Ricci flow with surgery on three-manifolds

$\ \ \ \ \ \ B(x,t,r)$ denotes the open metric ball of radius $r,$ with respect to the metric at time $t,$ centered at $x.$

$P(x,t,r,\triangle t)$ denotes a parabolic neighborhood, that is the set of all points $(x^{\prime},t^{\prime})$ with $x^{\prime}\in B(x,t,r)$ and $t^{\prime}\in[t,t+\triangle t]$ or $t^{\prime}\in[t+\triangle t,t],$ depending on the sign of $\triangle t.$

A ball $B(x,t,\epsilon^{-1}r)$ is called an $\epsilon$-neck, if, after scaling the metric with factor $r^{-2},$ it is $\epsilon$-close to the standard neck $\mathbb{S}^{2}\times\mathbb{I},$ with the product metric, where $\mathbb{S}^{2}$ has constant scalar curvature one, and $\mathbb{I}$ has length $2\epsilon^{-1};$ here $\epsilon$-close refers to $C^{N}$ topology, with $N>\epsilon^{-1}.$

A parabolic neighborhood $P(x,t,\epsilon^{-1}r,r^{2})$ is called a strong $\epsilon$-neck, if, after scaling with factor $r^{-2},$ it is $\epsilon$-close to the evolving standard neck, which at each time $t^{\prime}\in[-1,0]$ has length $2\epsilon^{-1}$ and scalar curvature $(1-t^{\prime})^{-1}.$

A metric on $\mathbb{S}^{2}\times\mathbb{I},$ such that each point is contained in some $\epsilon$-neck, is called an $\epsilon$-tube, or an $\epsilon$-horn, or a double $\epsilon$-horn, if the scalar curvature stays bounded on both ends, stays bounded on one end and tends to infinity on the other, and tends to infinity on both ends, respectively.

A metric on $\mathbb{B}^{3}$ or $\mathbb{RP}^{3}\setminus\bar{\mathbb{B}}^{3},$ such that each point outside some compact subset is contained in an $\epsilon$-neck, is called an $\epsilon$-cap or a capped $\epsilon$-horn, if the scalar curvature stays bounded or tends to infinity on the end, respectively.

We denote by $\epsilon$ a fixed small positive constant. In contrast, $\delta$ denotes a positive quantity, which is supposed to be as small as needed in each particular argument.

1.1 In this section we review some of the results, proved or quoted in [I,§11], correcting a few inaccuracies. We consider smooth solutions $g_{ij}(t)$ to the Ricci flow on oriented 3-manifold $M$, defined for $-\infty<t\leq 0$, such that for each $t$ the metric $g_{ij}(t)$ is a complete non-flat metric of bounded nonnegative sectional curvature, $\kappa$-noncollapsed on all scales for some fixed $\kappa>0;$ such solutions will be called ancient $\kappa$-solutions for short. By Theorem I.11.7, the set of all such solutions with fixed $\kappa$ is compact modulo scaling, that is from any sequence of such solutions $(M^{\alpha},g_{ij}^{\alpha}(t))$ and points $(x^{\alpha},0)$ with $R(x^{\alpha},0)=1$, we can extract a smoothly (pointed) convergent subsequence, and the limit $(M,g_{ij}(t))$ belongs to the same class of solutions. (The assumption in I.11.7. that $M^{\alpha}$ be noncompact was clearly redundant, as it was not used in the proof. Note also that $M$ need not have the same topology as $M^{\alpha}.)$ Moreover, according to Proposition I.11.2, the scalings of any ancient $\kappa$-solution $g_{ij}(t)$ with factors $(-t)^{-1}$ about appropriate points converge along a subsequence of $t\to-\infty$ to a non-flat gradient shrinking soliton, which will be called an asymptotic soliton of the ancient solution. If the sectional curvature of this asymptotic soliton is not strictly positive, then by Hamilton’s strong maximum principle it admits local metric splitting, and it is easy to see that in this case the soliton is either the round infinite cylinder, or its $\mathbb{Z}_{2}$ quotient, containing one-sided projective plane. If the curvature is strictly positive and the soliton is compact, then it has to be a metric quotient of the round 3-sphere, by [H 1]. The noncompact case is ruled out below.

1.2 Lemma. There is no (complete oriented 3-dimensional) noncompact $\kappa$-noncollapsed gradient shrinking soliton with bounded positive sectional curvature.

Proof. A gradient shrinking soliton $g_{ij}(t),-\infty<t<0,$ satisfies the equation

Differentiating and switching the order of differentiation, we get

Fix some $t<0,$ say $t=-1,$ and consider a long shortest geodesic $\gamma(s),0\leq s\leq\bar{s};$ let $x=\gamma(0),\bar{x}=\gamma(\bar{s}),X(s)=\dot{\gamma}(s).$ Since the curvature is bounded and positive, it is clear from the second variation formula that $\int_{0}^{\bar{s}}{\mathrm{Ric}(X,X)ds}\leq\mathrm{const}.$ Therefore, $\int_{0}^{\bar{s}}{|\mathrm{Ric}(X,\cdot)|^{2}ds}\leq\mathrm{const},$ and $\int_{0}^{\bar{s}}{|\mathrm{Ric}(X,Y)|ds}\leq\mathrm{const}(\sqrt{\bar{s}}+1)$ for any unit vector field $Y$ along $\gamma,$ orthogonal to $X.$ Thus by integrating (1.1) we get $X\cdot f(\gamma(\bar{s}))\geq\frac{\bar{s}}{2}+\mathrm{const},|Y\cdot f(\gamma(\bar{s}))|\leq\mathrm{const}(\sqrt{\bar{s}}+1).$ We conclude that at large distances from $x_{0}$ the function $f$ has no critical points, and its gradient makes small angle with the gradient of the distance function from $x_{0}.$

Now from (1.2) we see that $R$ is increasing along the gradient curves of $f,$ in particular, $\bar{R}=\mathrm{lim\ sup\ }R>0.$ If we take a limit of our soliton about points $(x^{\alpha},-1)$ where $R(x^{\alpha})\to\bar{R},$ then we get an ancient $\kappa$-solution, which splits off a line, and it follows from I.11.3, that this solution is the shrinking round infinite cylinder with scalar curvature $\bar{R}$ at time $t=-1.$ Now comparing the evolution equations for the scalar curvature on a round cylinder and for the asymptotic scalar curvature on a shrinking soliton we conclude that $\bar{R}=1.$ Hence, $R(x)<1$ when the distance from $x$ to $x_{0}$ is large enough, and $R(x)\to 1$ when this distance tends to infinity.

Now let us check that the level surfaces of $f,$ sufficiently distant from $x_{0},$ are convex. Indeed, if $Y$ is a unit tangent vector to such a surface, then $\nabla_{Y}\nabla_{Y}f=\frac{1}{2}-\mathrm{Ric}(Y,Y)\geq\frac{1}{2}-\frac{R}{2}>0.$ Therefore, the area of the level surfaces grows as $f$ increases, and is converging to the area of the round sphere of scalar curvature one. On the other hand, the intrinsic scalar curvature of a level surface turns out to be less than one. Indeed, denoting by $X$ the unit normal vector, this intrinsic curvature can be computed as

when $R$ is close to one and $|\nabla f|$ is large. Thus we get a contradiction to the Gauss-Bonnet formula.

1.3 Now, having listed all the asymptotic solitons, we can classify the ancient $\kappa$-solutions. If such a solution has a compact asymptotic soliton, then it is itself a metric quotient of the round 3-sphere, because the positive curvature pinching can only improve in time [H 1]. If the asymptotic soliton contains the one-sided projective plane, then the solution has a $\mathbb{Z}_{2}$ cover, whose asymptotic soliton is the round infinite cylinder. Finally, if the asymptotic soliton is the cylinder,then the solution can be either noncompact (the round cylinder itself, or the Bryant soliton, for instance), or compact. The latter possibility, which was overlooked in the first paragraph of [I.11.7], is illustrated by the example below, which also gives the negative answer to the question in the very end of [I.5.1].

1.4 Example. Consider a solution to the Ricci flow, starting from a metric on $\mathbb{S}^{3}$ that looks like a long round cylinder $\mathbb{S}^{2}\times\mathbb{I}$ (say, with radius one and length $L>>1$), with two spherical caps, smoothly attached to its boundary components. By [H 1] we know that the flow shrinks such a metric to a point in time, comparable to one (because both the lower bound for scalar curvature and the upper bound for sectional curvature are comparable to one) , and after normalization, the flow converges to the round 3-sphere. Scale the initial metric and choose the time parameter in such a way that the flow starts at time $t_{0}=t_{0}(L)<0,$ goes singular at $t=0,$ and at $t=-1$ has the ratio of the maximal sectional curvature to the minimal one equal to $1+\epsilon.$ The argument in [I.7.3] shows that our solutions are $\kappa$-noncollapsed for some $\kappa>0$ independent of $L.$ We also claim that $t_{0}(L)\to-\infty$ as $L\to\infty.$ Indeed, the Harnack inequality of Hamilton [H 3] implies that $\ \ R_{t}\geq\frac{R}{t_{0}-t},\ \ \ \$ hence $\ \ R\leq\frac{2(-1-t_{0})}{t-t_{0}}$ for $t\leq-1,\ \ \ \ \ \ \ \$ and then the distance change estimate $\ \ \ \frac{d}{dt}\mathrm{dist}_{t}(x,y)\geq-\mathrm{const}\sqrt{R_{\mathrm{max}}(t)}\ \ \ \$ from [H 2,§17] implies that the diameter of $g_{ij}(t_{0})$ does not exceed $-\mathrm{const}\cdot t_{0}$, which is less than $L\sqrt{-t_{0}}$ unless $t_{0}$ is large enough. Thus, a subsequence of our solutions with $L\to\infty$ converges to an ancient $\kappa$-solution on $\mathbb{S}^{3},$ whose asymptotic soliton can not be anything but the cylinder.

1.5 The important conclusion from the classification above and the proof of Proposition I.11.2 is that there exists $\kappa_{0}>0,$ such that every ancient $\kappa$-solution is either $\kappa_{0}$-solution, or a metric quotient of the round sphere. Therefore, the compactness theorem I.11.7 implies the existence of a universal constant $\eta,$ such that at each point of every ancient $\kappa$-solution we have estimates

Moreover, for every sufficiently small $\epsilon>0$ one can find $C_{1,2}=C_{1,2}(\epsilon),$ such that for each point $(x,t)$ in every ancient $\kappa$-solution there is a radius $r,0<r<C_{1}R(x,t)^{-\frac{1}{2}},$ and a neighborhood $B,B(x,t,r)\subset B\subset B(x,t,2r),$ which falls into one of the four categories:

(a) $B$ is a strong $\epsilon$-neck (more precisely, the slice of a strong $\epsilon$-neck at its maximal time), or

(b) $B$ is an $\epsilon$-cap, or

(c) $B$ is a closed manifold, diffeomorphic to $\mathbb{S}^{3}$ or $\mathbb{RP}^{3},$ or

(d) $B$ is a closed manifold of constant positive sectional curvature;

furthermore, the scalar curvature in $B$ at time $t$ is between $C_{2}^{-1}R(x,t)$ and $C_{2}R(x,t),$ its volume in cases (a),(b),(c) is greater than $C_{2}^{-1}R(x,t)^{-\frac{3}{2}},$ and in case (c) the sectional curvature in $B$ at time $t$ is greater than $C_{2}^{-1}R(x,t).$

Consider a rotationally symmetric metric on $\mathbb{R}^{3}$ with nonnegative sectional curvature, which splits at infinity as the metric product of a ray and the round 2-sphere of scalar curvature one. At this point we make some choice for the metric on the cap, and will refer to it as the standard cap; unfortunately, the most obvious choice, the round hemisphere, does not fit, because the metric on $\mathbb{R}^{3}$ would not be smooth enough, however we can make our choice as close to it as we like. Take such a metric on $\mathbb{R}^{3}$ as the initial data for a solution $g_{ij}(t)$ to the Ricci flow on some time interval $[0,T),$ which has bounded curvature for each $t\in[0,T).$

Claim 1. The solution is rotationally symmetric for all $t.$

Indeed, if $u^{i}$ is a vector field evolving by $u^{i}_{t}=\triangle u^{i}+R^{i}_{j}u^{j},$ then $v_{ij}=\nabla_{i}u_{j}$ evolves by $(v_{ij})_{t}=\triangle v_{ij}+2R_{ikjl}v_{kl}-R_{ik}v_{kj}-R_{kj}v_{ik}.$ Therefore, if $u^{i}$ was a Killing field at time zero, it would stay Killing by the maximum principle. It is also clear that the center of the cap, that is the unique maximum point for the Busemann function, and the unique point, where all the Killing fields vanish, retains these properties, and the gradient of the distance function from this point stays orthogonal to all the Killing fields. Thus, the rotational symmetry is preserved.

Claim 2. The solution converges at infinity to the standard solution on the round infinite cylinder of scalar curvature one. In particular, $T\leq 1.$

Claim 3. The solution is unique.

Indeed, using Claim 1, we can reduce the linearized Ricci flow equation to the system of two equations on $(-\infty,+\infty)$ of the following type

where the coefficients and their derivatives are bounded, and the unknowns $f,g$ and their derivatives tend to zero at infinity by Claim 2. So we get uniqueness by looking at the integrals $\int_{-A}^{A}{(f^{2}+g^{2})}$ as $A\to\infty.$

Claim 4. The solution can be extended to the time interval $[0,1).$

Indeed, we can obtain our solution as a limit of the solutions on $\mathbb{S}^{3},$ starting from the round cylinder $\mathbb{S}^{2}\times\mathbb{I}$ of length $L$ and scalar curvature one, with two caps attached; the limit is taken about the center $p$ of one of the caps, $L\to\infty.$ Assume that our solution goes singular at some time $T<1.$ Take $T_{1}<T$ very close to $T,\ \ T-T_{1}<<1-T.$ By Claim 2, given $\delta>0,$ we can find $\bar{L},\bar{D}<\infty,$ depending on $\delta$ and $T_{1},$ such that for any point $x$ at distance $\bar{D}$ from $p$ at time zero, in the solution with $L\geq\bar{L},$ the ball $B(x,T_{1},1)$ is $\delta$-close to the corresponding ball in the round cylinder of scalar curvature $(1-T_{1})^{-1}.$ We can also find $r=r(\delta,T),$ independent of $T_{1},$ such that the ball $B(x,T_{1},r)$ is $\delta$-close to the corresponding euclidean ball. Now we can apply Theorem I.10.1 and get a uniform estimate on the curvature at $x$ as $t\to T$, provided that $T-T_{1}<\epsilon^{2}r(\delta,T)^{2}.$ Therefore, the $t\to T$ limit of our limit solution on the capped infinite cylinder will be smooth near $x.$ Thus, this limit will be a positively curved space with a conical point. However, this leads to a contradiction via a blow-up argument; see the end of the proof of the Claim 2 in I.12.1.

The solution constructed above will be called the standard solution.

Claim 5. The standard solution satisfies the conclusions of 1.5 , for an appropriate choice of $\epsilon,\ \eta,C_{1}(\epsilon),C_{2}(\epsilon),$ except that the $\epsilon$-neck neighborhood need not be strong; more precisely, we claim that if $(x,t)$ has neither an $\epsilon$-cap neighborhood as in 1.5(b), nor a strong $\epsilon$-neck neighborhood as in 1.5(a), then $x$ is not in $B(p,0,\epsilon^{-1}),\ \ t<3/4,$ and there is an $\epsilon$-neck $B(x,t,\epsilon^{-1}r),$ such that the solution in $P(x,t,\epsilon^{-1}r,-t)$ is, after scaling with factor $r^{-2},\ \ \epsilon$-close to the appropriate piece of the evolving round infinite cylinder.

Moreover, we have an estimate $R_{\mathrm{min}}(t)\geq\mathrm{const}\cdot(1-t)^{-1}.$

Indeed, the statements follow from compactness and Claim 2 on compact subintervals of $[0,1),$ and from the same arguments as for ancient solutions, when $t$ is close to one.

Consider a smooth solution $g_{ij}(t)$ to the Ricci flow on $M\times[0,T),$ where $M$ is a closed oriented 3-manifold, $T<\infty.$ Assume that curvature of $g_{ij}(t)$ does not stay bounded as $t\to T.$ Recall that we have a pinching estimate $Rm\geq-\phi(R)R$ for some function $\phi$ decreasing to zero at infinity [H 4,§4], and that the solution is $\kappa$-noncollapsed on the scales $\leq r$ for some $\kappa>0,r>0$ [I, §4].Then by Theorem I.12.1 and the conclusions of 1.5 we can find $r=r(\epsilon)>0,$ such that each point $(x,t)$ with $R(x,t)\geq r^{-2}$ satisfies the estimates (1.3) and has a neighborhood, which is either an $\epsilon$-neck, or an $\epsilon$-cap, or a closed positively curved manifold. In the latter case the solution becomes extinct at time $T,$ so we don’t need to consider it any more.

If this case does not occur, then let $\Omega$ denote the set of all points in $M,$ where curvature stays bounded as $t\to T.$ The estimates (1.3) imply that $\Omega$ is open and that $R(x,t)\to\infty$ as $t\to T$ for each $x\in M\backslash\Omega.$ If $\Omega$ is empty, then the solution becomes extinct at time $T$ and it is entirely covered by $\epsilon$-necks and caps shortly before that time, so it is easy to see that $M$ is diffeomorphic to either $\mathbb{S}^{3}$, or $\mathbb{RP}^{3}$, or $\mathbb{S}^{2}\times\mathbb{S}^{1}$, or $\mathbb{RP}^{3}\ \sharp\ \mathbb{RP}^{3}.$

Otherwise, if $\Omega$ is not empty, we may (using the local derivative estimates due to W.-X.Shi, see [H 2,§13]) consider a smooth metric $\bar{g}_{ij}$ on $\Omega,$ which is the limit of $g_{ij}(t)$ as $t\to T.$ Let $\Omega_{\rho}$ for some $\rho<r$ denotes the set of points $x\in\Omega,$ where the scalar curvature $\bar{R}(x)\leq\rho^{-2}.$ We claim that $\Omega_{\rho}$ is compact. Indeed, if $\bar{R}(x)\leq\rho^{-2},$ then we can estimate the scalar curvature $R(x,t)$ on $[T-\eta^{-1}\rho^{2},T)$ using (1.3), and for earlier times by compactness, so $x$ is contained in $\Omega$ with a ball of definite size, depending on $\rho.$

Now take any $\epsilon$-neck in $(\Omega,\bar{g}_{ij})$ and consider a point $x$ on one of its boundary components. If $x\in\Omega\backslash\Omega_{\rho},$ then there is either an $\epsilon$-cap or an $\epsilon$-neck, adjacent to the initial $\epsilon$-neck. In the latter case we can take a point on the boundary of the second $\epsilon$-neck and continue. This procedure can either terminate when we reach a point in $\Omega_{\rho}$ or an $\epsilon$-cap, or go on indefinitely, producing an $\epsilon$-horn. The same procedure can be repeated for the other boundary component of the initial $\epsilon$-neck. Therefore, taking into account that $\Omega$ has no compact components, we conclude that each $\epsilon$-neck of $(\Omega,\bar{g}_{ij})$ is contained in a subset of $\Omega$ of one of the following types:

(a) An $\epsilon$-tube with boundary components in $\Omega_{\rho},$ or

(b) An $\epsilon$-cap with boundary in $\Omega_{\rho},$ or

(c) An $\epsilon$-horn with boundary in $\Omega_{\rho},$ or

(d) A capped $\epsilon$-horn, or

(e) A double $\epsilon$-horn.

Clearly, each $\epsilon$-cap, disjoint from $\Omega_{\rho},$ is also contained in one of the subsets above. It is also clear that there is a definite lower bound (depending on $\rho$) for the volume of subsets of types (a),(b),(c), so there can be only finite number of them. Thus we can conclude that there is only a finite number of components of $\Omega,$ containing points of $\Omega_{\rho},$ and every such component has a finite number of ends, each being an $\epsilon$-horn. On the other hand, every component of $\Omega,$ containing no points of $\Omega_{\rho},$ is either a capped $\epsilon$-horn, or a double $\epsilon$-horn.

Now, by looking at our solution for times $t$ just before $T,$ it is easy to see that the topology of $M$ can be reconstructed as follows: take the components $\Omega_{j},1\leq j\leq i$ of $\Omega$ which contain points of $\Omega_{\rho},$ truncate their $\epsilon$-horns, and glue to the boundary components of truncated $\Omega_{j}$ a collection of tubes $\mathbb{S}^{2}\times\mathbb{I}$ and caps $\mathbb{B}^{3}$ or $\mathbb{RP}^{3}\backslash\mathbb{B}^{3}.$ Thus, $M$ is diffeomorphic to a connected sum of $\bar{\Omega}_{j},1\leq j\leq i,$ with a finite number of $\mathbb{S}^{2}\times\mathbb{S}^{1}$ (which correspond to gluing a tube to two boundary components of the same $\Omega_{j}$), and a finite number of $\mathbb{RP}^{3};$ here $\bar{\Omega}_{j}$ denotes $\Omega_{j}$ with each $\epsilon$-horn one point compactified.

4.1 Suppose we are given a collection of smooth solutions $g_{ij}(t)$ to the Ricci flow, defined on $M_{k}\times[t_{k}^{-},t_{k}^{+}),$ which go singular as $t\to t_{k}^{+}.$ Let $(\Omega_{k},\bar{g}_{ij}^{k})$ be the limits of the corresponding solutions as $t\to t_{k}^{+},$ as in the previous section. Suppose also that for each $k$ we have $t_{k}^{-}=t_{k-1}^{+},$ and $(\Omega_{k-1},\bar{g}_{ij}^{k-1})$ and $(M_{k},g_{ij}^{k}(t_{k}^{-}))$ contain compact (possibly disconnected) three-dimensional submanifolds with smooth boundary, which are isometric. Then we can identify these isometric submanifolds and talk about the solution to the Ricci flow with surgery on the union of all $[t_{k}^{-},t_{k}^{+}).$

Fix a small number $\epsilon>0$ which is admissible in sections 1,2. In this section we consider only solutions to the Ricci flow with surgery, which satisfy the following a priori assumptions:

(pinching) There exists a function $\phi,$ decreasing to zero at infinity, such that $Rm\geq-\phi(R)R,$

(canonical neighborhood) There exists $r>0,$ such that every point where scalar curvature is at least $r^{-2}$ has a neighborhood, satisfying the conclusions of 1.5. (In particular, this means that if in case (a) the neighborhood in question is $B(x_{0},t_{0},\epsilon^{-1}r_{0}),$ then the solution is required to be defined in the whole $P(x_{0},t_{0},\epsilon^{-1}r_{0},-r_{0}^{2});$ however, this does not rule out a surgery in the time interval $(t_{0}-r_{0}^{2},t_{0}),$ that occurs sufficiently far from $x_{0}$.)

Recall that from the pinching estimate of Ivey and Hamilton, and Theorem I.12.1, we know that the a priori assumptions above hold for a smooth solution on any finite time interval. For Ricci flow with surgery they will be justified in the next section.

4.2 Claim 1. Suppose we have a solution to the Ricci flow with surgery, satisfying the canonical neighborhood assumption, and let $Q=R(x_{0},t_{0})+r^{-2}.$ Then we have estimate $R(x,t)\leq 8Q$ for those $(x,t)\in P(x_{0},t_{0},\frac{1}{2}\eta^{-1}Q^{-\frac{1}{2}},-\frac{1}{8}\eta^{-1}Q^{-1}),$ for which the solution is defined.

Indeed, this follows from estimates (1.3).

Claim 2. For any $A<\infty$ one can find $Q=Q(A)<\infty$ and $\xi=\xi(A)>0$ with the following property. Suppose we have a solution to the Ricci flow with surgery, satisfying the pinching and the canonical neighborhood assumptions. Let $\gamma$ be a shortest geodesic in $g_{ij}(t_{0})$ with endpoints $x_{0}$ and $x,$ such that $R(y,t_{0})>r^{-2}$ for each $y\in\gamma,$ and $Q_{0}=R(x_{0},t_{0})$ is so large that $\phi(Q_{0})<\xi.$ Finally, let $z\in\gamma$ be any point satisfying $R(z,t_{0})>10C_{2}R(x_{0},t_{0}).$ Then $\mathrm{dist}_{t_{0}}(x_{0},z)\geq AQ_{0}^{-\frac{1}{2}}$ whenever $R(x,t_{0})>QQ_{0}.$

The proof is exactly the same as for Claim 2 in Theorem I.12.1; in the very end of it, when we get a piece of a non-flat metric cone as a blow-up limit, we get a contradiction to the canonical neighborhood assumption, because the canonical neighborhoods of types other than (a) are not close to a piece of metric cone, and type (a) is ruled out by the strong maximum principle, since the $\epsilon$-neck in question is strong.

4.3 Suppose we have a solution to the Ricci flow with surgery, satisfying our a priori assumptions, defined on $[0,T),$ and going singular at time $T.$ Choose a small $\delta>0$ and let $\rho=\delta r.$ As in the previous section, consider the limit $(\Omega,\bar{g}_{ij})$ of our solution as $t\to T,$ and the corresponding compact set $\Omega_{\rho}.$

Lemma. There exists a radius $h,0<h<\delta\rho,$ depending only on $\delta,\rho$ and the pinching function $\phi,$ such that for each point $x$ with $h(x)=\bar{R}^{-\frac{1}{2}}(x)\leq h$ in an $\epsilon$-horn of $(\Omega,\bar{g}_{ij})$ with boundary in $\Omega_{\rho},$ the neighborhood $P(x,T,\delta^{-1}h(x),-h^{2}(x))$ is a strong $\delta$-neck.

Proof. An argument by contradiction. Assuming the contrary, take a sequence of solutions with limit metrics $(\Omega^{\alpha},\bar{g}_{ij}^{\alpha})$ and points $x^{\alpha}$ with $h(x^{\alpha})\to 0.$ Since $x^{\alpha}$ lies deeply inside an $\epsilon$-horn, its canonical neighborhood is a strong $\epsilon$-neck. Now Claim 2 gives the curvature estimate that allows us to take a limit of appropriate scalings of the metrics $g_{ij}^{\alpha}$ on $[T-h^{2}(x^{\alpha}),T]$ about $x^{\alpha},$ for a subsequence of $\alpha\to\infty.$ By shifting the time parameter we may assume that the limit is defined on $[-1,0].$ Clearly, for each time in this interval, the limit is a complete manifold with nonnegative sectional curvature; moreover, since $x^{\alpha}$ was contained in an $\epsilon$-horn with boundary in $\Omega^{\alpha}_{\rho},$ and $h(x^{\alpha})/\rho\to 0,$ this manifold has two ends. Thus, by Toponogov, it admits a metric splitting $\mathbb{S}^{2}\times\mathbb{R}.$ This implies that the canonical neighborhood of the point $(x^{\alpha},T-h^{2}(x^{\alpha}))$ is also of type (a), that is a strong $\epsilon$-neck, and we can repeat the procedure to get the limit, defined on $[-2,0],$ and so on. This argument works for the limit in any finite time interval $[-A,0],$ because $h(x^{\alpha})/\rho\to 0.$ Therefore, we can construct a limit on $[-\infty,0];$ hence it is the round cylinder, and we get a contradiction.