Comparison theorems in Riemannian geometry — Cartan-Hadamard, Bonnet-Myers, Bishop-Gromov

Intuition Beginner

Curvature is the bending of a space. A surface of positive curvature, like a sphere, bends inward: geodesics that start out moving apart are pulled back toward each other and meet again on the far side. A surface of negative curvature, like a saddle or a Pringles chip, bends the other way: geodesics that start near each other are driven apart, and the space spreads out forever.

Comparison theorems turn this picture into sharp mathematics. Their shared promise is: tell me the curvature, and I will bound the size of the space, the growth of its volume, and the shape of its topology. Curvature is a local measurement made at each point; the theorems reach all the way out to global facts. That leap from local to global is the whole point of the subject.

Three results anchor the theory. Cartan-Hadamard says that negative curvature opens the space out: its universal cover looks like flat space and cannot fold back on itself. Bonnet-Myers says that positive curvature pinches the space shut: a complete manifold of uniformly positive curvature has finite diameter and only finitely many topological loops. Bishop-Gromov caps how fast volume can grow under a lower bound on a curvature average called the Ricci curvature.

Visual Beginner

Two surfaces side by side: a sphere (positive curvature) whose geodesics converge and close into a finite ball, and a hyperbolic saddle (negative curvature) whose geodesics fan apart without end.

The sphere has a largest possible distance (its diameter); the saddle does not. The sign of curvature decides which picture you are in.

Worked example Beginner

Take a round sphere of radius $R=2$. Its sectional curvature is $1/R^2=1/4$. The Bonnet-Myers theorem promises an absolute ceiling on how far apart two points can be: the diameter is at most $\pi$ divided by the square root of the curvature.

Step 1. Compute the bound. The square root of the curvature is $\sqrt{1/4}=1/2$, so the bound is $\pi /(1/2)=2\pi$, which is about $6.28$.

Step 2. Check it against the real sphere. The longest walk on a sphere of radius $2$ is half of a great circle, namely $\pi R=2\pi$, again about $6.28$. The bound is met exactly.

Step 3. Read off the topology. The sphere is simply connected: every loop on it can be shrunk to a point. So its fundamental group has a single element, which is a finite group. Bonnet-Myers predicts a finite fundamental group, and one element counts as finite.

The lesson: a single number, the curvature $1/R^2$, fixed both the largest possible distance and the finiteness of the loop group.

Check your understanding Beginner

Formal definition Intermediate+

Let $(M,g)$ be a complete Riemannian manifold ^{[Lee Ch. 11]}. A Jacobi field along a geodesic $\gamma$ is a vector field $J$ along $\gamma$ solving the Jacobi equation

$${\nabla }_{\dot{\gamma }}{\nabla }_{\dot{\gamma }}J+R(J,\dot{\gamma })\dot{\gamma }=0.$$

Jacobi fields are the infinitesimal spread of families of geodesics: if ${\gamma }_s(t)$ is a variation of $\gamma$ through geodesics, then ${\partial }_t{\partial }_s{\gamma }_s{|}_{s=0}$ is a Jacobi field. They are the engine of every comparison theorem.

A point $q=\gamma (t_0)$ is conjugate to $p=\gamma (0)$ along $\gamma$ when there is a nonzero Jacobi field with $J(0)=0$ and $J(t_0)=0$; equivalently $d({\exp }_p{)}_{t_0\dot{\gamma }(0)}$ is singular. The index form on vector fields $V$ along $\gamma$ vanishing at the endpoints is

$$I(V,V)={\int }_0^L(|{\nabla }_{\dot{\gamma }}V{|}^2-⟨R(V,\dot{\gamma })\dot{\gamma },V⟩)dt.$$

A geodesic minimises length on $[0,L]$ only if $I(V,V)\ge 0$ for every such $V$, so a negative index forces a shorter curve to exist.

The model space $M_k^n$ is the unique simply connected complete $n$-manifold of constant sectional curvature $k$: Euclidean space for $k=0$, the round sphere of radius $1/\sqrt{k}$ for $k>0$, and hyperbolic space for $k<0$. Write $S_k(r)$ for the geometric factor governing the area of geodesic spheres in $M_k^n$: $S_k(r)=\sin (\sqrt{k}r)/\sqrt{k}$ for $k>0$, $S_0(r)=r$, and $S_k(r)=\sinh (\sqrt{-k}r)/\sqrt{-k}$ for $k<0$. Volume growth in the model is governed by $S_k(r{)}^{n-1}$.

Counterexamples to common slips

• Curvature sign is about planes, not points. Sectional curvature is a function of tangent 2-planes; the Ricci curvature is its average over planes containing a given direction.
• A lower Ricci bound is weaker than a lower sectional bound. Bishop-Gromov needs only Ricci; Bonnet-Myers in its modern form also needs only Ricci, which is why both are stated that way below.
• Conjugate points are not cut points. A cut point can occur without conjugacy (think two geodesics of equal length meeting); comparison theorems treat them separately.

Key theorem with proof Intermediate+

Theorem (Cartan-Hadamard). Let $(M,g)$ be a complete connected Riemannian manifold with sectional curvature $K\le 0$. Then for every $p\in M$ the exponential map ${\exp }_p:T_{p}M\to M$ is a covering map. In particular the universal cover of $M$ is diffeomorphic to ${\mathbb{R}}^n$, so ${\pi }_1(M)$ is torsion-free and $M$ has no conjugate points ^{[do Carmo Ch. 7]}.

Proof. Fix $p$ and let $J$ be a Jacobi field along a geodesic $\gamma$ with $J(0)=0$. The Jacobi equation gives $⟨J^{\prime \prime }(t),J(t)⟩=-⟨R(J,\dot{\gamma })\dot{\gamma },J⟩=-K(\sigma )|J\wedge \dot{\gamma }{|}^2$. Since $K\le 0$ this term is nonnegative, and

$$\frac{d^2}{d{t}^2}\frac{1}{2}|J{|}^2=|{\nabla }_{\dot{\gamma }}J{|}^2+⟨J^{\prime \prime },J⟩\ge |{\nabla }_{\dot{\gamma }}J{|}^2\ge 0.$$

So $f(t)=\frac{1}{2}|J(t){|}^2$ is convex. Because $J(0)=0$ we have $f(0)=0$ and $f^{\prime }(0)=⟨J(0),{\nabla }_{\dot{\gamma }}J(0)⟩=0$. A convex function on $[0,\infty )$ with value and slope zero at the origin that returns to zero at some $t_0>0$ must vanish identically on $[0,t_0]$, forcing $J\equiv 0$ there and, by uniqueness of the Jacobi ODE, everywhere. Hence no point is conjugate to $p$ along any geodesic, so $d({\exp }_p)$ is everywhere nonsingular and ${\exp }_p$ is a local diffeomorphism.

Completeness plus the Hopf-Rinow theorem makes ${\exp }_p$ surjective: every point of $M$ lies on a geodesic ray from $p$. Pull the metric back through ${\exp }_p$ so that it becomes a local isometry from the complete manifold $T_{p}M\cong {\mathbb{R}}^n$ onto $M$. The Hadamard covering lemma now applies: a surjective local isometry out of a complete connected Riemannian manifold is a covering map, because geodesics lift uniquely and their lifts build evenly covered neighbourhoods around every point. Therefore ${\exp }_p$ is a covering map, the universal cover is ${\mathbb{R}}^n$, and the deck group acts freely on a contractible space, which forces ${\pi }_1(M)$ to be torsion-free. $\square$

Bridge. Cartan-Hadamard builds toward 03.09.10 (the index theorem), where the contractibility of the universal cover under nonpositive curvature clears negative-curvature obstructions to the Bochner vanishing technique, and appears again in 03.06.04 (characteristic classes), where the topology pinned down by the curvature sign feeds the Chern-Weil invariants built from the Levi-Civita curvature 2-form of 48.01.01. The foundational reason the sign of curvature controls global topology is that Jacobi fields — the infinitesimal spread of geodesics — obey a second-order equation whose restoring or repelling force is exactly sectional curvature; this is the central insight that the derivative of the exponential map is governed by solutions of the Jacobi equation, and putting these together, the bridge is that local curvature data propagates through the index form into global covering-space, diameter, and volume information, a pattern that generalises to the Ricci-level Bishop-Gromov comparison below.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none records that Mathlib's Riemannian infrastructure, while substantial, has not yet absorbed the global comparison machinery. The natural first formalisation target is the no-conjugate-points lemma under $K\le 0$ (a convexity argument on $|J{|}^2$), which sits just above the existing Jacobi-field API; Cartan-Hadamard as a covering statement, Bonnet-Myers via the index form, and Bishop-Gromov monotonicity await upstream work documented in the unit metadata gap note.

Advanced results Master

Where Cartan-Hadamard governs the negative-curvature regime, two further theorems govern the positive and mixed regimes and together with it complete the programme that curvature controls global topology.

Theorem (Bonnet-Myers). Let $(M,g)$ be a complete $n$-manifold with $\mathrm{Ric}\ge (n-1)k$ for some $k>0$. Then $M$ is compact, $\mathrm{diam}(M)\le \pi /\sqrt{k}$, and ${\pi }_1(M)$ is finite ^{[Petersen Ch. 9]}. The round sphere $S^n(1/\sqrt{k})$ shows the bound is sharp: its diameter is exactly $\pi /\sqrt{k}$. The proof, given in full below, runs the index form against the test field $\sin (\pi t/L)$ along a presumed-minimising geodesic; if $L>\pi /\sqrt{k}$ the index goes negative, contradicting minimality.

Theorem (Bishop-Gromov volume comparison). Let $(M,g)$ be complete with $\mathrm{Ric}\ge (n-1)k$. For any $p\in M$ the ratio $\mathrm{Vol}(B(p,r))/V_k^n(r)$ is monotone nonincreasing in $r$; in particular $\mathrm{Vol}(B(p,r))\le V_k^n(r)$ for all $r$, with equality to first order at $r=0$ ^{[Bishop 1963]}. The model volume $V_k^n(r)$ is the volume of a radius-$r$ ball in $M_k^n$. This converts a lower Ricci bound into an absolute ceiling on volume growth, the input to Gromov's precompactness theorem and to almost-rigidity results.

Two rigidity refinements sharpen these into classification statements. Cheng's maximal diameter theorem says equality $\mathrm{diam}=\pi /\sqrt{k}$ in the Bonnet-Myers regime forces $M$ to be isometric to $S^n(1/\sqrt{k})$. The Bishop rigidity case says that if $\mathrm{Vol}(B(p,r_0))=V_k^n(r_0)$ for some $r_0$ (not merely a ratio limit), then a ball in $M$ is isometric to a ball in $M_k^n$. On the negative side, Preissman's theorem states that on a compact manifold with $K<0$, every element of ${\pi }_1(M)$ other than the identity generates an infinite cyclic subgroup — negative curvature forbids the kind of finite loop group that positive curvature forces.

Synthesis. The three comparison theorems together realise the programme that curvature controls global topology: Cartan-Hadamard builds toward the rigidity of nonpositive curvature and appears again in 03.09.10, where flat and negative-curvature hypotheses feed index-theoretic vanishing results; Bonnet-Myers shows that positive sectional (indeed Ricci) curvature closes space up — the central insight that a sphere's finite diameter is forced by its curvature — and the foundational reason is that Jacobi fields refocus at distance $\pi /\sqrt{k}$; Bishop-Gromov is exactly the volume-level refinement, bounding geodesic-ball growth by the constant-curvature model, and is dual to the Bonnet-Myers diameter bound in that Ricci curvature replaces the index form as the controlling quantity; putting these together, the bridge is a single Jacobi-field comparison engine that propagates local curvature data into global metric and topological rigidity, and the pattern generalises to the Cheeger finiteness theorem, the splitting theorem, and the Lorentzian comparisons of 13.03.01.

Full proof set Master

Proposition (Bonnet-Myers via the index form). Let $(M,g)$ be a complete $n$-manifold with $\mathrm{Ric}\ge (n-1)k$ for $k>0$. Then $\mathrm{diam}(M)\le \pi /\sqrt{k}$, $M$ is compact, and ${\pi }_1(M)$ is finite.

Proof. Let $\gamma :[0,L]\to M$ be a unit-speed minimising geodesic and suppose for contradiction that $L>\pi /\sqrt{k}$. Choose parallel orthonormal fields $E_2,\dots ,E_n$ along $\gamma$ perpendicular to $\dot{\gamma }$, and set $V_i(t)=\sin (\pi t/L)E_i(t)$. Each $V_i$ vanishes at the endpoints, so the index form applies:

$$I(V_i,V_i)={\int }_0^L\left(\right(\frac{\pi }{L}{)}^2{\cos }^2\frac{\pi t}{L}-{\sin }^2\frac{\pi t}{L}⟨R(E_i,\dot{\gamma })\dot{\gamma },E_i⟩)dt.$$

Summing over $i=2,\dots ,n$ and using ${\sum }_i⟨R(E_i,\dot{\gamma })\dot{\gamma },E_i⟩=\mathrm{Ric}(\dot{\gamma },\dot{\gamma })\ge (n-1)k$ gives

$$\sum _{i=2}^{n}I(V_i,V_i)\le (n-1)\left[\right(\frac{\pi }{L}{)}^2-k]{\int }_0^L{\sin }^2\frac{\pi t}{L}dt=(n-1)(\frac{{\pi }^2}{L^2}-k)\cdot \frac{L}{2}.$$

The hypothesis $L>\pi /\sqrt{k}$ makes the right-hand side strictly negative, so some $I(V_i,V_i)<0$. A negative index certifies a shorter curve between the endpoints, contradicting that $\gamma$ minimises. Hence every minimising geodesic has length $\le \pi /\sqrt{k}$, so $\mathrm{diam}(M)\le \pi /\sqrt{k}$.

Completeness plus a finite diameter bound forces compactness by the Hopf-Rinow theorem (closed bounded sets are compact). For the fundamental group, lift $g$ and the Ricci bound to the universal cover $\tilde{M}$; the inequality is local and survives lifting. Bonnet-Myers applied on the cover gives $\mathrm{diam}(\tilde{M})\le \pi /\sqrt{k}$, so $\tilde{M}$ is compact too. The deck group ${\pi }_1(M)$ acts freely and properly discontinuously by isometries on a compact space, so each orbit is finite; the fibre over any point is one such orbit, so ${\pi }_1(M)$ is finite. $\square$

Proposition (Bishop-Gromov monotonicity, sketch). Let $(M,g)$ be complete with $\mathrm{Ric}\ge (n-1)k$. In geodesic polar coordinates at $p$, write the volume density as $A(p,r,\theta )drd\theta$ on the unit sphere. The Laplacian of the distance function satisfies the comparison $\Delta r\le (n-1)S_k^{\prime }(r)/S_k(r)$ on the set where $r$ is smooth, because the Riccati equation for the shape operator of geodesic spheres, driven by the Ricci lower bound, yields $A^{\prime }(r,\theta )/A(r,\theta )\le (n-1)S_k^{\prime }(r)/S_k(r)$.

Integrating this differential inequality against $S_k(r{)}^{n-1}$ and comparing $A$ with $S_k(r{)}^{n-1}$ pointwise gives

$$\frac{A(p,r,\theta )}{S_k(r{)}^{n-1}}\text{is nonincreasing in }r,$$

and integrating once more over $\theta$ yields the monotonicity of the ratio $\mathrm{Vol}(B(p,r))/V_k^n(r)$. The limit of that ratio as $r\to 0$ is $1$ by the Euclidean blow-up of the metric, so the ratio never exceeds $1$, giving $\mathrm{Vol}(B(p,r))\le V_k^n(r)$. Equality at a single positive radius forces $A=S_k(r{)}^{n-1}$ identically on a ball, which is the Bishop rigidity case. $\square$

Connections Master

• Foundational Riemannian substrate 48.01.01. This unit presupposes the Levi-Civita connection, the curvature tensor, and the geodesic machinery developed in the foundational unit; every comparison theorem here is a statement about how those local objects control global geometry, so 48.01.01 is the indispensable prerequisite feeding the Jacobi-field and index-form arguments above.

• Index theory and vanishing theorems 03.09.10. Cartan-Hadamard and Bishop-Gromov supply the curvature conditions under which the Bochner technique kills cohomology: nonpositive curvature or positive Ricci curvature force harmonic forms to vanish in degrees that the Weitzenböck formula controls, which is the analytic input to the Atiyah-Singer index theorem's topological conclusions.

• Characteristic classes via Chern-Weil 03.06.04. The curvature 2-form of the Levi-Civita connection, the same object whose sign the comparison theorems bound, feeds the invariant polynomials producing Pontryagin and Euler classes; the global topology that Cartan-Hadamard and Bonnet-Myers pin down is exactly what those characteristic classes measure.

• Lorentzian and relativistic analogues 13.03.01. The comparison programme lifts to pseudo-Riemannian metrics in general relativity, where Raychaudhuri's equation — the Lorentzian cousin of the Bishop Riccati inequality — drives the singularity theorems by showing that positive energy density focuses geodesics.

Historical & philosophical context Master

The programme of reading global geometry off local curvature begins with Élie Cartan, whose 1928 Leçons sur la géométrie des espaces de Riemann established that a complete simply connected surface of nonpositive curvature is diffeomorphic to the plane, the seed of what is now called the Cartan-Hadamard theorem ^{[Cartan 1928]}. Cartan's insight reframed Riemann's local metric as a lever for global classification: the sign of curvature, a quantity measured infinitesimally, decides whether the space folds shut or unfolds forever.

The positive-curvature complement came from S. S. Myers in 1941, who proved in the Duke Mathematical Journal that a complete Riemannian manifold whose Ricci curvature is bounded below by a positive constant has finite diameter and finite fundamental group ^{[Myers 1941]}. Myers's use of the second-variation formula and parallel orthonormal frames — the index-form argument reproduced above — set the template for a century of geometric analysis. The volume-level refinement is due to Richard Bishop in 1963 (and in published form with Richard Crittenden), who showed that a Ricci lower bound alone forces the geodesic-ball volume ratio against a model space to be monotone ^{[Bishop 1963]}; Mikhail Gromov's 1981 Structures métriques pour les variétés riemanniennes then turned Bishop's inequality into a precompactness principle, founding modern metric geometry ^{[Gromov 1981]}. The recurring philosophy — local curvature controls global topology — is the load-bearing idea of 48.01.02 and of comparison geometry as a whole.

Bibliography Master

@book{Cartan1928,
  author = {Cartan, {\'E}lie},
  title = {Le{\c c}ons sur la g{\'e}om{\'e}trie des espaces de Riemann},
  publisher = {Gauthier-Villars},
  year = {1928},
}

@article{Myers1941,
  author = {Myers, Sumner B.},
  title = {Riemannian manifolds with positive mean curvature},
  journal = {Duke Mathematical Journal},
  volume = {8},
  number = {2},
  pages = {401--404},
  year = {1941},
}

@article{Bishop1963,
  author = {Bishop, Richard L.},
  title = {A relation between volume, mean curvature, and diameter},
  journal = {Notices of the American Mathematical Society},
  volume = {10},
  pages = {354},
  year = {1963},
}

@book{BishopCrittenden1964,
  author = {Bishop, Richard L. and Crittenden, Richard J.},
  title = {Geometry of Manifolds},
  publisher = {Academic Press},
  year = {1964},
}

@book{Gromov1981,
  author = {Gromov, Mikhail},
  title = {Structures m{\'e}triques pour les vari{\'e}t{\'e}s riemanniennes},
  publisher = {Cedic/Fernand Nathan},
  year = {1981},
}

@book{CheegerEbin1975,
  author = {Cheeger, Jeff and Ebin, David G.},
  title = {Comparison Theorems in Riemannian Geometry},
  publisher = {North-Holland},
  year = {1975},
}

@book{Petersen2016,
  author = {Petersen, Peter},
  title = {Riemannian Geometry},
  edition = {3},
  publisher = {Springer GTM 171},
  year = {2016},
}

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

@book{Lee1997,
  author = {Lee, Jeffrey M.},
  title = {Riemannian Manifolds: An Introduction to Curvature},
  publisher = {Springer GTM 176},
  year = {1997},
}