Does every magnification see the same mean-curvature singularity?

← All digests

Family 355 · Differential geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

A surface moving by mean curvature tends to smooth itself out, but it can also develop a neck pinch or collapse in finite time. Near such a singularity, one magnifies space and slows down time to see the local shape more clearly.

There is a subtle ambiguity. Different sequences of magnifications might conceivably reveal different limiting shapes. Even if the shape is always a cylinder, its axis could drift as the magnification changes. Existence of a limiting model along some sequence does not show that the singularity has one well-defined model.

Family 355 claims that this ambiguity disappears at the first singular time of every smooth closed embedded surface in $\mathbb R^3$. At each fixed singular point, every sufficiently large magnification approaches the same shrinking surface, with the same position and axes in the original coordinates. No mean-convexity assumption and no advance choice of a spherical, cylindrical or other model is required.

The result concerns what happens before the singularity. It does not assert a unique way to continue the original surface after the singular time.

2. Background and essential definitions

Mean-curvature flow and parabolic magnification

Let $M_t\subset\mathbb R^3$ be a smoothly evolving surface. Mean-curvature flow prescribes its normal velocity to be its mean curvature vector $\mathbf H$:

$$\partial_t X=\mathbf H,$$

up to tangential changes of parametrization. Here $X$ records points on the surface, and the convention uses the trace of the second fundamental form. A round sphere shrinks under this motion.

Suppose the flow first becomes singular at time $T$, and $x_0$ is a singular point. Magnifying lengths by $\lambda$ requires magnifying time by $\lambda^2$, because curvature has units of inverse length. The rescaled surfaces are

$$M_s^{\lambda}=\lambda\bigl(M_{T+\lambda^{-2}s}-x_0\bigr),\qquad s<0.$$

The new singular time is zero. For fixed negative $s$, increasing $\lambda$ examines the original flow closer to $T$. A tangent flow is a limit obtained along a sequence $\lambda_i\to\infty$.

The center $x_0$ remains fixed. Allowing the center to move with the scale is a different procedure and can reveal different geometric information.

Self-shrinkers and multiplicity

A self-shrinker is a surface $S$ satisfying

$$\mathbf H+\frac{X^\perp}{2}=0,$$

where $X^\perp$ is the normal component of the position vector. It generates a flow by simple dilation:

$$M_s=\sqrt{-s}\,S,\qquad s<0.$$

For example, in this convention the round shrinking sphere at time $-1$ has radius two, and the round cylindrical model has radius $\sqrt2$.

A limit has multiplicity one when it represents one sheet of area, rather than several sheets collapsing onto the same support. This is stronger than knowing only the visible set of points. Local smooth convergence means that on each fixed bounded spatial region the surfaces and their derivatives converge smoothly, using suitable local graphical descriptions.

A noncompact shrinker can have several ends, the parts remaining far outside a large ball. An end may approach a cone as it is dilated, or approach a round cylinder as one moves outward along it. Having a cylindrical end does not by itself mean that the entire surface is a cylinder.

3. Development history and previous partial results

Huisken’s monotonicity formula identified self-shrinkers as natural models for blowups of mean-curvature flow. Ilmanen’s work supplied smooth tangent supports at the first singular time for closed embedded surface flows in three dimensions. Smooth support, however, leaves multiplicity and uniqueness as additional issues.

Several major uniqueness cases were subsequently settled. Schulze treated smooth compact multiplicity-one shrinkers. Colding and Minicozzi proved uniqueness for round cylindrical tangents, including the cylinder’s axis. Chodosh and Schulze established uniqueness for multiplicity-one asymptotically conical hypersurface tangents.

These results already resolved large and important classes. They did not amount to a theorem for every first-time singularity without specifying its tangent geometry. In particular, a shrinker with cylindrical ends need not have been classified as an exact round cylinder, and it can require a simultaneous analysis of several ends.

Bamler and Kleiner’s multiplicity-one theory is a major geometric input for the present manuscript. In the setting used here, its tangent structure gives a smooth properly embedded multiplicity-one shrinker with finitely many separated conical and cylindrical ends. Lu Wang’s earlier work established the conical-or-cylindrical end alternative under finite-topology assumptions. The manuscript takes its required finite-end structure directly from Bamler–Kleiner; finite genus alone is not substituted for finite topology.

The analytic strategy has its own history. Simon developed convergence methods based on gradient inequalities, and the preceding tangent-uniqueness theorems adapted them to geometric flows. Zhu developed finite-order obstruction methods for certain exact product shrinkers. Family 355 proposes a stationary analysis covering all the ends supplied by the geometric structure theorem, without assuming a separate cylinder-rigidity conjecture.

4. What the new manuscript claims

Start with a smooth, compact, connected surface $M_0\subset\mathbb R^3$, embedded without boundary. Let $(M_t)_{0\le t<T}$ be its maximal smooth mean-curvature flow, and fix a singular point $(x_0,T)$ at its first singular time.

The theorem claims the existence of one smooth properly embedded self-shrinker $S$ such that

$$M_s^\lambda\longrightarrow\sqrt{-s}\,S \qquad\text{as }\lambda\to\infty.$$

This is convergence of the full family of scales, not merely a selected subsequence. It is locally smooth in space, with multiplicity one, uniformly for $s$ in every compact interval contained in $(-\infty,0)$.

The limit is unique in the original ambient frame. One does not rotate each rescaled surface separately to obtain convergence. The position and axes are part of the conclusion.

There is no mean-convexity condition—that is, no requirement that the mean curvature consistently point toward one chosen side of the initial surface. There is no separately imposed entropy or genus bound beyond those automatically supplied by the compact initial surface, and no assumed list of allowable tangent models.

The manuscript also states an all-negative-time conclusion for weak tangent flows in its specified integral Brakke convention. A Brakke flow describes evolving area measures through an inequality encoding motion by mean curvature. Here the convention requires the relevant local integrability and an inequality for every pair of time endpoints, together with convergence of the rescaled spacetime area measures. Under those conditions, every negative-time slice equals the area measure of $\sqrt{-s}\,S$, not just almost every slice.

That endpoint convention matters: spacetime measures alone do not determine what an arbitrarily chosen representative does at isolated times.

5. Why the advance matters

Importance: 5/5, conditional on the claim. Uniqueness of blowups is a fundamental question in singularity analysis. A tangent object is much more informative when it belongs intrinsically to the singularity, rather than depending on which sequence of scales was selected.

The claim unifies several previously separate geometric cases in the basic setting of closed embedded surfaces in three-dimensional space. Fixing the ambient frame strengthens the conclusion beyond uniqueness up to rotation and rules out persistent axis drift during magnification.

The broader methodological significance lies in handling noncompact models whose distant ends can change type under tiny perturbations. Such effects are difficult for ordinary local expansions to see. The paper proposes to combine finite Taylor information with exponentially small geometric effects, then convert that stationary information into actual convergence of the evolving surface.

This significance is conditional on the stated theorem. It does not imply that every singularity is a sphere or cylinder, or that uniqueness of a backward model resolves every issue in weak continuation.

6. The main idea

Rescaling turns the problem into convergence toward a critical point

Introduce logarithmic time and rescaled surfaces by

$$\tau=-\log(T-t),\qquad \widetilde M_\tau=(T-t)^{-1/2}(M_t-x_0).$$

As the original time approaches $T$, the rescaled time tends to infinity. A self-shrinker becomes stationary in these coordinates.

The relevant energy is Gaussian area,

$$F(S)=\int_S e^{-|X|^2/4}\,d\mu,$$

with an inessential normalization constant omitted. The Gaussian factor strongly suppresses contributions far from the origin. The rescaled flow decreases this energy, and its stationary defect is $\mathbf H+X^\perp/2$.

Energy decrease alone need not prevent a trajectory from drifting among nearby critical objects. A gradient inequality relates energy difference to the size of the stationary defect. When combined with the flow equation, it can show that the rescaled trajectory has finite total length in a suitable norm and hence approaches one limit.

The difficult directions open cylindrical ends

A Jacobi field is an infinitesimal deformation satisfying the linearized shrinker equation. Certain Jacobi fields grow quadratically along a cylindrical end and begin to change its type. They are not ordinary decaying perturbations.

For a circular comparison profile, an opening parameter $\beta<0$ gives

$$R_c(z)^2=2(1+\beta z^2).$$

Its radius reaches zero at distance $z=(-\beta)^{-1/2}$. At that distance the Gaussian factor is

$$\exp\!\left(-\frac1{4|\beta|}\right).$$

As $\beta$ tends to zero, this is smaller than every fixed power of $|\beta|$. A major geometric change far along the end can therefore be invisible to any single finite Taylor approximation of the Gaussian energy.

The manuscript does not simply assume that a power series in the opening parameters converges. It retains finite Taylor expansions, called jets, and analyzes the smaller effects separately.

Exclude a hidden family of stationary openings

Compactly supported observations near the core provide finitely many parameters for nearby surfaces. Some parameters preserve end types; others open cylindrical ends. Successive linear solves define a formal stationary action, an algebraic power series not presumed convergent in all directions.

The central proposed obstruction is that a formal curve of critical points cannot have a nonzero opening. To obtain it, the paper constructs two stationary comparison solutions using complex parameter contours passing on opposite sides of a collapsing radius. These are complex coefficients on real spatial domains; it does not analytically continue an arbitrary smooth surface in its spatial variables.

The two constructions share the same formal expansion but differ by an exponentially small Stokes jump. This term describes a difference that ordinary power-series coefficients miss. The collapsing end supplies a nonzero contribution, and the analysis of several ends shows that their relevant contributions cannot cancel. This contradicts the stronger smallness that a stationary opening curve would require.

Real-algebraic and analytic approximation tools then convert that exclusion into a finite-power estimate. One sufficiently high finite truncation is enough to control the opening parameters by the stationary defect. Convergence of the full formal series is unnecessary.

Localize carefully enough to control the actual flow

A noncompact limit cannot be compared globally to each compact rescaled surface without cutoffs. The manuscript derives a localized gradient inequality and estimates the discarded Gaussian tails at the scale of the defect.

Graphical improvement and short-time smoothing let the comparison radius grow. The radius is chosen using accumulated earlier energy drops, so the localization errors have a finite total sum. This closes the graphical comparison near the core and yields finite rescaled length.

The resulting convergence takes place at every intermediate rescaled time in the fixed coordinates, not just along the original extracting sequence. Finally, the all-endpoint Brakke inequality upgrades equality of spacetime weights to equality at each negative time under the stated weak-flow convention.

7. Scope, limitations, and open questions

The theorem is about two-dimensional embedded surfaces in $\mathbb R^3$, their smooth evolution before the first singular time, and rescalings about one fixed singular center. It does not assert the corresponding statement for arbitrary hypersurface dimensions, immersed initial surfaces, or moving-center blowups.

It establishes uniqueness of the tangent shrinker without classifying all such shrinkers. In particular, a model with a cylindrical end is not declared to be a round cylinder everywhere. The proof is designed to avoid needing that additional rigidity statement.

The convergence is local on compact negative-time intervals. The theorem does not describe a unique forward evolution through time zero, and the weak representative conclusion relies on its explicit endpoint and integrability conditions.

Questions about later singularities of continued weak flows, higher-dimensional analogues and a complete classification of possible shrinkers remain separate. The stationary obstruction and convergence argument are described here as the manuscript presents them, without independent proof verification.

8. Sources and further reading

The source’s introduction gives the precise fixed-center and weak-tangent conventions. Its Sections 3–7 develop the stationary obstruction; Sections 8–9 turn it into convergence.

Primary predecessors include Schulze on compact tangents, Colding–Minicozzi on cylindrical blowups, Chodosh–Schulze on conical tangents, and Bamler–Kleiner on multiplicity one. These supply distinct pieces of the context rather than interchangeable versions of the same theorem.

Source manuscripts

• M1. OpenAI, Uniqueness of tangent flows at the first singular time of embedded surface mean-curvature flow (September 24, 2026).

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