Mean curvature flow with surgery of mean convex surfaces in \(\mathbb R^3\) (Q5963146)

The goal of the paper is to prove the existence of an (appropriately defined) mean curvature flow with surgeries, which terminates after a finite length of time and finitely many surgeries, for the evolution of compact, mean convex, embedded surfaces in \(\mathbb{R}^3\). The second author and \textit{C. Sinestrari} [Invent. Math. 175, No. 1, 137--221 (2009; Zbl 1170.53042)] had already proved the existence of a mean curvature flow with surgeries for the evolution of compact, two convex immersed hypersurfaces of \(\mathbb{R}^{n+1}\) when \(n\geq 3\), but their analysis breaks down when \(n=2\). The reason is that their `cylindrical estimate' (and consequently their pointwise `gradient estimate' for the curvature) relies on the presence of at least three principal curvatures. Indeed, the estimate states that, close to a singularity, if one principal curvature is close to zero then the remaining principal curvatures are positive and close to each other. Combining this estimate with a clever maximum principle argument yields pointwise, scaling invariant estimates for the derivatives of the curvature (which hold up to the singular time). By a rather robust but delicate analysis, the cylindrical and gradient estimates can be combined to provide a qualitative diagnosis of singular regions needed to identify the `necks' which are to be removed by the surgery (at least at the first singular time. One also needs to prove that the estimates `survive' surgeries). It appears that the key to the present result is to find a replacement for these estimates which works for the evolution of mean convex surfaces in \(\mathbb{R}^3\). The crucial breakthroughs since the work of the second author and Sinestrari were the discovery by \textit{B. White} [J. Am. Math. Soc. 16, No. 1, 123--138 (2003; Zbl 1027.53078)] and \textit{W. Sheng} and \textit{X.-J. Wang} [Methods Appl. Anal. 16, No. 2, 139--156 (2009; Zbl 1184.53071)] that mean convex, embedded mean curvature flow is `noncollapsing', the qualitative refinements of these results by \textit{B. Andrews} [Geom. Topol. 16, No. 3, 1413--1418 (2012; Zbl 1250.53063)] and the author [Invent. Math. 202, No. 1, 217--237 (2015; Zbl 1330.53083)] and the local estimates of \textit{R. Haslhofer} and \textit{B. Kleiner} [Commun. Pure Appl. Math. 70, No. 3, 511--546 (2017; Zbl 1360.53069)]. We recall that a solution of the mean curvature flow is \textit{noncollapsing} if the pointwise ratio of the inscribed curvature (one over the radius of the largest ball enclosed by the surface which touches at the point in question) to the mean curvature is bounded uniformly in time. Andrews was able to derive an evolution equation for the ratio and deduce, by the maximum principle, that initial upper bounds (finite if the initial surface is embedded) are preserved. Refining this idea, the author was then able to show, using Stampacchia iteration, that the noncollapsing ratio improves to the optimal value, 1, at a singularity. This yields a replacement for the cylindrical estimate: If the smallest principal curvature is close to zero then the inscribed curvature is close to the remaining principal curvature (note that equality in both cases can only hold for a cylindrical region). The curvature gradient estimate is replaced by the remarkable estimate of Haslhofer-Kleiner, which provides a bound \(C\) (on a parabolic ball of radius \(\rho<1\)) for the gradient of the curvature, so long as the flow is noncollapsing and defined on a parabolic ball of radius 1 (where \(C\) and \(\rho\) depend only on the quality of the noncollapsing). One further difficulty needs to be overcome: Unlike the situation in [the second author and Sinestrari, loc. cit.] there is some difficulty in showing that these estimates survive surgeries. Indeed, the quality of noncollapsing can deteriorate slightly. This problem is overcome by performing the surgery after such time that the noncollapsing ratio was able to improve by an amount sufficient to compensate for any drop during the surgery. Finally, a new pseudolocality principle is introduced to control the derivatives of the curvature directly after a surgery (until such time that the Haslhofer-Kleiner estimate kicks in). Roughly speaking, if the initial hypersurface is sufficiently close to a plane in some ball then the curvature and its derivatives remain uniformly bounded in a smaller ball for a short time.