville measure whenever the metric is $C^{3+\mathrm{Lip}}$ and has nonpositive Gaussian curvature. More precisely, if $M$ is such a surface and $g^s:SM\to SM$ denotes its geodesic flow, then the invariant probability measure

$dm=\frac{d\bar a\,d\theta}{2\pi},$

where $\bar a(M)=1$, is ergodic for $g^s$ (2610.27697).

The result concerns the non-Anosov case, in which the curvature is allowed to vanish on nontrivial sets. For strictly negative curvature, ergodicity follows from uniform hyperbolicity. In contrast, nonpositive curvature permits geodesics that remain entirely in the zero-curvature locus and therefore have vanishing transverse Lyapunov exponents. The central issue is consequently to show that this singular family has zero Liouville measure.

The argument reduces the theorem to the following assertion. Let

$$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$

be the set of vectors whose entire geodesic remains in zero curvature. Pesin’s ergodicity theorem on the regular set implies that the restriction of the geodesic flow to $SM\setminus S$ is ergodic with respect to the restricted Liouville measure. Therefore, it suffices to establish $m(S)=0$. The paper proves this by assuming $m(S)>0$ and deriving an incompatible lower bound on the Fourier modes of the indicator function of $M$0.

Fourier decomposition of the singular set

Let $M$1. Since $M$2 is closed and invariant under the geodesic flow, $M$3 satisfies

$M$4

in the distributional sense, where $M$5 is the geodesic vector field. Because $M$6 is also invariant under velocity reversal, $M$7, so only even fiberwise Fourier modes occur.

Writing $M$8 for the $M$9th eigenspace of the vertical vector field $g^s:SM\to SM$0, the function decomposes as

$g^s:SM\to SM$1

with

$g^s:SM\to SM$2

The zeroth mode

$g^s:SM\to SM$3

is nonnegative. If

$g^s:SM\to SM$4

then every Fourier coefficient $g^s:SM\to SM$5, viewed as a section over $g^s:SM\to SM$6, is supported in $g^s:SM\to SM$7. Moreover, $g^s:SM\to SM$8 vanishes on the open set $g^s:SM\to SM$9, while it has positive integral if $dm=\frac{d\bar a\,d\theta}{2\pi},$0. Since Gauss–Bonnet and the genus assumption imply that $dm=\frac{d\bar a\,d\theta}{2\pi},$1 somewhere, $dm=\frac{d\bar a\,d\theta}{2\pi},$2 is nonconstant.

A key geometric observation is that $dm=\frac{d\bar a\,d\theta}{2\pi},$3 vanishes on $dm=\frac{d\bar a\,d\theta}{2\pi},$4. Indeed, a point in the support of $dm=\frac{d\bar a\,d\theta}{2\pi},$5 admits a direction belonging to $dm=\frac{d\bar a\,d\theta}{2\pi},$6, and the curvature along that geodesic is identically zero. Since $dm=\frac{d\bar a\,d\theta}{2\pi},$7, every point of $dm=\frac{d\bar a\,d\theta}{2\pi},$8 is consequently a maximum of $dm=\frac{d\bar a\,d\theta}{2\pi},$9, yielding

$\bar a(M)=1$0

The $\bar a(M)=1$1 regularity assumption gives $\bar a(M)=1$2, and hence, with $\bar a(M)=1$3,

$\bar a(M)=1$4

These quadratic and linear vanishing estimates are the precise way in which the metric regularity enters the proof.

Guillemin–Kazhdan operators and the connection Laplacian

Let $\bar a(M)=1$5. The frame vector fields satisfy

$\bar a(M)=1$6

The Guillemin–Kazhdan operators

$\bar a(M)=1$7

map $\bar a(M)=1$8 to $\bar a(M)=1$9 and $g^s$0, respectively. The equation $g^s$1 becomes the Fourier recurrence

$g^s$2

The commutator identity

$g^s$3

produces the energy formula

$g^s$4

for $g^s$5. Since $g^s$6, this implies

$g^s$7

This inequality propagates information from the zeroth Fourier mode toward arbitrarily high even modes.

The proof requires smoothing each mode while retaining quantitative control of its support near $g^s$8. For this purpose, the authors introduce

$g^s$9

On $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$0, this is the connection Laplacian

$$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$1

Its heat semigroup is denoted by

$$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$2

The semigroup preserves Fourier degree, is contractive on $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$3, and satisfies the uniform estimates

$$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$4

Uniformity in $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$5 is essential: the eventual contradiction requires simultaneous control of infinitely many Fourier modes.

Finite propagation and weighted heat estimates

Ordinary heat propagation is instantaneous, so heat smoothing does not preserve support in $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$6. The paper instead derives weighted estimates from the wave equation associated with $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$7: $$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$8 The conserved wave energy is

$$S=\{v\in SM:K(\pi(g^sv))=0\text{ for every }s\in\mathbb R\}$$9

Finite propagation speed implies that if $SM\setminus S$0 is supported in $SM\setminus S$1, then

$SM\setminus S$2

where

$SM\setminus S$3

The authors establish this using a weighted energy with weight $SM\setminus S$4. The argument remains valid for merely $SM\setminus S$5 initial data by approximation.

The heat semigroup is represented through the wave propagator by a Gaussian cosine transform. Combining this representation with finite propagation yields the weighted bounds

$SM\setminus S$6

and

$SM\setminus S$7

The constants are independent of the Fourier degree $SM\setminus S$8. Together with the estimates

$SM\setminus S$9

these bounds show that curvature-dependent commutator terms become small at short heat times.

The heat commutator

Heat smoothing does not commute with $m(S)=0$0, because the connection Laplacians in adjacent Fourier degrees interact differently with curvature. The relevant commutators are

$m(S)=0$1

and

$m(S)=0$2

on $m(S)=0$3.

Although these expressions appear to involve derivatives beyond the nominal regularity of the metric, they can be interpreted on the quadratic-form domain and estimated by

$m(S)=0$4

Applying the weighted heat estimates gives

$m(S)=0$5

Duhamel’s formula then yields the central commutator estimate: $m(S)=0$6

The exponent $m(S)=0$7 is decisive. The factor $m(S)=0$8 comes from the quadratic vanishing of curvature near $m(S)=0$9 and the weighted heat estimate, while integration in heat time contributes an additional factor of $m(S)>0$0. This error is sufficiently small when the heat time is chosen inversely proportional to the Fourier degree.

Propagation to high Fourier modes

Set

$m(S)>0$1

Because $m(S)>0$2 is nonconstant, $m(S)>0$3. Strong convergence of $m(S)>0$4 to $m(S)>0$5 therefore implies that there are constants $m(S)>0$6 and $m(S)>0$7 such that

$m(S)>0$8

The failure of the smoothed function to remain invariant under $m(S)>0$9 is measured in odd degree by

$M$00

The original recurrence for $M$01 and the heat commutator estimate imply

$M$02

The nonpositive-curvature energy inequality propagates the zeroth-mode lower bound: $M$03 By Cauchy–Schwarz and Parseval,

$M$04

Choosing

$M$05

with $M$06 sufficiently small makes this accumulated error at most $M$07, uniformly in $M$08. Consequently,

$M$09

The generic heat estimate gives

$M$10

Combining the lower and upper bounds yields

$M$11

with a constant independent of $M$12. This is incompatible with Parseval: $M$13 whereas $M$14. Thus $M$15 is impossible.

The contradiction establishes $M$16, and Pesin’s theorem then implies ergodicity of the geodesic flow with respect to Liouville measure.

Limitations and open questions

The theorem is restricted to closed oriented surfaces of genus at least two and requires $M$17 regularity. The proof uses this regularity through the facts that $M$18, $M$19 vanishes on $M$20, and the quadratic estimates

$M$21

hold globally. The paper does not determine whether the same conclusion remains valid under substantially weaker regularity.

The argument is specifically two-dimensional. It relies on the scalar curvature function, the fiberwise Fourier decomposition indexed by integers, and the Guillemin–Kazhdan raising and lowering operators. The authors state that the corresponding result for smooth rank-one manifolds of nonpositive sectional curvature in arbitrary dimension has also been obtained, but its proof is not included here. The higher-dimensional mechanisms replacing the surface Fourier recurrence and the associated commutator estimates therefore remain outside the scope of this paper.

The proof also establishes ergodicity through the vanishing of the singular set, rather than by deriving quantitative mixing or decay of correlations. No rate of convergence, spectral gap, or statistical property beyond ergodicity is obtained.

Conclusion

The paper proves ergodicity of the geodesic flow for every closed genus-at-least-two surface with a $M$22 nonpositively curved metric. Its principal contribution is a contradiction argument showing that a positive-measure family of geodesics confined to the zero-curvature locus would force the even Fourier coefficients of its indicator function to satisfy the non-summable lower bound $M$23. The proof combines Guillemin–Kazhdan energy identities, finite propagation for the connection wave equation, weighted heat estimates, and a degree-uniform heat commutator bound. This establishes that the singular set has zero Liouville measure, allowing Pesin’s theorem to yield the stated ergodicity result (2610.27697).