What it is
To take a limit of rescaled flows, Hamilton's compactness theorem needs curvature bounds and a lower bound on the injectivity radius. Cheeger, Gromov and Taylor showed that, with curvature bounded, an injectivity radius bound is equivalent to a lower bound on the volume of balls. A flow is κ-noncollapsed at scale ρ if every ball of radius r<ρ on which ∣Rm∣≤r−2 has volume at least κrn.
Perelman proved that any Ricci flow on a closed manifold is κ-noncollapsed on finite time intervals, with κ depending only on the initial metric and the time. The proof uses a lower bound on the W-entropy, and a second proof uses reduced volume. This rules out the cigar soliton, which is collapsed at large scales, as a blow-up limit. It was the missing ingredient in Hamilton's program.
Why it matters
Non-collapsing makes blow-up limits exist and be genuinely n-dimensional. That is why the singularity models in dimension three are κ-solutions and not collapsed objects. Collapsing matters too, in the long-time behaviour. As t→∞ the thick part of a 3-manifold becomes hyperbolic, while the thin part collapses with bounded curvature. Collapsing theory (Cheeger–Gromov, Shioya–Yamaguchi, Kleiner–Lott, Morgan–Tian) shows that the thin part is a graph manifold, which completes geometrization.
Where it stands
The non-collapsed theory now works in all dimensions (Chen–Wang; Bamler). Collapsed Ricci flows are much less understood: long-time behaviour of immortal flows (Lott's dimensional reduction), collapsed ancient solutions and steady solitons, and flows with only local curvature control. The thick–thin decomposition in Bamler's theory generalizes Perelman's long-time picture to higher dimensions.