the Cheeger isoperimetric constant
/ CHAY-gur /
Suppose you want to cut a manifold into two pieces. Some shapes are easy to cut with a small slice — a dumbbell pinched at a thin neck snaps apart cheaply — while others, like a round sphere, force you to make a big cut no matter how you try. The Cheeger constant measures exactly this: the cheapest possible cut, the smallest boundary area you need per unit of the smaller volume you separate off. A small Cheeger constant means a bottleneck; a large one means the space is robustly connected.
Precisely, for a compact Riemannian manifold M the Cheeger isoperimetric constant is h(M) = inf over hypersurfaces S that divide M into two pieces M_1 and M_2 of Area(S) / min(Vol(M_1), Vol(M_2)). You minimize the area of the cutting surface relative to the volume of the smaller side; the infimum is taken over all ways of slicing M in two. A thin neck gives a tiny numerator (small cut) with a substantial denominator, hence small h; a space with no bottlenecks keeps h bounded away from zero. The reason this combinatorial-looking quantity belongs to spectral geometry is Cheeger's inequality: the first nonzero Laplacian eigenvalue is bounded below by lambda_1 >= h(M)^2 / 4. So a hard-to-cut manifold necessarily vibrates at a high fundamental frequency, and conversely a small lambda_1 forces a bottleneck — analysis and geometry locked together. Buser's inequality provides a partial converse (an upper bound on lambda_1 in terms of h and a Ricci lower bound), showing the two quantities are genuinely comparable under curvature control.
The Cheeger constant is a cornerstone linking eigenvalues, isoperimetry, and connectivity, and its discrete analogue on graphs (the conductance / Cheeger constant of a graph) is fundamental in spectral graph theory, expander constructions, and the analysis of mixing times for random walks. The honest caveats: Cheeger's inequality is one-directional — a large h forces a large lambda_1, but a small h does not by itself force a small lambda_1 without further hypotheses, which is exactly why Buser's converse needs a Ricci bound. And the constant is genuinely a global, often hard-to-compute quantity (it is an infimum over all bisecting hypersurfaces), so in practice it is estimated, not evaluated; do not mistake the clean inequality for an easy formula.
A dumbbell — two round balls joined by a thin cylindrical neck of radius epsilon — can be cut across the neck with area proportional to epsilon^{n-1} while separating half the volume, so h is tiny; Cheeger's inequality then predicts a tiny lambda_1, and indeed the dumbbell has a near-zero fundamental tone.
The dumbbell's thin neck gives a small Cheeger constant and hence a small first eigenvalue.
Cheeger's inequality lambda_1 >= h^2/4 only bounds the eigenvalue from below; the reverse direction (Buser) requires a Ricci lower bound, so a small Cheeger constant alone does not, without curvature control, force a small eigenvalue.