diff 38956c6..370b521 bounded_scalar_ricci_flow_v2.txt
diff --git a/bounded_scalar_ricci_flow_v2.txt b/bounded_scalar_ricci_flow_v2.txt
index da8bc83..c6b616e 100644
--- a/bounded_scalar_ricci_flow_v2.txt
+++ b/bounded_scalar_ricci_flow_v2.txt
@@ -5,8 +5,7 @@
\usepackage{xcolor}
\usepackage[colorlinks=true,linkcolor=blue!45!black,citecolor=green!35!black,
urlcolor=blue!55!black,bookmarksnumbered=true]{hyperref}
-\numberwithin{equation}{section}
-\newtheorem{theorem}{Theorem}[section]
+\newtheorem{theorem}{Theorem}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
@@ -20,12 +19,15 @@
\DeclareMathOperator{\supp}{supp}
\DeclareMathOperator{\tr}{tr}
+
+\begin{document}
+
+
\title{Bounded scalar curvature and smooth extension of four-dimensional Ricci flow}
\author{OpenAI}
\date{September 24, 2026}
-
-\begin{document}
\maketitle
+
\begin{abstract}
We prove that a smooth Ricci flow on a closed real four-manifold extends on the
same manifold through every finite time at which its scalar curvature stays
@@ -35,22 +37,44 @@ of the degenerating slices with a renormalized Einstein--Hilbert functional and
a single monotone quantity, the spacetime $L^2$ norm of the Ricci tensor in the
outermost bubble.
\end{abstract}
-\tableofcontents
-\section{Introduction}\label{sec:intro}
-
-A Ricci flow is a family of Riemannian metrics solving
+\par\medskip\noindent\textbf{Strategy of the proof.}
+
+The proof follows one strategy. Fix a hypothetical concentration point
+where curvature blows up at $T$. Successive rescalings produce a finite
+tree of Ricci-flat limits; the outermost nonflat limit defines a scale
+$a(t)$ that cannot change by any fixed factor on short intervals. On the
+corresponding buffered intervals, parabolic rescaling places the Ricci
+energy $K$ of the outermost bubble in a limiting spacetime where the
+renormalized Einstein--Hilbert functional and the companion tree inequality
+force $K\to 0$; that pins $a(t)$ and rules out the concentration point.
+The mechanics are in four blocks. Block~I (``The spatial curvature
+radius'' to ``Buffered intervals'') estimates curvature at all scales,
+describes the slice limits, and selects the buffered intervals. Block~II
+(``Concentration estimates'' to ``The maximal function'') bounds the Ricci
+heat equation and the root energy. Block~III (``Collar extension'')
+builds and estimates the analytic exterior from a fixed collar. Block~IV
+(``Functional change and the contradiction'' to ``Continuation'') supplies
+the lower and upper bounds on the functional variation and closes the
+argument. A reader focused on the main idea should read Blocks~I and
+IV first, treating Block~III as a black box for the error analysis.
+
+\par\medskip\noindent\textbf{Introduction.}
+
+A Ricci flow is a one-parameter family of Riemannian metrics solving the
+heat-type equation
\begin{equation}\label{eq:flow}
\partial_t g=-2\Ric(g).
\end{equation}
-On a closed manifold it is smooth for a short time; the question is what makes
-the maximal smooth existence time $T<\infty$ finite. The full curvature tensor
-controls continuation: bounded $|\Rm|$ implies continuation
-\cite{Hamilton1982}. The scalar curvature $R=\tr_g\Ric$ only controls volume,
-$\partial_t\,d\mu_g=-R\,d\mu_g$, and a bound on $R$ is far weaker than one on
-$\Ric$ or $\Rm$. The scalar-curvature extension problem asks whether a uniform
-bound on $R$ alone already prevents a finite-time singularity. We answer it
-affirmatively in real dimension four.
+On a closed manifold it exists and is smooth for a short time; the central
+question is what makes the maximal smooth existence time $T<\infty$ finite.
+The full curvature tensor decides this: if $|\Rm|$ stays bounded on $[0,T)$,
+the flow continues past $T$ \cite{Hamilton1982}. The scalar curvature
+$R=\tr_g\Ric$ is only a trace, so a bound on $R$ is far weaker; it controls
+volume by $\partial_t\,d\mu_g=-R\,d\mu_g$ but leaves the full geometry free.
+The scalar-curvature extension problem asks whether a uniform bound on $R$
+alone already forbids a finite-time singularity. The answer is affirmative
+in real dimension four.
\begin{theorem}[Smooth extension]\label{thm:extension}
Let $M$ be a closed connected smooth four-manifold, $0<T<\infty$, and let
@@ -73,7 +97,22 @@ $R$ is smooth on the compact initial part, Theorem~\ref{thm:extension} applied
componentwise would contradict maximality of $T$.
\end{proof}
-\subsection{The geometric obstruction}
+The proof rules out contraction of a specific outermost bubble scale. A
+concentration point produces a finite rooted tree of Ricci-flat limits, one for
+each scale at which curvature blows up; the outermost nonflat limit has a
+length $a(t)$ that is positive before $T$ but tends to zero at the singular
+time. On short intervals $a$ cannot change by a fixed factor: rescaling would
+freeze the pointed geometry, so any fixed-factor change would contradict the
+identity of a fixed limiting tree. Parabolic rescaling on the buffered
+intervals therefore produces a limiting spacetime in which the Ricci energy $K$
+of the outermost bubble is comparable from start to end. The renormalized
+Einstein--Hilbert functional computed on that scaling, together with a sharp
+inequality from the companion paper, forces $K\to0$. But $K\to0$ means
+distances are preserved on the concentration scale, which pins $a(t)$ and
+contradicts its prescribed finite change. Hence no concentration point exists
+and the flow extends smoothly.
+
+\par\medskip\noindent\textbf{The geometric obstruction.}
Hamilton \cite{Hamilton1982} proved short-time existence; \v{S}e\v{s}um
\cite{Sesum2005} showed a uniform bound on $\Ric$ forces continuation. A scalar
@@ -90,22 +129,22 @@ Simon \cite{SimonExtension} continued through the limiting space by an orbifold
flow; Bamler's convergence theory \cite{BamlerConvergence} gives smooth
convergence off a codimension-four set; Liu--Simon \cite{LiuSimon} treat
four-dimensional convergence under scalar-integrability hypotheses. All allow
-concentrated curvature and orbifold points; our theorem instead continues the
-given flow on its original manifold. Blow-up is known for K\"ahler--Ricci flows
+concentrated curvature and orbifold points; the continuation below instead
+keeps the given flow on its original manifold. Blow-up is known for K\"ahler--Ricci flows
\cite{Zhang2010Kahler}, for global Type~I \cite{EndersMuellerTopping2011}, and
-for local Type~I points \cite{BuzanoDiMatteo2026}; we impose no rate.
+for local Type~I points \cite{BuzanoDiMatteo2026}; no rate is imposed.
The spatial picture descends from Einstein four-manifold theory: Anderson
\cite{Anderson1989}, Bando \cite{Bando1990,BandoCorrection1990},
Bando--Kasue--Nakajima \cite{BandoKasueNakajima1989}, Cheeger--Tian
-\cite{CheegerTian2006}. Our slices are not Einstein, so we reprove the needed
-statements under a scalar bound, retaining the Ricci forcing.
+\cite{CheegerTian2006}. The slices below are not Einstein, so the needed
+statements are reproved under a scalar bound, retaining the Ricci forcing.
-\subsection{The outermost scale and the static functional}
+\par\medskip\noindent\textbf{The outermost scale and the static functional.}
Successive rescalings at a concentration point produce a finite rooted tree of
Ricci-flat pieces; the outermost nonflat vertex has a length scale $a(t)$
-(Section~\ref{sec:geometry}). The proof rules out contraction of $a(t)$ while
+described below. The proof rules out contraction of $a(t)$ while
allowing arbitrarily small descendants.
Retain the true metric through an annular collar of radius $b=Na$ ($N$ large,
@@ -119,8 +158,8 @@ the mass term of Deruelle--Ozuch \cite[Definition~4.2, Proposition~4.4]
{DeruelleOzuch2025}. It scales by the square of the length unit and its first
variation pairs the velocity with $\tfrac12Rg-\Ric$. The companion article
\cite{StaticTrees} proves the sequential inequality $E=o(M)$ on each fixed
-limiting tree ($M$ the weighted, dimensionless Ricci error). Section~\ref{sec:static}
-records the input; we must verify the hypotheses for flow metrics and compare
+limiting tree ($M$ the weighted, dimensionless Ricci error). The static input is
+recorded below; the argument must verify the hypotheses for flow metrics and compare
the functional variation with spacetime Ricci energy.
Two estimates drive the argument. On a nearly flat annulus the contracted
@@ -131,25 +170,24 @@ built as an analytic function of the metric on the one fixed material collar,
with the longer exterior entering only an error estimate; hence the extension
differentiates in time on patches without a smoothly varying tree. This rests
on the Saint-Venant relation \cite{Khavkine2017} and Korn control modulo rigid
-motions \cite{CiarletCiarlet2005} (Section~\ref{sec:collar}).
-
-\subsection{Proof structure and conventions}
-
-Section~\ref{sec:static} fixes the static input. Section~\ref{sec:geometry}
-constructs scale estimates, genuine orbifold limits, finite trees, and the
-continuous scale $a(t)$, and selects buffered intervals on which $a$ changes by
-a fixed factor with root length tending to zero after parabolic normalization.
-Section~\ref{sec:parabolic} proves the Ricci-equation estimates, the exterior
-decay beyond the outermost scale, and the maximal/weighted interior estimates
-controlling the root energy. Section~\ref{sec:collar} builds the exterior
-extension and its error. Section~\ref{sec:contradiction} compares functional
+motions \cite{CiarletCiarlet2005}, both used in the collar construction below.
+
+\par\medskip\noindent\textbf{Proof structure and conventions.}
+
+The static input is fixed first. The scale estimates, genuine orbifold limits,
+finite trees, and the continuous scale $a(t)$ follow, together with buffered
+intervals on which $a$ changes by a fixed factor with root length tending to
+zero after parabolic normalization. The Ricci-equation estimates come next:
+the exterior decay beyond the outermost scale, and the maximal and weighted
+interior estimates controlling the root energy. The collar construction then
+builds the exterior extension and its error. A final comparison of functional
variation with Ricci energy: an explicit lower bound $cK^2$ and upper bound
$\zeta_i(K+1)^2$ force $K\to0$; then distances are preserved so the
curvature-radius maximum defining $a$ is preserved, contradicting its
prescribed change, while absence of a concentration point gives a smooth
terminal metric and continuation on $M$.
-We write $\Delta=\tr\nabla^2$ and $|A|_{k,r}=\sum_{j=0}^k r^j|\nabla^jA|$ (the
+Throughout, $\Delta=\tr\nabla^2$ and $|A|_{k,r}=\sum_{j=0}^k r^j|\nabla^jA|$ (the
pointwise magnitude unchanged, derivatives scaled). Quotient-chart components
are Cartesian components on the Euclidean cover; constructions are equivariant
and descend. Constants may depend on a fixed finite differentiation order and,
@@ -157,13 +195,13 @@ after extraction, on a fixed finite tree; uniformity in a sequence index, length
or collar factor $N$ is stated where needed, and a sequential little-oh is taken
after fixing the parameters in its statement.
-\section{The static input and its normalization}\label{sec:static}
+\par\medskip\noindent\textbf{The static input and its normalization.}
-We record the input from \cite[Theorem~2.2]{StaticTrees}. Constants and small
+The input from \cite[Theorem~2.2]{StaticTrees} is recorded here. Constants and small
exponents belong to one fixed limiting tree; no uniform exponent across trees is
needed.
-\subsection{Geometric data and the functional}
+\par\medskip\noindent\textbf{Geometric data and the functional.}
Let $\mathcal T$ be a finite rooted tree of four-dimensional Ricci-flat spaces
$(V_v,h_v)$. A vertex has one exterior end and one punctured end per child.
@@ -175,8 +213,8 @@ $\tau=-\log r$ at a puncture. In vertex units,
\begin{equation}\label{eq:end-decay}
h_v=\delta+O(e^{-\mu\tau})
\end{equation}
-on joined ends, with scaled derivatives, for some $\mu>0$; on the unjoined
-exterior end of the root we require $h_v-\delta=O(r^{-q})$ with $q>1$. Bounds
+on joined ends, with scaled derivatives, for some $\mu>0$; the unjoined
+exterior end of the root requires $h_v-\delta=O(r^{-q})$ with $q>1$. Bounds
through order $20$ suffice.
For an edge from a parent $p$ to a child $c$ truncate both end coordinates at
@@ -242,12 +280,12 @@ This is \cite[Theorem~2.2]{StaticTrees}. For the fixed tree and sequences above
and for $M=0$ it means $E(g)=0$.
\end{theorem}
-\section{Small-scale geometry and the outermost scale}\label{sec:geometry}
+\par\medskip\noindent\textbf{Small-scale geometry and the outermost scale.}
Throughout, $(M^4,g(t))$, $0\le t<T$, is the closed flow with $|R|\le C_R$.
Constants may depend on the flow, $T$, and a fixed differentiation order.
-\subsection{The spatial curvature radius}
+\par\medskip\noindent\textbf{The spatial curvature radius.}
\begin{definition}\label{def:radius}
For a small cap $l_0>0$ and a material point $x$,
@@ -280,7 +318,7 @@ unique outermost entry, and the joining regions carry quantitative neck charts.
The proof occupies this section.
-\subsection{Heat localization and two-sided persistence of the radius}
+\par\medskip\noindent\textbf{Heat localization and two-sided persistence of the radius.}
Fix the terminal time in the smooth part and use $h$-dependent conjugate heat
kernels of the fixed metric, then localize them on radius-$h$ balls by a cutoff
@@ -329,12 +367,8 @@ Chern--Gauss--Bonnet $\int|\Rm|^2\le C+4\int|\Ric|^2\le C+CF$
\cite{Chern1944,SimonIntegral2020} (orientation cover if needed) and Gronwall
bound $F$; this is \eqref{geom:volume-energy}.
-\emph{Bad set and fixed energy.} For small $\rho$, $\{l<\rho\}$ is covered by
-boundedly many radius-$C\rho$ balls; otherwise disjoint unit balls centered in
-$\{C\rho\le l<\rho\}$ with a primitive $x\mapsto l(x)\psi(\rho/l(x))$ give
-disjoint definite-energy balls, contradicting finite energy. A point with
-$r^2|\Rm|\ge\varepsilon$ in a controlled ball has energy at least
-$c\varepsilon$ on a smaller ball, so there is a fixed energy $e_0>0$.
+\emph{Bad set and fixed energy.} If $l(y)\ge r$ and $r^2|\Rm|(y)\ge\varepsilon$, then the derivative bound gives $|\Rm|\ge\varepsilon/(2r^2)$
+on $B(y,cr)$ and $\int_{B(y,c'r)}|\Rm|^2\ge c\varepsilon^2r^4$; volume comparability upgrades this to a definite Ricci energy $e_0>0$ on a smaller ball. Now fix small $\rho$ and suppose $\{l<\rho\}$ needs more than $M$ radius-$C\rho$ balls. Take a maximal family of disjoint unit balls centered at points $x_j\in\{C\rho\le l<\rho\}$ chosen by the primitive $x\mapsto l(x)\psi(\rho/l(x))$; the derivative bound on this primitive gives disjoint definite-energy balls, and the slice energy bound yields $M\le Ce_0^{-1}$. Hence $\{l<\rho\}$ is covered by boundedly many radius-$C\rho$ balls, with the bound independent of $\rho$.
\emph{Convergence of distances and volume comparison.} For $|t-s|\le cr^2$,
cover the bad set at the first time by bounded radius-$Cr$ balls, replace visits
@@ -344,6 +378,12 @@ curvature scale is $\ge cr$ and the length changes by a factor $1+Cr$ since
$|R|\le C/r$. Comparing both ways,
$|d_t(x,y)-d_s(x,y)|\le Cr(1+d_s+d_t)$, and taking $r\asymp\sqrt{|t-s|}$ gives
uniform convergence to a pseudodistance $d_T$ and a compact quotient $Z$.
+
+The slicewise $L^3$ bound $\int_{B(x,h)}|\Ric|^3\le Ch$ will also be needed. On
+the shell $\{r_j\le l\le r_{j-1}\}$, $|\Ric|\le Cr_j^{-1}$ and the bad-set
+measure bound established just above gives measure $\le Cr_j^4$, contributing
+$Cr_j$. Summing the dyadic shells up to $h$ yields $C\sum_jr_j\le Ch$.
+
In polar coordinates the polar Jacobian $P$ and $m=(\partial_r\log P-3/r)_+$
satisfy $(m^5P)'\le5|\Ric|m^4P-7m^5P/r-(2/3)m^6P$, giving
$\int m^6\le C\int|\Ric|^3\le Ch$ and hence, since $(P/r^3)'\le mP/r^3$,
@@ -353,7 +393,7 @@ $\int m^6\le C\int|\Ric|^3\le Ch$ and hence, since $(P/r^3)'\le mP/r^3$,
while polar integration including the cut measure gives $\Delta(d^2/2)\le4+dm$
weakly. Volume forms at two times differ by $e^{\pm C_R|t-s|}$.
-\subsection{Slice limits, cones, and quantitative necks}
+\par\medskip\noindent\textbf{Slice limits, cones, and quantitative necks.}
Rescaled slices $h_i^{-2}g(t_i)$ based at $x_i$ with $h_i\to0$, $t_i\uparrow T$
are precompact as length spaces (two-sided volume bounds bound the number of
@@ -361,7 +401,7 @@ disjoint balls); the bad-set cover has bounded cardinality so a diagonal limit
identifies a finite exceptional set $S$. Restricted maps are locally
bi-Lipschitz, extend by continuity, and preserve the injectivity radius bound, so
the limit is complete; smooth convergence off $S$ gives a smooth Ricci-flat
-metric there, which we call regular. Ball volumes converge and are regular
+metric there, called regular. Ball volumes converge and are regular
Riemannian volume. The polar defect vanishes and Bishop becomes exact on the
regular part of the limit.
@@ -413,7 +453,7 @@ contradiction. Derivatives follow by interior elliptic estimates; slowly varyin
boundary coefficients are absorbed by decreasing $\mu$.
\end{proof}
-\subsection{Removal of punctures}
+\par\medskip\noindent\textbf{Removal of punctures.}
Work near a punctured orbifold point. A constant rescaling makes $g(0)=\delta$;
\eqref{eq:neck-decay} gives $|g-\delta|=O(r^\mu)$, $|Dg|=O(r^{\mu-1})$,
@@ -452,7 +492,7 @@ develops to a global Euclidean quotient with at most one tip (finiteness of the
deck group from sheet counting and noncollapse; the only possible singular point
is the tip).
-\subsection{Terminal convergence, empty set, finite trees}
+\par\medskip\noindent\textbf{Terminal convergence, empty set, finite trees.}
For a compact set $K$ off the finite exceptional set, nested compactness and a
fixed radius $r$ control the whole remaining interval $[t_i,T)$ by spatial
@@ -477,7 +517,7 @@ Lemma~\ref{geom:neck} supplies the necks with positive decay and the matching
quotient groups (end variables $\tau_p=\log(A/r)$, $\tau_c=\log(r/(Ae^{-2L}))$,
$\tau_p+\tau_c=2L$ realize exactly the static half-neck weights).
-\subsection{The outermost scale}
+\par\medskip\noindent\textbf{The outermost scale.}
\begin{proposition}[A continuous outermost scale]\label{prop:root-scale}
Fix a small $c_*>0$ and a bounded continuous $\psi\ge0$ with $\psi=0$ on
@@ -506,7 +546,7 @@ nonflat limit has scale $\le Ca$ by the same detection, and the maximizing limit
shares its unique outer ancestor with the maximal list, so it is outermost.
\end{proof}
-\subsection{Motion on short intervals}
+\par\medskip\noindent\textbf{Motion on short intervals.}
\begin{lemma}[Curvature radius along minimizing segments]\label{lem:geodesic-radius}
For any minimizing segment $\gamma$ at a late time,
@@ -555,7 +595,7 @@ stops, maximizers stay in a common bounded root range, and the rescaled supremum
of \eqref{eq:root-def} agrees at both times, excluding a fixed-factor change.
\end{proof}
-\subsection{Buffered intervals, motion, and exterior geometry}
+\par\medskip\noindent\textbf{Buffered intervals, motion, and exterior geometry.}
\begin{proposition}[Buffered intervals]\label{prop:buffered-intervals}
Fix $u>0$ and choose $t_0\le t_1\le\cdots<t_N<T$ a maximal $u$-separated
@@ -604,7 +644,7 @@ shortened annuli. Since $\epsilon_iR_i^{\rm out}\to0$ and $\mu<1$,
$\delta_i/\epsilon_i\to\infty$.
\end{proof}
-\section{Concentration estimates for the Ricci equation}\label{sec:parabolic}
+\par\medskip\noindent\textbf{Concentration estimates for the Ricci equation.}
Let $G=\eta\Ric\in C^2_0$, $\eta$ a spatial cutoff of fixed width, maximal on the
$N_1a$-ball and supported in the $2N_1a$-ball, and
@@ -627,7 +667,7 @@ $G_r=r^{-2}\Phi_r^*G$; then \eqref{eq:ricci-improvement} gives
\end{equation}
so the cutoff source needs no division by $\eta$.
-The geometric inputs from Section~\ref{sec:geometry}, taken sequentially, are:
+The geometric inputs established above, taken sequentially, are:
(i) on regular charts with $l\ge ch$, coefficients move by
$O(\epsilon\Delta t/h)$ over windows; (ii) centers move $O(a)$ at root scale and
$o(h)$ for $a/h\to0$; (iii) distance annuli $Ca\le r\le C'$ carry the quotient
@@ -697,7 +737,7 @@ bounds $\overline S$ and its radial derivatives, hence the Cartesian derivatives
Conversion to covariant derivatives adds only metric-error terms.
\end{proof}
-\subsection{The exterior estimate}
+\par\medskip\noindent\textbf{The exterior estimate.}
For dyadic radii based on $a$ let
$\mathcal A_t(r)=\{r/2\le d_t(x_t,y)\le2r\}$ and
@@ -741,7 +781,7 @@ absorbing into $M/4$. Choices are made in order $H,T_0,N_0$, late index; then
$M\le C+M/2$ proves \eqref{eq:exterior-v} and \eqref{eq:exterior-a}.
\end{proof}
-\subsection{The maximal function controlling the root}
+\par\medskip\noindent\textbf{The maximal function controlling the root.}
Define the centered maximal function of $f=\eta|\Ric|_{j+2,a}$ (with
$f^p$ for $1<p\le2$): $H_*(t)=(\iint_{B_t(x_t,D_1a)}f^p)^{1/p}$ over the past
@@ -782,7 +822,12 @@ $D$-exponents $2\lambda,(4-2/p)\lambda,(2-2/p)\lambda,\delta-(2/p)\lambda$ yield
$r^2|\Ric|_{j,r}\le C\epsilon r\le o_i(1)e^{-\nu D}$ holds with $\nu<1$.
\end{proof}
-\section{An exterior extension determined by a fixed collar}\label{sec:collar}
+\par\medskip\noindent\textbf{An exterior extension determined by a fixed collar.}
+
+An exterior is attached to the true metric on a fixed annular collar, defined as
+an analytic function of the collar data. Smoothness in the collar metric lets
+the extension be differentiated in time without choosing a smoothly varying
+tree; the far exterior enters the error estimate only.
All lengths are in collar units; $C_1=\{1\le|x|\le2\}/\Gamma$,
$\mathcal E=\{|x|\ge1\}/\Gamma$, $\Gamma\subset O(4)$ finite acting freely on
@@ -806,7 +851,7 @@ independent of $R$, the weights, and the exterior. There is also an analytic
extension $\hat h$ equal to $g$ near the inner boundary with the same bounds.
\end{theorem}
-\subsection{Local linear estimates}
+\par\medskip\noindent\textbf{Local linear estimates.}
With $Sv,\mathcal K(v)$ the flat linearizations of Ricci and Riemann curvature
and $Bv=\partial^iv_{ij}-\tfrac12\partial_j\tr_\delta v$,
@@ -868,7 +913,7 @@ $|G(x,y)|\le C|x-y|^{-2}$; one shell contributes at most $Cc_r$ everywhere, so t
maximum principle and local estimate give the shell-uniform bound.
\end{proof}
-\subsection{A filled-space inverse and the analytic map}
+\par\medskip\noindent\textbf{A filled-space inverse and the analytic map.}
Let $\mathcal E_0$ contain the exterior and the filled unit ball with one smooth
metric piece; if $R$ is the outer radius use larger balls as needed. For
@@ -920,7 +965,7 @@ solvable, real analytic in the data. The metric $\bar h=(1-B\mathcal N)\delta$
depends analytically on the collar data with the stated bounds, and the extension
is independent of the exterior realizing the collar.
-\subsection{A comparison gauge and the residual}
+\par\medskip\noindent\textbf{A comparison gauge and the residual.}
For tensors on $1\le|x|\le R$ there are linear $P_0$, $\mathcal Dz=\mathcal L_{P_0z}\delta$,
zero on $C_1$, with
@@ -952,7 +997,7 @@ factors $O(\theta_r)$, last $r^2Y$), giving the residual bound
\le C\big(m_1+\sigma N_0+R^{-q-2}\big).
\end{equation}
-Run $L$ on $d_1=(p_1,f)$, $e=\|d_1\|_{\mathcal Y}$; $ALd_1=0$. We show
+Run $L$ on $d_1=(p_1,f)$, $e=\|d_1\|_{\mathcal Y}$; $ALd_1=0$. It remains to show
$d_1-ALd_1$ is small. With $u=\Delta^{-1}(-2f^\#)$,
$\|Bu\|_{H^{k-1}_{q+1}}\le CD_0$ ($D_0=m_1+\sigma e$), and
$Su-f^\#=\tfrac12\mathcal L_{(Bu)^\sharp}\delta$ gives
@@ -976,7 +1021,7 @@ $e\le Cm_1$ with length-independent constants, proving \eqref{eq:collar-extensio
An interpolation $\hat h=\bar h+\zeta(g-\bar h)$ with a fixed radial $\zeta$
gives the analytic extension $\hat h$.
-\section{Functional change and the contradiction}\label{sec:contradiction}
+\par\medskip\noindent\textbf{Functional change and the contradiction.}
Apply the results to a buffered interval $I=[-\theta,1+\theta]$ in rescaled
time, $\epsilon_i=\sqrt{d_i}$, $a=a_i\to0$; the root scale is comparable to $a$
@@ -1109,7 +1154,7 @@ $K_i^2\le\gamma_i(K_i+1)^2$, giving $K_i\le\sqrt{\gamma_i}/(1-\sqrt{\gamma_i})\t
and boundedness of $K$ in passing.
\end{proof}
-\subsection{Small energy fixes the root geometry}
+\par\medskip\noindent\textbf{Small energy fixes the root geometry.}
With $K_i\to0$, the pointed root geometry is preserved: for $x,y$ in a small
neighborhood of the concentration point and $d_t(x,y)\le Ca$, integration of
@@ -1121,7 +1166,7 @@ curvature-radius maximum defining $a$ is preserved across $I$, contradicting the
prescribed factor change. Hence a buffered interval with fixed-factor change
cannot exist, so no concentration point survives.
-\subsection{Continuation}
+\par\medskip\noindent\textbf{Continuation.}
Absence of a concentration point gives a smooth limiting metric on $M$ at $T$
(finitely many curvature-bounded charts, the DeTurck principal part, and the