Formation of Trapped Surfaces in General Relativity

Formation of Trapped Surfaces
in General Relativity
Xinliang An
A Dissertation
Presented to the Faculty
of Princeton University
in Candidacy for the Degree
of Doctor of Philosophy
Recommended for Acceptance
by the Department of
Mathematics
Adviser: Sergiu Klainerman
June 2014

c Copyright by Xinliang An, 2014.
All Rights Reserved

Abstract
In this thesis we present two results regarding the formation of trapped surfaces in general
relativity.
The first is a simplified approach to Christodoulou’s breakthrough result which showed that
trapped surfaces can form dynamically by the focusing of gravitational radiation from past null
infinity. We extend the methods of Klainerman-Rodnianski, who gave a simplified proof of this
result in a finite region.
The second result extends the theorem of Christodoulou by allowing for weaker initial data but
still guaranteeing that a trapped surface forms in the casual domain. In particular, we show that a
trapped surface can form dynamically from initial data which is merely “large” in a scale-invariant
way. The second result is obtained jointly with Luk.
iii
Acknowledgements
First and foremost, I would like to convey my gratitude to my advisor Professor Sergiu Klainerman
for bringing me into this field with great support and encouragement. His vision and insight for
mathematics always motivate me to explore the mathematical world.
I am very grateful to my collaborator and great friend, Jonathan Luk for all those enlightening
discussions and help in my PhD years.
I would like to thank Professors Mihalis Dafermos, Jérémie Szeftel, who shared their under-
standings and gave great suggestions and support to my study.
Sincere thanks also go to Professors Spyros Alexakis, Kung-Ching Chang, Peter Constantin,
Greg Galloway, Zheng-Chao Han, Gustav Holzegel, Alexandru Ionescu, Michael Kiessling, Yanyan
Li, Elliot Lieb, Hans Lindblad, Mao-Kang Luo, Shige Peng, Frans Pretorius, Jie Qing, Hans
Ringström, Igor Rodnianski, Avraham Soffer, Jared Speck, Shadi Tahvildar-Zadeh, Gang Tian,
Mu-Tao Wang for many insightful conversations and for their support.
I want to thank many friends in the analysis group at Princeton: John Anderson, Stefanos
Aretakis, Tom Beck, Yu Deng, Ross Granowski, Georgios Moschidis, Luc Nguyen, Hiro Oh, Sung-
Jin Oh, Oana Pocovnicu, Volker Schlue, Arick Shao, Yakov Shlapentokh-Rothman, John Stogin,
Martin Taylor, Qian Wang, Xuecheng Wang, Willie Wong, Polam Yang, Shiwu Yang, Pin Yu,
Ruixiang Zhang, Ruobing Zhang, who made my PhD years very joyful and unforgettable.
I also would like to thank Shiguang Ma, Jinxing Xu, Bohua Zhan, Ting Zhang and Xuefeng
Zhang for the friendship along the way.
I am grateful to our department Chair Professor Gabai, graduate administrator Jill LeClair and
Jennifer Johnson for their help and support.
Last, but certainly not least, I would like to thank my family. To my father Yuqing An and
my mother Pengqian Zhang for raising me and supporting me to chase my dreams; To my sister
Xiaozhu An for always being a loyal friend; To my fiancée Chen Chen, for her unconditional love
and company along these years.
iv
to my family.
v
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
1 Introduction 1
1.1 Formation of Trapped Surfaces from Past Null Infinity . . . . . . . . . . . . . . . . . 3
1.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Heuristic Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Formation of Trapped Surfaces in Vacuum from Mild Incoming Radiation
(Joint with Luk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.3 Heuristic Argument with ESFS-S . . . . . . . . . . . . . . . . . . . . . . . . . 19
2 Formation of Trapped Surfaces from Past Null Infinity 28
2.1 Setting, Equations and Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.1 Double Null Foliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1.2 The Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.1.3 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1.4 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.5 Signatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
vi
2.1.6 Scale Invariant Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.7 Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 Statement of Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.1 Structure of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.2.2 Strategy of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3 The Preliminary Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 Estimates for Metric Components . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.2 Estimates for Transport Equations . . . . . . . . . . . . . . . . . . . . . . . . 50
2.3.3 Sobolev Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.4 Commutation Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.5 General Elliptic Estimates for Hodge Systems . . . . . . . . . . . . . . . . . . 57
2.4 Ricci Coefficient Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4.1 Estimates for Zero Derivatives of Ricci Coefficients . . . . . . . . . . . . . . . 62
2.4.2 Estimates for First Derivatives of Ricci Coefficients . . . . . . . . . . . . . . . 72
2.4.3 L∞ (S) Estimates for Ricci Coefficients . . . . . . . . . . . . . . . . . . . . . . 88
2.5 Elliptic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.5.1 L4 (S) Estimates for Curvature Components . . . . . . . . . . . . . . . . . . . 95
2.5.2 L2 (S) Estimates for First Derivatives of Curvature Components . . . . . . . 98
2.5.3 L2 (S) Estimates for Third Derivatives of Ricci Coefficients . . . . . . . . . . 102
2.5.4 L4 (S) Estimates for Second Derivatives of Ricci Coefficients . . . . . . . . . . 135
2.6 Curvature Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
2.6.1 Curvature Estimates in the Scale Invariant Norms . . . . . . . . . . . . . . . 138
2.6.2 Curvature Estimate I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
2.6.3 Curvature Estimate II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
2.6.4 Curvature Estimate III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
2.7 Formation of Trapped Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
2.8 Retrieving Christodoulou’s Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
2.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
vii
2.9.1 Norms in standard Lp form . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
2.9.2 Equations for Elliptic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 180
2.9.3 Equations for Energy Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . 190
3 Formation of Trapped Surfaces in Vacuum from Mild Incoming Radiation
(Joint with Luk) 197
3.1 Setting, Equations and Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
3.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
3.1.2 Double Null Foliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
3.1.3 The Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
3.1.4 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.1.5 Schematic Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
3.1.6 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
3.1.7 Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
3.2 Statement of main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
3.2.1 Structure of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
3.2.2 Strategy of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
3.3 The Preliminary Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
3.3.1 Estimates for Metric Components . . . . . . . . . . . . . . . . . . . . . . . . 214
3.3.2 Estimates for Transport Equations . . . . . . . . . . . . . . . . . . . . . . . . 216
3.3.3 Sobolev Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
3.3.4 Commutation Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
3.3.5 General Elliptic Estimates for Hodge Systems . . . . . . . . . . . . . . . . . . 223
3.4 Estimates for Ricci Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
3.5 Elliptic estimates for the fifth derivatives of the Ricci coefficients . . . . . . . . . . . 244
3.6 Estimates for Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
3.7 Formation of Trapped Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
viii
Chapter 1
Introduction
My thesis is motivated by a long-standing and central question in general relativity: the dynamical
formation of black holes. This is a large data problem for the Einstein field equations
1
Rµν − gµν R = 8πTµν , (1.0.1)
2
where Rµν and R denote the Ricci and scalar curvatures of the metric gµν , and Tµν denotes the
stress-energy-momentum tensor of matter. The Einstein field equations are systems of supercrit-
ical quasilinear partial differential equations. The objects of study are 4-dimensional Lorentzian
manifolds (M, g) called spacetimes. These equations describe the gravitation that results from
spacetime being curved by matter and energy. When Tµν ≡ 0, (1.0.1) is named the Einstein vac-
uum equations (EVE). One of the most fascinating features of (1.0.1) is that it admits so called
black hole spacetimes. Informally, black holes are regions of spacetime where the gravitational field
is so strong that even light cannot escape and hence invisible to far-away observers. Famous explicit
solutions of EVE containing black hole regions include the Schwarzschild and Kerr families. The
explicit properties of these solutions are well-known to mathematicians and physicists. At present,
however, understanding the formation of black holes remains one of the most important unfinished
tasks in the theory of relativity. Nevertheless, encouraging progress has recently been made with
1
respect to formation of trapped surfaces. A trapped surface is defined as a two dimensional space-
like sphere whose outgoing and incoming expansions are negative. This is a local object which,
heuristically, indicates the formation of a black hole to the future.
In a 589-page breakthrough result [7] in 2008, Christodoulou proves: without any symmetry
assumption, trapped surfaces form in the Cauchy development of vacuum initial data which are
arbitrarily dispersed. In [7], Christodoulou presents a novel technique called the short pulse method.
He introduces a certain large amplitude in the data, the pulse, compensated by a small parameter
δ which controls the length of the outgoing direction. This gives a hierarchy of large and small
quantities, which are coupled via the EVE. In [15], Klainerman and Rodnianski extend the above
result and significantly simplify the proof in a finite region of spacetime. In my works [1] and [2], I
first extend the result of Klainerman and Rodnianski to an infinite region and use a new and more
direct way to re-prove the main theorem in Christodoulou’s monograph [7]. Second, jointly with
Luk, I use the methods developed in [1] to provide a scale invariant criterion for the formation of
trapped surfaces. This allows us to answer the following question: what is the minimum initial
“mass density” required to guarantee that a trapped surface forms in evolution?
In [1], I extend the results in [15] which are concerned with a finite region of spacetime by
showing analogous results from an ideal conformal boundary at infinity (so-called past null infinity).
One defines and studies gravitational waves from past null infinity by understanding decay rates
of various geometric quantities. For the EVE there are fourteen separate components (six from
curvature, eight from connection) which all behave differently. To overcome the difficulties coming
from the complexity of the EVE and to capture the decay properties of their components, I extend
the methods introduced by Klainerman and Rodnianski in [15] and assign to each component a
number which carries the information of decay, called the signature for decay rates. Using the
signature, I define a collection of scale invariant norms, which enable me to explore the hidden
smallness of interactions between various geometric quantities near past null infinity. Hence, except
for rare terms, which are called anomalies and left for further analysis, I treat most terms in a
simple and direct manner. This reduces the workload significantly.
Moreover, based on a relation between borderline (the most anomalous) terms’ coefficients and
2
their signatures for decay rates I extend a method employed by Holzegel, Luk and Rodnianski and
offer a more direct and intuitive way to obtain energy estimates. This approach is capable for
studying decay rates in far-from-symmetry cases.
The methods I exploit in [1] to study decay rates also fit well for studying blow-up rates close
to the center of gravitational collapse. In [2], together with Luk, I prove that with an initial
mass density of size δ, a trapped surface of radius δ forms dynamically. By introducing two new
parameters and corresponding hierarchies and a number of renormalizations, we extend the short
pulse method introduced by Christodoulou in [7].
The following subsections describe results in [1] and [2] in more details.
1.1 Formation of Trapped Surfaces from Past Null Infinity
1.1.1 Background
A celebrated solution to the Einstein vacuum equations
Ric(g) = 0 (1.1.1)
is the Schwarzschild metric:
−1
2M 2M
g =− 1− dt2 + 1 − dt2 + r2 dS,
r r
where dS denotes the standard metric on unit sphere and M and r represent mass and the radial
coordinate, respectively.
In the Schwarzschild spacetime, the three dimensional hypersurface r = 2M represents the
boundary of an interior region. Every timelike or null geodesic γ(s) starting from the interior
region (r < 2M ) is confined in the region r ≤ 2M . The interior region of Schwarzschild spacetime
is the most famous example of a black hole. Moreover, each γ(s) is future incomplete. γ(s) will
reach r = 0 and Rαβγδ Rαβγδ = ∞ at r = 0.Therefore, the Schwarzschild spacetime is future causally
3
geodescially incomplete.
Of course, the Schwarzschild metric is a very special solution to Einstein vacuum equations.
Thus, a fundamental question arises: do generic solutions to Einstein vacuum equations possess
any of the singular features of the Schwarzschild metric?
To some extend, this question is answered by Penrose in [20] with
Theorem 1.1.1. (Penrose’s Incompleteness Theorem) Let (M, g) satisfy reasonable topolog-
ical conditions (M is globally hyperbolic with a noncompact Cauchy hypersurface.) and physical
conditions (M satisfies Ric(V, V ) ≥ 0 for all null V .). It follows that if M contains a closed trapped
surface, then it is future causally geodesically incomplete.
By definition, 2-sphere S is called trapped if both its future expansions are negative. Let L, L
be null vector fields. L is the outgoing vectorfield and L is the incoming vectorfield. Define χ, χ
to be the null second fundamental forms of the hypersurfaces generated by L, L. If both trχ < 0
and trχ < 0 hold pointwise, then the 2-sphere S is called a trapped surface. As a comparison, a
standard 2-sphere S with radius r in Minkowski space possesses trχ = 2/r > 0 and trχ = −2/r < 0.
Thus, following Penrose’s Incompleteness Theorem, trapped surface formation implies geodesi-
cally incomplete Hence, one may formally equate the existence of a trapped surface with the exis-
tence of a black hole. This is very useful because trapped surfaces are local and concrete objects.
To prove that trapped surfaces can form dynamically requires a good understanding of the
solution to the EVE. Indeed, this problem was open for forty years, until the recent breakthrough
result of Christodoulou. In the 589-page monograph [7] he identifies an open set of smooth initial
data containing no trapped surfaces on a finite outgoing null hypersurface, with trivial data on an
incoming null hypersurface, whose evolution must form a trapped surface.
In [15], Klainerman and Rodnianski extend the above result and simplify Christodoulou’s proof
significantly in a finite region of spacetime. They enlarge the admissible set of initial conditions and
show that the corresponding propagation estimates are much easier to derive than the ones in [7].
Meanwhile, this relaxation is still enough to guarantee that a trapped surface forms dynamically.
Moreover, Klainerman and Rodnianski introduce a parabolic scaling in [15] and corresponding scale
4
invariant norms. These new norms allow them to capture the hidden smallness of the nonlinear
interactions in the EVE.
In the first part of this thesis, we will extend Klainerman and Rodnianski’s result to prove
the formation of trapped surfaces from past null infinity. We will then retrieve Christodoulou’s
estimates and re-prove the main theorem in [7].
1.1.2 Heuristic Argument
In this subsection, we demonstrate the heuristic argument for trapped surfaces formation. We
consider a region D = D(u, u) of a vacuum spacetime (M, g) generated by optical functions (u, u),
which are increasing toward the future, where u∞ ≤ u ≤ −c ≤ −1 and 0 ≤ u ≤ δ. Here u∞ and
c are fixed constants and δ is to be determined. Our results are independent of u∞ . Later we will
set u∞ to go to −∞ and obtain theorems from past null infinity. We denote by Hu the outgoing
null hypersurfaces generated by the level surfaces of u, and by H u the incoming null hypersurfaces
generated by the level surfaces of u. Hence Su,u = Hu ∩ H u is a 2-sphere.
• D(u, u) is the colored region on the left.
• The Optical functions (u, u) satisfy
g µν ∂µ u∂ν u = 0,
H g µν ∂µ u∂ν u = 0.
Hu δ (u
e4 =
δ) • One point stands for a 2-sphere.
e3
H
e3
0 (u )
= • (ea )a=1,2 is a frame tangent to the 2-sphere
0)
e4 u∞
=
(u Su,u .
u∞
H
• g(ea , eb ) = δab for a, b = 1, 2.
• e3 , e4 are a null pair.
• g(e3 , e4 ) = −2.
5
To create a trapped surface Su,u , we need that both trχ < 0 and trχ < 0 hold pointwise on
Su,u . For the initial data along H 0 , on each S(u, 0) we have
2 2
trχ(u, 0) = − , trχ(u, 0) = .
|u| |u|
For the initial data along Hu∞ , we have
2 2
trχ(u∞ , u) = − + l.o.t. < 0, trχ(u∞ , u) = + l.o.t. > 0.
|u∞ | |u∞ |
In the colored region D(u, u), trχ < 0 is always true due to the Raychaudhuri Equation:
1
∇3 trχ = − (trχ)2 − |χ̂|2 + l.o.t.
2
For χ, we have the following transport equations:
1
∇4 trχ + (trχ)2 = −|χ̂|2 + l.o.t., (1.1.2)
2
and
1
∇3 χ̂ + trχχ̂ = l.o.t. (1.1.3)
2
Here χ̂ denotes the traceless part of χ.
Employing (1.1.2), we derive
∇3 trχ ≤ −|χ̂|2 .
Hence, it follows that
Z u Z u
0 20
trχ(u, u) ≤ trχ(u, 0) − 2
|χ̂| (u, u )du = − |χ̂|2 (u, u0 )du0 .
0 |u| 0
Using the fact that trχ = −2/|u| + l.o.t. in D(u, u) as well as (1.1.3), we obtain
6
|u|2 |χ̂|2 (u, u) = |u∞ |2 |χ̂|2 (u∞ , u) + l.o.t.
Combining these together, along H−c we have
u u
|u∞ |2
Z Z
2
trχ(−c, u) ≤ trχ(−c, 0) − |χ̂|2 (−c, u0 )du0 = − |χ̂|2 (u∞ , u0 )du0 + l.o.t.
0 |c| |c|2 0
(0,δ)
In order to create a trapped surface along the hypersurface Hc , and to avoid trapped surfaces
(0,δ)
in the initial hypersurface Hu∞ , it is sufficient to require
Z u
4c 1
< |χ̂|2 (u∞ , u0 )du0 < .
|u∞ |2 0 |u∞ |
Thus, we expect
1
|u∞ |kχ̂kL∞ (Su∞ ,u ) ≈ δ − 2 ,
which is very large.
To rigorously verify this heuristic argument, we encounter two main difficulties.
1. With arbitrary dispersed initial data at past null infinity, we use a focusing mechanism to
create a trapped surface of radius c. The gravitational radiation needs to go sufficiently far inside.
Since this process takes a long time from past null infinity, we will essentially need a semi-global
existence result for the Einstein vacuum equations without symmetry assumption. We expect this
to be difficult because:
• with no symmetry assumptions, Einstein vacuum equations are energy supercritical;
• it is a large data problem, since small data will lead to the stability of Minkowski spacetime
(see [8]).
To deal with large data, in [7], Christodoulou introduces short pulse ansatz and the correspond-
ing hierarchy, which is called the short pulse method. Based on the smallness assumption on δ,
he establishes the initial data hierarchy for various geometric quantities at Hu∞ . Then he proves
preservation of this hierarchy in the whole colored region D(u, u) starting from Hu∞ . This enables
7
him to obtain the desired semi-global existence result.
We will also use the short pulse method, but with a different initial data hierarchy, and we will
employ a more direct and intuitive approach to the energy estimates.
In the proof, we will obtain results independent of u∞ . Thus, in the end we will set u∞ to go
to −∞ to obtain formation of trapped surfaces from past null infinity.
2. Explicitly refer to heuristic argument, we need to make sure that all of the lower order terms
are truly negligible compared with the main terms. Since Einstein vacuum equations are a coupled
system of many geometric quantities, this requires the understanding of detailed information about
all of the geometric quantities and their interactions.
In Chapter 2 of this thesis, by following and extending the ideas of Klainerman and Rodnianski
in [15], we introduce the notion of signatures. This associates a pair of numbers (s1 , s2 ) to every
quantities of interest, where s1 and s2 encode the information about short pulse and decay rates,
respectively. This allows for a systematic treatment of many terms and it significantly simplifies
the proof. There are however some terms which do not fall into this framework and these must be
tracked carefully.
1.1.3 Main Results
Signature and Scale Invariant Norms
We now turn to the explicit definition of signature and associate norms. We assign to each
geometric quantity φ two numbers s1 (φ) and s2 (φ). The former is introduced by Klainerman and
Rodnianski in [15]. The latter is called the signature for decay rates. Using them I define the scale
invariant norms:
1
kφkL∞
sc (Su,u )
= δ s1 (φ)− 2 |u|2s2 (φ)+1 kφkL∞ (Su,u ) ,
kφkL2sc (Su,u ) = δ s1 (φ)−1 |u|2s2 (φ) kφkL2 (Su,u ) .
Two identities follow from these definitions:
8
s1 (φ1 · φ2 ) = s1 (φ1 ) + s1 (φ2 ),
s2 (φ1 · φ2 ) = s2 (φ1 ) + s2 (φ2 ).
We therefore obtain Hölder’s inequality for scale invariant norms,
1
δ2
kφ1 · φ2 kL2sc (Su,u ) ≤ kφ1 kL∞
sc (Su,u )
kφ2 kL2sc (Su,u ) .
|u|
This inequality tells us that, if all terms are normal, then the nonlinear interactions can be
treated as lower order terms. Hence, only rare anomalous terms are left for further analysis. By
adapting the techniques in [8] and [12], I deal with all of the borderline terms for decay rate without
encountering a logarithmic divergence.
Energy Estimates without Bel-Robinson Tensor
Based on the relation between the new signatures and the coefficient in front of the borderline
terms I extend the methods used by Holzegel in [11] and Luk-Rodnianski in [18]. Without using the
Bel-Robinson tensor, I employ a more direct and intuitive approach to establish energy estimates.
Denote Ψ to be Curvature component and ψ to be Ricci coefficient. We observe that, by

0 1 0 1
separating Ψ into proper pairs, the pair Ψ(s,s ) and Ψ(s− 2 ,s + 2 ) satisfy
0 1 0 1 0 1
X 0 0
∇3 Ψ(s,s ) + ( + s0 )trχΨ(s,s ) = ∇Ψ(s− 2 ,s + 2 ) + ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) ,
2 s1 +s2 =s,
s01 +s02 =s0 +1
and
1 0 1 0 X 0 0
∇4 Ψ(s− 2 ,s + 2 ) = ∇Ψ(s,s ) + ψ (s̃1 ,s̃1 ) · Ψ(s̃2 ,s̃2 ) .
s̃1 +s̃2 =s+ 1 ,
2
s̃01 +s̃02 =s0 + 1
2
0 0
Here Ψ(s,s ) and ψ (s,s ) stand for S-tangent tensors Ψ and ψ with signatures s1 (Ψ) = s, s2 (Ψ) =
0
s0 and s1 (ψ) = s, s2 (ψ) = s0 , respectively. In the above equations, (1/2 + s0 )trχΨ(s,s ) is the
borderline term. Taking the fact trχ = −2/|u| + l.o.t in D(u, u) and using the connection between
0 0
the coefficient 1/2 + s0 and Ψ(s,s ) ’s signature for decay rates s2 (Ψ(s,s ) ) = s0 , we will cancel this
9
borderline term in our newly defined scale invariant norms. This observation avoids employing the
Bel-Robinson tensor and gives us a more direct and intuitive approach for energy estimates. Since
we do not need the symmetry property of the Bel-Robinson tensor, this approach can be used to
remove the Minkowskian initial data assumption along the incoming cone in [7].
Retrieving Christodoulou’s Estimates
The norms we use are consistent with the norms in [15], which are weaker than the norms in
[7]. With these weaker norms, we encounter many fewer borderline terms and it is less difficult to
establish the semi-global existence result. Furthermore, based on the existence results obtained in
weak norms together with initial data in strong norms, we can improve the estimates derived in
weak norms to strong norms and thus retrieve Christodoulou’s estimates in [7].
Let χ be the second fundamental form for Su,u with respect to Hu and let χ̂ be the traceless
part of χ. In Chapter 2 of this thesis, we re-prove the main theorem in Christodoulou’s monograph:
Theorem 1.1.2. (Christodoulou [7], 2008; A. [1], 2012)
Given c and B, there exists δ0 = δ0 (c, B) sufficiently small, such that for 0 < δ < δ0 , with initial
data:
1
δ 2 +k |u∞ |1+i k∇k4 ∇i χ̂∞ kL∞ (Su∞ ,u ) ≤ B along u = u∞
P
• i≤5,k≤3
• Minkowskian initial data along u = 0

Rδ
• 0
u2∞ |χ̂∞ |2 ≥ 4c for every direction along u = u∞
we have that S(c, δ) is a trapped surface.
Moreover, all of our proofs are independent of u∞ . Letting u∞ go to −∞, we obtain
Trapped surfaces can form dynamically for Einstein vacuum equations with initial data which
are dispersed at past null infinity.
10
1.2 Formation of Trapped Surfaces in Vacuum from Mild
Incoming Radiation
(Joint with Luk)
1.2.1 Background
A famous conjecture in general relativity is the Weak Cosmic Censorship Conjecture:
For generic asymptotically flat initial data, the maximal development of the Einstein field equa-
tions possesses a complete future null infinity.
Heuristically this conjecture states that a generic gravitational singularity is hidden from far-
away observers (Cosmic Censorship Principle).
In [3]-[6] Christodoulou proves this conjecture for the Einstein-scalar field system under spherical
symmetry (ESFS-S):
1
Rµν − gµν R = 8πTµν ,
2
g ψ = 0,
1
Tµν = ∂µ ψ∂ν ψ − gµν ∂ γ ψ∂γ ψ.
2
We employ u and u as coordinates. Denote r(u, u) to be the radius of 2-sphere Su,u and define
Ω through s
∂ ∂
Ω := −g( , ).
∂u ∂u
ESFS-S reduces to the following 1 + 1 dimensional partial differential equation system for the
functions r, Ω and ψ (see [3]):

ru φ2u
∂u ( ) = −4πr , (1.2.1)
Ω2 Ω2
11
ru φ2u
∂u ( ) = −4πr , (1.2.2)
Ω2 Ω2
Ω2 ru ru
ruu = − − , (1.2.3)
4r r
Ω2 ru ru
(log Ω)uu = + 2 − 4πφu φu , (1.2.4)
4r2 r
ru ru
φuu = − φu − φu . (1.2.5)
r r
We also define Hawking mass m through
m := r(1 + ∂u r∂u r)/2. (1.2.6)
And Hawking mass is an important quantity for ESFS-S.
Here, we briefly outline Christodoulou’s proof strategy. A spacetime with an incomplete future
null infinity corresponds to a naked singularity. In [5] Christodoulou shows the existence of regular
asymptotically flat initial data whose future developments have an incomplete future null infinity.
Hence, he has constructed an example of naked singularity in [5]. In [6] it is proved that the naked
singularity in [5] is unstable, that is, a small perturbation of the initial data corresponding to the
naked singularity will lead to the formation of trapped surfaces in the future. And moreover he
proves that all naked singularities are unstable (see also [4]), thus verifying weak cosmic censorship
for ESFS-S.
One essential step in Christodoulou’s proof of the instability of naked singularities is to establish
a sharp trapped surface formation criterion. This criterion gives sufficient condition on the “initial
mass density” such that a trapped surface forms in the future.
Christodoulou’s criterion also addresses a natural question: what is the least “initial mass den-
sity” (size of initial data) required to guarantee that a trapped surface forms in evolution?
12
In the below, we will list the known results on formation of trapped surfaces and carry out a
comparison.
Here is the specific criterion in [3]:
Christodoulou’s trapped surface formation criterion for ESFS-S
Theorem 1.2.1. (Christodoulou [3],1991)

For ESFS-S we prescribe characteris-
tic initial data along OA and AB. A

O and B are 2-spheres with radius r1 , r2
B
A and Hawking mass m1 , m2 respectively,
where m := r(1 + ∂u r∂u r)/2.

For data along AB we define δ := (r2 − r1 )/r2 and η := 2(m2 − m1 )/r2 . We call η the initial
mass density. For some fixed C0 , if η ≥ C0 δ log(1/δ), then there exists a trapped surface in the
causal future of the prescribed initial data.
In [7], [15], [1], we have:
Trapped Surface formation criterion for EVE
Theorem 1.2.2. (see [7], [15], [1])
H
Hc δ (u
L =
δ)
L
H
L
0 (u
= L 1)
0) =
(u
H1
13
Given c and B, there exists δ0 = δ0 (c, B) sufficiently small, such that for 0 < δ < δ0 , with initial
data:
1
δ 2 +k k∇k4 ∇i χ̂0 kL∞ (S1,u ) ≤ B
P
• i≤5,k≤3 along u = 1,
• Minkowskian initial data along u = 0,

Rδ
• 0
|χ̂0 |2 ≥ 4c for every direction along u = 1,
we have that S(c, δ) is a trapped surface.
Comparison with H s norms
We employ H s norms to carry out the comparison between ESFS-S and EVE.
ESFS-S
In [3] for ESFS-S by using the following equation for Hawking mass m
∂u m = r2 (∂u ψ)2 ru ,
we have
RB RB
2(m2 − m1 ) 2r2 (∂u ψ)2 ru 2 2 A ru r2 − r1
η= = A
≈ r (∂u ψ) ≈ r2 (∂u ψ)2 ≈ r2 (∂u ψ)2 δ.
r2 r2 r2 r2
1
To satisfy condition η ≥ C0 δ log(1/δ), we expect ∂u ψ is of size log 2 (1/δ)/r along AB. When
trying to acquire intuition for the EVE with no symmetry assumption, via results for ESFS-S, it is
helpful to use a formal correspondence between ψ and the metric g. ∂u ψ should correspond to ∂u g.
1
Thus, one would expect that the analogous criterion for EVE would require ∂u g ∼ log 2 (1/δ)/r.
1 1
Scaling consideration leads one formally to ∂u ∼ δ −1 and ∂u2 ∼ δ − 2 . Therefore, we have
Z Z δ Z
1 1
|∂u g|2 = |∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δ · log 2 (1/δ) · log 2 (1/δ) ≈ δ log(1/δ),
AB
Hu 0 Su,u
14
and
Z
3
Z δ Z
1 1 1 1 1
|∂u g| =
2 2
|∂u2 ∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δ · δ − 2 · log 2 (1/δ) · δ − 2 · log 2 (1/δ) ≈ log(1/δ).
AB
Hu 0 Su,u
3
Thus, we expect the H 1 norm of g along AB to be of size δ log(1/δ) and H 2 norm along AB to be
bounded by log(1/δ).
EVE
In the nonsymmetric case, χ̂ is considered to be ∂u g. To make trapped surfaces form dynamically,

1
in [7], [15] and [1], it is required that χ̂ is of size δ − 2 /r along AB, which formally means ∂u g is of
1 1 1
size δ − 2 /r along AB. Recalling ∂u2 ∼ δ − 2 , hence we have
Z Z δ Z
1 1
|∂u g| = 2
|∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δ · δ − 2 · δ − 2 ≈ 1,
AB
Hu 0 Su,u
and
Z
3
Z δ Z
1 1 1 1 1
|∂u g| =
2 2
|∂u2 ∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δ · δ − 2 · δ − 2 · δ − 2 · δ − 2 ≈ δ −1 .
AB
Hu 0 Su,u
The works in [7], [15] and [1] prescribe Minkowskian initial data along OA. Since ∂u g is the largest
component among all the 1st derivatives of g, the trapped surface formation criteria of [7], [15],
3
[1] require the H 1 norm and the H 2 norm of characteristic initial data are of size 1 and δ −1 ,
respectively.
Motivation
It is natural to ask whether in the context of δ-pulsed initial data, for EVE one can derive a
similar trapped surface formation criterion as in [3]. However, the proof of Theorem 1.2.1 in [3]
depends on a proof by contradiction argument. We do not know the precise location of trapped
surfaces and it requires a crucial monotonic property of Hawking mass for ESFS-S. Without spher-
ical symmetry, this nice monotonic property fails. Hence, the method of proof in [3] cannot be
generalized.
In the work [2], jointly with Luk, we overcome these difficulties and, without any symmetry
15
assumption, but still prescribing Minkowskian data along OA, we derive, in fact improve, a trapped
surface formation criterion similar to the one in [3] corresponding to mild incoming radiation. The
precise statement will be described in the next subsection.
1.2.2 Main Results
By taking the techniques developed in [1] for studying decay rates and instead using them to study
blow-up rates close to the center, in [2] jointly with Luk, we establish the formal analogous of
Theorem 1.2.1 for the EVE with no symmetry assumptions. We extend the short pulse method
introduced by Christodoulou in [7] as well by introducing two new parameters and corresponding
hierarchies and a number of renormalizations.
1
2
a
H bδ
a H a H
Hδ δ (u Hδ δ (u
L = L =
δ) δ)
L
L
H H
L
0 (u 0 (u
= L 1) = L 1)
0) = 0) =
(u (u
H1 H1
We study the colored regions above. We will state one theorem and two corollaries. The left
picture is for Theorem 1.2.3. The right picture is for Corollary 1.2.4. Recall that in [7] one small
1 1
parameter δ is exploited, we will use two more parameters a and b, where 1 ≤ b ≤ a 2 ≤ δ − 2 .
Theorem 1.2.3. (A.- Luk [2], 2014)
Given B, there exists a0 = a0 (B) and b0 = b0 (B) sufficiently large, such that for a0 ≤ a ≤ δ −1
1 1
and b0 ≤ b ≤ a 2 ≤ δ − 2 with initial data:
16
1
δ k a− 2 k∇k4 ∇i χ̂0 kL∞ (S1,u ) ≤ B
P
• i≤5,k≤3 along u = 1

Rδ 1
• 0
|χ̂0 |2 ≥ 4bδa 2 for every direction along u = 1
1
we have that S(bδa 2 , δ) is a trapped surface.
1
For the comparison with [3], [7], [15] and [1], it suffices to set b = ca 2 , which gives
Corollary 1.2.4. (A.- Luk [2], 2014)
Given c and B, there exists a0 = a0 (c, B) sufficiently large, such that for a > a0 , with initial
data:
1
δ k a− 2 k∇k4 ∇i χ̂0 kL∞ (S1,u ) ≤ B
P
• i≤5,k≤3 along u = 1

Rδ
• 0
|χ̂0 |2 ≥ 4cδa for every direction along u = 1
we have that S(cδa, δ) is a trapped surface.
1
Remark 1. In [2], note that ∂u g ∼ χ̂ and it is of size a 2 /r. By dimensional analysis ∂u ∼ δ −1 ,
1 1
∂us−1 ∼ δ −s+1 with 0 < s < 3
2 and ∂u2 ∼ δ − 2 . Hence, we have
Z Z δ Z
1 1
|∂u g| =2
|∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δa 2 a 2 ≈ δa,
(0,δ)
Hu=1 0 S1,u0
Z Z δ Z
1 1
|∂us g|2 = |∂us−1 ∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δδ −s+1 a 2 δ −s+1 a 2 ≈ δ 3−2s a,
(0,δ)
Hu=1 0 S1,u0
and
Z
3
Z δ Z
1 1 1 1 1
|∂u2 g|2 = |∂u2 ∂u g(u, u0 )|2 dθ1 dθ2 du0 ≈ δδ − 2 a 2 δ − 2 a 2 ≈ a.
(0,δ)
Hu=1 0 S1,u0
Here a is a large universal constant, which is independent of δ. Therefore, to make trapped surfaces
3
form dynamically, it suffices to choose initial data for which the H 2 norm of the characteristic
initial data is bounded below and above by a large universal constant a and all H s (s < 3/2) norms
17
are small. In particular, H 1 norm is of size δa, where a is a universal large constant independent
of δ. Note that formally this is an improvement of [3].
3
Remark 2. Recalling that H 2 norm is the critical norm for Einstein vacuum equation in 3 + 1
dimensions. To get formation of trapped surfaces, we only need the critical norm of initial data is of
size a universal large constant a. Therefore, we also call our improved trapped surfaces formation
criterion as a scale invariant trapped surfaces formation criterion.
We can further specialize the statement by coupling the choice of a and δ.
Corollary 1.2.5. (A.- Luk [2], 2014)
For a = a(δ), satisfying a(δ) → +∞ as δ → 0 and fixed c and B, there exists δ0 = δ0 (c, B), such
that for 0 ≤ δ ≤ δ0 , given initial data:
1
δ k (a(δ))− 2 k∇k4 ∇i χ̂0 kL∞ (S1,u ) ≤ B
P
• i≤5,k≤3 along u = 1

Rδ
• 0
|χ̂0 |2 ≥ 4cδa(δ) for every direction along u = 1
we have that S(cδa(δ), δ) is a trapped surface.
Setting a to δ −1 , log(1/δ) and log log(1/δ), we compare Corollary 1.2.4 with all the known
results:
Remark 3. Let a = δ −1 . By employing the corollary above, we reproduce the main results in [7],
where the initial mass density for each angle is of size c.
Remark 4. If we set a = log 1/δ, by using Corollary 1.2.4, we derive that S(cδ log 1/δ, δ) is a
Rδ
trapped surface. In this case our corresponding initial mass density η ∼ 0 |χ̂0 |2 and it is of size
cδ log 1/δ. In [3] for ESFS-S Christodoulou has proved that if the initial mass density η is larger
than C0 δ log 1/δ for some fixed C0 , then a trapped surface will form in future. These two results
are consistent with each other.
18
Remark 5. In order to go further, by employing Corollary 1.2.4 and taking a = log log 1/δ we
obtain that S(cδ log log 1/δ, δ) is a trapped surface. Under this situation, the corresponding initial
mass density η is of size cδ log log 1/δ. And for ESFS-S, to make a trapped surface form dynamically
with initial mass density of size cδ log log 1/δ is already beyond the conclusions in [3].
Remark 6. In Corollary 1.2.4, the parameter a can be chosen as a fixed large constant, which is
independent of δ. Hence, as δ → 0, the initial mass density is of size cδa, which is even smaller
than cδ log log(1/δ). In fact, Corollary 1.2.4 is sharp for δ with respect to the methods.
Remark 7. Among all the known results for Einstein’s equation, our result is of the least “initial
mass density” (size of initial data) to guarantee that a trapped surface forms in evolution.
Being back to Theorem 1.2.3, we have two more remarks.
1
Remark 8. Compared with [7], [15] and [1], we are able to reach the region u ≈ bδa 2 . This is
very close to the center of gravitational collapse (u = 0). And we have established sharp results for
both δ and a with respect to the methods.
Remark 9. The extra generality of Theorem 1.2.3 can be used to obtain a trapped surface for-
mation criterion with different bounds for χ̂0 from above and from below. Indeed, compared with
Corollary 1.2.4, under the same upper bound of initial data along u = 1, we obtain a larger region
for the existence of Einstein vacuum equations. Hence, we can make trapped surfaces form dynam-
ically with different upper and lower bounds of χ̂0 along u = 1. χ̂0 can be chosen in the range
1 1 1 1 1
2b 2 a 4 ≤ |χ̂0 | ≤ a 2 , where 1 ≤ b ≤ a 2 ≤ δ − 2 .
1.2.3 Heuristic Argument with ESFS-S
In this subsection we demonstrate the heuristic argument in [2] by establishing a corresponding
result for ESFS-S.

1 1
Let a and b be two large constants satisfying 1 ≤ b ≤ a 2 ≤ δ − 2 . In this subsection, we will
prove:
19
Theorem 1.2.6. (Formation of Trapped Surfaces for ESFS-S)
1
2
bδ
a H
H − δ (u
L =
δ)
L
H
L
0 (u
= L 1)
0) −
=
1
(u
H
We consider the characteristic initial value problem for ESFS-S.
• Along H 0 we prescribe trivial initial data, which corresponds to
r = −1/2u,
ru (u, 0) = −1/2, ru (u, 0) = 1/2, Ω(u, 0) = 1,
φ(u, 0) = 0, φu (u, 0) = 0, φu (u, 0) = 0.
• Along H−1 where u = −1 and u0 ∈ [0, δ], for Ω and φu , we set
Ω(−1, u0 ) = 1, (1.2.7)
1
|φu (−1, u0 )| ≤ a 2 , (1.2.8)
Z δ
1
φ2u (−1, u0 )du0 ≥ 4δa 2 b. (1.2.9)
0
Then S 1 is a trapped surface.

−bδa 2 ,δ
20
We first establish the existence of solutions to ESFS-S in the colored region.
Bootstrap Argument
1
Following the ideas in [2], in the colored region where bδa 2 ≤ |u| ≤ 1 and 0 ≤ u ≤ δ, for ESFS-S
we make bootstrap assumptions:

1
|Ω − 1| < ,
2
1 1 1
|ru + | ≤ , |ru | ≤ a 2 ,
2 b
1 1
δa 2 1 6a 2
|φu (u, u)| ≤ b4 , |φu (u, u)| ≤ .
|u|2 |u|
We prove the existence result by closing a bootstrap argument, which requires improving all the
bounds above.
We begin with a useful remark about r.
Remark 10. Under the bootstrap assumptions above, in the colored region, we have 41 |u| ≤ r ≤ |u|.
1
This is because for the initial data along H 0 , r = −1/2u. With bootstrap assumption |ru | ≤ a 2 ,
we derive
1 1 1
|r(u, u) + u| ≤ δa 2 ≤ |u|.
2 4
Therefore, we obtain 1/4|u| ≤ r ≤ |u|.
Using this remark, we can substitute |u| for r in the proofs below.
We then start from deriving better estimates for φu and φu with Equation 1.2.5:
ru ru
φuu = − φu − φu .
r r
For φu , we have
ru ru
(φu )u = − φ u − φu .
r r
Using Gronwall’s inequality and bootstrap assumptions for ru , ru and φu , we get
1
18δa 2
|φu (u, u)| ≤ . (1.2.10)
|u|2
21
For φu , we employ
ru ru
(φu )u = − φ u − φu .
r r
This is equivalent to
(rφu )u = ru φu + rφuu = −ru φu .
Using bootstrap assumptions for ru and φu , we obtain
1
δab 4
|(rφu )u (u, u)| ≤ .
|u|2
It follows
1 1
δa 2 1 1 a2
|rφu (u, u) − φu (−1, u)| ≤ a2 b4 ≤ 3 .
|u| b4
With the initial data condition

1
|φu (−1, u)| ≤ a 2 ,
we derive
1
5a 2
|φu (u, u)| ≤ . (1.2.11)
|u|
We move to Ω. Ω obeys Equation 1.2.4:
Ω2 ru ru
(log Ω)uu = 2
+ 2 − 4πφu φu .
4r r
With the fact that Ω = 1 along both H−1 and H 0 and the bootstrap assumptions above, we
have
1 1 1
16δ 16δa 2 4πδa 2 δa 2 1
|(log Ω)u (u, u)| ≤ + + b4 .
|u|2 |u|2 |u|2 |u|
Since (log Ω)u = Ω−1 ∂u Ω, we derive
1 1 1
32δ 32δa 2 4πδa 2 δa 2 1
|∂u Ω(u, u)| ≤ 2
+ 2
+ b4 .
|u| |u| |u|2 |u|
22
Therefore, we get
1 1 1
32δ 32δa 2 4πδa 2 δa 2 1 32 32 4π
|Ω(u, u) − 1| ≤ + + b4 ≤ 1 + + 7.
|u| |u| |u| |u| a2 b b b4
We now focus on ru and ru . ru satisfies Equation 1.2.3:
Ω2 ru ru
(ru )u = − − .
4r r
Using initial data condition

1
ru (u, 0) = − ,
2
together with bootstrap assumptions for Ω and ru and Gronwall’s inequality, we arrive at
1 2δ 2
|ru (u, u) + | ≤ ≤ 1 .
2 |u| a2 b
For ru , by employing Equation 1.2.2:
ru φ2u
∂u ( ) = −4πr ,
Ω2 Ω2
we have
ru δ φ2u
Z
1
| (u, u) − | ≤ 4π(r )(u, u0 )du0 . (1.2.12)
Ω2 2 0 Ω2
Using bootstrap assumptions for Ω and φu , it follows that
1
ru 1 10000πδa 1 10000πa 2
| 2
(u, u) − | ≤ ≤ + .
Ω 2 |u| 2 b
Recalling the bootstrap assumption |Ω(u, u) − 1| < 21 , we obtain
1
40000πa 2
|ru (u, u)| ≤ 2 + .
b
Hence by choosing a, b sufficiently large, we have improved all the estimates in bootstrap as-
23
sumptions. Thus, we establish the existence of ESFS-S in the whole colored region.

1 1
We want to show both ru (−bδa 2 , δ) and ru (−bδa 2 , d) are negative and then S 1 is a
−bδa 2 ,δ
trapped surface.
1
It is true that ru (−bδa 2 , δ) < 0, because we have already proved |ru + 1/2| ≤ 1/b.
1
For ru (−bδa 2 , δ), we first prove an useful lemma
Lemma 1.2.7. Under the same initial data condition as in Theorem 1.2.6, we have
Z δ Z δ
1 1
(r2 φ2u )(−bδa 2 , u0 )du0 ≥ φ2u (−1, u0 )du0 .
0 2 0
To prove this lemma, we make a new bootstrap assumption:
Z δ Z δ
sup (r2 φ2u )(u, u0 )du0 ≤ 2 φ2u (−1, u0 )du0 . (1.2.13)
1 0 0
|u|∈[bδa 2 ,1]
We then improve this bound.
Using Equation 1.2.5:

ru ru
φuu = − φu − φu .
r r
we derive

(rφu )2 = −2rφu ru φu .
u
Hence, it follows
Z δ Z δ Z u Z δ
| (r2 φ2u )(u, u0 )du0 − φ2u (−1, u0 )du0 | ≤ 2|rφu ru φu |(u0 , u0 )du0 du0 .
0 0 −1 0
For ru , with the bound in (1.2.12) and bootstrap assumption (1.2.13), we have
ru δ φ2u δ δ
Z Z Z
1 64π 128π
| (u, u) − | ≤ 4π(r )(u, u0 )du0 ≤ (r2 φ2u )(u, u0 )du0 ≤ φ2u (−1, u0 )du0 .
Ω2 2 0 Ω2 |u| 0 |u| 0
24
Together with the derived estimates for φu in (1.2.10) and φu in (1.2.11):
1 1
5a 2 18δa 2
|φu | ≤ , |φu | ≤ ,
|u| |u|2
we get
Z δ Z δ
| (r2 φ2u )(u, u0 )du0 − φ2u (−1, u0 )du0 |
0 0
Z u Z δ
≤ 2|rφu ru φu |(u0 , u0 )du0 du0
1 0
1 1 1 Z
1000δa 2 1 60000πδa 2 δa 2 u 2
≤ δa 2 + φ (−1, u0 )du0
|u| |u| |u| 0 u
60000π u 2
Z
1000 1
≤ δa 2 + φu (−1, u0 )du0 .
b b2 0
Recall for the initial condition we have
Z u
1
φ2u (−1, u0 )du0 ≥ 4bδa 2 .
0
Hence, for sufficiently large b, we arrive at
Z δ Z δ Z δ Z δ
1 1
− φ2u (−1, u0 )du0 ≤ (r2 φ2u )(u, u0 )du0 − φ2u (−1, u0 )du0 ≤ φ2u (−1, u0 )du0 ,
2 0 0 0 2 0
which is equivalent to
Z δ Z δ
3
(r2 φ2u )(u, u0 )du0 ≤ φ2u (−1, u0 )du0 ,
0 2 0
and
Z δ Z δ
1
(r2 φ2u )(u, u0 )du0 ≥ φ2u (−1, u0 )du0 .
0 2 0
Therefore, we have improved bootstrap assumption (1.2.13) and have also proved Lemma 1.2.7.
We are now ready to prove Theorem 1.2.6.
25
Employing Equation 1.2.2:
ru φ2u
∂u ( ) = −4πr ,
Ω2 Ω2
and together with the fact
1 1 1
ru (−bδa 2 , 0) = , Ω(−bδa 2 , 0) = 1,
2
we derive
ru δ φ2u
Z
1 1 1
2
(−bδa 2 , δ) = − 4π(r 2
)(−bδa 2 , u0 )du0
Ω 2 0 Ω
Z δ
1 π 1
≤ − 1 (r2 φ2u )(−bδa 2 , u0 )du0
2 4bδa 2 0
Z δ
1 π
≤ − φ2 (−1, u0 )du0
2 8bδa 12 0 u
1 π 1
≤ − 1 4bδa
2
2 8bδa 2
<0,
where we use Lemma 1.2.7:
Z δ Z δ
1 1
(r2 φ2u )(−bδa 2 , u0 )du0 ≥ (φ2u )(−1, u0 )du0 ,
0 2 0
and the initial data condition (1.2.9):
Z δ
1
φ2u (−1, u0 )du0 ≥ 4δa 2 b.
0
Therefore S 1 is a trapped surface.

−bδa 2 ,δ
Remark 11. In this subsection,the initial mass density η is of size δa due to the estimate
Z δ
2 0 2 0 0 0
η = 2 m(−1, δ) − m(−1, 0) /r(−1, δ) = 2r (−1, u )(∂u ψ) (−1, u )ru (−1, u )du /r(−1, δ) ≈ δa.
0
Remark 12. For EVE, there will be much more geometric quantities to control. We need to
26
design and derive correct hierarchies with respect to δ, a, b for all of them. Especially, we will have
curvature components. We control curvature components and their derivatives through L2 type
energy estimates.
In Chapter 3, we will demonstrate the detailed proofs of Theorem 1.2.3 for EVE.
27
Chapter 2
from Past Null Infinity
2.1 Setting, Equations and Notations
Our setting is the characteristic initial value problem with data given on the two characteristic
hypersurfaces Hu∞ and H 0 intersecting at the sphere Su∞ ,0 . The spacetime will be a solution to
the Einstein equations constructed in a neighborhood of Hu∞ and H 0 containing Su∞ ,0 . Here u∞
is a fixed constant. Our proofs are independent of u∞ . In the end of the paper, we will set u∞ to
go −∞ and obtain results from past null infinity.
28
2.1.1 Double Null Foliation
H
Hu δ (u
L =
δ)
L
H
L
0 (u )
= L u∞
0) =
(u
∞
Hu
For a spacetime in a neighborhood of Su∞ ,0 , we define a double null foliation as follows: Let u
and u be solutions to the eikonal equation
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0,
satisfying the initial conditions u = u∞ (u∞ << −1) on Hu∞ and u = 0 on H 0 . Both u and u are
increasing towards the future.
Let
L0µ = −2g µν ∂ν u, L0µ = −2g µν ∂ν u.
These are null and geodesic vector fields. Let
2Ω−2 = −g(L0 , L0 ).
29
Define
e3 = ΩL0 , e4 = ΩL0
to be the normalized null pair such that
g(e3 , e4 ) = −2
and
L = Ω2 L0 , L = Ω2 L0
to be the so-called equivariant vector fields.
In the sequel, we will consider spacetime solutions to the vacuum Einstein equations in the
gauge such that
Ω = 1, on Hu∞ and H 0 .
We denote the level sets of u as Hu and the level sets of u and H u . By virtue of the eikonal
equations, Hu and H u are null hypersurface. The sets defined by the intersections of the hyper-
surfaces Hu and H u are topologically 2-spheres, which we denote by Su,u . Notice that the integral
flows of L and L respect the foliation Su,u .
2.1.2 The Coordinate System
On a spacetime in a neighborhood of Su∞ ,0 , we define a coordinate system (u, u, θ1 , θ2 ) as follows:
On the sphere Su∞ ,0 , define a coordinate system (θ1 , θ2 ) such that on each coordinate patch the
metric γ is smooth, bounded and positive definite. Then we define the coordinates on the initial
hypersurfaces by requiring
∂ A ∂ A
θ = 0 on H 0 , and θ = 0 on Hu∞ .
∂u ∂u
30
We now define the coordinate system in the spacetime in a neighborhood of Su∞ ,0 by letting u and
u to be solutions to the eikonal equations:
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0,
and define θ1 , θ2 by
/ L θA = 0,
L
where L
/ L denote the restriction of the Lie derivative to T Su,u (See [7], Chapter 1). Relative to the
coordinate system (u, u, θ1 , θ2 ), the null pair e3 and e4 can be expressed as

−1 ∂ ∂ ∂
e3 = Ω + bA A , e4 = Ω−1 ,
∂u ∂θ ∂u
for some bA such that bA = 0 on H 0 , while the metric g takes the form
g = −2Ω2 (du ⊗ du + du ⊗ du) + γAB (dθA − bA du) ⊗ (dθB − bB du).
Consider a coordinate patch U on Su∞ ,0 and define Uu,0 to be a coordinate patch on Su,0 given
by the one-parameter diffeomorphism generated by L. Define Uu,u to be the image of Uu,0 under
S
the one-parameter diffeomorphism generated by L. Define D = u∞ ≤u≤−1,0≤u≤δ Uu,u and it is the
shadowed region.
2.1.3 Equations
We will recast the Einstein equations as a system for Ricci coefficients and curvature components
associated to a null frame e3 , e4 defined above and an orthonormal frame e1 , e2 tangent to the
2-spheres Su,u . Using the indices A, B to denote 1, 2, we define the Ricci coefficients relative to the
31
null fame:
χAB = g(DA e4 , eB ), χAB = g(DA e3 , eB ),

1 1
ηA = − g(D3 eA , e4 ), η A = − g(D4 eA , e3 )
2 2
(2.1.1)
1 1
ω = − g(D4 e3 , e4 ), ω = − g(D3 e4 , e3 ),
4 4
1
ζA = g(DA e4 , e3 )
2
where DA = De(A) . We also introduce the null curvature components,
αAB = R(eA , e4 , eB , e4 ), αAB = R(eA , e3 , eB , e3 ),

1 1
βA = R(eA , e4 , e3 , e4 ), β A = R(eA , e3 , e3 , e4 ), (2.1.2)
2 2
1 1
ρ = R(e4 , e3 , e4 , e3 ), σ = ∗ R(e4 , e3 , e4 , e3 ).
4 4
Here ∗ R denotes the Hodge dual of R. We denote by ∇ the induced covariant derivative operator
on Su,u and by ∇3 , ∇4 the projections to Su,u of the covariant derivatives D3 , D4 (see precise
definitions in [12]).
Observe that,
1 1
ω = − ∇4 (log Ω), ω = − ∇3 (log Ω),
2 2 (2.1.3)
ηA = ζA + ∇A (log Ω), η A = −ζA + ∇A (log Ω).
Let S-tensor φ(1) · φ(2) denote an arbitrary contraction of the tensor product of φ(1) and φ(2)
with respect to the metric γ. We also define
(1) (2) (1) (2) (1) (2)

(φ(1) ⊗φ
b (2) )AB := φA φB + φB φA − δAB (φ(1) · φ(2) ) for one forms φA , φA ,
(1) (2) (1) (2)

(φ(1) ∧ φ(2) )AB := / AB (γ −1 )CD φAC φBD for symmetric two tensors φAB , φAB ,
where / is the volume form associated to the metric γ. For totally symmetric tensors, the div and
32
curl operators are defined by the formulas
(div φ)A1 ...Ar := ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar := / BC ∇B φCA1 ...Ar .
Define also the trace to be
(trφ)A1 ...Ar−1 := (γ −1 )BC φBCA1 ...Ar−1 .
We separate the trace and traceless part of χ and χ. Let χ̂ and χ̂ be the traceless parts of χ
and χ respectively. Then χ and χ satisfy the following null structure equations:
1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ,
2
∇4 χ̂ + trχχ̂ = −2ω χ̂ − α,
1
∇3 trχ + (trχ)2 = −2ωtrχ − |χ̂|2 ,
2
∇3 χ̂ + trχ χ̂ = −2ωχ̂ − α,
1 (2.1.4)
∇4 trχ + trχtrχ = 2ωtrχ + 2ρ − χ̂ · χ̂ + 2div η + 2|η|2 ,
2
1 1
∇4 χ̂ + trχχ̂ = ∇⊗η
b + 2ωχ̂ − trχχ̂ + η ⊗η,b
2 2
1
2
1 1
∇3 χ̂ + trχχ̂ = ∇⊗η
b + 2ω χ̂ − trχχ̂ + η ⊗η.
b
2 2
The other Ricci coefficients satisfy the following null structure equations:
∇4 η = −χ · (η − η) − β,
∇3 η = −χ · (η − η) + β,
3 1 1 1 (2.1.5)
∇4 ω = 2ωω + |η − η|2 − (η − η) · (η + η) − |η + η|2 + ρ,
4 4 8 2
3 1 1 1
∇3 ω = 2ωω + |η − η|2 + (η − η) · (η + η) − |η + η|2 + ρ.
4 4 8 2
33
The Ricci coefficients also satisfy the following constraint equations
1 1 1
div χ̂ = ∇trχ − (η − η) · (χ̂ − trχ) − β,
2 2 2
1 1 1
div χ̂ = ∇trχ + (η − η) · (χ̂ − trχ) + β,
2 2 2 (2.1.6)
1
curl η = −curl η = σ + χ̂ ∧ χ̂,
2
1 1
K = −ρ + χ̂ · χ̂ − trχtrχ.
2 4
with K the Gauss curvature of the spheres Su,u . The null curvature components satisfy the following
null Bianchi equations:
1
b + 4ωα − 3(χ̂ρ +∗ χ̂σ) + (ζ + 4η)⊗β,
∇3 α + trχα = ∇⊗β b
2
∇4 β + 2trχβ = div α − 2ωβ + ηα,
∇3 β + trχβ = ∇ρ + 2ωβ +∗ ∇σ + 2χ̂ · β + 3(ηρ +∗ ησ),

3 1
∇4 σ + trχσ = −div ∗ β + χ̂ ·∗ α − ζ ·∗ β − 2η ·∗ β,
2 2
3 1
2 2 (2.1.7)
3 1
∇4 ρ + trχρ = div β − χ̂ · α + ζ · β + 2η · β,
2 2
3 1
∇3 ρ + trχρ = −div β − χ̂ · α + ζ · β − 2η · β,
2 2
∇4 β + trχβ = −∇ρ +∗ ∇σ + 2ωβ + 2χ̂ · β − 3(ηρ −∗ ησ),
∇3 β + 2trχ β = −div α − 2ωβ + η · α,

1
b + 4ωα − 3(χ̂ρ −∗ χ̂σ) + (ζ − 4η)⊗β
∇4 α + trχα = −∇⊗β b
2
∗
where denotes the Hodge dual on Su,u .
In the sequel, we will use capital Latin letters A ∈ {1, 2} for indices on the spheres Su,u and
Greek letters µ ∈ {1, 2, 3, 4} for indices in the whole spacetime.
34
2.1.4 Integration
Let U be a coordinate patch on Su∞ ,0 and pU be a partition of unity in DU such that pU is supported
in DU . Given a function φ, the integration on Su,u is given by the formula:
Z XZ ∞ Z ∞ p
φ := φpU det γdθ1 dθ2 .
Su,u U −∞ −∞
Let Du0 ,u0 by the region u∞ ≤ u ≤ u0 , 0 ≤ u ≤ u0 . The integration on Du,u is given by the formula
Z XZ u Z u Z ∞ Z ∞ p
φ := φpU − det gdθ1 dθ2 dudu
Du,u U 0 0 −∞ −∞
XZ u Z u Z ∞ Z ∞ p
=2 φpU Ω2 det γdθ1 dθ2 dudu.
U 0 0 −∞ −∞
Since there are no canonical volume forms on Hu and H u , we define integration by
Z XZ δ Z ∞ Z ∞ p
φ := φ2pU Ω det γdθ1 dθ2 du,
Hu U 0 −∞ −∞
and
Z XZ u Z ∞ Z ∞ p
φ := φ2pU Ω det γdθ1 dθ2 du.
Hu U u∞ −∞ −∞
With these notions of integration, we can define the norms that we will use. Let φ be an
arbitrary tensorfield. For 1 ≤ p < ∞, define
Z
||φ||pLp (Su,u ) := < φ, φ >p/2
γ ,
Su,u
Z
||φ||pLp (Hu ) := < φ, φ >p/2
γ ,
Hu
Z
||φ||pLp (H := < φ, φ >p/2
γ .
u)
Hu
35
Define also the L∞ norm by
||φ||L∞ (Su,u ) := sup < φ, φ >1/2

γ (θ).
θ∈Su,u
We will also use mixed norms defined by
! 12
Z u∗
2
||φ||L2u L∞ p
u L (S)
= ( sup ||φ||Lp (Su,u ) ) du ,
0 u∈[0,u∗ ]
! 12
Z u∗
||φ||L2u L∞ p
u L (S)
= ( sup ||∇i φ||Lp (Su,u ) )2 du .
0 u∈[0,]
Note that L∞ Lp is taken before taking L2 . In the sequel, we will frequently use
|| · ||L∞ 2 p
u Lu L (S)
≤ || · ||L2u L∞ p
u L (S)
.
With the above definition, ||φ||L2u L2 (Su,u ) and ||φ||L2 (H u ) differ by a factor of Ω. Nevertheless,
in view of Proposition 2.3.1, these norms are equivalent up to a factor of 2.
2.1.5 Signatures
To capture the structure of Einstein’s equation, we introduce the following definitions below.
Definition of signatures
To φ ∈ {α, β, ρ, σ, K, β, α, χ, χ, ζ, η, η, ω, ω, γ}, we assign signatures s(φ) according to the follow-
ing rules:
s(φ) := (s1 (φ), s2 (φ)),
where
1
s1 (φ) := 1 · N4 (φ) + · Na (φ) + 0 · N3 (φ) − 1,
2
36
and
1
s2 (φ) := 0 · N4 (φ) + · Na (φ) + 1 · N3 (φ) − 1.
2
N4 (φ) is the number of times e4 appears in the definition of φ. Similarly we define N3 (φ) and
Na (φ) where a = 1, 2.
By the definition above, we have
Signature table
α β ρ σ K β α χ ω ζ η η trχ χ̂ ω γ
s1 2 1.5 1 1 1 0.5 0 1 1 0.5 0.5 0.5 0 0 0 0
s2 0 0.5 1 1 1 1.5 2 0 0 0.5 0.5 0.5 1 1 1 0
Properties of signatures
s1 (∇4 φ) = s1 (φ) + 1, s2 (∇4 φ) = s2 (φ),
1 1
s1 (∇φ) = s1 (φ) + , s2 (∇φ) = s2 (φ) + ,
2 2
s1 (∇3 φ) = s1 (φ), s2 (∇3 φ) = s2 (φ) + 1.
Conservation of signatures
s1 (φ1 · φ2 ) = s1 (φ1 ) + s1 (φ2 ),
s2 (φ1 · φ2 ) = s2 (φ1 ) + s1 (φ2 ),
s(φ1 · φ2 ) = (s1 (φ1 · φ2 ), s2 (φ1 · φ2 )) = (s1 (φ1 ) + s1 (φ2 ), s2 (φ1 ) + s1 (φ2 )) = s(φ1 ) + s(φ2 ).
37
Remark: s1 is the same signature used in [15] and s2 is introduced to study the decay rate
near past null infinity.
2.1.6 Scale Invariant Norms
For any horizontal tensor-field φ with signature s(φ) = (s1 (φ), s2 (φ)), we give
Definition of scale invariant norms on Su,u
1
kφkL∞
sc (Su,u )
:= δ s1 (φ)− 2 |u|2s2 (φ)+1 kφkL∞ (Su,u ) ,
3 1
kφkL4sc (Su,u ) := δ s1 (φ)− 4 |u|2s2 (φ)+ 2 kφkL4 (Su,u ) ,
kφkL2sc (Su,u ) := δ s1 (φ)−1 |u|2s2 (φ) kφkL2 (Su,u ) ,
3
kφkL1sc (Su,u ) := δ s1 (φ)− 2 |u|2s2 (φ)−1 kφkL1 (Su,u ) .
More generally, for 1 ≤ p ≤ ∞, we define
1 1 2
kφkLpsc (Su,u ) := δ s1 (φ)− 2 − p |u|2s2 (φ)+1− p kφkLp (Su,u ) .
In scale invariant norms, we have
Hölder’s inequalities
1
δ2
sc (Su,u )
kφ2 kL1sc (Su,u ) ,
|u|
1
δ2
kφ1 · φ2 kL1sc (Su,u ) ≤ kφ1 kL2sc (Su,u ) kφ2 kL2sc (Su,u ) ,
|u|
38
1
δ2
sc (Su,u )
kφ2 kL2sc (Su,u ) ,
|u|
1
δ2
kφ1 · φ2 kL2sc (Su,u ) ≤ kφ1 kL4sc (Su,u ) kφ2 kL4sc (Su,u ) ,
|u|
1
δ2
sc (Su,u )
kφ2 kL4sc (Su,u ) .
|u|
(0,u)
For convenience, along the null hypersurfaces Hu and H (u
u
∞ ,u)
we also define
Scale invariant norms along a hypersurface
Z u
kφk2L2 (0,u) := δ −1 kφk2L2sc (Su,u0 ) du0 ,
sc (Hu )
0
Z u
1
kφk2L2 (u∞ ,u) := kφk2L2sc (Su0 ,u ) du0 .
sc (H u )
u∞ |u0 |2
2.1.7 Norms
We now define the following norms that we will work with
Ricci coefficient norms:
For any Su,u , we introduce Os,p (u, u).
O0,∞ (u, u) :=kωkL∞

sc (Su,u )
+ kχ̂kL∞
sc (Su,u )
+ ktrχkL∞
sc (Su,u )
+ kηkL∞
sc (Su,u )
1
1 δ2
+ kηkL∞
sc (Su,u )
+ kχ̂kL∞
sc (Su,u )
+ 2 ktrχkL∞
sc (Su,u )
+ kωkL∞
sc (Su,u )
,
|u| |u|
1
O0,4 (u, u) :=kωkL4sc (Su,u ) + δ 4 kχ̂kL4sc (Su,u ) + ktrχkL4sc (Su,u ) + kηkL4sc (Su,u )
1 3
δ4 δ4
+ kηkL4sc (Su,u ) + kχ̂kL4sc (Su,u ) + 2 ktrχkL4sc (Su,u ) + kωkL4sc (Su,u ) ,
|u| |u|
39
O1,4 (u, u) :=k∇ωkL4sc (Su,u ) + k∇χ̂kL4sc (Su,u ) + k∇trχkL4sc (Su,u )
1
+ k∇ηkL4sc (Su,u ) + k∇ηkL4sc (Su,u ) + k∇χ̂kL4sc (Su,u )
|u|
+ k∇trχkL4sc (Su,u ) + k∇ωkL4sc (Su,u ) ,
O2,4 (u, u) :=k∇2 ωkL4sc (Su,u ) + k∇2 χ̂kL4sc (Su,u ) + k∇2 trχkL4sc (Su,u )
1
+ k∇2 ηkL4sc (Su,u ) + k∇2 ηkL4sc (Su,u ) + k∇2 χ̂kL4sc (Su,u )
|u|
+ k∇2 trχkL4sc (Su,u ) + k∇2 ωkL4sc (Su,u ) ,
O3,2 (u, u) :=k∇3 ωkL2sc (Su,u ) + k∇3 χ̂kL2sc (Su,u ) + k∇3 trχkL2sc (Su,u )
1
+ k∇3 ηkL2sc (Su,u ) + k∇3 ηkL2sc (Su,u ) + k∇3 χ̂kL2sc (Su,u )
|u|
+ k∇3 trχkL2sc (Su,u ) + k∇3 ωkL2sc (Su,u ) .
We denote O0,4 , O0,∞ , O1,4 , O2,4 and O3,2 to be the supremum of the corresponding norms
over all values of u, u in our slab . Finally, we define the total Ricci norm O
O := O0,4 + O0,∞ + O1,4 + O2,4 + O3,2
and let O(0) be the corresponding norm of the initial hypersurface Hu∞ .
Curvature norms:
(0,u)
Along the null hypersurfaces H = Hu and H = H (u
u
∞ ,u)
, we introduce
1
R0 (u, u) := δ 2 kαkL2sc (H) + kβkL2sc (H) + kρkL2sc (H) + kσkL2sc (H) + kβkL2sc (H) ,
40
1
R0 (u, u) := δ 2 kβkL2sc (H) + kρkL2sc (H) + kσkL2sc (H) + kβkL2sc (H) + kαkL2sc (H) ,
R1 (u, u) := k∇αkL2sc (H) + k∇βkL2sc (H) + k∇ρkL2sc (H) + k∇σkL2sc (H) + k∇βkL2sc (H) ,
R1 (u, u) := k∇βkL2sc (H) + k∇ρkL2sc (H) + k∇σkL2sc (H) + k∇βkL2sc (H) + k∇αkL2sc (H) ,
R2 (u, u) := k∇2 αkL2sc (H) + k∇2 βkL2sc (H) + k∇2 ρkL2sc (H) + k∇2 σkL2sc (H) + k∇2 βkL2sc (H) ,
R2 (u, u) := k∇2 βkL2sc (H) + k∇2 ρkL2sc (H) + k∇2 σkL2sc (H) + k∇2 βkL2sc (H) + k∇2 αkL2sc (H) ,
We set R0 , R1 , R2 to be the supremum over u, u in our spacetime slab of R0 (u, u), R1 (u, u)
and R2 (u, u), respectively. Similarly, we define R0 , R1 and R2 . We write R := R0 + R1 + R2 and
R := R0 + R1 + R2 . Finally, we denote R(0) as the initial value for the norm R, i.e.,

(0)
R := sup R0 (u∞ , u) + R1 (u∞ , u) + R2 (u∞ , u) .
0≤u≤δ
Initial data assumptions:
We define the initial data quantity
I (0) := sup I (0) (u),

0≤u≤δ
41
where
1 1
X
I (0) (u) :=δ 2 |u∞ |kχ̂∞ kL∞ (Su∞ ,u ) + δ 2 k(δ∇4 )k χ̂∞ kL2 (Su∞ ,u )
0≤k≤2
1 1
X X
+ δ |u∞ |k(δ 2 |u∞ |∇)m−1 (δ∇4 )k ∇χ̂∞ kL2 (Su∞ ,u ) .
2
0≤k≤1 1≤m≤4
(0,u)
Here χ̂∞ denotes χ̂ along Hu∞ .
Norms in standard L∞ form:
For convenience, we list the norms in the standard L∞ form:
1 1 1
O0,∞ (u, u) =δ 2 |u|kχ̂kL∞ (Su,u ) + δ 2 |u|kωkL∞ (Su,u ) + δ 2 |u|ktrχkL∞ (Su,u )
1
+ |u|2 kη, ηkL∞ (Su,u ) + δ − 2 |u|2 kχ̂kL∞ (Su,u )
1
+ |u|ktrχkL∞ (Su,u ) + δ − 2 |u|3 kωkL∞ (Su,u ) .
The norms in the standard Lp form p = 2, 4 are listed in the appendix.
2.2 Statement of Main Theorem
We are now ready to state our main theorem
Theorem 2.2.1. (Main Theorem)
Consider the following characteristic initial value problem for the Einstein vacuum equations.
The initial incoming hypersurface H 0 is required to coincide with a backwards light cone in
Minkowski space with u∞ ≤ u ≤ 0. On the initial outgoing hypersurface Hu∞ , the data are
smooth and I (0) (u) is bounded by an arbitrary constant I (0) uniformly. Given I (0) and another
arbitrary constant c, there exists δ0 = δ0 (I (0) , c) > 0 sufficiently small, such that, for 0 < δ < δ0 ,
in the region u∞ ≤ u ≤ c, 0 ≤ u ≤ δ, we have
42
R + R + O ≤ I (0) .
Hence there exists a unique smooth spacetime solution (M, g) endowed with a double null
foliation (u, u) which solves the characteristic initial value problem to the vacuum Einstein equations
in the region u∞ ≤ u ≤ c, 0 ≤ u ≤ δ.
Remark 13. In the following, we will only prove the a priori estimates for O, Õ5,2 and R. The exis-
tence and uniqueness of solution and the propagation of regularity follow from standard arguments
(see, for example, [7]).
Remark 14. Following [7], one can solve the constraint ODEs and obtain bounds for the initial
data on Hu∞ from that of the initial shear. In particular, under the assumption of Theorem 2.2.1,
we have the following initial bounds for the Ricci coefficients and curvature components
R(0) + O(0) ≤ I (0) .
Once the existence theorem is established, the actual formation of trapped surfaces follows from
a simple ODE argument as in [7]:
Theorem 2.2.2. (Formation of Trapped Surfaces)
Given I (0) and c, there exist δ0 = δ0 (I (0) , c) sufficiently small, such that for 0 < δ < δ0 , with
initial data:
1
δ k δ 2 |u∞ |k∇k4 (|u∞ |∇)i χ̂∞ kL∞ (Su∞ ,u ) ≤ I (0)
P
• i≤5,k≤3 along u = u∞
1 1
δ 2 k(δ 2 |u∞ |∇)j χ̂∞ kL2 (Su∞ ,u ) ≤
P
• 2≤j≤7 along u = u∞
• Minkowski initial data along u = 0

Rδ
• 0
|u∞ |2 |χ̂∞ |2 ≥ 4c for every direction along u = u∞
we have that Sc,δ is a trapped surface.
Moreover, our proofs are independent of u∞ . After let u∞ go to −∞, we obtain
43
Trapped surfaces form in the Cauchy development of Einstein vacuum equations with initial
data which are arbitrarily dispersed at past null infinity.
2.2.1 Structure of the Proof
Let C denote a large fixed constant and R denote a constant to be determined. We briefly outline
the proof of Theorem 2.2.1:
STEP 0: Assuming that O0,∞ < ∞, in Section 2.3 we prove the bounds on the metric components,
from which we derive preliminary estimates such as Sobolev embedding and the estimates for
transport equations.
STEP 1: Assuming R + R < R, in Section 2.4.1, 2.4.2 and 2.4.3 we obtain that

O0,4 + O1,4 + O0,∞ ≤ 10 I (0) + R + R .
STEP 2: Assuming R + R < R, in Section 2.5 we show that

1
O3,2 ≤ 10 k∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + δ 4 (I (0) + R + R)5 ,
and
O2,4 ≤ Ck∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + C(I (0) + R + R)5 .
STEP 3: Using the previous steps, in Section 2.6 we derive the bound
R + R ≤ 2(R(0) + I (0) ).
By choosing R = C(R(0) +I (0) ), we close the bootstrap and finish the proof of Theorem 2.2.1.
44
2.2.2 Strategy of the Proof
In this subsection, we explain some strategies that we use in the proof.
Employing Scale Invariant Norms
Without symmetry assumptions, proving the existence and the uniqueness of solutions to Ein-
stein’s equation relies on L2 -type estimates for the curvature component and its derivatives. And
we are required to control many different curvature components and Ricci coefficients. Following
Klainerman-Rodnianski’s method in [15], we extend their definitions for scale invariant norms to
include decay rate u and we employ a systematical approach to control all the terms. The following
Hölder’s inequality holds in scale invariant norms
1
δ2
sc (Su,u )
kφ2 kL4sc (Su,u ) . (2.2.1)
|u|
This implies that, if all the terms are of size 1 in scale invariant norms (we call them normal
terms), the nonlinear terms behave as lower order terms. And therefore, we can treat nonlinear
terms uniformly. In fact, most of the Ricci coefficients and curvature components are normal terms.
Hence we only need to track a few anomalous terms.
Controlling Anomalous Terms
In this section, α, χ̂ and trχ are the anomalous terms, which can be observed via the definition
of norms and the expectation that all norms are of size 1. Therefore, through the definition of
1
1 1 δ2
R0 (u, u) and O0,∞ , we only expect that δ 2 kαkL2sc (H) , |u| kχ̂kLsc (Su,u )
∞ and |u|2 ktrχkLsc (Su,u )
∞ are
bounded by a fixed constant.
Here we list three typical cases about how to control anomalous terms.
Case 1: We constantly exploit Hölder’s inequalities in scale invariant norms, such as Hölder’s
1
inequality (2.2.1), to gain smallness in δ and |u| . We use the smallness to compensate for the
largeness arising from the anomalous terms. For example, to control k∇ηkL4sc (Su,u ) , we would like
to show that
Z u
1
|u0 |−3 ktrχηηkL4sc (Su0 ,u ) du0 ≤ .
u∞ |u|
45
Here trχ is anomalous. By employing Hölder’s inequality (2.2.1) twice and using the definition of
O0,∞ (u, u) and O0,4 (u, u), we derive the following inequality:
Z u
|u0 |−3 ktrχηηkL4sc (Su0 ,u ) du0
u∞
Zu
δ
≤ ktrχkL∞
sc (Su0 ,u )
kηkL∞
sc (Su0 ,u )
kηkL4sc (Su0 ,u ) du0
u∞ |u0 |5
Zu 1 1
δ2 δ2
≤ ktrχkL∞
sc (Su0 ,u )
kηkL∞
sc (Su0 ,u )
u∞ |u0 |3 |u0 |2
1
δ2 2
≤ 2 O0,∞ O0,4
|u|
1
≤ ,
|u|
1
δ2
where |u|2 is used to eliminate the largeness from ktrχkL∞
sc (Su,u )
.
Case 2: Let ψ denote Ricci coefficients and Ψ denote curvature components. ψ and Ψ obey null
structure equations
∇3 ψ + c[ψ]trχψ = Ψ + ψ 0 ψ 0 , ∇4 ψ = Ψ + ψ 0 ψ 0 ,
and null Bianchi equations
∇3 Ψ + c[Ψ]trχΨ = ∇Ψ0 + ψΨ0 , ∇4 Ψ = ∇Ψ0 + ψΨ0 .
Ru
In these equations c[ψ]trχψ and c[Ψ]trχΨ are borderline terms and may lead to 1
u∞ |u0 |
du0 , which
results in a logarithmic divergence when u∞ goes to −∞. We overcome this difficulty by using the
fact
2
trχ = − + l.o.t.,
|u|
2
and exploiting integrating factors to cancel − |u| terms. Thus, we get rid of the borderline terms.
Case 3: When deriving estimates for ∇trχ and ∇3 trχ, if we employ equations for ∇3 ∇trχ
and ∇3 ∇3 trχ, the borderline terms (η, η)trχtrχ and (∇2 η, ∇2 η)trχtrχ will appear. Those two
terms cannot be controlled by using integrating factors and will lead to a logarithmic divergence.
We circumvent the divergence by deducing equations for ∇3 ∇(Ωtrχ − u2 ) and ∇3 ∇3 (Ωtrχ − u2 ).
46
2
The borderline terms (η, η)trχtrχ and (∇2 η, ∇2 η)trχtrχ are replaced by (η, η)(trχ + |u| )trχ and
2
(∇2 , ∇2 )(trχ + |u| )trχ, respectively. Meanwhile, by obtaining bounds for ∇Ω, ∇2 Ω and ∇3 Ω, we
establish desired estimates for ∇trχ and ∇3 trχ.
Retrieving Christodoulou’s Estimates
Using the strategies above, under the assumptions for the initial data in Theorem 2.2.1, we can
establish the existence of solutions to Einstein vacuum equations in c ≤ |u| ≤ |u∞ | and 0 ≤ u ≤ δ.
An interesting question is, with smaller initial data as Christodoulou prescribes in [7], can we
retrieve Christodoulou’s estimates in [7]?
The answer is yes. The proof is outlined in Section 2.8. We demonstrate the following ideas:
first, with scale invariant norms in this section, we establish Theorem 2.2.1. As a consequence, we
derive bounds for Ricci coefficients in Section 2.4. With these bounds for Ricci coefficients and
the new initial data in [7], we improve L2 bounds for curvature components via the established
controls in energy estimates in Section 2.6. Second, by employing Sobolev Embedding, we improve
L∞ control of curvature components. Third, with null structure equations, we improve the bounds
for Ricci coefficients to retrieve all the estimates in [7].
We are ready to start the proof.
2.3 The Preliminary Estimates
We will make the following bootstrap assumptions on the Ricci coefficients:
O0,∞ ≤ ∆0 , (2.3.1)
O0,4 ≤ ∆1 , (2.3.2)
O1,4 ≤ ∆2 . (2.3.3)
47
Moreover, we set a bootstrap assumption on curvature components and their derivatives:
R + R ≤ R. (2.3.4)
Here, and in the rest of this section, we use the convention that A ≤ B denotes the
inequality A ≤ C0 B for some universal constant C0 that is independent of δ and u∞ .
2.3.1 Estimates for Metric Components
We first show that we can control Ω under the bootstrap assumption (2.3.1):
Proposition 2.3.1. Under the assumptions of Theorem 2.2.1 and bootstrap assumption (2.3.1),
we have
1
δ2
kΩ − 1kL∞ (Su,u ) ≤ ∆0 .
|u|
Proof. Consider the equation
1 1 1 ∂ −1
ω = − ∇4 log Ω = Ω∇4 Ω−1 = Ω . (2.3.5)
2 2 2 ∂u
Fix u. Notice that both ω and Ω are scalars and therefore the L∞ norm is independent of the
metric. We can integrate equation (2.3.5) using the fact that Ω−1 = 1 on H 0 to obtain
Z u 1
−1 δ2
||Ω − 1||L∞ (Su,u ) ≤ ||ω||L∞ (Su,u0 ) du0 ≤ ∆0 ,
0 |u|
where we have used the bootstrap assumption (2.3.1). Finally, notice that
1
δ2
kΩ − 1kL∞ (Su,u ) = kΩkL∞ (Su,u ) kΩ−1 − 1kL∞ (Su,u ) ≤ ∆0 .
|u|
We then show that we can control γ under the bootstrap assumption (2.3.1):
Proposition 2.3.2. Under the assumptions of Theorem 2.2.1 and the bootstrap assumptions
(2.3.1), (2.3.2) and (2.3.3), we have C and c depending only on initial data such that the following
48
pointwise bounds for γ in D hold:
c ≤ det γ ≤ C.
Moreover, in D,
|γAB |, |(γ −1 )AB | ≤ C.
Proof. The first variation formula states that
L
/ L γ = 2Ωχ.
In coordinates, this means

∂
γAB = 2ΩχAB .
∂u
From this we derive that

∂
log(det γ) = Ωtrχ.
∂u
Define γ0 (u, u, θ1 , θ2 ) = γ(u, 0, θ1 , θ2 ). Then
Z u 1
δ2
| det γ − det(γ0 )| ≤ C |trχ|du0 ≤ C ∆0 . (2.3.6)
0 |u|
This implies that the det γ is bounded above and below. Let Λ be the larger eigenvalue of γ.
Clearly,
Λ≤C sup γAB , (2.3.7)

A,B=1,2
and
X
|χAB |2 ≤ CΛ||χ||L∞ (Su,u ) .
A,B=1,2
Then
Z u 1
0 δ2
|γAB − (γ0 )AB | ≤ C |χAB |du ≤ CΛ ∆0 .
0 |u|
Using the upper bound (2.3.7), we thus obtain the upper bound for |γAB |. The upper bound for
|(γ −1 )AB | follows from the upper bound for |γAB | and the lower bound for det γ.
49
A consequence of the previous proposition is an estimate on the surface area of the two sphere
Su,u .
Proposition 2.3.3. In D we have
1
1 δ4
sup |Area(Su,u ) − Area(Su,0 )| ≤ C∆02 1 |u|2 .
u |u| 2
Proof. This follows from (2.3.6).
With the estimate on the volume form, we can now show that the Lp norms defined with respect
to the metric and the Lp norms defined with respect to the coordinate system are equivalent.
Proposition 2.3.4. Given a covariant tensor φA1 ...Ar on Su,u , we have
Z r
X X ZZ p
< φ, φ >p/2
γ ∼ |φA1 ...Ar |p det γdθ1 dθ2 .
Su,u i=1 Ai =1,2
2.3.2 Estimates for Transport Equations
In latter sections of the paper, we will derive the estimates for the Ricci coefficients and the null
curvature components from the null structure equations and the null Bianchi equations respectively.
These will be viewed as transport equations and we will need a way to obtain estimates from the
covariant null transport equations. For the transport equation in the e4 direction, we will need the
smallness of ktrχkL∞ 1 ∞
u Lu L (Su,u )
, which is a consequence of our bootstrap assumption (2.3.1). More
precisely, we have
(2.3.1), we have
Z u
||φ||L2 (Su,u ) ≤ ||φ||L2 (Su,u0 ) + ||∇4 φ||L2 (Su,u00 ) du00 , (2.3.8)
u0
Z u
||φ||L4 (Su,u ) ≤ ||φ||L4 (Su,u0 ) + ||∇4 φ||L4 (Su,u00 ) du00 (2.3.9)
u0
for an Su,u tangent tensor φ of arbitrary rank.
50
Proof. We first note that the following identity holds for any scalar f :
Z Z Z
d df
f= + Ωtrχf = Ω (e4 (f ) + trχf ) .
du Su,u Su,u du Su,u
Hence, taking f = |φ|2γ , we have
Z u Z
1
||φ||2L2 (Su,u ) =||φ||2L2 (Su,u0 ) + 2Ω < φ, ∇4 φ >γ + trχ|φ|2γ du00 .
u0 Su,u00 2
The inequality (2.3.8) can be concluded using Cauchy-Schwarz on the sphere and the L∞ bounds for
Ω and trχ which are provided by Proposition 2.3.1 and the bootstrap assumption (2.3.1) respectively.
With the same method, taking f = |φ|4γ , we obtain inequality (2.3.9).
(2.3.1), we have
Z u
||φ||L2 (Su,u ) ≤ ||φ||L2 (Su0 ,u ) + ||∇3 φ||L2 (Su00 ,u ) du00 , (2.3.10)
u0
Z u
||φ||L4 (Su,u ) ≤ ||φ||L4 (Su0 ,u ) + ||∇3 φ||L4 (Su00 ,u ) du00 (2.3.11)
u0
Z Z Z
d df
f= + Ωtrχf = Ω e3 (f ) + trχf .
Z u Z
1
||φ||2L2 (Su,u ) =||φ||2L2 (Su0 ,u ) + 2Ω < φ, ∇3 φ >γ + trχ|φ|2γ du00 .
u0 Su00 ,u 2
The inequality (2.3.10) can be concluded using Cauchy-Schwarz on the sphere and the L∞ bounds
for Ω and trχ which are provided by Proposition 2.3.1 and the bootstrap assumption (2.3.1) re-
51
spectively. With the same method, taking f = |φ|4γ , we obtain inequality (2.3.11).
We rewrite the above inequalities in scale invariant norms as follows:
Proposition 2.3.7. For an Su,u tangent tensor φ of arbitrary rank, we have
Z u
kφkL2sc (Su,u ) ≤ kφkL2sc (Su,0 ) + δ −1 k∇4 φkL2sc (Su,u0 ) du0 ,
0
Z u
1
kφkL2sc (Su,u ) ≤ kφkL2sc (Su∞ ,u ) + k∇3 φkL2sc (Su0 ,u ) du0 ,
u∞ |u0 |2
Z u
kφkL4sc (Su,u ) ≤ kφkL4sc (Su,0 ) + δ −1 k∇4 φkL4sc (Su,u0 ) du0 ,
0
Z u
1
kφkL4sc (Su,u ) ≤ kφkL4sc (Su∞ ,u ) + k∇3 φkL4sc (Su0 ,u ) du0 .
u∞ |u0 |2
For the ∇3 equation, we can get more precise estimates by incorporating the weights in the
norms. These weights depend on the coefficients in front of the linear term with a trχ factor. The
main observation is that under the bootstrap assumption (2.3.1), trχ can be viewed essentially as
2
− |u| . More clearly, we have
Proposition 2.3.8 (Evolution Lemma). We continue to work under the assumptions of Theorem
2.2.1 and the bootstrap assumptions (2.3.1). Let φ and F be Su,u -tangent tensor fields of rank k
satisfying the following transport equation:
∇3 φA1 ...Ak + λ0 trχφA1 ...Ak = FA1 ...Ak .
Set p ∈ {2, 4}. Denoting λ1 = 2(λ0 − p1 ), for φ we have
Z u
λ1 λ1
|u| kφkLp (Su,u ) . |u∞ | kφkLp (Su∞ ,u ) + |u0 |λ1 kF kLp (Su0 ,u ) du0 .
u∞
where the implicit constant is allowed to depend on λ0 .
52
Proof. To begin, we have the following identity for any scalar function f :
Z Z Z
d df
f= + Ωtrχf = Ω e3 (f ) + trχf .
Using this identity, we obtain
Z
d
( |u|λ1 p |φ|p )
du Su,u
Z
λ1 p−1 p λ1 p p−1 λ1 p p
= Ω − λ1 p|u| (e3 u)|φ| + p|u| < φ , ∇3 φ > +trχ|u| |φ|
Su,u
Z
= Ω p|u|λ1 p < φp−1 , ∇3 φ + λ0 trχφ >
Su,u
Z
λ1 p λ1 p(e3 u)
+ Ω|u| − + (1 − λ0 p)trχ |φ|p .
Su,u |u|
Observe that we have
λ1 p(e3 u)
− + (1 − λ0 p)trχ
|u|
λ1 pΩ−1
=− + (1 − λ0 p)trχ
|u|
λ1 p(Ω−1 − 1) 2 λ1 p + 2 − 2λ0 p
=− + (1 − λ0 p)(trχ + )−
|u| |u| |u|
1
δ2
≤ ∆0 .
|u|2
For the last inequality, we employ the bound for Ω in Proposition 2.3.1 and the bootstrap assumption
(2.3.1) together with the chosen parameters to satisfy λ1 p + 2 − 2λ0 p = 0.
Therefore,
Z Z
d λ1 p p λ1 p p−1 2λ1 −2 12 p
| ( |u| |φ| )| ≤ p|u| |φ| |F | + |u| δ ∆0 |φ| .
du Su,u Su,u
Using Cauchy-Schwarz for the first term and applying Gronwall’s inequality for the second term,
53
we obtain
|u|λ1 kφkLp (Su,u )

1
Z u
δ 2 ∆0 ku−2 kL1 λ1 0 λ1 0
≤e u |u∞ | kφkLp (Su∞ ,u ) + |u | kF kLp (Su0 ,u ) du
u∞
Z u
≤|u∞ |λ1 kφkLp (Su∞ ,u ) + |u0 |λ1 kF kLp (Su0 ,u ) du0 .
u∞
1 1
since δ 2 ∆0 ku−2 kL1u ≤ 1, when δ 2 ∆0 is small.
2.3.3 Sobolev Embedding
Using the upper and lower bounds of the volume form, Sobolev embedding theorems in our setting
follow from standard Sobolev embedding theorems (see [18]).
Proposition 2.3.9. (L4 Sobolev Embedding) Under the assumptions of Theorem 2.2.1 and the
bootstrap assumptions (2.3.1), (2.3.2) and (2.3.3), we have

1 1 1
||φ||L4 (Su,u ) ≤ C kφkL2 2 (Su,u ) k∇φkL2 2 (Su,u ) + 1 kφkL 2 (Su,u ) ,
|u| 2
and in scale invariant norms

1 1 1
kφkL4sc (Su,u ) ≤ C kφkL2 2 (Su,u ) k∇φkL2
2
(Su,u ) + δ 4 kφk 2
Lsc (Su,u ) .
sc sc
Similarly, for L∞ norm we obtain
Proposition 2.3.10. (L∞ Sobolev Embedding) Under the assumptions of Theorem 2.2.1 and the
bootstrap assumptions (2.3.1), (2.3.2) and (2.3.3), we have

1 1 1
kφkL∞ (Su,u ) ≤ C kφkL4 (Su,u ) k∇φkL4 (Su,u ) +
2 2
1 kφkL4 (Su,u ) ,
|u| 2
and in scale invariant norms
54

1 1 1
kφkL∞
sc (Su,u )
≤ C kφkL4 2
k∇φkL4
2
(Su,u ) + δ kφkLsc (Su,u ) .
4 4
sc (Su,u ) sc
In the same manner, we derive
1
(2.3.1), (2.3.2) and (2.3.3), let δ S u,u ⊂ Su,u denote a disk of radius δ 2 |u| relative to either θ or θ
coordinate system. There exist δ0 = δ0 (∆0 ), such that whenever δ ≤ δ0 , in D for any horizontal
tensor φ we have
1
δ− 4

1 1
kφkL∞ (Su,u ) . sup δ 4 |u| 2 k∇φkL4 (δ S u,u ) + 1 kφkL ( S u,u ) ,
4 δ
δS
u,u ⊂Su,u |u| 2
and in scale invariant norm

kφkL∞
sc
(Su,u ) . sup k∇φkL4sc (δ S u,u ) + kφkL4sc (δ S u,u ) .
δS
u,u ⊂Su,u
2.3.4 Commutation Formulas
We list the following formulas from [12]:
Proposition 2.3.12. For a scalar function f , we have
1
[∇4 , ∇]f = (η + η)D4 f − χ · ∇f,
2
1
[∇3 , ∇]f = (η + η)D3 f − χ · ∇f.
2
Proposition 2.3.13. For a 1-form Ub tangent to Su,u , we have
1
[D4 , ∇a ]Ub = −χac ∇c Ub + ac ∗ βb Uc + (ηa + η a )D4 Ub − χac η b Uc + χab η · U,
2
1
[D3 , ∇a ]Ub = −χac ∇c Ub + ac ∗ β b Uc + (ηa + η a )D3 Ub − χac ηb Uc + χab η · U.
2
55
Proposition 2.3.14. For a 2-form Vbc tangent to Su,u , we have
1
[D4 , ∇a ]Vbc = (ηa + η a )D4 Vbc − η b Vdc χad − η c Vbd χad − bd ∗ βa Vdc − cd ∗ βc Vbd
2
+ χac Vbd η d + χab Vdc η d − χad ∇d Vbc ,
1
[D3 , ∇a ]Vbc = (ηa + η a )D3 Vbc − ηb Vdc χad − ηc Vbd χad + bd ∗ β a Vdc + cd ∗ β c Vbd
2
+ χac Vbd ηd + χab Vdc ηd − χad ∇d Vbc .
By induction, we get the following schematic formula (see [18]):
Proposition 2.3.15. Suppose ∇4 φ = F0 . Let ∇4 ∇i φ = Fi . Then
X X
Fi ∼ ∇i1 (η + η)i2 ∇i3 F0 + ∇i1 (η + η)i2 ∇i3 χ∇i4 φ
i1 +i2 +i3 =i i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 β∇i4 φ.
i1 +i2 +i3 +i4 =i−1
where by ∇i1 (η + η)i2 we mean the sum of all terms which is a product of i2 factors, each factor
X
being ∇j (η + η) for some j and that the sum of all j’s is i1 , i.e., ∇i1 (η + η)i2 = ∇j1 (η +
j1 +...+ji2 =i1
η)...∇ji2 (η + η). Similarly, suppose ∇3 φ = G0 . Let ∇3 ∇i φ = Gi . Then
i X
Gi − trχ∇i φ ∼ ∇i1 (η + η)i2 ∇i3 G0
2 i 1 +i2 +i3 =i
X 2 i4
+ ∇i1 (η + η)i2 ∇i3 (χ̂, trχ + )∇ φ
i1 +i2 +i3 +i4 =i
u
X
+ ∇i1 (η + η)i2 ∇i3 β∇i4 φ.
i1 +i2 +i3 +i4 =i−1
56
2.3.5 General Elliptic Estimates for Hodge Systems
In this subsection, we prove elliptic estimates for general Hodge systems. To this end, we recall the
definition of the divergence and curl of a symmetric covariant tensor of an arbitrary rank:
(div φ)A1 ...Ar = ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar = / BC ∇B φCA1 ...Ar ,
where / is the volume form associated to the metric γ. Recall also that the trace is defined to be
(tr φ)A1 ...Ar−1 = (γ −1 )BC φBCA1 ...Ar−1 .
The following is the main L2 (Su,u ) elliptic estimate that we will use:
Proposition 2.3.16. We continue to work under the assumptions of Theorem 2.2.1 and the boot-
strap assumptions (2.3.1), (2.3.2) and (2.3.3). Let φ be a totally symmetric r+1 covariant tensorfield
on Su,u satisfying
div φ = f, curl φ = g, trφ = h.
Then we have
k∇3 φkL2sc (Su,u )
≤k∇2 f kL2sc (Su,u ) + k∇2 gkL2sc (Su,u ) + δk∇φ, ∇h, f kL2sc (Su,u )
1 3
+ δ 2 k∇f, ∇g, ∇2 hkL2sc (Su,u ) + δ 2 kφ, hkL2sc (Su,u )
1
δ2 1
+ k∇(K − 2 )kL2sc (Su,u ) kφ, hkL∞
sc (Su,u )
|u| |u|
1
δ2 1
+ kK − 2 kL4sc (Su,u ) k∇φ, ∇h, f kL4sc (Su,u )
|u| |u|
1
δ8 1 12
+ kK − k 4 k∇f, ∇g, ∇2 hkL4sc (Su,u )
|u|
1
2 |u|2 Lsc (Su,u )
5
δ8 1 32
+ kK − k 4 kφ, hkL∞
sc (Su,u )
.
|u|
3
2 |u|2 Lsc (Su,u )
57
Proof. Recall the following identity from Chapter 2 in [8] and Chapter 7 in [7] that for φ, f , g
and h as above, we have
Z Z
2 2 2 2 2
|∇φ| + (r + 1)K|φ| = |f | + |g| + K|h| . (2.3.12)
Su,u Su,u
To get higher order elliptic estimates, we recall again from [8] and [7] that the symmetrized angular
derivative of φ defined by
r+1
1 X
(∇φ)sBA1 ...Ar+1 := (∇B φA1 ...Ar + ∇Ai φA1 ...<Ai >B...Ar+1 )
r+2 i=1
satisfies the div-curl system
1 ∗ 2K
div (∇φ)s =(∇f )s − ( ∇g)s + (r + 1)Kφ − (γ ⊗s h)
r+2 r+1
r+1
curl (∇φ)s = (∇g)s + (r + 1)K(∗ φ)s
r+2
2 r
tr (∇φ)s = f+ (∇h)s ,
r+2 r+2
where
X
(γ ⊗s h)A1 ...Ar+1 := γAi Aj hA1 ...<Ai >...<Aj >...Ar+1
i<j=1,...,r+1
and
r+1
1 X
(∗ φ)sA1 ...Ar+1 := / A B φA1 ...<Ai >B...Ar .
r + 1 i=1 i
Using (2.3.12), we obtain
k∇2 φk2L2 (Su,u )
≤k∇f k2L2 (Su,u ) + k∇gk2L2 (Su,u ) + kK(|∇φ|2 + |f |2 + |∇h|2 )kL1 (Su,u )
+ kKφk2L2 (Su,u ) + kKhk2L2 (Su,u ) .
Iterating the procedure and applying (2.3.12), we thus derive
58
≤k∇2 f k2L2 (Su,u ) + k∇2 gk2L2 (Su,u ) + k∇K(φ, h)k2L2 (Su,u ) + kK(∇φ, ∇h)k2L2 (Su,u )
+ kKf k2L2 (Su,u ) + kK(∇f )2 kL1 (Su,u ) + kK(∇g)2 kL1 (Su,u ) + kK(∇2 φ)2 kL1 (Su,u )
+ kK(∇2 h)2 kL1 (Su,u ) + kK 3 (φ2 , h2 )kL1 (Su,u ) .
1 1
By decomposing K = K − |u|2 + |u|2 , we deduce

1
≤k∇2 f k2L2 (Su,u ) + k∇2 gk2L2 (Su,u ) + k∇(K − )(φ, h)k2L2 (Su,u )
|u|2
1 1
+ k(K − )(∇φ, ∇h)k2L2 (Su,u ) + k 2 (∇φ, ∇h)k2L2 (Su,u )
|u|2 |u|
1 1 1
+ k(K − 2 )f k2L2 (Su,u ) + k(K − 2 )(∇f )2 kL1 (Su,u ) + k(K − 2 )(∇2 φ, ∇g)2 kL1 (Su,u )
|u| |u| |u|
1 1 1
+ k 2 f k2L2 (Su,u ) + k 2 (∇f )2 kL1 (Su,u ) + k 2 (∇2 φ, ∇g)2 kL1 (Su,u )
|u| |u| |u|
1 1
+ k(K − 2 )(∇2 h)2 kL1 (Su,u ) + k(K − 2 )3 (φ2 , h2 )kL1 (Su,u )
|u| |u|
1 1
+ k 2 (∇2 h)2 kL1 (Su,u ) + k 6 (φ2 , h2 )kL1 (Su,u ) .
|u| |u|
With Hölder’s inequality, we obtain
59
1
≤k∇2 f k2L2 (Su,u ) + k∇2 gk2L2 (Su,u ) + + k∇φ, ∇hk2L2 (Su,u )
|u|4
1 1 1
+ 2 k∇2 φ, ∇f, ∇g, ∇2 hk2L2 (Su,u ) + 6 kφ, hk2L2 (Su,u ) + 4 kf k2L2 (Su,u )
|u| |u| |u|
1 1
+ k∇(K − 2 )k2L2 (Su,u ) k(φ, h)k2L∞ (Su,u ) + kK − 2 k2L4 (Su,u ) k∇φ, ∇h, f k2L4 (Su,u )
|u| |u|
1 1
+ |u| 2 kK − 2 kL4 (Su,u ) k∇2 φ, ∇f, ∇g, ∇2 hk2L4 (Su,u )
|u|
1 1
+ |u| 2 kK − 2 k3L4 (Su,u ) kφ, hk2L∞ (Su,u ) .
|u|
In scale invariant norms, we deduce
k∇3 φk2L2sc (Su,u )
≤k∇2 f k2L2sc (Su,u ) + k∇2 gk2L2sc (Su,u ) + δ 2 k∇φ, ∇h, f k2L2sc (Su,u )
+ δk∇2 φ, ∇f, ∇g, ∇2 hk2L2sc (Su,u ) + δ 3 kφ, hk2L2sc (Su,u )

δ 1
+ 2
k∇(K − 2 )k2L2sc (Su,u ) kφ, hk2L∞
sc (Su,u )
|u| |u|
δ 1
+ 2 kK − 2 k2L4sc (Su,u ) k∇φ, ∇h, f k2L4sc (Su,u )
|u| |u|
1
δ4 1
+ kK − 2 kL4sc (Su,u ) k∇2 φ, ∇f, ∇g, ∇2 hk2L4sc (Su,u )
|u| |u|
5
δ4 1
+ 3 kK − 2 k3L4sc (Su,u ) kφ, hk2L∞
sc (Su,u )
,
|u| |u|
which is equivalent to
60
≤k∇2 f kL2sc (Su,u ) + k∇2 gkL2sc (Su,u ) + δk∇φ, ∇h, f kL2sc (Su,u )
1 3
+ δ 2 k∇2 φ, ∇f, ∇g, ∇2 hkL2sc (Su,u ) + δ 2 kφ, hkL2sc (Su,u )
1
δ2 1
+ k∇(K − 2 )kL2sc (Su,u ) kφ, hkL∞
sc (Su,u )
|u| |u|
1
δ2 1
+ kK − 2 kL4sc (Su,u ) k∇φ, ∇h, f kL4sc (Su,u )
|u| |u|
1
δ8 1 12
+ kK − k 4 k∇2 φ, ∇f, ∇g, ∇2 hkL4sc (Su,u )
|u|
1
2 |u|2 Lsc (Su,u )
5
δ8 1 32
+ kK − k 4 kφ, hkL∞
sc (Su,u )
.
|u|
3
2 |u|2 Lsc (Su,u )
For the special case that φ a symmetric traceless 2-tensor, we only need to know its divergence:
Proposition 2.3.17. Suppose φ is a symmetric traceless 2-tensor satisfying
div φ = f.
Then, under the assumptions of Theorem 2.2.1 and the bootstrap assumptions (2.3.1), (2.3.2) and
(2.3.3), we have

1
≤k∇2 f kL2sc (Su,u ) + δk∇φ, f kL2sc (Su,u ) + δ 2 k∇2 φ, ∇f kL2sc (Su,u )
1
3 δ2 1
+ δ 2 kφkL2sc (Su,u ) + k∇(K − 2 )kL2sc (Su,u ) kφkL∞
sc (Su,u )
|u| |u|
1 1
δ2 1 δ8 1 12
+ kK − 2 kL4sc (Su,u ) k∇φ, f kL4sc (Su,u ) + 1 kK − k
2 L4sc (Su,u )
k∇2 φ, ∇f kL4sc (Su,u )
|u| |u| |u| 2 |u|
5
δ8 1 32
+ kK − k 4 kφkL∞
sc (Su,u )
.
|u|
3
2 |u|2 Lsc (Su,u )
61
Proof. In view of Proposition 2.3.16, this conclusion follows from
curl φ =∗ f.
This is a direct computation using that fact that φ is both symmetric and traceless.
2.4 Ricci Coefficient Estimates
In this section, we prove estimates for the Ricci coefficients and their first angular derivatives in
L4 (Su,u ).
2.4.1 Estimates for Zero Derivatives of Ricci Coefficients
We start with the L4 (Su,u ) estimate for the Ricci coefficients. We first notice a fact that for all null
structure equations, the Ricci coefficients ψ with signature (s, s0 ) satisfies either transport equation
0 X 0 0 0
∇4 ψ (s,s ) = ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + Ψ(s+1,s ) , (2.4.1)
s1 +s2 =s+1,
s01 +s02 =s0
or
0 0 0 X 0 0 0
∇3 ψ (s,s ) + λ[ψ (s,s ) ]trχψ (s,s ) = ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + Ψ(s,s +1) . (2.4.2)
s1 +s2 =s,
s01 +s02 =s0 +1
0 0
In this section, ψ (s,s ) and Ψ(s,s ) stand for arbitrary Ricci coefficient with signature (s, s0 ) and
0
null curvature component with signature (s, s0 ), respectively. And λ[ψ (s,s ) ] is a constant depending
0
on ψ (s,s ) .
For Ricci coefficients ψ satisfying (2.4.1), we have
62
Proposition 2.4.1.
0
kψ (s,s ) kL4sc (Su,u )
1
(s,s0 )
X δ 2 (s1 ,s01 ) 0
≤kψ kL4sc (Su,0 ) + sup kψ kL∞
sc (Su,u0 )
kψ (s2 ,s2 ) kL4sc (Su,u0 )
u0 ∈[0,δ] s1 +s2 =s+1,
|u|
s01 +s02 =s0
0 1 0 1 1 0
+ kΨ(s+1,s ) k 2 2 (0,u) k∇Ψ(s+1,s ) k 2 2 (0,u) + δ 4 kΨ(s+1,s ) kL2 (0,u) .
Lsc (Hu ) Lsc (Hu ) sc (Hu )
Proof. Using Proposition 2.3.7 for transport equations, we have
0
Z u
(s,s0 ) 0
≤kψ kL4sc (Su,0 ) + δ −1 k∇4 ψ (s,s ) kL4sc (Su,u0 ) du0
Z0 u
0 X 0 0
≤kψ (s,s ) kL4sc (Su,0 ) + δ −1 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su,u0 ) du0
0 s1 +s2 =s+1,
s01 +s02 =s0
Z u
0
+ δ −1 kΨ(s+1,s ) kL4sc (Su,u0 ) du0
0
Z u 1
(s,s0 ) −1
X δ 2 (s1 ,s01 ) 0
≤kψ kL4sc (Su,0 ) + δ kψ kL∞
sc (Su,u0 )
kψ (s2 ,s2 ) kL4sc (Su,u0 ) du0
0 s1 +s2 =s+1,
|u|
s01 +s02 =s0
0 1 0 1 1 0
+ kΨ(s+1,s ) k 2 2 (0,u) k∇Ψ(s+1,s ) k 2 2 (0,u) + δ 4 kΨ(s+1,s ) kL2 (0,u)
1
0 X δ 2 (s1 ,s01 ) 0
≤kψ (s,s ) kL4sc (Su,0 ) + sup kψ kL∞
sc (Su,u0 )
kψ (s2 ,s2 ) kL4sc (Su,u0 )
u0 ∈[0,δ] s1 +s2 =s+1,
|u|
s01 +s02 =s0
0 1 0 1 1 0
+ kΨ(s+1,s ) k 2 2 (0,u) k∇Ψ(s+1,s ) k 2 2 (0,u) + δ 4 kΨ(s+1,s ) kL2 (0,u) ,
where we employ L4 Sobolev Embedding (Proposition 2.3.9) and Hölder’s inequality.
As a consequence, we can prove
(2.3.1), we have
1
1 1 1 δ2
kηkL4sc (Su,u ) ≤ R02 R12 + δ 4 R0 + ∆0 O0,4 ,
|u|
63
1
1 1 δ2 1
kωkL4sc (Su,u ) ≤ R0 R1 + δ R0 +
2 2
∆0 O0,4 ,
4
|u|
1 1
δ4 δ2
ktrχkL4sc (Su,u ) ≤ ktrχkL4sc (Su,0 ) + ∆0 [χ̂]O0,4 [χ̂] + ∆0 O0,4 ,
|u| |u|
3
1 1 1 δ4
δ kχ̂kL4sc (Su,u )
4 ≤ R0 [α]R1 [α] + R0 [α] +
2 2
∆0 O0,4 .
|u|
Proof. The null Ricci coefficients η and ω satisfy
∇4 η = −χ · (η − η) − β,
and
1
∇4 ω = 2ωω + ρ + (η, η)(η, η).
2
Since there are no anomalous terms, the conclusion is guaranteed by Proposition 2.4.1.
The null Ricci coefficient trχ obeys
1
∇4 trχ = − (trχ)2 − |χ̂|2 − 2ωtrχ.
2
With Proposition 2.4.1, we deduce
ktrχkL4sc (Su,u )
1
δ4 1
≤ktrχkL4sc (Su,0 ) + sup kχ̂kL∞
sc (Su,u0 )
δ 4 kχ̂kL4sc (Su,u0 )
u0 ∈[0,δ] |u|
1
δ2
+ sup ktrχ, ωkL∞
sc (Su,u0 )
ktrχkL4sc (Su,u0 )
u0 ∈[0,δ] |u|
1 1
δ4 δ2
≤ktrχkL4sc (Su,0 ) + ∆0 [χ̂]O0,4 [χ̂] + ∆0 O0,4 .
|u| |u|
For χ̂, we have
∇4 χ̂ + trχχ̂ = −2ω χ̂ − α.
64
Proposition 2.4.1 implies
1
δ 4 kχ̂kL4sc (Su,u )
3
δ4
≤ sup kχ̂kL∞
sc (Su,u0 )
ktrχ, ωkL4sc (Su,u0 )
0
u ∈[0,δ] |u|
1 1 1 1
+ δ 4 kαk 2 2 (0,u) k∇αk 2 2 (0,u) + δ 2 kαkL2 (0,u)
3
1 1 δ4
≤R02 [α]R12 [α] + R0 [α] + ∆0 O0,4 .
|u|
0
All the other Ricci coefficients ψ (s,s ) satisfy
0 0 0 X 0 0 0
∇3 ψ (s,s ) + λ[ψ (s,s ) ]trχψ (s,s ) = ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + Ψ(s,s +1) ,
s1 +s2 =s,
s01 +s02 =s0 +1
0
and for these ψ (s,s ) we have
Proposition 2.4.3.
0 0
|u|2λ−2s −1 kψ (s,s ) kL4sc (Su,u )
Z u
2λ−2s0 −1 (s,s0 ) 0 0
≤|u∞ | kψ k L4sc (Su∞ ,u ) + |u0 |2λ−2s −3 kΨ(s,s +1) kL4sc (Su0 ,u ) du0
u∞
Z u
0 X 0 0
+ |u0 |2λ−2s −3 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 .
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Proof. With the help of evolution lemma (Proposition 2.3.8), for
1
p = 4, λ1 = 2λ − ,
2
65
we derive
1 0
|u|2λ− 2 kψ (s,s ) kL4 (Su,u )
Z u
2λ− 12 (s,s0 ) 1 0
≤|u∞ | kψ kL4 (Su∞ ,u)) + |u0 |2λ− 2 k∇3 ψ (s,s ) kL4 (Su0 ,u ) du0
u∞
Z u
2λ− 21 (s,s0 ) 1 0
≤|u∞ | kψ kL4 (Su∞ ,u) + |u0 |2λ− 2 kΨ(s,s +1) kL4 (Su0 ,u ) du0
u∞
Z u
1
X 0 0
+ |u0 |2λ− 2 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4 (Su0 ,u ) du0 .
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
In views of the definition
0 3 0 1 0
kψ (s,s ) kL4sc (Su,u ) = δ s− 4 |u|2s + 2 kψ (s,s ) kL4 (Su,u ) ,
we rewrite the estimate above in scale invariant norms:
0 0
|u|2λ−2s −1 kψ (s,s ) kL4sc (Su,u )
Z u
0 0 0 0
≤|u∞ |2λ−2s −1 kψ (s,s ) kL4sc (Su∞ ,u ) + |u0 |2λ−2s −3 kΨ(s,s +1) kL4sc (Su0 ,u ) du0
u∞
Z u
0 2λ−2s0 −3 (s1 ,s01 ) 0
X
+ |u | kψ · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 .
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
0
Furthermore, it is observed that for ψ (s,s ) ∈ {ω, χ̂, η, trχ} in the null structure equations
0 0
∇3 ψ (s,s ) , we have λ[ψ (s,s ) ] ≤ s0 . With this observation, we derive
0
Proposition 2.4.4. Under the assumption λ ≤ s0 , for ψ (s,s ) satisfying equation (2.4.2), we have
0
0 1 0 1 0 1
≤kψ (s,s ) kL4sc (Su∞ ,u ) + 1 kΨ(s,s +1) k 2 2 (u ) k∇Ψ(s,s +1) k 2 2 (u∞,u )
|u| 2 Lsc (H u ∞,u ) Lsc (H u )
1 Z u
δ 4 0 X 0 0
+ 1 kΨ(s,s +1) kL2 (u∞,u ) + |u0 |−2 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 ,
|u| 2 sc (H u )
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
66
or
1 0
|u|
1 0 1 (s,s0 +1) 2
1 0 1
≤ kψ (s,s ) kL4sc (Su∞ ,u ) + 3 kΨ k 2 (u∞,u ) k∇Ψ(s,s +1) k 2 2 (u∞,u )
|u∞ | |u| 2 Lsc (H u ) Lsc (H u )
1 Z u
δ4 (s,s0 +1)
X 0 0
+ 3 kΨ kL2 (H (u∞,u ) ) + |u0 |−3 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 .
|u| 2 sc u
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
0 0
Proof. With the aid of assumption λ[ψ (s,s ) ] ≤ s0 , we multiply |u|−2λ+2s +1 on both sides of
Proposition 2.4.3. Employing the fact 1 ≤ |u| ≤ |u0 | ≤ |u∞ |, we conclude
0
Z u
(s,s0 ) 0
≤kψ kL4sc (Su∞ ,u ) + |u0 |−2 kΨ(s,s +1) kL4sc (Su0 ,u ) du0
u∞
Z u X 0 0
+ |u0 |−2 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Z u
0 0 1 0 1
≤kψ (s,s ) kL4sc (Su∞ ,u ) + |u0 |−2 kΨ(s,s +1) kL2 2 (S 0 ) k∇Ψ(s,s +1) kL2 2 (S 0 ) du0
sc u ,u sc u ,u
u∞
Z u
1 0
+ |u0 |−2 δ 4 kΨ(s,s +1) kL2sc (Su0 ,u ) du0
u
Z u∞ X 0 0
+ |u0 |−2 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
0 1 0 1 0 1
≤kψ (s,s ) kL4sc (Su∞ ,u ) + 1 kΨ(s,s +1) k 2 2 (u ) k∇Ψ(s,s +1) k 2 2 (u∞,u )
|u| 2 Lsc (H u ∞,u ) Lsc (H u )
1 Z u
δ 4 0 X 0 0
+ 1 kΨ(s,s +1) kL2 (u∞,u ) + |u0 |−2 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 ,
|u| 2 sc (H u )
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
67
0
Similarly, if we multiply |u|−2λ+2s on both sides of Proposition 2.4.3, we derive
1 0
|u|
1 0 1 (s,s0 +1) 2
1 0 1
≤ kψ (s,s ) kL4sc (Su∞ ,u ) + 3 kΨ k 2 (u∞,u ) k∇Ψ(s,s +1) k 2 2 (u∞,u )
|u∞ | |u| 2 Lsc (H u ) Lsc (H u )
1 Z u
δ4 (s,s0 +1)
X 0 0
+ 3 kΨ kL2 (H (u∞,u ) ) + |u0 |−3 kψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 .
|u| 2 sc u
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Consequently, we obtain
Proposition 2.4.5. Under the assumptions of Theorem 2.2.1 and bootstrap assumption (2.3.1),
the following inequalities hold
1 1
1 1 1 δ4 δ2
kωkL4sc (Su,u ) ≤ kωkL4sc (Su∞ ,u ) + 1 R0 R1 +
2 2
1 R0 + ∆0 O0,4 ,
|u| 2 |u| 2 |u|2
1 1 1 1 3
δ4 δ4 δ4 1 1 δ2 δ4
kχ̂kL4sc (Su,u ) ≤ kχ̂kL4sc (Su∞ ,u ) + 3 R0 R1 +
2 2
3 R0 + ∆0 O0,4 ,
|u| |u∞ | |u| 2 |u| 2 |u|2
1 1
1 1 1 1 1 1 δ4 δ2
kηkL4sc (Su,u ) ≤ kηkL4sc (Su∞ ,u ) + R02 R12 + δ 4 R0 + 1 R0 R1 +
2 2
1 R0 + ∆0 O0,4 .
|u| 2 |u| 2 |u|
3 3 3
δ4 δ4 δ4
2
ktrχkL4sc (Su,u ) ≤ 2
ktrχkL4sc (Su,u ) + ∆0 O0,4 .
|u| |u∞ | |u|
Proof. The null Ricci coefficient ω satisfies
1
∇3 ω = 2ωω + ρ + (η, η)(η, η).
2
Since no anomalous term appears in the equation above. The conclusion follows by using Proposition
2.4.4.
68
The null Ricci coefficient χ̂ obeys
∇3 χ̂ + trχ χ̂ = −2ωχ̂ − α.
With Proposition 2.4.4, we derive
1
kχ̂kL4sc (Su,u )
|u|
1 1 1 1
≤ kχ̂kL4sc (Su∞ ,u ) + 3 kαk 2
2
(u∞,u ) kαk 2
2
(u )
|u∞ | |u| 2 Lsc (H u ) Lsc (H u ∞,u )
1 Z u
δ4
+ 3 kαk 2
Lsc (H u
(u∞,u ) +
)
|u0 |−3 kωχ̂kL4sc (Su0 ,u ) du0
|u| 2 u∞
1 1 1 1
2
(u∞,u ) kαk 2
2
(u )
|u∞ | |u| 2 Lsc (H u ) Lsc (H u ∞,u )
1 Z u 1
δ4 δ2 1
+ 3 kαk 2 (u∞,u ) +
0 3
kωkL4sc (Su0 ,u ) 0 kχ̂kL∞ sc (Su0 ,u )
du0
u∞ |u | |u |
L (H )
|u| 2 sc u
1 1 1 1
2
(u∞,u ) k∇αk 2
2
(u )
|u∞ | |u| 2 Lsc (H u ) Lsc (H u ∞,u )
1 1
δ4 δ2
+ 3 kαkL2 (u∞,u ) + ∆0 [χ̂]O0,4 [ω]du0 .
|u| 2 sc (H u ) |u|2
1
Multiplying δ 4 on both sides, we obtain the desired estimate.
For η, we have null structure equation
1 1
∇3 η + trχ η = trχη − χ̂ · (η − η) + β.
2 2
There is a borderline term trχ η. With evolution lemma (Proposition 2.3.8) and
1
λ[η] = s2 (η) =
2
69
we obtain
|u|−1 kηkL4sc (Su,u )

Z u
−1
≤|u∞ | kηkL4sc (Su∞ ,u ) + |u0 |−3 kβkL4sc (Su0 ,u ) du0
u∞
Z u
+ |u0 |−3 kχ̂ · (η − η)kL4sc (Su0 ,u ) du0
u∞
Z u
+ |u0 |−3 ktrχηkL4sc (Su0 ,u ) du0
u∞
1 1 1
≤|u∞ |−1 kηkL4sc (Su∞ ,u ) + 3 kβk 2 2 (u∞,u ) k∇βk 2 2 (u∞,u )
|u| 2 Lsc (H u ) Lsc (H u )
1 1
δ 4 δ 2 1
+ 3 kβkL2 (u∞,u ) + ∆0 [χ̂]O0,4 [η, η] + O0,4 [η].
|u| 2 sc (H u ) |u|2 |u|
Multiply |u| on both sides, we deduce
1 1
1δ2 1 1 δ4
kηkL4sc (Su,u ) ≤ kηkL4sc (Su∞ ,u ) + O0,4 (η) + 1 R0 R1 + 1 R0 + ∆0 O0,4 .
2 2
|u| 2 |u| 2 |u|
In Proposition 2.4.2, we have proved
1
1 1 1 δ2
kηkL4sc (Su,u ) ≤ R02 R12 + δ 4 R0 + ∆0 O0,4 .
|u|
Putting these together, we get
1 1
1 1 1 1 1 1 δ4 δ2
kηkL4sc (Su,u ) ≤ kηkL4sc (Su∞ ,u ) + R02 R12 + δ 4 R0 + 1 R0 R1 +
2 2
1 R0 + ∆0 O0,4 .
|u| 2 |u| 2 |u|
The null Ricci coefficient trχ obeys
1
∇3 trχ + trχtrχ = −2ωtrχ − |χ̂|2 .
2
With evolution lemma (Proposition 2.3.8) and
1
λ[trχ] = , s2 (trχ) = 1,
2
70
we derive
|u|−2 ktrχkL4sc (Su,u )

Z u
−2
≤|u∞ | ktrχkL4sc (Su∞ ,u ) + |u0 |−4 kχ̂ χ̂kL4sc (Su0 ,u ) du0
u∞
Z u
+ |u0 |−4 ktrχωkL4sc (Su0 ,u ) du0
u∞
Z u 1 1
δ4 1 δ4
≤|u∞ |−2 ktrχkL4sc (Su∞ ,u ) + 0 0
kχ̂kL∞
sc (Su0 ,u )
kχ̂kL4sc (Su0 ,u ) du0
u∞
2
|u | |u | |u0 |
Z u 1
δ2
+ |u0 |−3 0 2 ktrχkL∞
sc (Su0 ,u )
kωkL4sc (Su0 ,u ) du0
u∞ |u |
1
δ4 1
≤|u∞ |−2 ktrχkL4sc (Su∞ ,u ) + ∆0 [χ̂]O0,4 [χ̂] + 2 ∆0 [trχ]O0,4 [ω].
|u| |u|
3
Multiplying δ 4 on both sides, we finish the proof.
Collecting all the estimates in this section, we conclude
Proposition 2.4.6. Assume
R < ∞, R < ∞, O0,∞ < ∞,
then there exist δ0 = δ0 (I (0) , R(0) , R, R, O0,∞ ), such that whenever 0 < δ < δ0 , in D we have

(0)
O0,4 ≤ C I +R+R ,
where C is a large universal constant.
Proof. As a consequence of the estimates above, by taking δ0 sufficient small we derive
1
1 1 1 1 1 δ4
O0,4 ≤ I (0) + R02 R12 + R0 + 1 R0 R1 +
2 2
R .
|u| 2 |u| 0
By setting
1
1 1 1 1 1 δ4
∆1 I (0) + R02 R12 + R0 + 1 R0 R1 +
2 2
R ,
|u| 2 |u| 0
71
we have improved bootstrap assumption (2.3.2).
2.4.2 Estimates for First Derivatives of Ricci Coefficients
We now move to the L4 (Su,u ) estimate for first angular derivatives of the Ricci coefficients. With
the help of commutation formulas (Propositions 2.3.12, 2.3.13 and 2.3.14), it is easy to verify that
the null Ricci coefficient ψ with signature (s, s0 ) satisfies either transport equation
0
∇4 ∇ψ (s,s )
X 0 0 0
= ∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + ∇Ψ(s+1,s )
s1 +s2 =s+1,
s01 +s02 =s0 (2.4.3)
X 0 0 X 0 0 0
+ ψ (s3 ,s3 ) · Ψ(s4 ,s4 ) + ψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) ,
s3 +s4 =s+ 3 , s5 +s6 +s7 =s+ 3 ,
2 2
s03 +s04 =s0 + 1 s05 +s06 +s07 =s0 + 1
2 2
or

0 0 1 0
∇3 ∇ψ (s,s ) + λ[ψ (s,s ) ] + trχ∇ψ (s,s )
2
X 0 0 0
= ∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + ∇Ψ(s,s +1)
s1 +s2 =s,
s01 +s02 =s0 +1
(2.4.4)
X 0 0 X 0 0 0
+ ψ (s3 ,s3 ) · Ψ(s4 ,s4 ) + ψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) .
s3 +s4 =s+ 1 , s5 +s6 +s7 =s+ 1 ,
2 2
s03 +s04 =s0 + 3 s05 +s06 +s07 =s0 + 3
2 2
As a consequence, we have
0
Proposition 2.4.7. For ψ (s,s ) satisfying (2.4.3), the following inequality holds
72
0
k∇ψ (s,s ) kL4sc (Su,u )
1
0 X δ2 0 0
≤k∇ψ (s,s ) kL4sc (Su,0 ) + sup k∇ψ (s1 ,s1 ) kL4sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
u0 ∈[0,δ] s1 +s2 =s+1,
|u|
s01 +s02 =s0
0 1 0 1 1 0
+ k∇Ψ(s+1,s ) k 2 2 (0,u) k∇2 Ψ(s+1,s ) k 2 2 (0,u) + δ 4 k∇Ψ(s+1,s ) kL2 (0,u)
1
δ4 X 0 1 0 1 0 1
+ sup kψ (s3 ,s3 ) kL∞
sc (Su,u0 )
δ 4 kΨ(s4 ,s4 ) k 2 2 (0,u) k∇Ψ(s4 ,s4 ) k 2 2 (0,u)
|u| u0 ∈[0,δ] Lsc (Hu ) Lsc (Hu )
s3 +s4 =s+ 3 ,
2
s03 +s04 =s0 + 1
2
1
δ4 X 0 1 0
sc (Su,u0 )
δ 2 kΨ(s4 ,s4 ) kL2 (0,u)
|u| u0 ∈[0,δ] sc (Hu )
s3 +s4 =s+ 3 ,
2
s03 +s04 =s0 + 1
2
δ X 0 0 0
sc (Su,u0 )
kψ (s6 ,s6 ) kL∞
sc (Su,u0 )
kψ (s7 ,s7 ) kL4sc (Su,u0 ) .
|u|2 u0 ∈[0,δ]
s5 +s6 +s7 =s+ 3 ,
2
s05 +s06 +s0 7=s0 + 1
2
Proof. Using Proposition 2.3.7 for transport equations, we obtain
0
k∇ψ (s,s ) kL4sc (Su,u )
Z u
0 0
≤k∇ψ (s,s ) kL4sc (Su,0 ) + δ −1 k∇4 ∇ψ (s,s ) kL4sc (Su,u0 ) du0
0
(s,s0 ) 0 0
X
≤k∇ψ kL4sc (Su,0 ) + sup k∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su,u0 )
u0 ∈[0,δ] s1 +s2 =s+1,
s01 +s02 =s0
Z u
0
+ δ −1 k∇Ψ(s+1,s ) kL4sc (Su,u0 ) du0
Z0 u X 0 0
+ δ −1 kψ (s3 ,s3 ) · Ψ(s4 ,s4 ) kL4sc (Su,u0 ) du0
0 s3 +s4 =s+ 3 ,
2
s03 +s04 =s0 + 1
2
Z u X 0 0 0
+ δ −1 kψ (s5 ,s5 ) · ψ (s6 ,s6 ) · ψ (s7 ,s7 ) kL4sc (Su,u0 ) du0
0 s5 +s6 +s7 =s+ 3 ,
2
s05 +s06 +s0 7=s0 + 1
2
73
1
0 X δ2 0 0
≤k∇ψ (s,s ) kL4sc (Su,0 ) + sup k∇ψ (s1 ,s1 ) kL4sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
u0 ∈[0,δ] s1 +s2 =s+1,
|u|
s01 +s02 =s0
0 1 0 1 1 0
+ k∇Ψ(s+1,s ) k 2 2 (0,u) k∇2 Ψ(s+1,s ) k 2 2 (0,u) + δ 4 k∇Ψ(s+1,s ) kL2 (0,u)
1
δ4 X 0 1 0 1 0 1
sc (Su,u0 )
δ 4 kΨ(s4 ,s4 ) k 2 2 (0,u) k∇Ψ(s4 ,s4 ) k 2 2 (0,u)
|u| u0 ∈[0,δ] Lsc (Hu ) Lsc (Hu )
s3 +s4 =s+ 3 ,
2
s03 +s04 =s0 + 1
2
1
δ4 X 0 1 0
sc (Su,u0 )
δ 2 kΨ(s4 ,s4 ) kL2 (0,u)
|u| u0 ∈[0,δ] sc (Hu )
s3 +s4 =s+ 3 ,
2
s03 +s04 =s0 + 1
2
δ X 0 0 0
sc (Su,u0 )
kψ (s6 ,s6 ) kL∞
sc (Su,u0 )
kψ (s7 ,s7 ) kL4sc (Su,u0 ) ,
|u|2 u0 ∈[0,δ]
s5 +s6 +s7 =s+ 3 ,
2
s05 +s06 +s0 7=s0 + 1
2
With this preparation, it follows that
Proposition 2.4.8. Under the assumptions of Theorem 2.2.1 and the bootstrap assumptions (2.3.1)
and (2.3.2), we have
1 3
1 1 1 δ2 1 1 δ4
k∇η, ∇ω, ∇trχkL4sc (Su,u ) ≤R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 ,
|u| |u| 0
and
1 1
1 1δ4 1 11 δ4
k∇χ̂kL4sc (Su,u ) ≤R1 R2 + δ R1 +
2 2
∆0 R02 R12 +
4 ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 .
|u| |u| 0
Proof. The null Ricci coefficients ∇η, ∇ω, ∇trχ satisfy
74
∇4 ∇η = ∇β + (η, η)β + (η, η)∇χ + χ(∇η, ∇η) + χ(η, η)(η, η),
∇4 ∇ω =∇ρ + (η, η)ρ + ω∇ω + (χ, ω)∇ω + (η, η)(∇η, ∇η)
+ (η, η)ωω + (η, η)(η, η)(η, η),
and
∇4 ∇trχ = (ω, χ)∇trχ + χ̂∇χ̂ + trχ∇ω + (η, η)(ω, χ)(ω, χ).
There is no anomalous term in these equations. The conclusion follows from Proposition 2.4.7.
For χ̂, we have
∇4 ∇χ̂ =∇α + (η, η)α + χ̂β + χ̂∇trχ + (ω, χ)∇χ̂ + χ̂∇ω
+ (η, η)(χ, ω)χ̂.
The anomalous term is due to α. And the terms including ∇α are normal. Recalling that
1
δ 2 kαkL2sc (H) ≤ R0 ,
we obtain
1 1
1 1 1 δ4 1 1 δ4
k∇χ̂kL4sc (Su,u ) ≤R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 .
|u| |u| 0
0
We still need to derive estimates for the other Ricci coefficients ψ (s,s ) , and it is easy to verify
75
0
that these ψ (s,s ) satisfy

0 0 1 0
∇3 ∇ψ (s,s ) + λ[ψ (s,s ) ] + trχ∇ψ (s,s )
2
X 0 0 0
= ∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) + ∇Ψ(s,s +1)
s1 +s2 =s,
s01 +s02 =s0 +1
(2.4.5)
X 0 0 X 0 0 0
s3 +s4 =s+ 1 , s5 +s6 +s7 =s+ 1 ,
2 2
s03 +s04 =s0 + 3 s05 +s06 +s07 =s0 + 3
2 2
0
Proposition 2.4.9. For ψ s,s obeys (2.4.5), we have
0 0
|u|2λ−2s −1 k∇ψ (s,s ) kL4sc (Su,u )
Z u
0 0 0 0
≤|u∞ |2λ−2s −1 k∇ψ (s,s ) kL4sc (Su∞ ,u ) + |u0 |2λ−2s −3 k∇Ψ(s,s +1) kL4sc (Su0 ,u ) du0
u∞
Z u
0 2λ−2s0 −3 (s1 ,s01 ) 0
X
+ |u | k∇ψ · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Z u
0 X 0 0
+ |u0 |2λ−2s −3 kψ (s3 ,s3 ) · Ψ(s4 ,s4 ) kL4sc (Su0 ,u ) du0
u∞ s3 +s4 =s+ 1 ,
2
s03 +s04 =s0 + 3
2
Z u
0 X 0 0 0
+ |u0 |2λ−2s −3 kψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) kL4sc (Su0 ,u ) du0 .
u∞ s5 +s6 +s7 =s+ 1 ,
2
s05 +s06 +s07 =s0 + 3
2
Proof. Using evolution lemma (Proposition 2.3.8) for
1
p = 4, λ1 = 2λ + 1 − ,
2
76
we derive
1 0
|u|2λ+ 2 k∇ψ (s,s ) kL4 (Su,u )
Z u
2λ+ 12 (s,s0 ) 1 0
≤|u∞ | k∇ψ
kL4 (Su∞ ,u)) + |u0 |2λ+ 2 k∇3 ∇ψ (s,s ) kL4 (Su0 ,u ) du0
u∞
Z u
2λ+ 21 (s,s0 ) 1 0
≤|u∞ | k∇ψ kL4 (Su∞ ,u) + |u0 |2λ+ 2 k∇Ψ(s,s +1) kL4 (Su0 ,u ) du0
u∞
Z u
1
X 0 0
+ |u0 |2λ+ 2 k∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4 (Su0 ,u ) du0
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Z u
1
X 0 0
+ |u0 |2λ+ 2 kψ (s3 ,s3 ) · Ψ(s4 ,s4 ) kL4 (Su0 ,u ) du0
u∞ s3 +s4 =s+ 1 ,
2
s03 +s04 =s0 + 3
2
Z u
1
X 0 0 0
+ |u0 |2λ+ 2 kψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) kL4 (Su0 ,u ) du0 .
u∞ s5 +s6 +s7 =s+ 1 ,
2
s05 +s06 +s07 =s0 + 3
2
With the definition

0 3 0 1 0
kψ (s,s ) kL4sc (Su,u ) = δ s− 4 |u|2s + 2 kψ (s,s ) kL4 (Su,u ) ,
we deduce the following estimate in scale invariant norms:
0 0
|u|2λ−2s −1 k∇ψ (s,s ) kL4sc (Su,u )
Z u
0 0 0 0
≤|u∞ |2λ−2s −1 k∇ψ (s,s ) kL4sc (Su∞ ,u ) + |u0 |2λ−2s −3 k∇Ψ(s,s +1) kL4sc (Su0 ,u ) du0
u∞
Z u
0 2λ−2s0 −3 (s1 ,s01 ) 0
X
+ |u | k∇ψ · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =s,
s01 +s02 =s0 +1
Z u
0 X 0 0
+ |u0 |2λ−2s −3 kψ (s3 ,s3 ) · Ψ(s4 ,s4 ) kL4sc (Su0 ,u ) du0
u∞ s3 +s4 =s+ 1 ,
2
s03 +s04 =s0 + 3
2
Z u
0 X 0 0 0
+ |u0 |2λ−2s −3 kψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) kL4sc (Su0 ,u ) du0 .
u∞ s5 +s6 +s7 =s+ 1 ,
2
s05 +s06 +s07 =s0 + 3
2
As a consequence, we conclude
77
(2.3.1) and (2.3.2), the following inequalities hold
1 1
1 1 1 δ4 δ2 1 1
k∇ωkL4sc (Su,u ) ≤k∇ωkL4sc (Su∞ ,u ) + 1 R12 R22 + 1 R1 + 3 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1
δ 4 δ 2 δ
+ 3 ∆0 R0 + ∆0 O1,4 + 3 O0,4 ∆20 .
|u| 2 |u| |u|
1 3
1 1 1 δ2 1 1 δ4
k∇ηkL4sc (Su,u ) ≤k∇ηkL4sc (Su∞ ,u ) + R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1 1
1 1 1 δ4 δ2 1 1
+ 1 R1 R2 +
2 2
1 R1 + 3 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1 1
δ 4 δ 2 δ2
+ 3 ∆0 R0 + ∆0 O1,4 + O0,4 ∆20 .
|u| 2 |u| |u|
1 1
1 1 1 1 δ4 δ2 1 1
k∇χ̂kL4sc (Su,u ) ≤|u∞ |−1 k∇χ̂kL4sc (Su∞ ,u ) + 3 R1 R2 +
2 2
3 R1 + 3 ∆0 R0 R1
2 2
|u| |u| 2 |u| 2 |u| 2

3 1 1
δ4 δ2 δ2
+ 3 ∆0 R0 + ∆0 O1,4 + O0,4 ∆20 ,
|u| 2 |u| |u|
Proof. The null Ricci coefficient ω satisfies
1
∇3 ∇ω + trχ∇ω =∇ρ + (η, η)ρ + ω∇ω + χ̂∇ω + ω∇ω + (η, η)(∇η, ∇η)
2
+ (η, η)ωω + (η, η)(η, η)(η, η).
The anomalous term on the right hand side is due to χ̂. With the fact that
λ(ω) = s2 (ω) = 0,
78
and employing Proposition 2.4.9, we deduce
|u|−1 k∇ωkL4sc (Su,u )

Z u
≤|u∞ |−1 k∇ωkL4sc (Su∞ ,u ) + |u0 |−3 k∇Ψ(1,1) kL4sc (Su0 ,u ) du0
u∞
Z u X 0 0
+ |u0 |−3 k∇ψ (s1 ,s1 ) · ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =1,
s01 +s02 =1
Z u
1 1
+ |u0 |−3 δ 2 k∇ωkL4sc (Su0 ,u ) kχ̂kL∞
sc (Su0 ,u )
du0
u∞ |u0 |
Z u X 0 0
+ |u0 |−3 kψ (s3 ,s3 ) · Ψ(s4 ,s4 ) kL4sc (Su0 ,u ) du0
u∞ s3 +s4 = 3 ,
2
s03 +s04 = 3
2
Z u X 0 0 0
+ |u0 |−3 kψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) kL4sc (Su0 ,u ) du0
u∞ s5 +s6 +s7 = 3 ,
2
s05 +s06 +s07 = 3
2
1 1
1 1 1 δ4 δ2 1 1
≤|u∞ |−1 k∇ωkL4sc (Su∞ ,u ) + 3 R12 R22 + 3 R1 + 5 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1
δ 4 δ2 δ
+ 5 ∆0 R0 + ∆0 O1,4 + 4 O0,4 ∆20 ,
|u| 2 |u|2 |u|
where we use L4 Sobolev Embedding (Proposition 2.3.9) and Hölder’s inequality. Multiplying |u|
on both sides, the conclusion follows.
The null Ricci coefficient η obeys
∇3 ∇η + trχ∇η =trχ∇η + ∇β + (η, η)β + (η, η)∇trχ + (η, η)∇χ̂
+ χ̂∇(η, η) + trχηη + χ̂ηη.
Recalling that
1
λ(η) = s2 (η) = ,
2
79
Using Proposition 2.4.9, we obtain
|u|−1 k∇ηkL4sc (Su,u )

Z u
≤|u∞ |−1 k∇ηkL4sc (Su∞ ,u ) + |u0 |−3 ktrχ∇ηkL4sc (Su0 ,u ) du0
u∞
Z u Z u
+ |u0 |−3 k∇βkL4sc (Su0 ,u ) du0 + |u0 |−3 k(η, η)βkL4sc (Su0 ,u ) du0
u u∞
Z u∞
+ |u0 |−3 k(η, η)∇trχ + (η, η)∇χ̂ + χ̂∇(η, η)kL4sc (Su0 ,u ) du0
u
Z u∞
+ |u0 |−3 ktrχηη + χ̂ηηkL4sc (Su0 ,u ) du0
u∞
Z u 1
−1 1 δ2
≤|u∞ | k∇ηkL4sc (Su∞ ,u ) + ktrχkL∞
sc (Su0 ,u )
k∇ηkL4sc (Su0 ,u ) du0
u∞ |u0 |2 |u0 |2
Z u Z u
0 −3 0
+ |u | k∇βkL4sc (Su0 ,u ) du + |u0 |−3 k(η, η)βkL4sc (Su0 ,u ) du0
u∞ u∞
Z u 1
δ2 1
+ kη, ηkL∞
sc (Su0 ,u )
k∇trχ, ∇χ̂kL4sc (Su0 ,u ) du0
u∞ |u0 |3 |u0 |
Z u 1
δ2 1
+ kχ̂kL∞
sc (Su0 ,u )
k∇(η, η)kL4sc (Su0 ,u ) du0
u∞ |u0 |3 |u0 |
Z u 1 1
δ2 δ2
+ ktrχkL∞ sc (Su0 ,u )
kηkL∞ sc (Su0 ,u )
u∞ |u0 |3 |u0 |2
Z u
δ 1
+ kχ̂kL∞sc (Su0 ,u )
kηkL∞ sc (Su0 ,u )
u∞ |u|4 |u0 |
1 1
1 1 1 1 δ4 δ2 1 1
≤|u∞ |−1 k∇ηkL4sc (Su∞ ,u ) + O1,4 (η) + 3 R1 R2 +
2 2
3 R1 + 5 ∆0 R0 R1
2 2
|u| |u| 2 |u| 2 |u| 2

3 1 1
δ4 δ2 δ2
+ 5 ∆0 R0 + 2
∆0 O1,4 + 2 O0,4 ∆20 .
|u| 2 |u| |u|
Multiplying |u| on both sides, we obtain
1 1
1 1 1 δ4 δ2 1 1
k∇ηkL4sc (Su,u ) ≤k∇ηkL4sc (Su∞ ,u ) + O1,4 (η) + 1 R12 R22 + 1 R1 + 3 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1 1
δ 4 δ2 δ 2
+ ∆0 R0 + ∆0 O1,4 + ∆2 O0,4 .
|u|
3
2 |u| |u| 0
80
In Proposition 2.4.8, we have proved
1 3
1 1δ2 1
1 1 δ4
k∇ηkL4sc (Su,u ) ≤R1 R2 + δ R1 +
2 2
∆0 R02 R12 +
4 ∆0 R 0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 ,
|u| |u| 0
Collecting these together, we derive
1 3
1 1 1 δ2 1 1 δ4
k∇ηkL4sc (Su,u ) ≤k∇ηkL4sc (Su∞ ,u ) + R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1 1
1 1 1 δ4 δ2 1 1
+ 1 R12 R22 + 1 R1 + 3 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1 1
δ 4 δ δ2 2
+ 3 ∆0 R0 + ∆0 O1,4 + O0,4 ∆20 .
|u| 2 |u| |u|
For χ̂, by a straightforward computation, we derive
3
∇3 ∇χ̂ + trχ∇χ̂ =∇α + (η, η)α + χ̂ β + χ̂∇trχ + χ̂∇ω + (ω, χ̂)∇χ̂
2
+ (η, η)trχ χ̂ + (η, η)ωχ̂ + (η, η)χ̂ χ̂.
Since
λ(χ̂) = s2 (χ̂) = 1,
by employing Proposition 2.4.9, we deduce
|u|−1 k∇χ̂kL4sc (Su,u )

Z u
≤|u∞ |−1 k∇χ̂kL4sc (Su∞ ,u ) + |u0 |−3 k∇αkL4sc (Su0 ,u ) du0
u∞
81
Z u Z u
0 −3 0
+ |u | k(η, η)αkL4sc (Su0 ,u ) du + |u0 |−3 kχ̂ βkL4sc (Su0 ,u ) du0
u u∞
Z u∞
+ |u0 |−3 kχ̂∇trχ + χ̂∇ω + (ω, χ̂)∇χ̂kL4sc (Su0 ,u ) du0
u
Z u∞
+ |u0 |−3 k(η, η)trχ χ̂ + (η, η)χ̂(χ̂, ω)kL4sc (Su0 ,u ) du0
u∞
Z u
−1 1
≤|u∞ | k∇χ̂kL4sc (Su∞ ,u ) + 0 3
k∇αkL4sc (Su0 ,u ) du0
u∞ |u |
Z u Z u
0 −3 0 1 1
+ |u | k(η, η)αkL4sc (Su0 ,u ) du + |u0 |−3 δ 2 kβkL4sc (Su0 ,u ) 0 kχ̂kL∞
sc (Su0 ,u )
du0
u∞ u∞ |u |
Z u 1
δ2 1
+ 0 |3 |u0 |
kχ̂kL∞sc (Su0 ,u )
k∇trχ, ∇ωkL4sc (Su0 ,u ) du0
u∞ |u
Z u 1
δ2 1 1
+ 0 0
kχ̂, ωkL∞ sc (Su0 ,u )
k∇χ̂kL4sc (Su0 ,u ) du0
2
u∞ |u | |u | |u0 |
Z u 1 1
δ2 δ2 1
+ 0 |2 |u0 |2
ktrχkL∞ sc (Su0 ,u ) 0|
kχ̂kL∞
sc (Su0 ,u )
kη, ηkL4sc (Su0 ,u ) du0
u∞ |u |u
Z u
δ 1 1
+ 0 |3 |u0 |
kχ̂kL∞sc (Su0 ,u ) 0|
kχ̂, ωkL∞ sc (Su0 ,u )
u∞ |u |u
1 1
1 1 1 δ4 δ2 1 1
≤|u∞ |−1 k∇χ̂kL4sc (Su∞ ,u ) + 3 R1 R2 +
2 2
3 R1 + 3 ∆0 R02 R12
|u| 2 |u| 2 |u| 2
3 1 1
δ 4 δ2 δ 2
+ 3 ∆0 R 0 + ∆0 O1,4 + O0,4 ∆20 ,
|u| 2 |u| |u|
which is the desired estimate.
Estimate for trχ
We now focus on L4 (Su,u ) estimate for ∇trχ. If we trivially employ transport equation for
∇3 ∇trχ, there will be a (η, η)trχtrχ term. After applying evolution lemma (Proposition 2.3.8),
this term will lead to a logarithmic divergence. To cure this problem, we substitute ∇(Ωtrχ − u2 )
for ∇trχ. In the transport equation for ∇3 ∇(Ωtrχ − u2 ), the borderline term (η, η)trχtrχ is replace
by (η, η)trχ(Ωtrχ − u2 ).
By introducing a new bootstrap assumption
1 2
kΩtrχ − kL∞ (S ) ≤ ∆4 , (2.4.6)
|u| u sc u,u
82
we can prove
Proposition 2.4.11. Under the assumption of Theorem 2.2.1 and bootstrap assumptions (2.3.1),(2.3.2),
(2.3.3), (2.4.6), we have
1 3
1 1 1 δ2 1 1 δ4
k∇trχkL4sc (Su,u ) ≤k∇trχkL4sc (Su∞ ,u ) + R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R 0
|u| |u|
1 1
δ2 δ2 1
+ ∆0 O1,4 + O0,4 ∆20 + δ 2 (∆0 + ∆4 )(O0,4 + O1,4 )
|u| |u|
1
+ δ 2 O0,4 (∆0 + ∆4 )2 .
Proof. Thanks to the fact

1
ω=− ∇3 Ω,
2Ω
we calculate
2
∇3 (Ωtrχ − )
u
2
=∇3 Ωtrχ + Ω∇3 trχ − ∇3 ( )
u
2Ω−1

1 2 2
= − 2Ωωtrχ + Ω − (trχ) − 2ωtrχ − |χ̂| +
2 u2
−1

Ω 4
= − 4Ωωtrχ − Ω|χ̂|2 − (Ωtrχ)2 − 2
2 u
Ω−1 2 2
= − 4Ωωtrχ − Ω|χ̂|2 − (Ωtrχ − )(Ωtrχ + ).
2 u u
Therefore, we obtain
2 2 Ω−1 2
∇3 (Ωtrχ − ) + trχ(Ωtrχ − ) = −4Ωωtrχ − Ω|χ̂|2 + (Ωtrχ − )2 . (2.4.7)
u u 2 u
83
Applying (2.9.2) and commutation formula (Proposition 2.3.12), we compute
2 3 2
∇3 ∇(Ωtrχ − ) + trχ∇(Ωtrχ − )
u 2 u
2 2
=χ̂∇(Ωtrχ − ) + ∇trχ(Ωtrχ − ) + ∇Ωωtrχ + Ω∇ωtrχ + Ωω∇trχ
u u
2 2 2
+ ∇Ωχ̂ χ̂ + Ωχ̂∇χ̂ + ∇ΩΩ−2 (Ωtrχ − )2 + Ω−1 ∇(Ωtrχ − )(Ωtrχ − )
u u u
2 2
+ (η, η)trχ(Ωtrχ − ) + (η, η)Ωωtrχ + (η, η)Ωχ̂ χ̂ + (η, η)Ω−1 (Ωtrχ − )2 .
u u
Combining the fact
ηA + η A = 2∇A (log Ω),
the bound for Ω in Proposition 2.3.1, bootstrap assumption (2.4.6), evolution lemma (Proposition
2.3.8) and Gronwall’s inequality, we conclude
2
|u|−1 k∇(Ωtrχ − )kL4sc (Su,u )
u
Z u 1
2 1 δ2
≤|u∞ |−1 k∇(Ωtrχ − )kL4sc (Su∞ ,u ) + 0 |2 |u0 |2
ktrχkL∞ sc (Su0 ,u )
k∇ωkL4sc (Su0 ,u ) du0
u u∞ |u
Z u 1
δ2 1 2
+ 0 3 0
kΩtrχ − , ωkL∞ sc (Su0 ,u )
k∇trχkL4sc (Su0 ,u ) du0
u∞ |u | |u | u
Z u 1
δ2 1 1
+ 0 |2 |u0 |
kχ̂kL∞
sc (Su0 ,u ) 0|
k∇χ̂kL4sc (Su0 ,u ) du0
u∞ |u |u
Z u 1 1
δ2 δ2 1 2 0
+ ktrχkL∞ sc (Su0 ,u )
kη, ηkL4sc (Su0 ,u ) 0 kΩtrχ − kL∞ (S ) du
0 2
u∞ |u | |u |
0 2 |u | u sc u0 ,u
Z u 1 1
δ2 δ2
+ 0 3 0 2
kη, ηkL∞ sc (Su0 ,u )
kωkL4sc (Su0 ,u ) du0
u∞ |u | |u |
Z u
δ 1 2 1 2
+ 0 |2 |u0 |
kΩtrχ − , χ̂kL∞ sc (Su0 ,u ) 0|
kΩtrχ − , χ̂kL∞ sc (Su0 ,u )
u∞ |u u |u u
1 1
1 δ2 δ2
≤|u∞ |−1 k∇trχkL4sc (Su∞ ,u ) + O1,4 [ω] + (∆0 + ∆4 )O1,4 + O0,4 (∆0 + ∆4 )2 .
|u| |u| |u|
Multiplying |u| on both sides, we obtain
1 1
k∇(Ωtrχ)kL4sc (Su,u ) ≤ k∇trχkL4sc (Su∞ ,u ) + O1,4 [ω] + δ 2 (∆0 + ∆4 )O1,4 + δ 2 O0,4 (∆0 + ∆4 )2 .
84
To get desired bound for k∇trχkL4sc (Su,u ) , we need derive estimate for k∇ΩkL4sc (Su,u ) . And we
have the following lemma:
Lemma 2.4.12. Under the assumption of Theorem 2.2.1 and bootstrap assumptions (2.3.1)-(2.3.3),
the following inequality holds
1
δ2
|u|k∇ΩkL4sc (Su,u ) ≤ O1,4 [ω] + ∆0 O0,4 .
|u|
Proof. For Ω, we have
∇3 Ω = −2Ωω.
Applying commutation formula (Proposition 2.3.12), we obtain
1
∇3 ∇Ω + trχ∇Ω = (χ̂, ω)∇Ω + Ω∇ω + (η, η)Ωω.
2
With Gronwall’s inequality, we get
Z u
1
|u|−1 k∇ΩkL4sc (Su,u ) ≤|u∞ |−1 k∇ΩkL4sc (Su∞ ,u ) + k∇ωkL4sc (Su0 ,u ) du0
u∞ |u0 |3
Z u 1
δ 2
+ kη, ηkL4sc (Su0 ,u ) kωkL∞
sc (Su0 ,u )
du0
u∞ |u0 |4
1
1 δ2
≤ 2
O1,4 [ω] + 3 ∆0 O0,4 ,
|u| |u|
(0,u)
where we employ ∇Ω = 0 along Hu∞ . Multiplying |u|2 on both sides, we derive
1
δ2
|u|k∇ΩkL4sc (Su,u ) ≤ O1,4 [ω] + ∆0 O0,4 .
|u|
85
Therefore, together with Lemma 2.4.12, we infer
k∇trχkL4sc (Su,u )
≤kΩ∇trχkL4sc (Su,u )
≤k∇(Ωtrχ)kL4sc (Su,u ) + k(∇Ω)trχkL4sc (Su,u )

1
δ2
≤k∇(Ωtrχ)kL4sc (Su,u ) + |u|k∇ΩkL4sc (Su,u ) ktrχkL∞
sc (Su,u )
|u|2
1 1
≤k∇trχkL4sc (Su∞ ,u ) + O1,4 [ω] + δ 2 (∆0 + ∆4 )(O0,4 + O1,4 ) + δ 2 O0,4 (∆0 + ∆4 )2 .
Recall the estimate obtained in Proposition 2.4.8
1 3
1 1 δ2 11 1 δ4
k∇ωk L4sc (Su,u ) ≤R1 R2 + δ R1 +
2 2
∆0 R02 R12 +
4 ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 .
|u| |u| 0
Putting these estimates altogether, we derive
1 3
1 1 1 δ2 1 1 δ4
k∇trχkL4sc (Su,u ) ≤k∇trχkL4sc (Su∞ ,u ) + R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R 0
|u| |u|
1 1
δ2 δ2 1
+ ∆0 O1,4 + O0,4 ∆20 + δ 2 (∆0 + ∆4 )(O0,4 + O1,4 )
|u| |u|
1
+ δ 2 O0,4 (∆0 + ∆4 )2 .
Thanks to the proposition above, we are ready to derive bound for kΩtrχ − u2 kL∞
sc (Su,u )
.
Proposition 2.4.13. Under the assumption of Theorem 2.2.1 and bootstrap assumptions (2.3.1)-
(2.3.3), we have
1 2 (0)
kΩtrχ − kL∞ (S ) ≤ C(I + R + R)5 ,
|u| u sc u,u
where C is a universal large constant.
86
Proof. Recall (2.9.2)
2 2 Ω−1 2
∇3 (Ωtrχ − ) + trχ(Ωtrχ − ) = −4Ωωtrχ − Ω|χ̂|2 + (Ωtrχ − )2 .
u u 2 u
Since
2 2
λ(Ωtrχ − ) = s2 (Ωtrχ − ) = 1,
u u
with the bound for Ω in Proposition 2.3.1, bootstrap assumptions (2.3.1) and (2.4.6) and Gronwall’s
inequality, we deduce
1 2
kΩtrχ − kL4sc (Su,u )
|u| u
Z u 1
1 2 1 δ2
≤ kΩtrχ − kL4sc (Su∞ ,u ) + 0 2
kωk L 4 (S 0 ) ktrχkL∞
sc (Su0 ,u )
du0
|u∞ | u u∞ |u |
sc u ,u
|u0 |2
Z u 1 1
δ4 δ4 1
+ 0 |2 |u0 |
kχ̂kL4sc (Su0 ,u ) 0 kχ̂kL∞ sc (Su0 ,u )
du0
u∞ |u |u |
1
1 2 1 δ4
≤ kΩtrχ − kL4sc (Su∞ ,u ) + O0,4 [ω] + ∆0 O0,4 .
|u∞ | u |u| |u|
Employing conclusions in Proposition 2.4.11 for ∇(Ωtrχ − u2 ) and L4 Sobolev Embedding (Propo-
sition 2.3.9) yields

1 2 (0)
kΩtrχ − kL∞ (S ) ≤ (I + R + R)5 .
|u| u sc u,u
By choosing ∆4 (I (0) + R + R)5 , we have improved bootstrap assumption (2.4.6) and obtain
desired estimate.
Gathering all the estimates above, we obtain
R < ∞, R < ∞, O0,∞ < ∞,
87
then there exist δ0 = δ0 (I (0) , R(0) , R, R, O0,∞ ), such that whenever δ ≤ δ0 , in D we have

(0)
O1,4 ≤ C I +R+R ,
Proof. From the estimates above, by taking δ0 sufficient small we have
1
1 1 1 1 1 δ4
O1,4 ≤ I (0) + R02 R12 + R0 + 1 R0 R1 +
2 2
1 R0 .
|u| 2 |u| 2
Let ∆2 I (0) + R + R. We have improved bootstrap assumption (2.3.3).
2.4.3 L∞ (S) Estimates for Ricci Coefficients
We then show how to bound the L∞ (Su,u ) norm of the Ricci coefficients.
O0,∞ Estimates for ω, trχ, η, η, ω
For ω, trχ, η, η, ω, we have
(2.3.3), it follows that

(0)
kω, trχ, η, η, ωkL∞
sc (Su,u )
≤C I +R+R ,
Proof. From L∞ Sobolev Embedding (Proposition 2.3.10) for any horizontal tensor φ in D, we
have
1 1 1
kφkL∞
sc (Su,u )
≤ kφkL2 4 k∇φkL4 (Su,u )
2
+ δ 4 kφkL4sc (Su,u ) .
sc (Su,u ) sc
Thanks to this inequality and Propositions 2.4.6 and 2.4.14, the conclusion follows.
88
O0,∞ Estimates for χ̂, χ̂
We complement the anomalous norms of χ̂, χ̂ by the local, non-anomalous, scale invariant norms
O0δ [χ̂](u, u) = sup kχ̂kL4sc (δ S u,u ) , O0δ [χ̂](u, u) = sup kχ̂kL4sc (δ S u,u )
δ S⊂S δ S⊂S
1
where δ S u,u is a disk of radius δ 2 |u| obtained by transporting from the initial data embedded in
Su,0 or Su∞ ,u .
With the aid of these norms, we can prove
(2.3.3), it follows that
1
1 1 1 δ4
kχ̂kL∞
sc (Su,u )
≤kαkL4sc (δ S u∞ ,u ) + 1 R12 R22 + 1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R 0 R 1 +
2 2
3 ∆0 R 0
|u| 2 |u| 2
1 1
1 1 1 δ4 1 1 δ4
+ R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 .
|u| |u| 0
Proof. For α, we have the null Bianchi equation
1
b + 4ωα − 3(χ̂ρ + ∗ χ̂σ) + (ζ + 4η)⊗β,
∇3 α + trχα = ∇⊗β b
2
Applying L4 Sobolev Embedding (Proposition 2.3.9), Gronwall’s inequality and evolution lemma
89
(Proposition 2.3.8), we have
Z u
1
kαkL4sc (δ S u,u ) ≤kαkL4sc (δ S u∞ ,u ) + 0 |2
k∇βkL4sc (δ S u,u ) du0
u∞ |u
Z u
1
+ 0 |2
kχ̂ · (ρ, σ) + (η, η)βkL4sc (δ S u,u ) du0
u∞ |u
Z u
1
≤kαkL4sc (δ S u∞ ,u ) + 0 |2
k∇βkL4sc (Su,u ) du0
u∞ |u
Z u
1
+ 0 |2
kχ̂ · (ρ, σ) + (η, η)βkL4sc (Su,u ) du0
u∞ |u
1
1 1 1 δ4
≤kαkL4sc (δ S u∞ ,u ) + 1 R1 R2 +
2 2
1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R0 R1 +
2 2
3 ∆0 R0 .
|u| 2 |u| 2
For χ̂, we have
∇4 χ̂ + trχχ̂ = −2ω χ̂ − α.
Using Gronwall’s inequality and Proposition 2.3.7 for transport equations, we derive
Z u
kχ̂kL4sc (δ S u,u ) ≤ δ −1 kαkL4sc (δ S u,u0 ) du0
0
Together with the estimate for kαkL4sc (δ S u,u ) above, we obtain
1
1 1 1 δ4
kχ̂kL4sc (δ S u,u ) ≤kαkL4sc (δ S u∞ ,u ) + 1 R1 R2 + 2 2
1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R0 R1 + ∆0 R0 .
2 2
3
|u| 2 |u| 2
90
Employing L∞ Sobolev Embedding (Proposition 2.3.10) leads to

kχ̂kL∞
sc (Su,u )
≤ sup k∇χ̂kL4sc (2δ S u,u ) + kχ̂kL4sc (2δ S u,u )
δ S⊂S
1
1 1 1 δ4
≤O1,4 [χ̂] + kαkL4sc (δ S u∞ ,u ) + 1 R1 R2 +
2 2
1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R02 R12 + 3 ∆0 R0 .
|u| 2 |u| 2
The following conclusion in Proposition 2.4.8:
1 1
1 δ41 1 11 δ4
k∇χ̂kL4sc (Su,u ) ≤R1 R2 + δ R1 +
2 2
∆0 R02 R12 +
4 ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 ,
|u| |u| 0
immediately implies
1
1 1 1 δ4
kχ̂kL∞
sc (Su,u )
≤kαkL4sc (δ S u∞ ,u ) + 1 R1 R2 +
2 2
1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R 0 R 1 + ∆0 R 0
2 2
3
|u| 2 |u| 2
1 1
1 1 1 δ4 1 1 δ4
+ R12 R22 + δ 4 R1 + ∆0 R02 R12 + ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 .
|u| |u| 0
For χ̂, we can prove
(2.3.3), we have
1 1 1 1 1 1
kχ̂kL∞
sc (Su,u )
≤ k∇χ̂kL4sc (Su∞ ,u ) + kχ̂kL4sc (δ S u∞ ,u ) + 3 R1 R2
2 2
|u| |u∞ | |u∞ | |u| 2
1 1 1
δ4 δ2 1 1 1 1 1 δ4
+ 3 R1 + 3 ∆0 R0 R1 + 3 R0 R1 + R0
2 2 2 2
3
|u| 2 |u| 2 |u| 2 |u| 2
3 1 1
δ4 δ2 δ2
+ 3 ∆0 R 0 + ∆0 O1,4 + O0,4 ∆20 .
|u| 2 |u| |u|
91
Proof. The null coefficient χ̂ satisfies
∇3 χ̂ + trχ χ̂ = −2ωχ̂ − α.
Using Gronwall’s inequality and evolution lemma (Proposition 2.3.8), we derive
Z u
1 1 1
kχ̂kL4sc (δ S u,u ) ≤ kχ̂kL4sc (δ S u∞ ,u ) + 0 3
kαkL4sc (δ S u,u ) du0
|u| |u∞ | u∞ |u |
Z u
1 1
≤ kχ̂kL4sc (δ S u∞ ,u ) + 0 3
kαkL4sc (Su,u ) du0
|u∞ | u∞ |u |
1
1 1 1 1 δ4
≤ kχ̂kL4sc (δ S u∞ ,u ) + 3 R0 R1 +
2 2
3 R0 .
|u∞ | |u| 2 |u| 2
Employing L∞ Sobolev Embedding (Proposition 2.3.10), we have

1 1 1
kχ̂kL∞
sc (Su,u )
≤ sup k∇χ̂kL4sc (2δ S u,u ) + kχ̂kL4sc (2δ S u,u )
|u| δ S⊂S |u| |u|
1
1 1 1 1 δ4
≤O1,4 [χ̂] + kχ̂kL4sc (δ S u∞ ,u ) + 3 R0 R1 +
2 2
3 R0 .
|u∞ | |u| 2 |u| 2
Hence the following conclusion in Proposition 2.4.10:
1 1
1 1 1 1 δ4 δ2 1 1
k∇χ̂kL4sc (Su,u ) ≤|u∞ |−1 k∇χ̂kL4sc (Su∞ ,u ) + 3 R1 R2 +
2 2
3 R1 + 3 ∆0 R0 R1
2 2
|u| |u| 2 |u| 2 |u| 2

3 1 1
δ4 δ2 δ2
+ 3 ∆0 R0 + ∆0 O1,4 + O0,4 ∆20 ,
|u| 2 |u| |u|
implies
1 1 1 1 1 1
kχ̂kL∞
sc (Su,u )
≤ k∇χ̂kL4sc (Su∞ ,u ) + kχ̂kL4sc (δ S u∞ ,u ) + 3 R1 R2
2 2
|u| |u∞ | |u∞ | |u| 2

1 1 1
δ4 δ2 1 1 1 1 1 δ4
+ 3 R1 + 3 ∆0 R0 R1 + 3 R0 R1 + R0
2 2 2 2
3
|u| 2 |u| 2 |u| 2 |u| 2
3 1 1
δ4 δ2 δ2
+ 3 ∆0 R 0 + ∆0 O1,4 + O0,4 ∆20 .
|u| 2 |u| |u|
92
Gathering all the estimates in this section, we obtain
R < ∞, R < ∞,
then there exist δ0 = δ0 (I (0) , R(0) , R, R), such that whenever δ ≤ δ0 , in D we have

O0,∞ ≤ C I (0) + R + R ,
where C is a universal constant.
Proof. From the estimates above, by taking δ0 sufficient small we have
O0,∞ ≤ I (0) + R + R.
Let ∆3 I (0) + R + R. We have improved bootstrap assumption (2.3.1).
Improved L∞ (S) Estimates for trχ
For trχ, we can prove the following improved estimate, which will be used to prove formation of
trapped surfaces in Section 2.7.
(2.3.3), we have
1 1
δ − 2 ktrχkL∞
sc (Su,u )
≤δ − 2 ktrχkL∞
sc (Su,0 )
+ kαkL4sc (δ S u∞ ,u )
1 1 3 2
1 12 12 δ4 δ2 1 1 δ4 1 1
+ R R + R + ∆0 R0 R1 + 2 ∆0 R0 + R1 R2
2 2 2 2
|u| 1 2 |u| 1 |u|2 |u|

1 1 1 2
1 δ4 1 1 δ4 δ2 δ 2
+ δ 2 R1 + ∆0 R02 R12 + ∆0 R0 + ∆0 O1,4 + ∆0 O0,4 .
|u| |u| |u| |u|
93
Proof. The null coefficient trχ obeys
1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ.
2
Applying Gronwall’s inequality, we obtain
Z u
1 1 1
δ − 2 ktrχkL∞
sc (Su,u )
sc (Su,0 )
+ δ− 2 δ −1 kχ̂ · χ̂kL∞
sc (Su,u0 )
du0
0
1 1 2
sc (Su,0 )
+ O [χ̂].
|u| 0,∞
The following conclusion in Proposition 2.4.16:
1
1 1 1 δ4
kχ̂kL∞
sc (Su,u )
≤kαkL4sc (δ S u∞ ,u ) + 1 R1 R2 +
2 2
1 R1
|u| 2 |u| 2
1 3
δ2 1 1 δ4
+ 3 ∆0 R02 R12 + 3 ∆0 R 0
|u| 2 |u| 2
1 1
1 δ4
1 1
1 1 δ4
+ R1 R2 + δ R1 +
2 2
∆0 R02 R12 +
4 ∆0 R0
|u| |u|
1
δ2 δ 2
+ ∆0 O1,4 + ∆ O0,4 ,
|u| |u| 0
implies
1 1
δ − 2 ktrχkL∞
sc (Su,u )
≤δ − 2 ktrχkL∞ sc (Su,0 )
+ kαk2L4sc (δ S u∞ ,u )
1 1 3 2
1 1 1 δ4 δ2 1 1 δ4 1 1
+ 1 R 2
1 R 2
2 + 1 R 1 + 3 ∆ R
0 0
2
R 2
1 + 3 ∆ R
0 0 + R 2
1 R 2
2
|u| 2 |u| 2 |u| 2 |u| 2
1 1 1 2
1 δ4 1 1 δ4 δ2 δ 2
+ δ 4 R1 + ∆0 R02 R12 + ∆0 R0 + ∆0 O1,4 + ∆0 O0,4 .
|u| |u| |u| |u|
2.5 Elliptic Estimates
Establishing energy estimate requires L4 (Su,u ) estimates for the second angular derivatives of Ricci
coefficients. We obtain the desired L4 (Su,u ) bounds via deriving L2 (Su,u ) estimate for the third
94
angular derivatives of Ricci coefficients and employing Sobolev Embedding.
We now estimate the third angular derivatives of Ricci coefficients. This is achieved by a
combination of transport estimates and elliptic estimates. To start with, we first prove two useful
propositions for curvature components.
2.5.1 L4 (S) Estimates for Curvature Components
For curvature components β, ρ, σ, β, we have the following null Bianchi equations
∇4 β + 2trχβ = div α − 2ωβ + η · α,
3 1
2 2
3 1
∇4 σ + trχσ = −div ∗ β + χ̂ · ∗ α − ζ · ∗ β − 2η · ∗ β,
2 2
∇4 β + trχβ = −∇ρ + ∗ ∇σ + 2ωβ + 2χ̂β − 3(ηρ − ∗ ησ).
With these equations we derive
Proposition 2.5.1. Under the assumption of Theorem 2.2.1, it follows that
1 1 1 1 1 1
kβ, ρ, σ, βkL4sc (Su,u ) ≤ R12 R22 + δ 4 R1 + δ 4 (I (0) + R + R)(R12 R22 + R0 ).
0 0
Proof. For Ψ(s,s ) ∈ {β, ρ, σ, β}, Ψ(s,s ) satisfies the following systematical equation
0 1 0 1
X 0 0
∇4 Ψ(s,s ) = ∇Ψ(s+ 2 ,s − 2 ) + ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) .
s1 +s2 =s+1,
s01 +s02 =s0
95
Employing Proposition 2.3.7 for transport equations, we have
0
kΨ(s,s ) kL4sc (Su,u )
Z u
1 0 1
≤ δ −1 k∇Ψ(s+ 2 ,s − 2 ) kL4sc (Su,u0 ) du0
0
Z u 1
X δ 2 (s1 ,s01 ) 0
+ δ −1 kψ kL∞
sc (Su,u0 )
kΨ(s2 ,s2 ) kL4sc (Su,u0 ) du0
s1 +s2 =s+1, 0 |u|
s01 +s02 =s0
Z u 1 1
1 0 1 1 0 1
≤ δ −1 k∇Ψ(s+ 2 ,s − 2 ) kL2 2 k∇ 2
Ψ(s+ 2 ,s − 2 ) kL2 2 du0
sc (Su,u0 ) sc (Su,u0 )
0
Z u
1 1 0 1
+ δ −1 δ 4 k∇Ψ(s+ 2 ,s − 2 ) kL2sc (Su,u0 ) du0
0
Z u 1
X δ 2 (s1 ,s01 ) 0 1 0 1
+ δ −1 kψ kL∞
sc (Su,u0 )
kΨ(s2 ,s2 ) kL2 2 (S 0 ) k∇Ψ(s2 ,s2 ) kL2 2 (S 0 ) du0
s1 +s2 =s+1, 0 |u| sc u,u sc u,u
s01 +s02 =s0
Z u 1
−1 δ
X 2 0 1 0
+ δ kψ (s1 ,s1 ) kL∞
sc (Su,u0 )
δ 4 kΨ(s2 ,s2 ) kL2sc (Su,u0 ) du0
s1 +s2 =s+1, 0 |u|
s01 +s02 =s0
1 0 1 1 1 0 1 1 1 1 0 1
≤k∇Ψ(s+ 2 ,s − 2 ) k 2 2 (0,u) k∇2 Ψ(s+ 2 ,s − 2 ) k 2 2 (0,u) + δ 4 k∇Ψ(s+ 2 ,s − 2 ) kL2 (0,u)
X 1 1 0 1 0 1 0 1
+ δ 4 sup kψ (s1 ,s1 ) kL∞
sc (Su,u )
δ 4 kΨ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s2 ,s2 ) k 2 2 (0,u)
s1 +s2 =s+1,
u,u |u| Lsc (Hu ) Lsc (Hu )
s01 +s02 =s0
X 1 1 0 1 0
+ δ 4 sup kψ (s1 ,s1 ) kL∞
sc (Su,u )
δ 2 kΨ(s2 ,s2 ) kL2 (H (0,u) ) ,
s1 +s2 =s+1,
u,u |u| sc u
s01 +s02 =s0
where we employ L4 Sobolev Embedding (Proposition 2.3.9) and Hölder’s inequality. The anoma-
lous term we encounter are among {η · α, χ̂ · α, χ̂ · ∗ α, χ̂ · β}. We control them through the
estimates above by adding suitable δ or |u| weights in front of anomalous terms. Hence we obtain
1 1 1 1 1 1 1
kβ, ρ, σ, βkL4sc (Su,u ) ≤ R12 R22 + δ 4 R1 + δ 4 O0,∞ R02 R12 + δ 4 O0,∞ R0 .
Combining the result in Proposition 2.4.18, we derive
1 1 1 1 1 1
kβ, ρ, σ, βkL4sc (Su,u ) ≤ R12 R22 + δ 4 R1 + δ 4 C(I (0) + R + R)(R02 R12 + R0 ).
96
1
For K − |u|2 , we have
1
kK − kL4 (S )
|u|2 sc u,u
1 3 1
1 1 1 δ2 1 1 δ4 δ2
≤R12 R22 + δ 4 R1 + O0,∞ R02 R12 + O0,∞ R0 + O0,∞ O1,4
|u| |u| |u|
1
δ 3 δ4 1
(0) 3
+ 2 O0,∞ + 1 O0,∞ 1 + R + δ (I + R + R) .
4
|u| |u| 2
Proof. The curvature component K obeys
1 1
∇4 K + trχK = − div β − ζ · β − 2η · β + χ̂ · ∇⊗η
b + χ̂ · (η ⊗η)
b
2 2
1 1
− trχdiv η − trχ|η|2 .
2 2
Employing Proposition 2.3.7 for transport equations and Gronwall’s inequality, we derive
kKkL4sc (Su,u )
1 1 1
≤kKkL4sc (Su,0 ) + k∇βk 2 2 (0,u) k∇2 βk 2 2 (0,u) + δ 4 k∇βkL2 (0,u)
1
δ2 1 1
+ sup kη, ηkL∞
sc (Su,u )
kβk 2 2 (0,u) k∇βk 2 2 (0,u)
|u| u,u Lsc (Hu ) Lsc (Hu )
3 1
δ4 δ2
+ sup kη, ηkL∞
sc (Su,u )
kβkL2 (H (0,u) ) + sup kχkL∞
sc (Su,u )
k∇ηkL4sc (Su,u )
|u| u,u sc u |u| u,u
δ
+ sup kχkL∞
sc (Su,u )
kηk2L∞
sc (Su,u )
.
|u|2 u,u
1
For kKkL4sc (Su,0 ) , since K = |u|2 on Su,0 , we have
1
kKkL4sc (Su,0 ) = δ 4 |u|.
Therefore, combining Propositions 2.4.18,2.4.6 and 2.4.14 we get
1 1 1
kKkL4sc (Su,u ) ≤δ 4 + R + δ 4 (I (0) + R + R)3 .
|u|
97
1 1
This bound implies an improved estimate for K − |u|2 . For K − |u|2 , we have
1 1 1
∇4 (K − 2
) = − div β − ζ · β − 2η · β + χ̂ · ∇⊗η
b + χ̂ · (η ⊗η)
b
|u| 2 2
1 1
− trχdiv η − trχ|η|2 − trχK.
2 2
Applying Proposition 2.3.7 for transport equations and the estimate above, we derive
1
kK − kL4 (S )
|u|2 sc u,u
1 1 1
≤k∇βk 2 2 (0,u) k∇2 βk 2 2 (0,u) + δ 4 k∇βkL2 (0,u)
1
δ2 1 1
+ sup kη, ηkL∞
sc (Su,u )
kβk 2 2 (0,u) k∇βk 2 2 (0,u)
|u| u,u Lsc (Hu ) Lsc (Hu )
3 1
δ4 δ2
+ sup kη, ηkL∞
sc (Su,u )
kβkL2 (H (0,u) ) + sup kχkL∞
sc (Su,u )
k∇ηkL4sc (Su,u )
|u| u,u sc u |u| u,u
δ 1 1
+ sup kχkL∞
sc (Su,u )
kηk2L∞
sc (Su,u )
+ δ 2 sup ktrχkL∞
sc (Su,u )
kKkL4sc (Su,u ) .
|u|2 u,u u,u |u|
Putting this and the estimate for kKkL4sc (Su,u ) together, we deduce
1
kK − kL4 (S )
|u|2 sc u,u
1 3 1
1 1 δ2 1 1 1 δ4 δ2
≤R12 R22 + δ 4 R1 + O0,∞ R02 R12 + O0,∞ R0 + O0,∞ O1,4
|u| |u| |u|

δ 3 1 1 1
(0) 3
+ 2 O0,∞ + δ O0,∞ δ + R + δ (I + R + R) .
2 4 4
|u|
2.5.2 L2 (S) Estimates for First Derivatives of Curvature Components
For ∇β, ∇ρ, ∇σ, ∇β, the following estimate holds
98
Proposition 2.5.3. Under the assumption of Theorem 2.2.1, we have
1 1 1
k∇β, ∇ρ, ∇σ, ∇βkL2sc (Su,u ) ≤R2 + δ 4 (I (0) + R + R)(R02 R12 + R0 )
1
1
(0) δ 2 (0)
+ δ (I 2 + R + R)R1 + (I + R + R)2 R0
|u|
1 1 1
+ (R02 R12 + δ 4 R0 )2 .
0 0
Proof. For Ψ(s,s ) ∈ {β, ρ, σ, β}, Ψ(s,s ) satisfies the systematical equation:
0
∇4 ∇Ψ(s,s )
1 0 1
X 0 0 X 0 0
=∇2 Ψ(s+ 2 ,s − 2 ) + ∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) + ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 )
s1 +s2 =s+1, s1 +s2 =s+1,
s01 +s02 =s0 s01 +s02 =s0
X 0 0 0 X 0 0
+ ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) + Ψ(s6 ,s6 ) · ∇Ψ(s7 ,s7 ) .
s3 +s4 +s5 =s+ 3 , s6 +s7 =s+1,
2
s06 +s07 =s0
s03 +s04 +s05 =s0 + 1
2
Employing Proposition 2.3.7 for transport equations and L4 Sobolev Embedding (Proposition
2.3.9), we infer
0
k∇Ψ(s,s ) kL2sc (Su,u )
Z u
1 0 1
≤ δ −1 k∇2 Ψ(s+ 2 ,s − 2 ) kL2sc (Su,u0 ) du0
0
Z u 1
X δ2 0 0
+ δ −1 k∇ψ (s1 ,s1 ) kL4sc (Su,u0 ) kΨ(s2 ,s2 ) kL4sc (Su,u0 ) du0
s1 +s2 =s+1, 0 |u|
s01 +s02 =s0
Z u 1
X δ 2 (s1 ,s01 ) 0
+ δ −1 kψ kL∞
sc (Su,u0 )
k∇Ψ(s2 ,s2 ) kL2sc (Su,u0 ) du0
s1 +s2 =s+1, 0 |u|
s01 +s02 =s0
Z u
X δ 0 0 0
+ δ −1 kψ (s3 ,s3 ) kL∞
sc (Su,u0 )
kψ (s4 ,s4 ) kL∞
sc (Su,u0 )
kΨ(s5 ,s5 ) kL2sc (Su,u0 ) du0
0 |u|2
s3 +s4 +s5 =s+ 3 ,
2
s03 +s04 +s05 =s0 + 1
2
Z u
X 0 0
+ δ −1 kΨ(s6 ,s6 ) kL4sc (Su,u0 ) k∇Ψ(s7 ,s7 ) kL4sc (Su,u0 ) du0
s6 +s7 =s+1, 0
s06 +s07 =s0
99
1 0 1
≤k∇2 Ψ(s+ 2 ,s − 2 ) kL2 (H (0,u) )
sc u
1
X 1 0 1 0 1 0 1
+ δ4 sup k∇ψ (s1 ,s1 ) kL4sc (Su,u ) δ 4 kΨ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s2 ,s2 ) k 2 2 (0,u)
s +s =s+1,
u,u |u| Lsc (Hu ) Lsc (Hu )
1 2
s01 +s02 =s0
1
X 1 0 1 0
+ δ4 sup k∇ψ (s1 ,s1 ) kL4sc (Su,u ) δ 2 kΨ(s2 ,s2 ) kL2 (H (0,u) )
s1 +s2 =s+1,
u,u |u| sc u
s01 +s02 =s0
1
X 1 0 0
+ δ2 sup kψ (s1 ,s1 ) kL∞
sc (Su,u )
k∇Ψ(s2 ,s2 ) kL2 (H (0,u) )
s1 +s2 =s+1,
u,u |u| sc u
s01 +s02 =s0
1
δ2 X 1 0 0 1 0
sc (Su,u )
kψ (s4 ,s4 ) kL∞
sc (Su,u )
δ 2 kΨ(s5 ,s5 ) kL2 (H (0,u) )
|u| u,u |u| sc u
s3 +s4 +s5 =s+ 3 ,
2
s03 +s04 +s05 =s0 + 1
2

1 1
(s6 ,s06 ) (s6 ,s06 ) 1
(s6 ,s06 )
X
+ kΨ k 2
(0,u) k∇Ψ k 2
(0,u) + δ kΨ
4 kL2 (0,u)
L2sc (Hu ) L2sc (Hu ) sc (Hu )
s6 +s7 =s+1,
s06 +s07 =s0

0 1 1
(s7 ,s07 ) 1
(s7 ,s07 )
× k∇Ψ(s7 ,s7 ) k 2 2 (0,u)
2
k∇ Ψ k 2
(0,u) + δ k∇Ψ
4 kL2 (0,u) .
Lsc (Hu ) L2sc (Hu ) sc (Hu )
The anomalous term we have are among
{∇ηα, ∇χ̂α, ∇χ̂β, χ̂∇α, χ̂∇β, (η, η)ηα, (η, η)χ̂α, (η, η)χ̂β}.
All of them are controlled through the estimates above. Hence we obtain
1
1 1 1 1 δ2 2
k∇β, ∇ρ, ∇σ, ∇βkL2sc (Su,u ) ≤R2 + δ 4 O1,4 (R02 R12 + R0 ) + δ 2 O0,∞ R1 + O R0
|u| 0,∞
1 1 1
+ (R02 R12 + δ 4 R0 )2 .
100
Combining the result in Propositions 2.4.18 and 2.4.14, we demonstrate
1 1 1
k∇β, ∇ρ, ∇σ, ∇βkL2sc (Su,u ) ≤R2 + δ 4 (I (0) + R + R)(R02 R12 + R0 )
1
1
(0) δ 2 (0)
+ δ (I
2 + R + R)R1 + (I + R + R)2 R0
|u|
1 1 1
+ (R02 R12 + δ 4 R0 )2 .
The curvature component K satisfies
3
1
(0)
k∇KkL2sc (Su,u ) ≤R + δ I + R + R .4
Proof. For K, we have

1 1
K = −ρ + χ̂ · χ̂ − trχtrχ.
2 4
1 1 1 2 1 2
K− = −ρ + χ̂ · χ̂ − trχ(trχ + )+ (trχ − ).
|u|2 2 4 |u| 2|u| |u|
Hence, we get
k∇KkL2sc (Su,u )
1
=k∇(K − )kL2sc (Su,u )
|u|2
≤k∇ρkL2sc (Su,u ) + kχ̂∇χ̂kL2sc (Su,u ) + k∇χ̂χ̂kL2sc (Su,u )

2
+ k∇trχ(trχ + )kL2sc (Su,u ) + ktrχ∇trχkL2sc (Su,u )
|u|
1
+ k∇trχkL2sc (Su,u )
|u|
101
1
1 1 1 1 δ4
≤k∇ρkL2sc (Su,u ) + δ 4 δ 4 kχ̂kL4sc (Su,u ) k∇χ̂kL4sc (Su,u ) + δ 4 k∇χ̂kL4sc (Su,u ) kχ̂kL4sc (Su,u )
|u| |u|
1
1 1 2 δ2
+ δ k∇trχkL2sc (Su,u ) ktrχ +
2 kL∞ (S ) + ktrχkL∞
sc (Su,u )
|u| |u| sc u,u |u|
1
+ k∇trχkL2sc (Su,u ) .
|u|
Let φ be Su,u -tangent tensor field. With the definition of scale invariant norms and Hölder’s
inequality, we have
1
kφkL2sc (Su,u ) ≤ δ 4 kφkL4sc (Su,u ) .
Therefore, we deduce
k∇KkL2sc (Su,u )
1
1 1 1 1 δ4
≤k∇ρkL2sc (Su,u ) + δ δ kχ̂kL4sc (Su,u ) k∇χ̂kL4sc (Su,u ) + δ 4 k∇χ̂kL4sc (Su,u ) kχ̂kL4sc (Su,u )
4 4
|u| |u|
3
3 1 2 δ4
+ δ 4 k∇trχkL4sc (Su,u ) ktrχ + kL∞
sc (Su,u )
+ ktrχkL∞
sc (Su,u )
|u| |u| |u|
1
δ4
+ k∇trχkL4sc (Su,u ) .
|u|
From Propositions 2.5.3, 2.4.6, 2.4.14, 2.4.18 and 2.4.13, we have
3
1
(0)
k∇KkL2sc (Su,u ) ≤R + δ I + R + R .
4
2.5.3 L2 (S) Estimates for Third Derivatives of Ricci Coefficients
Deriving estimates for the third angular derivatives of Ricci coefficients by transport equations may
require third derivatives of curvature components. It will cause losing of derivatives. To cure this
102
0
trouble, we introduce new renormalized quantities Θ(s,s ) with signature (s, s0 ). We have
0
Θ(s,s ) ∈ {∇trχ, ∇trχ, µ, µ, κ, κ},
where
µ := −div η − ρ,
µ := −div η − ρ.
To define κ, we set ω † to be the solution to
1
∇3 ω † = σ,
2
with zero initial data on Hu∞ . And
1
κ := ∇ω + ∗ ∇ω † − β.
2
To define κ, we set ω † to be the solution to
1
∇4 ω † = σ,
2
with zero initial data on H 0 . And
1
κ := −∇ω + ∗ ∇ω † − β.
2
Using null structure equations, null Bianchi equations and commutation formulas (Proposi-
0
tions 2.3.12-2.3.15), it can be demonstrated that Θ(s,s ) obeys the following systematical transport
103
equations:
0
∇4 Θ(s,s )
0 X 0 0
=χ(1,0) · Θ(s,s ) + ∇ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
X 0 0 X 0 0 0
+ ψ (s3 ,s3 ) · Ψ(s4 ,s4 ) + ψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 ) ,
s3 +s4 =s+1, s5 +s6 +s7 =s+1,
s03 +s04 =s0 s05 +s06 +s07 =s0
or
0 0 0
∇3 Θ(s,s ) + λ[Θ(s,s ) ]trχΘ(s,s )
0 X 0 0
=χ̂(0,1) · Θ(s,s ) + ∇ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
X 0 0 X 0 0 0
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
0 0
In this section, ψ (s,s ) and Ψ(s,s ) represent an arbitrary Ricci coefficient with signature (s, s0 ) and
0
a null curvature component with signature (s, s0 ), respectively. λ[ψ (s,s ) ] is a constant depending
0
on ψ (s,s ) .
104
0
Similarly, we can check Θ(s,s ) satisfies the following systematical transport equations:
0
∇4 ∇Θ(s,s )
0 X 0 0
=χ(1,0) · ∇Θ(s,s ) + ∇2 ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
X 0 0 X 0 0
+ ∇ψ (s1 ,s1 ) · ∇ψ (s2 ,s2 ) + ∇ψ (s3 ,s3 ) · Ψ(s4 ,s4 )
s1 +s2 =s+ 1 , s3 +s4 =s+1,
2
s03 +s04 =s0
s01 +s02 =s0 − 1
2
X 0 0 X 0 0 0
+ ψ (s3 ,s3 ) · ∇Ψ(s4 ,s4 ) + ∇ψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 )
s3 +s4 =s+1, s5 +s6 +s7 =s+1,
s03 +s04 =s0 s05 +s06 +s07 =s0
X 0 0 0 X 0 0
+ ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · Ψ(s10 ,s10 ) + Ψ(s11 ,s11 ) · Ψ(s12 ,s12 )
s8 +s9 +s10 =s+ 3 , s11 +s12 =s+ 3 ,
2 2
s08 +s09 +s010 =s0 + 1 s011 +s012 =s0 + 1
2 2
X 0 0 0 0
+ ψ (s13 ,s13 ) · ψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 ) ,
s13 +s14 +s15 +s16 =s+ 3 ,
2
s013 +s014 +s015 +s016 =s0 + 1
2
or

0 0 1 0
∇3 ∇Θ(s,s ) + λ[Θ(s,s ) ] + trχΘ(s,s )
2
0 X 0 0
=χ̂(0,1) · ∇Θ(s,s ) + ∇2 ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
X 0 0 X 0 0
+ ∇ψ (s1 ,s1 ) · ∇ψ (s2 ,s2 ) + ∇ψ (s3 ,s3 ) · Ψ(s4 ,s4 )
s1 +s2 =s− 1 , s3 +s4 =s,
2
s03 +s04 =s0 +1
s01 +s02 =s0 + 1
2
X 0 0 X 0 0 0
+ ψ (s3 ,s3 ) · ∇Ψ(s4 ,s4 ) + ∇ψ (s4 ,s4 ) · ψ (s5 ,s5 ) · ψ (s6 ,s6 )
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
X 0 0 0 X 0 0
+ ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · Ψ(s10 ,s10 ) + Ψ(s11 ,s11 ) · Ψ(s12 ,s12 )
s8 +s9 +s10 =s+ 1 , s11 +s12 =s+ 1 ,
2 2
s08 +s09 +s010 =s0 + 3 s011 +s012 =s0 + 3
2 2
X 0 0 0 0
+ ψ (s13 ,s13 ) · ψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 ) .
s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
105
0
And Θ(s,s ) obeys
0
∇4 ∇2 Θ(s,s )
0 X 0 0
=χ(1,0) · ∇2 Θ(s,s ) + ∇3 ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
X 0 0 X 0 0
+ ∇2 ψ (s1 ,s1 ) · ∇ψ (s2 ,s2 ) + ∇2 ψ (s3 ,s3 ) · Ψ(s4 ,s4 )
s1 +s2 =s+ 1 , s3 +s4 =s+1,
2
s03 +s04 =s0
s01 +s02 =s0 − 1
2
X 0 0 X 0 0 0
+ ∇ψ (s3 ,s3 ) · ∇Ψ(s4 ,s4 ) + ∇2 ψ (s5 ,s5 ) · ψ (s6 ,s6 ) · ψ (s7 ,s7 )
s3 +s4 =s+1, s5 +s6 +s7 =s+1,
s03 +s04 =s0 s05 +s06 +s07 =s0
0 0 0 0 0
(2.5.1)
X X
+ ψ (s3 ,s3 ) · ∇2 Ψ(s4 ,s4 ) + ∇ψ (s5 ,s5 ) · ∇ψ (s6 ,s6 ) · ψ (s7 ,s7 )
s3 +s4 =s+1, s5 +s6 +s7 =s+1,
s03 +s04 =s0 s05 +s06 +s07 =s0
X 0 0 0
+ ∇ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · Ψ(s10 ,s10 )
s8 +s9 +s10 =s+ 3 ,
2
s08 +s09 +s010 =s0 + 1
2
X 0 0 0 X 0 0
+ ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · ∇Ψ(s10 ,s10 ) + ∇Ψ(s11 ,s11 ) · Ψ(s12 ,s12 )
s8 +s9 +s10 =s+ 3 , s11 +s12 =s+ 3 ,
2 2
s08 +s09 +s010 =s0 + 1 s011 +s012 =s0 + 1
2 2
X 0 0 0 0
+ ∇ψ (s13 ,s13 ) · ψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 )
s13 +s14 +s15 +s16 =s+ 3 ,
2
s013 +s014 +s015 +s016 =s0 + 1
2
X 0 0 0 0
+ ψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) · Ψ(s20 ,s20 )
s17 +s81 +s19 +s20 =s+2,
s017 +s018 +s019 +s020 =s0 +1
X 0 0 0
+ ψ (s21 ,s21 ) · Ψ(s22 ,s22 ) · Ψ(s23 ,s23 )
s21 +s22 +s23 =s+2,
s021 +s022 +s023 =s0 +1
X 0 0 0 0 0
+ ψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) · ψ (s28 ,s28 ) ,
s24 +s25 +s26 +s27 +s28 =s+2,
s024 +s025 +s026 +s027 +s028 =s0 +1
or
106

2 (s,s0 ) (s,s0 ) 0
∇3 ∇ Θ + λ[Θ ] + 1 trχ∇2 Θ(s,s )
0 X 0 0
=χ̂(0,1) · ∇2 Θ(s,s ) + ∇3 ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
X 0 0 X 0 0
+ ∇2 ψ (s1 ,s1 ) · ∇ψ (s2 ,s2 ) + ∇2 ψ (s3 ,s3 ) · Ψ(s4 ,s4 )
s1 +s2 =s− 1 , s3 +s4 =s,
2
s03 +s04 =s0 +1
s01 +s02 =s0 + 1
2
X 0 0 X 0 0 0
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
X 0 0 X 0 0 0
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
X 0 0 0
+ ∇ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · Ψ(s10 ,s10 )
s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2
X 0 0 0 X 0 0
s8 +s9 +s10 =s+ 1 , s11 +s12 =s+ 1 ,
2 2
s08 +s09 +s010 =s0 + 3 s011 +s012 =s0 + 3
2 2
X 0 0 0 0
+ ∇ψ (s13 ,s13 ) · ψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 )
s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
X 0 0 0 0
+ ψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) · Ψ(s20 ,s20 )
s17 +s81 +s19 +s20 =s+1,
s017 +s018 +s019 +s020 =s0 +2
X 0 0 0
+ ψ (s21 ,s21 ) · Ψ(s22 ,s22 ) · Ψ(s23 ,s23 )
s21 +s22 +s23 =s+1,
s021 +s022 +s023 =s0 +2
X 0 0 0 0 0
+ ψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) · ψ (s28 ,s28 ) .
s24 +s25 +s26 +s27 +s28 =s+1,
s024 +s025 +s026 +s027 +s028 =s0 +2
0
Now we are ready to estimate the second derivative of Θ(s,s ) .
0
Proposition 2.5.5. For Θ(s,s ) satisfying (2.5.1), we have
107
u 1
δ− 2
Z
0 X 0 0
k∇2 Θ(s,s ) kL2sc (Su,u ) ≤ k∇3 ψ (s1 ,s1 ) kL2sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
du0
0 |u|
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
1 1 3
δ2 δ2 1 1 δ4
+ O2,4 O1,4 + O2,4 R02 R12 + O2,4 R0
|u| |u| |u|
1 3
δ2 1 1 δ4 δ
+ O1,4 R12 R22 + O1,4 R1 + ∆0 O2,4 O0,4
|u| |u| |u|
5
δ δ 1 1 δ4
+ ∆0 O1,4 O1,4 + ∆0 O1,4 R12 R22 + ∆0 O1,4 R1
|u| |u| |u|
1 3 3
δ 2 δ2 1 1 δ 4 12 12 δ 4 32 12 δ
+ ∆0 R 1 + R1 R22 R02 + R1 R2 R0 + R R + R1 R0
|u| |u| |u| |u| 1 0 |u|
3 3
δ2 2 δ2 δ
+ 3
∆0 O1,4 O0,4 + 3 ∆30 R0 + 2 ∆0 R0 R1
|u| |u| |u|
5 3
δ4 3 1 δ2 δ2
+ 2
∆0 R02 R12 + 2 ∆0 R20 + 4 ∆40 R20 .
|u| |u| |u|
Proof. Employing Proposition 2.3.7 for transport equations, it follows that
0
k∇2 Θ(s,s ) kL2sc (Su,u )
≤E1 + E2 + E3 + E4 + E5 + E6 + E7 + E8 + E9 + E10
+ +E11 + E12 + E13 + E14
where E1 − E14 are given as follows:
u 1
δ− 2
Z X 0 0
E1 = k∇3 ψ (s1 ,s1 ) kL2sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
du0 ,
0 |u|
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
u 1
δ− 2
Z X 0 0
E2 = k∇2 ψ (s1 ,s1 ) kL4sc (Su,u0 ) k∇ψ (s2 ,s2 ) kL4sc (Su,u0 ) du0 ,
0 |u|
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
108
u 1
δ− 2
Z X 0 0
E3 = k∇2 ψ (s3 ,s3 ) kL4sc (Su,u0 ) kΨ(s4 ,s4 ) kL4sc (Su,u0 ) du0 ,
0 |u| s3 +s4 =s+1,
s03 +s04 =s0
u 1
δ− 2
Z X 0 0
E4 = k∇ψ (s3 ,s3 ) kL4sc (Su,u0 ) k∇Ψ(s4 ,s4 ) kL4sc (Su,u0 ) du0 ,
0 |u| s3 +s4 =s+1,
s03 +s04 =s0
u 1
δ− 2
Z X 0 0 0
E5 = k∇2 ψ (s5 ,s5 ) kL4sc (Su,u0 ) kψ (s6 ,s6 ) · ψ (s7 ,s7 ) kL4sc (Su,u0 ) du0 ,
0 |u| s5 +s6 +s7 =s+1,
s05 +s06 +s07 =s0
Z u
1 X 0 0 0
E6 = k∇ψ (s5 ,s5 ) kL4sc (Su,u0 ) k∇ψ (s6 ,s6 ) kL4sc (Su,u0 ) kψ (s7 ,s7 ) kL∞
sc (Su,u0 )
du0 ,
0 |u|2 s5 +s6 +s7 =s+1,
s05 +s06 +s07 =s0
Z u
1 X 0 0 0
E7 = k∇ψ (s8 ,s8 ) kL4sc (Su,u0 ) kψ (s9 ,s9 ) kL∞
sc (Su,u0 )
kΨ(s10 ,s10 ) kL4sc (Su,u0 ) du0 ,
0 |u|2
s8 +s9 +s10 =s+ 3 ,
2
s08 +s09 +s010 =s0 + 1
2
u 1
δ− 2
Z X 0 0 0
E8 = kψ (s8 ,s8 ) · ψ (s9 ,s9 ) kL4sc (Su,u0 ) k∇Ψ(s10 ,s10 ) kL4sc (Su,u0 ) du0 ,
0 |u|
s8 +s9 +s10 =s+ 3 ,
2
s08 +s09 +s010 =s0 + 1
2
u 1
δ− 2
Z X 0 0
E9 = k∇Ψ(s11 ,s11 ) kL4sc (Su,u0 ) kΨ(s12 ,s12 ) kL4sc (Su,u0 ) du0 ,
0 |u|
s11 +s12 =s+ 3 ,
2
s011 +s012 =s0 + 1
2
u 1
δ− 2
Z X 0
E10 = k∇ψ (s13 ,s13 ) kL4sc (Su,u0 )
0 |u|
s13 +s14 +s15 +s16 =s+ 3 ,
2
s013 +s014 +s015 +s016 =s0 + 1
2
0 0 0
× kψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 ) kL4sc (Su,u0 ) du0 ,
u 1
δ− 2
Z X 0 0 0
E11 = kψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) kL4sc (Su,u0 )
0 |u| s17 +s81 +s19 +s20 =s+2,
s017 +s018 +s019 +s020 =s0 +1
0
× kΨ(s20 ,s20 ) kL4sc (Su,u0 ) du0 ,
Z u
1 X 0 0
E12 = 2
kψ (s21 ,s21 ) kL∞
sc (Su,u0 )
kΨ(s22 ,s22 ) kL4sc (Su,u0 )
0 |u| s +s +s =s+2,
21 22 23
s021 +s022 +s023 =s0 +1
0
× kΨ(s23 ,s23 ) kL4sc (Su,u0 ) du0 ,
u 1
δ− 2
Z X 0 0 0 0
E13 = kψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) kL4sc (Su,u0 )
0 |u| s24 +s25 +s26 +s27 +s28 =s+2,
s024 +s025 +s026 +s027 +s028 =s0 +1
0
× kψ (s28 ,s28 ) kL4sc (Su,u0 ) du0 ,
109
u 1
δ− 2
Z X 0 0
E14 = kψ (s3 ,s3 ) kL∞
sc (Su,u0 )
k∇2 Ψ(s4 ,s4 ) kL2sc (Su,u0 ) du0 .
0 |u| s3 +s4 =s+1,
s03 +s04 =s0
Using L4 Sobolev Embedding (Proposition 2.3.9) and Hölder’s inequality, we demonstrate the
estimates below:
1
X 1 0 0
E2 ≤ δ 2 sup k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇ψ (s2 ,s2 ) kL4sc (Su0 ,u0 ) ,
u0 ,u0 |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
1
X 1 0 1 0 1 0 1
E3 ≤δ 4 sup 0
k∇2 ψ (s3 ,s3 ) kL4sc (Su0 ,u0 ) δ 4 kΨ(s4 ,s4 ) k 2 2 (0,u) k∇Ψ(s4 ,s4 ) k 2 2 (0,u)
s3 +s4 =s+1, u0 ,u0 |u | L sc (Hu ) L sc (Hu )
s03 +s04 =s0
1
X 1 0 1 0
+ δ4 sup k∇2 ψ (s3 ,s3 ) kL4sc (Su0 ,u0 ) δ 2 kΨ(s4 ,s4 ) kL2 (H (0,u) ) ,
s3 +s4 =s+1, u0 ,u0 |u0 | sc u
s03 +s04 =s0
1
X 1 0 0 1 0 1
E4 ≤δ 2 sup 0
k∇ψ (s3 ,s3 ) kL4sc (Su0 ,u0 ) k∇Ψ(s4 ,s4 ) k 2 2 (0,u) k∇2 Ψ(s4 ,s4 ) k 2 2 (0,u)
s3 +s4 =s+1, u0 ,u0 |u | Lsc (Hu ) Lsc (Hu )
s03 +s04 =s0
3
δ4 X 0 0
+ sup k∇ψ (s3 ,s3 ) kL4sc (Su0 ,u0 ) k∇Ψ(s4 ,s4 ) kL2 (0,u) ,
|u| u0 ,u0 sc (Hu )
s3 +s4 =s+1,
s03 +s04 =s0
δ X 0 0 0
E5 ≤ sup k∇2 ψ (s5 ,s5 ) kL4sc (Su0 ,u0 ) kψ (s6 ,s6 ) kL4sc (Su0 ,u0 ) kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
,
|u|2 s5 +s6 +s7 =s+1, u0 ,u0
s05 +s06 +s07 =s0
δ X 0 0 0
E6 ≤ sup k∇ψ (s5 ,s5 ) kL4sc (Su0 ,u0 ) k∇ψ (s6 ,s6 ) kL4sc (Su0 ,u0 ) kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
,
|u|2 s5 +s6 +s7 =s+1, u0 ,u0
s05 +s06 +s07 =s0
110
3
δ4 X 1 0 0
E7 ≤ sup 0
k∇ψ (s8 ,s8 ) kL4sc (Su0 ,u0 ) kψ (s9 ,s9 ) kL∞
sc (Su0 ,u0 )
|u| u0 ,u0 |u |
s8 +s9 +s10 =s+ 3 ,
2
s08 +s09 +s010 =s0 + 1
2

0 1 1
1
(s10 ,s010 ) 1
(s10 ,s010 )
× δ 4 kΨ(s10 ,s10 ) k 2 2 (0,u) k∇Ψ k 2
(0,u) + δ kΨ 2 kL2 (0,u) ,
Lsc (Hu ) L2sc (Hu ) sc (Hu )
δ X 0 1 0
E8 ≤ sup kψ (s8 ,s8 ) kL4sc (Su0 ,u0 ) kψ (s9 ,s9 ) kL∞
sc (Su0 ,u0 )
|u| u0 ,u0 |u0 |
s8 +s9 +s10 =s+ 3 ,
2
s08 +s09 +s010 =s0 + 1
2

0 1 0 1 1 0
× k∇Ψ(s10 ,s10 ) k 2 2 (0,u) k∇2 Ψ(s10 ,s10 ) k 2 2 (0,u) + δ 4 k∇Ψ(s10 ,s10 ) kL2 (0,u) ,
1
δ2 X 0 1 0 1
E9 ≤ k∇Ψ(s11 ,s11 ) k 2 2 (0,u) k∇2 Ψ(s11 ,s11 ) k 2 2 (0,u)
|u| Lsc (Hu ) Lsc (Hu )
s11 +s12 =s+ 3 ,
2
s011 +s012 =s0 + 1
2

1 0
+ δ 4 k∇Ψ(s12 ,s12 ) kL2 (0,u)
sc (Hu )

0 1 0 1 1 0
× kΨ(s12 ,s12 ) k 2 2 (0,u) k∇Ψ(s12 ,s12 ) k 2 2 (0,u) + δ 4 kΨ(s12 ,s12 ) kL2 (0,u) ,
3
δ2 X 0 0
E10 ≤ 3 k∇ψ (s13 ,s13 ) kL4sc (Su0 ,u0 ) kψ (s14 ,s14 ) kL4sc (Su0 ,u0 )
|u|
s13 +s14 +s15 +s16 =s+ 3 ,
2
s013 +s014 +s015 +s016 =s0 + 1
2
0 0
× kψ (s15 ,s15 ) kL∞
sc (Su0 ,u0 )
kψ (s16 ,s16 ) kL∞
sc (Su,u0 )
,
3
δ2 X 0 0
E11 ≤ 3 sup kψ s17 ,s17 kL∞
sc (Su0 ,u0 )
kψ (s18 ,s18 ) kL∞
sc (Su0 ,u0 )
|u| s17 +s81 +s19 +s20 =s+2, u0 u0
s017 +s018 +s019 +s020 =s0 +1
0
× kψ (s19 ,s19 ) kL4sc (Su0 ,u0 )

0 1 0 1 1
(s20 ,s020 )
× kΨ(s20 ,s20 ) k 2 2 (0,u) k∇Ψ(s20 ,s20 ) k 2 2 (0,u) + δ kΨ4 kL2 (0,u) ,
111
δ X 0
E12 ≤ kψ (s21 ,s21 ) kL∞
sc (Su,u0 )
|u|2 s21 +s22 +s23 =s+2,
s021 +s022 +s023 =s0 +1

0 1 0 1 1 0
× kΨ(s22 ,s22 ) k 2 2 (0,u) k∇Ψ(s22 ,s22 ) k 2 2 (0,u) + δ 4 kΨ(s22 ,s22 ) kL2 (H (0,u) )
Lsc (Hu ) Lsc (Hu ) sc u

0 1 0 1 1 0
× kΨ(s23 ,s23 ) k 2 2 (0,u) k∇Ψ(s23 ,s23 ) k 2 2 (0,u) + δ 4 kΨ(s23 ,s23 ) kL2 (H (0,u) ) ,
Lsc (Hu ) Lsc (Hu ) sc u
δ2 X 0 0
E13 ≤ kψ (s24 ,s24 ) kL∞
sc (Su0 ,u0 )
kψ (s25 ,s25 ) kL∞
sc (Su0 ,u0 )
|u|4 s24 +s25 +s26 +s27 +s28 =s+2,
s024 +s025 +s026 +s027 +s028 =s0 +1
0 0 0
× kψ (s26 ,s26 ) kL∞
sc (Su0 ,u0 )
kψ (s27 ,s27 ) kL4sc (Su0 ,u0 ) kψ (s28 ,s28 ) kL4sc (Su0 ,u0 ) du0 ,
1
X 1 0 0
E14 ≤ δ 2 sup 0
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
k∇2 Ψ(s4 ,s4 ) kL2 (H (0,u) ) .
s3 +s4 =s+1, u0 ,u0 |u | sc u
s03 +s04 =s0
When Θ ∈ {∇trχ, κ}, there is no anomalous terms, which implies
112
Proposition 2.5.6. Under the assumption of Theorem 2.2.1, for Θ ∈ {∇trχ, κ} we have
u 1
δ− 2
Z
(s,s0 ) 0 0
X
2
k∇ Θ kL2sc (Su,u ) ≤ k∇3 ψ (s1 ,s1 ) kL2sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
du0
0 |u|
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
1 1 3
δ2 δ2 1 1 δ4
+ O2,4 O1,4 + O2,4 R02 R12 + O2,4 R0
|u| |u| |u|
1 3
δ2 1 1 δ4 δ
+ O1,4 R12 R22 + O1,4 R1 + ∆0 O2,4 O0,4
|u| |u| |u|
5
δ δ 1 1 δ4
+ ∆0 O1,4 O1,4 + ∆0 O1,4 R12 R22 + ∆0 O1,4 R1
|u| |u| |u|
5
δ 1 1 δ4
+ 2
O0,4 ∆0 R12 R22 + 2 O0,4 ∆0 R1
|u| |u|
3 3 1
δ4 1 1 δ 4 21 12 δ 2 32 12 δ
+ R1 R22 R02 + R1 R2 R0 + R R + R1 R0
|u| |u| |u| 1 0 |u|
3 3
δ2 2 δ2 δ 1 1 1
+ 3
∆0 O1,4 O0,4 + 3 ∆20 R0 O0,4 + 2 ∆0 (R02 R12 + δ 4 R0 )2
|u| |u| |u|
3 1
δ2 2 1 1 1 δ2 δ2
+ 3
∆0 O0,4 (R02 R12 + δ 4 R0 ) + 4 ∆30 O0,4
2
+ ∆0 R2 .
|u| |u| |u|
0
When Θ(s,s ) = µ, in the equation of ∇4 ∇2 µ, there are several anomalous terms. We en-
counter ∇χ̂∇α and it is bounded through estimate for E4 . We have χ̂∇2 α and it is bounded
through estimate for E14 . There is ∇2 χ̂α and we bound it through estimate for E3 . We also get
(∇η, ∇η)χ̂α, (η, η)∇χ̂α and we treat them through estimate for E7 . The term (η, η)χ̂∇α shows up,
and it is bounded through estimate for E8 . Therefore, we deduce
113
Proposition 2.5.7. Under the assumption of Theorem 2.2.1, it follows that for Θ = µ we obtain
u 1
δ− 2
Z
(s,s0 ) 0 0
X
2
k∇ Θ kL2sc (Su,u ) ≤ k∇3 ψ (s1 ,s1 ) kL2sc (Su,u0 ) kψ (s2 ,s2 ) kL∞
sc (Su,u0 )
du0
0 |u|
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
1
δ2 1 1 1 1
+ O2,4 O1,4 + δ 4 O2,4 R02 R12 + δ 4 O2,4 R0
|u|
1 1 1 3 δ
+ δ 2 O1,4 R12 R22 + δ 4 O1,4 R1 + ∆0 O2,4 O0,4
|u|
δ 3 1 1 3
+ ∆0 O1,4 O1,4 + δ 4 ∆0 O1,4 R02 R12 + δ 4 ∆0 O1,4 R0
|u|
1 3 3
δ 2 δ2 1 1 δ 4 12 12 δ 4 32 12 δ
+ ∆0 R 1 + R1 R22 R02 + R1 R2 R0 + R1 R0 + R1 R0
|u| |u| |u| |u| |u|
3 3
δ2 2 δ2 δ δ 1 1
+ 3
∆0 O1,4 O0,4 + 3 ∆30 R0 + 2 ∆0 R0 R1 + ∆0 O0,4 R12 R22
|u| |u| |u| |u|
5 3
δ δ4 3 1 δ2
+ ∆0 O0,4 R1 + 2 ∆0 R02 R12 + 2 ∆0 R20
|u| |u| |u|
δ2 1
+ 4 ∆40 R20 + δ 2 ∆0 R2 .
|u|
114
0
In the case Θ(s,s ) obeys

0 0 0
∇3 ∇2 Θ(s,s ) + λ[Θ(s,s ) ] + 1 trχ∇2 Θ(s,s )
0 X 0 0
=χ̂(0,1) · ∇2 Θ(s,s ) + ∇3 ψ (s1 ,s1 ) · ψ (s2 ,s2 )
s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
X 0 0 X 0 0
+ ∇2 ψ (s1 ,s1 ) · ∇ψ (s2 ,s2 ) + ∇2 ψ (s3 ,s3 ) · Ψ(s4 ,s4 )
s1 +s2 =s− 1 , s3 +s4 =s,
2
s03 +s04 =s0 +1
s01 +s02 =s0 + 1
2
X 0 0 X 0 0 0
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
X 0 0 X 0 0 0
s3 +s4 =s, s5 +s6 +s7 =s,
s03 +s04 =s0 +1 s05 +s06 +s07 =s0 +1
X 0 0 0
+ ∇ψ (s8 ,s8 ) · ψ (s9 ,s9 ) · Ψ(s10 ,s10 )
s8 +s9 +s10 =s+ 1 ,
2 (2.5.2)
s08 +s09 +s010 =s0 + 3
2
X 0 0 0 X 0 0
s8 +s9 +s10 =s+ 1 , s11 +s12 =s+ 1 ,
2 2
s08 +s09 +s010 =s0 + 3 s011 +s012 =s0 + 3
2 2
X 0 0 0 0
+ ∇ψ (s13 ,s13 ) · ψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 )
s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
X 0 0 0 0
+ ψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) · Ψ(s20 ,s20 )
s17 +s81 +s19 +s20 =s+1,
s017 +s018 +s019 +s020 =s0 +2
X 0 0 0
+ ψ (s21 ,s21 ) · Ψ(s22 ,s22 ) · Ψ(s23 ,s23 )
s21 +s22 +s23 =s+1,
s021 +s022 +s023 =s0 +2
X 0 0 0 0 0
+ ψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) · ψ (s28 ,s28 ) ,
s24 +s25 +s26 +s27 +s28 =s+1,
s024 +s025 +s026 +s027 +s028 =s0 +2
from evolution lemma (Proposition 2.3.8), Gronwall’s inequality and the definition of scale invariant
norms, we obtain the following proposition.
115
0
Proposition 2.5.8. When Θ(s,s ) satisfies (2.5.2), we get
0 0
|u|2λ−2s −1 k∇2 Θ(s,s ) kL2sc (Su,u )
Z u
1 0 X 0 0
≤ δ 2 |u0 |2λ−2s −4 k∇3 ψ (s1 ,s1 ) kL2sc (Su0 ,u ) kψ (s2 ,s2 ) kL∞
sc (Su0 ,u )
du0
u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
Z u
1 0 X 0 0
+ δ 2 |u0 |2λ−2s −4 k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u ) k∇ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0
u∞ s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
Z u
1 0 X 0 0
+ δ 2 |u0 |2λ−2s −4 k∇2 ψ (s3 ,s3 ) kL4sc (Su0 ,u ) kΨ(s4 ,s4 ) kL4sc (Su0 ,u ) du0
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
Z u
1 0 X 0 0
+ δ 2 |u0 |2λ−2s −4 k∇ψ (s3 ,s3 ) kL4sc (Su0 ,u ) k∇Ψ(s4 ,s4 ) kL4sc (Su0 ,u ) du0
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
Z u
1 0 X 0 0
+ δ 2 |u0 |2λ−2s −4 kψ (s3 ,s3 ) kL∞
sc (Su0 ,u )
k∇2 Ψ(s4 ,s4 ) kL2sc (Su0 ,u ) du0
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
Z u
1 0 X 0 0 0
+ δ 2 |u0 |2λ−2s −4 k∇2 ψ (s5 ,s5 ) kL4sc (Su0 ,u ) kψ (s6 ,s6 ) · ψ (s7 ,s7 ) kL4sc (Su0 ,u ) du0
u∞ s5 +s6 +s7 =s,
s05 +s06 +s07 =s0 +1
Z u
0 X 0 0
+ δ|u0 |2λ−2s −5 k∇ψ (s5 ,s5 ) kL4sc (Su0 ,u ) k∇ψ (s6 ,s6 ) kL4sc (Su0 ,u )
u∞ s5 +s6 +s7 =s,
s05 +s06 +s07 =s0 +1
0
× kψ (s7 ,s7 ) kL∞
sc (Su0 ,u )
du0
Z u
0 X 0 0
+ δ|u0 |2λ−2s −5 k∇ψ (s8 ,s8 ) kL4sc (Su0 ,u ) kψ (s9 ,s9 ) kL∞
sc (Su0 ,u )
u∞ s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2
0
× kΨ(s10 ,s10 ) kL4sc (Su0 ,u ) du0
Z u
1 0 X 0 0 0
+ δ 2 |u0 |2λ−2s −4 kψ (s8 ,s8 ) · ψ (s9 ,s9 ) kL4sc (Su0 ,u ) k∇Ψ(s10 ,s10 ) kL4sc (Su0 ,u ) du0
u∞ s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2
Z u
1 0 X 0 0
+ δ 2 |u0 |2λ−2s −4 k∇Ψ(s11 ,s11 ) kL4sc (Su0 ,u ) kΨ(s12 ,s12 ) kL4sc (Su0 ,u ) du0
u∞ s11 +s12 =s+ 1 ,
2
s011 +s012 =s0 + 3
2
116
Z u
1 0 X 0
+ δ 2 |u0 |2λ−2s −4 k∇ψ (s13 ,s13 ) kL4sc (Su0 ,u )
u∞ s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
0 0 0
× kψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 ) kL4sc (Su0 ,u ) du0
Z u
1 0 X 0 0 0
+ δ 2 |u0 |2λ−2s −4 kψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) kL4sc (Su0 ,u )
u∞ s17 +s81 +s19 +s20 =s+1,
s017 +s018 +s019 +s020 =s0 +2
0
× kΨ(s20 ,s20 ) kL4sc (Su0 ,u ) du0
Z u
0 X 0 0
+ δ|u0 |2λ−2s −5 kψ (s21 ,s21 ) kL∞
sc (Su0 ,u )
kΨ(s22 ,s22 ) kL4sc (Su0 ,u )
u∞ s21 +s22 +s23 =s+1,
s021 +s022 +s023 =s0 +2
0
× kΨ(s23 ,s23 ) kL4sc (Su0 ,u ) du0
Z u
1 0 X 0
+ δ 2 |u0 |2λ−2s −4 kψ (s28 ,s28 ) kL4sc (Su0 ,u )
u∞ s24 +s25 +s26 +s27 +s28 =s+1,
s024 +s025 +s026 +s027 +s028 =s0 +2
0 0 0 0
× kψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) kL∞
sc (Su0 ,u )
du0 .
Remark: Using the equations for elliptic estimates in appendix, it is easy to verify that, when
Θ ∈ {µ, κ}, ∇2 Θ obeys (2.5.2). In the case of Θ = ∇(Ωtrχ − u2 ), by employing the bound for Ω in
Proposition 2.3.1 and the fact that
ηA + η A = 2Ω−1 ∇A Ω,
0
it can be verified that ∇3 (Ωtrχ − u2 ) obeys the same bound for ∇2 Θ(s,s ) above.
For Θ ∈ {∇(Ωtrχ − u2 ), µ, κ}, we have λ[∇(Ωtrχ − u2 )] = s2 (∇(Ωtrχ − u2 )) = 32 , λ[µ] = s2 (µ) = 1
and λ[κ] = s2 (κ) = 12 . This implies
117
Proposition 2.5.9. For Θ ∈ {∇(Ωtrχ − u2 ), µ, κ}, we infer
0
|u|−1 k∇2 Θ(s,s ) kL2sc (Su,u )
Z u
1
X 0 0
≤ δ 2 |u0 |−4 k∇3 ψ (s1 ,s1 ) kL2sc (Su0 ,u ) kψ (s2 ,s2 ) kL∞
sc (Su0 ,u )
du0
u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
+ F1 + F2 + F3 + F4 + F5 + F6 + F7 + F8 + F9 + F10
+ F11 + F12 + F13 ,
where F1 − F13 are listed below:

Z u
1
X 0 0
F1 = δ 2 |u0 |−4 k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u ) k∇ψ (s2 ,s2 ) kL4sc (Su0 ,u ) du0 ,
u∞ s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
Z u
1
X 0 0
F2 = δ 2 |u0 |−4 k∇2 ψ (s3 ,s3 ) kL4sc (Su0 ,u ) kΨ(s4 ,s4 ) kL4sc (Su0 ,u ) du0 ,
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
Z u
1
X 0 0
F3 = δ 2 |u0 |−4 k∇ψ (s3 ,s3 ) kL4sc (Su0 ,u ) k∇Ψ(s4 ,s4 ) kL4sc (Su0 ,u ) du0 ,
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
Z u
1
X 0 0 0
F4 = δ 2 |u0 |−4 k∇2 ψ (s5 ,s5 ) kL4sc (Su0 ,u ) kψ (s6 ,s6 ) · ψ (s7 ,s7 ) kL4sc (Su0 ,u ) du0 ,
u∞ s5 +s6 +s7 =s,
s05 +s06 +s07 =s0 +1
Z u X 0 0
F5 = δ|u0 |−5 k∇ψ (s5 ,s5 ) kL4sc (Su0 ,u ) k∇ψ (s6 ,s6 ) kL4sc (Su0 ,u )
u∞ s5 +s6 +s7 =s,
s05 +s06 +s07 =s0 +1
0
× kψ (s7 ,s7 ) kL∞
sc (Su0 ,u )
du0 ,
Z u X 0 0
F6 = δ|u0 |−5 k∇ψ (s8 ,s8 ) kL4sc (Su0 ,u ) kψ (s9 ,s9 ) kL∞
sc (Su0 ,u )
u∞ s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2
0
× kΨ(s10 ,s10 ) kL4sc (Su0 ,u ) du0 ,
118
Z u
1
X 0 0 0
F7 = δ 2 |u0 |−4 kψ (s8 ,s8 ) · ψ (s9 ,s9 ) kL4sc (Su0 ,u ) k∇Ψ(s10 ,s10 ) kL4sc (Su0 ,u ) du0 ,
u∞ s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2
Z u
1
X 0 0
F8 = δ 2 |u0 |−4 k∇Ψ(s11 ,s11 ) kL4sc (Su0 ,u ) kΨ(s12 ,s12 ) kL4sc (Su0 ,u ) du0 ,
u∞ s11 +s12 =s+ 1 ,
2
s011 +s012 =s0 + 3
2
Z u
1
X 0
F9 = δ 2 |u0 |−4 k∇ψ (s13 ,s13 ) kL4sc (Su0 ,u )
u∞ s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
0 0 0
× kψ (s14 ,s14 ) · ψ (s15 ,s15 ) · ψ (s16 ,s16 ) kL4sc (Su0 ,u ) du0 ,
Z u
1
X 0 0 0
F10 = δ 2 |u0 |−4 kψ s17 ,s17 · ψ (s18 ,s18 ) · ψ (s19 ,s19 ) kL4sc (Su0 ,u )
u∞ s17 +s81 +s19 +s20 =s+1,
s017 +s018 +s019 +s020 =s0 +2
0
× kΨ(s20 ,s20 ) kL4sc (Su0 ,u ) du0 ,
Z u X 0 0
F11 = δ|u0 |−5 kψ (s21 ,s21 ) kL∞
sc (Su0 ,u )
kΨ(s22 ,s22 ) kL4sc (Su0 ,u )
u∞ s21 +s22 +s23 =s+1,
s021 +s022 +s023 =s0 +2
0
× kΨ(s23 ,s23 ) kL4sc (Su0 ,u ) du0 ,
Z u
1
X 0
F12 = δ 2 |u0 |−4 kψ (s28 ,s28 ) kL4sc (Su0 ,u )
u∞ s24 +s25 +s26 +s27 +s28 =s+1,
s024 +s025 +s026 +s027 +s028 =s0 +2
0 0 0 0
× kψ (s24 ,s24 ) · ψ (s25 ,s25 ) · ψ (s26 ,s26 ) · ψ (s27 ,s27 ) kL4sc (Su0 ,u ) du0
Z u
1
X 0 0
F13 = δ 2 |u0 |−4 kψ (s3 ,s3 ) kL∞sc (Su0 ,u )
k∇2 Ψ(s4 ,s4 ) kL2sc (Su0 ,u ) du0 .
u∞ s3 +s4 =s,
s03 +s04 =s0 +1
With the aid of L4 Sobolev Embedding (Proposition 2.3.9) and Hölder’s inequality, we demon-
strate the following estimate.
For F1 , we obtain
1
δ2 X 1 0 1 0
F1 ≤ sup k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) 0 k∇ψ (s2 ,s2 ) kL4sc (Su0 ,u0 ) .
|u| u0 ,u0 |u0 | |u |
s1 +s2 =s− 1 ,
2
s01 +s02 =s0 + 1
2
119
For F2 , we get
1
δ2 X 1 0
F2 ≤ 3 sup k∇2 ψ (s3 ,s3 ) kL4sc (Su0 ,u0 )
|u| 2
s3 +s4 =s, u0 ,u0 |u0 |
s03 +s04 =s0 +1

0 1 1
(s4 ,s04 ) 1
(s4 ,s04 )
× kΨ(s4 ,s4 ) k 2 2 (u∞ ,u) k∇Ψ k 2
(u∞ ,u) + δ kΨ
4 kL2 (u∞ ,u) .
Lsc (H u ) L2sc (H u ) sc (H u )
The anomalous term of F2 type is ∇2 χ̂ρ, which is bounded as above.
We estimate F3 via
1
δ2 X 1 0
F3 ≤ 3 sup 0
k∇ψ (s3 ,s3 ) kL4sc (Su0 ,u0 )
|u| 2
s3 +s4 =s, u0 ,u0 |u |
s03 +s04 =s0 +1

1 1
(s4 ,s04 ) 2 (s4 ,s04 ) 1
(s4 ,s04 )
× k∇Ψ k 2
(u∞ ,u) k∇ Ψ k 2
(u∞ ,u) + δ k∇Ψ
4 kL2 (u∞ ,u) .
L2sc (H u ) L2sc (H u ) sc (H u )
The anomalous terms of F3 type are ∇χ̂∇ρ and ∇χ̂∇β, which is bounded as above.
For F4 , we have
1 1
δ4 X 0 δ4 0
F4 ≤ sup k∇2 ψ (s5 ,s5 ) kL4sc (Su0 ,u0 ) kψ (s6 ,s6 ) kL4sc (Su0 ,u0 )
|u| s5 +s6 +s7 =s, u0 ,u0 |u0 |
s05 +s06 +s07 =s0 +1
1
δ2 0
× 0 2 kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
.
|u |
Of F4 type, a potential most anomalous term is (∇2 η, ∇2 η)trχtrχ. This term can be avoided by
using equation for ∇3 ∇3 (Ωtrχ− u2 ). The anomalous terms of F4 type left are (∇2 η, ∇2 η)trχ(Ωtrχ−
2 2 2 2 2 2 2
u ), (∇ η, ∇ η)ωtrχ, (∇ η, ∇ η)χ̂ χ̂, (∇ ω, ∇ χ)(η, η)(ω, χ).
These terms are bounded through the estimate above.
For F5 , we obtain
1
δ2 X 0 1 0
F5 ≤ sup k∇ψ (s5 ,s5 ) kL4sc (Su0 ,u0 ) k∇ψ (s6 ,s6 ) kL4sc (Su0 ,u0 )
|u| s5 +s6 +s7 =s, u0 ,u0 |u0 |
s05 +s06 +s07 =s0 +1
1
δ2 0
× 0 2 kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
.
|u |
120
For F6 , we derive
1 1
δ2 X
(s8 ,s08 ) δ2 0
F6 ≤ 3 sup k∇ψ kL4sc (Su0 ,u0 ) 0 2 kψ (s9 ,s9 ) kL∞
sc (Su0 ,u0 )
|u| 2 u0 ,u0 |u |
s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2

0 1 0 1 1 0
× kΨ(s10 ,s10 ) k 2 2 (u∞ ,u) k∇Ψ(s10 ,s10 ) k 2 2 (u∞ ,u) + δ 4 kΨ(s10 ,s10 ) kL2 (u ,u)
(H ∞ )
,
Lsc (H u ) Lsc (H u ) sc u
The anomalous term of F6 type are (∇η, ∇η)χρ and (∇η, ∇η)χβ, which are bounded through the
estimates above.
For F7 , we get
1 1
δ2 X 0 δ2 0
F7 ≤ 3 sup kψ (s8 ,s8 ) kL4sc (Su0 ,u0 ) 0 2
kψ (s9 ,s9 ) kL∞
sc (Su0 ,u0 )
|u| 2 u0 ,u0 |u |
s8 +s9 +s10 =s+ 1 ,
2
s08 +s09 +s010 =s0 + 3
2

0 1 0 1
× k∇Ψ(s10 ,s10 ) k 2 2 (u∞ ,u) k∇2 Ψ(s10 ,s10 ) k 2 2 (u∞ ,u)
Lsc (H u ) Lsc (H u )

1 0
× +δ 4 k∇Ψ(s10 ,s10 ) kL2 (u ,u)
(H ∞ )
.
sc u
The anomalous term of F7 type is (η, η)χ∇β, which is bounded as above.
For F8 , when Θ = ∇(Ωtrχ − u2 ), there is no F8 term. When Θ = µ, the F8 terms are ∇β · ρ and
β · ∇ρ. When Θ = κ, the F8 terms are β∇β and β∇β. From Proposition 2.5.2, we have
1 1 1 1 1 1
kρ, βkL4sc (Su,u ) ≤ R12 R22 + δ 4 R1 + δ 4 (I (0) + R + R)(R12 R22 + R0 ).
Therefore, we derive
1
δ2 X 0 1 0 1
F8 ≤ 2 k∇Ψ(s11 ,s11 ) k 2 2 (u ,u) k∇2 Ψ(s11 ,s11 ) k 2 2 (u∞ ,u)
|u| Lsc (H u ∞ ) Lsc (H u )
s11 +s12 =s+ 1 ,
2
s011 +s012 =s0 + 3
2

1 0
+ δ 4 k∇Ψ(s11 ,s11 ) kL2 (u∞ ,u)
sc (H u )

0 1 0 1 1 0
× kΨ(s12 ,s12 ) k 2 2 (u∞ ,u) k∇Ψ(s12 ,s12 ) k 2 2 (u∞ ,u) + δ 4 kΨ(s12 ,s12 ) kL2 (u∞ ,u) .
Lsc (H u ) Lsc (H u ) sc (H u )
121
For F9 and F10 , we have
1
δ2 X 0
F9 ≤ sup k∇ψ (s13 ,s13 ) kL4sc (Su0 ,u0 )
|u| u0 ,u0
s13 +s14 +s15 +s16 =s+ 1 ,
2
s013 +s014 +s015 +s016 =s0 + 3
2
1 1
0 δ2 0 δ2 0
× kψ (s14 ,s14 ) kL4sc (Su0 ,u0 ) 0
kψ (s15 ,s15 ) kL∞
sc (Su0 ,u0 )
kψ (s16 ,s16 ) kL∞
sc (Su0 ,u0 )
,
|u |2 |u0 |2
and
δ X 0 0
F10 ≤ 5 sup kψ s17 ,s17 kL4sc (Su0 ,u0 ) kψ (s18 ,s18 ) kL∞
sc (Su0 ,u0 )
|u| 2
s17 +s81 +s19 +s20 =s+1, u0 ,u0
s017 +s018 +s019 +s020 =s0 +2
1
δ2 0
× 0 2
kψ (s19 ,s19 ) kL∞
sc (Su0 ,u0 )
|u |

1 1
(s20 ,s020 ) (s20 ,s020 ) 1
(s20 ,s020 )
× kΨ k 2
(u∞ ,u) k∇Ψ k 2
(u∞ ,u) + δ kΨ 4 kL2 (u∞ ,u) .
L2sc (H u ) L2sc (H u ) sc (H u )
For F11 , when Θ = ∇(Ωtrχ − u2 ), there is no F11 term. When Θ = µ, the F11 term is (η, η)βρ.
When Θ = κ, the F11 term is (η, η)ββ. Hence, we derive
δ X 0
F11 ≤ sup kψ (s21 ,s21 ) kL∞
sc (Su0 ,u0 )
|u|3 s21 +s22 +s23 =s+1, u0 ,u0
s021 +s022 +s023 =s0 +2

1 1
(s22 ,s022 ) (s22 ,s022 ) 1
(s22 ,s022 )
× kΨ k 2 (u∞ ,u) k∇Ψ 2
k 2 (u∞ ,u) + δ kΨ kL2 (H (u∞ ,u) )
2 4

1 1
(s23 ,s023 ) 2 (s23 ,s023 ) 2 1
(s23 ,s023 )
× kΨ k 2 (u∞ ,u) k∇Ψ k 2 (u∞ ,u) + δ kΨ
4 kL2 (H (u∞ ,u) ) .
We bound F12 through
3
δ2 X 0 0
F12 ≤ sup kψ (s28 ,s28 ) kL4sc (Su0 ,u0 ) kψ (s24 ,s24 ) kL4sc (Su0 ,u0 )
|u|4 s24 +s25 +s26 +s27 +s28 =s+1, u0 ,u0
s024 +s025 +s026 +s027 +s028 =s0 +2
1
0 0 δ2 0
× kψ (s25 ,s25 ) kL∞
sc (Su0 ,u0 )
kψ (s26 ,s26 ) kL∞
sc (Su0 ,u0 ) 0 2
kψ (s27 ,s27 ) kL∞
sc (Su0 ,u )
du.
|u |
122
Finally, F13 obeys
1
δ2 0 0
F13 ≤ 5 sup sup kψ (s3 ,s3 ) kL∞
sc (Su,u )
k∇2 Ψ(s4 ,s4 ) kL2 (u∞ ,u) .
|u| 2 s3 +s4 =s, u,u sc (H u )
s03 +s04 =s0 +1
Gathering all the estimate, with Propositions 2.4.6, 2.4.14 and 2.4.18 we deduce
Proposition 2.5.10. Under the assumption of Theorem 2.2.1, for Θ ∈ {∇(Ωtrχ − u2 ), µ, κ}, we
have
0
|u|−1 k∇2 Θ(s,s ) kL2sc (Su,u )
Z u
1
X 0 0
≤ δ 2 |u0 |−4 k∇3 ψ (s1 ,s1 ) kL2sc (Su0 ,u ) kψ (s2 ,s2 ) kL∞
sc (Su0 ,u )
du0
u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 − 1
2
1
δ 2 (0)
+ (I + R + R)4 .
|u|
In the estimates above we encounter O2,4 , which can be bounded via the inequality
O2,4 ≤ O3,2 + O1,4 .
This is guaranteed by the following Gagliardo-Nirenberg’s inequality
2 1
k∇2 φkL4sc (Su,u ) ≤ k∇3 φkL3 2 k∇φkL3 4 ,
sc (Su,u ) sc (Su,u )
where φ is a Su,u -tangent tensor.
To get the desired O3,2 estimates, we need to make another bootstrap assumption.
Recall
O3,2

= sup k∇3 χ̂, ∇3 trχ, ∇3 ω, ∇3 ω † , ∇3 η, ∇3 η, ∇3 trχ, ∇3 ω, ∇3 ω † kL2sc (Su,u )
u,u

1
+ k∇3 χ̂kL2sc (Su,u ) ,
|u|
123
and we set up a new bootstrap assumption
O3,2 ≤ ∆5 . (2.5.3)
Proposition 2.5.11. Under the assumption of Theorem 2.2.1 and the bootstrap assumption
(2.5.3), we demonstrate
1
δ2 1
3
k∇ trχkL2sc (Su,u ) ≤ ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ),
|u|
and
1
δ2 1
k∇3 χ̂kL2sc (Su,u ) ≤k∇2 βkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5
|u|
1
δ2 2
+ (O0,∞ O2,2 + O1,4 ).
|u|
0
Proof. From the derived estimates for ∇2 Θ(s,s ) , we have
u 1
δ− 2
Z
k∇3 trχkL2sc (Su,u ) ≤ kχ, ωkL∞
sc (Su,u0 )
k∇3 χ, ∇3 ωkL2sc (Su,u0 ) du0
0 |u|
1
+ δ 4 ((I (0) )3 + R3 + R3 )
1
δ2 1
≤ ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
|u|
Recalling the null structure equation
1 1 1
div χ̂ = ∇trχ − (η − η) · (χ̂ − trχ) − β,
2 2 2
124
we apply the elliptic estimates in Proposition 2.3.17 to obtain
k∇3 χ̂kL2sc (Su,u ) ≤k∇3 trχkL2sc (Su,u ) + k∇2 βkL2sc (Su,u )

1 1
δ2 δ2
+ k∇2 (η, η)kL2sc (Su,u ) kχkL∞
sc (Su,u )
+ kη, ηkL∞
sc (Su,u )
k∇2 χkL2sc (Su,u )
|u| |u|
1
δ2
+ k∇η, ∇ηkL4sc (Su,u ) k∇χkL4sc (Su,u )
|u|
1
+ δ 4 (I (0) + R + R)5
1
δ2 1
≤k∇2 βkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5
|u|
1
δ2 2
+ (O0,∞ O2,2 + O1,4 ).
|u|
(2.5.3), we have
1
δ2 1
k∇3 ηkL2sc (Su,u ) ≤k∇2 ρ, ∇2 σkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5 .
|u|
0
Proof. The estimates derived for ∇2 Θ(s,s ) implies
u 1
δ− 2
Z
k∇2 µkL2sc (Su,u ) ≤ kχkL∞
sc (Su,u0 )
k∇3 η, ∇3 ηkL2sc (Su,u0 ) du0
0 |u|
1
+ δ 4 ((I (0) )3 + R3 + R3 )
1
δ2 1
≤ ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
|u|
Thanks to the definition µ = −div η − ρ, η obeys the following elliptic system:
div η = −µ − ρ,
1
curl η = σ + χ̂ ∧ χ̂.
2
125
We apply the elliptic estimates in Proposition 2.3.17 to obtain
k∇3 ηkL2sc (Su,u ) ≤k∇2 µkL2sc (Su,u ) + k∇2 ρ, ∇2 σkL2sc (Su,u )

1 1 1 1
+ δ 2 k∇2 χ̂kL2sc (Su,u ) kχ̂kL∞
sc (Su,u )
+ δ 2 kχ̂kL∞
sc (Su,u )
k∇2 χ̂kL2sc (Su,u )
|u| |u|
1 1
+ δ 2 k∇χ̂kL4sc (Su,u ) k∇χ̂kL4sc (Su,u )
|u|
1
+ δ 4 (I (0) + R + R)5
1
δ2 1
≤k∇2 ρ, ∇2 σkL2sc (Su,u ) + ∆0 ∆6 + δ 4 (I (0) + R + R)5
|u|
1
2
+ δ 2 (O2,2 O0,∞ + O1,4 ).
(2.5.3), we have
1
δ2 1
k∇3 ω, ∇3 ω † kL2sc (Su,u ) ≤ k∇2 βkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5 .
|u|
0
Proof. The following inequality follows from the estimates for ∇2 Θ(s,s )
u 1
δ− 2
Z
k∇2 κkL2sc (Su,u ) ≤ kχ, ωkL∞
sc (Su,u0 )
k∇3 ω, ∇3 ω † kL2sc (Su,u0 ) du0
0 |u|
u 1
δ− 2
Z
+ kωkL∞
sc (Su,u0 )
k∇3 ωkL2sc (Su,u0 ) du0
0 |u|
u 1
δ− 2
Z
+ kη, ηkL∞
sc (Su,u0 )
k∇3 η, ∇3 ηkL2sc (Su,u0 ) du0
0 |u|
1
+ δ 4 ((I (0) )3 + R3 + R3 )
1
δ2 1
≤ ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
|u|
Recalling ω † is defined to be the solution of
1
∇4 ω † = σ
2
126
with zero data. And κ is defined as
1
κ := −∇ω + ∗ ∇ω † − β.
2
By the following div − curl system:
1
div ∇ω = −div κ − div β,
2
curl ∇ω = 0,
1
curl ∇ω † = curl κ + curl β,
2
div ∇ω † = 0,
and elliptic estimates in Proposition 2.3.16, we demonstrate
k∇3 ω, ∇3 ω † kL2sc (Su,u ) ≤k∇2 κkL2sc (Su,u ) + k∇2 βkL2sc (Su,u )

1
+ δ 4 (I (0) + R + R)5
1
δ2 1
≤k∇2 βkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5 .
|u|
(2.5.3), we obtain
1
δ2 1
k∇3 ηkL2sc (Su,u ) ≤k∇2 ρ, ∇2 σkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5
|u|
1
2
+ δ 2 (O2,2 O0,∞ + O1,4 ).
127
0
Proof. From the estimates for ∇2 Θ(s,s ) above, we have
Z u 1
−1 δ2
|u| 2
k∇ µkL2sc (Su,u ) ≤ ktrχkL∞
sc (Su0 ,u )
k∇3 ηkL2sc (Su0 ,u ) du0
u∞ |u0 |4
Z u 1
δ2
+ kη, ηkL∞
sc (Su0 ,u )
k∇3 χkL2sc (Su0 ,u ) du0
u∞ |u0 |4
1
δ4
+ ((I (0) )3 + R3 + R3 )
|u|
1 1
1 δ2 δ4
≤ O3,2 [η] + 2 ∆0 ∆5 + ((I (0) )3 + R3 + R3 ).
|u| |u| |u|
Hence the result in Proposition 2.5.12 implies
1
δ2 1
2
k∇ µkL2sc (Su,u ) 2 2
≤k∇ ρ, ∇ σkL2sc (Su,u ) + ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 )
|u|
1
2
+ δ 2 (O2,2 O0,∞ + O1,4 ).
Thanks to the definition µ = −div η − ρ, η obeys the following elliptic system:
div η = −µ − ρ,
1
curl η = −σ − χ̂ ∧ χ̂.
2
Employing the elliptic estimates in Proposition 2.3.16, we obtain
k∇3 ηkL2sc (Su,u ) ≤k∇2 µkL2sc (Su,u ) + k∇2 ρ, ∇2 σkL2sc (Su,u )

1 1 1 1
+ δ 2 k∇2 χ̂kL2sc (Su,u ) kχ̂kL∞
sc (Su,u )
+ δ 2 kχ̂kL∞
sc (Su,u )
|u| |u|
1 1 1
+ δ 2 k∇χ̂kL4sc (Su,u ) k∇χ̂kL4sc (Su,u ) + δ 4 (I (0) + R + R)5
|u|
1
δ2 1
≤k∇2 ρ, ∇2 σkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5
|u|
1
2
+ δ 2 (O2,2 O0,∞ + O1,4 ).
128
(2.5.3), we have
1
3 3 † 2 δ2 1
k∇ ω, ∇ ω k L2sc (Su,u ) ≤k∇ βk L2sc (Su,u ) + 2 ∆0 ∆5 + δ 4 (I (0) + R + R)5 .
|u|
0
Proof. Estimates obtained for ∇2 Θ(s,s ) lead to
Z u 1
δ2
|u|−1 k∇2 κkL2sc (Su,u ) ≤ kχ̂, ωkL∞
sc (Su0 ,u )
k∇3 ω, ∇3 ω † kL2sc (Su0 ,u ) du0
u∞ |u0 |4
Z u 1
δ2
+ kωkL∞
sc (Su0 ,u )
k∇3 ωkL2sc (Su0 ,u ) du0
u∞ |u0 |4
Z u 1
δ2
+ kη, ηkL∞
sc (Su0 ,u )
k∇3 η, ∇3 ηkL2sc (Su0 ,u ) du0
u∞ |u0 |4
1
δ4
+ ((I (0) )3 + R3 + R3 ).
|u|
Multiplying |u| on both sides, we derive
1
δ2 1
2
k∇ κkL2sc (Su,u ) ≤ ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
|u|2
Recalling ω † solves
1
∇3 ω † = σ
2
with zero initial data on Hu∞ . And κ is defined:
1
κ := ∇ω + ∗ ∇ω † − β.
2
By the following div − curl system:
1
div ∇ω = div κ + div β,
2
curl ∇ω = 0,
1
2
129
div ∇ω † = 0,
and elliptic estimates in Proposition 2.3.16, we demonstrate
k∇3 ω, ∇3 ω † kL2sc (Su,u ) ≤k∇2 κkL2sc (Su,u ) + k∇2 βkL2sc (Su,u )

1
+ δ 4 (I (0) + R + R)5
1
δ2 1
≤k∇2 βkL2sc (Su,u ) + 2
∆0 ∆5 + δ 4 (I (0) + R + R)5 .
|u|
We are left to derive bounds for k∇3 trχkL2sc (Su,u ) and k∇3 χ̂kL2sc (Su,u ) .
(2.5.3), we have
1 1
k∇3 trχkL2sc (Su,u ) ≤ k∇2 βkL2sc (Su,u ) + δ 2 ∆0 ∆5 + δ 4 ((I (0) ) + R + R)5 ,
and
1
|u|
1
≤ k∇2 βkL2sc (Su,u ) + k∇2 ρ, ∇2 σkL2sc (Su,u )
|u|
1
δ2 1 1
+ ∆0 ∆5 + δ 4 (I (0) + R + R)5 + δ 2 (O2,2 O0,∞ + O1,4
2
).
|u|
0
Proof. The derived estimates for ∇2 Θ(s,s ) guarantee that
2
|u|−1 k∇3 (Ωtrχ − )kL2sc (Su,u )
u
Z u 1
δ2
≤ 0 |4
k∇3 ωkL2sc (Su0 ,u ) du0
u∞ |u
Z u 1
δ2 2 3 0
+ kχ̂, ω, Ωtrχ − kL∞ (S ) k∇ χkL2sc (Su0 ,u ) du
u∞ |u |
0 4 u sc u0 ,u
1
+ δ 4 ((I (0) )3 + R3 + R3 )
1
1 δ2 1
≤ O3,2 [ω] + ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
|u| |u|
130
Hence the result in Proposition 2.5.13 implies
2 1 1
k∇3 (Ωtrχ − )kL2sc (Su,u ) ≤O3,2 [ω] + δ 2 ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 )
u
1 1
≤k∇2 βkL2sc (Su,u ) + δ 2 ∆0 ∆5 + δ 4 ((I (0) )3 + R3 + R3 ).
To obtain estimate for ∇3 trχ, we employ
k∇3 trχkL2sc (Su,u )

1
3 2 δ2
≤k∇ (Ωtrχ − )kL2sc (Su,u ) + 2 ktrχkL∞
sc (Su,u )
|u|k∇3 ΩkL2sc (Su,u )
u |u|
1 1
δ2 δ2
+ k∇2 ΩkL4sc (Su,u ) k∇trχkL4sc (Su,u ) + k∇ΩkL4sc (Su,u ) k∇2 trχkL4sc (Su,u )
|u| |u|
1
2 δ2
≤k∇3 (Ωtrχ − )kL2sc (Su,u ) + 2 ktrχkL∞
sc (Su,u )
|u|k∇3 ΩkL2sc (Su,u )
u |u|
1 1
δ2 δ2
+ k∇η, ∇ηkL4sc (Su,u ) k∇trχkL4sc (Su,u ) + kη, ηkL4sc (Su,u ) k∇2 trχkL4sc (Su,u )
|u| |u|
δ
+ 2 kη, ηkL∞sc (Su,u )
kη, ηkL4sc (Su,u ) k∇trχkL4sc (Su,u ) ,
|u|
where we use the bound for Ω in Proposition 2.3.1 and the fact ∇Ω = Ω(η, η).
As a consequence, we need estimate for k∇3 ΩkL2sc (Su,u ) . Using commutation formulas (Propo-
sitions 2.3.12-2.3.15), and
∇3 Ω = −2ω,
it can be verified that for Ω, we have
1
∇3 ∇Ω + trχ∇Ω = (χ̂, ω)∇Ω + Ω∇ω + (η, η)Ωω,
2
131
∇3 ∇2 Ω + trχ∇2 Ω
=χ̂∇2 Ω + (∇χ, ∇ω)∇Ω + (χ̂, ω)∇2 Ω + Ω∇2 ω + (∇η, ∇η)Ωω
+ (η, η)∇Ωω + (η, η)Ω∇ω + χ(η, η)∇Ω + β∇Ω + (η, η)(η, η)Ωω,
and
3
∇3 ∇3 Ω + trχ∇3 Ω
2
=∇χ∇2 Ω + (χ̂, ω)∇3 Ω + ∇ω∇2 Ω + (∇2 χ, ∇2 ω)∇Ω + Ω∇3 ω + (∇2 η, ∇2 η)Ωω
+ (∇η, ∇η)∇Ωω + (∇η, ∇η)Ω∇ω + (η, η)∇2 Ωω + (η, η)∇Ω∇ω
+ (η, η)Ω∇2 ω + ∇χ(η, η)∇Ω + χ(∇η, ∇η)∇Ω + χ(η, η)∇2 Ω + ∇β∇Ω + β∇2 Ω
+ (∇η, ∇η)(η, η)Ωω + (η, η)(η, η)∇Ωω + (η, η)(η, η)Ω∇ω + χ(η, η)(η, η)∇Ω
+ β∇Ω(η, η) + (η, η)(η, η)(η, η)Ωω.
∇3 Ω has signature ( 32 , 32 ). With Gronwall’s inequality and evolution lemma (Proposition 2.3.8),
we obtain
1
k∇3 ΩkL2sc (Su,u )
|u|
Z u Z u 1
1 0 δ2 1
≤ 0 3
k∇ 3
ωkL 2 (S 0 ) du + 0 3 0
k∇χ, ∇ωkL4sc (Su0 ,u ) k∇2 ΩkL4sc (Su0 ,u ) du0
u∞ |u | u∞ |u | |u |
sc u ,u
Z u 1
δ2 1
+ 0 3 0
k∇2 χ, ∇2 ωkL4sc (Su0 ,u ) k∇ΩkL4sc (Su0 ,u ) du0
u∞ |u | |u |
Z u 1
δ2
+ 0 4
k∇2 η, ∇2 ηkL4sc (Su0 ,u ) kωkL4sc (Su0 ,u ) du0
u∞ |u |
Z u 1
δ2
+ 0 4
k∇η, ∇ηkL4sc (Su0 ,u ) k∇ωkL4sc (Su0 ,u ) du0
u∞ |u |
132
Z u 1
δ2
+ kη, ηkL4sc (Su0 ,u ) k∇2 ωkL4sc (Su0 ,u ) du0
u∞ |u0 |4
Zu 1
δ2
+ k∇βkL4sc (Su0 ,u ) k∇ΩkL4sc (Su0 ,u ) du0
u∞ |u0 |4
Zu 1
δ2
+ kβkL4sc (Su0 ,u ) k∇2 ΩkL4sc (Su0 ,u ) du0
u∞ |u0 |4
Zu
δ
+ k∇η, ∇ηkL4sc (Su0 ,u ) k∇ΩkL4sc (Su0 ,u ) kωkL∞ sc (Su0 ,u )
du0
u∞ |u0 |5
Zu
δ
+ kη, ηkL4sc (Su0 ,u ) k∇2 ΩkL4sc (Su0 ,u ) kωkL∞
sc (Su0 ,u )
du0
u∞ |u0 |5
Zu
δ 1
+ kη, ηkL∞ sc (Su0 ,u )
k∇ΩkL4sc (Su0 ,u ) 0 k∇ω, ∇χkL4sc (Su0 ,u ) du0
u∞ |u0 |4 |u |
Zu 1 1
δ2 δ2
+ kχkL∞
sc (Su0 ,u )
k∇ΩkL4sc (Su0 ,u ) k∇η, ∇ηkL4sc (Su0 ,u ) du0
u∞ |u0 |3 |u0 |2
Zu 1 1
δ2 δ2
+ kχkL∞sc (Su0 ,u )
k∇2 ΩkL4sc (Su0 ,u ) kη, ηkL4sc (Su0 ,u ) du0
u∞ |u0 |3 |u0 |2
Zu
δ
+ k∇η, ∇ηkL4sc (Su0 ,u ) kη, ηkL4sc (Su0 ,u ) kωkL∞ sc (Su0 ,u )
du0
u∞ |u0 |5
Zu
δ
+ k∇ωkL4sc (Su0 ,u ) kη, ηkL4sc (Su0 ,u ) kη, ηkL∞ sc (Su0 ,u )
du0
u∞ |u0 |5
Zu
δ
+ kβkL2sc (Su0 ,u ) k∇ΩkL∞ sc (Su0 ,u )
kη, ηkL∞
sc (Su0 ,u )
du0
u∞ |u0 |5
Zu 1
δ δ2
+ kχ, ωkL∞
sc (Su0 ,u )
k∇Ω, η, ηkL∞
sc (Su0 ,u )
kη, ηkL∞
sc (Su0 ,u )
kη, ηkL∞
sc (Su0 ,u )
du0 .
u∞ |u0 |4 |u0 |2
1
1 1 δ2
k∇3 ΩkL2sc (Su,u ) ≤ 2
O3,2 [ω] + 2 (I (0) + R + R)5 .
|u| |u| |u|
Multiplying |u|2 on both sides, we obtain
1
|u|k∇3 ΩkL2sc (Su,u ) ≤ O3,2 [ω] + δ 2 (I (0) + R + R)5 .
133
Putting the estimates above altogether and using the result in Proposition 2.5.13, we derive
1 1
k∇3 trχkL2sc (Su,u ) ≤O3,2 [ω] + δ 2 ∆0 ∆5 + δ 4 ((I (0) ) + R + R)5
1 1
≤k∇2 βkL2sc (Su,u ) + δ 2 ∆0 ∆5 + δ 4 ((I (0) ) + R + R)5 .
To obtain the estimates for ∇3 χ̂, we employ the Codazzi equation
1 1 1
div χ̂ = β + ∇trχ − (η − η) · (χ̂ − trχ)
2 2 2
and elliptic estimates in Proposition 2.3.17 to derive
1
|u|
1
1 1 δ2
≤ k∇3 trχkL2sc (Su,u ) + k∇2 βkL2sc (Su,u ) + k∇2 η, ∇2 ηkL2sc (Su,u ) 2 kχkL∞
sc (Su,u )
|u| |u| |u|
1 1
δ2 1 δ2 1
+ k∇η, ∇ηkL4sc (Su,u ) k∇χkL4sc (Su,u ) + kη, ηkL∞
sc (Su,u )
k∇2 χkL2sc (Su,u )
|u| |u| |u| |u|
1
+ δ 4 (I (0) + R + R)5
1
1 δ2 1
≤ k∇2 βkL2sc (Su,u ) + k∇2 η, ∇2 ηkL2sc (Su,u ) + ∆0 ∆5 + δ 4 (I (0) + R + R)5
|u| |u|
1
δ2 1
+ (O2,2 O0,∞ + O1,4 O1,4 ) + O3,2 [ω].
|u| |u|
For k∇2 η, ∇2 ηkL2sc (Su,u ) , we have
1
k∇2 η, ∇2 ηkL2sc (Su,u ) ≤ δ 4 k∇2 η, ∇2 ηkL4sc (Su,u ) .
The conclusions in Propositions 2.5.12 and 2.5.14 immediately imply
134
1
|u|
1
≤ k∇2 βkL2sc (Su,u ) + k∇2 ρ, ∇2 σkL2sc (Su,u )
|u|
1
δ2 1 1
+ ∆0 ∆5 + δ 4 (I (0) + R + R)5 + δ 2 (O2,2 O0,∞ + O1,4
2
).
|u|
Gathering all the estimates above, we obtain
1
O3,2 ≤ k∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + δ 4 (I (0) + R + R)5 .
Let
1
∆5 k∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + δ 4 (I (0) + R + R)5 .
Then we improve bootstrap assumption (2.5.3) and derive

1
O3,2 ≤ C k∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + δ 4 (I (0) + R + R)5 ,
2.5.4 L4 (S) Estimates for Second Derivatives of Ricci Coefficients
We recall the following norms:
O2,4

= sup k∇2 χ̂, ∇2 trχ, ∇2 ω, ∇2 η, ∇2 η, ∇2 trχ, ∇2 ωkL4sc (Su,u )
u,u

1
+ k∇2 χ̂kL4sc (Su,u ) .
|u|
135
For any Su,u -tangent tensor φ, we have the following Gagliardo-Nirenberg’s inequality
2 1
k∇2 φkL4sc (Su,u ) ≤ k∇3 φkL3 2 k∇φkL3 4 .
With this inequality, for O2,4 we deduce
O2,4 ≤ Ck∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2sc (Su,u ) + C(I (0) + R + R)5 .
Moreover, in the end of of this section, we prove a useful proposition for energy estimate.
Z u 1
δ2 1
k∇2 χ, ∇2 ω, ∇2 η, ∇2 η, ∇2 χ, ∇2 ωkL4sc (Su0 ,u ) du0
u∞ |u0 |2 |u0 |
1 1
δ2 δ2
≤ 1 Ck∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2 (u∞ ,u) + C(I (0) + R + R)5 ,
|u| 2 sc (H u ) |u|
and
Z u
1 1
δ− 2 k∇2 χ, ∇2 ω, ∇2 η, ∇2 η, ∇2 χ, ∇2 ωkL4sc (Su,u0 ) du0
0 |u|
1 1
≤δ Ck∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2
2 (0,u) + δ 2 C(I (0) + R + R)5 .
sc (Hu )
Proof. Employing the results in Proposition 2.5.17, we derive
Z u 1
δ2 1
k∇2 χ, ∇2 ω, ∇2 η, ∇2 η, ∇2 χ, ∇2 ωkL4sc (Su0 ,u ) du0
u∞ |u0 |2 |u0 |
Z u 1
δ2
≤ C k∇β, ∇ρ, ∇σ, ∇βk 2
Lsc (Su,u ) + (I (0)
+ R + R) 5
du0
u∞ |u0 |2
1 1
δ2 δ2
≤ 1 Ck∇β, ∇ρ, ∇σ, ∇βkL2 (u∞ ,u) + C(I (0) + R + R)5 ,
|u| 2 sc (H u ) |u|
136
and
Z u
1 1
δ− 2k∇2 χ, ∇2 ω, ∇2 η, ∇2 η, ∇2 χ, ∇2 ωkL4sc (Su,u0 ) du0
0 |u|
Z u
− 12 (0) 5
≤ δ C k∇β, ∇ρ, ∇σ, ∇βkL2sc (Su,u ) + (I + R + R)
0
1 1
≤δ Ck∇β, ∇ρ, ∇σ, ∇βkL2
2 (0,u) + δ 2 C(I (0) + R + R)5 .
sc (Hu )
2.6 Curvature Estimates
In this section, we derive and prove the energy estimates for the curvature components and their
first two angular derivatives. We will show in particular R + R ≤ I (0) . Together with the estimates
in the previous sections, we will therefore improve all of the bootstrap assumptions (2.3.1), (2.3.2),
(2.3.3) and (2.3.4) to obtain Theorem 2.2.1.
The estimates in this section are based on an key observation:
Both curvature components Ψ with signature s(Ψ) = (s, s0 ), and the kth-order angular deriva-
tives of curvature components Ψ with signature s(∇k Ψ) = (s, s0 ), satisfy the following transport
equations:
0 1 0 1 0 1
X 0 0
∇3 Ψ(s,s ) + ( + s0 )trχΨ(s,s ) = DΨ(s− 2 ,s + 2 ) + ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) , (2.6.1)
2 s1 +s2 =s,
s01 +s02 =s0 +1
or
1 0 1 0 X 0 0
∇4 Ψ(s− 2 ,s + 2 ) = ∗ DΨ(s,s ) + ψ (s̃1 ,s̃1 ) · Ψ(s̃2 ,s̃2 ) , (2.6.2)
s̃1 +s̃2 =s+ 1 ,
2
s̃01 +s̃02 =s0 + 1
2
0 0
where Ψ(s,s ) and ψ (s,s ) stand for S-tangent tensor fields with signatures (s, s0 ). D and ∗ D are
Hodge operators listed below. And ∗ D is the L2 adjoint of D.
• The operator D1 takes any 1-form φ into the pairs of functions (div φ, curl φ).
• The operator D2 takes any S tangent symmetric traceless tensor φ into the S-tangent one
form div φ.
137
• The operator ∗ D1 takes the pair of scalar functions (ρ, σ) into the S-tangent 1-form −∇ρ +
∗
∇σ.
• The operator ∗ D2 takes 1-form φ on S into the 2-covariant, symmetric, traceless tensors

1
− 2 ∇b φa + ∇a φb − (div φ)γab .
When k = 0, we can simply check null Bianchi equations. When k ≥ 1, this observation can be
verified by taking kth order angular derivatives of null Bianchi equations and using commutation
formulas (Propositions 2.3.12-2.3.15).
For Hodge operators D and ∗ D, we address a useful proposition:
0 1 0 1
Proposition 2.6.1. For Ψ(s,s ) and Ψ(s− 2 ,s + 2 ) satisfying (2.6.1) and (2.6.2), we have
Z
0 0 1 0 1 0 1 0 1 0
|u0 |4s Ψ(s,s ) DΨ(s− 2 ,s + 2 ) + |u0 |4s Ψ(s− 2 ,s + 2 )∗ DΨ(s,s ) du0 du0 = 0. (2.6.3)
Du,u
Proof. This follows from the definitions of D, ∗ D and integrating by parts.
2.6.1 Curvature Estimates in the Scale Invariant Norms
To derive energy estimates, we will need the following integration by parts formula, which can be
proved by a direct computation:
Proposition 2.6.2. Suppose φ1 and φ2 are r tensorfields, then we have
Z Z
φ 1 ∇4 φ 2 + φ 2 ∇4 φ 1
Du,u Du,u
Z Z Z
= φ1 φ2 − φ1 φ2 + (2ω − trχ)φ1 φ2 .
(u∞ ,u) (u∞ ,u)
Hu H0 Du,u
Proposition 2.6.3. Suppose φ1 and φ2 are r tensorfields, then we have
Z Z
φ 1 ∇3 φ 2 + φ 2 ∇3 φ 1
Du,u Du,u
Z Z Z
= φ1 φ2 − φ1 φ2 + (2ω − trχ)φ1 φ2 .
(0,u) (0,u)
Hu Hu∞ Du,u
138
(1) (2)
Proposition 2.6.4. Suppose we have an r tensorfield φ and an r − 1 tensorfield φ. It follows
that
Z Z
(1) A1 A2 ...Ar (2)
φ ∇A r φA1 ...Ar−1 + ∇Ar (1) φA1 A2 ...Ar (2) φA1 ...Ar−1
Du,u Du,u
Z
=− (η + η)(1) φ(2) φ.
Du,u
With these propositions above, we obtain
0 1 0 1
Proposition 2.6.5. Assuming a pair of S-tangent tensor fields Ψ(s,s ) and Ψ(s− 2 ,s + 2 ) satisfy
0 1 0 1 0 1
∇3 Ψ(s,s ) + ( + s0 )trχΨ(s,s ) = DΨ(s− 2 ,s + 2 ) + J3 ,
2
and
1 0 1 0
∇4 Ψ(s− 2 ,s + 2 ) = ∗ DΨ(s,s ) + J4 ,
then we have
0 1 0 1
kΨ(s,s ) k2L2 (0,u) + kΨ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu ) sc (H u )
0 1 0 1
≤kΨ(s,s ) k2L2 (H (0,u) ) + kΨ(s− 2 ,s + 2 ) k2L2 (H (u∞ ,u) )
sc u∞ sc 0
Z u Z u −3
δ 2 0 1 0 1
+ 0|
k(η + η) · Ψ(s,s ) · Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0
0 u∞ |u
Z u Z u −3
δ 2 0
+ 0
kΨ(s,s ) · J3 kL1sc (Su0 ,u0 ) du0 du0
0 u∞ |u |
Z u Z u −3
δ 2 1 0 1
+ 0|
kΨ(s− 2 ,s + 2 ) · J4 kL1sc (Su0 ,u0 ) du0 du0 .
0 u∞ |u
139
Proof. With the aid of equations and employing Propositions 2.6.2, 2.6.3 and 2.6.4, we compute
Z
0 1 0 1 0 1 0 1
< |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s ∇4 Ψ(s− 2 ,s + 2 ) >γ
Du,u
Z
0 1 0 1 0 0
= < |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s ∗ DΨ(s,s ) >γ
Du,u
Z Z
2s0 (s− 12 ,s0 + 21 ) 2s0 0 0 1 0 1
+ < |u| Ψ , |u| J4 >γ + |u|4s (η + η)Ψ(s,s ) Ψ(s− 2 ,s + 2 )
Du,u Du,u
Z
0
(s− 12 ,s0 + 21 ) 0 0
= < −|u|2s DΨ , |u|2s Ψ(s,s ) >γ
Du,u
Z Z
0 1 0 1 0 0 0 1 0 1
+ < |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s J4 >γ + |u|4s (η + η)Ψ(s,s ) Ψ(s− 2 ,s + 2 )
Du,u Du,u
Z
0 0 0 0
=− < |u|2s ∇3 Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u
Z
0 1 0 0 0
− < |u|2s ( + s0 )trχΨ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u 2
Z Z
0 0 0 0 1 0 1 0
+ < |u|2s Ψ(s,s ) , |u|2s J3 >γ + < |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s J4 >γ
Du,u Du,u
Z
0 0
(s− 21 ,s0 + 12 )
+ |u|4s (η + η)Ψ(s,s ) Ψ
Du,u
Z
0 0 0 0
=− < ∇3 (|u|2s Ψ(s,s ) ), |u|2s Ψ(s,s ) >γ
Du,u
2s0 −1 2s0 (s,s0 )
Z
0 0
− < Ω |u| Ψ , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z
0 1 0 0 0
− < |u|2s ( + s0 )trχΨ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u 2
Z Z
2s0 (s,s0 ) 2s0 0 1 0 1 0
+ < |u| Ψ , |u| J3 >γ + < |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s J4 >γ
Du,u Du,u
Z
0 0
(s− 21 ,s0 + 12 )
+ |u|4s (η + η)Ψ(s,s ) Ψ
Du,u
Z
0 0 0 0
=− < ∇3 (|u|2s Ψ(s,s ) ), |u|2s Ψ(s,s ) >γ
Du,u
2s0 −1
Z
0 0 0 0
− < (Ω − 1)|u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z
1 0 0 0 0
+ < |u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z
0 1 2 0 0 0
− < |u|2s ( + s0 )(trχ + )Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u 2 |u|
Z Z
2s0 (s,s0 ) 2s0 0 1 0 1 0
+ < |u| Ψ , |u| J3 >γ + < |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s J4 >γ
Du,u D
Z 140u,u
0 0 1 0 1
+ |u|4s (η + η)Ψ(s,s ) Ψ(s− 2 ,s + 2 ) .
Du,u
On the other hand, applying Proposition 2.6.3 gives
Z
0 0 0 0
− < ∇3 (|u|2s Ψ(s,s ) ), |u|2s Ψ(s,s ) >γ
Du,u
Z Z
1 0 0 1 0 0
=− |u|4s (Ψ(s,s ) )2 + |u|4s (Ψ(s,s ) )2
2 (0,u)
Hu 2 (0,u)
Hu∞
Z Z
0 0 1 0 0
− ω|u|4s (Ψ(s,s ) )2 + trχ|u|4s (Ψ(s,s ) )2
Du,u 2 Du,u
Z Z
1 1 0 0 0 0
=− |u|4s (Ψ(s,s ) )2 + |u|4s (Ψ(s,s ) )2
2 Hu
(0,u) 2 Hu∞(0,u)
Z Z
0 0 1 2 0 0
− ω|u|4s (Ψ(s,s ) )2 + (trχ + )|u|4s (Ψ(s,s ) )2
Du,u 2 Du,u |u|
Z
1 0 0
− |u|4s (Ψ(s,s ) )2 .
Du,u |u|
Combining the two identities above together, we arrive at
Z
0 1 0 1 0 1 0 1
< |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s ∇4 Ψ(s− 2 ,s + 2 ) >γ
Du,u
2s0 −1
Z
0 0 0 0
=− < (Ω − 1)|u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z
2 0 0 0 0
− < s0 (trχ + )|u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z Z
0 0 0 0 1 0 1 0
Du,u Du,u
Z Z
0 0 1 0 1 0 0
+ |u|4s (η + η)Ψ(s,s ) Ψ(s− 2 ,s + 2 ) − ω|u|4s (Ψ(s,s ) )2
Du,u Du,u
Z Z
1 0 0 1 0 0
− |u|4s (Ψ(s,s ) )2 + |u|4s (Ψ(s,s ) )2 .
2 (0,u)
Hu 2 (0,u)
Hu∞
Meanwhile, using Proposition 2.6.2, we get
Z
0 1 0 1 0 1 0 1
< |u|2s Ψ(s− 2 ,s + 2 ) , |u|2s ∇4 Ψ(s− 2 ,s + 2 ) >γ
Du,u
Z Z
1 0 1 0 1 1 0 1 0 1
= |u|4s (Ψ(s− 2 ,s + 2 ) )2 − |u|4s (Ψ(s− 2 ,s + 2 ) )2
2 (u ,u)
Hu ∞ 2 (u ,u)
H0 ∞
Z
1 0 1 0 1
+ (ω − trχ)|u|4s (Ψ(s− 2 ,s + 2 ) )2 .
Du,u 2
141
Therefore, we conclude
Z Z
1 0 1 0 1 1 0 0
|u|4s (Ψ(s− 2 ,s + 2 ) )2 + |u|4s (Ψ(s,s ) )2
2 (u ,u)
Hu ∞ 2 (0,u)
Hu
Z Z
1 0 1 0 1 1 0 0
= |u|4s (Ψ(s− 2 ,s + 2 ) )2 + |u|4s (Ψ(s,s ) )2
2 (u ,u)
H0 ∞ 2 (0,u)
Hu∞
2s0 −1
Z
0 0 0 0
− < (Ω − 1)|u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z
2 0 0 0 0
− < s0 (trχ + )|u|2s Ψ(s,s ) , |u|2s Ψ(s,s ) >γ
Du,u |u|
Z Z
0 0 0 0 1 0 1 0
Du,u Du,u
Z Z
0 0
4s (s,s ) (s− 12 ,s0 + 12 ) 0 0
+ |u| (η + η)Ψ Ψ − ω|u|4s (Ψ(s,s ) )2
Du,u Du,u
Z
1 0 1 0 1
− (ω − trχ)|u|4s (Ψ(s− 2 ,s + 2 ) )2 .
Du,u 2
Multiplying δ 2s−3 on both sides and using the Gronwall’s inequality, we obtain desired estimate.
2.6.2 Curvature Estimate I
With all the preparations, we are ready to prove energy estimates for curvature components. The
0
curvature component Ψ(s,s ) satisfies
0 1 0 1 0 1
X 0 0
∇3 Ψ(s,s ) + ( + s0 )trχΨ(s,s ) = DΨ(s− 2 ,s + 2 ) + ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) ,
2 s1 +s2 =s,
s01 +s02 =s0 +1
and
1 0 1 0 X 0 0
∇4 Ψ(s− 2 ,s + 2 ) = ∗ DΨ(s,s ) + ψ (s̃1 ,s̃1 ) · Ψ(s̃2 ,s̃2 ) .
s̃1 +s̃2 =s+ 1 ,
2
s̃01 +s̃02 =s0 + 1
2
Applying Proposition 2.6.5 and using Hölder’s and Gronwall’s inequality, we conclude
0 1 0 1
Proposition 2.6.6. Assuming a pair of S-tangent tensor fields Ψ(s,s ) and Ψ(s− 2 ,s + 2 ) satisfy
142
(2.6.1) and (2.6.2), then we have
0 1 0 1
sc (Hu ) sc (H u )
0 1 0 1
≤kΨ(s,s ) k2L2 (0,u) + kΨ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu∞ ) sc (H 0 )
1
δ2 X 1 0 0
sc (Su0 ,u0 )
kΨ(s2 ,s2 ) kL2 (H (0,u) )
|u| s1 +s2 +s3 =2s,
0≤u0 ≤u, |u0 | sc u
u∞ ≤u0 ≤u
s01 +s02 +s03 =2s0 +1
0
× kΨ(s3 ,s3 ) kL2 (0,u) ,
sc (Hu )
or
0 1 0 1
sc (Hu ) sc (H u )
0 1 0 1
sc (Hu∞ ) sc (H 0 )
1
X 1 0 0
+δ 2 sup kψ (s1 ,s1 ) kL∞
sc (Su0 ,u0 )
kΨ(s2 ,s2 ) kL2 (H (u∞ ,u) )
s1 +s2 +s3 =2s,
0≤u0 ≤u, |u0 | sc u
u∞ ≤u0 ≤u
s01 +s02 +s03 =2s0 +1
0
× kΨ(s3 ,s3 ) kL2 (u∞ ,u) .
sc (H u )
With this, we arrive at
Proposition 2.6.7. Under the assumption of Theorem 2.2.1 and the bootstrap assumption (2.3.4),
we have
1
R[α] + R[β] ≤ R(0) + δ 4 R.
Proof. Using Proposition 2.6.6 and Hölder’s inequality, for the pair
0 1 0 1
Ψ(s,s ) = α, Ψ(s− 2 ,s + 2 ) = β,
143
we have
0 1 0 1
δkΨ(s,s ) k2L2 (0,u) + δkΨ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu ) sc (H u )
0 1 0 1
≤δkΨ(s,s ) k2L2 (0,u) + δkΨ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu∞ ) sc (H 0 )
1
δ 2 X 1 0 1 0
sc (Su0 ,u0 )
δ 2 kΨ(s2 ,s2 ) kL2 (H (0,u) )
|u| s1 +s2 +s3 =2s,
0≤u0 ≤u, |u0 | sc u
u∞ ≤u0 ≤u
s01 +s02 +s03 =2s0 +1
1 0
× δ 2 kΨ(s3 ,s3 ) kL2 (0,u) .
sc (Hu )
We do not have α term in the null Bianchi equations for ∇3 α and ∇4 β. Hence all the curvature
(0,u)
components are bounded in L2sc (Hu ) norms. The anomalous terms we have are due to χ̂ and α.
And they are bounded through the estimates above. The conclusion of proposition follows.
we have
1
R[ρ, σ, β] + R[β, α] ≤ R(0) + δ 4 R.
Proof. Applying Proposition 2.6.6 and Hölder’s inequality, for the pairs
0 1 0 1
Ψ(s,s ) = ρ, σ, Ψ(s− 2 ,s + 2 ) = β,
0 1 0 1
Ψ(s,s ) = β, Ψ(s− 2 ,s + 2 ) = α,
we derive
0 1 0 1
sc (Hu ) sc (H u )
0 1 0 1
sc (Hu∞ ) sc (H 0 )
1
X 1 0 0
+δ 2 sup kψ (s1 ,s1 ) kL∞
sc (Su0 ,u0 )
kΨ(s2 ,s2 ) kL2 (H (u∞ ,u) )
s1 +s2 +s3 =2s,
0≤u0 ≤u, |u0 | sc u
u∞ ≤u0 ≤u
s01 +s02 +s03 =2s0 +1
0
× kΨ(s3 ,s3 ) kL2 (u∞ ,u) .
sc (H u )
We do not have α term in the null Bianchi equations for ∇3 ρ, ∇3 σ, ∇3 β and ∇4 β, ∇4 α. Hence
144
all the curvature components are bounded in L2sc (H (u
u
∞ ,u)
) norms. The anomalous terms we have
are due to χ̂. And they are bounded through the estimates above. The conclusion of proposition
follows.
we have
1 1 1 1
R[β] + R[ρ, σ] ≤ (I (0) ) + (R(0) ) + δ 8 (R + R) + (I (0) ) 2 (R(0) ) 4 R14 [α].
Proof. Appealing to L4 Sobolev Embedding (Proposition 2.3.9), Proposition 2.6.6 for energy
estimate and Hölder’s inequality, for the pair
0 1 0 1
Ψ(s,s ) = β, Ψ(s− 2 ,s + 2 ) = ρ, σ
we deduce
0 1 0 1
sc (Hu ) sc (H u )
0 1 0 1
sc (Hu∞ ) sc (H 0 )
1
δ2 X 1 0 0
sc (Su0 ,u0 )
kΨ(s2 ,s2 ) kL2 (H (0,u) )
|u| s1 +s2 +s3 =2s,
0≤u0 ≤u, |u0 | sc u
u∞ ≤u0 ≤u
s01 +s02 +s03 =2s0 +1
0
× kΨ(s3 ,s3 ) kL2 (0,u)
sc (Hu )
1
1 δ 4 1 1 1
+ sup kχ̂kL4sc (Su,u ) kρ, σkL2 (H (0,u) ) δ 4 kαk 2 2 (0,u) k∇αk 2 2 (0,u)
|u| u,u |u| sc u Lsc (Hu ) Lsc (Hu )
1
1 δ4 1
+ sup kχ̂kL4sc (Su,u ) kρ, σkL2 (H (0,u) ) δ 2 kαkL2 (H (0,u) ) .
|u| u,u |u| sc u sc u
For this pair, we encounter anomalous term (ρ, σ)χ̂α, and it is bounded above. Recalling for χ̂,
from Proposition 2.4.5, we have
1 1 1 1 3
δ4 δ4 δ4 1 1 δ2 δ4
kχ̂kL4sc (Su,u ) ≤ kχ̂kL4sc (Su∞ ,u ) + 3 R0 R1 +
2 2
3 R0 + ∆0 O0,4 .
|u| |u∞ | |u| 2 |u| 2 |u|2
145
Using Hölder’s inequality again, therefore, we prove
1 1 1
R2 [β] + R2 [ρ, σ] ≤ (R(0) )2 + δ 4 (R2 + R2 ) + I (0) R02 [α]R12 [α].
With Proposition 2.6.7, we arrive at
1 1 1
R2 [β] + R2 [ρ, σ] ≤ (I (0) )2 + (R(0) )2 + δ 4 (R2 + R2 ) + I (0) (R(0) ) 2 R12 [α].
The desired estimate follows.
Gathering all the estimates in this section, we conclude
Proposition 2.6.10. Under the assumption of Theorem 2.2.1 and bootstrap assumption 2.3.4,
there exists δ0 = δ0 (R, R), when 0 < δ < δ0 , we have
1
R0 + R0 ≤ R(0) + R1 [α] + δ 8 (I (0) + R(0) + R + R)5 .
2.6.3 Curvature Estimate II
The aim of this subsection is to derive energy estimate for first angular derivatives of curvature
0 1 0 1
components. And the pair ∇Ψ(s,s ) and ∇Ψ(s− 2 ,s + 2 ) satisfy
0 0
∇3 ∇Ψ(s,s ) + (1 + s0 )trχ∇Ψ(s,s )
1 0 1 0 0 0 0
X X
=∇DΨ(s− 2 ,s + 2 ) + ∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) + ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 )
s1 +s2 =s, s1 +s2 =s,
s01 +s02 =s0 +1 s01 +s02 =s0 +1 (2.6.4)
X 0 0 0 X 0 0
+ ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) + Ψ(s6 ,s6 ) · Ψ(s7 ,s7 ) ,
s3 +s4 +s5 =s+ 1 , s6 +s7 =s+ 1 ,
2 2
s03 +s04 +s05 =s0 + 3 s06 +s07 =s0 + 3
2 2
146
or
1 0 1
∇4 ∇Ψ(s− 2 ,s + 2 )
0 X 0 0 X 0 0
=∇∗ DΨ(s,s ) + ∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) + ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 )
s1 +s2 =s+ 1 , s1 +s2 =s+ 1 ,
2 2 (2.6.5)
s01 +s02 =s0 + 1 s01 +s02 =s0 + 1
2 2
X 0 0 0 X 0 0
+ ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) + Ψ(s6 ,s6 ) · Ψ(s7 ,s7 ) .
s3 +s4 +s5 =s+1, s6 +s7 =s+1,
s03 +s04 +s05 =s0 +1 s06 +s07 =s0 +1
Using equations and Hölder’s inequality, we arrive at
0 1 0 1
Proposition 2.6.11. For curvature component ∇Ψ(s,s ) and ∇Ψ(s− 2 ,s + 2 ) obeying (2.6.4) and
(2.6.5), we have
0 1 0 1
k∇Ψ(s,s ) k2L2 (0,u) + k∇Ψ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu ) sc (H u )
0 1 0 1
≤k∇Ψ(s,s ) k2L2 (0,u) + k∇Ψ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu∞ ) sc (H 0 )
+ M 1 + M 2 + M 3 + M 4 + M5 + M6 + M7 + M8 ,
where M1 − M8 are listed below:
u u 3
δ− 2
Z Z X 0 0 0
M1 = k∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 0
M2 = kψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 1 0 1
M3 = k∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
u u 3
δ− 2
Z Z X 0 0 1 0 1
M4 = kψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
147
u u 3
δ− 2
Z Z X 0 0 0 0
M5 = kψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
u u
δ −1
Z Z X 0 0 0
M6 = kψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) kL2sc (Su0 ,u0 )
0 u∞ |u0 |2 s3 +s4 +s5 =s+1,
s03 +s04 +s05 =s0 +1
1 0 1
× k∇Ψ(s− 2 ,s + 2 ) kL2sc (Su0 ,u0 ) du0 du0 ,
Z u Z u −3
δ 2 X 0 0 0
M7 = 0
kΨ(s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u | 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
u u 3
δ− 2
Z Z X 0 0 1 0 1
M8 = kΨ(s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 .
0 u∞ |u0 | s1 +s2 =s+1,
s01 +s02 =s0 +1
We will do estimate term by term. For M1 , it is easy to verify that for equations (2.6.4) and
(2.6.5), α and α terms will not appear at the same time. If there is no α term, we have
u u 3
δ− 2
Z Z X 0 0 0
M1 = k∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 1
δ− 2
Z Z X 0 0
≤ k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 )
0 u∞ |u0 |3 s1 +s2 =s,
s01 +s02 =s0 +1
0
× k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
148
u u 1
δ− 2
Z Z X 0 0 1
≤ k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL2 2
|u0 |3 sc (Su0 ,u0 )
0 u∞ s1 +s2 =s,
s01 +s02 =s0 +1
0 1 0
× k∇Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc u ,u
Z u Z u −1
δ 4 X 0 0
+ 0 |3
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 )
0 u∞ |u s +s =s, 1 2
s01 +s02 =s0 +1
0
Z uZ u 1
δ2 X 1 0 1 0 1
≤ 0 2 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) δ − 4 kΨ(s2 ,s2 ) kL2 2 (S 0 0 )
u∞ |u | |u | sc u ,u
0 s +s =s, 1 2
s01 +s02 =s0 +1
1 0 1 1 0
× δ − 4 k∇Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) δ − 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc u ,u
Z uZ u 3
δ 4 X 1 0 1 0
+ 0 2 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) δ − 2 kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 )
0 u∞ |u | s +s =s,
|u |
1 2
s01 +s02 =s0 +1
1 0
× δ − 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
1
δ4 X 1 0 1 0 1
≤ sup 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) δ 4 kΨ(s2 ,s2 ) k 2 2 (0,u)
|u| s1 +s2 =s, u0 u0 |u | Lsc (Hu )
s01 +s02 =s0 +1
0 1 0
× k∇Ψ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
Lsc (Hu ) sc (Hu )
1
δ4 X 1 0
+ sup 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
|u| s1 +s2 =s, u0 u0 |u |
s01 +s02 =s0 +1
1 0 0
× δ 2 kΨ(s2 ,s2 ) kL2 (0,u) k∇Ψ(s,s ) kL2 (0,u) ,
sc (Hu ) sc (Hu )
where we use L4 Sobolev Embedding (Proposition 2.3.9), definition of scale invariant norm and
Hölder’s inequality.
149
Similarly, if there is no a term in M1 , we derive
u u 3
δ− 2
Z Z X 0 0 0
M1 = k∇ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 1
δ− 2
Z Z X 0 0
≤ k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 )
0 u∞ |u0 |3 s1 +s2 =s,
s01 +s02 =s0 +1
0
Z u Z u −1
δ 2 X 0 0 1
≤ 0 3
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL2 2 (S 0 0 )
u∞ |u |
sc u ,u
0 s +s =s, 1 2
s01 +s02 =s0 +1
0 1 0
× k∇Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc u ,u
Z u Z u −1
δ 4 X 0 0
+ 0 |3
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 )
0 u∞ |u s +s =s, 1 2
s01 +s02 =s0 +1
0
Z uZ u
1
X 1 0 1 1
(s2 ,s02 ) 2
≤ δ− 2 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) 1 kΨ kL2 (S 0 0 )
0 u∞ s +s =s,
|u | |u0 | 2 sc u ,u
1 2
s01 +s02 =s0 +1
1 0 1 1 0
× k∇Ψ(s2 ,s2 ) kL2 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc (Su0 ,u0 )
1 0
|u0 | 2 |u |
Z u Z u
1
X 1 0 1 0
+ δ− 4 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) 0 kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 )
0 u∞ s1 +s2 =s,
|u | |u |
s01 +s02 =s0 +1
1 0
× 0
k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u |
1
X 1 0 0 1
≤δ 2 sup 0 k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) k 2 2 (u∞ ,u)
0
s +s =s, u u
0 |u | L sc (H u )
1 2
s01 +s02 =s0 +1
0 1 0
× k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
Lsc (H u ) sc (H u )
3
X 1 0
+ δ4 sup 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
s1 +s2 =s, u0 u0 |u |
s01 +s02 =s0 +1
0 0
× kΨ(s2 ,s2 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) ,
sc (H u ) sc (H u )
where we use L4 Sobolev Embedding (Proposition 2.3.9), definition of scale invariant norm and
150
Hölder’s inequality. The anomalous terms of M1 type are (∇χ, ∇ω, ∇η)α∇Ψ and ∇χ(β, ρ, σ)∇Ψ,
where Ψ is a curvature component. These terms are bounded through the estimates.
In the same fashion, we bound M2 , M3 , M4 , M5 and M6 . For M2 , we have either
1
δ2 X 1 0 0 1
M2 ≤ sup kψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇Ψ(s2 ,s2 ) k 2 2 (0,u)
|u| s1 +s2 =s, u0 u0 |u0 | Lsc (Hu )
s01 +s02 =s0 +1
0 1 0
× k∇2 Ψ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
Lsc (Hu ) sc (Hu )
3
δ4 X 1 0
+ sup kψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
|u| s1 +s2 =s, u0 u0 |u0 |
s01 +s02 =s0 +1
0 0
× k∇Ψ(s2 ,s2 ) kL2 (0,u) k∇Ψ(s,s ) kL2 (0,u) ,
sc (Hu ) sc (Hu )
or
1
X 1 0 0 1
M2 ≤δ 2 sup 0
kψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u)
s1 +s2 =s, u0 u0 |u | L sc (H u )
s01 +s02 =s0 +1
0 1 0
× k∇2 Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
3
X 1 0
+ δ4 sup 0
kψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
s1 +s2 =s, u0 u0 |u |
s01 +s02 =s0 +1
0 0
× k∇Ψ(s2 ,s2 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u ) sc (H u )
We encounter χ̂(∇β, ∇ρ, ∇σ∇β)∇Ψ as anomalous term of M2 type. They are bounded through
the estimates.
M3 and M4 can be bounded in the same manner as for M1 and M2 , respectively.
151
The term M5 satisfies either
1 1
δ4 X δ2 0
(s4 ,s04 ) 1
(s5 ,s05 ) 2
1
M5 ≤ sup 0 2 kψ (s3 ,s3 ) kL∞ (S 0 0 ) kψ k L4 (S 0 0 ) δ 4 kΨ k (0,u)
|u| u0 u0 |u |
sc u ,u sc u ,u L2sc (Hu )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
0 1 0
× k∇Ψ(s5 ,s5 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
Lsc (Hu ) sc (Hu )
1 1
δ4 X δ2 0 0
+ sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
|u| u0 u0 |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 1
2
1 0 0
sc (Hu ) sc (Hu )
or
1
1
X δ2 0 0 0 1
M5 ≤δ 2 sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 ) kΨ(s5 ,s5 ) k 2 2 (u∞ ,u)
u0 u0 |u | Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
0 1 0
1
3
X δ2 0 0
+ δ4 sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
u0 u0 |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 1
2
0 0
× kΨ(s5 ,s5 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u ) sc (H u )
We control M6 in the same method. With these estimates, M5 and M6 type anomalous terms
(η, η)(χ, ω, η, η)α∇Ψ, (η, η)χ(β, ρ, σ, β)∇Ψ are clearly bounded.
152
We bound M7 by
Z u Z u 1
δ2 X 1 0 1 1 0 1
M7 ≤ kΨ(s1 ,s1 ) kL2 2 k∇Ψ(s1 ,s1 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
0 u∞ |u0 |2 |u0 | 2 |u0 | 2
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
1 0 1 0 1
− 41
× δ − 4 kΨ(s2 ,s2 ) kL2 2 δ k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 0
× δ − 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z uZ u 3
δ4 X 1 0
+ 0 2 0
kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1
|u |
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× δ − 4 kΨ(s2 ,s2 ) kL2 2 δ − 4 k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
0
− 12
× δ k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z uZ u 3
δ4 X 1 1
(s1 ,s01 ) 2 1 1
(s1 ,s01 ) 2
+ 0 2 1 kΨ kL2 (S 0 0 ) 1 k∇Ψ kL2 (S 0 0 )
u∞ |u | |u0 | 2 |u0 | 2
sc u ,u sc u ,u
0 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× δ − 2 kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 ) δ − 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z uZ u
δ X 1 0
+ 0 2
kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1
|u|
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× δ − 2 kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 ) δ − 2 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 .
Hence, we have
1
δ2 X 0 1 0 1
M7 ≤ 3 kΨ(s1 ,s1 ) k 2 2 (u ,u) k∇Ψ(s1 ,s1 ) k 2 2 (u∞ ,u)
|u| 2 Lsc (H u ∞ ) Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× kΨ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
3
δ4 X 0 0 1
+ 3 kΨ(s1 ,s1 ) kL2 (u∞ ,u) kΨ(s2 ,s2 ) k 2 2 (0,u)
|u| 2 sc (H u ) Lsc (Hu )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0
× k∇Ψ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
Lsc (Hu ) sc (Hu )
153
3
δ4 X 0 1 0 1
+ 3 kΨ(s1 ,s1 ) k 2 2 (u ,u) k∇Ψ(s1 ,s1 ) k 2 2 (u∞ ,u)
|u| 2 Lsc (H u ∞ ) Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 0
× kΨ(s2 ,s2 ) kL2 (H (0,u) ) k∇Ψ(s,s ) kL2 (H (0,u) )
sc u sc u
δ X 0 0 0
+ 3 kΨ(s1 ,s1 ) kL2 (H (u∞ ,u) ) kΨ(s2 ,s2 ) kL2 (H (0,u) ) k∇Ψ(s,s ) kL2 (H (0,u) ) .
|u| s +s =s+ 1 ,
2 sc u sc u sc u
1 2 2
s01 +s02 =s0 + 3
2
Alternatively M7 is bounded by
Z u Z u
1 X 1 0 1
− 14 0 1
M7 ≤ δ − 4 kΨ(s1 ,s1 ) kL2 2 δ k∇Ψ(s1 ,s1 ) kL2 2
|u0 | sc (Su0 ,u0 ) sc (Su0 ,u0 )
0 u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× kΨ(s2 ,s2 ) kL2 2 k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
|u0 | 2 |u0 | 2
1 0
0
|u |
Z uZ u 1
δ4 X 1 0
+ 0|
δ − 2 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× kΨ(s2 ,s2 ) kL2 2 k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 )
sc (Su0 ,u0 )
1 1
|u0 | 2 |u0 | 2
1 0
× 0 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u |
Z uZ u 1
δ4 X 1 0 1 1 0 1
+ 0|
δ − 4 kΨ(s1 ,s1 ) kL2 2 (S 0 0 ) δ − 4 k∇Ψ(s1 ,s1 ) kL2 2 (S 0 0 )
0 u∞ |u 1
sc u ,u sc u ,u
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 ) 0 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u0 | |u |
Z uZ u 1
δ2 X 1 0
+ 0
δ − 2 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× kΨ(s2 ,s2 ) kL2sc (Su0 ,u0 ) 0 k∇Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 .
|u0 | |u |
154
1
δ4 X 1 0 1 0 1
M7 ≤ δ 4 kΨ(s1 ,s1 ) k 2 2 (0,u) k∇Ψ(s1 ,s1 ) k 2 2 (0,u)
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× kΨ(s2 ,s2 ) k 2 2 (u ,u) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
Lsc (H u ∞ ) Lsc (H u ) sc (H u )
1
δ4 X 1 0 0 1
+ δ 2 kΨ(s1 ,s1 ) kL2 (0,u) kΨ(s2 ,s2 ) k 2 2 (u∞ ,u)
|u| sc (Hu ) Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0
1
δ2 X 1 0 1 0 1
+ δ 4 kΨ(s1 ,s1 ) k 2 2 (0,u) k∇Ψ(s1 ,s1 ) k 2 2 (0,u)
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 0
× kΨ(s2 ,s2 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
sc (H u ) sc (H u )
1
δ 2 X 1 0 0 0
+ δ 2 kΨ(s1 ,s1 ) kL2 (0,u) kΨ(s2 ,s2 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) .
|u| sc (Hu ) sc (H u ) sc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
In the same manner, M8 is controlled as well.
Putting all the estimates in this section altogether, we arrive at
Proposition 2.6.12. Under the assumption of Theorem 2.2.1 and bootstrap assumption (2.3.4),
1
R1 + R1 ≤ R(0) + δ 8 (I (0) + R(0) + R + R)5 .
155
2.6.4 Curvature Estimate III
Our goal in this subsection is to derive estimates for the second angular derivatives of curvature
0 1 0 1
components and the pair ∇2 Ψ(s,s ) and ∇2 Ψ(s− 2 ,s + 2 ) obey
0 3 0
∇3 ∇2 Ψ(s,s ) + ( + s0 )trχ∇2 Ψ(s,s )
2
1 0 1 0 0 0 0
X X
=∇3 Ψ(s− 2 ,s + 2 ) + ∇2 ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) + ∇ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 )
s1 +s2 =s, s1 +s2 =s,
s01 +s02 =s0 +1 s01 +s02 =s0 +1
X 0 0 0 X 0 0 0
+ ∇ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) + ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · ∇Ψ(s5 ,s5 )
s3 +s4 +s5 =s+ 1 , s3 +s4 +s5 =s+ 1 ,
2 2
s03 +s04 +s05 =s0 + 3 s03 +s04 +s05 =s0 + 3
2 2
X 0 0 X 0 0 (2.6.6)
+ Ψ(s1 ,s1 ) · ∇Ψ(s2 ,s2 ) + ψ (s1 ,s1 ) · ∇2 Ψ(s2 ,s2 )
s1 +s2 =s+ 1 , s1 +s2 =s,
2
s01 +s02 =s0 +1
s01 +s02 =s0 + 3
2
X 0 0 0 0
+ ψ (s6 ,s6 ) · ψ (s7 ,s7 ) · ψ (s8 ,s8 ) · Ψ(s9 ,s9 )
s6 +s7 +s8 +s9 =s+1,
s06 +s07 +s08 +s09 =s0 +2
X 0 0 0
+ ψ (s10 ,s10 ) · Ψ(s11 ,s11 ) · Ψ(s12 ,s12 ) ,
s10 +s11 +s12 =s+1,
s010 +s011 +s012 =s0 +2
or
1 0 1
∇4 ∇2 Ψ(s− 2 ,s + 2 )
0 X 0 0 X 0 0
=∇3 Ψ(s,s ) + ∇2 ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) + ∇ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 )
s1 +s2 =s+ 1 , s1 +s2 =s+ 1 ,
2 2
s01 +s02 =s0 + 1 s01 +s02 =s0 + 1
2 2
X 0 0 0 X 0 0 0 (2.6.7)
+ ∇ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) + ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · ∇Ψ(s5 ,s5 )
s3 +s4 +s5 =s+1, s3 +s4 +s5 =s+1,
s03 +s04 +s05 =s0 +1 s03 +s04 +s05 =s0 +1
X 0 0 X 0 0
+ Ψ(s1 ,s1 ) · ∇Ψ(s2 ,s2 ) + ψ (s1 ,s1 ) · ∇2 Ψ(s2 ,s2 )
s1 +s2 =s+1, s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 +1
s01 +s02 =s0 + 1
2
156
X 0 0 0 0
+ ψ (s6 ,s6 ) · ψ (s7 ,s7 ) · ψ (s8 ,s8 ) · Ψ(s9 ,s9 )
s6 +s7 +s8 +s9 =s+ 3 ,
2
s06 +s07 +s08 +s09 =s0 + 3
2
X 0 0 0
+ ψ (s10 ,s10 ) · Ψ(s11 ,s11 ) · Ψ(s12 ,s12 ) .
s10 +s11 +s12 =s+ 3 ,
2
s010 +s011 +s012 =s0 + 3
2
With equations above and Hölder’s inequality, we arrive at
0 1 0 1
Proposition 2.6.13. For curvature component ∇2 Ψ(s,s ) , ∇2 Ψ(s− 2 ,s + 2 ) satisfying (2.6.6) and
(2.6.7), we have
0 1 0 1
k∇2 Ψ(s,s ) k2L2 (0,u) + k∇2 Ψ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu ) sc (H u )
0 1 0 1
≤k∇2 Ψ(s,s ) k2L2 (0,u) + k∇2 Ψ(s− 2 ,s + 2 ) k2L2 (u∞ ,u)
sc (Hu∞ ) sc (H 0 )
+ N1 + N2 + N3 + N4 + N5 + N6 + N7 + N8 + N9
+ N10 + N11 + N12 + N13 + N14 + N15 + N16 ,
where N1 − N16 are given as follows:
u u 3
δ− 2
Z Z X 0 0 0
N1 = k∇2 ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 0
N2 = k∇ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 0
N3 = kΨ(s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
u u 3
δ− 2
Z Z X 0 0 0
N4 = kψ (s1 ,s1 ) · ∇2 Ψ(s2 ,s2 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 1 0 1
N5 = k∇2 ψ (s1 ,s1 ) · Ψ(s2 ,s2 ) · ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
157
u u 3
δ− 2
Z Z X 0 0 1 0 1
N6 = k∇ψ (s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
u u 3
δ− 2
Z Z X 0 0 1 0 1
N7 = kΨ(s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 | s1 +s2 =s+1,
s01 +s02 =s0 +1
u u 3
δ− 2
Z Z X 0 0 1 0 1
N8 = kψ (s1 ,s1 ) · ∇2 Ψ(s2 ,s2 ) · ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 1
2
u u 3
δ− 2
Z Z X 0 0 0 0
N9 = k∇ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
u u 3
δ− 2
Z Z X 0 0 0 0
N10 = kψ (s3 ,s3 ) · ψ (s4 ,s4 ) · ∇Ψ(s5 ,s5 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0 ,
0 u∞ |u0 |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
u u 3
δ− 2
Z Z X 0 0 0
N11 = k∇ψ (s3 ,s3 ) · ψ (s4 ,s4 ) · Ψ(s5 ,s5 )
0 u∞ |u0 | s3 +s4 +s5 =s+1,
s03 +s04 +s05 =s0 +1
1 0 1
× ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
u u 3
δ− 2
Z Z X 0 0 0
N12 = kψ (s3 ,s3 ) · ψ (s4 ,s4 ) · ∇Ψ(s5 ,s5 )
0 u∞ |u0 | s3 +s4 +s5 =s+1,
s03 +s04 +s05 =s0 +1
1 0 1
× ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0 ,
u u 3
δ− 2 δ
Z Z X 0 0 0
N13 = kψ (s6 ,s6 ) · ψ (s7 ,s7 ) · ψ (s8 ,s8 ) kL4sc (Su0 ,u0 )
0 u∞ |u0 | |u0 |2 s6 +s7 +s8 +s9 =s+1,
s06 +s07 +s08 +s09 =s0 +2
0 0
× kΨ(s9 ,s9 ) kL4sc (Su0 ,u0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 ,
Z u Z u −3
δ 2 δ X 0 0 0
N14 = 0 | |u0 |2
kψ (s6 ,s6 ) · ψ (s7 ,s7 ) · ψ (s8 ,s8 ) kL4sc (Su0 ,u0 )
0 u∞ |u 3
s6 +s7 +s8 +s9 =s+ ,
2
s06 +s07 +s08 +s09 =s0 + 3
2
0 1 0 1
× kΨ(s9 ,s9 ) kL4sc (Su0 ,u0 ) k∇2 Ψ(s− 2 ,s + 2 ) kL2sc (Su0 ,u0 ) du0 du0
Z u Z u −3
δ 2 X 0 0 0
N15 = 0
kψ (s10 ,s10 ) · Ψ(s11 ,s11 ) · Ψ(s12 ,s12 )
0 u∞ |u | s +s +s =s+1, 10 11 12
s010 +s011 +s012 =s0 +2
0
× ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0
u u 3
δ− 2
Z Z X 0 0 0
N16 = kψ (s10 ,s10 ) · Ψ(s11 ,s11 ) · Ψ(s12 ,s12 )
0 u∞ |u0 |
s10 +s11 +s12 =s+ 3 ,
2
s010 +s011 +s012 =s0 + 3
2
158
1 0 1
× ∇2 Ψ(s− 2 ,s + 2 ) kL1sc (Su0 ,u0 ) du0 du0
We will bound these term by term.
In the same fashion as in previous subsection, with L4 Sobolev Embedding (Proposition 2.3.9)
and Hölder’s inequality, we have the following bounds.
We control N1 either by
u u 1
δ− 2
Z Z
0 0 0
N1 ≤ 0 3
k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
0 u∞ |u |
Z u 1 Z u
δ2 2 (s1 ,s01 ) 1 0
≤ k∇ ψ kL4sc (Su0 ,u0 ) δ − 2 kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 )
u∞ |u0 |3 0

1 0
−2
× δ k∇ Ψ 2 (s,s )
kL2sc (Su0 ,u0 ) du du0
0
1 1
δ4 2 2 2 2 δ4
≤ 1 Ck∇ β, ∇ ρ, ∇ σ, ∇ βkL2 (u∞ ,u) + C(I (0) + R + R)5
|u| 2 sc (H u ) |u|

1 0 1 0 1
× δ 4 kΨ(s2 ,s2 ) k 2 2 (0,u) k∇Ψ(s2 ,s2 ) k 2 2 (0,u)
Lsc (Hu ) Lsc (Hu )

1 0 0
+ δ 2 kΨ(s2 ,s2 ) kL2 (H
(0,u)
)
k∇2 Ψ(s,s ) kL2 (H (0,u) ) ,
sc u sc u
where we use Proposition 2.5.19, or by
u u 1
δ− 2
Z Z
0 0 0
N1 ≤ 0 |3
k∇2 ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
0 u∞ |u
Z u −1 Z u
δ 2 2 (s1 ,s01 ) 1 0
≤ 0
k∇ ψ k 4
Lsc (Su0 ,u0 ) 0
kΨ(s2 ,s2 ) kL4sc (Su0 ,u0 )
0 |u | u∞ |u |

1 0
× 0 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u |

1 1
≤ δ 2 Ck∇2 β, ∇2 ρ, ∇2 σ, ∇2 βkL2 (H (0,u) ) + δ 2 C(I (0) + R + R)5
sc u

0 1 0 1
× kΨ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u)
Lsc (H u ) Lsc (H u )

1 0 0
+ δ 4 kΨ(s2 ,s2 ) kL2 (H (u∞ ,u) ) k∇2 Ψ(s,s ) kL2 (H (u∞ ,u) ) ,
sc u sc u
where we use Proposition 2.5.19.
Anomalous terms (∇2 χ, ∇2 ω, ∇2 η, ∇2 η)α∇2 Ψ, ∇2 χ̂(β, ρ, σ)∇2 Ψ and (∇2 χ, ∇2 ω)α∇2 Ψ of N1
159
type can be treated through the estimate above.
For N2 , we derive
1
δ2 X 1 0 0 1
N2 ≤ sup k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇Ψ(s2 ,s2 ) k 2 2 (0,u)
|u| s1 +s2 =s, u0 u0 |u0 | Lsc (Hu )
s01 +s02 =s0 +1
0 1 0
× k∇2 Ψ(s2 ,s2 ) k 2 2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u)
Lsc (Hu ) sc (Hu )
3
δ4 X 1 0
+ sup k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
|u| s1 +s2 =s, u0 u0 |u0 |
s01 +s02 =s0 +1
0 0
× k∇Ψ(s2 ,s2 ) kL2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u) ,
sc (Hu ) sc (Hu )
or
1
X 1 0 0 1
N2 ≤δ 2 sup 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u)
s1 +s2 =s, u0 u0 |u | L sc (H u )
s01 +s02 =s0 +1
0 1 0
× k∇2 Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u)
3
X 1 0
+ δ4 sup 0
k∇ψ (s1 ,s1 ) kL4sc (Su0 ,u0 )
s1 +s2 =s, u0 u0 |u |
s01 +s02 =s0 +1
0 0
× k∇Ψ(s2 ,s2 ) kL2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u ) sc (H u )
The anomalous term of N2 type is ∇χ̂(∇α, ∇β, ∇ρ, ∇σ, ∇β)∇2 Ψ and it is bounded through the
estimate above.
We bound N3 through
u u 3
δ− 2
Z Z X 0 0 0
N3 = kΨ(s1 ,s1 ) · ∇Ψ(s2 ,s2 ) · ∇2 Ψ(s,s ) kL1sc (Su0 ,u0 ) du0 du0
0 u∞ |u0 |
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
u u 1
δ− 2
Z Z X 0 0
≤ kΨ(s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇Ψ(s2 ,s2 ) kL4sc (Su0 ,u0 )
0 u∞ |u0 |3
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0
× k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
160
u u 1
δ− 2
Z Z
0 1 1
(s1 ,s01 )
X
≤ kΨ(s1 ,s1 ) kL2 2 k∇Ψ kL2 2
|u0 |3 sc (Su0 ,u0 ) sc (Su0 ,u0 )
0 u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× kΨ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc u ,u sc u ,u
Z u Z u −1
δ 2 X 1 0
+ 0 |3
δ 4 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× k∇Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇2 Ψ(s2 ,s2 ) kL2 2 (S 0 0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
sc u ,u sc u ,u
Z u Z u −1
δ 2 X 0 1 0 1
+ 0 3
kΨ(s1 ,s1 ) kL2 2 (S 0 0 ) k∇Ψ(s1 ,s1 ) kL2 2 (S 0 0 )
u∞ |u |
sc u ,u sc u ,u
0 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 0
× δ 4 k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z u Z u −1
δ 2 X 1 0
+ 0 3
δ 4 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1 s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 0
× δ 4 k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 .
Hence, we have
Z u Z u 1
δ2 X 1 0 1 1 0 1
N3 ≤ kΨ(s1 ,s1 ) kL2 2 k∇Ψ(s1 ,s1 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
0 u∞ |u0 |2 |u0 | 2 |u0 | 2
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× δ − 4 kΨ(s2 ,s2 ) kL2 2 δ − 4 k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
0
− 12
× δ k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z uZ u 3
δ4 X 1 0
+ 0 |2 0|
kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u 1
|u
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0 1
− 41
× δ − 4 k∇Ψ(s2 ,s2 ) kL2 2 δ k∇2 Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 0
× δ − 2 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
161
Z u Z u 3
δ4 X 1 0 1 1 0 1
+ kΨ(s1 ,s1 ) kL2 2 k∇Ψ(s1 ,s1 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
0 u∞ |u0 |2 |u0 | 2 |u0 | 2
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× δ − 2 k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) δ − 2 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
Z uZ u
δ X 1 0
+ 0 2
kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1
|u|
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× δ − 2 k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) δ − 2 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 .
Therefore, we deduce
1
δ2 X 0 1 0 1
N3 ≤ 3 kΨ(s1 ,s1 ) k 2 2 (u ,u) k∇Ψ(s1 ,s1 ) k 2 2 (u∞ ,u)
|u| 2 Lsc (H u ∞ ) Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× k∇Ψ(s2 ,s2 ) k 2 2 (0,u) k∇2 Ψ(s2 ,s2 ) k 2 2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u)
3
δ 4 X 0 1 0 1
+ 3 kΨ(s1 ,s1 ) k 2 2 (u∞ ,u) k∇Ψ(s1 ,s1 ) k 2 2 (u∞ ,u)
|u| 2 Lsc (H u ) Lsc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 0
× k∇Ψ(s2 ,s2 ) kL2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u)
sc (Hu ) sc (Hu )
3
δ4 X 0
+ 3 kΨ(s1 ,s1 ) kL2 (u∞ ,u)
|u| 2 sc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× k∇Ψ(s2 ,s2 ) k 2 2 (0,u) k∇2 Ψ(s2 ,s2 ) k 2 2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u)
δ X 0 0 0
+ 3 kΨ(s1 ,s1 ) kL2 (u∞ ,u) k∇Ψ(s2 ,s2 ) kL2 (0,u) k∇2 Ψ(s,s ) kL2 (0,u) .
|u| 2 sc (H u ) sc (Hu ) sc (Hu )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
162
Alternatively, we get
Z u Z u
1 X 1 0 1
− 41 0 1
N3 ≤ δ − 4 kΨ(s1 ,s1 ) kL2 2 δ k∇Ψ(s1 ,s1 ) kL2 2
|u0 | sc (Su0 ,u0 ) sc (Su0 ,u0 )
0 u∞ s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× kΨ(s2 ,s2 ) kL2 2 k∇Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
|u0 | 2 |u0 | 2
1 0
× 0
k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u |
Z uZ u 1
δ4 X 1 0
+ 0
δ − 2 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u | 1
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 1 0 1
× k∇Ψ(s2 ,s2 ) kL2 2 k∇2 Ψ(s2 ,s2 ) kL2 2
sc (Su0 ,u0 ) sc (Su0 ,u0 )
1 1
|u0 | 2 |u0 | 2
1 0
× 0
k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u |
Z uZ u 1
δ4 X
− 14 (s1 ,s01 ) 2
1
− 14 (s1 ,s01 ) 2
1
+ 0
δ kΨ kL 2 (S 0 0 ) δ k∇Ψ k L 2 (S 0 0 )
u∞ |u |
sc u ,u sc u ,u
0 1
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) 0 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0
|u0 | |u |
Z uZ u 1
δ2 X 1 0
+ 0|
δ − 2 kΨ(s1 ,s1 ) kL2sc (Su0 ,u0 )
0 u∞ |u 1
s1 +s2 =s+ ,
2
s01 +s02 =s0 + 3
2
1 0 1 0
× k∇Ψ(s2 ,s2 ) kL2sc (Su0 ,u0 ) 0 k∇2 Ψ(s,s ) kL2sc (Su0 ,u0 ) du0 du0 .
|u0 | |u |
163
Therefore, we obtain
1
δ2 X 0 1 0 1
N3 ≤ kΨ(s1 ,s1 ) k 2 2 (0,u) k∇Ψ(s1 ,s1 ) k 2 2 (0,u)
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× kΨ(s2 ,s2 ) k 2 2 (u ,u) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u)
3
δ4 X 0 1 0 1
+ kΨ(s1 ,s1 ) k 2 2 (0,u) k∇Ψ(s1 ,s1 ) k 2 2 (0,u)
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 0
× kΨ(s2 ,s2 ) kL2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u)
sc (H u ) sc (H u )
3
δ 4 X 0
+ kΨ(s1 ,s1 ) kL2 (0,u)
|u| sc (Hu )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
0 1 0 1 0
× kΨ(s2 ,s2 ) k 2 2 (u ,u) k∇Ψ(s2 ,s2 ) k 2 2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u)
1
δ2 X 1 0 0 0
+ δ 2 kΨ(s1 ,s1 ) kL2 (0,u) k∇Ψ(s2 ,s2 ) kL2 (u∞ ,u) k∇2 Ψ(s,s ) kL2 (u∞ ,u) .
|u| sc (Hu ) sc (H u ) sc (H u )
s1 +s2 =s+ 1 ,
2
s01 +s02 =s0 + 3
2
The anomalous term of N3 type is ∇βα∇2 Ψ, and it is bounded through the estimate above.
The term N4 is controlled through
1
δ2 X 1 0 0
N4 ≤ sup 0
kψ (s1 ,s1 ) kL∞
sc (Su0 ,u0 )
k∇2 Ψ(s2 ,s2 ) kL2 (H (0,u) )
|u| s1 +s2 =s, u0 u0 |u | sc u
s01 +s02 =s0 +1
0
× k∇2 Ψ(s,s ) kL2 (0,u) ,
sc (Hu )
or
1
X 1 0 0
N4 ≤δ 2 sup kψ (s1 ,s1 ) kL4sc (Su0 ,u0 ) k∇2 Ψ(s2 ,s2 ) kL2 (H (u∞ ,u) )
s1 +s2 =s, u0 u0 |u0 | sc u
s01 +s02 =s0 +1
0
× k∇2 Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u )
The anomalous term of N4 type is χ̂(∇2 α, ∇2 β, ∇2 ρ, ∇2 σ, ∇2 β)∇2 Ψ, which is bounded through the
estimate above.
164
N5 , N6 , N7 and N8 are treated in the same manner as for N1 , N2 , N3 and N4 , respectively.
In N9 − N14 , χ appears at most once in a term. Hence we derive the following estimates. For
N9 we have
1
1
X δ2 0 1 0
N9 ≤δ 4 sup 0 2 kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 ) 0|
k∇ψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
0
uu 0 |u | |u
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
1 0 1 0 1 0
× δ 4 kΨ(s5 ,s5 ) k 2 2 (0,u) k∇Ψ(s5 ,s5 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
1 1
δ4 X δ2 0 1 0
+ sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 ) 0
k∇ψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
|u| u0 u0 |u | |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
1 0 0
sc (Hu ) sc (Hu )
or
1
1
X δ2 0 0 0 1
N9 ≤δ 2 sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
k∇ψ (s4 ,s4 ) kL4sc (Su0 ,u0 ) kΨ(s5 ,s5 ) k 2 2 (u∞ ,u)
u0 u0 |u | Lsc (H u )
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
0 1 0
1
3
X δ2 0 0
+ δ4 sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
k∇ψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
u0 u0 |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
0 0
sc (H u ) sc (H u )
The anomalous terms of N9 type are
(χ, η, η, ω)(∇χ, ∇ω, ∇η, ∇η)α∇2 Ψ,
(∇η, ∇η)χ(α, β, ρ, σ, β)∇2 Ψ,
and
(η, η)∇χ̂(α, β, ρ, σ, β)∇2 Ψ.
165
All of them are bounded through the estimate above.
We bound N10 by
1 1
δ2 X δ2 0 0
N10 ≤ sup 0 2 kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
|u| 0
uu 0 |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
0 1 0 1 0
× k∇Ψ(s5 ,s5 ) k 2 2 (0,u) k∇2 Ψ(s5 ,s5 ) k 2 2 (0,u) k∇Ψ(s,s ) kL2 (0,u)
3 1
δ4 X δ2 0 0
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
|u| u0 u0 |u0 |2
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 3
2
0 0
× k∇Ψ(s5 ,s5 ) kL2 (0,u) k∇Ψ(s,s ) kL2 (0,u) ,
sc (Hu ) sc (Hu )
or
1
1
X δ2 0 0
N10 ≤δ 2 sup kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
u0 u0 |u0 |2
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 1
2
0 1 0 1 0
× k∇Ψ(s5 ,s5 ) k 2 2 (u∞ ,u) k∇2 Ψ(s5 ,s5 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
1
3
X δ2 0 0
+ δ4 sup 0 2
kψ (s3 ,s3 ) kL∞
sc (Su0 ,u0 )
kψ (s4 ,s4 ) kL4sc (Su0 ,u0 )
u0 u0 |u |
s3 +s4 +s5 =s+ 1 ,
2
s03 +s04 +s05 =s0 + 1
2
0 0
× k∇Ψ(s5 ,s5 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u ) sc (H u )
The anomalous term of N10 type is (η, η)χ(∇α, ∇β, ∇ρ, ∇σ, ∇β)∇2 Ψ and it is bounded through
the estimate above.
The terms N11 and N12 are bounded in the same manner as for N9 and N10 , respectively.
166
The term N13 obeys
1
δ X δ2 0
N13 ≤ 2 sup 0 2 kψ (s6 ,s6 ) kL∞
sc (Su0 ,u0 )
|u| s6 +s7 +s8 +s9 =s+1,
0
uu 0 |u |
s06 +s07 +s08 +s09 =s0 +2
0 0
× kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
kψ (s8 ,s8 ) kL4sc (Su0 ,u0 )
0 1 0 1 0
5 1
δ4 X δ2 0 0
+ sup 0
kψ (s6 ,s6 ) kL∞
sc (Su0 ,u0 )
kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
|u|2 s6 +s7 +s8 +s9 =s+1, u0 u0 |u |2
s06 +s07 +s08 +s09 =s0 +2
0 0 0
× kψ (s8 ,s8 ) kL4sc (Su0 ,u0 ) kΨ(s9 ,s9 ) kL2 (0,u) k∇Ψ(s,s ) kL2 (0,u) ,
sc (Hu ) sc (Hu )
or
1
δ X δ2 0
N13 ≤ sup 0 2
kψ (s6 ,s6 ) kL∞
sc (Su0 ,u0 )
|u| s6 +s7 +s8 +s9 =s+1, u0 u0 |u |
s06 +s07 +s08 +s09 =s0 +2
0 0
× kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
kψ (s8 ,s8 ) kL4sc (Su0 ,u0 )
0 1 0 1 0
× kΨ(s9 ,s9 ) k 2 2 (u∞ ,u) k∇Ψ(s9 ,s9 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
5 1
δ4 X δ 2 0 0
sc (Su0 ,u0 )
kψ (s7 ,s7 ) kL∞
sc (Su0 ,u0 )
|u| s6 +s7 +s8 +s9 =s+1, u0 u0 |u0 |2
s06 +s07 +s08 +s09 =s0 +2
0 0 0
× kψ (s8 ,s8 ) kL4sc (Su0 ,u0 ) kΨ(s9 ,s9 ) kL2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u) .
sc (H u ) sc (H u )
The anomalous terms of N13 type are (η, η)(η, η)(η, η, χ, ω)α and (η, η)(η, η)χ(β, ρ, σ)∇2 Ψ, which
are bounded through the estimate above.
We use the same method to control N14 .
For N15 , since there is no ψαα, ψαα or ψα α term. Employing Proposition 2.5.2, one of the
167
curvature components can be bounded in k · kL4sc (Su,u ) norm, hence we derive
1 1
δ4 X δ2 0 1 0
N15 ≤ sup 0 2
kψ (s10 ,s10 ) kL∞
sc (Su0 ,u0 )
δ 4 kΨ(s11 ,s11 ) kL4sc (Su0 ,u0 )
|u| s10 +s11 +s12 =s+1, u0 u0 |u |
s010 +s011 +s012 =s+2
0 1 0 1 0
1 1
δ4 X δ2 0 0
+ sup 0 2
kψ (s10 ,s10 ) kL∞
sc (Su0 ,u0 )
kΨ(s11 ,s11 ) kL4sc (Su0 ,u0 )
|u| s10 +s11 +s12 =s+1, u0 u0 |u |
s010 +s011 +s012 =s+2
1 0 0
sc (Hu ) sc (Hu )
or
1
1
X δ2 0 0
N15 ≤δ 2 sup 0 2
kψ (s10 ,s10 ) kL∞
sc (Su0 ,u0 )
kΨ(s11 ,s11 ) kL4sc (Su0 ,u0 )
s10 +s11 +s12 =s+1, u0 u0 |u |
s010 +s011 +s012 =s+2
0 1 0 1 0
× kΨ(s12 ,s12 ) k 2 2 (u∞ ,u) k∇Ψ(s12 ,s12 ) k 2 2 (u∞ ,u) k∇Ψ(s,s ) kL2 (u∞ ,u)
1
3
X δ 2 0 0
+ δ4 sup kψ (s10 ,s10 ) kL∞
sc (Su0 ,u0 )
kΨ(s11 ,s11 ) kL4sc (Su0 ,u0 )
s10 +s11 +s12 =s+1, u0 u0 |u0 |2
s010 +s011 +s012 =s+2
0 0
sc (H u ) sc (H u )
The anomalous term of N15 type is (η, η)βα∇2 Ψ, and it is bounded through the estimate above.
Gathering all the estimates in this subsection, we conclude
Proposition 2.6.14. Under the assumption of Theorem 2.2.1 and bootstrap assumption (2.3.4),
1
R2 + R2 ≤ R(0) + δ 8 (I (0) + R(0) + R + R)5 .
Therefore, we arrive at
Proposition 2.6.15. Under the assumption of Theorem 2.2.1, we get
R + R ≤ C(R(0) + I (0) ),
168
1
Proof. By choosing δ, such that δ − 200 R holds, we have improved the bootstrap assumption
(2.3.4). The conclusion follows.
With the estimates in this section, we can prove another useful proposition.
Denote a small constant satisfying δ 1. Let Ĩ (0) = −1 I (0) . We have
Proposition 2.6.16. Under the assumption of Theorem 3.2.2, we obtain
R + R ≤ Ĩ (0) .
Proof. Multiplying −2 on both sides of all the estimates in this section, we derive
1 1
−2 R2 + −2 R2 ≤ −2 R20 + δ 4 −2 (I (0) )5 ≤ (Ĩ (0) )2 + δ 4 3 (Ĩ (0) )5 .
From the assumptions of Theorem 3.2.2, we conclude that Ĩ (0) = −1 I (0) ≤ C, where C is a
universal large constant. Times 2 on both sides, we deduce the desired estimate.
2.7 Formation of Trapped Surfaces
In the section, we will prove
Theorem 3.2.2 (Formation of Trapped Surfaces)
Given I (0) and c, there exist δ0 = δ0 (I (0) , c) sufficiently small, such that for 0 < δ < δ0 , with
initial data:
1
δ k δ 2 k∇k4 (|u∞ |∇)i u∞ χ̂∞ kL∞ (Su∞ ,u ) ≤ I (0)
P
• i≤5,k≤3 along u = u∞
1 1
δ 2 k(δ 2 |u∞ |∇)j χ̂∞ kL2 (Su∞ ,u ) ≤
P
• 2≤j≤7 along u = u∞
• Minkowski initial data along u = 0

Rδ
• 0
|u∞ |2 |χ̂|2 ≥ 4c for every direction along u = u∞
169
Then S−c,δ is a trapped surface.
Proof. We fix (θ1 , θ2 ) ∈ S 2 and consider the following null structure equation
1 1
∇3 χ̂ + trχχ̂ = ∇⊗η
b + 2ω χ̂ − trχχ̂ + η ⊗η.
b
2 2
Contracting this two tensor with χ̂, we have
1 1 1
∇3 |χ̂|2γ + trχ|χ̂|2γ − 2ω|χ̂|2γ = χ̂(∇⊗η
b − trχχ̂ + η ⊗η).
b
2 2 2
In coordinates, we obtain
1 ∂ ∂ 1 1 2
( + bA A )|χ̂|2γ − |χ̂|2 + (trχ + )|χ̂|2γ − 2ω|χ̂|2γ
2Ω ∂u ∂θ Ω|u| γ 2 Ω|u|
1
=χ̂(∇⊗ηb − trχχ̂ + η ⊗η).
b
2
Employing the fact ω = − 12 ∇3 (log Ω), we get
Z u Z u
2 ∂ 0 0 −2 2
Ω exp(− Ωtrχdu ) exp(− Ωtrχdu )Ω |χ̂|γ
1 ∂u 1
∂ ∂Ω
= − bA A |χ̂|2γ − bA A + 2Ωχ̂(∇⊗η b − Ωtrχχ̂ + 2Ωη ⊗η).
b
∂θ ∂θ
Using the identity
2 2 2 2 2
Ωtrχ = Ω(trχ + )−Ω = Ω(trχ + ) − (Ω − 1) −
|u| |u| |u| |u| |u|
and the bounds for Ω and trχ in Propositions 2.3.1 and 2.4.13, we deduce

∂ ∂ A ∂Ω
(Ω−1 u2 |χ̂|2γ ) ≤ 2u2 −b A 2
|χ̂| − b + 2Ωχ̂(∇⊗η − Ωtrχχ̂ + 2Ωη ⊗η) ,
∂θA γ
b b
∂u ∂θA
when b is sufficiently large.
170
For bA , we have equation
∂bA
= −4Ω2 ζ A ,
∂u
which is from
∂bA ∂
[L, L] = .
∂u ∂θA
Applying the bounds for Ω, η and η in Propositions 2.3.1 and 2.4.18 implies in Du,u
δ
kbA kL∞ (Su,u ) ≤ .
|u|2
For ∇χ̂, we have
1 1 1
k∇χ̂kL∞ (Su,u ) ≤k∇χ̂kL2 4 (Su,u ) k∇2 χ̂kL2 4 (S(u,u)) + 1 k∇χ̂kL4 (Su,u )
|u| 2
3 5 3
δ− 4 1 δ− 4 1
(0) δ− 4
≤( 3 ) (
2
5 ) (I
2 )+ 3 I (0)
|u| 2 |u| 2 |u| 2
3
δ −1 (0) δ − 4 (0)
= 2
(I ) + 3 I .
|u| |u| 2
Hence, we derive
3
∂ δ −1 (0) δ − 4 (0)
k A χ̂kL∞ (Su,u ) ≤ |u|k∇χ̂kL∞ (Su,u ) ≤ (I ) + 1 I .
∂θ |u| |u| 2
For ∇η, appealing to L4 and L∞ Sobolev Embedding (Propositions 2.3.9 and 2.3.10), we obtain
1 1 1
k∇ηkL∞
sc (Su,u )
≤k∇ηkL2 4 k∇2 ηkL2 4 + δ 4 k∇ηkL4sc (Su,u )
2 1 1
≤k∇ηkL3 4 k∇3 ηkL3 2 (Su,u ) + δ 4 k∇ηkL4sc (Su,u ) .
sc (Su,u ) sc
From Proposition 2.5.12, we have
1
k∇3 ηkL2sc (Su,u ) ≤ k∇2 ρ, ∇2 σkL2sc (Su,u ) + δ 4 (I (0) )5 .
171
With this bound, we derive
Z u
u2 kχ̂kL∞ (Su0 ,u ) k∇ηkL∞ (Su0 ,u ) du0
u∞
Z u
δ −1
≤ kχ̂kL∞
sc (Su0 ,u )
k∇ηkL∞
sc (Su0 ,u )
du0
u∞ |u0 |2
u
δ −1
Z
2 1
≤ k χ̂k L∞ (S 0 ) k∇ηk 3
L 4 (S ) k∇3
ηkL
3
2 (S 0 du0
|u0 |2 sc u ,u sc u0 ,u sc u ,u )
u∞
u
δ −1
Z
1
+ 0 2
kχ̂kL∞
sc (Su0 ,u )
δ 4 k∇ηkL4sc (Su0 ,u ) du0
u∞ |u |
Z u −1
δ 2 1
≤ 0 2
kχ̂kL∞
sc (Su0 ,u )
k∇ηkL3 4 (S 0 ) k∇2 ρ, ∇2 σkL3 2 (S 0 ) du0
u∞ |u | sc u ,u sc u ,u
u
δ −1
Z
2 1 5
+ 0 2
kχ̂kL∞
sc (Su0 ,u )
k∇ηkL3 4 (S 0 ) δ 12 (I (0) ) 3 du0
u∞ |u | sc u ,u
u
δ −1
Z
1
+ kχ̂kL∞
sc (Su0 ,u )
δ 4 k∇ηkL4sc (Su0 ,u ) du0
u∞ |u0 |2
δ −1 2 1
≤ sup kχ̂kL∞
sc (Su,u )
k∇ηkL3 4 k∇2 ρ, ∇2 σk 3 2 (u∞ ,u)
sc (Su,u )
5
|u| 6 u,u Lsc (H u )
1
δ −1 δ 12 2 5
+ sup kχ̂kL∞
sc (Su,u )
k∇ηkL3 4 (I (0) ) 3
|u| sc (Su,u )
u,u
1
δ −1 δ 4
+ sup kχ̂kL∞
sc (Su,u )
k∇ηkL4sc (Su,u )
|u| u,u
δ −1 2 1 1
≤ sup kχ̂kL∞
sc (Su,u )
k∇ηkL3 4 3 (Ĩ (0) ) 3
sc (Su,u )
5
|u| 6 u,u
1
−1
δ δ 12 2 5
+ sup kχ̂kL∞
sc (Su,u )
k∇ηkL3 4 (Su,u ) (I (0) ) 3
|u| u,u sc
1
δ −1 δ 4
+ sup kχ̂kL∞
sc (Su,u )
k∇ηkL4sc (Su,u ) ,
|u| u,u
where we employ Proposition 2.6.16 in the last inequality.
Combining estimates in Proposition 2.4.18, for fixed (θ1 , θ2 ) ∈ S 2 we conclude
1 1
δ −1 3 1 δ −1 δ 12 (0) 5
Ω−1 |u|2 |χ̂|2γ (u, u, θ1 , θ2 ) ≥ |u∞ |2 |χ̂|2γ (u∞ , u, θ1 , θ2 ) − 5 (Ĩ (0) ) 3 − (I ) .
|u| 6 |u|
172
This yields
Z δ
Ω−1 |u|2 |χ̂|2γ (u, u0 , θ1 , θ2 )du0
0
Z δ 1 1
3 1 δ 12 (0) 5
≥ |u∞ |2 |χ̂|2γ (u∞ , u0 , θ1 , θ2 )du0 − 5 (Ĩ (0) ) 3 (I (0) )5 − (I ) ≥ 3c.
0 |u| 6 |u|
Considering another null structure equation
1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ,
2
and after employing ω = − 21 ∇4 (log Ω), we have
1
∇4 trχ + (trχ)2 = − |χ̂|2 − 2ωtrχ
2
= − |χ̂|2 + ∇4 (log Ω)trχ
1
= − |χ̂|2 + ∇4 Ωtrχ.
Ω
∇4 (Ω−1 trχ) = − Ω−2 ∇4 Ω · trχ + Ω−1 ∇4 trχ
=Ω−1 (∇4 trχ − Ω−1 ∇4 Ω · trχ)

1
=Ω−1 (− (trχ)2 − |χ̂|2 ).
2
Applying the fact e4 = Ω−1 ∂u

∂
, for every (θ1 , θ2 ) ∈ S 2 we conclude
Ω−1 c2 trχ(c, δ, θ1 , θ2 )
Z δ
≤Ω−1 c2 trχ(c, 0, θ1 , θ2 ) − Ω−1 c2 |χ̂|2 (c, u0 , θ1 , θ2 )du0
0
≤2c − 3c
<0.
173
Since in Du,u , the following inequality holds
1
2 δ2
ktrχ + kL∞ (Su,u ) ≤ .
|u| |u|2
Therefore S(c, δ) is a trapped surface.
Moreover, our proofs are independent of u∞ and we can prove
Theorem 2.7.1. (Christodoulou [7], 2008; A. [1], 2012) Trapped surfaces form in the Cauchy
development of Einstein vacuum equations with initial data which are arbitrarily dispersed at past
null infinity.
2.8 Retrieving Christodoulou’s Estimates
In this section, we claim that with the same initial data as in [7], we can retrieve the same bounds
for curvature components and null Ricci coefficients as in [7].
Comparing with the norms in [7], the norms we employ in previous subsections are weaker. So
far we have established a semi-global existence result in these weaker norms. With initial data in
[7], we will promote the estimates in weaker norms to stronger norms.
We demonstrate the ideas here: first, with the aid of scale invariant norms, we establish Theorem
2.2.1. As a byproduct, we have derived bounds for Ricci coefficients in Section 2.4. With these
bounds for Ricci coefficients and the new initial data in [7], we improve L2 bounds for curvature
components via the established controls in energy estimates in Section 2.6. Second, by employing
Sobolev Embedding, we improve L∞ control of curvature components. Third, with null structure
equations, we improve the bounds for Ricci coefficients to retrieve all the estimates in [7].
Initial data
The prescribed initial data along Hu∞ in [7] are demonstrated below:
1
ω = 0, (kωkL∞ (Su∞ ,u ) ≤ ),
|u∞ |2
174
1 1
δ− 2 2 1 δ2
kχ̂kL∞ (Su∞ ,u ) ≤ , ktrχ − kL∞ (Su∞ ,u ) ≤ , kη, ηkL∞ (Su∞ ,u ) ≤ ,
|u∞ | |u∞ | |u∞ |2 |u∞ |2
1
δ2 2 δ δ
kχ̂kL∞ (Su∞ ,u ) ≤ , ktrχ + kL∞ S(u∞ ,u) ≤ , kωkL∞ (Su∞ ,u ) ≤ ,
|u∞ |2 |u∞ | |u∞ |2 |u∞ |3
1 3 1
1 δ2 δ− 2 δ− 2
kK − kL∞ (Su∞ ,u ) ≤ , kαkL∞ (Su∞ ,u ) ≤ , kβkL∞ (Su∞ ,u ) ≤ ,
|u∞ |2 |u∞ |3 |u∞ | |u∞ |2
3
1 δ δ2
kρ, σkL∞ (Su∞ ,u ) ≤ , kβkL∞ (Su∞ ,u ) ≤ , kαkL∞ (Su∞ ,u ) ≤ .
|u∞ |3 |u∞ |4 |u∞ |5
Energy Estimates
The curvature norms R∗ and R∗ used in [7] are as below:
1 1
R∗0 + R∗0 :=δ 2 kαkL2 (0,u) + kβ, ρ, σkL2 (0,u) + δ − 2 kβkL2 (0,u)
sc (Hu ) sc (Hu ) sc (Hu )
1 1
+ δ 2 kβkL2 (u∞ ,u) + kρ, σ, βkL2 (u∞ ,u) + δ − 2 kαkL2 (u∞ ,u) ,
sc (H u ) sc (H u ) sc (H u )
1
R∗1 + R∗1 :=k∇αkL2 (0,u) + δ − 2 k∇β, ∇ρ, ∇σkL2 (0,u) + δ −1 k∇βkL2 (0,u)
sc (Hu ) sc (Hu ) sc (Hu )
1
+ k∇βkL2 (u∞ ,u) + δ − 2 k∇ρ, ∇σ, ∇βkL2 (u∞ ,u) + δ −1 k∇αkL2 (u∞ ,u) ,
sc (H u ) sc (H u ) sc (H u )
1
R∗2 + R∗2 :=δ − 2 k∇2 αkL2 (0,u) + δ −1 k∇2 β, ∇2 ρ, ∇2 σkL2 (0,u)
sc (Hu ) sc (Hu )
3 1
+ δ − 2 k∇2 βkL2 (0,u) + δ − 2 k∇2 βkL2 (u∞ ,u)
sc (Hu ) sc (H u )
3
+ δ −1 k∇2 ρ, ∇2 σ, ∇2 βkL2 (u∞ ,u) + δ − 2 k∇2 αkL2 (u∞ ,u) .
sc (H u ) sc (H u )
With the results in previous sections and these new norms, we claim
Proposition 2.8.1. Under the assumption of Theorem 2.2.1 and initial data in [7], we have
2 2
R∗0 2 + R∗0 ≤ (R∗ (0) ) + C(I (0) )20 ,
175
Proof. Since for

0 1 0 1
Ψ(s,s ) = α, Ψ(s− 2 ,s + 2 ) = β,
0 1 0 1
Ψ(s,s ) = β, Ψ(s− 2 ,s + 2 ) = ρ, σ,
0 1 0 1
Ψ(s,s ) = ρ, σ, Ψ(s− 2 ,s + 2 ) = β,
the desired estimates hold trivially. We are left to consider pairs
0 1 0 1
Ψ(s,s ) = β, Ψ(s− 2 ,s + 2 ) = α.
From Proposition 2.6.8, we have
kβk2L2 (0,u) + kαk2L2 (u∞ ,u)

sc (Hu ) sc (H u )
≤kβk2L2 (0,u) + kαk2L2 (u∞ ,u)

sc (Hu∞ ) sc (H 0 )
1 1
+δ 2 sup kχ̂, η, η, ωkL∞
sc (Su0 ,u0 )
kρ, σ, β, αkL2 (H (u∞ ,u) ) kβ, αkL2 (H (u∞ ,u) )
u0 ,u0 |u0 | sc u sc u
1
δ2 1 1
+ sup kχ̂, η, η, ωkL∞
sc (Su0 ,u0 )
kρ, σ, βkL2 (H (0,u) ) δ − 2 kβkL2 (H (0,u) ) .
|u| u0 ,u0 |u0 | sc u sc u
Multiplying δ −1 on both sides, we have
δ −1 kβk2L2 (0,u) + δ −1 kαk2L2 (u∞ ,u)

sc (Hu ) sc (H u )
≤δ −1 kβk2L2 (0,u) + δ −1 kαk2L2 (u∞ ,u)

sc (Hu∞ ) sc (H 0 )
1 1
+ sup kχ̂, η, η, ωkL∞
sc (Su0 ,u0 )
kρ, σ, β, αkL2 (H (u∞ ,u) ) δ − 2 kαkL2 (H (u∞ ,u) )
u0 ,u0 |u0 | sc u sc u
1 1 1
+ sup 0 kχ̂, η, η, ωkL∞
sc (Su0 ,u0 )
kρ, σ, βkL2 (H (0,u) ) δ − 2 kβkL2 (H (0,u) ) .
|u| u ,u |u |
0 0 sc u sc u
Thus, we obtain
R∗ [β]2 + R∗ [α]2 ≤ (R∗ (0) )2 + O0,∞ RR∗ [α] + O0,∞ RR∗ [β].
176
Using Hölder’s inequality, we arrive at
2 (0) 2
R∗0 2 + R∗0 ≤ (R∗ 0 ) + O0,∞
2
R2 + O0,∞
2
R2 .
Applying the conclusion in Proposition 2.6.15, we have proved desired estimate.
For R∗1 and R∗1 , we get
Proposition 2.8.2. Under the assumption of Theorem 2.2.1 and initial data in [7], the following
inequality holds
2 2
R∗1 2 + R∗1 ≤ (R∗ (0) ) + C(I (0) )20 ,
Proof.
2
For α, we have R∗ 21 [α] + R∗ 1 [β] = R21 [α] + R21 [β].
For β, ρ, σ, using estimates in Proposition 2.6.12 and similar methods in Proposition 2.8.1, we
derive
2 (0) 2
R∗1 2 [β, ρ, σ] + R∗1 [ρ, σ, β] ≤(R∗ 0 ) + C(I (0) + R + R)5 RR∗1 [ρ, σ, β]
+ C(I (0) + R + R)5 RR∗1 [β, ρ, σ].
With Hölder’s inequality, we obtain desired estimates.
Similarly, for β we deduce
2 (0) 2
R∗1 2 [β] + R∗1 [α] ≤(R∗ 0 ) + C(I (0) + R + R)5 R∗1 [ρ, σ, β]R∗1 [α]
+ C(I (0) + R + R)5 R∗1 [ρ, σ]R∗1 [β].
Since, we have already estimated R∗1 [ρ, σ] and R∗1 [ρ, σ, β]. After employing Hölder’s inequality and
Proposition 2.6.15, we finish the proof.
In the same fashion, we conclude
177
Proposition 2.8.3. Under the assumption of Theorem 2.2.1 and initial data in [7], we get
2 2
R∗2 2 + R∗2 ≤ (R∗ (0) ) + C(I (0) )20 ,
Coefficients Estimates
We define R∗0,∞ as
3 1
R∗0,∞ :=δ 2 |u|kαkL∞ (Su,u ) + δ 2 |u|2 kβkL∞ (Su,u ) + |u|3 kρ, σkL∞ (Su,u )
3
+ δ −1 |u|4 kβkL∞ (Su,u ) + δ − 2 |u|5 kαkL∞ (Su,u ) .
With the energy estimates obtained above, it can be verified that
R∗0,∞ ≤ C(I (0) )20 ,
where C is a universal large constant. With the aid of this bound, we improve several L∞ (Su,u )
estimates for Ricci coefficients.

1
δ2 (0) 20
For η, using ∇4 η = −χ · (η − η) − β, we have kηkL∞ (Su,u ) ≤ |u|2 C(I ) .
1
δ2 (0) 20
For η, with ∇3 η = −χ · (η − η) − β, we obtain kηkL∞ (Su,u ) ≤ |u|2 C(I ) .
For ω, employing ∇3 ω = 2ωω + (η, η)(η, η) + 21 ρ, we derive kωkL∞ (Su,u ) ≤ 1

|u|2 C(I
(0) 20
) .
For ω, using ∇4 ω = 2ωω + (η, η)(η, η) + 12 ρ, we conclude kωkL∞ (S) ≤ δ

|u|3 C(I
(0) 20
) .
In summary, starting with initial data in [7], by employing an alternative strategy, which is
outlined in this section, we arrive at
Theorem 2.8.4. Given Minkowskian initial data on H (u

u
∞ ,u)
and Christodoulou’s initial data on
(0,u)
Hu∞ as in [7], we can derive estimates that are consistent with [7].
178
2.9 Appendix
2.9.1 Norms in standard Lp form
For convenience, we list the norms in the standard Lp form:
1 1
O0,∞ (u, u) =δ 2 (|u|kχ̂kL∞ (Su,u ) + |u|kωkL∞ (Su,u ) ) + δ 2 |u|ktrχkL∞ (Su,u )
1
+ |u|2 kη, ηkL∞ (Su,u ) + δ − 2 |u|2 kχ̂kL∞ (Su,u )
1
+ |u|ktrχkL∞ (Su,u ) + δ − 2 |u|3 kωkL∞ (Su,u ) ,
1 1 1 1 1 1
O0,4 (u, u) =δ 2 |u| 2 kχ̂kL4 (Su,u ) + δ 4 |u| 2 kωkL4 (Su,u ) + δ 4 |u| 2 ktrχkL4 (Su,u )
1 3 1 3
+ δ − 4 |u| 2 kη, ηkL4 (Su,u ) + δ − 2 |u| 2 kχ̂kL4 (Su,u )
1 3 5
+ |u| 2 ktrχkL4 (Su,u ) + δ − 4 |u| 2 kωkL4 (Su,u ) ,
1 1
O0,2 (u, u) =δ 2 kχ̂kL2 (Su,u ) + kωkL2 (Su,u ) + ktrχkL2 (Su,u ) + δ − 2 |u|kη, ηkL2 (Su,u )
1
+ δ − 2 |u|kχ̂kL2 (Su,u ) + ktrχkL2 (Su,u ) + δ −1 |u|2 kωkL2 (Su,u ) ,
3 3 3 3 3 3
O1,4 (u, u) =δ 4 |u| 2 k∇χ̂kL4 (Su,u ) + δ 4 |u| 2 k∇trχkL4 (Su,u ) + δ 4 |u| 2 k∇ωkL4 (Su,u )
1 5 1 5
+ δ 4 |u| 2 k∇(η, η)kL4 (Su,u ) + δ − 4 |u| 2 k∇χ̂kL4 (Su,u )
1 7 1 7
+ δ − 4 |u| 2 k∇trχkL4 (Su,u ) + δ − 4 |u| 2 k∇ωkL4 (Su,u ) ,
1 1 1
O1,2 (u, u) =δ 2 |u|k∇χ̂kL2 (Su,u ) + δ 2 |u|k∇trχkL2 (Su,u ) + δ 2 |u|k∇ωkL2 (Su,u )
1
+ |u|2 k∇(η, η)kL2 (Su,u ) + δ − 2 |u|2 k∇χ̂kL2 (Su,u )
1 1
+ δ − 2 |u|3 k∇trχkL2 (Su,u ) + δ − 2 |u|3 k∇ωkL2 (Su,u ) ,
179
1
R0 (u, u) = δkαkL2 (H (0,u) ) + |u|kβkL2 (H (0,u) ) + δ − 2 |u|2 kρ, σkL2 (H (0,u) )
u u u
−1 3
+δ |u| kβkL2 (H (0,u) ) ,
u
1
R0 (u, u) =δkβkL2 (H (u∞ ,u) ) + k|u0 |(ρ, σ)kL2 (H (u∞ ,u) ) + δ − 2 k|u0 |2 βkL2 (H (u∞ ,u) )
u u u
+ δ −1 |k|u0 |3 αkL2 (H (u∞ ,u) ) ,

u
1
R1 (u, u) =δ|u|k∇αkL2 (H (0,u) ) + δ 2 |u|2 k∇βkL2 (H (0,u) ) + |u|3 k∇(ρ, σ)kL2 (H (0,u) )
u u u
− 12 4
+δ |u| k∇βkL2 (H (0,u) ) ,
u
1
R1 (u, u) =δk|u0 |∇βkL2 (H (u∞ ,u) ) + δ 2 k|u0 |2 ∇(ρ, σ)kL2 (H (u∞ ,u) ) + k|u0 |3 ∇βkL2 (H (u∞ ,u) )
u u u
− 21
+δ ||u0 |4 ∇αkL2 (H (u∞ ,u) ) .
u
2.9.2 Equations for Elliptic Estimates
In this section, we write down the transport equations for Θ ∈ {∇trχ, ∇trχ, µ, µ, κ, κ}.
For ∇trχ, we have
∇4 ∇trχ = (∇χ, ∇ω)(χ, ω) + (η, η)(ω, χ)(ω, χ),
∇4 ∇2 trχ
=(∇2 χ, ∇2 ω)(χ, ω) + (∇χ, ∇ω)(∇χ, ∇ω) + (∇η, ∇η)(ω, χ)(ω, χ)
+ (η, η)(∇ω, ∇χ)(ω, χ) + β∇trχ + (η, η)(η, η)(ω, χ)(ω, χ),
180
∇4 ∇3 trχ
=(∇3 χ, ∇3 ω)(χ, ω) + (∇2 χ, ∇2 ω)(∇χ, ∇ω) + (∇2 η, ∇2 η)(ω, χ)(ω, χ)
+ (∇η, ∇η)(∇ω, ∇χ)(ω, χ) + (η, η)(∇2 ω, ∇2 χ)(ω, χ) + ∇β∇trχ + β∇2 χ
+ (∇η, ∇η)(η, η)(ω, χ)(ω, χ) + (η, η)(η, η)(∇ω, ∇χ)(ω, χ) + β(η, η)∇trχ
+ (η, η)(∇χ, ∇ω)(∇χ, ∇ω) + (η, η)(η, η)(η, η)(ω, χ)(ω, χ).
For µ, we obtain
∇4 µ = χµ + χ(∇η, ∇η) + (η, η)∇χ + χ̂α + (η, η)β + χρ + χ(η, η)(η, η),
∇4 ∇µ
=χ∇µ + ∇χµ + ∇χ(∇η, ∇η) + χ(∇2 η, ∇2 η) + (η, η)∇2 χ + χ̂∇α
+ ∇χ̂α + (∇η, ∇η)β + (η, η)∇β + ∇χρ + χ∇ρ + ∇χ(η, η)(η, η)
+ χ(∇η, ∇η)(η, η) + (η, η)µχ + (η, η)χ̂α + (η, η)(η, η)β,
+ (η, η)χρ + χ(η, η)(η, η)(η, η)
181
∇4 ∇2 µ
=χ∇2 µ + ∇χ∇µ + ∇2 χµ + ∇2 χ(∇η, ∇η) + ∇χ(∇2 η, ∇2 η)
+ χ(∇3 η, ∇3 η) + ∇χ̂∇α + χ̂∇2 α + ∇2 χ̂α + (∇2 η, ∇2 η)β + (∇η, ∇η)∇β
+ (η, η)∇2 β + ∇2 χρ + ∇χ∇ρ + χ∇2 ρ + ∇2 χ(η, η)(η, η) + ∇χ(∇η, ∇η)(η, η)
+ χ(∇2 η, ∇2 η)(η, η) + (∇η, ∇η)µχ + (η, η)∇µχ + (η, η)µ∇χ + (η, η)(η, η)χ̂α
+ (∇η, ∇η)χ̂α + (η, η)∇χ̂α + (η, η)χ̂∇α + (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β
+ (∇η, ∇η)χρ + (η, η)∇χρ + (η, η)χ∇ρ + ∇χ(η, η)(η, η)(η, η) + χ(∇η, ∇η)(η, η)(η, η)
+ η∇µχ + β∇µ + (η, η)(η, η)(η, η)β + (η, η)(η, η)χρ + χ(η, η)(η, η)(η, η)(η, η).
For µ, we demonstrate
∇3 µ + trχ µ = trχ∇η + χ̂(∇η, ∇η) + (η, η)∇χ + χ̂α + (η, η)β + χρ + χ(η, η)(η, η),
3
∇3 ∇µ + trχ∇µ
2
=χ̂∇µ + trχ∇2 η + ∇χ(∇η, ∇η) + (η, η)∇2 χ
+ ∇χ̂α + χ̂∇α + (∇η, ∇η)β + (η, η)∇β + ∇χρ + χ∇ρ + ∇χ(η, η)(η, η)
+ χ(∇η, ∇η)(η, η) + (η, η)µ χ + (η, η)χ̂α + ∇χ̂(∇η, ∇η) + χ̂(∇2 η, ∇2 η)
+ (η, η)(η, η)β + χ(η, η)ρ + χ(η, η)(η, η)(η, η) + (η, η)trχ∇η + (η, η)(η, η)∇χ,
182
∇3 ∇2 µ + 2trχ∇2 µ
=χ̂∇2 µ + trχ∇3 η + ∇χ(∇2 η, ∇2 η) + ∇2 χ(∇η, ∇η)
+ (η, η)∇3 χ + ∇χ̂∇α + ∇2 χ̂α + χ̂∇2 α + (∇2 η, ∇2 η)β + (∇η, ∇η)∇β + (η, η)∇2 β
+ ∇2 χ̂ρ + ∇χ∇ρ + χ∇2 ρ + ∇2 χ̂(η, η)(η, η) + ∇χ(∇η, ∇η)(η, η) + (η, η)trχ∇2 η
+ χ(∇2 η, ∇2 η)(η, η) + (∇η, ∇η)µ χ + (η, η)∇µ χ + (η, η)µ∇χ + (η, η)(η, η)trχ∇η
+ (∇η, ∇η)χ̂α + (η, η)∇χ̂α + (η, η)χ̂∇α + (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β + ∇χ(η, η)ρ
+ χ(∇η, ∇η)ρ + χ(η, η)∇ρ + ∇χ(η, η)(η, η)(η, η) + χ(∇η, ∇η)(η, η)(η, η) + β∇µ
+ (η, η)(η, η)µ χ + (η, η)(η, η)χ̂α + (η, η)(η, η)(η, η)β
+ χ(η, η)(η, η)ρ + χ(η, η)(η, η)(η, η)(η, η).
For κ, we derive
∇4 κ =χκ + ωβ + χ̂β + (η, η)(ρ, σ) + χ(∇ω, ∇ω † ) + ∇ωω + ω∇ω + (∇η, ∇η)(η, η)
+ (η, η)ωω + (η, η)(η, η)(η, η),
∇4 ∇κ
=χ∇κ + ∇χκ + ∇ωβ + ω∇β + ∇χ̂β + χ̂∇β
+ (∇η, ∇η)(ρ, σ) + (η, η)(∇ρ, ∇σ) + ∇χ̂(∇ω, ∇ω † ) + χ̂(∇2 ω, ∇2 ω † )
+ ∇2 ωω + ∇ω∇ω + ω∇2 ω + (∇2 η, ∇2 η)(η, η) + (∇η, ∇η)(∇η, ∇η)
+ (∇η, ∇η)ωω + (η, η)∇ωω + (η, η)ω∇ω + (∇η, ∇η)(η, η)(η, η) + (η, η)κχ + βκ
+ (η, η)ωβ + (η, η)χ̂β + (η, η)(η, η)(ρ, σ) + (η, η)χ̂(∇ω, ∇ω † ) + (η, η)(η, η)ωω
+ (η, η)(η, η)(η, η)(η, η),
183
∇4 ∇2 κ
=χ∇2 κ + ∇χ∇κ + ∇2 χκ + ∇2 ωβ
+ ∇ω∇β + ω∇2 β + ∇2 χ̂β + ∇χ̂∇β + χ̂∇2 β + (∇2 η, ∇2 η)(ρ, σ)
+ (∇η, ∇η)(∇ρ, ∇σ) + (η, η)(∇2 ρ, ∇2 σ) + ∇2 χ̂(∇ω, ∇ω † )
+ ∇χ̂(∇2 ω, ∇2 ω † ) + χ̂(∇3 ω, ∇3 ω † ) + ∇3 ωω + ∇2 ω∇ω + ∇ω∇2 ω
+ ω∇3 ω + (∇3 η, ∇3 η)(η, η) + (∇2 η, ∇2 η)(∇η, ∇η) + (∇2 η, ∇2 η)ωω
+ (∇η, ∇η)∇ωω + (∇η, ∇η)ω∇ω + (η, η)∇2 ωω + (η, η)ω∇2 ω + (∇2 η, ∇2 η)(η, η)(η, η)
+ (∇η, ∇η)(∇η, ∇η)(η, η) + (∇η, ∇η)κχ + (η, η)∇κχ + (η, η)∇ω∇ω
+ (η, η)κ∇χ + ∇βκ + β∇κ + (∇η, ∇η)ωβ
+ (η, η)∇ωβ + (η, η)ω∇β + (∇η, ∇η)χ̂β + (η, η)∇χ̂β + (η, η)χ̂∇β + (∇η, ∇η)(η, η)(ρ, σ)
+ (η, η)(η, η)(∇ρ, ∇σ) + (∇η, ∇η)χ̂(∇ω, ∇ω † ) + (η, η)∇χ̂(∇ω, ∇ω † )
+ (η, η)χ̂(∇2 ω, ∇2 ω † ) + (∇η, ∇η)(η, η)ωω + (η, η)(η, η)∇ωω + (η, η)(η, η)ω∇ω
+ (∇η, ∇η)(η, η)(η, η)(η, η) + (η, η)∇κχ + β∇κ + (η, η)(η, η)κχ
+ (η, η)βκ + (η, η)(η, η)ωβ + (η, η)χ̂β + (η, η)(η, η)(η, η)(ρ, σ) + (η, η)(η, η)χ̂(∇ω, ∇ω † )
+ (η, η)(η, η)(η, η)ωω + (η, η)(η, η)(η, η)(η, η)(η, η).
For κ, we have
1
∇3 κ + trχκ =ωβ + χ̂β + (η, η)(ρ, σ) + χ̂(∇ω, ∇ω † ) + ∇ωω + ω∇ω
2
+ (∇η, ∇η)(η, η) + (η, η)ωω + (η, η)(η, η)(η, η),
184
∇3 ∇κ + trχ∇κ
=χ̂∇κ + ∇trχκ + ∇ωβ + ω∇β + ∇χ̂β + χ̂∇β + (η, η)(η, η)(ρ, σ)
+ (∇η, ∇η)(ρ, σ) + (η, η)(∇ρ, ∇σ) + ∇χ̂(∇ω, ∇ω † ) + χ̂(∇2 ω, ∇2 ω † )
+ ∇2 ωω + ∇ω∇ω + ω∇2 ω + (∇2 η, ∇2 η)(η, η) + (∇η, ∇η)(∇η, ∇η)
+ (∇η, ∇η)ωω + (η, η)∇ωω + (η, η)ω∇ω + (∇η, ∇η)(η, η)(η, η) + (η, η)κχ + βκ
+ (η, η)ωβ + (η, η)χ̂β + (η, η)χ̂(∇ω, ∇ω † ) + (η, η)(η, η)ωω + (η, η)(η, η)(η, η)(η, η),
3
∇3 ∇2 κ + trχ∇2 κ
2
=χ̂∇2 κ + ∇χ̂∇κ + ∇2 trχκ + ∇trχ∇κ
+ ∇2 ωβ + ∇ω∇β + ω∇2 β + ∇2 χ̂β + ∇χ̂∇β + χ̂∇2 β + (∇2 η, ∇2 η)(ρ, σ)
+ (∇η, ∇η)(∇ρ, ∇σ) + (η, η)(∇2 ρ, ∇2 σ) + ∇2 χ̂(∇ω, ∇ω † )
+ ∇χ̂(∇2 ω, ∇2 ω † ) + χ̂(∇3 ω, ∇3 ω † ) + ∇3 ωω + ∇2 ω∇ω
+ ∇ω∇2 ω + ω∇3 ω + (∇3 η, ∇3 η)(η, η) + (∇2 η, ∇2 η)(∇η, ∇η)
+ (∇2 η, ∇2 η)ωω + (∇η, ∇η)∇ωω + (∇η, ∇η)ω∇ω + (η, η)∇2 ωω + (η, η)ω∇2 ω
+ (η, η)∇ω∇ω + (∇2 η, ∇2 η)(η, η)(η, η) + (∇η, ∇η)(∇η, ∇η)(η, η) + (∇η, ∇η)κχ
+ (η, η)∇κχ + (η, η)κ∇χ + ∇βκ + β∇κ + (∇η, ∇η)ωβ + (η, η)∇ωβ
+ (η, η)ω∇β + (∇η, ∇η)χ̂β + (η, η)∇χ̂β + (η, η)χ̂∇β
+ (∇η, ∇η)χ̂(∇ω, ∇ω † ) + (η, η)∇χ̂(∇ω, ∇ω † ) + (η, η)χ̂(∇2 ω, ∇2 ω † )
185
+ (∇η, ∇η)(η, η)ωω + (η, η)(η, η)∇ωω + (η, η)(η, η)ω∇ω
+ (∇η, ∇η)(η, η)(η, η)(η, η) + (η, η)(∇η, ∇η)(ρ, σ)
+ (η, η)(η, η)(∇ρ, ∇σ) + (η, η)(η, η)κχ + (η, η)βκ + (η, η)(η, η)ωβ
+ (η, η)(η, η)χ̂β + (η, η)(η, η)χ̂(∇ω, ∇ω † ) + (η, η)(η, η)(η, η)ωω
+ (η, η)(η, η)(η, η)(η, η)(η, η).
For Ωtrχ − u2 , we obtain
2 2 Ω−1 2
∇3 (Ωtrχ − ) + trχ(Ωtrχ − ) = −4Ωωtrχ − Ω|χ̂|2 + (Ωtrχ − )2 ,
u u 2 u
2 3 2
∇3 ∇(Ωtrχ − ) + trχ∇(Ωtrχ − )
u 2 u
2 2
=χ̂∇(Ωtrχ − ) + ∇trχ(Ωtrχ − ) + ∇Ωωtrχ + Ω∇ωtrχ + Ωω∇trχ
u u
2 2 2
+ ∇Ωχ̂ χ̂ + Ωχ̂∇χ̂ + ∇ΩΩ−2 (Ωtrχ − )2 + Ω−1 ∇(Ωtrχ − )(Ωtrχ − )
u u u
2 −1 2
+ (η, η)trχ(Ωtrχ − ) + (η, η)Ωωtrχ + (η, η)Ωχ̂ χ̂ + (η, η)Ω (Ωtrχ − )2 ,
u u
186
2 2
∇3 ∇2 (Ωtrχ − ) + 2trχ∇2 (Ωtrχ − )
u u
2 2 2 2
=χ̂∇2 (Ωtrχ − ) + ∇χ̂∇(Ωtrχ − ) + ∇2 trχ(Ωtrχ − ) + ∇trχ∇(Ωtrχ − )
u u u u
+ ∇2 Ωωtrχ + ∇Ω∇ωtrχ + ∇Ωω∇trχ + Ω∇2 ωtrχ + Ω∇ω∇trχ
2 2
+ Ωω∇2 trχ + ∇2 Ωχ̂ χ̂ + ∇Ω∇χ χ̂ + Ωχ̂∇2 χ̂ + ∇2 ΩΩ−2 (Ωtrχ − )
u
2 2 2 2
+ ∇Ω∇ΩΩ−3 (Ωtrχ − ) + ∇ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − )
u u u
−1 2 2 2 2
+ Ω ∇ (Ωtrχ − )(Ωtrχ − ) + (∇η, ∇η)trχ(Ωtrχ − )
u u u
2 −1 2 2
+ (η, η)∇trχ(Ωtrχ − ) + Ω ∇(Ωtrχ − )∇(Ωtrχ − )
u u u
2
+ (η, η)trχ∇(Ωtrχ − ) + (∇η, ∇η)Ωωtrχ + (η, η)∇Ωωtrχ
u
+ (η, η)Ω∇ωtrχ + (η, η)Ωω∇trχ + (∇η, ∇η)Ωχ̂ χ̂ + (η, η)∇Ωχ̂ χ̂
2 2 2
+ (η, η)Ωχ̂∇χ̂ + (∇η, ∇η)Ω−1 (Ωtrχ − ) + (η, η)∇ΩΩ−2 (Ωtrχ − )2
u u
2 2 2 2
+ (η, η)Ω−1 ∇(Ωtrχ − )(Ωtrχ − ) + (η, η)χ∇(Ωtrχ − ) + β∇(Ωtrχ − )
u u u u
2 2
+ (η, η)χ̂∇(Ωtrχ − ) + (η, η)∇trχ(Ωtrχ − ) + (η, η)∇Ωωtrχ
u u
+ (η, η)Ω∇ωtrχ + (η, η)Ωω∇trχ + (η, η)∇Ωχ̂ χ̂ + (η, η)Ωχ̂∇χ̂
2 2 2
+ (η, η)∇ΩΩ−2 (Ωtrχ − ) + (η, η)(η, η)trχ(Ωtrχ − ) + (η, η)(η, η)Ωωtrχ
u u
2
+ (η, η)(η, η)Ωχ̂ χ̂ + (η, η)(η, η)Ω−1 (Ωtrχ − )2 ,
u
187
2 5 2
∇3 ∇3 (Ωtrχ − ) + trχ∇3 (Ωtrχ − )
u 2 u
2 2 2
=χ̂∇3 (Ωtrχ − ) + ∇2 χ̂∇(Ωtrχ − ) + ∇χ̂∇2 (Ωtrχ − )
u u u
3 2 2 2
+ ∇ trχ(Ωtrχ − ) + ∇ trχ∇(Ωtrχ − )
u u
2
+ ∇trχ∇ (Ωtrχ − ) + ∇ Ωωtrχ + ∇ Ω∇ωtrχ + ∇2 Ωω∇trχ + ∇Ω∇2 ωtrχ
2 3 2
u
+ ∇Ω∇ω∇trχ + ∇Ωω∇2 trχ + Ω∇3 ωtrχ + Ω∇2 ω∇trχ + Ω∇ω∇2 trχ
+ Ωω∇3 trχ + ∇3 Ωχ̂ χ̂ + ∇2 Ω∇χ̂ χ̂ + ∇Ω∇2 χ̂ χ̂ + ∇Ω∇χ̂∇χ̂

2 2
+ Ω∇χ̂∇2 χ̂ + Ωχ̂∇3 χ̂ + ∇3 ΩΩ−2 (Ωtrχ − )
u
2 2 2 2
+ ∇2 Ω∇ΩΩ−3 (Ωtrχ − ) + ∇2 ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − )
u u u
2 −3 2 2 −4 2 2
+ ∇ Ω∇ΩΩ (Ωtrχ − ) + ∇Ω∇Ω∇ΩΩ (Ωtrχ − )
u u
−3 2 2 2 −2 2 2
+ ∇Ω∇ΩΩ ∇(Ωtrχ − )(Ωtrχ − ) + ∇ ΩΩ ∇(Ωtrχ − )(Ωtrχ − )
u u u u
2 2 2
+ ∇ΩΩ−2 ∇2 (Ωtrχ − )(Ωtrχ − ) + (η, η)∇2 ΩΩ−2 (Ωtrχ − )2
u u u
2 2 2 2
+ ∇ΩΩ−2 ∇(Ωtrχ − )∇(Ωtrχ − ) + ∇ΩΩ−2 ∇2 (Ωtrχ − )(Ωtrχ − )
u u u u
−1 3 2 2 2 2 2
+ Ω ∇ (Ωtrχ − )(Ωtrχ − ) + (∇ η, ∇ η)trχ(Ωtrχ − )
u u u
2 −1 2 2 2
+ (∇η, ∇η)∇trχ(Ωtrχ − ) + Ω ∇ (Ωtrχ − )∇(Ωtrχ − )
u u u
2 2 2 2
+ ∇ΩΩ−2 ∇(Ωtrχ − )∇(Ωtrχ − ) + Ω−1 ∇2 (Ωtrχ − )∇(Ωtrχ − )
u u u u
2 2 2
+ (∇η, ∇η)trχ∇(Ωtrχ − ) + (η, η)∇2 trχ(Ωtrχ − ) + (η, η)∇χ∇(Ωtrχ − )
u u u
2
+ (η, η)χ∇2 (Ωtrχ − ) + (∇2 η, ∇2 η)Ωωtrχ + (∇η, ∇η)∇Ωωtrχ + (∇η, ∇η)Ω∇ωtrχ
u
+ (∇η, ∇η)Ωω∇trχ + (η, η)∇Ω∇ωtrχ + (η, η)∇Ωω∇trχ + (∇2 η, ∇2 η)Ωωtrχ
+ (η, η)∇2 Ωωtrχ + (∇η, ∇η)Ω∇ωtrχ + (η, η)Ω∇2 ωtrχ + (η, η)Ω∇ω∇trχ
+ (η, η)Ωω∇2 trχ + (∇2 η, ∇2 η)Ωχ̂ χ̂ + (∇η, ∇η)∇Ωχ̂ χ̂ + (∇η, ∇η)Ω∇χ̂ χ̂

2 2
+ (η, η)∇2 Ωχ̂ χ̂ + (η, η)∇Ω∇χ̂ χ̂ + (η, η)Ω∇χ̂∇χ̂ + (∇η, ∇η)∇ΩΩ−2 (Ωtrχ − )
u
2 2 2 2
+ (η, η)Ωχ̂∇2 χ̂ + (∇2 η, ∇2 η)Ω−1 (Ωtrχ − ) + (∇η, ∇η)Ω−1 ∇(Ωtrχ − )(Ωtrχ − )
u u u
2 2
+ (∇η, ∇η)∇ΩΩ−2 (Ωtrχ − )
u 188
2 2 2 2
+ (η, η)∇Ω∇ΩΩ−3 (Ωtrχ − ) + (η, η)∇ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − )
u u u
2 2 2 2
+ (∇η, ∇η)Ω−1 ∇(Ωtrχ − )(Ωtrχ − ) + (η, η)∇ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − )
u u u u
−1 2 2 2 −1 2 2
+ (η, η)Ω ∇ (Ωtrχ − )(Ωtrχ − ) + (η, η)Ω ∇(Ωtrχ − )∇(Ωtrχ − )
u u u u
2 2 2 2
+ (∇η, ∇η)χ∇(Ωtrχ − ) + (η, η)∇χ∇(Ωtrχ − ) + (η, η)χ∇ (Ωtrχ − )
u u u
2 2 2 2 2
+ ∇β∇(Ωtrχ − ) + β∇ (Ωtrχ − ) + (∇η, ∇η)χ̂∇(Ωtrχ − ) + (η, η)∇χ̂∇(Ωtrχ − )
u u u u
2 2 2
+ (η, η)χ̂∇2 (Ωtrχ − ) + (∇η, ∇η)∇trχ(Ωtrχ − ) + (η, η)∇2 trχ(Ωtrχ − )
u u u
2
+ (η, η)∇trχ∇(Ωtrχ − ) + (∇η, ∇η)∇Ωωtrχ + (η, η)∇2 Ωωtrχ + (η, η)∇Ω∇ωtrχ
u
+ (η, η)∇Ωω∇trχ + (∇η, ∇η)Ω∇ωtrχ + (η, η)Ω∇2 ωtrχ + (η, η)Ω∇ω∇trχ
+ (∇η, ∇η)Ωω∇trχ + (η, η)Ω∇ω∇trχ + (η, η)Ωω∇2 trχ
+ (∇η, ∇η)∇Ωχ̂ χ̂ + (η, η)∇2 Ωχ̂ χ̂ + (η, η)∇Ω∇χ̂ χ̂ + (∇η, ∇η)Ωχ̂∇χ̂

2 2
+ (η, η)Ω∇χ̂∇χ̂ + (η, η)Ωχ̂∇2 χ̂ + (∇η, ∇η)∇ΩΩ−2 (Ωtrχ − )
u
2 2 2
+ (η, η)∇2 ΩΩ−2 (Ωtrχ − ) + (η, η)∇Ω∇ΩΩ−3 (Ωtrχ − )2
u u
2 2 2
+ (η, η)∇ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − ) + (∇η, ∇η)∇ΩΩ−2 (Ωtrχ − )2
u u u
2 2
+ (η, η)∇2 ΩΩ−2 (Ωtrχ − )2 + (η, η)∇Ω∇ΩΩ−3 (Ωtrχ − )2
u u
2 2 2
+ (η, η)∇ΩΩ−2 ∇(Ωtrχ − )(Ωtrχ − ) + (∇η, ∇η)(η, η)trχ(Ωtrχ − )
u u u
2 2
+ (η, η)(η, η)∇trχ(Ωtrχ − ) + (η, η)(η, η)χ∇(Ωtrχ − )
u u
+ (∇η, ∇η)(η, η)Ωωtrχ + (η, η)(η, η)∇Ωωtrχ + (η, η)(η, η)Ω∇ωtrχ + (η, η)(η, η)Ωω∇trχ
+ (∇η, ∇η)(η, η)Ωχ̂ χ̂ + (η, η)(η, η)∇Ωχ̂ χ̂ + (η, η)(η, η)Ω∇χ̂ χ̂

2 2 2 2
+ (∇η, ∇η)(η, η)Ω−1 (Ωtrχ − ) + (η, η)(η, η)∇ΩΩ−2 (Ωtrχ − )2 + (η, η)β∇(Ωtrχ − )
u u u
2 2 2 2
+ (η, η)(η, η)Ω−1 ∇(Ωtrχ − )(Ωtrχ − ) + (η, η)χ∇2 (Ωtrχ − ) + β∇2 (Ωtrχ − )
u u u u
2
+ (η, η)(η, η)(η, η)trχ(Ωtrχ − ) + (η, η)(η, η)(η, η)Ωωtrχ
u
2
+ (η, η)(η, η)(η, η)Ωχ̂ χ̂ + (η, η)(η, η)(η, η)Ω−1 (Ωtrχ − )2 .
u
189
2.9.3 Equations for Energy Estimates
In this section, we demonstrate the null Bianchi equations for angular derivatives of curvature
components. For α, β, we have
∇3 ∇α + trχ∇α
ˆ + ω∇α + (η, η)∇β + χ̂∇α + βα

=∇∇⊗β
+ ∇trχα + ∇ωα + ∇χ̂(ρ, σ) + (∇η, ∇η)β
+ (η, η)χα + (η, η)ωα + (η, η)χ̂(ρ, σ) + (η, η)(η, η)β,
3
∇3 ∇2 α + trχ∇α
2
=∇2 ∇⊗β
ˆ + (η, η)∇2 β + χ̂∇2 α + ω∇2 α + α∇β + β∇α
+ ∇χ∇α + ∇ω∇α + ∇χ̂∇(ρ, σ) + (∇η, ∇η)∇β + (η, η)χ∇α
+ (η, η)ω∇α + (η, η)χ̂(∇ρ, ∇σ) + (η, η)(η, η)∇β
+ ∇2 trχα + ∇2 ωα + ∇2 χ̂(ρ, σ) + (∇2 η, ∇2 η)β
+ (∇η, ∇η)χα + (η, η)∇χα + (∇η, ∇η)ωα + (η, η)∇ωα
+ (∇η, ∇η)χ̂(ρ, σ) + (η, η)∇χ̂(ρ, σ) + (∇η, ∇η)(η, η)β
+ (η, η)βα + (η, η)(η, η)(χ, ω)α + (η, η)(η, η)χ̂(ρ, σ)
+ (η, η)(η, η)(η, η)β,
∇4 ∇β
=∇div α + ω∇β + χ∇β + (η, η)∇α + ββ
+ ∇trχβ + ∇ωβ + ∇ηα
+ χηβ + (η, η)trχβ + (η, η)ωβ + (η, η)ηα,
190
∇4 ∇2 β
=∇2 div α + (η, η)∇2 α + (ω, χ)∇2 β + β∇β
+ (∇χ, ∇ω)∇β + (∇η, ∇η)∇α + (η, η)(χ, ω)∇β + (η, η)(η, η)∇α
+ (∇2 trχ, ∇2 ω)β + (∇2 η, ∇2 η)α + (∇η, ∇η)χβ + (η, η)∇χβ
+ (∇η, ∇η)ωβ + (η, η)∇ωβ + (∇η, ∇η)(η, η)α
+ (η, η)ββ + (ω, χ)(η, η)(η, η)β + (η, η)(η, η)(η, η)α.
For β, ρ, σ, we obtain
3
∇3 ∇β + trχ∇β
2
=∇2 ρ + ∇∗ ∇σ + χ̂∇β + ∇ωβ + ω∇β + ∇χ̂β + χ̂∇β + ∇trχβ + (η, η)trχβ
+ (∇η, ∇η)(ρ, σ) + (η, η)(∇ρ, ∇σ) + ηβχ + ββ + (η, η)ωβ + (η, η)χ̂β + (η, η)(η, η)(ρ, σ),
∇3 ∇2 β + 2trχ∇2 β
=∇3 ρ + ∇∇∗ ∇σ + ∇χ̂∇β + χ̂∇2 β + ∇2 ωβ + ∇ω∇β + ω∇2 β
+ ∇2 χ̂β + ∇χ̂∇β + χ̂∇2 β + (∇2 η, ∇2 η)(ρ, σ) + (∇η, ∇η)(∇ρ, ∇σ) + ∇trχ∇β + ∇2 trχβ
+ (η, η)(∇2 ρ, ∇2 σ) + (∇η, ∇η)χβ + (η, η)∇χβ + (η, η)χ∇β + ∇ββ + β∇β
+ (∇η, ∇η)ωβ + (η, η)∇ωβ + (η, η)ω∇β + (∇η, ∇η)χ̂β + (η, η)∇χ̂β + (η, η)χ̂∇β
+ (∇η, ∇η)(η, η)(ρ, σ) + (η, η)(η, η)(∇ρ, ∇σ) + (η, η)∇βχ + (η, η)(η, η)χβ + (η, η)ββ
+ (η, η)(η, η)ωβ + (η, η)(η, η)χ̂β + (η, η)(η, η)(η, η)(ρ, σ),
191
∇4 ∇ρ
=∇div β + ∇χ̂α + χ̂∇α + χ(∇ρ, ∇σ) + ∇χ(ρ, σ) + (∇η, ∇η)β
+ (η, η)∇β + (η, η)χ(ρ, σ) + β(ρ, σ) + (η, η)χ̂α + (η, η)(η, η)β,
∇4 ∇σ
= − ∇div ∗ β + ∇χ̂α + χ̂∇α + χ(∇ρ, ∇σ) + ∇χ(ρ, σ) + (∇η, ∇η)β
+ (η, η)∇β + (η, η)χ(ρ, σ) + β(ρ, σ) + (η, η)χ̂α + (η, η)(η, η)β,
∇4 ∇2 ρ
=∇2 div β + ∇2 χ̂α + ∇χ̂∇α + χ̂∇2 α + χ̂∇2 α + χ(∇2 ρ, ∇2 σ)
+ ∇χ(∇ρ, ∇σ) + ∇2 χ(ρ, σ) + (∇2 η, ∇2 η)β + (∇η, ∇η)∇β + (η, η)∇2 β
+ (∇η, ∇η)χ(ρ, σ) + (η, η)∇χ(ρ, σ) + (η, η)χ(∇ρ, ∇σ)
+ ∇β(ρ, σ) + β(∇ρ, ∇σ) + (∇η, ∇η)χ̂α + (η, η)∇χ̂α + (η, η)χ̂∇α
+ (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β + (η, η)(∇ρ, ∇σ)χ + χ(∇2 ρ, ∇2 σ) + (η, η)(η, η)∇β
+ (η, η)(η, η)χ(ρ, σ) + (η, η)β(ρ, σ) + (η, η)(η, η)χ̂α + (η, η)(η, η)(η, η)β,
192
∇4 ∇2 σ
= − ∇2 div ∗ β + ∇2 χ̂α + ∇χ̂∇α + χ̂∇2 α + χ̂∇2 α + χ(∇2 ρ, ∇2 σ)
+ ∇χ(∇ρ, ∇σ) + ∇2 χ(ρ, σ) + (∇2 η, ∇2 η)β + (∇η, ∇η)∇β + (η, η)∇2 β
+ (∇η, ∇η)χ(ρ, σ) + (η, η)∇χ(ρ, σ) + (η, η)χ(∇ρ, ∇σ)
+ ∇β(ρ, σ) + β(∇ρ, ∇σ) + (∇η, ∇η)χ̂α + (η, η)∇χ̂α + (η, η)χ̂∇α
+ (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β + (η, η)(∇ρ, ∇σ)χ + χ(∇2 ρ, ∇2 σ) + (η, η)(η, η)∇β
+ (η, η)(η, η)χ(ρ, σ) + (η, η)β(ρ, σ) + (η, η)(η, η)χ̂α + (η, η)(η, η)(η, η)β.
For ρ, σ, β, we derive
∇3 ∇ρ + 2trχ∇ρ
= −∇div β + ∇trχ(ρ, σ) + ∇χ̂α + χ̂∇α + (∇η, ∇η)β + (η, η)∇β + χ̂(∇ρ, ∇σ)
+ (η, η)χ(ρ, σ) + (η, η)χ̂α + (η, η)(η, η)β,
∇3 ∇σ + 2trχ∇σ
= −∇div ∗ β + ∇trχ(ρ, σ) + ∇χ̂α + χ̂∇α + (∇η, ∇η)β + (η, η)∇β + χ̂(∇ρ, ∇σ)
+ (η, η)χ(ρ, σ) + (η, η)χ̂α + (η, η)(η, η)β,
193
5
∇3 ∇2 σ + trχ∇2 σ
2
= −∇2 div ∗ β + ∇2 trχ(ρ, σ) + ∇trχ(∇ρ, ∇σ) + ∇2 χ̂α + ∇χ̂∇α + χ̂∇2 α
+ (∇2 η, ∇2 η)β + (∇η, ∇η)∇β + (η, η)∇2 β + ∇χ̂(∇ρ, ∇σ)
+ χ̂(∇2 ρ, ∇2 σ) + (∇η, ∇η)χ(ρ, σ) + (η, η)∇χ(ρ, σ) + (η, η)χ(∇ρ, ∇σ)
+ (∇η, ∇η)χ̂α + (η, η)∇χ̂α + (η, η)χ̂∇α + (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β + χ̂(∇2 ρ, ∇2 σ)
+ β(∇ρ, ∇σ) + χ(η, η)(∇ρ, ∇σ) + (η, η)χ̂(∇ρ, ∇σ)
+ (η, η)(η, η)χ(ρ, σ) + (η, η)(η, η)χ̂α + (η, η)(η, η)(η, η)β,
∇4 ∇β
= − ∇2 ρ + ∇∗ ∇σ + ∇trχβ + trχ∇β + ∇ωβ + ω∇β + ∇χ̂β + χ̂∇β
+ (η, η)(∇ρ, ∇σ) + (∇η, ∇η)(ρ, σ) + χ∇β + ββ + χ(η, η)β + (η, η)ωβ
+ (η, η)χ̂β + (η, η)(η, η)(ρ, σ),
∇4 ∇ 2 β
= − ∇3 ρ + ∇∇∗ ∇σ + ∇2 trχβ + ∇trχ∇β + trχ∇2 β + ∇2 ωβ
+ ∇ω∇β + ω∇2 β + ∇2 χ̂β + ∇χ̂∇β + χ̂∇2 β + (∇η, ∇η)(∇ρ, ∇σ)
+ (η, η)(∇2 ρ, ∇2 σ) + (∇2 η, ∇2 η)(ρ, σ) + ∇χ∇β + χ∇2 β + ∇ββ + β∇β
+ ∇χ(η, η)β + χ(∇η, ∇η)β + χ(η, η)∇β + (∇η, ∇η)ωβ + (η, η)∇ωβ + (η, η)ω∇β
+ (∇η, ∇η)χ̂β + (η, η)∇χ̂β + (η, η)χ̂∇β + (∇η, ∇η)(η, η)(ρ, σ) + (η, η)(η, η)(∇ρ, ∇σ)
+ (η, η)∇βχ + χ∇2 β + (η, η)(η, η)(∇ρ, ∇σ) + (η, η)(∇η, ∇η)(ρ, σ)
+ (η, η)χ∇β + (η, η)ββ + χ(η, η)(η, η)β + (η, η)(η, η)ωβ
+ (η, η)(η, η)χ̂β + (η, η)(η, η)(η, η)(ρ, σ).
194
For β, α, we have
5
∇3 ∇β + trχ∇β
2
= − ∇div α + χ̂∇β + ∇ωβ + ω∇β + (∇η, ∇η)α + (η, η)∇α + ∇trχ β
+ β β + χ̂∇β + χ(η, η)β + (η, η)ωβ + (η, η)(η, η)α,
∇3 ∇2 β + 3trχ∇2 β
= − ∇2 div α + χ̂∇2 β + ∇χ̂∇β + ∇2 ωβ + ∇ω∇β + ω∇2 β + ∇2 trχ β + ∇trχ∇β
+ (∇2 η, ∇2 η)α + (∇η, ∇η)∇α + (η, η)∇2 α + ∇β β + ∇χ̂∇β
+ χ̂∇2 β + ∇χ(η, η)β + χ(∇η, ∇η)β + χ(η, η)∇β + (∇η, ∇η)ωβ
+ (η, η)∇ωβ + (η, η)ω∇β + (∇η, ∇η)(η, η)α + (η, η)(η, η)∇α + χ(η, η)∇β + χ̂∇2 β
+ (η, η)trχ∇β + (η, η)β β + (η, η)χ̂∇β + χ(η, η)(η, η)β
+ (η, η)(η, η)ωβ + (η, η)(η, η)(η, η)α,
∇4 ∇α
= − ∇∇⊗β
b + ∇trχα + χ∇α + ∇ωα + ω∇α + ∇χ̂(ρ, σ) + χ̂(∇ρ, ∇σ)
+ (∇η, ∇η)β + (η, η)∇β + (η, η)χα + (η, η)ωα + (η, η)χ̂(ρ, σ) + (η, η)(η, η)β,
195
∇4 ∇ 2 α
= − ∇2 ∇⊗β
b + ∇2 trχα + ∇trχ∇α + χ∇2 α + ∇2 ωα + ∇ω∇α + ω∇2 α
+ ∇2 χ̂(ρ, σ) + ∇χ̂(∇ρ, ∇σ) + χ̂(∇2 ρ, ∇2 σ) + (∇2 η, ∇2 η)β
+ (∇η, ∇η)∇β + (η, η)∇2 β + (∇η, ∇η)χα + (η, η)∇χα + (η, η)χ∇α
+ (∇η, ∇η)ωα + (η, η)∇ωα + (η, η)ω∇α + (∇η, ∇η)χ̂(ρ, σ) + (η, η)∇χ̂(ρ, σ)
+ (η, η)χ̂(∇ρ, ∇σ) + (∇η, ∇η)(η, η)β + (η, η)(η, η)∇β
+ (η, η)χ∇α + β∇α + χ∇2 α + (η, η)(η, η)χα + (η, η)(η, η)ωα
+ (η, η)(η, η)χ̂(ρ, σ) + (η, η)(η, η)(η, η)β.
196
Chapter 3
Formation of Trapped Surfaces in
Vacuum from Mild Incoming
Radiation
(Joint with Luk)
3.1 Setting, Equations and Notations
3.1.1 Definitions
Our setting is the characteristic initial value problem with data given on the two characteristic
hypersurfaces H1 and H 0 intersecting at the sphere S1,0 . The spacetime will be a solution to the
Einstein equations constructed in a neighborhood of H1 and H 0 containing S1,0 .
197
3.1.2 Double Null Foliation
1
2
a
H bδ
a H
Hδ δ (u
L =
δ)
L
H
L
0 (u
= L 1)
0) =
(u
H1
For a spacetime in a neighborhood of S1,0 , we define a double null foliation as follows: Let u
and u be solutions to the eikonal equation
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0,
satisfying the initial conditions u = 0 on H0 and u = 0 on H 0 . u is increasing towards the future
while u is decreasing towards the future.
Let
L0µ = −2g µν ∂ν u, L0µ = −2g µν ∂ν u.
These are null and geodesic vector fields. Let
2Ω−2 = −g(L0 , L0 ).
198
Define
e3 = ΩL0 , e4 = ΩL0
to be the normalized null pair such that
g(e3 , e4 ) = −2
and
L = Ω2 L0 , L = Ω2 L0
to be the so-called equivariant vector fields.
In the sequel, we will consider spacetime solutions to the vacuum Einstein equations in the
gauge such that
Ω = 1, on H1 and H 0 .
We denote the level sets of u as Hu and the level sets of u and H u . By virtue of the eikonal
equations, Hu and H u are null hypersurface. The sets defined by the intersections of the hyper-
surfaces Hu and H u are topologically 2-spheres, which we denote by Su,u . Notice that the integral
flows of L and L respect the foliation Su,u .
3.1.3 The Coordinate System
On a spacetime in a neighborhood of S1,0 , we define a coordinate system (u, u, θ1 , θ2 ) as follows:
On the sphere S1,0 , define a coordinate system (θ1 , θ2 ) such that on each coordinate patch the
metric γ is smooth, bounded and positive definite. Then we define the coordinates on the initial
hypersurfaces by requiring
∂ A ∂ A
θ = 0 on H 0 , and θ = 0 on H1 .
∂u ∂u
199
We now define the coordinate system in the spacetime in a neighborhood of S1,0 by letting u and
u to be solutions to the eikonal equations:
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0,
and define θ1 , θ2 by
/ L θA = 0,
L
where L
/ L denote the restriction of the Lie derivative to T Su,u (See [7], Chapter 1). Relative to the
coordinate system (u, u, θ1 , θ2 ), the null pair e3 and e4 can be expressed as

−1 ∂ ∂ ∂
e3 = −Ω + dA A , e4 = Ω−1 ,
∂u ∂θ ∂u
for some dA such that dA = 0 on H 0 , while the metric g takes the form
g = −2Ω2 (du ⊗ du + du ⊗ du) + γAB (dθA − dA du) ⊗ (dθB − dB du).
Consider a coordinate patch U on S1,0 and define Uu,0 to be a coordinate patch on Su,0
given by the one-parameter diffeomorphism generated by L. Define Uu,u to be the image of

S
Uu,0 under the one-parameter diffeomorphism generated by L. Define D1 = δa≤u≤1,0≤u≤δ Uu,u ,
S S
D2 = 1 Uu,u . We denote D1 D2 as D, that is the shadowed region.
bδa ≤u≤δa,0≤u≤δ
2
3.1.4 Equations
We will recast the Einstein equations as a system for Ricci coefficients and curvature components
associated to a null frame e3 , e4 defined above and an orthonormal frame e1 , e2 tangent to the
2-spheres Su,u . Using the indices A, B to denote 1, 2, we define the Ricci coefficients relative to the
200
null fame:
χAB = g(DA e4 , eB ), χAB = g(DA e3 , eB ),

1 1
ηA = − g(D3 eA , e4 ), η A = − g(D4 eA , e3 )
2 2
(3.1.1)
1 1
ω = − g(D4 e3 , e4 ), ω = − g(D3 e4 , e3 ),
4 4
1
ζA = g(DA e4 , e3 )
2
where DA = De(A) . We also introduce the null curvature components,
αAB = R(eA , e4 , eB , e4 ), αAB = R(eA , e3 , eB , e3 ),

1 1
βA = R(eA , e4 , e3 , e4 ), β A = R(eA , e3 , e3 , e4 ), (3.1.2)
2 2
1 1
ρ = R(e4 , e3 , e4 , e3 ), σ = ∗ R(e4 , e3 , e4 , e3 ).
4 4
Here ∗ R denotes the Hodge dual of R. We denote by ∇ the induced covariant derivative operator
on Su,u and by ∇3 , ∇4 the projections to Su,u of the covariant derivatives D3 , D4 (see precise
definitions in [12]).
Observe that,
1 1
ω = − ∇4 (log Ω), ω = − ∇3 (log Ω),
2 2 (3.1.3)
ηA = ζA + ∇A (log Ω), η A = −ζA + ∇A (log Ω).
Let φ(1) · φ(2) denote an arbitrary contraction of the tensor product of φ(1) and φ(2) with respect
to the metric γ. We also define
(1) (2) (1) (2) (1) (2)

(φ(1) ⊗φ
b (2) )AB := φA φB + φB φA − δAB (φ(1) · φ(2) ) for one forms φA , φA ,
(1) (2) (1) (2)

(φ(1) ∧ φ(2) )AB := / AB (γ −1 )CD φAC φBD for symmetric two tensors φAB , φAB ,
where / is the volume form associated to the metric γ. For totally symmetric tensors, the div and
201
curl operators are defined by the formulas
(div φ)A1 ...Ar := ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar := / BC ∇B φCA1 ...Ar .
Define also the trace to be
(trφ)A1 ...Ar−1 := (γ −1 )BC φBCA1 ...Ar−1 .
We separate the trace and traceless part of χ and χ. Let χ̂ and χ̂ be the traceless parts of χ
and χ respectively. Then χ and χ satisfy the following null structure equations:
1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ,
2
∇4 χ̂ + trχχ̂ = −2ω χ̂ − α,
1
∇3 trχ + (trχ)2 = −2ωtrχ − |χ̂|2 ,
2
∇3 χ̂ + trχ χ̂ = −2ωχ̂ − α,
1 (3.1.4)
2
1 1
∇4 χ̂ + trχχ̂ = ∇⊗η
b + 2ωχ̂ − trχχ̂ + η ⊗η,b
2 2
1
2
1 1
∇3 χ̂ + trχχ̂ = ∇⊗η
b + 2ω χ̂ − trχχ̂ + η ⊗η.
b
2 2
The other Ricci coefficients satisfy the following null structure equations:
∇4 η = −χ · (η − η) − β,
∇3 η = −χ · (η − η) + β,
3 1 1 1 (3.1.5)
∇4 ω = 2ωω + |η − η|2 − (η − η) · (η + η) − |η + η|2 + ρ,
4 4 8 2
3 1 1 1
∇3 ω = 2ωω + |η − η|2 + (η − η) · (η + η) − |η + η|2 + ρ.
4 4 8 2
202
The Ricci coefficients also satisfy the following constraint equations
1 1 1
div χ̂ = ∇trχ − (η − η) · (χ̂ − trχ) − β,
2 2 2
1 1 1
div χ̂ = ∇trχ + (η − η) · (χ̂ − trχ) + β,
2 2 2 (3.1.6)
1
curl η = −curl η = σ + χ̂ ∧ χ̂,
2
1 1
K = −ρ + χ̂ · χ̂ − trχtrχ.
2 4
with K the Gauss curvature of the spheres Su,u . The null curvature components satisfy the following
null Bianchi equations:
1
b + 4ωα − 3(χ̂ρ +∗ χ̂σ) + (ζ + 4η)⊗β,
∇3 α + trχα = ∇⊗β b
2
∇4 β + 2trχβ = div α − 2ωβ + ηα,
∇3 β + trχβ = ∇ρ + 2ωβ +∗ ∇σ + 2χ̂ · β + 3(ηρ +∗ ησ),

3 1
2 2
3 1
2 2 (3.1.7)
3 1
2 2
3 1
∇3 ρ + trχρ = −div β − χ̂ · α + ζ · β − 2η · β,
2 2
∇4 β + trχβ = −∇ρ +∗ ∇σ + 2ωβ + 2χ̂ · β − 3(ηρ −∗ ησ),
∇3 β + 2trχβ = −div α − 2ωβ + η · α,

1
b + 4ωα − 3(χ̂ρ −∗ χ̂σ) + (ζ − 4η)⊗β,
∇4 α + trχα = −∇⊗β b
2
∗
where denotes the Hodge dual on Su,u .
We now define the renormalized curvature components and rewrite the Bianchi equations in
terms of them. Let

1
σ̌ = σ + χ̂ ∧ χ̂.
2
The Bianchi equations expressed in terms of K and σ̌ instead of ρ and σ are as follows:
203
∇3 β + trχβ = − ∇K +∗ ∇σ̌ + 2ωβ + 2χ̂ · β − 3(ηK −∗ ησ̌)
1 3
+ (∇(χ̂ · χ̂) −∗ ∇(χ̂ ∧ χ̂)) − ηtrχtrχ
2 4
1
+ 3(η χ̂ · χ̂ −∗ η χ̂ ∧ χ̂) − (∇trχtrχ + trχ∇trχ),
4
3 1 1
∇4 σ̌ + trχσ̌ = − div ∗ β − ζ ·∗ β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)b + χ̂ ·∗ (η ⊗η),
b
2 2 2
1 1
∇4 K + trχK = − div β − ζ · β − 2η · β + χ̂ · ∇⊗η b + χ̂ · (η ⊗η)
b
2 2
1 1
− trχdiv η − trχ|η|2 ,
2 2 (3.1.8)
3 ∗ 1 1
∇3 σ̌ + trχσ̌ = − div β − ζ · β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)
∗ b + χ̂ ·∗ (η ⊗η),
b
2 2 2
1 1
∇3 K + trχK =div β − ζ · β + 2η · β + χ̂ · ∇⊗η b + χ̂ · (η ⊗η)
b
2 2
1 1
− trχdiv η − trχ|η|2 ,
2 2
∇4 β + trχβ =∇K +∗ ∇σ̌ + 2ωβ + 2χ̂ · β + 3(ηK +∗ ησ̌)
1 1
− (∇(χ̂ · χ̂) −∗ ∇(χ̂ ∧ χ̂)) + (∇trχtrχ + trχ∇trχ)
2 4
3
− 3(η χ̂ · χ̂ +∗ η χ̂ ∧ χ̂) + ηtrχtrχ.
4
Notice that we have obtained a system for the renormalized curvature components in which the
singular curvature components α and α do not appear.
In the sequel, we will use capital Latin letters A ∈ {1, 2} for indices on the spheres Su,u and
Greek letters µ ∈ {1, 2, 3, 4} for indices in the whole spacetime.
3.1.5 Schematic Notation
We introduce a schematic notation as follow: Let φ denote an arbitrary tensorfield. For the Ricci
coefficients, we use the notation
2
ψ ∈ {χ̂, trχ, ω}, ψ ∈ {η, η, χ̂, trχ + , ω}. (3.1.9)
u
204
The set of all Ricci coefficients can therefore be represented by either ψ, ψ or trχ.
3.1.6 Integration
Let U be a coordinate patch on S1,0 and pU be a partition of unity in DU such that pU is supported
in DU . Given a function φ, the integration on Su,u is given by the formula:
Z XZ ∞ Z ∞ p
φ := φpU det γdθ1 dθ2 .
Su,u U −∞ −∞
Let Du0 ,u0 by the region u0 ≤ u ≤ 1, 0 ≤ u ≤ u0 . The integration on Du,u is given by the formula
Z XZ u Z u Z ∞ Z ∞ p
φ := φpU − det gdθ1 dθ2 dudu
Du,u U 0 0 −∞ −∞
XZ u Z u Z ∞ Z ∞ p
=2 φpU Ω2 det γdθ1 dθ2 dudu.
U 0 0 −∞ −∞
Since there are no canonical volume forms on Hu and H u , we define integration by
Z XZ Z ∞ Z ∞ p
φ := φ2pU Ω det γdθ1 dθ2 du,
Hu U 0 −∞ −∞
and
Z XZ Z ∞ Z ∞ p
φ := φ2pU Ω det γdθ1 dθ2 du.
Hu U 0 −∞ −∞
With these notions of integration, we can define the norms that we will use. Let φ be an
arbitrary tensorfield. For 1 ≤ p < ∞, define
Z
||φ||pLp (Su,u ) := < φ, φ >p/2
γ ,
Su,u
Z
||φ||pLp (Hu ) := < φ, φ >p/2
γ ,
Hu
Z
||φ||pLp (H := < φ, φ >p/2
γ .
u)
Hu
205
Define also the L∞ norm by
||φ||L∞ (Su,u ) := sup < φ, φ >1/2

γ (θ).
θ∈Su,u
We will also use mixed norms defined by
! 12
Z u∗
2
||φ||L2u L∞ p
u L (S)
= ( sup ||φ||Lp (Su,u ) ) du ,
0 u∈[0,u∗ ]
! 12
Z u∗
||φ||L2u L∞ p
u L (S)
= ( sup ||∇i φ||Lp (Su,u ) )2 du .
0 u∈[0,]
Note that L∞ Lp is taken before taking L2 . In the sequel, we will frequently use
|| · ||L∞ 2 p
u Lu L (S)
≤ || · ||L2u L∞ p
u L (S)
.
With the above definition, ||φ||L2u L2 (Su,u ) and ||φ||L2 (H u ) differ by a factor of Ω. Nevertheless,
in view of Proposition 3.3.1, these norms are equivalent up to a factor of 2.
3.1.7 Norms
We now define the norms that we will work with. Let
X
1

i+1 i
R := sup 1 1 ||u ∇ β||L2 (Hu )
i≤4
u δ 2 a2

1 i+1 i 1
+ sup 1 1 ||u ∇ (K − , σ̌)||L2 (H u )
δ 2 a2
u |u|2
X
1 1

i+2 i
+ sup 3 3 ||u ∇ (K − , σ̌)||L2 (Hu )
u δ 2 a4 |u|2
1≤i≤4

1 i+2 i
+ sup 3 3 ||u ∇ β||L2 (H u )
u δ 2 a4

1 12
+ sup 3 3 ||u (K − )||L2 (Hu ) ,
u δ 2 a4 |u|2
206

1 5 5 1 5 5
Õ5,2 := sup 1 1 ||u ∇ (χ̂, trχ, ω)||L2 (Hu ) + sup 1 1 ||u ∇ η||L2 (H u )
u δ 2 a2 u δ 2 a2

1 6 5
+ sup 3 3 ||u ∇ (η, η)||L2 (Hu )
u δ 2 a4
1
|u| 2 5 5
+ sup 1 ||u ∇ (trχ, χ̂, ω)||L (H u ) ,
2
u,u δa 2

1 i i
Oi,2 := sup 1 ku ∇ (χ̂, ω, trχ)kL2 (Su,u )
u,u a 2

|u| i i 2
+ 1 ku ∇ (η, η, ∇ log Ω, χ̂, trχ + , ω)kL2 (Su,u ) ,
δa 2 u

1 i i
Oi,∞ := sup 1 |u|ku ∇ (χ̂, ω, trχ)kL∞ (Su,u )
u,u a 2
|u|2 i i

2
+ 1 ku ∇ (η, η, χ̂, trχ + , ω)kL∞ (Su,u ) ,
δa 2 u
X X
O := Oi,2 + Oi,∞ .
i≤4 i≤2
Remark 15. For convenience, we list the L∞ (Su,u ) norms for curvature components:
|u|3

1 2 1
R0,∞ := sup 1 |u| kβkL∞ (Su,u ) + 1 kK − , σ̌, βkL∞ (Su,u ) ,
u,u a2 δa 2 |u|2
and
|u|4

1 3
R1,∞ := sup 1 |u| k∇βkL∞ (Su,u ) + 1 k∇K, ∇σ̌, ∇βkL (Su,u ) .
∞
u,u a2 δa 2
With the aid of constraint equations (3.1.6), by employing Sobolev Embedding it is straight forward
that we have
X X
2 2
R0,∞ + R1,∞ ≤ Oi,2 + Oi,∞ .
i≤4 i≤2
207
3.2 Statement of main theorem
We are now ready to state our main theorem
Theorem 3.2.1. Consider the following characteristic initial value problem for the Einstein vacuum
equations. The initial incoming hypersurface H 0 is required to coincide with a backwards light cone
in Minkowski space with 0 ≤ u ≤ 1. On the initial outgoing hypersurface H1 , the data are smooth
and the initial shear satisfies

1
X
k∇i χ̂0 kL∞ 2
u L (Su,u )
≤ a2
i≤7
for 0 ≤ u ≤ δ.
1
Then there exists a universal large constant b0 such that if b0 ≤ b ≤ a and δa 2 b < 1, there exists
a unique smooth spacetime solution (M, g) endowed with a double null foliation (u, u) which solves
1
the characteristic initial value problem to the vacuum Einstein equations in the region δa 2 b ≤ u ≤ 1,
0 ≤ u ≤ δ.
Moreover, the solution satisfies the estimates
O, Õ5,2 , R ≤ 1,
where the implicit constant is universal and independent of a, b and δ.
Remark 16. In the following, we will only prove the a priori estimates for O, Õ5,2 and R. The exis-
tence and uniqueness of solution and the propagation of regularity follow from standard arguments
(see, for example, [7]).
Remark 17. Following [7], one can solve the constraint ODEs and obtain bounds for the initial
data on H1 from that of the initial shear. In particular, under the assumption of Theorem 3.2.1,
we have the following initial bounds for the Ricci components
X 1 1

i i
1 k∇ (χ̂, ω, trχ)kL2 (S1,u ) + 1 k∇ (η, η, χ̂, trχ + 2, ω)kL2 (S1,u ) ≤ 1,
i≤5
a2 δa 2
208
and the following initial bounds for the curvature components
X 1 1

i i
1 k∇ βkL∞ 2
u L (S1,u )
+ 1 k∇ (K − 1, σ̌, β)kL∞ 2
u L (S1,u )
≤ 1.
i≤4
a2 δa 2
Once the existence theorem is established, the actual formation of trapped surfaces follows from
a simple ODE argument as in [7]:

Rδ 1
Theorem 3.2.2. If moreover, 0
|χ̂|2 ≥ 4bδa 2 for every direction θ along u = 1, then S 1 :=
bδa 2 ,δ
1
{(u, u, θ1 , θ2 ) : u = bδa 2 , u = δ} is a trapped surface.
Our existence theorem can easily be combined with the result of [13] to obtain the anisotropic
formation of trapped surfaces.
Remark 18. The original theorem of Christodoulou can be recovered by choosing a = δ −1 and
1
b = r∗ δ − 2 , where r∗ ∈ (0, 1) is arbitrary. It is obvious that for any fixed r∗ , δ can be chosen
sufficiently small such that b0 ≤ b ≤ a.
Remark 19. Even in the setting of Christodoulou’s theorem, we can improve his result to get
a larger region of existence. This can be seen by choosing b = b0 instead, where b0 is the large
constant in the statement of Theorem 3.2.1. In particular, the lower bound required for the data is
much smaller than the upper bound in terms of δ.
3.2.1 Structure of the Proof
We briefly outline the proof of Theorem 3.2.1:
P 1 P 1
STEP 0: Assuming that i≤4 Oi,2 ≤ b 4 holds, we derive i≤2 Oi,∞ ≤ b 4 by applying Sobolev
Embedding.
P 1
STEP 1: With the bound i≤2 Oi,∞ ≤ b 4 , in Section 3.3 we control the metric components, from
which we derive preliminary estimates.
1 P
STEP 2: Assuming Õ5,2 + R ≤ b 4 , in Section 3.4 we prove that i≤4 Oi,2 ≤ 1. This implies
P
i≤2 Oi,∞ ≤ 1.
209
1
STEP 3: Assuming R ≤ b 4 , in Section 3.5 we show that Õ5,2 ≤ 1.
STEP 4: Using the previous steps, in Section 3.6 we establish the estimate R ≤ 1.
Thus we have improved all the estimates and finish the proof of Theorem 3.2.1.
3.2.2 Strategy of the Proof
In this subsection, we list some strategies that used in the proof.
Weighted Estimates
One main difficulty in this section is that, when the spacetime is close to the center of gravita-
1
tional collapse, all the Ricci coefficients and curvature components grow like |u|e , where e = 1, 2, 3, 4
and u is the distance to the center. We employ u-weighted estimates to control the growth. The
method to putting u-weights is consistent with the structure of Einstein’s equations. Let ψ denote
Ricci coefficients and Ψ denote curvature components. They obey null structure equations
∇3 ψ + c[ψ]trχψ = Ψ + ψ 0 ψ 0 , ∇4 ψ = Ψ + ψ 0 ψ 0 ,
and null Bianchi equations
∇3 Ψ + c[Ψ]trχΨ = ∇Ψ0 + ψΨ0 , ∇4 Ψ = ∇Ψ0 + ψΨ0 ,
where c[ψ]trχψ and c[Ψ]trχΨ are borderline terms. With the fact
2
trχ = − + l.o.t.,
|u|
we apply integrating factors to cancel borderline terms containing trχ. Meanwhile, this method
provide us with the correct weighted estimates.
Furthermore, to control angular derivatives of Ricci coefficients and curvature components, we
exploit the identities:
∇3 ∇ψ = ∇∇3 ψ + [∇3 , ∇]ψ,
210
and
∇3 ∇Ψ = ∇∇3 Ψ + [∇3 , ∇]Ψ.
Commutators [∇3 , ∇]ψ and [∇3 , ∇]Ψ provide us with additional borderline terms 21 trχ∇ψ and
1
2 trχ∇Ψ. Thus, we obtain
1
∇3 ∇ψ + (c[ψ] + )trχ∇ψ = ∇Ψ + ∇ψ 0 ψ 0 ,
2
and
1
∇3 ∇Ψ + (c[Ψ] + )trχ∇Ψ = ∇2 Ψ0 + ∇ψΨ0 + ψ∇Ψ0 .
2
1 1
Coefficients c[ψ] + 2 and c[Ψ] + 2 provide us with the desired weight for ∇ψ and ∇Ψ through
integrating factors. In the same fashion, we obtain weighted estimates for higher order angular
derivatives of Ricci coefficients and curvature components.
Getting Smallness with Parameter b
During the proof, we constantly apply the inequality
1
δa 2 1
≤ , (3.2.1)
|u| b
1
for the region δa 2 b ≤ u ≤ 1, 0 ≤ u ≤ δ. Here b is large parameter.
For example, in Section 3.3.1 we would like to show that kΩ−1 − 1kL∞ (Su,u ) is small. With
equation
∂ −1
Ω = 2ω,
∂u
and bootstrap assumption

1 1
a2 b4
kωkL∞ (Su,u ) ≤ ,
|u|
we derive
Z u 1 1
δa 2 b 4 1
kΩ−1 − 1kL∞ (Su,u ) ≤ kωkL∞ (Su,u0 ) du0 ≤ ≤ 3.
0 |u| b4
For the last step, we exploit inequality (3.2.1). Since b is large parameter. We have proved that
211
kΩ−1 − 1kL∞ (Su,u ) is small.
Estimate by Groups and Improved Estimates
The other challenge is that we need to control many different Ricci coefficients and curvature
components, whose behaviors are governed by three parameters δ, u and a. In [7], Christodoulou
bounds these components separately with different δ and u weights.
In our proof, we divide Ricci coefficients into two groups:{χ̂, ω, trχ} and {η, η, ∇ log Ω, χ̂, trχ +
2
P P
|u| , ω}. In bootstrap assumptions for i≤4 Oi,2 and i≤2 Oi,∞ , we expect coefficients within each
group obey the same bound. With the aid of constraint equations, we control curvature components
via the bounds for Ricci coefficients. Estimating by groups reduces a lot of work.
P P
After controlling i≤4 Oi,2 and i≤2 Oi,∞ , we take one step further and improve the estimates
for
X 2 X X X
∇i (trχ + ), ∇i ω, ∇i trχ, ∇i µ,
|u|
1≤i≤4 1≤i≤4 1≤i≤4 1≤i≤4
which are useful for establishing elliptic estimates and energy estimates.
Renormalization
In this section, with initial data given in Theorem 3.2.1, we will establish an existence result in
1
the region δa 2 b ≤ u ≤ 1 and 0 ≤ u ≤ δ, which is sharp with respect to δ and a. Compared with
1
[7], in the norms we replace ρ by K − |u|2 . Here we demonstrate the reasons.
Our results are independent of α and α. Following [19], we first replace ρ and σ by ρ̌ and σ̌ to
get rid of α and α, where

1 1
ρ̌ := ρ − χ̂ · χ̂, σ̌ := σ + χ̂ ∧ χ̂.
2 2
With these renormalizations, we are able to establish existence result in the region δa ≤ u ≤ 1 and
0 ≤ u ≤ δ. However, when we go further toward the center of gravitational collapse, ρ̌ only obeys
bound
δa
kρ̌kL∞ (Su,u ) ≤ .
|u|3
While for σ̌, we have

1
δa 2
kσ̌kL∞ (Su,u ) ≤ .
|u|3
212
1
Under this situation, ρ̌ is not good enough. By noticing the fact that K − |u|2 obeys bound
1
1 δa 2
kK − 2 kL∞ (Su,u ) ≤
|u| |u|3
and satisfies the following constraint equation
1 1 1
K− = −ρ̌ − trχtrχ − 2 ,
|u|2 4 |u|
1
we make a further renormalization to substitute K − |u|2 for ρ̌, which enables us to obtain desired
results.
We start our proof in the below.
3.3 The Preliminary Estimates
We will make the following bootstrap assumptions on the first four derivatives of the Ricci coeffi-
cients:
X 1 X 1 1
1 kui+1 ∇i ψkL2 (Su,u ) + 1 kui+2 ∇i ψkL∞ (Su,u ) ≤ b 4 (3.3.1)
i≤4
δa 2
i≤2
δa 2
and
X 1 X 1 1
1 kui ∇i ψkL2 (Su,u ) + 1 kui+1 ∇i ψkL∞ (Su,u ) ≤ b 4 . (3.3.2)
i≤4
a 2
i≤2
a 2
We will also make the bootstrap assumptions on Õ5,2 and R:
1
Õ5,2 + R ≤ b 4 . (3.3.3)
Finally, we make a bootstrap assumption on K and its derivatives, which will be useful for elliptic
estimates:
X 1
kui+1 ∇i (K − )kL∞ ∞ 2
u Lu L (Su,u )
≤ 1. (3.3.4)
|u|2
i≤3
In the following sections, we will show that given the bootstrap assumptions, the bounds in fact
213
hold with a better constant. More precisely, we will show
X 1 X 1
1 kui+1 ∇i ψkL2 (Su,u ) + 1 kui+2 ∇i ψkL∞ (Su,u ) ≤ 1,
i≤4
δa 2
i≤2
δa 2
X 1 X 1
1 kui ∇i ψkL2 (Su,u ) + 1 kui+1 ∇i ψkL∞ (Su,u ) ≤ 1,
i≤4
a 2
i≤2
a2
Õ5,2 + R ≤ 1
and
X 1 1
kui+1 ∇i (K − 2
)kL∞ ∞ 2
u Lu L (Su,u )
≤ 3.
|u| b4
i≤3
Here, and in the rest of this section, we use the convention that A ≤ B denotes the
1
inequality A ≤ CB for some universal constant C that is independent of δ, a 2 and b.
The bounds that we derive will therefore improve over the bootstrap assumptions (3.3.1), (3.3.2),
(3.3.3) and (3.3.4) after choosing b0 to be sufficiently large.
3.3.1 Estimates for Metric Components
We first show that we can control Ω under the bootstrap assumptions:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1 1
δa 2 b 4
kΩ−1 − 1kL∞ (Su,u ) ≤ .
|u|
Proof. Consider the equation
1 1 1 ∂ −1
ω = − ∇4 log Ω = Ω∇4 Ω−1 = Ω . (3.3.5)
2 2 2 ∂u
Fix u. Notice that both ω and Ω are scalars and therefore the L∞ norm is independent of the
214
metric. We can integrate equation (3.3.5) using the fact that Ω−1 = 1 on H 0 to obtain
Z u 1 1
−1 δa 2 b 4
||Ω − 1||L∞ (Su,u ) ≤ ||ω||L∞ (Su,u0 ) du0 ≤ ,
0 |u|
where we have used the bootstrap assumption (3.3.2).
We then show that we can control γ under the bootstrap assumptions. This follows from an
argument similar to that in [7].
strap assumptions (3.3.1), (3.3.2), (3.3.3) and (3.3.4). Fix a point (u, ϑ) on the initial hypersurface
H 0 . Along the outgoing characteristic emanating from (u, ϑ), define Λ(u) and λ(u) to be the larger
and smaller eigenvalue of γ −1 (u, u, ϑ)γ(u, u, ϑ). Then
1 1
δa 2 b 4
|Λ(u) − 1| + |λ(u) − 1| ≤
|u|
for every u ∈ [0, δ]. As a consequence, we also have
1 1
δa 2 b 4
|µ(u) − 1| ≤
|u|
for every u ∈ [0, δ], where

dvolγu
µ(u) = .
dvolγ0
Proof. The first variation formula states that
L
/ L γ = 2Ωχ,
which translates to
∂
γAB = 2ΩχAB (3.3.6)
∂u
215
in coordinates. From this we derive that
∂
log µ = Ωtrχ.
∂u
Since by definition µ(0) = 1, we have
1 1
δa 2 b 4
|µ(u) − 1| ≤ , (3.3.7)
|u|
using the bound for trχ in the bootstrap assumption (3.3.2) and the estimate for Ω from Proposition
3.3.1.
q
Λ(u) Λ(u)
Define ν(u) = µ(u) = λ(u) . Following the derivation of (5.93) in [7], we can use (3.3.6) to
derive the estimate

Z u
ν(u) ≤ 1 + |(Ωχ̂)(u0 )|γ ν(u0 )du0 .
0
This implies via Gronwall’s inequality that
1 1
δa 2 b 4
|ν(u) − 1| ≤ . (3.3.8)
|u|
The desired conclusion follows from (3.3.7) and (3.3.8).
A direct consequence of the previous proposition is an estimate on the surface area of the two
sphere Su,u .
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1 1
δa 2 b 4
sup |Area(Su,u ) − Area(Su,0 )| ≤ .
u,u |u|
3.3.2 Estimates for Transport Equations
In latter sections of the paper, we will derive the estimates for the Ricci coefficients and the null
curvature components from the null structure equations and the null Bianchi equations respectively.
216
These will be viewed as transport equations and we will need a way to obtain estimates from the
covariant null transport equations. For the transport equation in the e4 direction, we will need the
smallness of ktrχkL∞ 1 ∞
u Lu L (Su,u )
, which is a consequence of our bootstrap assumption (3.3.2). More
precisely, we have
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
Z u
||φ||L2 (Su,u ) ≤ ||φ||L2 (Su,u0 ) + ||∇4 φ||L2 (Su,u00 ) du00
u0
Z Z Z
d df
f= + Ωtrχf = Ω (e4 (f ) + trχf ) .
Z u Z
1
||φ||2L2 (Su,u ) =||φ||2L2 (Su,u0 ) + 2Ω < φ, ∇4 φ >γ + trχ|φ|2γ du00 . (3.3.9)
u0 Su,u00 2
The proposition can be concluded using the Cauchy-Schwarz inequality on the sphere and the L∞
bounds for Ω and trχ which are provided by Proposition 3.3.1 and the bootstrap assumption (3.3.2)
respectively.
On the other hand, in order to use the ∇3 equation, we need to incorporate the weights in the
norms. These weights depend on the coefficients in front of the linear term with a trχ factor. The
main observation is that under the bootstrap assumption (3.3.1), trχ can be viewed essentially as
2
− |u| . More precisely, we have
strap assumptions (3.3.1), (3.3.2), (3.3.3) and (3.3.4). Let φ and F be Su,u -tangent tensor fields of
217
rank k satisfying the following transport equation:
∇3 φA1 ...Ak + λ0 trχφA1 ...Ak = FA1 ...Ak ,
Denoting λ1 = 2(λ0 − 12 ), we have
Z 1
|u|λ1 kφkL2 (Su,u ) ≤ kφkL2 (S1,u ) + |u0 |λ1 kF kL2 (Su0 ,u ) du0 ,
u
where the implicit constant is allowed to depend on λ0 .
Proof. To begin, we have the following identity for any scalar function f :
Z Z Z
d df
− f= − + Ωtrχf = Ω e3 (f ) + trχf .
Using this identity, we obtain
Z
d
− ( |u|2λ1 |φ|2 )
du Su,u
Z
= Ω 2λ1 |u|2λ1 −1 (e3 u)|φ|2 + 2|u|2λ1 < φ, ∇3 φ > +trχ|u|2λ1 |φ|2
Su,u
Z (3.3.10)
2λ1
= Ω 2|u| < φ, ∇3 φ + λ0 trχφ >
Su,u
Z
2λ1 (e3 u)
+ |u|2λ1 Ω( + (1 − 2λ0 )trχ)|φ|2 .
Su,u |u|
Observe that1 we have
2λ1 (e3 u)
+ (1 − 2λ0 )trχ
|u|
2λ1 − 4λ0 + 2 2λ1 (Ω−1 − 1) 2
=− − + (1 − 2λ0 )(trχ + )
|u| |u| |u|
1 1
δa 2 b 4
≤
|u|2
1 Note that in the following formula we need the exact cancellation and thus we do not use the schematic notation
for this equation.
218
using Proposition 3.3.1 and the bootstrap assumption (3.3.1), since we have chosen the parameters
to satisfy 2λ1 − 4λ0 + 2 = 0.
Therefore,
Z Z
d 2λ1 2 2λ1 2λ1 −2 1 1
2
|− ( |u| |φ| )| ≤ 2|u| |φ||F | + |u| δa 2 b 4 |φ| .
du Su,u Su,u
Using Cauchy-Schwarz for the first term and applying Gronwall’s inequality for the second term,
we obtain
|u|λ1 kφkL2 (Su,u )

1 1
Z 1
δa 2 b 4 ku−2 kL1
≤e u (kφkL2 (S1,u ) + |u0 |λ1 kF kL2 (Su0 ,u ) du0 )
u
Z 1
≤kφkL2 (S1,u ) + |u0 |λ1 kF kL2 (Su0 ,u ) du0 .
u
1 1
1 1
since δa 2 b 4 ku−2 kL1u ≤ δa 2 b 4
|u| ≤ 1
3 .
b4
3.3.3 Sobolev Embedding
Using the estimates for the metric γ given in Proposition 3.3.2, we can follow [7] to obtain a bound
on the isoperimetric constant
min{Area(U ), Area(U c )}
I(S) = sup ,
U (Perimeter(∂U ))2
∂U ∈C 1
where S is one of the 2-spheres Su,u adapted to the double null foliation. This will then allow us
to obtain the necessary Sobolev embedding theorems. We first have the following estimate:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), the isoperimetric constant obeys the bound
1
I(Su,u ) ≤
π
219
1
for 0 ≤ u ≤ δ and δa 2 b ≤ u ≤ 1.
Proof. Fix u. For a domain Uu ⊂ Su,u , define U0 ⊂ Su,0 to be the image of Uu under the
diffeomorphism generated by the equivariant vector field L. Using the notations introduced in
Proposition 3.3.2 and its proof, we have
Perimeter(∂Uu ) r
≥ inf λ(u)
Perimeter(∂U0 ) S0,u
and
Area(Uu ) Area(Uuc )
≤ sup µ(u), ≤ sup µ(u).
Area(U0 ) S0,u Area(U0c ) S0,u
1
Now, the conclusion follows from the fact that S0,u = 2π and the bounds in Proposition 3.3.2.
We will only need an L2 − L∞ Sobolev embedding in this paper. In order to derive it, we will
use the following two propositions, quoted directly from [7]. The first is an L2 − Lp embedding
theorem:
Proposition 3.3.7 ([7], Lemma 5.1). For any Riemannian 2-manifold (S, γ), we have the estimate
1 1
(Area(S))− p kξkLp (S) ≤ Cp
p
max{I(S), 1}(k∇ξkL2 (S) + (Area(S))− 2 kξkL2 (S) )
for 2 < p < ∞ and for any tensor ξ.
The second is an Lp − L∞ embedding:
Proposition 3.3.8 ([7], Lemma 5.2). For any Riemannian 2-manifold (S, γ), we have the estimate
1 1 1
max{I(S), 1}(Area(S)) 2 − p (k∇ξkLp (S) + (Area(S))− 2 kξkLp (S) )
p
kξkL∞ (S) ≤ Cp
for p > 2 and for any tensor ξ.
Recall from Proposition 3.3.3 that Area(Su,u ) ∼ |u|2 . We can now combine Propositions 3.3.6,
3.3.7 and 3.3.8 to obtain
220
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X
kφkL∞ (Su,u ) ≤ kui−1 ∇i φkL2 (Su,u ) . (3.3.11)
i≤2
3.3.4 Commutation Formula
In this section, we derive general commutation formulae. We have the following formula from [12]:
Proposition 3.3.10. The commutator [∇4 , ∇] acting on an (0, r) S-tensor is given by
[∇4 , ∇B ]φA1 ...Ar
=[D4 , DB ]φA1 ...Ar + (∇B log Ω)∇4 φA1 ...Ar − (γ −1 )CD χBD ∇C φA1 ...Ar
r
X r
X
− (γ −1 )CD χBD η A φA1 ...Âi C...Ar + (γ −1 )CD χAi B η D φA1 ...Âi C...Ar .
i
i=1 i=1
Proposition 3.3.11. The commutator [∇3 , ∇] acting on an (0, r) S-tensor is given by
[∇3 , ∇B ]φA1 ...Ar
=[D3 , DB ]φA1 ...Ar + (∇B log Ω)∇3 φA1 ...Ar − (γ −1 )CD χBD ∇C φA1 ...Ar
r
X r
X
− (γ −1 )CD χBD ηAi φA1 ...Âi C...Ar + (γ −1 )CD χA ηD φA1 ...Âi C...Ar .
iB
i=1 i=1
By induction, we get the following schematic formula for repeated commutations (see [18]):
X X
Fi = ∇i1 (η + η)i2 ∇i3 F0 + ∇i1 (η + η)i2 ∇i3 χ∇i4 φ
i1 +i2 +i3 =i i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 β∇i4 φ.
i1 +i2 +i3 +i4 =i−1
where by ∇i1 (η + η)i2 we mean the sum of all terms which is a product of i2 factors, each factor
221
X
being ∇j (η + η) for some j and that the sum of all j’s is i1 , i.e., ∇i1 (η + η)i2 = ∇j1 (η +
j1 +...+ji2 =i1
η)...∇ji2 (η + η). Similarly, suppose ∇3 φ = G0 . Let ∇3 ∇i φ = Gi . Then
i X
Gi − trχ∇i φ = ∇i1 (η + η)i2 ∇i3 G0
2 i 1 +i2 +i3 =i
X 2 i4
+ ∇i1 (η + η)i2 ∇i3 (χ̂, trχ + )∇ φ
i1 +i2 +i3 +i4 =i
u
X
+ ∇i1 (η + η)i2 ∇i3 β∇i4 φ.
i1 +i2 +i3 +i4 =i−1
The following further simplified version is useful in the latter sections:
X X
Fi = ∇ i 1 ψ i 2 ∇ i 3 F0 + ∇i1 ψ i2 ∇i3 ψ∇i4 φ.
i1 +i2 +i3 =i i1 +i2 +i3 +i4 =i
Similarly, suppose ∇3 φ = G0 . Let ∇3 ∇i φ = Gi . Then
i X X
Gi − trχ∇i φ = ∇i1 ψ i2 ∇i3 G0 + ∇i1 ψ i2 ∇i3 φ
2 i 1 +i2 +i3 =i i1 +i2 +i3 =i
X
+ trχ∇i1 ψ i2 ∇i3 φ.
i1 +i2 +i3 =i,i3 ≤i−1
Proof. We first replace β and β using the schematic Codazzi equations:
β = ∇ψ + ψψ,
β = ∇ψ + ψ(trχ + ψ).
2
We then replace trχ and χ̂ by ψ, as well as substitute η, η, χ̂ and trχ + u with ψ.
222
3.3.5 General Elliptic Estimates for Hodge Systems
In this subsection, we prove elliptic estimates for general Hodge systems. To this end, we recall the
definition of the divergence and curl of a symmetric covariant tensor of an arbitrary rank:
(div φ)A1 ...Ar = ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar = / BC ∇B φCA1 ...Ar ,
where / is the volume form associated to the metric γ. Recall also that the trace is defined to be
(tr φ)A1 ...Ar−1 = (γ −1 )BC φBCA1 ...Ar−1 .
The following is the main L2 (Su,u ) elliptic estimate that we will use:
strap assumptions (3.3.1), (3.3.2), (3.3.3) and (3.3.4). Let φ be a totally symmetric r + 1 covariant
tensorfield on a 2-sphere (S2 , γ) satisfying
div φ = f, curl φ = g, trφ = h.
Then, for 1 ≤ i ≤ 4, we have
kui ∇i φ||L2 (Su,u )

i−1
X
≤ (||uj+1 ∇j (f, g)||L2 (Su,u ) + kuj ∇j hkL2 (Su,u ) + ||uj ∇j φ||L2 (Su,u ) ).
j=0
Proof. Recall the following identity from [8] and Chapter 7 in [7] that for φ, f , g and h as
above
Z Z
2 2 2 2 2
|∇φ| + (r + 1)K|φ| = |f | + |g| + K|h| . (3.3.12)
Su,u Su,u
1
Notice that kKkL∞ ∞ ∞
u Lu L (Su,u )
≤ |u|2 by the bootstrap assumption (3.3.4) and the Sobolev embed-
223
ding theorem (Proposition 3.3.9). This implies the conclusion for i = 1 after multiplying (3.3.12)
by u2 . For i > 1, we recall from [8] and [7] that the symmetrized angular derivative of φ defined by
r+1
1 X
(∇φ)sBA1 ...Ar+1 := (∇B φA1 ...Ar + ∇Ai φA1 ...<Ai >B...Ar+1 )
r+2 i=1
satisfies the div-curl system
1 ∗ 2K
div (∇φ)s =(∇f )s − ( ∇g)s + (r + 1)Kφ − (γ ⊗s h)
r+2 r+1
r+1
curl (∇φ)s = (∇g)s + (r + 1)K(∗ φ)s
r+2
2 r
tr (∇φ)s = f+ (∇h)s ,
r+2 r+2
where
X
(γ ⊗s h)A1 ...Ar+1 := γAi Aj hA1 ...<Ai >...<Aj >...Ar+1
i<j=1,...,r+1
and
r+1
∗ 1 X
( φ)sA1 ...Ar+1 := / A B φA1 ...<Ai >B...Ar .
r + 1 i=1 i
Using (3.3.12), we therefore obtain that for i = 2, we have
≤k∇f k2L2 (Su,u ) + k∇gk2L2 (Su,u ) + kK(|∇φ|2 + |f |2 + |∇h|2 )kL1 (Su,u )
+ kKφk2L2 (Su,u ) + kKhk2L2 (Su,u ) .
1
Using again kKkL∞ ∞ ∞
u Lu L (Su,u )
≤ |u|2 , we have
k∇2 φk2L2 (Su,u ) ≤k∇(f, g)k2L2 (Su,u )

1 1
+ (k∇(φ, h)k2L2 (Su,u ) + kf k2L2 (Su,u ) ) + 4 k(φ, h)k2L2 (Su,u ) ,
|u|2 |u|
224
which implies, after multiplying by |u|4 , that
ku2 ∇2 φk2L2 (Su,u )
≤ku2 ∇(f, g)k2L2 (Su,u ) + ku∇(φ, h)k2L2 (Su,u ) + kuf k2L2 (Su,u ) + k(φ, h)k2L2 (Su,u )
X
≤ (kuj+1 ∇j (f, g)k2L2 (Su,u ) + kuj ∇j (φ, g)kL2 (Su,u ) ).
j≤1
Iterating the procedure and applying (3.3.12), we thus obtain that for i ≤ 4,
k∇i φk2L2 (Su,u )
≤k∇i−1 (f, g)k2L2 (Su,u ) + kK(|∇i−1 (φ, h)|2 + |∇i−2 (f, g)|2 )kL1 (Su,u )
X X
+ kK( ∇i1 K∇i2 (φ, h))2 kL1 (Su,u ) + kK( ∇i1 K∇i2 f )2 kL1 (Su,u )
i1 +i2 =i−3 i1 +i2 =i−4
X X
+ k∇i1 K∇i2 (φ, h)k2L2 (Su,u ) + k∇i1 K∇i2 (f, g)k2L2 (Su,u )
i1 +i2 =i−2 i1 +i2 =i−3
X
+ kK∇i1 K∇i2 (φ, h)k2L2 (Su,u )
i1 +i2 =i−4
P
where we have used the convention that i≤−1 = 0. By bootstrap assumption (3.3.4) and Sobolev
embedding (Proposition 3.3.9), we have
X
kui ∇i KkL∞ (Su,u ) ≤ 1.
i≤2
Therefore, for i ≤ 4, we have
X
kui ∇i φk2L2 (Su,u ) ≤ kuj+1 ∇j (f, g)k2L2 (Su,u ) + kuj ∇j (φ, h)k2L2 (Su,u ) .
j≤i−1
For the special case that φ a symmetric traceless 2-tensor, we only need to know its divergence:
Proposition 3.3.15. Suppose φ is a symmetric traceless 2-tensor satisfying
div φ = f.
225
Then, under the assumptions of Theorem 3.2.1 and the bootstrap assumptions (3.3.1), (3.3.2),
(3.3.3) and (3.3.4), for 1 ≤ i ≤ 4, we have
i−1
X
||ui ∇i φ||L2 (Su,u ) ≤ (||uj+1 ∇j f ||L2 (Su,u ) + ||uj ∇j φ||L2 (Su,u ) ).
j=0
Proof. In view of Proposition 3.3.14, this Proposition follows from
curl φ =∗ f.
This is a straightforward computation using the fact that φ is both symmetric and traceless.
3.4 Estimates for Ricci Coefficients
In this section, we prove estimates for the Ricci coefficients and their first four angular derivatives
in L2 (S).
Before we proceed to prove estimates for the Ricci coefficients, we first make a preliminary
observation regarding the bounds for ψ and its derivatives and products. These follow directly
from the bootstrap assumption (3.3.1).
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X X
kui1 +i2 +1 ∇i1 ψ i2 +1 kL2 (Su,u ) ≤ kui1 +1 ∇i1 ψkL2 (Su,u )
i1 +i2 ≤4 i1 ≤4
and
X X
kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ (Su,u ) ≤ kui1 +2 ∇i1 ψkL∞ (Su,u ) .
i1 +i2 ≤2 i1 ≤2
226
In particular, by (3.3.1), we have
1 1
X X
kui1 +i2 +1 ∇i1 ψ i2 +1 kL2 (Su,u ) + kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ (Su,u ) ≤ δa 2 b 4 .
i1 +i2 ≤4 i1 +i2 ≤2
Proof. For the L2 (Su,u ) estimates, we have
X
kui1 +i2 +1 ∇i1 ψ i2 +1 kL2 (Su,u )
i1 +i2 ≤4
X X X
≤( kui1 +1 ∇i1 ψkL2 (Su,u ) )( kui3 +1 ∇i3 ψkiL2∞ (Su,u ) )
i1 ≤4 i2 ≤4 i3 ≤2
X
≤ kui1 +1 ∇i1 ψkL2 (Su,u ) ,
i1 ≤4
since by (3.3.1) and Sobolev embedding, we have
1 1
X
i3 +1 i3 δa 2 b 4 1
ku ∇ ψkL∞ (Su,u ) ≤ ≤ 3.
|u| b4
i3 ≤2
Similarly, we have
X
kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ (Su,u )
i1 +i2 ≤2
X X X
≤( kui1 +2 ∇i1 ψkL∞ (Su,u ) )( kui3 +1 ∇i3 ψkiL2∞ (Su,u ) )
i1 ≤2 i2 ≤2 i3 ≤2
X
≤ kui1 +2 ∇i1 ψkL∞ (Su,u ) .
i1 ≤2
We now proceed to the estimates for the Ricci coefficients and their derivatives. We will first
bound the terms we denote as ψ, first with χ̂ (Proposition 3.4.2), then trχ (Proposition 3.4.3) and
ω (Proposition 3.4.4). We will then turn to the estimates for ψ in Proposition 3.4.6. We begin with
χ̂:
227
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1
X
kui ∇i χ̂kL2 (Su,u ) ≤ a 2 .
i≤4
Proof. We use the null structure equation
1
∇3 χ̂ + trχχ̂ = ∇η + ψψ + ψψ.
2
Commuting this equation with i angular derivatives, we have
i+1
∇3 ∇i χ̂ + trχ∇i χ̂
2
X X
=∇i+1 η + ∇i1 ψ i2 +2 + ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 =i i1 +i2 +i3 =i
X 1 i1 i2 +1 i3
+ ∇ ψ ∇ ψ.
i1 +i2 +i3 =i−1
u
i+1
We apply Proposition 3.3.5 with λ0 = 2 , which shows that the quantity kui ∇i χ̂kL∞ ∞ 2
u Lu L (Su,u )
can be controlled by the sum of its initial value and the kui · kL∞ 1 2
u Lu L (Su,u )
norm of the right hand
side. We now estimate each of the terms on the right hand side of the equation for i ≤ 4. We first
control the linear term in η for i ≤ 3:
X X 1
kui ∇i+1 ηkL1u L2 (Su,u ) ≤ k kL1 kui+2 ∇i+1 ψkL∞ 2
u L (Su,u )
|u|2 u
i≤3 i≤3
1 1
δa 2 b 4
≤ .
|u|
using the bootstrap assumption (3.3.1).
For the highest order derivative, i.e., when i = 4, we have
1 1 1 1 1
1 δ 2 a2 δ 2 a2 b4
ku4 ∇5 ηkL1u L2 (Su,u ) ≤ k kL2u ku5 ∇5 ηkL2u L2 (Su,u ) ≤ 1 Õ5,2 ≤ 1 .
|u| |u| 2 |u| 2
228
We then control the second and third terms together. Here, we use the estimates derived in
Proposition 3.4.1.
X X
k ui ∇i1 ψ i2 +1 ∇i3 (ψ, ψ)kL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
X X
≤ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L2 (Su,u ) kui3 +1 ∇i3 (ψ, ψ)kL∞ ∞
u L (Su,u )
i1 +i2 ≤4 i3 ≤2
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L∞ (Su,u ) kui3 +1 ∇i3 (ψ, ψ)kL∞ 2
u L (Su,u )
(3.4.1)
i1 +i2 ≤2 i3 ≤4
1 1 1 1
δa 2 b 4 1 1 δa 2 b 4
≤ (a 2 b 4 + )
|u| |u|
1
δab 2
≤ .
|u|
Finally, we control the last term by
X X
k ui−1 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i−1
X X
≤ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L2 (Su,u ) kui3 +1 ∇i3 ψkL∞ ∞
u L (Su,u )
i1 +i2 ≤3 i3 ≤1
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L∞ (Su,u ) kui3 +1 ∇i3 ψkL∞ 2
u L (Su,u )
(3.4.2)
i1 +i2 ≤1 i3 ≤3
1 1 1 1
δa 2 b 4 1 1 δa 2 b 4
≤ (a 2 b 4 + )
|u| |u|
1
δab 2
≤ .
|u|
1
We now apply the condition |u| ≥ δa 2 b. For b sufficiently large and a ≥ b, it is easy to see that all
the terms above are of size

1 3 1
δ 2 a4 a2 1
≤ 1 ≤ 1 ≤ a2 .
|u| 2 b 2
Recalling that initially we have
1
X
k∇i χ̂0 kL∞ 2
u L (S0,u )
≤ a2
i≤4
229
and using the above estimates, we get
1
X
kui ∇i χ̂kL∞ 2
u L (Su,u )
≤ a2 .
i≤4
We next prove the estimates for trχ and its derivatives. It will be useful to show not only that
∇i trχ obey the required estimates for ∇i ψ but also to obtain a slightly more refined bound as
follows:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 2
kui+1 ∇i (trχ − )kL2 (Su,u ) ≤ δa.
|u|
i≤4
Proof. Using the null structure equation, we have
2
∇4 (trχ − ) = χ̂χ̂ + ωtrχ + trχtrχ.
|u|
2
Notice that we have put in the term − |u| in the equation for ∇4 trχ. The ∇4 derivative of this term
2
is 0. We put in this additional term because the initial value for trχ − |u| on H 0 is 0. Commuting
this equation with i angular derivatives, we have
2
∇4 ∇i (trχ − )
|u|
X X
= ∇i1 ψ i2 ∇i3 χ̂∇i4 χ̂ + ∇i1 ψ i2 ∇i3 ψ∇i4 trχ
i1 +i2 +i3 +i4 =i i1 +i2 +i3 +i4 =i
X 1 i1 i2 i3
+ ∇ ψ ∇ ψ.
i1 +i2 +i3 =i
|u|
2
We now use Proposition 3.3.4 to bound kui+1 ∇i (trχ − |u| )kLu Lu L (Su,u )
∞ ∞ 2 by controlling the right
hand side in the kui+1 · kL∞ 1 2

u Lu L (Su,u )
norm.
We first bound each of the three terms in the case where i2 = 0. We first consider the contri-
230
∇i1 ψ i2 ∇i3 χ̂∇i4 χ̂ where i2 = 0 (and as a consequence i1 = 0). In this
P
bution from i1 +i2 +i3 +i4 =i
term, we will see estimates that are “borderline”, but we will use the fact that the bound derived
for ∇i χ̂ in Proposition 3.4.2 is independent of the bootstrap assumption (3.3.1) and does not suffer
1
a loss of b 4 . More precisely, we have the bound
X X
k ui+1 ∇i3 χ̂∇i4 χ̂kL1u L2 (Su,u )
i≤4 i3 +i4 =i
X X
≤δ kui3 +1 ∇i3 χ̂kL∞ ∞ ∞
u Lu L (Su,u )
kui4 ∇i4 χ̂kL∞ ∞ 2
u Lu L (Su,u )
i3 ≤2 i4 ≤4
≤δa.
∇i1 ψ i2 ∇i3 ψ∇i4 trχ with i2 = 0.

P
We now consider the contribution from i1 +i2 +i3 +i4 =i
X X
k ui+1 ∇i3 ψ∇i4 trχkL1u L2 (Su,u )
i≤4 i3 +i4 =i
X X
≤δ kui3 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
kui4 +1 ∇i4 trχkL∞ ∞ 2
u Lu L (Su,u )
i3 ≤2 i4 ≤4
X X
+δ kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
kui4 ∇i4 trχkL∞ ∞ ∞
u Lu L (Su,u )
i3 ≤4 i4 ≤4
1 1
δa b X i+1 i
2 4 2 1 1
≤ ku ∇ (trχ − )kL∞ ∞ 2
u Lu L (Su,u )
+ δa 2 b 4 ,
|u| |u|
i≤4
where we have used the bootstrap assumption (3.3.2) together with Sobolev embedding in Propo-
sition 3.3.9.
1 i1 i2 i3
P
The contribution of the third term, i.e., i1 +i2 +i3 =i |u| ∇ ψ ∇ ψ, where i2 = 0 can be con-
trolled by
1 1
X X
kui ∇i ψkL1u L2 (Su,u ) ≤ δ kui ∇i ψkL∞ ∞ 2
u Lu L (Su,u )
≤δa 2 b 4 ,
i≤4 i≤4
using the bootstrap assumption (3.3.2).
We now move to the terms where i2 ≥ 1. It turns out that the ψ factors provide extra smallness
and we can use the bootstrap assumption (3.3.2) together with Proposition 3.4.1 to deal with the
231
first two terms2 to get
X X
k ui+1 ∇i1 ψ i2 +1 ∇i3 ψ∇i4 ψkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 +i4 =i−1
δ X X
≤ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
|u|
i1 +i2 ≤3 i3 ≤2
i1 ≤2
X
× kui4 ∇i4 ψkL∞ ∞ 2
u Lu L (Su,u )
i4 ≤3
δ
+ ku4 ∇3 ψkL∞ ∞ 2
u Lu L (Su,u )
kuψkL∞ ∞ ∞
u Lu L (Su,u )
kuψkL∞ ∞ ∞
u Lu L (Su,u )
|u|
3 3
δ2 a 2 b 4
≤ .
|u|
The third term can be handled in a similar way in the case i2 ≥ 1 using Proposition 3.4.1 and the
bootstrap assumption (3.3.2)
X X
k ui ∇i1 ψ i2 +1 ∇i3 ψkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i−1
δ X X
≤ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
kui3 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
|u|
i1 +i2 ≤3 i3 ≤3
i1 ≤2
δ
+ ku4 ∇3 ψkL∞ ∞ 2
u Lu L (Su,u )
kuψkL∞ ∞ ∞
u Lu L (Su,u )
|u|
1
δ 2 ab 2
≤ .
|u|
2
Recall that the initial data for trχ − |u| is vanishing. Therefore, by Proposition 3.3.4 and using
1
the condition |u| ≥ δa 2 b, we have
X 2
kui+1 ∇i (trχ − )kL∞ ∞ 2
u Lu L (Su,u )
|u|
i≤4
1 X 2
≤δa + 3 kui+1 ∇i (trχ − )kL∞ ∞ 2
u Lu L (Su,u )
.
b 4 |u|
i≤4
2 Notice that we have relabeled i to simplify the notation. We have also used the schematic notation to write χ̂
2
and trχ as ψ.
232
For b sufficiently large, we can subtract the last term from both sides to get
X 2
kui+1 ∇i (trχ − )kL∞ ∞ 2
u Lu L (Su,u )
≤ δa.
|u|
i≤4
We next prove the desired estimates for ω and its derivatives. We will show that it obeys a
slightly better estimate than a general component ψ. Moreover, the bound improves if we take at
least one angular derivative.
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1 3
X
i i δ 2 a4
ku ∇ ωkL2 (Su,u ) ≤ 1 + 1 .
i≤4
|u| 2
If i 6= 0, we get the improved bound
1 3
X δ 2 a4
kui ∇i ωkL2 (Su,u ) ≤ 1 .
1≤i≤4
|u| 2
Proof. We use the following schematic null structure equation for ω:
∇3 ω = K + ψψ + ψψ + trχtrχ.
Commuting it with angular derivative for i times, we have
i
∇3 ∇i ω + trχ∇i ω
2
X X X
= ∇i1 ψ i2 ∇i3 K + ∇i1 ψ i2 +2 + ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 +i3 ≤i i1 +i2 =i i1 +i2 +i3 =i
X 1 i1 i2 +1 i3 1
+ ∇ ψ ∇ ψ + ∇i trχ.
i1 +i2 +i3 =i−1
u u
We apply Proposition 3.3.5 with λ0 = 2i . In particular, since the initial data for ω vanishes, we
can estimate kui−1 ∇i ωkL∞ ∞ 2

u Lu L (Su,u )
by the kui−1 · kL∞ 1 2
u Lu L (Su,u )
norm of the right hand side. To
233
estimate each of the terms in the equation, we note that all terms except for the K term and the
1 i
u ∇ trχ term have been controlled in the proof of Proposition 3.4.2 and 3.4.3. More precisely, by
(3.4.1) and (3.4.2), we have
1 1 1 1 3
X X 1 δ 2 a2 b4 δ 2 a4
kui−1 Fi kL∞ 1 2
u Lu L (Su,u )
≤ kui Fi kL∞ L 1 L2 (S
u,u ) ≤ 3 ≤ 3 ,
|u| u u
|u| 2 |u| 2
i≤4 i≤4
where Fi is defined to be
X X
Fi = ∇i1 ψ i2 +2 + ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 =i i1 +i2 +i3 =i
X 1 i1 i2 +1 i3
+ ∇ ψ ∇ ψ.
i1 +i2 +i3 =i−1
u
1 i
There are two remaining terms: the term with K and the term u ∇ trχ. We first estimate the term
containing the Gauss curvature K. We split up the term into
1 1
K ≤ (K − 2
) + 2.
|u| |u|
1
For the term with (K − |u|2 ), if i2 = 0, we have
X 1
kui−1 ∇i (K − )kL1u L2 (Su,u )
|u|2
i≤4
X 1
≤ kui+1 ∇i (K − )kL2u L2 (Su,u ) ku−2 kL2u
|u|2
i≤4
1 1 1 1 3
δ 2 a2 b4 δ 2 a4
≤ 3 ≤ 3 ,
|u| 2 |u| 2
where in the second inequality, we have used the bootstrap assumption (3.3.3). For i2 ≥ 1, we get3
3 Notice that we have relabeled i2
234
the better estimate
X X 1
k ui−1 ∇i1 ψ i2 +1 ∇i3 (K − )kL1u L2 (Su,u )
i1 +i2 +i3 =i−1
|u|2
i≤4
X X 1
≤ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +2 ∇i3 (K − )kL2u L∞ (Su,u ) ku−3 kL2u
|u|2
i1 +i2 ≤4 i3 ≤2
X
+ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞
u L (Su,u )
i1 +i2 ≤2
X 1
× kui3 +1 ∇i3 (K − )kL2u L2 (Su,u ) ku−3 kL2u
|u|2
i3 ≤4
X X 1
≤ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 (K − )kL2u L2 (Su,u ) ku−3 kL2u
|u|2
i1 +i2 ≤4 i3 ≤4
1 1 1 1
δa b2 4 1 1 δ a
2 2
≤ 5 · δ 2 a2 R ≤ 3 ,
|u| 2 |u| 2
where in the second inequality, we have used the Sobolev embedding in Proposition 3.3.9; in the
third inequality, we have applied Proposition 3.4.1 and in the last estimate, we have used the
1
bootstrap assumption (3.3.3) and |u| ≥ δa 2 b.
1
We then move to the contribution arising from |u|2 . Notice that for this term, the only possibility
for having i2 = 0 is when i = 0. In this case, the term can be controlled by
1 1
ku−1 kL1 L2 (Su,u ) ≤ . (3.4.3)
|u|2 u |u|
For i2 ≥ 1, we have
X X 1
k ui−1 ∇i1 ψ i2 +1 kL1 L2 (Su,u )
i1 +i2 =i−1
|u|2 u
i≤4
X
≤ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
ku−3 kL1u
i1 +i2 ≤3
1 1 3
δa 2 b 4 δa 4
≤ 2
≤ ,
|u| |u|2
where in the second inequality, we have used the bootstrap assumption (3.3.1). Combining all the
235
above estimates, we have
1 3
X X 1 1 δ 2 a4
k ui−1 ∇i1 ψ i2 +1 ∇i3 (K − 2
)kL1u L2 (Su,u ) ≤ + 3 . (3.4.4)
i1 +i2 +i3 =i−1
|u| |u| |u| 2
i≤4
1
Notice moreover that the appearance of the term |u| is only due to the contribution of (3.4.3) and
is only present when i = 0. We now estimate the remaining term by:
1 3
X δa 1 δ 2 a4 1
kui−2 ∇i trχkL1u L2 (Su,u ) ≤ 2
+ ≤ 3 + , (3.4.5)
|u| |u| |u| 2 |u|
i≤4
2
using the bound for ∇i trχ in Proposition 3.4.3. Moreover, for i ≥ 1, since ∇i trχ = ∇i (trχ − |u| ),
we have the improved bound
1 3
X δ 2 a4
kui−2 ∇i trχkL1u L2 (Su,u ) ≤ 3
1≤i≤4
|u| 2
using Proposition 3.4.3.
Collecting the above estimates, we have
1 3
X 1 δ 2 a4
kui−1 ∇i ωkL2 (Su,u ) ≤ + 3 .
|u| |u| 2
i≤4
Multiplying by |u|, we get

1 3
X δ 2 a4
kui ∇i ωkL2 (Su,u ) ≤ 1 + 1 .
i≤4
|u| 2
Moreover, if i 6= 0, the worst terms in (3.4.4) and (3.4.5) contributing to the bound 1 are absent.
Therefore, we have
1 3
X δ 2 a4
kui ∇i ωkL2 (Su,u ) ≤ 1 .
1≤i≤4
|u| 2
We summarize the estimates that we have already proved for ψ and its derivatives:
236
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 1
1 kui ∇i ψkL2 (Su,u ) ≤ 1.
i≤4
a2
We now estimate the L2 (Su,u ) norms of the first four derivatives of the remaining Ricci coeffi-
cients, i.e., the Ricci coefficients that we call ψ:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 1
1 kui+1 ∇i ψkL2 (Su,u ) ≤ 1 + Õ5,2 + R.
i≤4
δa 2
We also have the more precise bound
X 1 1
1 kui+1 ∇i ψkL2 (Su,u ) ≤ 1 + 1 1 ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
+ R,
i≤4
δa 2 δ a22
1
i.e., the only dependence on Õ5,2 is through the term 1 1 ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
.
δ2 a2
Proof. In order to unify the exposition, we will not consider the equation ∇3 η, but will instead
use the equation ∇4 ∇ log Ω. Since η = −η + 2∇ log Ω, the desired estimate for η can be recovered
from the bounds for η and ∇ log Ω.
For ψ ∈ {η, ∇ log Ω, trχ + u2 , χ̂, ω}, we have the schematic transport equation
1
∇4 ψ = β + ∇ω + K + ∇η + ψψ + ψψ + ψ.
u
Notice moreover that by the assumption of Minkowskian data on the initial incoming cone, all of
these quantities ψ are initially 0.
237
Commuting the above equation with angular derivatives, we get
∇4 ∇i ψ
X X
=∇i β + ∇i+1 (ω, η) + ∇i1 ψ i2 ∇i3 K + ∇i1 ψ i2 +1 ∇i3 (ψ, ψ)
i1 +i2 +i3 =i i1 +i2 +i3 =i
X 1 i1 i2 i3
+ ∇ ψ ∇ ψ.
i1 +i2 +i3 =i
u
By Proposition 3.3.4 and the triviality of ψ on H 0 , in order to estimate the quantity kui ∇i ψkL∞ 2
u L (Su,u )
,
it suffices to bound the kui · kL1u L2 (Su,u ) norm of the right hand side. We now estimate each of the
terms in the equation. We first control the β term
1 1
X
i i δ 2 X i+1 i δa 2
ku ∇ βkL1u L2 (Su,u ) ≤ ku ∇ βkL2u L2 (Su,u ) ≤ R. (3.4.6)
|u| |u|
i≤4 i≤4
We now turn to the term ∇i+1 ω. For i ≤ 3, we apply the estimates in Proposition 3.4.5, while for
i = 4, we use Õ5,2 to get
X
kui ∇i+1 ωkL1u L2 (Su,u )
i≤4
1
δ X i+1 i+1 δ2 5 5
≤ ku ∇ ψkL∞ 2
u L (Su,u )
+ ku ∇ ωkL2u L2 (Su,u ) (3.4.7)
|u| |u|
i≤3
1
δa 2
≤ (1 + Õ5,2 ).
|u|
For the ∇i+1 η term, we have
X
kui ∇i+1 ηkL1u L2 (Su,u )
i≤4
1
δ X i+2 i+1 δ2
≤ ku ∇ ψk ∞ 2
Lu L (Su,u ) + ku6 ∇5 ηkL2u L2 (Su,u ) (3.4.8)
|u|2 |u|2
i≤3
1 1 3 1 3 1
δ2 a 2 b 4 δ2 a 4 b 4 δ2 a 4 b 4
≤ + ≤ ,
|u|2 |u|2 |u|2
238
where we have used the bootstrap assumptions (3.3.1) and (3.3.3).
We control the term containing the Gauss curvature K as follows:
X X
k ui ∇i1 ψ i2 ∇i3 KkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
1
δ2 X 1
≤ kui+2 ∇i (K − )kL2u L2 (Su,u ) + δku−2 kL2 (Su,u )
|u|2 |u|2
i≤4
1
δ2 X X 1
+ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞
u L (Su,u )
kui3 +2 ∇i3 (K − )kL2u L2 (Su,u )
|u|3 |u|2
i1 +i2 ≤2 i3 ≤4
1
δ 2 X
(3.4.9)
+ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
|u|3
i1 +i2 ≤4
X 1 1
×( kui3 +1 ∇i3 (K − 2
)kL2u L∞ (Su,u ) + δ 2 |u|)
|u|
i3 ≤2
3 1 5 1 1 1
2
δ a b4 δ4 δ3 a 4 b 2 δ2 a 2 b 4
≤ + + +
|u|2 |u| |u|3 |u|2
3 1
δ2 a 4 b 4 δ
≤ 2
+ ,
|u| |u|
where in the second inequality we have used the bootstrap assumptions (3.3.1) and (3.3.3), and in
1
the final inequality we have used |u| ≥ δa 2 b and b ≤ a. Moreover, notice that the contribution
δ 1
for |u| comes from only from the term i = 0. For i ≥ 1, since ∇i K = ∇i (K − |u|2 ), we have the
improved bound
3 1
X X
i i1 i2 i3 δ2 a 4 b 4
k u ∇ ψ ∇ KkL1u L2 (Su,u ) ≤ . (3.4.10)
i1 +i2 +i3 =i
|u|2
1≤i≤4
239
The fourth term can be bounded as follows:
X X
k ui ∇i1 ψ i2 +1 ∇i3 (ψ, ψ)kL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
δ X X
≤ 2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 (ψ, ψ)kL∞ ∞
u L (Su,u )
|u|
i1 +i2 ≤4 i3 ≤2
(3.4.11)
δ X
i1 +i2 +2 i1 i2 +1
X
i3 i3
+ 2 ku ∇ ψ kL∞ ∞
u L (Su,u )
ku ∇ (ψ, ψ)kL∞ 2
u L (Su,u )
|u|
i1 +i2 ≤2 i3 ≤4
1 1 1
δ 2 ab 4 δ 3 ab 2 δ 2 ab 4
≤ 2
+ 3
≤ ,
|u| |u| |u|2
where in the last line we have used Proposition 3.4.5 to control ψ and used Proposition 3.4.1 to
control the product of ψ, as well as using Sobolev embedding in Proposition 3.3.9. We then control
the final term as follows:
X X
k ui−1 ∇i1 ψ i2 ∇i3 ψkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
δ X i i
≤ ku ∇ ψkL∞ 2
u L (Su,u )
|u|
i≤4
δ X X
+ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 ψkL∞ ∞
u L (Su,u ) (3.4.12)
|u|2
i1 +i2 ≤3 i3 ≤2
δ X X
+ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞
u L (Su,u )
kui3 ∇i3 ψkL∞ 2
u L (Su,u )
|u|2
i1 +i2 ≤2 i3 ≤4
1 1
2
δa 2 δ ab 4
≤ + ,
|u| |u|2
where as before we have used Proposition 3.4.5 to control ψ and used Proposition 3.4.1 to estimate
the product of ψ, as well as using Sobolev embedding in Proposition 3.3.9.

1
Therefore, applying Proposition 3.3.4, using the estimates above and using |u| ≥ δa 2 b, we get
1
X δa 2
kui ∇i ψkL∞ 2
u L (Su,u )
≤ (1 + Õ5,2 + R).
|u|
i≤4
Moreover, notice that the term Õ5,2 only comes from controlling ∇5 ω in (3.4.7). Therefore, we
240
have the improved estimate
1
X δa 2 1
kui ∇i ψkL2 (Su,u ) ≤ (1 + 1 1 ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
+ R).
|u| δ 2 a2
i≤4
2
From the proof of Proposition 3.4.6, we also get the following additional bound for ∇i (trχ + |u| )
for i ≥ 1:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
3 3
X 2 δ 2 a4
kui ∇i (trχ + )kL2 (Su,u ) ≤ 3 .
|u| |u| 2
1≤i≤4
2
Proof. Notice that trχ + |u| satisfies the equation
2 1
∇4 (trχ + ) = K + ∇η + ψψ + ψψ + (trχ, ω).
|u| u
This equation can be derived from the usual equation for ∇4 trχ and noting that ∇4 u = 0. In
particular, compared to the equation for a general component ψ, we do not have the terms β, ∇ω
1
and u χ̂.
Commuting the above equation with angular derivatives, we get
2
∇4 ∇i (trχ + )
|u|
X X
=∇i+1 η + ∇i1 ψ i2 ∇i3 K + ∇i1 ψ i2 +1 ∇i3 (ψ, ψ)
i1 +i2 +i3 =i i1 +i2 +i3 =i
X 1 i1 i2 i3
+ ∇ ψ ∇ (trχ, ω).
i1 +i2 +i3 =i
u
We now revisit the proof of Proposition 3.4.6. Notice that except for the terms in (3.4.6), (3.4.7),
1
δ 2 ab 4
(3.4.9) and (3.4.12), the error terms in the proof of Proposition 3.4.6 satisfy the better bound |u|2 .
2
Since there are no ∇i β and ∇i+1 ω terms in the equation for ∇i (trχ + |u| ), the contributions of
241
1
δa 2 δ
|u| (1 + Õ3,2 + R) from (3.4.6) and (3.4.7) are absent. Moreover, since we have i ≥ 1, the term |u|
from (3.4.9) is also absent (and we have the bound (3.4.10) instead). We now revisit the estimate
in (3.4.12), using the fact that we only have trχ and ω and that χ̂ is absent:
X X
k ui−1 ∇i1 ψ i2 ∇i3 (trχ, ω)kL1u L2 (Su,u )
1≤i≤4 i1 +i2 +i3 =i
δ X
≤ kui ∇i (trχ, ω)kL∞ 2
u L (Su,u )
|u|
1≤i≤4
δ X X
+ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 (trχ, ω)kL∞ ∞
u L (Su,u )
|u|2
i1 +i2 ≤4 i3 ≤2
δ X X
+ 2 kui1 +i2 ∇i1 +2 ψ i2 +1 kL∞ ∞
u L (Su,u )
kui3 ∇i3 (trχ, ω)kL∞ 2
u L (Su,u )
|u|
i1 +i2 ≤2 i3 ≤4
3 3 1 3 3
2 2
δ a δ a 2 δ ab
4 δ a 4 2 4
≤ + 3 + ≤ 3 ,
|u|2 |u| 2 |u| 2
|u| 2
where in addition to using Proposition 3.4.1, we have used the improved bounds for ∇i trχ and
∇i ω derived in Propositions 3.4.3 and 3.4.4 respectively . Combining this with (3.4.8), (3.4.10) and
(3.4.11), we obtain the desired conclusion.

2
Besides ∇i (trχ + |u| ), we also need an improved bound for ∇i log Ω for 1 ≤ i ≤ 4:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
3 3
X
i−1 i δ 2 a4
ku ∇ log ΩkL∞ ∞ 2
u Lu L (Su,u )
≤ 3 .
1≤i≤4
|u| 2
Proof. Recall that
X
∇4 ∇i log Ω = ∇i ω + ∇i1 ψ i2 +1 ∇i3 ψ.
i1 +i2 +i3 =i−1
We would like to control ∇i log Ω in the norm kui−1 ∇i log ΩkL∞ ∞ 2

u Lu L (Su,u )
. By Proposition 3.3.4,
we thus have to estimate the right hand side in the norm kui−1 · kL∞ 1 2
u Lu L (Su,u )
. The first term can
be bounded with the estimate for ∇i ω for 1 ≤ i ≤ 4 in Proposition 3.4.4. We also need to take
242
advantage of the improved bound that we achieve for i 6= 0. More precisely, we have
3 3
X δ 2 a4
kui−1 ∇i ωkL∞ 1 2
u Lu L (Su,u )
≤ 3 .
1≤i≤4
|u| 2
The second term can be controlled with the aid of (3.4.11):
X X
k ui−1 ∇i1 ψ i2 +1 ∇i3 ψkL∞ 1 2
u Lu L (Su,u )
1≤i≤4 i1 +i2 +i3 =i−1
X X
= k ui ∇i1 ψ i2 +1 ∇i3 ψkL∞ 1 2
u Lu L (Su,u )
i≤3 i1 +i2 +i3 =i
3 1 3 3
2
δ a b4 δ 2 a4
4
≤ 2
≤ 3
|u| |u| 2
1
since |u| ≥ δa 2 b.
1
We also note that the ∇4 equation for K − |u|2 contains exactly the same type of terms as the
1
∇4 equation for ∇ψ. Therefore, ∇i (K − |u|2 ) obeys the same estimates as ∇i+1 ψ for i ≤ 3. More
precisely, we have
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 1 1
kui+2 ∇i (K − )kL∞ ∞ 2
u Lu L (Su,u )
≤ δa 2 (1 + Õ5,2 + R).
|u|2
i≤3
In particular,
X 1 1
kui+1 ∇i (K − 2
)kL∞ ∞ 2
u Lu L (Su,u )
≤ 3.
|u| b4
i≤3
Proof. As mentioned above, by repeating the proof of Proposition 3.4.6, we obtain the bound
X 1 1
1 kui+2 ∇i (K − )kL∞ ∞ 2
u Lu L (Su,u )
≤ 1 + Õ5,2 + R.
δa 2 |u|2
i≤3
243
1
δa 2
Multiplying by |u| , we obtain
1 1
X 1 δa 2 b 4 1
kui+1 ∇i (K − 2
)kL∞ ∞ 2
u Lu L (Su,u )
≤ ≤ 3.
|u| |u| b4
i≤3
3.5 Elliptic estimates for the fifth derivatives of the Ricci
coefficients
We now estimate the fifth angular derivatives of the Ricci coefficients. As in [8], [12], [7] and [15],
this is achieved by a combination of transport estimates and elliptic estimates. This is because
curvature enters as source terms in the transport equations for the Ricci coefficients. As a result,
with the bounds for four derivatives of the curvature components, we can only obtain estimates for
the four derivatives of the Ricci coefficients. The standard approach is therefore to control some
chosen linear combination of some derivatives of the Ricci coefficients and curvature components
via transport estimates. We then recover the bounds for the remaining highest derivatives of the
Ricci coefficients by L2 elliptic estimates on Su,u .
We now begin with the estimates for trχ and χ̂:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
3 1
ku6 ∇5 trχkL2u L2 (Su,u ) ≤ δ 2 ab 4
and
1 1
ku5 ∇5 χ̂kL2u L2 (Su,u ) ≤ δ 2 a 2 (1 + R).
Proof. Consider the following equation:
1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ
2
244
Commuting with angular derivatives for i times, we get
X
∇4 ∇5 trχ = ∇i1 ψ i2 ∇i3 ψ∇i4 ψ.
i1 +i2 +i3 +i4 =5
We will apply Proposition 3.3.4 to estimate ku6 ∇5 trχkL∞ ∞ 2

u Lu L (Su,u )
by the ku6 · kL∞ 1 2
u Lu L (Su,u )
norm of the right hand side. We separately estimate the terms with 5 derivatives on ψ and the
remaining terms.
X
ku6 ∇i1 ψ i2 ∇i3 ψ∇i4 ψkL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 +i4 =5
1
≤δ kuψkL∞
2 ∞
u L (Su,u )
ku5 ∇5 ψkL∞ 2 2
u Lu L (Su,u )
X X
+δ kui+1 ∇i ψkL∞ u L∞ L∞ (S
u u,u ) kui ∇i ψkL∞ ∞ 2
u Lu L (Su,u )
i1 ≤2 i2 ≤4
X
+δ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X X
× kui3 +1 ∇i3 ψkL∞ ∞
u L (Su,u )
u Lu L (Su,u )
i3 ≤2 i4 ≤4
X
+δ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X X
× kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
k∇i4 +1 ψkL∞ ∞ ∞
u Lu L (Su,u )
i3 ≤2 i4 ≤2
1
≤δab , 4
where we have used Proposition 3.4.1 to control the product of ψ and Proposition 3.4.5 to bound
ψ.
Recall that ∇5 trχ = 0 initially on H 0 . Therefore, we have
1
k|u|6 ∇5 trχkL∞ 2
u L (Su,u )
≤ δab 4 .
Taking L2 in u, we get
3 1
k|u|6 ∇5 trχkL2u L2 (Su,u ) ≤ δ 2 ab 4 . (3.5.1)
245
Recalling the schematic form of the Codazzi equation
1
div χ̂ − ∇trχ + β = ψψ,
2
we can apply the elliptic estimates in Proposition 3.3.15 to obtain
X X
ku5 ∇5 χ̂kL2 (Su,u ) ≤ kui ∇i trχkL2 (Su,u ) + kui+1 ∇i βkL2 (Su,u )
i≤5 i≤4
X X X
+ kui+1 ∇i1 ψ∇i2 ψkL2 (Su,u ) + kui ∇i χ̂kL2 (Su,u ) .
i≤4 i1 +i2 =i i≤4
Taking L2 in u, using the control of ψ from Proposition 3.4.5, the bound for ψ in (3.3.1), as well
as the bound (3.5.1) for ∇5 trχ that we have just achieved above, we obtain
ku5 ∇5 χ̂kL2u L2 (Su,u )

X X
≤ kui ∇i trχkL2u L2 (Su,u ) + kui+1 ∇i βkL2u L2 (Su,u )
i≤5 i≤4
X X X
+ kui+1 ∇i1 ψ∇i2 ψkL2u L2 (Su,u ) + kui ∇i χ̂kL2u L2 (Su,u )
i≤4 i1 +i2 =i i≤4
3 1
δ ab
2 4
1 1 1 1
≤ + δ 2 a2 + δ 2 a2 R
|u|
1 1
≤δ 2 a 2 (1 + R),
1
since |u| ≥ δa 2 b.
We now turn to the estimates for the highest derivative of ω. ∇5 ω obeys the same bounds
as ∇5 χ̂. However, the proof will proceed in a slightly different manner as we need to control a
transport equation in the ∇3 direction.
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1 1
ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 2 (1 + R).
246
Proof. Define ω † to be the solution to
1
∇3 ω † = σ̌,
2
with zero initial data on H1 and let
1
κ := ∇ω +∗ ∇ω † − β.
2
It is easy to see using Proposition 3.3.5 and the bootstrap assumptions (3.3.1) and (3.3.3) that we
have
1 3
X δ 2 a4 1
kui ∇i ω † kL∞ ∞ 2
u Lu L (Su,u )
≤1+ 1 ≤ a2 .
i≤4
|u| 2
The proof of this estimate is similar to that for ω in Proposition 3.4.4. Hence for ω † and its first
four derivatives, ω † obeys the same estimates as ψ as given in Proposition 3.4.5. Therefore, in the
remainder of the proof of this proposition, we will include ω † as one of the ψ terms and use the
notation ψ ∈ {trχ, χ̂, ω, ω † }.
With this modified notation, κ obeys the following transport equation
1
∇3 κ + trχκ
2
X 2 1 1
= ψ i1 ∇i2 (trχ + , χ̂, ω)∇i3 ψ + ψ∇(η, η) + ψψ + β + ψK.
i +i +i =1
|u| |u| |u|
1 2 3
Commuting with angular derivatives for 4 times, and applying the schematic Codazzi equation for
β:
X
β= ψ i1 ∇i2 ψ,
i1 +i2 =1
247
we have
5
∇3 ∇4 κ + trχ∇4 κ
2
X 2 X
= ∇i1 ψ i2 ∇i3 (trχ + , χ̂, ω)∇i4 ψ + ∇i1 ψ i2 +1 ∇i3 +1 (η, η)
i +i +i +i =5
|u| i
1 2 3 4 1 +i2 +i3 =4
1 4 1 X
+ ∇ β+ ∇i1 ψ i2 +1 ∇i3 ψ
|u| |u| i
1 +i2 +i3 =4
X
+ ∇i1 ψ i2 ∇i3 ψ∇i4 K.
i1 +i2 +i3 +i4 =4
We estimate ∇4 κ using Proposition 3.3.5. Since we will only need to estimate ∇4 κ after integrated
in u, we will directly control it in the ku4 ∇4 κkL2u L∞ 2

u L (Su,u )
norm. Applying Proposition 3.3.5 with
λ0 = 52 , it suffices to estimate the initial data and the ku4 · kL2u L1u L2 (Su,u ) norm of the right hand
side. We now estimate each of the terms in the equation.
For the term i1 +i2 +i3 +i4 =5 ∇i1 ψ i2 ∇i3 (trχ + |u|
2
, χ̂, ω)∇i4 ψ, we estimate separately the contri-
P
butions with the highest order derivatives. These include the case where there are 5 derivatives on
2
(trχ + |u| , χ̂, ω) and the case where there are 5 derivatives on ψ. In the first case, we have
2
ku4 ∇5 (trχ + , χ̂, ω)ψkL2u L1u L2 (Su,u )
|u|
1 2
≤δ 2 kuψkL∞ ∞ ∞
u Lu L (Su,u )
ku−2 kL2u ku5 ∇5 (trχ + , χ̂, ω)kL∞ 2 2
u Lu L (Su,u )
|u|
1 1 3 1
1 1 1 δa 4 b 4 δ 2 ab 4
≤δ 2 · a 2 · 3 · 1 ≤ ,
|u| 2 |u| 2 |u|2
where we have used the estimate in Proposition 3.4.5 and the bootstrap assumption (3.3.3).
248
In the second case, we have
2
ku4 (trχ + , χ̂, ω)∇5 ψkL2u L1u L2 (Su,u )
|u|
≤ku2 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−3 kL1u
× (ku5 ∇5 (trχ, χ̂, ω)kL∞ 2 2

u Lu L (Su,u )
+ ku5 ∇5 ω † kL∞ 2 2
u Lu L (Su,u )
)
1 1 1 1 1
δa 2 b 4 · δ 2 a 2 b 4 1
≤ (1 + 1 1 ku5 ∇5 ω † kL∞ 2 2
u Lu L (Su,u )
)
|u|2 δ 2 a2
3 1
δ 2 ab 2 1
≤ (1 + 1 1 ku5 ∇5 ω † kL∞ 2 2
u Lu L (Su,u )
),
|u|2 δ a2
2
For the lower order term where each factor has at most 4 derivatives, we have
X
ku4 ∇i1 ψ i2 +1 ∇i3 ψkL2u L1u L2 (Su,u )
i1 +i2 +i3 =5
i1 ,i3 ≤4
1
X X
≤δ 2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
u Lu L (Su,u )
ku−3 kL1u
i1 +i2 ≤5 i3 ≤4
i1 ≤2
1
X
+ δ2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
(3.5.2)
i1 +i2 ≤5
i1 ≤4
X
× kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−3 kL1u
i3 ≤2
3 1
δ ab 4
2
≤ ,
|u|2
where we have used Propositions 3.4.1 and 3.4.5.
We now move to the term i1 +i2 +i3 =4 ∇i1 ψ i2 +1 ∇i3 +1 (η, η). As for the previous term, we will
P
first control the contribution where one of the Ricci coefficients has 5 derivatives. More precisely,
249
we have
ku4 ψ∇5 (η, η)kL2u L1u L2 (Su,u )
≤ku2 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−4 kL1u ku6 ∇5 (η, η)kL∞ 2 2
u Lu L (Su,u )
1 1 3 3 1 5 5 1
δa 2 b 4 · δ 2 a 4 b 4 δ 2 a4 b2
≤ 3
≤ ,
|u| |u|3
using the bootstrap assumptions (3.3.1) and (3.3.3).
For the lower order term, since ∇i ψ satisfies all the estimates that ∇i ψ for i ≤ 4, we can follow
the proof of (3.5.2) to get
3 1
X
4 i1 i2 +1 i3 δ 2 ab 4
ku ∇ ψ ∇ ψkL2u L1u L2 (Su,u ) ≤ .
i1 +i2 +i3 =5
|u|2
i1 ,i3 ≤4
We now move to the third term, which is the term containing β:
1 1
3 4 −2 5 4 δ 2 a2
ku ∇ βkL2u L1u L2 (Su,u ) ≤ku kL1u ku ∇ βkL∞ 2 2
u Lu L (Su,u )
≤ R.
|u|
For the fourth term, we have the estimate:
X
i1 +i2 +i3 =4
1
X X
≤δ 2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
u Lu L (Su,u )
ku−3 kL1u
i1 +i2 ≤2 i3 ≤4
1
X
+ δ2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X
× kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−3 kL1u
i3 ≤2
3 1
δ ab 4
2
≤ ,
|u|2
using the bounds in Propositions 3.4.1 and 3.4.5.
250
Finally, for the fifth term, i.e., the term containing K, we have
X
ku4 ∇i1 ψ i2 +1 ∇i3 KkL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
X
≤ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X 1
× kui3 +2 ∇i3 (K − )kL∞ 2 2
u Lu L (Su,u )
ku−4 kL1u
|u|2
i3 ≤4
X
+ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X 1 1
×( kui3 +3 ∇i3 (K − 2
)kL∞ 2 ∞
u Lu L (Su,u )
ku−4 kL1u + δ 2 ku−3 kL1u )
|u|
i3 ≤2
3 3 1 1 3 3
1 1 δ 2 a4 b4 δ2 δ 2 a4
≤δa 2 b 4 ( + 2) ≤ ,
|u|3 |u| |u|2
using Proposition 3.4.1 and the bootstrap assumption (3.3.3).
Combining the above estimates and using the bootstrap assumption (3.3.3), we have
1 1 1 1
δ 2 a2 δ 2 a2 1 5 5 †
ku4 ∇4 κkL2u L∞ 2
u L (Su,u )
≤ (1 + R) + 1 1 1 ku ∇ ω kL∞ 2 2
u Lu L (Su,u )
,
|u| b |u| δ a
2 2 2
which implies
1 1 1
ku5 ∇4 κkL2u L∞ 2
u L (Su,u )
≤ δ 2 a 2 (1 + R + 1 1 1 ku5 ∇5 ω † kL∞ 2 2
u Lu L (Su,u )
).
δ a2 b2
2
By the following div -curl system:
1
div ∇ω = div κ + div β,
2
curl ∇ω = 0,
1
2
div ∇ω † = 0.
251
and the elliptic estimates from Proposition 3.3.14, we have
ku5 ∇5 (ω, ω † )kL2 (Su,u )

X
≤ (kuj+1 ∇j κkL2 (Su,u ) + kuj+1 ∇j βkL2 (Su,u ) + kuj ∇j (ω, ω † )kL2 (Su,u ) )
j≤4
X
≤ku5 ∇4 κkL2 (Su,u ) + (kuj+1 ∇j βkL2 (Su,u ) + kuj ∇j (ω, ω † )kL2 (Su,u ) ).
j≤4
This implies, after taking L2 in u,
ku5 ∇5 (ω, ω † )kL2u L2 (Su,u )

1 1 1
≤δ 2 a 2 (1 + R + 1 1 1 ku5 ∇5 ω † kL∞ 2 2
u Lu L (Su,u )
).
δ a2 b2
2
For b sufficiently large, we can absorb the last term to the left hand side to get
1 1
ku5 ∇5 (ω, ω † )kL2u L2 (Su,u ) ≤δ 2 a 2 (1 + R).
In the following proposition, we prove the highest order derivative estimates for η. We need in
particular an improved bound for ∇i µ compared to ∇i+1 η for 1 ≤ i ≤ 4.
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
3 3
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 4 (1 + R)
and
1 1
ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 2 (1 + R).
For 1 ≤ i ≤ 4, the derivatives of the mass aspect function ∇i µ obey the following improved
estimates:
5 1
X
kui+3 ∇i µkL∞ ∞ 2
u Lu L (Su,u )
≤ δ2 a 4 b 4 .
1≤i≤4
252
Proof.
Define µ by
1
µ = −div η + K − .
|u|2
Thus η obeys the following elliptic system
1
div η = −µ + K − , curl η = σ̌
|u|2
and µ satisfies the equation
∇4 µ = ψ∇(η, η) + ψψ ψ + ψ∇(trχ, χ̂) + trχK.
Commuting with angular derivatives for i times with 1 ≤ i ≤ 4, we get
∇4 ∇ i µ
X
=ψ∇i+1 (η, η) + ψ∇i+1 (trχ, χ̂) + ∇i1 ψ i2 +1 ∇i3 ψ∇i4 ψ
i1 +i2 +i3 =i+1
i1 ,i3 ≤i
X
+ ∇i1 ψ i2 ∇i3 trχ∇i4 K.
i1 +i2 +i3 +i4 =i
Notice that ∇i µ vanishes initially on H 0 . By Proposition 3.3.4, in order to estimate kui+2 ∇i µkL∞ ∞ 2
u Lu L (Su,u )
,
it suffices to estimate the kui+2 · kL∞ 1 2

u Lu L (Su,u )
norm of the right hand side. We now estimate each
of the terms in the equation.
For the term containing highest derivative of η, we have
X
kui+2 ψ∇i+1 (η, η)kL1u L2 (Su,u )
i≤4
1
δ2 1
X
≤ kuψkL∞ L ∞ (S
u,u ) (δ 2 kui+2 ∇i+1 (η, η)kL∞ 2
u L (Su,u )
+ ku6 ∇5 (η, η)kL2u L2 (Su,u ) )
|u| u
i≤3
1 5 1
2
δ 1 3 3 1
2 δ a b 4 4
≤ a2 δ 2 a4 b4 ≤ ,
|u| |u|
253
where we have used the bounds in Proposition 3.4.5 and the bootstrap assumption (3.3.3).
For the other term with the highest order derivative, i.e., the term containing the highest
derivative of (trχ, χ̂), we get
X
kui+2 ψ∇i+1 (trχ, χ̂)kL1u L2 (Su,u )
i≤4
1
δ2 2
≤ ku ψkL∞ ∞
u L (Su,u )
|u|
X
×( kui+1 ∇i+1 (trχ, χ̂)kL2u L2 (Su,u ) + ku5 ∇5 (trχ, χ̂)kL2u L2 (Su,u ) )
i≤3
1 1
δ 2 1 1 1 1 1 δ 2 ab 2
≤ δa 2 b 4 · δ 2 a 2 b 4 ≤
|u| |u|
where we have used the bootstrap assumptions (3.3.1), (3.3.2) and (3.3.3).
We now estimate the lower order terms in the Ricci coefficients. Using Proposition 3.4.1 together
with the bootstrap assumption (3.3.2), we obtain
X X
kui+2 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
i1 ,i3 ≤i
δ X X
≤ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞
u L (Su,u )
kui3 ∇i3 ψkL∞ 2
u L (Su,u )
|u|
i1 +i2 ≤2 i3 ≤4
δ X X
+ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 ψkL∞ ∞
u L (Su,u )
|u|
i1 +i2 ≤5 i3 ≤2
i1 ≤4
1
δ 1 1 1 δ 2 ab 4
≤ δa 2 b 4 · a 2 ≤ .
|u| |u|
Finally, we control the term containing the Gauss curvature K. For this term, we need to make use
of the fact i ≥ 1 and apply the improved estimate for ∇i trχ from Proposition 3.4.3. Using Hölder’s
254
inequality and Sobolev embedding in Proposition 3.3.9, we have
X X
kui+2 ∇i1 ψ i2 ∇i3 trχ∇i4 KkL1u L2 (Su,u )
1≤i≤4 i1 +i2 +i3 =i
1
δ2 X X
≤ kui1 +i2 +1 ∇i1 ψ i2 kL∞ ∞
u L (Su,u )
kui3 +1 ∇i3 trχkL∞ ∞
u L (Su,u )
|u|2
i1 +i2 ≤2 i3 ≤2
X 1
× kui4 +2 ∇i4 (K − 2 )kL2u L2 (Su,u )
|u|
i4 ≤4
1
δ 2 X X
+ kui1 +i2 +1 ∇i1 ψ i2 kL∞ ∞
u L (Su,u )
kui3 ∇i3 trχkL∞ 2
u L (Su,u )
|u|2
i1 +i2 ≤2 i3 ≤4
X 1
×( kui4 +3 ∇i4 (K − )kL2u L∞ (Su,u ) + kukL2u L∞ (Su,u ) )
|u|2
i4 ≤2
1
δ 2 X X
+ kui1 +i2 ∇i1 ψ i2 kL∞ 2
u L (Su,u )
kui3 +1 ∇i3 trχkL∞ ∞
u L (Su,u )
|u|2
i1 +i2 ≤4 i3 ≤2
X 1
×( kui4 +3 ∇i4 (K − 2 )kL2u L∞ (Su,u ) + kukL2u L∞ (Su,u ) )
|u|
i4 ≤2
1
δ2 1 1 δa 3 3 1 1
≤ (|u| + δa 2 b 4 ) (δ 2 a 4 b 4 + δ 2 |u|)
|u|2 |u|
7 1 9 1 3 1 5
δ3 a 4 b 4 δ2 a δ4 a 4 b 2 δ3 a 2 b 4 δ2 a 4
≤ + + + ≤
|u|2 |u| |u|3 |u|2 |u|
using Propositions 3.4.1 and 3.4.3 and the bootstrap assumption (3.3.3).
1
Combining all the estimates above and using |u| ≥ δa 2 b, we have
5 1
X δ2 a 4 b 4
kui+2 ∇i µkL∞ 2
u L (Su,u )
≤ . (3.5.3)
|u|
1≤i≤4
By the div-curl system
1
div η = −µ + K − , curl η = σ̌
|u|2
255
and elliptic estimates from Proposition 3.3.14, we have
ku6 ∇5 ηkL2 (Su,u )

X 1
≤ (kui+2 ∇i µkL2 (Su,u ) + kui+2 ∇i (K − 2 , σ̌)kL2 (Su,u ) + kui+1 ∇i ηkL2 (Su,u ) )
|u|
i≤4
X 1 1 1
≤ku6 ∇4 µkL2 (Su,u ) + kui+2 ∇i (K − 2
, σ̌)kL2 (Su,u ) + δa 2 b 4 ,
|u|
i≤4
where we have used the bootstrap assumption (3.3.1). Hence, using (3.5.3) and taking L2 in u and
u respectively, we obtain
3 3
6 5 δ 2 a4 3 1 1 3 3 3 3
ku ∇ ηkL∞ 2 2
u Lu L (Su,u )
≤ 1 + δ 2 a 2 b 4 + δ 2 a 4 R ≤ δ 2 a 4 (1 + R)
b 2
and
3
δa 4 1 1 1 1 1 1
ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ ku−1 kL2u ( 1 + δa 2 b 4 ) + δ 2 a 2 R ≤ δ 2 a 2 (1 + R),
b 2
1
since |u| ≥ δa 2 b. This concludes the proof of the proposition.
We then turn to the estimates for ∇5 η. Unlike that for ∇5 η, we only achieve an estimate that
is integrated in the u direction but we lose the bound that is integrated in the u direction.
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
3 3
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 4 (1 + R).
Proof.
Let µ be defined by
1
µ = −div η + K − .
|u|2
Thus, we have the Hodge system for η:
1
div η = −µ + K − , curl η = −σ̌.
|u|2
256
Moreover, µ obeys the following equation:
1
∇3 µ + trχ µ = ψ∇(η, η) + ψ ψ ψ + ψ∇(χ̂, trχ) + trχdiv η + trχK + .
|u|3
Commuting with angular derivatives for 4 times, we get
∇3 ∇4 µ + 3trχ∇4 µ
1 5
=ψ∇5 (η, η, trχ, χ̂) + ∇ η
|u|
1 X 1 X
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 ∇i3 K
|u| i |u| i
1 +i2 +i3 =4 1 +i2 +i3 =4
X X
+ ∇i1 ψ i2 +2 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i4 K.
i1 +i2 +i3 =4 i1 +i2 +i3 +i4 =4
We apply Proposition 3.3.5 with λ0 = 3 which shows that for every fixed u, ku5 ∇4 µkL∞ 2
u L (Su,u )
can be estimated by the ku5 · kL1u L2 (Su,u ) norm of the right hand side. Since we will only need to
control ∇i µ after taking L2 norm in u, we will directly control ku5 · kL2u L1u L2 (Su,u ) of the right hand
side. We now estimate each of the terms in the equation. First, we start with terms with 5 angular
1
derivatives on the Ricci coefficients. We have two contributions from ∇5 η, one multiplied by |u|
and one multiplied by ψ. In the former case, we have, by Proposition 3.5.3,
ku4 ∇5 ηkL2u L1u L2 (Su,u )
≤ku−2 kL1u ku6 ∇5 ηkL∞ 2 2

u Lu L (Su,u )
3 3
δ 2 a4
≤ (1 + R).
|u|
Notice that in this bound, we do not gain an extra smaller factor compared to the desired estimate.
It is therefore important that we have already obtained the sharp estimates in Proposition 3.5.3
and do not have to resort to the bootstrap assumption (3.3.3).
The other contribution from ∇5 η can be estimated together with that from ∇5 η. For these
257
terms, we have
ku5 ψ∇5 (η, η)kL2u L1u L2 (Su,u )
≤ku−3 kL1u ku2 ψkL∞ ∞ ∞

u Lu L (Su,u )
ku6 ∇5 (η, η)kL∞ 2 2
u Lu L (Su,u )
5 5 1
δ 2 a4 b2
≤ ,
|u|2
where we have used the bootstrap assumptions (3.3.1) and (3.3.3) in the last inequality.
For the contributions from ∇5 (trχ, χ̂), we have
ku5 ψ∇5 (trχ, χ̂)kL2u L1u L2 (Su,u )

1
≤δ 2 ku−2 kL2u ku2 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku5 ∇5 (trχ, χ̂)kL∞ 2 2
u Lu L (Su,u )
1 1 1 1 5 1
1 δa 2 b 4 δa 2 b 4 δ 2 ab 2
≤δ 2 · 3 · 1 ≤ ,
|u| 2 |u| 2 |u|2
We then move to the lower order terms. First, we have
1 X
|u| i
1 +i2 +i3 =4
1
X
≤δ 2 ku−3 kL1u kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X
× kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
i3 ≤4
1
X
+ δ 2 ku−3 kL1u kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X
× kui3 +2 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
i3 ≤2
5 1
δ ab 2
2
≤ ,
|u|2
where we have used Proposition 3.4.1 and the bootstrap assumption (3.3.1).
258
Then, for the term with K, using Sobolev embedding in Proposition 3.3.9, we have
1 X
ku5 ∇i1 ψ i2 ∇i3 KkL2u L1u L2 (Su,u )
|u| i
1 +i2 +i3 =4
1
≤ku−2 kL1u ku6 ∇4 (K − )kL∞ 2 2
u Lu L (Su,u )
|u|2
1
X
+ δ 2 ku−2 kL2u kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X 1
× kui3 +1 ∇i3 (K − )kL∞ 2 2
u Lu L (Su,u )
|u|2
i3 ≤4
1
X
+δ 2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X
× (ku−2 kL2u kui3 +2 ∇i3 KkL∞ 2 ∞
u Lu L (Su,u )
+ ku−2 kL1u )
i3 ≤2
3 3 1 3 1 1 3 3 3 3
2
δ a
2 δ ab 2
4 δ 2 a2 b4 δ 2 a4 δ 2 a4
≤ R+ 3 + ≤ R+ ,
|u| |u| 2 |u| |u| |u|
where we have used Proposition 3.4.1.
The remaining two terms are actually better behaved than the two terms that we have just
1
estimates since ψ obeys better bounds that |u| . More precisely, we have
X
k u5 ∇i1 ψ i2 +2 ∇i3 ψkL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
1
X
≤δ 2 ku−4 kL1u kui1 +i2 +4 ∇i1 ψ i2 +2 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X
× kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
i3 ≤4
1
X
+ δ 2 ku−4 kL1u kui1 +i2 +3 ∇i1 ψ i2 +2 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X
× kui3 +2 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
i3 ≤4
1 1 1 1 7 3 3
δ 2 · δ 2 ab 2 · δa 2 b 4 δ 2 a2 b4
≤ 3
≤
|u| |u|3
using Proposition 3.4.1 and the bootstrap assumption (3.3.1).
259
Finally, the last term can be bounded by
X
ku5 ∇i1 ψ i2 +1 ∇i3 KkL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
1
X
≤δ ku−2 kL2u
2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X 1
× kui3 +1 ∇i3 (K − )kL∞ 2 2
u Lu L (Su,u )
|u|2
i3 ≤4
1
X
+δ 2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X 1
× (ku−2 kL2u kui3 +2 ∇i3 (K − )kL∞ 2 ∞
u Lu L (Su,u )
+ ku−2 kL1u )
|u|2
i3 ≤2
1 3 1 1 3 3
2
δ ab 2 δ a b4
2 2 δ 2 a4
≤ 3 + ≤ ,
|u| 2 |u| |u|
Therefore, combining the above estimates, we get
3 3
δ 2 a4
ku5 ∇4 µkL2u L∞ 2
u L (Su,u )
≤ (1 + R)
|u|
1
using |u| ≥ δa 2 b and b ≤ a. This implies, after multiplying by u, that
3 3
ku6 ∇4 µkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 4 (1 + R).
Therefore, using the div-curl system
1
div η = −µ + K − , curl η = −σ̌
|u|2
260
and applying elliptic estimates from Proposition 3.3.14, we have
ku6 ∇5 ηkL2 (Su,u )

X 1
≤ (kui+2 ∇i µkL2 (Su,u ) + kui+2 ∇i (K − 2 , σ̌)kL2 (Su,u ) + kui+1 ∇i ηkL2 (Su,u ) )
|u|
i≤4
X 1
≤ku6 ∇4 µkL2 (Su,u ) + (kui+1 ∇i ηkL2 (Su,u ) + kui+2 ∇i (K − 2 , σ̌)kL2 (Su,u ) )
|u|
i≤4
X 1 1 1
≤ku6 ∇4 µkL2 (Su,u ) + kui+2 ∇i (K − 2
, σ̌)kL2 (Su,u ) + δa 2 b 4 ,
|u|
i≤4
where we have used the bootstrap assumption (3.3.1) in the last step. This implies, after taking
L2u norm, that

3 3
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ δ 2 a 4 (1 + R).
We now prove the highest order bounds for ω:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1
δa 2
ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
≤ 1 (1 + R).
|u| 2
Proof. Recall that in the proof of Proposition 3.5.2 we have defined an auxiliary function ω †
in order to apply elliptic estimates to obtain the highest order estimate for ω. Here, we similarly
define an auxiliary function ω † by

1
∇4 ω † = σ̌
2
with zero initial data on H 0 . We then define κ by
1
κ := −∇ω +∗ ∇ω † − β.
2
261
Before we proceed, observe that the proof of Proposition 3.4.6 implies that
X 1 1
kui+1 ∇i ω † kL∞ ∞ 2
u Lu L (Su,u )
≤ δa 2 (1 + 1 1 ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
+ R). (3.5.4)
i≤4
δ a2
2
1
(In fact, the stronger bound with δa 2 on the right hand side holds. We will not need this refinement.)
In view of this bound, we will allow ψ to also denote ω † in the remainder of the proof of this
proposition. With this new convention, it is easy to check that κ obeys the equation
X 1 1 1 1
∇4 κ = ∇i1 ψ i2 +1 ∇i3 (ψ, ψ) + ψ(K − , σ̌) + 2 ψ + ∇trχ + ψψ.
i1 +i2 +i3 =1
|u|2 |u| |u| |u|
Commuting with angular derivatives for i times, we get
∇ 4 ∇4 κ
X X 1
= ∇i1 ψ i2 +1 ∇i3 (ψ, ψ) + ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)
i1 +i2 +i3 =5 i1 +i2 +i3 =4
|u|2
X 1 i1 i2 +1 1 5 X 1 i1 i2 +1 i3
+ 2
∇ ψ + ∇ trχ + ∇ ψ ∇ ψ.
i1 +i2 =4
|u| |u| i
|u|
1 +i2 +i3 =4
Using Proposition 3.3.4 and the fact that ∇5 κ = 0 on H 0 , we can estimate ku5 ∇4 κkL∞ 2 2
u Lu L (S)
by controlling the ku5 · kL2u L1u L2 (S) norm of the right hand side. We now estimate each of the terms
in the equation.
We first estimate the term with highest order derivatives of the Ricci coefficients, i.e., the terms
1 5
(ψ, ψ)∇5 ψ, ψ∇5 ψ, ∇ trχ.
|u|
For (ψ, ψ)∇5 ψ, we will separately consider the terms (ψ, ψ)∇5 (trχ, χ̂, ω) and (ψ, ψ)∇5 (η, η). For
262
the contributions from (ψ, ψ)∇5 (trχ, χ̂, ω), we have
ku5 (ψ, ψ)∇5 (trχ, χ̂, ω)kL2u L1u L2 (Su,u )
≤ku5 ∇5 (trχ, χ̂, ω)kL∞ 2 2

u Lu L (Su,u )
k(ψ, ψ)kL1u L∞ ∞
u L (Su,u )
1 1 1 1 1
δa 2 b 4 δa 2 b 4 δ 2 ab 2
≤ 1 ≤ 3 ,
|u| 2 |u| |u| 2
where we have used bootstrap assumptions (3.3.1), (3.3.2) and (3.3.3).
For the contributions from (ψ, ψ)∇5 (η, η), we have
ku5 (ψ, ψ)∇5 (η, η)kL2u L1u L2 (Su,u )
≤ku6 ∇5 (η, η)kL∞ 2 2

u Lu L (Su,u )
ku2 ψkL2u L∞ ∞
u L (Su,u )
ku−3 kL2u
3 1 1 5 1
3 3 1 δ 2 a2 b4 δ3 a 4 b 2
≤δ 2 a 4 b 4 5 ≤ 5 ,
|u| 2 |u| 2
We then control the term u5 ψ∇5 ψ, for which we have
ku5 ψ∇5 ψkL2u L1u L2 (Su,u )
≤ku5 ∇5 ψkL∞ 2 2
u Lu L (Su,u )
kψkL2u L2u L∞ (Su,u )
3 1 1 1
1 1 1 δ 2 a2 b4 δ 2 ab 2
≤δ 2 a 2 b 4 3 ≤ 3 .
|u| 2 |u| 2
Here, we have used the bootstrap assumptions (3.3.1) and (3.3.3).
To estimate the remaining highest order term u4 ψ∇5 trχ, we use the bound from Proposition
3.5.1 which is stronger than that in the bootstrap assumptions. More precisely, we have
ku4 ∇5 trχkL2u L1u L2 (Su,u )

1
≤ku5 ∇5 trχkL∞ 2 2
u Lu L (Su,u )
k kL2 L2 L∞ (Su,u )
|u| u u
3 1 1 1
δ 2 ab 4 δ 2 δ 2 ab 4
≤ 1 ≤ 3
|u| |u| 2 |u| 2
263
where we have used Proposition 3.5.1.
After estimating the highest order Ricci coefficient term, we now control the curvature terms,
1
i.e., the terms with (K − |u|2 , σ̌). For these terms, we have the estimate
X 1
ku5 ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)kL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
|u|2
X 1 X
≤ kui3 +1 ∇i3 (K − , σ̌)kL ∞ L2 L2 (S ) kui1 +i2 ∇i1 ψ i2 +1 kL∞ 1 ∞
u Lu L (Su,u )
|u|2 u u u,u
i3 ≤4 i1 +i2 ≤2
X 1
+ kui3 +3 ∇i3 (K − 2 , σ̌)kL∞ 2 ∞
u Lu L (Su,u )
|u|
i3 ≤1
X
× kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ 2 2
u Lu L (Su,u )
ku−3 kL2u
i1 +i2 ≤4
1 3 1 1
1 δ2 a 2 1
1 1 3 3 1 δ 2 a2 b4
≤δ a b
2 2 b4 + δ 2 a4 b4
4
5
|u|2 |u| 2
5 1 5 1 5 1
δ 2 ab 2 δ3 a 4 b 2 δ 2 ab 2
≤ 2
+ 5 ≤ ,
|u| |u| 2 |u|2
where we have used the Sobolev embedding in Proposition 3.3.9 to control the curvature terms in
L∞ (Su,u ) by the curvature norm R. In the above, we have also used Proposition 3.4.1 to bound
the product of derivatives of ψ and used bootstrap assumption (3.3.3) to estimate R.
We then move to the lower order terms, i.e., the terms containing only Ricci coefficients and
such that there are at most 4 angular covariant derivatives on ψ and ψ. These are the terms
X X 1 i1 i2 +1
∇i1 ψ i2 +1 ∇i3 (ψ, ψ), ∇ ψ
i1 +i2 +i3 =5, i1 +i2 =4
|u|2
i1 ,i3 ≤4
and
X 1 i1 i2 +1 i3
∇ ψ ∇ ψ.
i1 +i2 +i3 =4
|u|
264
We estimate these terms according to the order above. First, we have
X
k u5 ∇i1 ψ i2 +1 ∇i3 (ψ, ψ)kL2u L1u L2 (Su,u )
i1 +i2 +i3 =5,
i1 ,i3 ≤4
X X
≤ kui1 +i2 ∇i1 ψ i2 +1 kL2u L∞ ∞
u L (Su,u )
kui3 ∇i3 (ψ, ψ)kL∞ 1 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL2u L∞ 2
u L (Su,u )
kui3 +1 ∇i3 (ψ, ψ)kL∞ 1 ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
1
2
δ ab 4
≤ 3
|u| 2
using Proposition 3.4.1 and 3.4.5, the bootstrap assumptions (3.3.1) and (3.3.3) and the condition
1
|u| ≥ δa 2 b. The second term can be controlled as follows:
X
k u3 ∇i1 ψ i2 +1 kL2u L1u L2 (Su,u )
i1 +i2 =4
X X
≤ kui1 +i2 −1 ∇i1 ψ i2 kL2u L∞ ∞
u L (Su,u )
kui3 ∇i3 ψkL∞ 1 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
1 1
2
δ a b 2 4
≤ 3
|u| 2
using Proposition 3.4.1 and the bootstrap assumption (3.3.1). Finally, we bound the remaining
term by
X
k u4 ∇i1 ψ i2 +1 ∇i3 ψkL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
X X
≤ kui1 +i2 ∇i1 ψ i2 +1 kL2u L∞ ∞
u L (Su,u )
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL2u L∞ 2
u L (Su,u )
kui3 +1 ∇i3 ψkL∞ 1 ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
1
2
δ ab 4
≤ 3
|u| 2
using Proposition 3.4.1 and 3.4.5.
265
Therefore, combining the above estimates, we get
1
δa 2
ku5 ∇4 κkL∞ 2 2
u Lu L (Su,u )
≤ 1 . (3.5.5)
|u| 2
By the following div-curl system:
1
div ∇ω = −div κ − div β,
2
curl ∇ω = 0,
1
2
div ∇ω † = 0.
and Proposition 3.3.14, we have
ku5 ∇5 (ω, ω † )kL2 (Su,u )

X
i+1 i i+1 i i i †
≤ ku ∇ κkL2 (Su,u ) + ku ∇ βkL2 (Su,u ) + ku ∇ (ω, ω )kL2 (Su,u )
i≤4
X
5 4 i+1 i i i †
≤ku ∇ κkL2 (Su,u ) + ku ∇ βkL2 (Su,u ) + ku ∇ (ω, ω )kL2 (Su,u )
i≤4
This implies, after taking L2 norm in u that
ku5 ∇5 (ω, ω † )kL2u L2 (Su,u )

1 3 3 1 1
δa 2
δ 2 a4 b4 δa 2
≤ 1 + + 1 (1 + R)
|u| 2 |u| |u| 2
1
δa 2
≤ 1 (1 + R).
|u| 2
after substituting in the bound (3.5.5) and using bootstrap assumption (3.3.3) together with (3.5.4),
and also Propositions 3.4.6 and 3.5.2.
266
Finally, we prove the highest order estimates for the remaining Ricci coefficients, trχ and χ̂:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1
δa 2
ku5 ∇5 (trχ, χ̂)kL∞ 2 2
u Lu L (Su,u )
≤ 1 (1 + R).
|u| 2
Proof.
Consider the following equation for trχ:
1
∇3 trχ + (trχ)2 = −2ωtrχ − |χ̂|2 .
2
Commuting the equation with angular derivatives for 5 times, we get
∇3 ∇5 trχ + 3trχ∇5 trχ

1 5 X
=ψ∇5 (trχ, χ̂, ω) + ∇ ω+ ∇i1 ψ i2 +1 ∇i3 ψ
|u| i 1 +i2 +i3 =5
i1 ,i3 ≤4
X 1 i1 i2 +1 i3 X 1 i1 i2 +1
+ ∇ ψ ∇ ψ+ ∇ ψ .
i1 +i2 +i3 =4
|u| i +i =4
|u|2
1 2
We apply Proposition 3.3.5 with λ0 = 3. This allows us to estimate the quantity ku5 ∇5 trχkL∞ ∞ 2
u Lu L (Su,u )
by bounding the ku5 · kL∞ 1 2

u Lu L (Su,u )
norm of the right hand side of the equation above. We now
estimate each of the terms in the equation. The first term can be controlled by
ku5 ψ∇5 (trχ, χ̂, ω)kL∞ 1 2

u Lu L (Su,u )
≤ku−2 kL2u ku2 ψkL∞ ∞ ∞

u Lu L (Su,u )
ku5 ∇5 (trχ, χ̂, ω)kL∞ 2 2
u Lu L (Su,u )
1
δ 2 ab 2
≤ ,
|u|2
where we have used the bootstrap assumption (3.3.1) and (3.3.3). For the second term, we use
267
Proposition 3.5.5 to get
1 1
δa 2 δa 2
ku4 ∇5 ωkL∞ 1 2
u Lu L (Su,u )
≤ku−1 kL2u 1 (1 + R) ≤ (1 + R).
|u| 2 |u|
We now turn to the lower order terms in the Ricci coefficients. For the third term, we have
X
ku5 ∇i1 ψ i2 +1 ∇i3 ψkL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 =5
i1 ,i3 ≤4
X X
≤ku−3 kL1u kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
kui3 +1 ∇i ψkL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
1
2
δ ab 2
≤ ,
|u|2
For the fourth term, we have the estimate
X 1 i1 i2 +1 i3
ku5 ∇ ψ ∇ ψkL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 =4
|u|
X
≤ku−3 kL1u kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X
× kui3 +1 ∇i ψkL∞ ∞ 2
u Lu L (Su,u )
i3 ≤3
1
2
δ ab 2
≤ ,
|u|2
where we have used Proposition 3.4.1 the bootstrap assumption (3.3.1).
268
Finally, the last term can be controlled by
X 1 i1 i2 +1
ku5 ∇ ψ kL∞ 1 2
u Lu L (Su,u )
i1 +i2 =4
|u|2
X
≤ku−2 kL1u kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
1
δa 2 1
≤ (1 + 1 1 ku5 ∇5 ωkL∞ 2 2
u Lu L (Su,u )
+ R)
|u| δ a2
2
1
δa 2
≤ (1 + R),
|u|
using Propositions 3.4.1, 3.4.6 and 3.5.2. Notice that we do not have extra smallness in the above
estimate and it is important that we applied the sharp estimates in Propositions 3.4.6 and 3.5.2
instead of using the bootstrap assumptions.
Collecting all the above terms, we have
1
5 5 δa 2
ku ∇ trχkL∞ ∞ 2
u Lu L (Su,u )
≤ (1 + R),
|u|
which implies
1
1 δa 2
ku5 ∇5 trχkL∞ 2 2
u Lu L (Su,u )
≤ δa 2 (1 + R)ku−1 kL2u ≤ 1 (1 + R). (3.5.6)
|u| 2
In order to obtain the estimates for the fifth angular derivatives of χ̂, we use the Codazzi equation
1 1 1
div χ̂ = β + ∇trχ − (η − η) · (χ̂ − trχ)
2 2 2
269
and Proposition 3.3.15 to derive elliptic estimates for ∇5 χ̂:
ku5 ∇5 χ̂kL2 (Su,u )

X
≤ (kui+1 ∇i+1 trχkL2 (Su,u ) + kui+1 ∇i βkL2 (Su,u ) )
i≤4
X X X
+ kui+1 ∇i1 ψ∇i2 ψkL2 (Su,u ) + kui ∇i χ̂kL2 (Su,u )
i≤4 i1 +i2 =i i≤4
X
≤ (kui+1 ∇i+1 trχkL2 (Su,u ) + kui+1 ∇i βkL2 (Su,u ) + kui ∇i χ̂kL2 (Su,u ) )
i≤4
1 X i1 +2 i1 X
+ 2
ku ∇ ψkL∞ (Su,u ) kui2 +1 ∇i2 ψkL2 (Su,u )
|u|
i1 ≤2 i2 ≤4
X
≤ (kui+1 ∇i+1 trχkL2 (Su,u ) + kui+1 ∇i βkL2 (Su,u ) )
i≤4
1 1
δa 2 δ 2 ab 2
+ (1 + R) + ,
|u| |u|2
where we have used the bootstrap assumption (3.3.1) together with Propositions 3.4.1, 3.4.6 and
3.5.2 in the last step. Taking L2 norm in u and using (3.5.6), we obtain
ku5 ∇5 χ̂kL∞ 2 2
u Lu L (Su,u )
1 1
X
i+1 i+1 i+1 i δa 2 δ 2 ab 2
≤ (ku ∇ trχkL∞ 2 2
u Lu L (Su,u )
+ ku ∇ βkL∞ 2 2
u Lu L (Su,u )
)+ 1 + 3
i≤4
|u| 2 |u| 2
1
δa 2
≤ 1 (1 + R)
|u| 2
using (3.5.6) and the definition for the R norm.
We conclude this section by the following proposition, summarizing all the estimates for the
Õ5,2 norm derived in this section:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
Õ5,2 ≤ 1 + R.
270
3.6 Estimates for Curvature
In this section, we derive and prove the energy estimates for the renormalized curvature components
and their first four angular derivatives. We will show that R ≤ 1. Together with the estimates in
the previous sections, we will therefore improve all of the bootstrap assumptions (3.3.1), (3.3.2),
(3.3.3) and (3.3.4) and obtain Theorem 3.2.1.
To derive the energy estimates, we will need the following integration by parts formula, which
can be proved by direct computations:
Proposition 3.6.1. Suppose φ1 and φ2 are r tensorfields, then
Z Z
φ1 ∇ 4 φ2 + φ2 ∇ 4 φ1
Du,u Du,u
Z Z Z
= φ1 φ2 − φ1 φ2 + (2ω − trχ)φ1 φ2 .
H u (1,u) H 0 (1,u) Du,u
Here, we have used the convention that H u (1, u) = {(u0 , u, θ1 , θ2 ) : u ≤ u0 ≤ 1}.
(1) (2)
Proposition 3.6.2. Suppose we have an r tensorfield φ and an r − 1 tensorfield φ.
Z Z
(1) A1 A2 ...Ar (2)
φ ∇A r φA1 ...Ar−1 + ∇Ar (1) φA1 A2 ...Ar (2) φA1 ...Ar−1
Du,u Du,u
Z
=− (η + η)(1) φ(2) φ.
Du,u
To derive the energy estimates from the ∇3 equations, we also need the following analogue of
Proposition 3.6.1 in the ∇3 direction. Moreover, we need to obtain an analogue which incorporates
the weights in u. More precisely, we have
Proposition 3.6.3. Suppose φ is an r tensorfield and let λ1 = 2(λ0 − 21 ). Then
Z
2 |u|2λ1 φ(∇3 + λ0 trχ)φ
Du,u
Z Z Z
= |u|2λ1 |φ|2 − |u|2λ1 |φ|2 + |u|2λ1 f |φ|2 ,
Hu (0,u) H0 (0,u) Du,u
271
where f obeys the estimate
1 1
δa 2 b 4
|f | ≤ .
|u|2
Similar to Proposition 3.6.1, we have used the convention that Hu (0, u) = {(u, u0 , θ1 , θ2 ) : 0 ≤ u0 ≤
u}.
Proof. Slightly modifying (3.3.10), we have
Z
d
− ( |u|2λ1 Ω|φ|2 )
du Su,u
Z
= Ω2 2|u|2λ1 < φ, ∇3 φ + λ0 trχφ >
Su,u
Z
2 2λ1 2λ1 (e3 u) 2
+ Ω |u| ( + (1 − 2λ0 )trχ − 2ω)|φ| .
Su,u |u|
The proposition follows after integrating with respect to du du, applying the fundamental theorem
of calculus in u and noting that
1 1
2λ1 (e3 u) δa 2 b 4
+ (1 − 2λ0 )trχ − 2ω ≤ ,
|u| |u|2
using Propositions 3.3.1 and the bootstrap assumption 3.3.1.

1
We now derive energy estimates for ∇i (K − |u|2 , σ̌) in L2u L2 (Su,u ) and for ∇i β in L2u L2 (Su,u ).
1
Notice that by the definition of our norms, we also need to obtain bounds for K − |u|2 in the
case i = 0. We will leave this case to later (see Proposition 3.6.6). We now prove estimates for
1
∇i (K − |u|2 , σ̌) in L2u L2 (Su,u ) and for ∇i β in L2u L2 (Su,u ) in the case i ≥ 1.
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 1 3 3
(kui+2 ∇i (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
+ kui+2 ∇i βkL∞ 2 2
u Lu L (Su,u )
) ≤δ 2 a 4 .
|u|2
1≤i≤4
272
1
Proof. We begin with the following schematic Bianchi equations for σ̌, K − |u|2 and β:
3 X
∇3 σ̌ + div ∗ β + trχσ̌ = ψ i1 +1 ∇i2 ψ,
2 i +i =1 1 2
and
1 3 1
∇3 (K − ) + div β + trχ(K − 2 )
|u|2 2 |u|
X 1 1 2 1 (3.6.1)
= ψ i1 +1 ∇i2 ψ + µ + 2 (trχ + ) + 3 (Ω−1 − 1),
i +i =1
|u| |u| u |u|
1 2
and
∇4 β − ∇K −∗ ∇σ̌
X 2 X 1
=ψ(K, σ̌) + ψ i1 ∇i2 (trχ + , χ̂, ω)∇i3 ψ + ψ i1 ∇i2 trχ.
i1 +i2 +i3 =1
|u| i +i =1
|u|
1 2
Notice that in the above, we have used a special cancellation to obtain (3.6.1). More precisely, we
start with
1 X
∇3 K + div β + trχK + trχdiv η = ψ i1 +1 ∇i2 ψ.
2 i +i =1 1 2
Now, note4 that

1 2Ω−1
∇3 (K − ) = ∇ 3 K − .
|u|2 |u|3
On the other hand, we have
1 1 1 1
trχdiv η = trχ(K − 2 ) − trχµ.
2 2 |u| 2
4 In the three displayed equations below, since we need to capture the cancellation, we will not use the schematic
notation but track the exact coefficients in each of the terms.
273
Therefore,
1
∇3 K + trχK + trχdiv η
2
1 3 1 1 1 2Ω−1
=∇3 (K − 2 ) + trχ(K − )− trχµ + trχ 2 +
|u| 2 |u|2 2 |u| |u|3
1 3 1 1 2 1 2(1 − Ω−1 )
=∇3 (K − 2 ) + trχ(K − )− trχµ + (trχ + ) 2− ,
|u| 2 |u|2 2 |u| |u| |u|3
from which (3.6.1) follows. Commuting the equations with ∇ for i times, we have
3+i
∇3 ∇i σ̌ + div ∗ ∇i β + trχ∇i σ̌ = F1,i , (3.6.2)
2
where F1,i is given by
X 1 X
F1,i = ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i3 σ̌
i1 +i2 +i3 =i+1
|u| i
1 +i2 +i3 =i−1
and the equation
1 3+i 1
∇3 ∇i (K − ) − div ∇i β + trχ∇i (K − 2 ) = F2,i , (3.6.3)
|u|2 2 |u|
where F2,i is defined as
X 1 i 1 X
F2,i = ∇i1 ψ i2 +1 ∇i3 ψ + ∇ µ+ ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 +i3 =i+1
|u| |u| i
1 +i2 +i3 =i
1 X 2
+ ∇i1 ψ i2 ∇i3 (trχ + )
|u|2 i1 +i2 +i3 =i
|u|
1 X
+ 3 ∇i1 ψ i2 ∇i3 (1 − Ω−1 )
|u| i1 +i2 +i3 =i
X 1
+ ∇i1 ψ i2 +1 ∇i3 (K − )
i1 +i2 +i3 =i
|u|2
X 1 i1 i2 +1 i3 1
+ ∇ ψ ∇ (K − 2 )
i1 +i2 +i3 =i−1
|u| |u|
274
and also the equation
1
∇4 ∇i β − div ∇i (K − ) −∗ ∇∇i σ̌ = F3,i , (3.6.4)
|u|2
with F3,i given by
X 2
F3,i = ∇i1 ψ i2 ∇i3 (trχ + , χ̂, ω)∇i4 ψ
i1 +i2 +i3 +i4 =i+1
|u|
X 1 X
+ ∇i1 ψ i2 +1 ∇i3 (K, σ̌) + ∇i1 ψ i2 ∇i3 trχ.
i1 +i2 +i3 =i
|u| i
1 +i2 +i3 =i+1
Using the equations (3.6.2), (3.6.3) and (3.6.4), we can derive the energy estimates. More
precisely, using Proposition 3.6.1 and the equation (3.6.4), we have
Z 2
1 i+2 i
u ∇β
2 H u (1,u)
Z 2 Z
1 i+2 i
= u ∇β + < ui+2 ∇i β, ui+2 ∇4 ∇i β >γ
2 H 0 (1,u) Du,u
Z 2
1 i+2 i (3.6.5)
− (ω − trχ) u ∇ β
Du,u 2
Z 2 Z
1 i+2 i
= u ∇β + < ui+2 ∇i β, ui+2 F3,i >γ
2 H 0 (1,u) Du,u
Z
1
+ < ui+2 ∇i β, ui+2 (∇∇i (K − 2 ) + ∗ ∇∇i σ̌) >γ .
Du,u |u|
2
1 i+2 i
R
Here, we have abused notation to drop the term − Du,u
(ω − 2 trχ) u ∇ β since it has the
same schematic form as one of the terms represented by
Z
< ui+2 ∇i β, ui+2 F3,i >γ .
Du,u
275
Now, applying Proposition 3.6.3 and equation (3.6.2), we obtain
Z 2
1 i+2 i
u ∇ σ̌
2 Hu (0,u)
Z 2 Z
1 i+2 i 3+i
= u ∇ σ̌ + < ui+2 ∇i σ̌, ui+2 (∇3 + trχ)∇i σ̌ >γ
2 H1 (0,u) Du,u 2
Z 2
1
− f ui+2 ∇i σ̌ (3.6.6)
2 Du,u
Z 2 Z
1 i+2 i
= u ∇ σ̌ + < ui+2 ∇i σ̌, ui+2 F1,i >γ
2 H1 (0,u) Du,u
Z Z 2
i+2 i i+2 ∗ i 1 i+2 i
− <u ∇ σ̌, u (div ∇ β) >γ − f u ∇ σ̌ .
Du,u 2 Du,u
Similarly, using Proposition 3.6.3 and the equation (3.6.3), we get
Z 2
1 i+2 i 1
u ∇ (K − 2 )
2 Hu (0,u) |u|
Z 2 Z
1 i+2 i 1
= u ∇ (K − 1) + < ui+2 ∇i (K − 2 ), ui+2 F2,i >γ
2 H1 (0,u) Du,u |u|
Z (3.6.7)
1
+ < ui+2 ∇i (K − 2 ), ui+2 (div ∇i β) >γ
Du,u |u|
Z 2
1 i+2 i 1
− f u ∇ (K − 2 ) .
2 Du,u |u|
Now, we can integrate by parts on the spheres Su,u using Proposition 3.6.2 to show that the sum of
the terms with highest order angular derivatives in (3.6.5), (3.6.6) and (3.6.7) cancel up to a lower
276
order error term:
Z
1
< ui+2 ∇i β, ui+2 (∇∇i (K − ) + ∗ ∇∇i σ̌) >γ
Du,u |u|2
Z
− < ui+2 ∇i σ̌, ui+2 (div ∗ ∇i β) >γ
Du,u
Z
1
+ < ui+2 ∇i (K − ), ui+2 (div ∇i β) >γ (3.6.8)
Du,u |u|2
1
≤ku2i+4 ∇i (K − , σ̌)ψ∇i βkL1u L1u L1 (Su,u )
|u|2
3 1 1
δ 2 a2 b4 1
≤ 3 kui+2 ∇i (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
kui+2 ∇i βkL∞ 2 2
u Lu L (Su,u )
,
|u| 2 |u|2
where in the last line we have used the bootstrap assumption (3.3.1). Therefore, adding the identities
(3.6.5), (3.6.6), (3.6.7), using (3.6.8) and the bound for f in Proposition 3.6.3, we obtain
1
kui+2 ∇i ((K − ), σ̌)k2L2u L2 (Su,u ) + kui+2 ∇i βk2L2u L2 (Su,u )
|u|2
1
≤k∇i (K − , σ̌)k2L2u L2 (S0,u ) + ku2i+4 ∇i σ̌F1,i k2L1u L1u L1 (Su,u )
|u|2
1
+ ku2i+4 ∇i (K − 2 )F2,i kL1u L1u L1 (Su,u ) + ku2i+4 ∇i βF3,i kL1u L1u L1 (Su,u ) (3.6.9)
|u|
1 1
δa b 2 4 1 1
+ ku2i+4 ∇i (K − 2 , σ̌)∇i (K − 2 , σ̌)kL1u L1u L1 (Su,u )
|u|2 |u| |u|
3 1 1
δ 2 a2 b4 1
+ 3 kui+2 ∇i (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
kui+2 ∇i βkL∞ 2 2
u Lu L (Su,u )
|u| 2 |u|2
For the second, third and fourth terms, we can apply Cauchy-Schwarz in either the H or the H
1
hypersurface so that the terms kui+2 ∇i ((K − |u|2 ), σ̌)kLu L (Su,u )
2 2 and kui+2 ∇i βkL2u L2 (Su,u ) can be
absorbed to the left and we only need to bound the weighted norms of F1,i , F2,i and F3,i . For the
fifth term, noticing that

1 1
δa 2 b 4 1
k kL1u ≤ 7 ,
|u|2 b4
we see that it can be controlled by Gronwall’s inequality. For the final term, since
3 1 1
δ 2 a2 b4 1
k 3 kL∞
u
≤ 5 ,
|u| 2 b4
277
the term can be absorbed to the left hand side after using Schwarz’s inequality. Therefore, (3.6.9)
implies that
1
kui+2 ∇i ((K − ), σ̌)kL2u L2 (Su,u ) + kui+2 ∇i βkL2u L2 (Su,u )
|u|2
1
≤k∇i (K − , σ̌)kL2u L2 (S0,u ) + kui+2 F1,i kL1u L2u L2 (Su,u )
|u|2
X
+ kui+2 F2,i kL1u L2u L2 (Su,u ) + kui+2 F3,i kL1u L2u L2 (Su,u ) .
i≤4
Summing over 1 ≤ i ≤ 4 and using the fact that
1 1 1
k∇i (K − , σ̌)kL2u L2 (S0,u ) ≤ δ 2 a 2 ,
|u|2
we obtain
X 1
(kui+2 ∇i ((K − ), σ̌)kL2u L2 (Su,u ) + kui+2 ∇i βkL2u L2 (Su,u ) )
|u|2
1≤i≤4
1 1
X
≤δ 2 a 2 + (kui+2 F1,i kL1u L2u L2 (Su,u ) + kui+2 F2,i kL1u L2u L2 (Su,u ) ) (3.6.10)
1≤i≤4
X
+ kui+2 F3,i kL1u L2u L2 (Su,u ) .
1≤i≤4
We will estimate the right hand side of (3.6.10) term by term.5 We first estimate the term F1,i .
For the first term in F1,i , we can assume without loss of generality that i1 ≤ i3 . We bound separately
the contributions where there are at most 4 derivatives falling on any of the Ricci coefficients, where
5 We now remark on a convention that we will use in this proof. While it is important that we only have 1 ≤ i ≤ 4
in the sum in some of the error terms to take advantage of the improved estimates, we will simply write i ≤ 4 in the
terms where this restriction is not necessary.
278
5 derivatives fall on (trχ, χ̂, ω) and where 5 derivatives fall on (η, η). More precisely, we have
X X
kui+2 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
X
≤ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤5
i1 ≤2
1
X
× (δ 2 kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
(3.6.11)
i3 ≤4
1
+ δ 2 ku5 ∇5 (trχ, χ̂, ω)kL∞ 2 2
u Lu L (Su,u )
ku−1 kL2u
+ ku6 ∇5 (η, η)kL∞ 2 2

u Lu L (Su,u )
ku−2 kL1u )
3 1 1 3 3 5 5 1
1 1 δ 2 a2 b4 δ 2 a4 δ 2 a4 b4
≤δa 2 b 4 ( + )≤ ,
|u| |u| |u|
For the remaining contributions in F1,i , we will prove the slightly more general bound where we
1
allow (K − |u|2 , σ̌) in place of σ̌. Using Sobolev embedding in Proposition 3.3.9, we have
X X 1
kui+2 ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
|u|2
i≤4
X
≤ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
i1 ≤2 (3.6.12)
X 1
×( ku i3 +2 i3
∇ (K − 2 , σ̌)kL∞ 2 2
u Lu L (Su,u )
ku−2 kL1u
|u|
i3 ≤4
5 5 1
1 1 3 3 1 1 δ 2 a4 b2
≤δa 2 b 4 δ 2 a 4 b 4 ≤ .
|u| |u|
279
The final term in F1,i can be controlled in a similar fashion as (3.6.12):
X X 1
kui+1 ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i−1
|u|2
i≤4
X
≤ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤3
i1 ≤2 (3.6.13)
X 1
×( ku i3 +2 i3
∇ (K − 2 , σ̌)kL∞ 2 2
u Lu L (Su,u )
ku−2 kL1u
|u|
i3 ≤3
5 5 1
1 1 3 3 1 1 δ 2 a4 b2
≤δa 2 b 4 δ 2 a 4 b 4 ≤ .
|u| |u|
We now move to the estimates for F2,i . Notice that the first, sixth and seventh terms are already
estimated above in (3.6.11), (3.6.12) and (3.6.13). For the second term, we need to use the improved
estimates for ∇i µ derived in Proposition 3.5.3. In particular, we need to use the fact that i ≥ 1.
X
kui+1 ∇i µkL1u L2u L2 (Su,u )
1≤i≤4
1
X
≤δ 2 kui+3 ∇i µkL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
1≤i≤4
5 5 1
δ a4 b4
2
≤ .
|u|
For the third term in F2,i , we have
X X
kui+1 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
1
X X
≤δ 2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
i1 +i2 ≤4 i3 ≤4
i1 ≤2
5 1
δ 2 ab 2
≤
|u|
For the fourth term in F2,i , we need to use i ≥ 1 and apply the improvement in the bounds for
280
2
∇i (trχ + |u| ) from Proposition 3.4.7. More precisely, we have
X X 2
kui ∇i1 ψ i2 ∇i3 (trχ + )kL1u L2u L2 (Su,u )
i1 +i2 =i
|u|
1≤i≤4
1
X 3 2 3
≤δ 2 kui+ 2 ∇i (trχ + )kL∞ ∞ 2
u Lu L (Su,u )
ku− 2 kL1u
|u|
1≤i≤4
1
X
+ δ2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X
× kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
i3 ≤4
3 5 1
2
δ a 4 δ 2 ab 2
≤ 1 + ,
|u| 2 |u|
where in addition to using Proposition 3.4.1, we have used Proposition 3.4.7 for the first term and
the bootstrap assumptions (3.3.1) for the second term.
For the fifth term in F2,i , we also need to use i ≥ 1 and apply the improvement in the bounds
for ∇i (1 − Ω−1 ) from Proposition 3.4.8. More precisely, we have
X X
kui−1 ∇i1 ψ i2 ∇i3 (1 − Ω−1 )kL1u L2u L2 (Su,u )
1≤i≤4 i1 +i2 =i
1 1 3
X
≤δ 2 kui+ 2 ∇i log ΩkL∞ ∞ 2
u Lu L (Su,u )
ku− 2 kL1u
1≤i≤4
1
X
+ δ2 kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2
X
× kui3 +1 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
i3 ≤3
1
X
+δ 2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
× ku(1 − Ω−1 )kL∞ ∞ ∞

u Lu L (Su,u )
ku−2 kL1u
3 5 1
δ2 a 4 δ 2 ab 2
≤ 1 + .
|u| 2 |u|
Here, we have used Propositions 3.3.1, 3.4.1 and 3.4.8.
281
We now estimate the contributions from F3,i . For the first term, we have
X X 2
kui+2 ∇i1 ψ i2 ∇i3 (trχ + , χ̂, ω)∇i4 ψkL1u L2u L2 (Su,u )
i1 +i2 +i3 +i4 =i+1
|u|
i≤4
X X
≤δ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
u Lu L (Su,u )
ku−1 kL2u
i1 +i2 ≤5 i3 ≤4
i1 ≤2
1
+ δ 2 ku2 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku5 ∇5 ψkL∞ 2 2
u Lu L (Su,u )
ku−1 kL2u
X
+δ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤5
i1 ≤4
X
× kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−1 kL2u
i3 ≤2
2
+ δku5 ∇5 (trχ + , χ̂, ω)kL∞ 2 2
u Lu L (Su,u )
kuψkL∞ ∞ ∞
u Lu L (Su,u )
|u|
1
δ 2 ab 4
≤ 1 ,
|u| 2
where we have used Proposition 3.4.1, the bootstrap assumption (3.3.3), as well as the bound for
∇i ψ derived in Proposition 3.4.5.
For the second term, we use Sobolev embedding (Theorem 3.3.9) to get
X X
kui+2 ∇i1 ψ i2 +1 ∇i3 (K, σ̌)kL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
1
X
≤δ 2 kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4
X 1
× kui3 +2 ∇i3 (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
ku−2 kL2u
|u|2
i3 ≤4
X
+δ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
ku−1 kL2u
i1 +i2 ≤4
5 1 1 1
3
δ a b24 δ2 a 2 b 4
≤ 3 + 1 ,
|u| 2 |u| 2
For the third term of Fi,3 , we need to use the improved bounds for ∇i trχ for i ≥ 1 given by
282
Propositions 3.4.3 and 3.5.1. More precisely, we have the estimate
X X
kui+1 ∇i1 ψ i2 ∇i3 trχkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
1
X
≤(δ ku6 ∇5 trχkL∞
2 2 2
u Lu L (Su,u )
+δ kui+1 ∇i trχkL∞ ∞ 2
u Lu L (Su,u )
)ku−1 kL2u
i≤4
X
+δ kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤5
i1 ≤2
X
× kui3 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
ku−1 kL2u
i3 ≤4
X
+δ kui1 +i2 +1 ∇i1 ψ i2 +1 kL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤5
i1 ≤4
X
× kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
ku−1 kL2u
i3 ≤2
1
2
δ ab 4
≤ 1 ,
|u| 2
where we have used Propositions 3.4.3, 3.4.5 and 3.5.1, as well as applied Proposition 3.4.1 to
control the product of ψ.

1
Returning to (3.6.10), collecting all the above estimates and using the condition |u| ≥ δa 2 b, we
get
X 1 3 3
(kui+2 ∇i ((K − 2 ), σ̌)kL2u L2 (Su,u ) + kui+2 ∇i βkL2u L2 (Su,u ) ) ≤ δ 2 a 4 .
|u|
i≤4
We now turn to the remaining energy estimates, i.e., we derive the bounds for kui+1 ∇i βkL∞ 2 2
u Lu L (Su,u )
1
and kui+1 ∇i (K − |u|2 , σ̌)kLu Lu L (Su,u )
∞ 2 2 for i ≤ 4:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
X 1 1 1
(kui+1 ∇i βkL∞ 2 2
u Lu L (Su,u )
+ kui+1 ∇i (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
) ≤ δ 2 a2 .
|u|2
i≤4
Proof. As in the proof of Proposition 3.6.4, we derive the energy estimates from the Bianchi
283
equations. We consider the three schematic Bianchi equations below. First, the equation for ∇3 β:
1
∇3 β − ∇(K − ) −∗ ∇σ̌ + trχβ
|u|2
X 1 1 (3.6.14)
=ψ(K, σ̌) + ψ i1 +1 ∇i2 ψ + ψ∇(trχ, χ̂) + ψψ + ∇trχ.
i +i =1
|u| |u|
1 2
We also have the equation for ∇4 σ̌
X
∇4 σ̌ + div ∗ β = ψσ̌ + ψ∇η + ψ i1 +1 ∇i2 ψ (3.6.15)
i1 +i2 =1
1
and the equation for ∇4 (K − |u|2 )
1 1 X 1
∇4 (K − ) + div β = ψ(K − , σ̌) + ψ∇η + ψ i1 +1 ∇i2 ψ + 2 ψ. (3.6.16)
|u|2 |u|2 i +i =1
|u|
1 2
We commute the above equations with ∇i . From (3.6.14), we obtain
1 2+i
∇3 ∇i β − ∇∇i (K − ) −∗ ∇∇i σ̌ + trχ∇i β = G1,i (3.6.17)
|u|2 2
where G1,i is given by
1 5 X 1
G1,i =ψ∇5 ψ + ∇ trχ + ψ∇5 (χ̂, trχ) + ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)
|u| i
|u|2
1 +i2 +i3 =i
X 1 X
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 +i3 =i+1
|u| i
1 +i2 +i3 =i
i1 ,i3 ≤i
1 X
+ ∇i1 ψ i2 +1 .
|u|2 i1 +i2 =i
Using (3.6.15), we get
∇4 ∇i σ̌ + div ∗ ∇i β = G2,i , (3.6.18)
284
where
X
G2,i =ψ∇5 ψ + ψ∇5 η + ∇i1 ψ i2 ∇i3 ψ∇i4 σ̌
i1 +i2 +i3 +i4 =i
X
+ ∇i1 ψ i2 +1 ∇i3 ψ.
i1 +i2 +i3 =i+1
i1 ,i3 ≤i
Also, by (3.6.16), we have

1
∇4 ∇i (K − ) + div ∇i β = G3,i , (3.6.19)
|u|2
where
X 1
G3,i =ψ∇5 ψ + ∇i1 ψ i2 ∇i3 ψ∇i4 (K − , σ̌)
i1 +i2 +i3 +i4 =i
|u|2
X 1 X
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +1 .
i1 +i2 +i3 =i+1
|u|2 i1 +i2 =i
i1 ,i3 ≤i
As in the proof of Proposition 88, we use equations (3.6.17), (3.6.18),(3.6.19) and apply Propo-
sitions 3.6.1, 3.6.2 and 3.6.3 to obtain
X 1

kui+1 ∇i βk2L2u L2 (Su,u ) + kui+1 ∇i (K − )k2 2 + kui+1 ∇i σ̌k2L2u L2 (Su,u )
|u|2 L (H)
i≤4
X X
≤ kui+1 ∇i βk2L2u L2 (S0,u ) + kui+1 G1,i kL1u L2u L2 (Su,u ) + kui+1 G2,i kL1u L2u L2 (Su,u ) (3.6.20)
i≤4 i≤4
X
+ kui+1 G3,i kL1u L2u L2 (Su,u ) .
i≤4
2+i
Notice that the weight ui+1 is dictated by term i
2 trχ∇ β in (3.6.17). The proof of (3.6.20) is
otherwise analogous to that of (3.6.10) in Proposition 88 and is omitted.
We now estimate each of the terms on the right hand side (3.6.20). We begin with the term
G1,i . Among the terms in G1,i , we first bound the contributions where 5 derivatives fall on one of
285
the Ricci coefficients. More precisely, these are the terms
1 5
ψ∇5 ψ, ∇ trχ, ψ∇5 (χ̂, trχ). (3.6.21)
|u|
For the first term in (3.6.21), we have the estimate
≤ku5 ∇5 ψkL∞ 2 2
u Lu L (Su,u )
kψkL1u L∞ ∞
u L (Su,u )
1 3 1
1 1 δa 2 1 δ 2 ab 2
≤δ 2 a 2 b2 ≤ ,
|u| |u|
where we have used bootstrap assumptions (3.3.1) and (3.3.3). The second term of (3.6.21) can be
controlled by
ku4 ∇5 trχkL1u L2u L2 (Su,u )

1
≤ku6 ∇5 trχkL∞ 2 2
u Lu L (Su,u )
k kL1
|u|2 u
3 1
δ 2 ab 4
≤ .
|u|
Here, we have used the bound for ∇5 trχ in Proposition 3.5.1. Notice that it is important that
Proposition 3.5.1 gives a better bound for ∇5 trχ than other ∇5 ψ components. Then, turning to
the third term in (3.6.21), we use Proposition 3.4.5 and the bootstrap assumption (3.3.3) to obtain
the following estimate:
ku5 ψ∇5 (χ̂, trχ)kL1u L2u L2 (Su,u )
≤ku5 ∇5 (χ̂, trχ)kL∞ 2 2

u Lu L (Su,u )
1 1 1 3 1
δa 2 1 δ 2 a2 δ 2 ab 4
≤ 1 b
4
1 ≤ .
|u| 2 |u| 2 |u|
After bounding all the highest derivative Ricci coefficient terms, we now move to the term containing
1
(K − |u|2 , σ̌). Using Sobolev embedding in Proposition 3.3.9 together with Proposition 3.4.1 and
286
the bootstrap assumption (3.3.3), we have
X X 1
k ui+1 ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
|u|2
i≤4
X X 1
≤ kui1 +i2 ∇i1 ψ i2 +1 kL2u L2u L∞ (Su,u ) kui3 +1 ∇i3 (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
|u|2
i1 +i2 ≤2 i3 ≤4
X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL2u L2u L2 (Su,u )
i1 +i2 ≤4
X 1
× kui3 +2 ∇i3 (K − , σ̌)kL∞ 2 ∞
u Lu L (Su,u )
|u|2
i3 ≤2
1 1
1 1 1 δa 2 1 1 1 δ 2 ab 2
≤δ 2 a 2 b 4 |u| 2 δ 2 b4 ≤
3 .
|u|2 |u| 2
We then control the Ricci coefficient terms where there are at most 4 derivatives. For each i ≤ 4,
there are three terms to be bounded, namely
X X 1 i1 i2 +1 i3 X 1 i1 i2 +1
∇i1 ψ i2 +1 ∇i3 ψ, ∇ ψ ∇ ψ, ∇ ψ . (3.6.22)
i1 +i2 +i3 =i+1 i1 +i2 +i3 =i
|u| i1 +i2 =i
|u|2
i1 ,i3 ≤i
The first terms in (3.6.22) can be estimated as follows
X X
k ui+1 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
i1 ,i3 ≤i
X X
≤ kui1 +i2 ∇i1 ψ i2 +1 kL1u L∞ ∞
u L (Su,u )
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L∞ 2
u L (Su,u )
kui3 +1 ∇i3 ψkL∞ 2 ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
1 1 3 1
δa b 1 1
2 4 δ ab 2 4
≤ δ 2 a2 ≤ ,
|u| |u|
where we have used Propositions 3.4.1 and 3.4.5 and the bootstrap assumption (3.3.1).
287
The second term in (3.6.22) can be controlled in a very similar fashion:
X X
k ui ∇i1 ψ i2 +1 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
X X
≤ kui1 +i2 ∇i1 ψ i2 +1 kL1u L∞ ∞
u L (Su,u )
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L∞ 2
u L (Su,u )
kui3 +1 ∇i3 ψkL∞ 2 ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
1 1 3 1
δa b 1 1
2 4 δ ab 2 4
≤ δ 2 a2 ≤ ,
|u| |u|
where we have again used Proposition 3.4.1 and 3.4.5 and the bootstrap assumption (3.3.1).
We now turn to the last term of (3.6.22), for which we have the following bound:
X X
k ui−1 ∇i1 ψ i2 +1 kL1u L2u L2 (Su,u )
i≤4 i1 +i2 =i
X X
≤ kui1 +i2 −1 ∇i1 ψ i2 kL2u L∞ ∞
u L (Su,u )
kui3 ∇i3 ψkL2u L2u L2 (Su,u )
i1 +i2 ≤2 i3 ≤4
3 1 1
δ a b
2 2 4
≤ ,
|u|
after using Proposition 3.4.1 and bootstrap assumption (3.3.1).
This concludes the estimates for G1,i We now move to the estimates for G2,i and G3,i . It is
easy to observe that all the terms of G2,i are in fact contained in the expression for G3,i . Hence, it
suffices to control the terms in G3,i . As in the estimates for G1,i , we first bound the contributions
where there are 5 derivatives on one of the Ricci coefficients. There are two terms of this type,
namely,
ψ∇5 ψ, ψ∇5 η.
288
For the first term, we have
≤ku5 ∇5 ψkL∞ 2 2
u Lu L (Su,u )
3 1 1 1
1 1 1 δ 2 a2 b4 δ 2 ab 2
≤δ 2 a 2 b 4 3 ≤ 3 ,
|u| 2 |u| 2
where we have used the bootstrap assumptions (3.3.1) and (3.3.3). The second term can be con-
trolled after using Proposition 3.4.5 and the bootstrap assumption (3.3.3):
ku5 ψ∇5 ηkL1u L2u L2 (Su,u )

1
≤ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
k ψkL2u L2u L∞ (Su,u )
|u|
1 1 5 1
3 3 1 δ 2 a2 δ2 a 4 b 4
≤δ 2 a 4 b 4 3 ≤ 3 .
|u| 2 |u| 2
1
We now move to the curvature term containing (K − |u|2 , σ̌). We have
X X 1
k ui+1 ∇i1 ψ i2 ∇i3 ψ∇i4 (K − , σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 +i4 =i
|u|2
i≤4
X X
≤δ kui1 +i2 −1 ∇i1 ψ i2 kL∞ ∞ ∞
u Lu L (Su,u )
k∇i3 +1 ψkL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤2
X 1
× kui4 +1 ∇i4 (K − , σ̌)kL∞ 2 2
u Lu L (Su,u )
|u|2
i4 ≤4
X X
+δ kui1 +i2 −1 ∇i1 ψ i2 kL∞ ∞ ∞
u Lu L (Su,u )
k∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X 1
× kui4 +2 ∇i4 (K − , σ̌)kL∞ 2 ∞
u Lu L (Su,u )
|u|2
i4 ≤2
X X
+δ kui1 +i2 −2 ∇i1 ψ i2 kL∞ ∞ 2
u Lu L (Su,u )
k∇i3 +1 ψkL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
X 1
× kui4 +2 ∇i4 (K − 2 , σ̌)kL∞ 2 ∞
u Lu L (Su,u )
|u|
i4 ≤2
3 1
δ 1 1 1 1 δ 2 ab 4
≤ a2 · δ 2 a2 b4 ≤ ,
|u| |u|
289
where we have used Sobolev embedding in Proposition 3.3.9 and also Propositions 3.4.1 and 3.4.5
and the bootstrap assumptions (3.3.1) and (3.3.3). We finally turn to the remaining two terms, i.e.,
the Ricci coefficient terms with at most 4 derivatives. These are the terms
X X 1 i1 i2 i3
∇i1 ψ i2 +1 ∇i3 ψ, ∇ ψ ∇ ψ
i1 +i2 +i3 =i+1 i1 +i2 +i3 =i
|u|2
i1 ,i3 ≤i
for i ≤ 4. For the first term, we have
X X
k ui+1 ∇i1 ψ i2 +1 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
i1 ,i3 ≤i
X X
≤ kui1 +i2 ∇i1 ψ i2 +1 kL1u L2u L∞ (Su,u ) kui3 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −1 ∇i1 ψ i2 +1 kL1u L2u L2 (Su,u ) kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
3 1 1 1
2
δ a b
2 2 4 1 1 δ ab 4
≤ 3 δ 2 a2 ≤ 3 ,
|u| 2 |u| 2
where we have used Propositions 3.4.1 and 3.4.5 and the bootstrap assumption (3.3.1). Finally,
using again Propositions 3.4.1 and 3.4.5 and the bootstrap assumption (3.3.1), we obtain
X X
k ui−1 ∇i1 ψ i2 ∇i3 ψkL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i
X X
≤ kui1 +i2 −1 ∇i1 ψ i2 kL1u L2u L∞ (Su,u ) kui3 ∇i3 ψkL∞ ∞ 2
u Lu L (Su,u )
i1 +i2 ≤2 i3 ≤4
X X
+ kui1 +i2 −2 ∇i1 ψ i2 kL1u L2u L2 (Su,u ) kui3 +1 ∇i3 ψkL∞ ∞ ∞
u Lu L (Su,u )
i1 +i2 ≤4 i3 ≤2
3 1 1 1
2
δ a b
2 2 4 1 1 δ ab 4
≤ 3 δ 2 a2 ≤ 3 .
|u| 2 |u| 2
1
We have thus estimated all of the error terms. Notice that since |u| ≥ δa 2 b, all the error terms can
be controlled by
1 1
≤ δ 2 a2 .
290
Returning to (3.6.20), we therefore conclude the proof of the proposition.
1
With Propositions 3.6.4 and 3.6.5, the only remaining energy estimate is that for K − |u|2 in
the case i = 0. To obtain the desired bounds, we will simply integrate the estimates we obtained
in Proposition 3.4.9 to get the following proposition:
(3.3.1), (3.3.2), (3.3.3) and (3.3.4), we have
1 3 3
ku2 (K − 2
)kL∞ 2 2
u Lu L (S)
≤δ 2 a 4 .
|u|
Proof. This follows immediately from Proposition 3.4.9:
1 1 1
ku2 (K − 2
)kL∞ 2 2
u Lu L (S)
≤ δ 2 · δa 2 (1 + Õ5,2 + R)
|u|
and the bootstrap assumption (3.3.3).
Combining Propositions 3.6.4, 3.6.5 and 3.6.6, we have thus obtained
R ≤ 1. (3.6.23)
Substituting the bound (3.6.23) into Proposition 3.5.7, we have
Õ5,2 ≤ 1. (3.6.24)
(3.6.23) and (3.6.24) together improve over the bootstrap assumption (3.3.3) for b sufficiently large.
Now, using (3.6.23) and (3.6.24) together with Propositions 3.4.5 and 3.4.6 and the Sobolev em-
bedding in Proposition 3.3.9, we obtain
X 1 X 1
1 kui+1 ∇i ψkL2 (Su,u ) + 1 kui+2 ∇i ψkL∞ (Su,u ) ≤ 1
i≤4
δa 2
i≤2
δa 2
291
and
X 1 X 1
1 kui ∇i ψkL2 (Su,u ) + 1 kui+1 ∇i ψkL∞ (Su,u ) ≤ 1,
i≤4
a 2
i≤2
a2
which improve over (3.3.1) and (3.3.2) respectively, after choosing b to be sufficiently large. Finally,
recall from Proposition 3.4.9 that we have
X 1 1
kui+1 ∇i (K − 2
)kL∞ ∞ 2
u Lu L (Su,u )
≤ 3,
|u| b4
i≤3
which improves over (3.3.4) for b sufficiently large. We have thus recovered all the bootstrap
assumptions and showed that the bound
O, Õ5,2 , R ≤ 1
holds.
This concludes the proof of Theorem 3.2.1.
3.7 Formation of Trapped Surfaces
In the section, we will prove Theorem 3.2.2. The proof will make use of the bounds in Theorem
3.2.1 and follows from an ODE argument as in [7]. We first need to obtain some refined estimates
compared to that in the proof Theorem 3.2.1.
First, using the bound for ω proved in Proposition 3.4.8 instead of applying the bootstrap
assumption, we can revisit the proof of Proposition 3.3.1 to obtain the following improved estimate
for Ω:
Proposition 3.7.1. Under the assumptions of Theorem 3.2.1, we have
1
δa 2
kΩ−1 − 1kL∞ (Su,u ) ≤ .
|u|
292
Proof. Using (3.3.5), we have
Z u 1
−1 δa 2
kΩ − 1kL∞ (Su,u ) ≤ kωkL∞ (Su,u0 ) du0 ≤ ,
0 |u|
where in the last step we have used the estimate for ω derived in Theorem 3.2.1.
We then also need the following improved estimate for trχ. More precisely, we show that while
the bound in Proposition 3.4.3 is in general sharp, the main contribution comes from the |χ̂|2γ term.
This will allow us to obtain an upper bound for |trχ(u, u, θ1 , θ2 )| depending on the integral of |χ̂|2γ
along the characteristic in the (θ1 , θ2 ) direction. This improvement in the bound for trχ will in
turn enable us to prove a lower bound for the integral of |χ̂|2γ in the proof of Theorem 3.2.2.
Proposition 3.7.2. Under the assumptions of Theorem 3.2.1, we have
Z δ
1
|trχ(u, u, θ1 , θ2 )| ≤ + |χ̂|2γ (u, u0 , θ1 , θ2 )du0
|u| 0
for every (θ1 , θ2 ).
Proof. Fix (θ1 , θ2 ). Recall that
∇4 trχ + |χ̂|2 = ψtrχ.
2
Integrating this equation and recalling that initially trχ = |u| , we obtain
1 Z δ
1 δa 2
1 2
|trχ(u, u, θ , θ )| ≤ + sup |trχ(u, u0 , θ1 , θ2 )| + |χ̂|2γ (u, u0 , θ1 , θ2 )du0 .
|u| |u| u0 ≤u 0
1
δa 2
Since |u| ≤ 1b , we can choose b to be sufficiently large to obtain
Z δ
1
1 2
|trχ(u, u, θ , θ )| ≤ + |χ̂|2γ (u, u0 , θ1 , θ2 )du0 .
|u| 0
With these estimates, we are now ready to prove Theorem 3.2.2.
Proof. [Proof of Theorem 3.2.2]
293
In order to show that S 1 is a trapped surface, we need to prove that trχ < 0 and trχ < 0
bδa 2 ,δ
everywhere on S 1 . The fact that trχ < 0 follows directly from the estimates we have proved.
bδa 2 ,δ
More precisely, recall from Theorem 3.2.1 that we have the following bound for trχ in the region
1
δa 2 b ≤ |u| ≤ 1:
1
2 δa 2
ktrχ + kL∞ (Su,u ) ≤ .
|u| |u|2
In particular,
trχ < 0 (3.7.1)
on the sphere S 1 .
bδa 2 ,δ
It now remains to show that trχ < 0. Fix a point p = (θ1 , θ2 ) ∈ S1,0 . We will show that
1
trχ(u = δa 2 b, u = δ, θ1 , θ2 ) < 0. To achieve this, we make the following bootstrap assumption for
1
every u ∈ [δa 2 b, 1]:
Z δ Z δ
2
u |χ̂|2γ (u, u, θ1 , θ2 )du ≤2 |χ̂|2γ (1, u, θ1 , θ2 )du. (3.7.2)
0 0
1
Our strategy is to obtain a lower bound for |χ̂|2γ (u = δa 2 b, u, θ1 , θ2 ) under the assumption (3.7.2).
1
This will in turn be used to get an upper bound for trχ(u = δa 2 b, u = δ, θ1 , θ2 ). First, as an
immediate consequence of the bootstrap assumption (3.7.2) and Proposition 3.7.2, we have
Z δ
1 1
|trχ(u, u, θ1 , θ2 )| ≤ + 2 |χ̂|2γ (1, u0 , θ1 , θ2 )du0 . (3.7.3)
|u| |u| 0
Now consider the null structure equation
1 1
∇3 χ̂ + trχχ̂ = ∇⊗η
b + 2ω χ̂ − trχχ̂ + η ⊗η.
b
2 2
Contracting this equation with χ̂, we have
1 1 1
∇3 |χ̂|2γ + trχ|χ̂|2γ − 2ω|χ̂|2γ = χ̂(∇⊗η
b − trχχ̂ + η ⊗η).
b
2 2 2
294
In the coordinate system introduced in Section 3.1.3, we have
1 ∂ ∂ 1 1 2
− ( + dA A )|χ̂|2γ − |χ̂|2γ + (trχ + )|χ̂|2γ − 2ω|χ̂|2γ
2Ω ∂u ∂θ Ω|u| 2 Ω|u|
1
b − trχχ̂ + η ⊗η).
=χ̂(∇⊗η b
2
Using ω = − 12 ∇3 (log Ω), we can rewrite this as
∂ 2 −2 2
u−2 (u Ω |χ̂|γ )
∂u
∂ 2
= (Ω−2 |χ̂|2γ ) + 2 |χ̂|2γ (3.7.4)
∂u Ω |u|
1 2 ∂
= − 2Ω−1 χ̂(∇⊗η b + Ω−1 (trχ +
b − trχχ̂ + η ⊗η) )|χ̂|2γ + dA A (Ω−2 |χ̂|2 ).
2 Ω|u| ∂θ
We now obtain pointwise bounds for the terms on the right hand side. After bounding Ω−1 in L∞
using Proposition 3.3.1, the first term can be re-written schematically as
X
ψψ i1 ∇i2 ψ + ψψtrχ.
i1 +i2 =1
Using the bounds in Theorem 3.2.1 together with (3.7.3), we have
X
k ψψ i1 ∇i2 ψkL∞ (Su,u ) + kψψtrχkL∞ (Su,u )
i1 +i2 =1
(3.7.5)
Z δ
δa δa
≤ 4+ 5 |χ̂|2γ (1, u0 , θ1 , θ2 )du0 .
|u| |u| 0
For the second term on the right hand side of (3.7.4), note that
2 2 2
ktrχ + kL∞ (Su,u ) =ktrχ + kL∞ (Su,u ) + k (Ω−1 − 1)kL∞ (Su,u )
Ω|u| |u| |u|
1
δa 2
≤ ,
|u|2
where we have used the bounds proved in Theorem 3.2.1 together with Proposition 3.7.1. Therefore,
295
1
for every u ∈ [δa 2 b, 1], u ∈ [0, δ], we have
1
2 δa 2 2
|Ω−1 (trχ + )|χ̂|2γ |(u, u, θ1 , θ2 ) ≤ |χ̂| (u, u, θ1 , θ2 ). (3.7.6)
Ω|u| |u|2 γ
For the third term on the right hand side of (3.7.4), we have
∂
kdA (Ω−2 |χ̂|2 )kL∞ (Su,u )
∂θA
≤kdkL∞ (Su,u ) k∇χ̂kL∞ (Su,u ) kψkL∞ (Su,u ) + kdkL∞ (Su,u ) kψψψkL∞ (Su,u ) (3.7.7)
a
≤ kdkL∞ (Su,u ) .
|u|3
To obtain the estimate for d, we use the fact that
∂dA ∂
[L, L] =
∂u ∂θA
which implies
∂dA
= −4Ω2 ζ A .
∂u
Integrating this in the u direction and applying the bounds in Theorem 3.2.1 thus give
1
δ2 a 2
kdkL∞ (Su,u ) ≤ .
|u|2
Combining this with (3.7.7), we thus have
3
∂ −2 δ2 a 2
kdA (Ω | χ̂| 2
)k L∞ (S ) ≤ . (3.7.8)
∂θA u,u
|u|5
Multiplying (3.7.4) by u2 , integrating in u and using the bounds (3.7.5), (3.7.6) and (3.7.8), we
thus have
|u2 |χ̂|2γ (u, u, θ1 , θ2 ) − |χ̂|2γ (1, u, θ1 , θ2 )|

1
Cδa δ 2
Z Z 1
Ca 2 0 1 2 0 1
≤ + |χ̂|γ (1, u , θ , θ )du + Cδa 2 |χ̂|2γ (u0 , u, θ1 , θ2 )du0 ,
b |u|2 0 u
296
for some universal constant C. Integrating in u from 0 to δ, we obtain
Z δ Z δ
| u 2
|χ̂|2γ (u, u0 , θ1 , θ2 )du0 − |χ̂|2γ (1, u0 , θ1 , θ2 )du0 |
0 0
1
Cδ 2 a δ 2
Z Z δZ 1
Cδa 2 0 1 2 0 1
≤ + | χ̂| γ (1, u , θ , θ )du + Cδa 2 |χ̂|2γ (u0 , u, θ1 , θ2 )du0 du0
b |u|2 0 0 u
1
C δ 2
Z
Cδa 2
≤ + |χ̂|γ (1, u0 , θ1 , θ2 )du0
b b 0
Z 1 Z δ
1 1 0
+ Cδa (
2
0 2
du )(sup (u0 )2 |χ̂|2γ (u0 , u, θ1 , θ2 )du0 )
u |u | u0 0
1 1 Z
C δ 2 2Cδa 2 δ 2
Z
Cδa 2
≤ + |χ̂|γ (1, u0 , θ1 , θ2 )du0 + |χ̂|γ (1, u, θ1 , θ2 )du0
b b 0 |u| 0
1
3C δ 2
Z
Cδa 2
≤ + |χ̂|γ (1, u0 , θ1 , θ2 )du0 ,
b b 0
where we have used the bootstrap assumption (3.7.2). After choosing b sufficiently large so that
C 1
b ≤ 15 , this yields the upper bound
Z δ Z δ 1
6 2Cδa 2
u2 |χ̂|2γ (u, u0 , θ1 , θ2 )du0 ≤ |χ̂|2γ (1, u0 , θ1 , θ2 )du0 + .
0 5 0 b
and the lower bound
Z δ Z δ 1
4 2Cδa 2
u2 |χ̂|2γ (u, u0 , θ1 , θ2 )du0 ≥ |χ̂|2γ (1, u0 , θ1 , θ2 )du0 − .
0 5 0 b
Recalling that the following assumption in Theorem 3.2.2
Z δ
1
|χ̂0 |2 (1, u0 , θ1 , θ2 )du0 ≥ 4bδa 2 ,
0
we can thus choose b to be sufficiently large to guarantee that
Z δ Z δ
3
2
u |χ̂|2γ (u, u0 , θ1 , θ2 )du0 ≤ |χ̂|2γ (1, u0 , θ1 , θ2 )du0 . (3.7.9)
0 2 0
297
and
Z δ Z δ
3 1
u2 |χ̂|2γ (u, u0 , θ1 , θ2 )du0 > |χ̂|2γ (1, u0 , θ1 , θ2 )du0 ≥ 3bδa 2 . (3.7.10)
0 4 0
In particular, (3.7.9) improves over the bootstrap assumption (3.7.2). Therefore, (3.7.10) holds. We
now use (3.7.10) to obtain the desired upper bound for trχ. To this end, we combine the transport
equation for trχ

1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ωtrχ
2
and ω = − 21 ∇4 (log Ω) to get
Ω−1
∇4 (Ω−1 trχ) = − (trχ)2 − Ω−1 |χ̂|2 .
2
Using the fact e4 = Ω−1 ∂u

∂
, we integrate this equation to obtain
1
Ω−1 trχ(bδa 2 , δ, θ1 , θ2 )
Z δ
1 1
−1
≤Ω 1
trχ(bδa , 0, θ , θ ) −
2 2
|χ̂|2 (bδa 2 , u0 , θ1 , θ2 )du0
0 (3.7.11)
2 3
≤ 1 − 1
bδa 2 bδa 2
<0.
Here, we have used the lower bound (3.7.10) for the integral of |χ̂|2γ .
Therefore, by (3.7.1) and (3.7.11), S 1 is a trapped surface.

bδa 2 ,δ
298
Bibliography
[1] X. An, Formation of trapped surfaces from past null infinity, preprint (2012), arXiv:1207.5271.
[2] X. An, J. Luk, Trapped surfaces in vacuum arising dynamically from mild incoming radiation,
in preparation (2014).
[3] D. Christodoulou, The formation of black holes and singularities in spherically symmetric
gravitational collapse, Comm. Pure Appl. Math. 44 (1991), no. 3, 339–373.
[4] D. Christodoulou, Bounded variation solutions of the spherically symmetric Einstein-scalar
field equations Comm. Pure Appl. Math. 46 (1993), no. 8, 1131-1220
[5] D. Christodoulou, Examples of naked singularity formation in the gravitational collapse of a
scalar field, Ann. of Math. (2) 140 (1994), no. 3, 607–653.
[6] D. Christodoulou, The instability of naked singularities in the gravitational collapse of a scalar
field, Ann. of Math. (2) 149 (1999), no. 1, 183–217.
[7] D. Christodoulou, The formation of black holes in general relativity, Monographs in Mathe-
matics, European Mathematical Soc. (2009).
[8] D. Christodoulou, S. Klainerman, The global nonlinear stability of the Minkowski space, Prince-
ton mathematical series 41 (1993).
[9] M. Dafermos, The formation of black holes in general relativity, Astrisque 352 (2013).
299
[10] M. Dafermos, G. Holzegel, I. Rodnianski, A scattering theory construction of dynamical vacuum
black holes, preprint (2013), arXiv:1306.5364.
[11] G. Holzegel, Ultimately Schwarzschildean spacetimes and the black hole stability problem,
preprint (2010), arXiv:1010.3216.
[12] S. Klainerman, F. Nicolo, The evolution problem in general relativity, Progress in Mathematical
Physics, Birkhaüser (2003).
[13] S. Klainerman, J. Luk, I. Rodnianski, A fully anisotropic mechanism for formation of trapped
surfaces in vacuum, to appear in Invent. Math. (2013).
[14] S. Klainerman, I. Rodnianski, On emerging scarred surfaces for the Einstein vacuum equations,
Discrete Contin. Dyn. Syst. , 28 (2010), no. 3, 1007–1031.
[15] S. Klainerman, I. Rodnianski, On the the formation of trapped surfaces, Acta Math. 208 (2012),
no.2, 211–333.
[16] J. Li, P. Yu, Construction of Cauchy data of vacuum Einstein field equations evolving to black
holes, preprint (2012), arXiv:1207.3164.
[17] J. Luk, Weak null singularities in general relativity, preprint (2013), arXiv:1311.4970.
[18] J. Luk, I. Rodnianski, Local propagation of impulsive gravitational waves, preprint (2012),
arXiv:1209.1130.
[19] J. Luk, I. Rodnianski, Nonlinear interactions of impulsive gravitational waves for the vacuum
Einstein equations, preprint (2013), arXiv:1301.1072.
[20] R. Penrose, Gravitational collapse and space-time singularities, Phys. Rev. Lett. 14 (1965),
57–59.
[21] M. Reiterer, E. Trubowitz, Strongly focused gravitational waves Comm. Math. Phys. 307 (2011),
no. 2, 275–313.
300
[22] P. Yu, Energy estimates and gravitational collapse, Comm. Math. Phys. 317 (2013), no. 2,
273–316.
[23] P. Yu, Dynamical formation of black holes due to the condensation of matter field, preprint
(2011), arXiv:1105.5898.
301

Formation of Trapped Surfaces in General Relativity

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Formation of Trapped Surfaces in General Relativity

Caricato da

Copyright:

Formati disponibili

Formation of Trapped Surfaces

Presented to the Faculty

in Candidacy for the Degree

Recommended for Acceptance

Adviser: Sergiu Klainerman

All Rights Reserved

result in a finite region.

way. The second result is obtained jointly with Luk.

mathematics always motivate me to explore the mathematical world.

discussions and help in my PhD years.

standings and gave great suggestions and support to my study.

Zhang for the friendship along the way.

Jennifer Johnson for their help and support.

and company along these years.

1.1 Formation of Trapped Surfaces from Past Null Infinity . . . . . . . . . . . . . . . . . 3

1.1.2 Heuristic Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1.3 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.2 Formation of Trapped Surfaces in Vacuum from Mild Incoming Radiation

(Joint with Luk) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.2.2 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.2.3 Heuristic Argument with ESFS-S . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Formation of Trapped Surfaces from Past Null Infinity 28

2.1 Setting, Equations and Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.1.1 Double Null Foliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.1.2 The Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2 Statement of Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.2.1 Structure of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.2.2 Strategy of the Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.3 The Preliminary Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.3.1 Estimates for Metric Components . . . . . . . . . . . . . . . . . . . . . . . . 48

2.3.2 Estimates for Transport Equations . . . . . . . . . . . . . . . . . . . . . . . . 50

2.3.3 Sobolev Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.3.4 Commutation Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.3.5 General Elliptic Estimates for Hodge Systems . . . . . . . . . . . . . . . . . . 57

2.4 Ricci Coefficient Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.4.1 Estimates for Zero Derivatives of Ricci Coefficients . . . . . . . . . . . . . . . 62

2.4.2 Estimates for First Derivatives of Ricci Coefficients . . . . . . . . . . . . . . . 72

2.4.3 L∞ (S) Estimates for Ricci Coefficients . . . . . . . . . . . . . . . . . . . . . . 88

2.5 Elliptic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

2.5.1 L4 (S) Estimates for Curvature Components . . . . . . . . . . . . . . . . . . . 95

2.5.2 L2 (S) Estimates for First Derivatives of Curvature Components . . . . . . . 98

2.5.3 L2 (S) Estimates for Third Derivatives of Ricci Coefficients . . . . . . . . . . 102

2.5.4 L4 (S) Estimates for Second Derivatives of Ricci Coefficients . . . . . . . . . . 135

2.6 Curvature Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

2.6.1 Curvature Estimates in the Scale Invariant Norms . . . . . . . . . . . . . . . 138

2.6.2 Curvature Estimate I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

2.6.3 Curvature Estimate II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

2.6.4 Curvature Estimate III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

2.7 Formation of Trapped Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

2.8 Retrieving Christodoulou’s Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

2.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

2.9.2 Equations for Elliptic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 180

2.9.3 Equations for Energy Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . 190

3 Formation of Trapped Surfaces in Vacuum from Mild Incoming Radiation

(Joint with Luk) 197

3.1 Setting, Equations and Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

3.1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

3.1.2 Double Null Foliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

3.1.3 The Coordinate System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

3.1.4 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

3.1.5 Schematic Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

3.1.6 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

3.1.7 Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206

3.2 Statement of main theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208