Skip to content

Singularities and Blow-up Limits

Summary

The Ricci flow breaks down in finite time as soon as curvature diverges somewhere. To understand how it breaks down, one zooms in on the singularity: Hamilton's blow-up procedure produces parabolically rescaled limit flows. Classifying these limit models — in particular the notion of a \(\kappa\)-solution — is the key to Perelman's surgery program.

1. When does the flow break down?

Short-time existence (see Article 03) gives a maximal existence interval \([0, T)\). The terminal time \(T\) is characterised by a curvature blow-up:

Lemma (Hamilton 1982). If \(T < \infty\) is the maximal existence time, then \(\displaystyle \limsup_{t \to T} \max_M \lvert\mathrm{Rm}(\cdot, t)\rvert = +\infty.\)

Equivalently: as long as the full Riemann curvature stays bounded the flow can be extended. A finite singular time is therefore always a curvature concentration point.

2. Type-I and Type-II singularities

Hamilton (1995) classified singularities by the growth rate of curvature relative to the remaining time \(T - t\):

Type Condition Model picture
I \(\sup_{M\times[0,T)} (T-t)\,\lvert\mathrm{Rm}\rvert < \infty\) round shrinking cylinder, \(S^3\) shrinker
II \(\sup (T-t)\,\lvert\mathrm{Rm}\rvert = \infty\) degenerate "cigar" and Bryant solitons
III \(\sup t\,\lvert\mathrm{Rm}\rvert < \infty\), \(T = \infty\) infinite time, hyperbolic pieces

In dimension three Perelman shows that all finite-time singularities are Type I or II with \(\kappa\)-solution model — Type III appears only after all surgeries, as long-time behaviour.

3. The neckpinch as model singularity

The prototypical example: a dumbbell-shaped \(S^3\) with a thin neck. Under the Ricci flow the neck shrinks faster than the bells; the metric converges locally to a round cylinder \(S^2 \times \mathbb{R}\) with vanishing \(S^2\)-radius (Angenent–Knopf 2004 give a rigorous construction).

In suitable coordinates the neck radius behaves as \(r(t) \sim \sqrt{2(T-t)}\) — the rescaled flow is a shrinking cylinder, i.e. a gradient shrinking soliton.

4. Parabolic rescaling (blow-up)

Hamilton's idea for "looking inside" a singularity: choose a sequence \((p_k, t_k)\) with \(t_k \to T\) and \(Q_k := \lvert\mathrm{Rm}(p_k, t_k)\rvert \to \infty\), and define the parabolically rescaled metrics

\[g_k(s) := Q_k \cdot g\!\left(t_k + \frac{s}{Q_k}\right),\qquad s \in [-Q_k\, t_k,\, 0].\]

This rescaling is the unique one that preserves the Ricci flow (Article 03, §5): if \(g\) satisfies \(\partial_t g = -2\,\mathrm{Ric}(g)\), so does each \(g_k\). Curvature at the base point is normalised to 1.

5. Hamilton's compactness theorem

To extract a limit flow from \((M, g_k(s), p_k)\) one needs two hypotheses:

  1. Curvature bounds on every backward time interval (uniform in \(k\)).
  2. A lower injectivity-radius bound \(\mathrm{inj}(p_k) \ge \iota > 0\) with respect to \(g_k\).

Theorem (Hamilton, Compactness, 1995). Under these hypotheses there is a subsequence converging in \(C^\infty_{\mathrm{loc}}\) (in the sense of pointed Cheeger–Gromov convergence) to a complete Ricci flow \((M_\infty, g_\infty(s), p_\infty)\) on an interval \((-\infty, 0]\).

This is the machine that turns a singularity into an "infinitely-old" limit flow — a so-called ancient solution.

6. Where the compactness theorem fails: collapse

Hypothesis 2 is the delicate one: without a lower injectivity-radius bound the limit can collapse (locally drop dimension). Example: a thin torus whose \(S^1\)-factor shrinks — only the two-dimensional factor survives the limit.

Perelman's breakthrough fixes precisely this gap: the \(\kappa\)-non-collapsing theorem (see Article 06) enforces a universal lower injectivity-radius bound along the singularity, and only then is Hamilton's compactness theorem applicable.

7. Ancient \(\kappa\)-solutions

The limit objects of the blow-up procedure are, in dimension three, extremely constrained:

Definition. An ancient \(\kappa\)-solution is a complete, non-flat Ricci flow \((M, g(s))\) on \((-\infty, 0]\) with non-negative curvature, bounded curvature on every compact time interval, and \(\kappa\)-non-collapsing.

Perelman (2002, §11) classifies all ancient \(\kappa\)-solutions in dim 3: they are quotients of the round sphere, the round cylinder \(S^2 \times \mathbb{R}\), or special Bryant solitons (rotationally symmetric, asymptotically cylindrical solitons).

Consequence: every singularity in dim 3 looks, up close, like one of these three models — the canonical neighbourhood on which Perelman's surgery construction is based.

8. Solitons: stationary model solutions

Self-similar solutions of the Ricci flow are called Ricci solitons:

\[\mathrm{Ric}(g) + \nabla^2 f = \lambda\, g,\qquad \lambda \in \{-1, 0, +1\}.\]
  • \(\lambda > 0\): shrinking soliton (model for Type-I singularities), e.g. the round sphere, the round cylinder, or the Gaussian soliton \((\mathbb{R}^n, g_{\text{flat}}, f = \lvert x \rvert^2/4)\).
  • \(\lambda = 0\): steady soliton, e.g. Hamilton's cigar soliton (model for Type-II in dim 2).
  • \(\lambda < 0\): expanding soliton, model for the resolution of a singularity after surgery.

Solitons form the phase space of the Ricci flow; Perelman's \(\mathcal{W}\)-functional (see Article 05) identifies shrinking solitons as critical points.

9. From model to surgery

The red thread leading to the proof of the geometrisation conjecture:

  1. Flow runs up to a singularity at \(t = T_1\).
  2. Blow-up at the singularity yields an ancient \(\kappa\)-solution.
  3. Classification implies: locally the picture is a sphere/quotient, a round neck, or a Bryant cap.
  4. Surgery: cut at a neck, glue in standard caps.
  5. The remaining manifold has simpler topology; the flow restarts.
  6. Iterate; Act 3 shows that this process terminates in finite time and reconstructs the topology.

Sources

  • Richard S. Hamilton, The formation of singularities in the Ricci flow, Surveys Diff. Geom. 2 (1995), 7–136.
  • Sigurd Angenent & Dan Knopf, An example of neckpinching for Ricci flow on \(S^{n+1}\), Math. Res. Lett. 11 (2004), 493–518.
  • Grigori Perelman, The entropy formula for the Ricci flow and its geometric applications, arXiv:math/0211159, §11.
  • John W. Morgan & Gang Tian, Ricci Flow and the Poincaré Conjecture, AMS (2007), §§9–12.
  • Bruce Kleiner & John Lott, Notes on Perelman's papers, Geom. Topol. 12 (2008), 2587–2855, §§37–43.

Cross-references