Translators2760875656EF

Translation as a geometric evolution

A translator is a submanifold whose mean curvature flow moves with a constant ambient velocity. For a smooth $n$-dimensional submanifold $\Sigma\subset\mathbb R^N$ and a constant vector $V$, the equation and associated flow are

Only the normal component of the velocity changes the evolving image. We use $\vec H=\Delta_\Sigma X$ and, for hypersurfaces, $\vec H=-H\nu$. After a rotation and a spatial scaling, a nonzero translation velocity can be normalized to $V=e_{n+1}$ in $\mathbb R^{n+1}$. The magnitude of the velocity is not preserved by dilation: dilating space by $a>0$ changes the speed to $|V|/a$. The examples below use unit upward translation unless another convention is stated.

Translators are eternal solutions, but eternal existence does not imply translation. They are stationary when viewed in a frame moving with $V$. A minimal hypersurface invariant under translation in the $V$ direction is a stationary special case, since $V^\perp=0$. There is no compact boundaryless Euclidean translator with $V\ne0$: the height $h=\langle X,V\rangle$ satisfies $\Delta_\Sigma h=|V^\perp|^2$, so integration on a compact domain would force $\vec H=0$, which is impossible for a compact boundaryless positive-dimensional Euclidean immersion.

Type II models and backward limits

A finite-time singularity is Type I if $\sup_{M_t}|A|^2\leq C/(T-t)$ near its singular time, and Type II if no such bound holds. In Type II analysis, one magnifies around high-curvature points at their curvature scale. Suitable compactness estimates can produce ancient or eternal limits. Translators are important examples of such limits, but a curvature rescaling alone does not prove that the limit translates. Convexity, noncollapsing, Harnack inequalities, or a classification theorem provide the additional information.

The degenerate neckpinches of Angenent–Velázquez provide a concrete link between finite-time singularity formation and translating tip models (AV97, Introduction). The bowl is particularly important in convex flow. Brendle–Choi prove that every noncompact ancient solution in $\mathbb R^3$ that is strictly convex and uniformly two-sided noncollapsed is a bowl flow, up to scaling and ambient isometries (BrendleChoi19, Theorem 1.1). Noncollapsing means that interior and exterior tangent balls have radii at least $\alpha/H$ for a uniform $\alpha>0$. Dropping these hypotheses changes the classification problem.

The introduction of Sun–Wang emphasizes another connection: a finite-entropy translator can be studied by taking a parabolic limit of its eternal flow as time tends to $-\infty$. Its backward tangent flow is self-shrinking and splits off the translation direction, in the precise weak formulation of their Corollary 2.4. Thus the large-scale geometry of a translator is linked to a shrinker of one lower dimension. This limit may have multiplicity, and the statement does not assert that every translator has a smooth or uniquely determined shrinker profile (SunWang2025, Sections 1–2).

Graphs, weighted minimality, and basic models

For unit upward translation, a graph $(x,u(x)+t)$ over a domain $\Omega\subset\mathbb R^n$ satisfies

In dimension one this yields the exact grim reaper $u(x)=-\log\cos x$, with $|x|<\pi/2$. The finite strip width is compatible with completeness because the height diverges at the boundary. In dimensions $n\geq2$, the rotational entire graph $u(x)=f(|x|)$ satisfies

This regular ODE defines the bowl. Its quadratic leading term, $r^2/(2(n-1))$, describes its growth but does not give the exact surface. Clutterbuck–Schnürer–Schulze establish the profile asymptotics and dynamical stability for graphical initial data whose height difference from the bowl tends to zero at spatial infinity (CSS07, Theorem 1.1 and Section 2). A theorem for this asymptotic class is not stability under unrestricted perturbations.

Translators are also critical points, for compactly supported variations, of the weighted area

Indeed, the first variation in a normal direction $W$ is $\int_\Sigma\langle V^\perp-\vec H,W\rangle e^{\langle V,X\rangle}\,d\mu$. The integral may be infinite on a complete translator; the local variational equation still makes sense. For a hypersurface, the normal Jacobi operator is $\Delta_\Sigma+\langle V^T,\nabla_\Sigma\cdot\rangle+|A|^2$. On a translating graph, the positive vertical normal component provides the stability function. This is weighted stability, distinct from stability for ordinary area or from convergence of the time-dependent flow (HIMWGraphical18, Section 2).

Classification and the range of constructions

Hoffman–Ilmanen–Martín–White classify complete translating graphs in $\mathbb R^3$: after normalizing the velocity and allowing the appropriate rigid motions, the possibilities are the bowl, grim-reaper cylinders and their tilted versions, and strictly convex $\Delta$-wings. A $\Delta$-wing exists on each strip of width greater than $\pi$ and is unique up to translation for that width. The theorem concerns graphs over domains perpendicular to the translation direction; it is not a classification of all complete embedded translators (HIMWGraphical18, Introduction and Theorem 1.1).

Outside the graphical class, annular translating catenoids, Scherk-type surfaces, pitchforks, and gluing constructions have different topology and end geometry. The analogy with minimal surfaces is useful because both problems have elliptic variational formulations, but a minimal surface does not become a translator merely by adding a constant velocity. The survey HIMWNotes19 explains these constructions and the limits of the analogy. Higher dimensions also admit convex translators beyond the rotational bowl, as in Xu-Jia Wang's work (XJWang11, Introduction). Dimension, convexity, and graphicality must therefore accompany every uniqueness statement.

Prescribing the shrinker at infinity

Sun–Wang turn the backward-limit question into a construction. Given a smooth closed embedded shrinker $S^{n-1}\subset\mathbb R^n$ of Gaussian Morse index $I$, they construct an $I$-parameter family of complete embedded translators in $\mathbb R^{n+1}$ with unique multiplicity-one backward tangent flow $S\times\mathbb R$. The construction is smooth for $n\leq6$; in higher dimensions it allows a singular set of dimension at most $n-7$. The prescribed asymptotics do not determine a unique translator, and their fattening construction gives further nonuniqueness (SunWang2025, Theorem 1.1 and Section 1.3).

For this correspondence, the cross-sectional shrinker and the translator have different dimensions. A genus-one surface in $\mathbb R^3$ supplies asymptotic data for a three-dimensional translator in $\mathbb R^4$; its surface genus is not a genus assigned to the entire translator. Likewise, a backward parabolic limit is different from an unscaled cylindrical end. The catalogue records the precise end description, regularity range, topology, and entropy information for each construction. These distinctions make it possible to compare examples without turning a particular asymptotic classification into a universal one.