Can scalar curvature detect a Ricci-flow singularity?

← All digests

Family 351 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

Ricci flow changes a Riemannian metric according to its curvature. It is often compared to a heat equation for geometry: irregularities tend to smooth out, but some regions can also collapse and develop singularities.

Several levels of curvature information are available. The full curvature tensor records all directional bending. Ricci curvature averages some of that information, and scalar curvature is a further trace. If the scalar curvature stays bounded, must the whole flow remain smooth?

Family 351 claims that the answer depends on dimension. One manuscript proves extension on every closed four-manifold. Another constructs closed flows in sufficiently high dimension whose scalar curvature stays uniformly bounded while the full curvature diverges at a finite time. A third provides the static analytic estimate used by the four-dimensional proof.

These conclusions concern smooth continuation on the original manifold. Continuing from a singular limit by an orbifold flow is a different result and does not by itself settle this extension question.

2. Background and essential definitions

The equation and its curvature tensors

A Ricci flow is a family of metrics satisfying

$$\frac{\partial g}{\partial t}=-2\operatorname{Ric}_{g(t)}.$$

Write $\operatorname{Rm}$ for the full curvature tensor, $\operatorname{Ric}$ for its Ricci contraction, and

$$R=\operatorname{tr}_g\operatorname{Ric}$$

for scalar curvature. A bound on a tensor controls its trace, but a bound on the trace need not control the tensor. Ricci-flat geometry can still have nonzero full curvature.

On a closed manifold—compact without boundary—a smooth flow extends while full curvature stays uniformly bounded. Its finite maximal time $T$ is the endpoint beyond which no smooth continuation agreeing with the original flow exists.

Scalar curvature satisfies

$$(\partial_t-\Delta)R=2|\operatorname{Ric}|^2.$$

It also controls volume change through $\partial_t d\mu=-R\,d\mu$. These identities make bounded scalar curvature powerful, but turning that information into full regularity is the central issue.

Rates and geometric scales

The usual parabolic singularity scale is curvature of order $(T-t)^{-1}$, corresponding to lengths of order $\sqrt{T-t}$. A finite-time singularity is Type I if a uniform bound of that order holds. It is Type II if

$$\limsup_{t\uparrow T}(T-t)\max_M|\operatorname{Rm}|=\infty.$$

This is a limsup condition, not necessarily a monotone limit. The examples in M1 have stronger, two-sided power-law estimates.

When curvature concentrates, magnifying different regions can reveal several nested scales. In dimension four, such limits may be Ricci-flat spaces that approach a Euclidean quotient at infinity. These are called asymptotically locally Euclidean, or ALE, spaces. Several nested limits can be organized as a finite bubble tree, with large pieces containing smaller concentration regions.

3. Development history and previous partial results

Hamilton established Ricci flow and its basic existence and continuation theory. Šešum proved that a uniform bound on the Ricci tensor already suffices for extension, even without initially bounding the full curvature tensor. The scalar question asks whether one can replace that tensor bound by a bound on its trace. Šešum.

Bamler–Zhang developed heat-kernel, curvature, and compactness estimates under scalar bounds. In dimension four, their theory gives curvature-energy control and orbifold limits, as well as extension under additional topological assumptions. Bamler’s later convergence theory permits singular sets of codimension at least four. These results constrain possible degeneration but do not automatically remove every singular point from the original manifold. Bamler–Zhang, Bamler.

Positive extension or scalar-blowup conclusions were also known in special settings, including closed Kähler–Ricci flow and Type I singularities. Local Type I results address specified concentration regimes. M2 claims to remove such rate restrictions in real dimension four.

The counterexample direction builds on symmetric singularity constructions. Angenent–Knopf produced neckpinches, Gu–Zhu constructed Type II singularities, and Angenent–Isenberg–Knopf obtained degenerate neckpinches with prescribed rates. These show rapid full-curvature concentration, but do not by themselves give bounded scalar curvature. Degenerate neckpinches.

Stolarski constructed closed doubly warped flows with conical singularities and arbitrarily fast blowup, approaching a Ricci-flat cone at the parabolic scale. He identified bounded scalar curvature as a possible feature. M1 imports quantitative profile estimates from that construction and claims the additional scalar bound for suitable large dimensions and mode indices. Stolarski’s source version.

The four-dimensional static companion has a separate ancestry in Einstein desingularization and analytic inequalities near Ricci-flat ALE metrics. It develops an estimate allowing several neck lengths to diverge simultaneously, rather than staying near one fixed smooth ALE background.

4. What the new manuscripts claim

High-dimensional failure: M1

For some sufficiently large integer $q\geq10$, M1 constructs a smooth Ricci flow on

$$M=S^2\times S^{q+1}$$

with finite maximal time $T$, such that

$$\sup_{M\times[0,T)}|R|<\infty, \qquad \lim_{t\uparrow T}\max_M|\operatorname{Rm}|=\infty.$$

The real dimension is $q+3$. The theorem does not claim that the construction works already at the smallest displayed lower bound on $q$.

More precisely, in one sufficiently high fixed dimension, the examples have an unbounded discrete collection of exponents $\alpha$ with

$$c(T-t)^{-\alpha} \leq\max_M|\operatorname{Rm}| \leq C(T-t)^{-\alpha}$$

for all sufficiently late times. The flow and constants may vary with the exponent. This is not an assertion that every real exponent is realizable.

Four-dimensional extension: M2

Let $g(t)$, $0\leq t<T<\infty$, be a smooth Ricci flow on a closed smooth real four-manifold. M2 claims that a uniform absolute scalar bound implies a smooth extension on the same manifold to $T+\epsilon$, agreeing with the original flow at every earlier time.

No orientation assumption, curvature-rate restriction, or additional topology is imposed. Disconnected closed manifolds are handled componentwise. Consequently, at a finite maximal singular time in dimension four,

$$\limsup_{t\uparrow T}\max_M R(\cdot,t)=+\infty.$$

The conclusion is stated as a limsup. The scalar evolution equation supplies a lower bound on finite intervals, so an upper bound would be enough to contradict maximality.

The static estimate: M3

For metrics degenerating along one fixed finite tree of four-dimensional Ricci-flat pieces, M3 proves a sequential inequality

$$\mathcal E=o(\mathcal M).$$

Here $\mathcal E$ is a renormalized Einstein–Hilbert functional in fixed root units, and $\mathcal M$ is a scale-neutral weighted Ricci error controlling ten scaled derivatives. The end geometry has uniform bounds through twenty derivatives; the root end decays faster than inverse radius, and joined quotient groups are nontrivial.

The notation means $\mathcal E/\mathcal M\to0$ as the error tends to zero and all joining lengths diverge under the stated hypotheses. It is not a universal estimate from a small unweighted Ricci norm on arbitrary spaces. M2 uses this companion and supplies the flow-specific constructions needed to apply it.

5. Why the advance matters

Importance: 5/5. If correct, the family would separate the four-dimensional extension theorem from a false unrestricted all-dimensional statement. It would identify a genuine limitation of scalar curvature as a monitor of geometric singularity.

The high-dimensional examples are particularly strong because they keep scalar curvature bounded while allowing arbitrarily fast power-law full-curvature blowup in one fixed dimension. Their two-sided estimates describe the scale of the singularity, not just an isolated lower bound along a sequence.

The four-dimensional result would close the gap between compactness with possible orbifold points and smooth continuation on the original manifold. Its need to control nested Ricci-flat bubbles explains why a scalar estimate alone is not a short route to extension.

The rating reflects the stated scope and consequences, not verification of the constructions or analytic inequalities.

6. The main idea

Separate the cone scale from the cap scale

M1 uses metrics of the form

$$g=ds^2+\phi(s,t)^2g_{S^2}+r(s,t)^2g_{S^q}$$

away from two pole orbits. The sphere radii vary along the interval coordinate $s$. At each pole, the $S^q$ factor collapses smoothly, leaving a positive-radius two-sphere that shrinks as the singular time approaches.

There are two relevant length scales. The larger is $\sqrt{T-t}$, where the geometry approaches a Ricci-flat cone. The smaller cap scale is $(T-t)^\sigma$, with $\sigma>1/2$. Near-flat Ricci behavior on the larger annuli does not control the smaller smooth cap automatically.

The paper uses one-sided cone comparisons and a slope estimate to bound the cap, then curvature point selection to control full curvature without assuming convergence to a particular complete cap metric. Dimension-explicit parabolic estimates yield a Ricci bound of the form $|\operatorname{Ric}|\leq Cr^{-e}$ with $0<e<1$.

The square of this estimate can be dominated in the scalar evolution equation by a bounded supersolution. Full curvature still diverges because a tangent plane to the shrinking pole two-sphere has curvature $\phi(0,t)^{-2}$. Thus the scalar control and the full-curvature blowup are obtained from separate estimates at the two scales.

In four dimensions, isolate the outermost bubble

Under a hypothetical bounded-scalar singularity, M2 extracts a finite tree of Ricci-flat pieces from spatial rescalings. It tracks the outermost nonflat piece at a scale $a(t)$, while permitting arbitrarily smaller descendants.

Attach an ALE exterior beyond a collar of radius $Na$, where $N$ is large and fixed. The renormalized functional combines an integral of scalar curvature with a boundary-flux subtraction. The two terms may diverge separately; the subtraction is what makes their combined limit meaningful.

The functional’s first variation pairs metric velocity with the Einstein tensor. Along the flow, this relates its change to Ricci energy. The main difficulty is to construct and differentiate the exterior using one fixed material collar, without assuming that the whole bubble tree varies smoothly in time.

Why a multiscale analytic path is needed

M3 reduces the elliptic problem to finite parameters while retaining kernels and cokernels of the linearized equations. Long necks introduce exponential and polynomial-logarithmic terms. An asymptotic expansion to every finite order is not automatically a convergent power series at infinite neck length.

The manuscript therefore proves a path-selection theorem for a specified class of such expansions. A path can retain finite sign conditions and prescribed limits while having finite weighted variation in all length and ordinary parameters. The lengths may diverge at unrelated rates, and no assumption says that infinitesimal Ricci-flat deformations integrate to a smooth moduli space.

Combining this path control with the first variation gives the little-oh improvement $\mathcal E=o(\mathcal M)$ on each fixed limiting tree. No uniform exponent across all possible trees is asserted or needed by M2.

Small energy prevents the prescribed scale change

M2 selects time intervals on which the outermost scale changes by a fixed factor. A divergence constraint removes a problematic radial mode in the Ricci equation on nearly flat annuli, improving exterior decay. A time maximal-function estimate converts spacetime Ricci energy into the weighted single-time error required by M3.

The static smallness of the functional conflicts with the amount of variation that a contracting root would require. The resulting estimate forces normalized Ricci energy to tend to zero. Dense material point sets then preserve distances at the root scale between the interval endpoints, producing a pointed isometry that preserves the geometric quantity defining that scale.

But the intervals were chosen to change the scale by a fixed factor. This contradiction rules out concentration. The flow consequently reaches a smooth terminal metric and extends on the original manifold.

7. Scope, limitations, and open questions

M2 treats closed four-manifolds; M1 supplies counterexamples only in sufficiently high dimension. The family does not identify the smallest dimension where bounded-scalar finite-time singularities occur or settle every intermediate dimension.

A scalar bound must be uniform over the full spacetime interval. The theorems do not replace it by an arbitrary averaged or spatial integral condition.

The static companion applies to a fixed finite limiting tree with detailed decay and regularity hypotheses. Its error is weighted and includes derivatives. It should not be simplified into a general scalar-energy inequality without those assumptions.

Finally, the high-dimensional construction explicitly builds on Stolarski’s singular flows, while the four-dimensional extension explicitly depends on M3. This digest preserves those dependencies and reports the claims without auditing the underlying estimates.

8. Sources and further reading

The linked primary sources explain tensor extension, scalar-bound compactness, and the singular-flow construction. Optional background appears under Ricci flow, scalar curvature, and Einstein–Hilbert action.

Source manuscripts

• M1. OpenAI, A closed Ricci flow with bounded scalar curvature and finite-time curvature blowup (September 24, 2026).
• M2. OpenAI, Bounded scalar curvature and smooth extension of four-dimensional Ricci flow (September 24, 2026).
• M3. OpenAI, Path selection and an elliptic inequality on degenerating Ricci-flat trees (September 24, 2026).

Sources consulted on 26 October 2026 at repository revision 1581d4c. Download the manuscript citations.