One idea, three theorems: compare to a model
The previous four guides built every tool you now own: Guide 1 gave completeness and Hopf-Rinow, Guide 2 the Jacobi fields and conjugate points that measure how geodesics spread, Guide 3 the positive-curvature theorems (Bonnet-Myers, Synge), and Guide 4 the nonpositive side (Cartan-Hadamard). Every one of those results was secretly the same move: take a quantity on your manifold M, take the corresponding quantity on a constant-curvature space form of curvature kappa — the round sphere, flat space, or hyperbolic space — and prove an inequality between them. This final guide makes the comparison itself the hero.
Why a constant-curvature model is the right yardstick is no accident. On a space form everything is computable in closed form: geodesics, distances, the spreading of Jacobi fields, the area of triangles. So if you can bound the sectional curvature K of M between two constants, K_lower and K_upper, you can sandwich the genuinely unknown geometry of M between two known model geometries. The two theorems of this guide do exactly this at two scales. Rauch comparison works infinitesimally, controlling how fast a single Jacobi field grows along one geodesic. Toponogov comparison works globally, controlling the shape of an entire geodesic triangle.
Rauch: racing two Jacobi fields
Recall from Guide 2 that a Jacobi field J along a geodesic gamma measures the separation between gamma and an infinitesimally nearby geodesic; it obeys J'' + R(J, gamma') gamma' = 0, where the prime is covariant differentiation along gamma. The curvature term is the whole story: a large positive sectional curvature acts like a restoring force pulling neighboring geodesics back together, while negative curvature acts like an anti-spring pushing them apart. Rauch's theorem turns this differential equation into a clean inequality by comparing the size |J(t)| against the size of the model Jacobi field on a space form.
Here is the statement in its most useful form. Let gamma on M and a model geodesic gamma_kappa on the space form of curvature kappa both start at the same point, and let J, J_kappa be Jacobi fields vanishing at t = 0 with |J'(0)| = |J_kappa'(0)|. If every sectional curvature of M along gamma is at most kappa (so M curves LESS than the model), then |J(t)| is at least |J_kappa(t)| — the Jacobi field on M grows FASTER, because less focusing lets geodesics spread more freely. Reverse the curvature inequality and the conclusion reverses too. The proof is a slick application of the index form you met in Guide 2: compare the energy of J to that of J_kappa using the index-form inequality.
A worked sanity check makes Rauch click. On the unit sphere (kappa = 1), the model Jacobi field vanishing at 0 is J_1(t) = sin(t), which returns to zero at t = pi — that is exactly why antipodal points are conjugate at distance pi. On flat space (kappa = 0) the model field is J_0(t) = t, never returning. So if your manifold has K at most 1 everywhere, Rauch says its Jacobi fields grow at least as fast as sin(t), hence its first conjugate point along any geodesic is at distance at least pi. That single clause is the engine of the sphere theorem, and it is also a back-door proof of Bonnet-Myers: a lower curvature bound forces conjugate points to appear by a fixed distance, capping the diameter.
Toponogov: fat triangles, thin triangles
Rauch is infinitesimal — it lives on a single geodesic. Toponogov is its global, finite-scale partner, and it speaks the language of triangles. Take three points in M joined by minimizing geodesics to form a geodesic triangle with side lengths a, b, c and an angle alpha between two of the sides. Now build the comparison triangle in the model space form of curvature kappa: a triangle with the SAME three side lengths, and read off its corresponding angle alpha_kappa. Toponogov compares alpha to alpha_kappa.
The conclusion is a single inequality with a memorable picture. If every sectional curvature of M is at least kappa (a LOWER bound, the hypothesis that survives even when M is only an Alexandrov space with no smoothness), then the angles of your real triangle are at least as big as the model angles: alpha is at least alpha_kappa. In words: lower curvature bounds make triangles FATTER than the model. The dual hinge version compares the third side: fix two sides and the angle between them, and the opposite side in M is no longer than in the model. Fat triangles, short opposite sides — that is the whole content, and it generalizes the law of cosines you knew in flat trigonometry.
curvature LOWER bound K >= kappa ==> triangles are FAT
real angle alpha >= model angle alpha_kappa
real third side <= model third side (hinge form)
curvature UPPER bound K <= kappa ==> triangles are THIN
real angle alpha <= model angle alpha_kappa (CAT(kappa))
flat model (kappa = 0): alpha_0 from the ordinary law of cosines
c^2 = a^2 + b^2 - 2 a b cos(alpha_0)
Rauch = this same comparison, shrunk to one geodesic
|J(t)| vs |J_kappa(t)| = (1/sqrt k) sin(sqrt(k) t)The sphere theorem: pinching forces a sphere
Now the payoff. The sphere theorem is one of the crown jewels of the subject, and Rauch built the cage. The statement (the topological version, Berger-Klingenberg, 1960s): if M is a complete simply-connected manifold whose sectional curvature is pinched in the half-open band 1/4 < K is at most 1, then M is homeomorphic to the sphere S^n. The numbers are sharp and the band is the entire content — pinch your curvature tightly enough around a constant and the manifold has no choice but to be a topological sphere.
Why 1/4, and how does Rauch enter? The upper bound K is at most 1 lets Rauch guarantee no conjugate points before distance pi, so geodesics keep minimizing a long way. The lower bound K > 1/4 controls the injectivity radius — the cut locus cannot be closer than pi — via Klingenberg's estimate. Together these two distances trap the geometry so tightly that two metric balls around antipodal-like points cover M, and a ball is a disk; gluing two disks along their boundary sphere gives exactly S^n. The 1/4 is forced by a genuine example: the complex projective spaces CP^n carry a metric pinched in [1/4, 1] (closed band) that are NOT spheres, so the strict inequality cannot be relaxed.
The splitting theorem: a line that cuts the manifold
The last great theorem of the rung turns the comparison philosophy in a different direction. The Cheeger-Gromoll splitting theorem asks: what if M has nonnegative Ricci curvature AND contains a line — a geodesic gamma: R -> M that minimizes distance between ANY two of its points, all the way out to infinity in both directions? Such a line is a strong demand; on a sphere no geodesic minimizes past distance pi, so spheres have no lines. The theorem says that if a complete manifold of nonnegative Ricci does contain one line, then it splits as a metric product M = R x N, where R is the line's direction and N is a complete manifold of nonnegative Ricci one dimension lower.
The mechanism is beautiful and worth sketching, because it ties the whole rung together. For each end of the line build a Busemann function b, the renormalized limit of distance-to-a-faraway-point along the line; it measures 'how far along the line's direction' you are. Nonnegative Ricci makes each Busemann function superharmonic via the Laplacian comparison that Bishop-Gromov supplies, while the line being two-sided makes the sum of the two Busemann functions subharmonic. A function that is both super- and sub-harmonic is harmonic, the maximum principle forces it to be linear, and a globally linear function whose gradient is a parallel unit field is exactly a flat R-direction to peel off. That parallel splitting is the product structure.
State the hypothesis with care, because it is exactly the kind that is easy to drop. The splitting theorem needs nonnegative RICCI, not nonnegative sectional — Cheeger and Gromoll's achievement was precisely lowering the requirement from the sectional version (proved earlier by Toponogov) to the Ricci version. And it needs a genuine line, not merely a ray: a paraboloid has rays going to infinity but no two-sided line, and indeed it does not split. When the hypotheses do hold, the consequence is rigid and global — a single infinite minimizing geodesic, plus a curvature sign, dictates the topology of the entire space. That is the comparison philosophy of this rung at its most dramatic: local curvature, read through one geodesic, decides everything.