Constant-curvature spaces and Killing-Hopf

Intuition [Beginner]

A space of constant curvature looks the same in every direction at every point. The curvature at each point does not depend on which 2-plane you measure, and it does not vary from point to point.

There are exactly three such geometries. Flat space (Euclidean ${\mathbb{R}}^n$): zero curvature everywhere, where parallel lines never meet and triangles have angles summing to 180 degrees. Spherical space (the $n$-sphere): positive curvature, where "parallel" lines eventually converge and triangles have angles summing to more than 180 degrees. Hyperbolic space: negative curvature, where parallel lines diverge and triangles have angles summing to less than 180 degrees.

These three model geometries are the building blocks from which all constant-curvature spaces are assembled. A "space form" is a quotient of one of these models by a discrete group of symmetries --- like taking a strip of wallpaper and rolling it into a cylinder (a flat quotient of ${\mathbb{R}}^2$).

Why does this concept exist? The Killing-Hopf theorem says these three models and their quotients exhaust all possibilities for constant-curvature geometry. This classification is the starting point for Thurston's geometrisation programme, which classifies all 3-manifolds.

Visual [Beginner]

A drawing of three 2-dimensional geometries side by side: a flat plane with a triangle whose angles sum to exactly 180 degrees; a sphere with a triangle (bounded by great-circle arcs) whose angles sum to more than 180 degrees; and a saddle-shaped hyperbolic disk (Poincare disk model) with a triangle whose angles sum to less than 180 degrees. Each triangle is labelled with its angle sum.

Every simply-connected complete space of constant curvature is isometric to one of these three models.

Worked example [Beginner]

Consider the flat torus $T^2={\mathbb{R}}^2/{\mathbb{Z}}^2$, obtained by identifying opposite sides of a unit square. We verify that $T^2$ is a space of constant curvature zero.

Step 1. The torus inherits its metric from ${\mathbb{R}}^2$ via the quotient map. At each point of $T^2$, the tangent space is identified with ${\mathbb{R}}^2$, and the metric is the standard flat metric $d{x}^2+d{y}^2$.

Step 2. The curvature of flat ${\mathbb{R}}^2$ is zero. Since the metric descends without modification, the curvature of $T^2$ is also zero at every point. So $T^2$ is a space of constant zero curvature.

Step 3. The torus is not simply-connected: a loop going once around the torus cannot be shrunk to a point. So the Killing-Hopf theorem says $T^2$ is a quotient of ${\mathbb{R}}^2$ (the simply-connected model) by ${\mathbb{Z}}^2$ (a discrete group of translations acting freely).

What this tells us: the flat torus is a space form --- a quotient of flat space by a discrete symmetry group.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Definition (Space of constant curvature). A Riemannian manifold $(M,g)$ has constant sectional curvature $K$ if the sectional curvature of every 2-plane at every point equals $K$. The three model spaces are:

• ${\mathbb{R}}^n$ with the flat metric (curvature $K=0$)
• $S^n(R)$, the sphere of radius $R$ in ${\mathbb{R}}^{n+1}$ (curvature $K=1/R^2$)
• ${\mathbb{H}}^n(R)$, hyperbolic $n$-space of curvature $K=-1/R^2$

Definition (Space form). A space form is a complete Riemannian manifold of constant sectional curvature. A space form is called spherical, flat, or hyperbolic according to whether $K>0$, $K=0$, or $K<0$.

The space forms are precisely the quotients of the simply-connected models by discrete groups of isometries acting freely. A group $\Gamma$ acts freely on $M$ if $\gamma \cdot p=p$ implies $\gamma =\mathrm{id}$ for all $\gamma \in \Gamma$ and $p\in M$.

Counterexamples to common slips

• A quotient of $S^n$ by a non-free action is not a manifold. The quotient $S^2/\{\pm 1\}={\mathbb{RP}}^2$ is a manifold because the antipodal map acts freely. But $S^2/\{$rotation by 180 degrees about the $z$-axis$\}$ is not a manifold at the fixed poles.

• Not every flat manifold is a torus. In dimension 3, there are exactly 10 compact flat manifolds (6 orientable, 4 non-orientable), classified by Bieberbach. Only one of these is the 3-torus $T^3$. The others involve glide reflections and screw motions.

• Constant curvature does not imply simply-connected. The flat torus $T^2$, real projective space ${\mathbb{RP}}^n$, and lens spaces $S^{2n+1}/{\mathbb{Z}}_m$ are all space forms that are not simply-connected.

Key theorem with proof [Intermediate+]

Theorem (Killing-Hopf classification). Let $M$ be a simply-connected complete Riemannian manifold of dimension $n\ge 2$ with constant sectional curvature $K$. Then $M$ is isometric to:

• ${\mathbb{R}}^n$ if $K=0$
• $S^n(1/\sqrt{K})$ if $K>0$
• ${\mathbb{H}}^n(1/\sqrt{|K|})$ if $K<0$

Every other space form is a quotient of one of these model spaces by a discrete group of isometries acting freely.

Proof. The proof has two parts: (1) showing that the model spaces have constant curvature and are simply-connected and complete, and (2) the uniqueness argument.

Part 1: Models. The flat metric on ${\mathbb{R}}^n$ has $R=0$ (all Christoffel symbols vanish in Cartesian coordinates), hence constant zero curvature. The sphere $S^n(R)\subset {\mathbb{R}}^{n+1}$ inherits a Riemannian metric with curvature $K=1/R^2$ (by the Gauss equation: the second fundamental form is $(1/R)g$, so $R(X,Y)Y=(1/R^2)(g(Y,Y)X-g(X,Y)Y)$). Hyperbolic space ${\mathbb{H}}^n(R)$ has curvature $K=-1/R^2$ by explicit computation in the Poincare ball or upper half-space model. All three are simply-connected and complete.

Part 2: Uniqueness. Let $M$ be a simply-connected complete Riemannian $n$-manifold with constant sectional curvature $K$. Fix a point $p\in M$ and an orthonormal frame $\{e_1,\dots ,e_n\}$ at $T_{p}M$. Choose the corresponding point $\tilde{p}$ and frame $\{{\tilde{e}}_i\}$ in the model space $\tilde{M}$ (the appropriate one of ${\mathbb{R}}^n$, $S^n$, or ${\mathbb{H}}^n$).

Define a map $F:M\to \tilde{M}$ by sending $q={\exp }_p(v)$ to $\tilde{q}={\exp }_{\tilde{p}}(d{F}_p(v))$, where $d{F}_p$ is the linear isometry $T_{p}M\to T_{\tilde{p}}\tilde{M}$ sending $e_i$ to ${\tilde{e}}_i$. Both ${\exp }_p$ and ${\exp }_{\tilde{p}}$ are defined on all of $T_{p}M$ and $T_{\tilde{p}}\tilde{M}$ respectively (by completeness and simply-connectedness, both are diffeomorphisms onto their images).

The map $F$ is well-defined: if ${\exp }_p(v_1)={\exp }_p(v_2)$ for distinct $v_1,v_2\in T_{p}M$, then $M$ has a conjugate point. But in constant curvature, the injectivity radius of ${\exp }_p$ depends only on $K$: it is $\pi /\sqrt{K}$ for $K>0$ and infinite for $K\le 0$. For $K\le 0$, ${\exp }_p$ is injective on all of $T_{p}M$ (Cartan-Hadamard). For $K>0$, since $M$ is simply-connected and complete with positive constant curvature, the Bonnet-Myers theorem gives $\mathrm{diam}(M)\le \pi /\sqrt{K}$, and $M$ is a sphere of radius $1/\sqrt{K}$, with ${\exp }_p$ injective on the open ball of radius $\pi /\sqrt{K}$.

To show $F$ is an isometry, compute the pushforward of tangent vectors along radial geodesics. At a point $q={\exp }_p(v)$, the tangent space $T_{q}M$ splits into the radial direction $\dot{\gamma }(1)$ (where $\gamma$ is the geodesic from $p$ to $q$) and the perpendicular subspace. In constant curvature $K$, the Jacobi fields perpendicular to $\gamma$ have the form $J(t)=f(t)E(t)$ where $E(t)$ is a parallel field perpendicular to $\dot{\gamma }$ and:

$$f(t)=\{\begin{array}{ll}t& K=0\\ \frac{\sin (t\sqrt{K})}{\sqrt{K}}& K>0\\ \frac{\sinh (t\sqrt{|K|})}{\sqrt{|K|}}& K<0\end{array}$$

This function $f$ is the same for $M$ and $\tilde{M}$ (both have constant curvature $K$). So $d{F}_q$ preserves the length of perpendicular vectors. It also preserves the radial direction by construction. Hence $d{F}_q$ is an isometry $T_{q}M\to T_{\tilde{q}}\tilde{M}$, and $F$ is a local isometry.

A local isometry between complete, simply-connected Riemannian manifolds is a covering map (by the path-lifting property for local isometries). Since both are simply-connected, the covering is one-sheeted, so $F$ is a diffeomorphism and hence a global isometry. $\square$

Bridge. This classification builds toward the Thurston geometrisation programme that classifies all 3-manifolds, and appears again in the theory of symmetric spaces 07.04.07 where constant-curvature spaces are the simplest examples. The foundational reason is that the Jacobi equation in constant curvature has a closed-form solution, and this is exactly the bridge between the local curvature data and the global isometry type. Putting these together with Schur's theorem from 03.02.05, any space where sectional curvature depends only on the point (not the 2-plane) automatically has globally constant curvature in dimension $n\ge 3$, so the Killing-Hopf classification applies to a broad class of manifolds.

Exercises [Intermediate+]

Advanced results [Master]

Theorem 1 (Cartan-Hadamard). If $(M,g)$ is a complete, simply-connected Riemannian manifold with non-positive sectional curvature, then ${\exp }_p:T_{p}M\to M$ is a diffeomorphism for every $p$. In particular, $M$ is diffeomorphic to ${\mathbb{R}}^n$.

Theorem 2 (Space forms and discrete groups). The space forms of constant curvature $K$ are classified by conjugacy classes of discrete subgroups $\Gamma$ of the isometry groups of the model spaces acting freely:

• Flat forms: $\Gamma \subset E(n)=O(n)⋉{\mathbb{R}}^n$ (Bieberbach groups)
• Spherical forms: $\Gamma \subset O(n+1)$, finite subgroups acting freely on $S^n$
• Hyperbolic forms: $\Gamma \subset O^{+}(n,1)$, discrete torsion-free subgroups

Theorem 3 (Bieberbach's theorems). For flat manifolds: (1) the fundamental group of a compact flat $n$-manifold is a torsion-free subgroup of $E(n)$ with a free abelian subgroup of rank $n$ and finite index; (2) there are only finitely many diffeomorphism classes of compact flat $n$-manifolds; (3) two compact flat manifolds are affinely equivalent if and only if their fundamental groups are isomorphic.

Theorem 4 (Thurston's eight model geometries). A model geometry in dimension 3 is a simply-connected complete Riemannian 3-manifold $X$ with a transitive action of a Lie group $G$ by isometries, such that $G$ is maximal with this property and $X$ admits a compact quotient. The eight model geometries are: $S^3$, ${\mathbb{R}}^3$, ${\mathbb{H}}^3$ (constant curvature), $S^2\times \mathbb{R}$, ${\mathbb{H}}^2\times \mathbb{R}$ (products), and $\widetilde{SL(2,\mathbb{R})}$, Nil, Sol (twisted products). Thurston's geometrisation conjecture (proved by Perelman, 2003) states that every closed orientable 3-manifold decomposes along spheres and tori into pieces each modelled on one of these eight geometries.

Theorem 5 (Preissmann's theorem). If $(M,g)$ is a compact Riemannian manifold with strictly negative sectional curvature, then every non-identity abelian subgroup of ${\pi }_1(M)$ is infinite cyclic. In particular, ${\pi }_1(M)$ has no subgroup isomorphic to $\mathbb{Z}\times \mathbb{Z}$.

Synthesis. The Killing-Hopf classification is the foundational reason that constant-curvature geometry reduces to the group theory of discrete subgroups of isometry groups. The central insight is that the model spaces are homogeneous and isotropic, so their quotients are determined by the group action, and this is exactly the bridge between local Riemannian geometry and global topology.

The pattern generalises from the three constant-curvature models to Thurston's eight geometries, where the additional five model spaces incorporate product and twisted-product structures. Putting these together with Preissmann's theorem, the fundamental group of a hyperbolic manifold is severely constrained, and this appears again in Mostow rigidity, which identifies hyperbolic manifolds of dimension $n\ge 3$ by their fundamental group alone.

Full proof set [Master]

Proposition 1 (Preissmann's theorem). Let $(M,g)$ be a compact Riemannian manifold with $K<0$ everywhere. Then any abelian subgroup of ${\pi }_1(M)$ is either the identity or infinite cyclic.

Proof. Let $\alpha ,\beta \in {\pi }_1(M)$ be commuting elements (so $\alpha \beta {\alpha }^{-1}{\beta }^{-1}=1$). Lift them to deck transformations $\tilde{\alpha },\tilde{\beta }$ of the universal cover $\tilde{M}$. The deck transformations are isometries of $\tilde{M}$.

Since $\tilde{\alpha }$ and $\tilde{\beta }$ commute, $\tilde{\alpha }$ maps the axis of $\tilde{\beta }$ to itself (the axis of an isometry of negative-curvature space is the unique geodesic preserved by it). In negative curvature, each isometry has at most one invariant geodesic: if $\tilde{\alpha }$ preserves two geodesics, it acts by translation on both, but in negative curvature, two distinct geodesics diverge, making this impossible unless $\tilde{\alpha }$ is the identity.

So both $\tilde{\alpha }$ and $\tilde{\beta }$ preserve the same axis $\gamma$. They both act by translation on $\gamma$. Two commuting translations of a line generate a cyclic group (the subgroup $\mathbb{Z}$ of $\mathbb{R}$ generated by their translation lengths). So $⟨\tilde{\alpha },\tilde{\beta }⟩\cong \mathbb{Z}$, which means $⟨\alpha ,\beta ⟩\cong \mathbb{Z}$ in ${\pi }_1(M)$. $\square$

Proposition 2 (Bieberbach: finiteness of flat space forms). In each dimension $n$, there are only finitely many compact flat manifolds up to affine equivalence.

Proof. By Bieberbach's first theorem, the fundamental group ${\pi }_1(M)$ of a compact flat $n$-manifold fits into $0\to {\mathbb{Z}}^n\to {\pi }_1(M)\to F\to 0$ where $F$ is a finite subgroup of $O(n)$. The holonomy group $F$ is a subgroup of $O(n)$ acting faithfully on ${\mathbb{Z}}^n$.

There are finitely many finite subgroups of $O(n)$ (up to conjugacy). For each such $F$, the extension ${\mathbb{Z}}^n\to {\pi }_1\to F$ is classified by $H^2(F,{\mathbb{Z}}^n)$, which is finite. So there are finitely many choices for ${\pi }_1$ up to isomorphism, hence finitely many flat space forms up to affine equivalence. $\square$

Connections [Master]

• Sectional and Ricci curvature 03.02.05. The Killing-Hopf classification applies to manifolds of constant sectional curvature, the simplest case of the curvature hierarchy developed in 03.02.05. Schur's theorem ensures that pointwise-constant sectional curvature implies global constancy in dimension $n\ge 3$, feeding directly into the Killing-Hopf theorem.

• Riemannian symmetric spaces 07.04.07. The three model geometries ($S^n$, ${\mathbb{R}}^n$, ${\mathbb{H}}^n$) are the simplest Riemannian symmetric spaces. They have maximum symmetry: the isometry group acts transitively on frames. The Killing-Hopf quotients are the locally symmetric spaces of constant curvature.

• Frobenius theorem and integrability 03.02.04. The Jacobi field computation in the Killing-Hopf proof uses the fact that the distribution of radial directions integrates to geodesic spheres, an instance of the Frobenius theorem applied to the exponential map's level sets. The integrability of the perpendicular distribution in constant curvature is what makes the classification work.

Historical & philosophical context [Master]

Killing studied the geometry of spaces of constant curvature in his 1891 paper (J. Reine Angew. Math. 109) ^{[Killing1891]}, extending earlier work of Clifford and Klein on the "Clifford-Klein space problem": classifying all manifolds of constant curvature. Killing's contribution was to formulate the problem in terms of discrete groups of isometries acting on the simply-connected model spaces.

Hopf resolved the simply-connected case in 1926 (Math. Ann. 97) ^[Hopf1926], proving that the simply-connected complete space forms are exactly ${\mathbb{R}}^n$, $S^n$, and ${\mathbb{H}}^n$. Wolf's monograph Spaces of Constant Curvature (first edition 1967, revised 2011) ^[Wolf2011] gives the definitive treatment. Thurston's geometrisation programme, announced in 1982 (Bull. AMS 6) ^{[Thurston1982]} and proved by Perelman in 2003 using Ricci flow, extends the classification from constant-curvature spaces to all 3-manifolds.

Bibliography [Master]

@article{Killing1891,
  author = {Killing, Wilhelm},
  title = {Ueber die {C}lifford-{K}leinschen Raumformen},
  journal = {J. Reine Angew. Math.},
  volume = {109},
  year = {1891},
  pages = {121--159},
}

@article{Hopf1926,
  author = {Hopf, Heinz},
  title = {Zum {C}lifford-{K}leinschen Raumproblem},
  journal = {Math. Ann.},
  volume = {97},
  year = {1926},
  pages = {151--184},
}

@article{Thurston1982,
  author = {Thurston, William P.},
  title = {Three-dimensional manifolds, {K}leinian groups and hyperbolic geometry},
  journal = {Bull. Amer. Math. Soc.},
  volume = {6},
  year = {1982},
  pages = {357--381},
}

@book{Wolf2011,
  author = {Wolf, Joseph A.},
  title = {Spaces of Constant Curvature},
  edition = {6},
  publisher = {AMS Chelsea},
  year = {2011},
}

@book{doCarmo1992,
  author = {do Carmo, Manfredo P.},
  title = {Riemannian Geometry},
  publisher = {Birkh\"{a}user},
  year = {1992},
}