Naked Singularity Censoring With Anisotropic Apparent Horizon

NAKED SINGULARITY CENSORING
WITH ANISOTROPIC APPARENT HORIZON
XINLIANG AN
Abstract. Employing the Einstein-scalar field system, we demonstrate an approach for prov-
ing high co-dimensional nonlinear instability of naked-singularity solutions as constructed by
Christodoulou in [18]. We further investigate the censorship of Christodoulou’s naked singu-
larity and show that a tiny anisotropic perturbation arising from the outgoing characteristic
initial data would lead to the emergence of an anisotropic apparent horizon, which covers and
censors the naked singularity. Our approach advances the hyperbolic short-pulse method by
not requiring the aid of additional large parameters, by permitting the use of initial pertur-
bations for the shear tensor and the derivative of scalar field to be with finite BV and C 0
arXiv:2401.02003v1 [gr-qc] 3 Jan 2024
norms, and by allowing the initial perturbation to be arbitrarily small in scale-critical norms.
New elliptic arguments based on non-perturbative methods are also developed.
1. Introduction
In 1994 Christodoulou [18] constructed his famous naked-singularity solution in 3 + 1 dimen-
sions for the Einstein-scalar field system
1
Rµν − Rgµν = Tµν ,
2 (1.1)
1 λ
Tµν =Dµ φDν φ − gµν D φDλ φ.
2
3+1
Here φ : R → R is a real-valued scalar function and (M, g) is the 3+1 dimensional spacetime.
The stability or instability of these naked singularities is closely related to the weak cosmic
censorship [20, 34]:
Conjecture 1. (Weak cosmic censorship conjecture). For generic asymptotically flat initial
data, the maximal development of Einstein’s field equations possesses a complete future null
infinity I + and hides the possibly formed singularities in a (black hole) region causally discon-
nected from I + .
The above is one of the greatest open problems in classic general relativity. When Penrose
formulated the original weak cosmic censorship conjecture in [34], the caveat generic was not
included. Christodoulou’s example in [18] surprisingly shows that naked singularity could form
in the evolution of Einstein’s field equations. Within spherical symmetry, for the Einstein-scalar
field system (1.1), by a celebrated series of works [16,17,19] Christodoulou proved that the initial
datum leading to the naked-singularity formation is rare, and for generic initial data the possibly
formed singularities are indeed hidden in the trapped (black hole) region. This motivates him
to add the term generic for Conjecture 1 in [20].
Christodoulou’s proofs in [16,17,19] reveal novel mathematical and physical properties of the
spherically symmetric gravitational collapse. For the initial datum leading to naked singularity,
in [19] he identified 2 instability mechanisms and deduced co-dimensional 2 instability. These
instability theorems show that each rare initial datum, which leads to naked singularity forma-
tion, is always associated with 2-dimensional perturbations and each of the perturbations would
give rise to a trapped surface and hence a black hole formation. Here a trapped surface is a
2-sphere in M with both null expansions negative.
In above arguments, the spherically symmetric perturbation required in [19] is a bit restric-
tive, since it imposes a global condition in all angles. Can we relax this requirement by only
placing a small anisotropic-in-angle perturbation? If it is possible, this may also significantly
Date: January 5, 2024.

1
2 XINLIANG AN
raise the co-dimensions of the instability theorems. In this article, we give such an anisotropic
result. In particular, we prove that for Christodoulou’s constructed naked singularity in [18], by
imposing a tiny perturbation from any angle, we can trigger a trapped region and an apparent
horizon to censor the naked singularity. Compared to Christodoulou’s instability argument that
the singularity is covered by later-formed trapped surfaces, here we further show that the naked
singularity is censored by an achronal apparent horizon emerging from it. Subjected to a general
anisotropic perturbation, the naked singularity in [18] is not naked even locally. It is completely
censored for observers outside the apparent horizon.
1.1. Main Theorem. Here we use the double-null foliation for the spacetime (M, g). Let u
and u be optical functions increasing towards the future and satisfy the eikonal equations
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0.
Set Hu to be the level sets of u and H u to be the level sets of u. And we require that Hu and H u
correspond to the outgoing and incoming null cones, respectively. The intersections of Hu and
H u are topologically 2-spheres and are denoted by Su,u . We further fix a frame e1 , e2 tangent
to Su,u and denote ∇ to be the angular derivative on it. In this paper, we employ the double
null coordinates (u, u, θ1 , θ2 ) for (M, g), where ∂/∂θ1 and ∂/∂θ2 are tangential to Su,u .1 For
A, B = 1, 2, the spacetime metric g takes the form
g = −2Ω2 (du ⊗ du + du ⊗ du) + g/AB (dθA − dA du) ⊗ (dθB − dB du).
Here Ω(u, u, θ1 , θ2 ) is the lapse function, g/AB is the induced metric on Su,u , and A, B = 1, 2.
For notational simplicity, in this article we also use ω ∈ S2 to denote (θ1 , θ2 ). With the help of
Ω, we choose the null pair e3 , e4 in this article as

∂ ∂ ∂
e3 = Ω−1 , e4 = Ω−1 + dA A
∂u ∂u ∂θ
with dA obeying dA = 0 on H−1 . And it holds
g(e3 , e3 ) = 0, g(e4 , e4 ) = 0, g(e3 , e4 ) = −2.
Define
χAB := g(DeA e4 , eB )
and set χ̂ to be its traceless part, i.e., the shear tensor. In this article, we first prove
Theorem 1.1. Consider the characteristic initial value problem for the Einstein-scalar field
system (1.1). Assigning Christodoulou’s naked-singularity initial data in [18] along H 0 with
−1 ≤ u ≤ 0 and prescribing perturbed initial data along H−1 satisfying
X
uj k(∂u )j ∇i χ̂kL∞ 2
u L (S−1,u )
+ uj k(∂u )j ∇i ∇4 φkL∞ 2
u L (S−1,u )
≤ B, (1.2)
i≤5,j≤3
then for each B there exist a sufficient small δ = δ(B) and the Einstein-scalar field system
admits a unique regular solution in the spacetime region (u, u, ω) satisfying 0 ≤ u ≤ δ and
u ≤ |u|Ω2−δ̃ (u, 0) ≤ 1 with 0 < δ̃ ≪ 1 a fixed constant.
Moreover, if we further assume
Z u
|χ̂|2 (−1, u′ , ω) + [∇4 φ(−1, u′ , ω) − ∇4 φ(−1, 0, ω)]2 du′ = f (ω, u)u, (1.3)
0
with f (ω, u) obeying
0 ≤ f ≤ 1 on S2 × (0, δ] and f (·, u) ≥ m on Bp (ǫ) (1.4)
2
for a point p ∈ S and constants m ∈ (0, 1) and ǫ ∈ (0, π/2), then within the solved hyperbolic
region, there exists a unique MOTS (marginally outer trapped surface) Mu on each H u with
0 < u ≤ δ and the collection of MOTS {Mu } emerges and censors the central singularity and
forms an achronal apparent horizon.
1See Chapter 1 and Chapter 2 in [21] for a detailed discussion of the double null coordinates.
NAKED SINGULARITY CENSORING WITH ANISOTROPIC APPARENT HORIZON 3
For the shear tensor and the derivative of scalar field, the above theorem prescribes initial
[0,δ]
perturbation along H−1 with a finite BV norm satisfying (1.3). Our next theorem further
[0,δ]
allows initial perturbation along H−1 to be with finite C 0 norm.
Theorem 1.2. The above statements in Theorem 1.1 still hold, if we replace the requirements
for f (ω, u) in (1.3) and (1.4) by
1 1
f (ω, u) = g(u)f˜(ω, u) with g(u) ≈ [ln(ln )]− 2 obeying lim+ g(u) = 0,
u u→0
and f˜(ω, u) satisfying 0 ≤ f˜ ≤ 1 on S2 × (0, δ], f˜(·, u) ≥ m on Bp (ǫ) for a point p ∈ S2 with
constants m ∈ (0, 1) and ǫ ∈ (0, π/2).
,ũ
,δ
µ∗
µ∗
1−
1−
a)
a)
(ũ
(δ
−
−
S
S
O
Hδ
(u
H ũ
=
δ)
(u
=
H0
ũ)
1)
(u
−
=
=
0)
1(
u
−
H
Figure 1
Remark 1. To set the above-mentioned C 0 perturbation, we can let

1 1 1
|χ̂|(−1, u, ω) + |∇4 φ(−1, u, ω) − ∇4 φ(−1, 0, ω)| = [ln(ln )]− 4 · f˜(ω) 2
u
˜
with f (ω) being a smooth function depending only on ω. By Proposition 13.4, we have
1 1 1 1
k[ln(ln )]− 4 k 21 . [ln ]− 2 ≪ 1.
u Ḣ ([0,δ]) δ
Hence, with Theorem 1.2, to trigger the censoring of naked-singularity, we only need the scale-
critical perturbation of g in k · k 23 [0,δ] norm, i.e.,
Ḣ (H−1 )
the k · k 1 [0,δ] =k·k 1 +k·k 1 norm

Ḣ 2 (H−1 ) L2ω Ḣu2 ([0,δ]) L2u Ḣω2 (S−1,u )
1
of χ̂(−1, u, ω) and ∇4 φ(−1, u, ω) − ∇4 φ(−1, 0, ω) to be of order [ln δ1 ]− 2 ≪ 1. And as δ → 0+ ,
we allow the size of the perturbation to shrink to 0.
Remark 2. Compared with the previous short-pulse results in the scale-critical regime, here
we don’t need the additional large parameter a and we can view that the parameter a is set to
be a = a(B) = 1. This is an important advance for the short-pulse method invented in [21] by
1
Christodoulou with a = δ − 2 ≫ 1 and generalized in [2] and [4] with a = a(B) ≫ 1 by the author
and Luk, and by the author, respectively. With the aid of this parameter a, we could consider the
1
strength of the short pulse to be of order a 2 . The larger the short-pulse perturbation, the easier
to form a trapped surface. A novel insight of this paper is that, via employing the instability
4 XINLIANG AN
nature of the interior naked-singularity solution, to form an anisotropic trapped surface and the
corresponding apparent horizon, we can set the upper bound of the short-pulse perturbation to
1
be of order a(B) 2 = 1, and according to (1.4) we could even set the size of short pulse to be
a universal small number m. As pointed out in Remark 1, we can further choose its size to be
g(u) with g(u) → 0 as u → 0+ . The perturbation allowed can be arbitrarily small measured at
the scale-critical level. This is markedly different from [2] by the author and Luk, and [6] by the
author and Han, where the constant a = a(B) has to be a large parameter and the fact 1/a ≪ 1
is utilized crucially throughout [2, 6] to control the nonlinear terms. This paper contains other
hyperbolic and elliptic advancements. To capture the key features of the anisotropic instability
mechanism, we also develop quite a few new ingredients. These are summarized in Section 1.4.
Remark 3. In this paper, all the estimates still hold if we restore sharp a-weights and set
a = a(B) to be a parameter greater than or equal to 1. In particular, we allow a = a(B) ≫ 1,
but we do not need a ≫ 1 to close the arguments. To be consistent with [2,6] and to contain more
analytic information for further use, in below sections we work under the following assumptions
for initial perturbation:
X 1 1
uj a− 2 k(∂u )j ∇i χ̂kL∞ 2
u L (S−1,u )
+ uj a− 2 k(∂u )j ∇i ∇4 φkL∞ 2
u L (S−1,u )
≤ B, (1.5)
i≤5,j≤3
and
Z u
|χ̂|2 (−1, u′ , ω) + [∇4 φ(−1, u′ , ω) − ∇4 φ(−1, 0, ω)]2 du′ = f (ω, u)ua. (1.6)
0
Setting a = a(B) = 1, we then retrieve assumptions (1.2), (1.3).
Remark 4. In below, we also give a stereoscopic perspective, which demonstrates naked-

singularity censoring subject to an anisotropic perturbation. Note that, as portrayed in the
following picture, the anisotropic apparent horizon censors the naked singularity even locally.
No local observer before the apparent horizon can detect the naked singularity. What we prove
in this paper is a more precise statement than showing tiny trapped surfaces formed around O.
O
Anisotropic Apparent Horizon
Anisotropic Perturbation
along u = −1
S−1,0
Figure 2
Our next result further exhibits a high co-dimensional nonlinear instability:
Theorem 1.3. Consider the space of prescribed initial data for χ̂(−1, u, ω)q
and ∂u φ(−1, u, ω)
√ 1−µ∗
along u = −1 with u ≥ 0. With |χ̂|(−1, 0, ω) = 0, ∂u φ(−1, 0, ω) = 2 · 1−µ
∗
µ∗ · 4 cor-
responding to their values of Christodoulou’s naked-singularity solution in [18], then for any
k ∈ Z+ , we can find functions f1 , f2 , ..., fk and g1 , g2 , ..., gk , which are continuous with respect
to the u variable and smooth with respect to the ω variable, such that the below prescribed
initial data for |χ̂| and ∂u φ with u ≥ 0

|χ̂|(−1, u, ω), ∂u φ(−1, u, ω)

′ ′ ′
= 0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ1 g1 + λ2 g2 + · · · + λk gk
would lead to the trapped surface and the apparent horizon formation when
k
X k
X 2
λ2i + λ′i 6= 0 with λk , λ′k ∈ R.
1 1
Moreover, if it holds

0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ′1 g1 + λ′2 g2 +···+ λ′k gk

= 0 + λ̃1 f1 + λ̃2 f2 + · · · + λ̃k fk , ∂u φ(−1, 0, ω) + λ̃′1 g1 + λ̃′2 g2 +···+ λ̃′k gk ,
then it renders
λi = λ̃i and λ′i = λ̃i with 1 ≤ i ≤ k.
+
Hence, we may say that, for any k ∈ Z , Christodoulou’s naked-singularity solution has at least
co-dimension 2k nonlinear instability subject to outgoing continuous characteristic perturbations
at (−1, 0, ω).
Remark 5. In Section 13, we obtain two instability theorems. With respect to the shear tensor
and the derivative of the scalar field, Theorem 13.1 proves instability subject to perturbations
with finite BV norms and Theorem 1.3 proves instability subject to perturbations with finite
C 0 norms. In [2, 3, 6], to trigger the short pulse, BV perturbations for ∂u g/|S−1,0 are prescribed.
In [2, 3, 6], when u ≤ 0, the shear tensor χ̂ vanishes, and when u > 0, the shear tensor χ̂ obeys
a large lower bound related to the large parameter a. For this article, in Section 13.1 we first
prove Theorem 13.1, which is consistent with [2, 3, 6] by imposing the BV perturbations. In
Section 13.2, we further allow the perturbations to be C 0 . Additional gains in both hyperbolic
and elliptic parts enable us to do so. Detailed discussions are provided in Section 13.
Remark 6. Our above theorems prove arbitrarily large co-dimensional nonlinear instability for
a naked-singularity solution. This is not reported before and we spot the importance of the
anisotropic arguments. Together with Remark 1, we can see that this paper contains a desired
scale-critical nonlinear instability result.
Remark 7. The proofs of above two theorems and the approach we demonstrate here have
the potential to be generalized to other Einstein field equations. For the hyperbolic part, we
employ Ω(u, 0) → 0 as u → 0. For the elliptic part, we utilize Ωω(u, 0) = −∂u log Ω(u, 0)/2 >
0, which also holds for Rodnianski-Shlapentokh-Rothman’s naked-singularity in [35] for the
Einstein vacuum equations.
Our proofs of above two theorems are rooted in recent progress on trapped surface formation
and apparent horizon emergence. In below, we review these developments.
1.2. Formation of Trapped Surface. With a 589-page proof, Christodoulou [21] proved the
first trapped-surface-formation result for Einstein’s equations with no symmetry. As depicted in
Figure 3 below, Christodoulou considered a characteristic initial-value problem for the Einstein
vacuum equations in the shadowed region. Minkowskian initial data are prescribed along the
incoming null cone H 0 and short-pulse initial data are prescribed along the outgoing null cone
H−1 . Along H−1 , denoting χ to be its associated outgoing null second fundamental form and
setting χ̂ to be its traceless part, in 2008 Christodoulou proved the following2:
2In [21] Christodoulou’s short-pulse initial data are prescribed at past null infinity. Here we focus on its
correspondence in a finite region.
6 XINLIANG AN
Hδ
∗
u
H
(u
=
δ)
H0
1)
(u
−
=
=
0)
1(
u
−
H
Figure 3
Theorem (Christodoulou [21]). Set H 0 to coincide with a backward light cone in the Minkowskian
spacetime for −1 ≤ u ≤ 0. For every B > 0 and −1 < u∗ < 0, there exists a δ = δ(B, u∗ ) > 0
sufficiently small such that the following holds: for 0 ≤ u ≤ δ, if the prescribed χ̂ on H−1
satisfies X 1
δ 2 +j k(∂u )j ∇i χ̂kL∞ 2
u L (S−1,u )
≤B (1.7)
i≤5, j≤3
with ∇ being the angular derivative on S−1,u , then the solution to Einstein vacuum equations
remains regular in the spacetime region −1 ≤ u ≤ u∗ , 0 ≤ u ≤ δ. Furthermore, if the below
lower-bound assumption also satisfies
Z δ
inf2 |χ̂|2 (−1, u′ , ω) du′ ≥ M∗ > 2|u∗ |, (1.8)
ω∈S 0
then the 2-sphere Su∗ ,δ is a trapped surface.

Using similar upper and lower bounds as in [21], extensions of Theorem 1.2 via introducing the
signatures for short pulse and for decay rates were obtained in [29] and [1], respectively. The
trapped surfaces formed are of radius close to 1.
An important result was later obtained in [28]. Via deforming the double null foliation and
solving a quasilinear elliptic inequality, the lower bound assumption in [21] is vastly relaxed to
allow fully anisotropic initial data.
Theorem (Klainerman-Luk-Rodnianski [28]). Suppose that the initial data for Einstein vacuum
equations satisfy the condition (1.7) in Theorem 1.2 and the lower bound assumption
Z δ
sup |χ̂|2 (−1, u′ , ω) du′ ≥ M∗ > 0.
ω∈S2 0
Then, for sufficiently small δ, a trapped surface forms in the evolution.

We also remark that, with a more geometric approach, Le in [30] gave a different proof of the
main theorem in [28].
1 1
Via introducing two more large parameters a, b satisfying 1 ≪ b ≤ a 2 ≤ δ − 2 and designing
a new scale-critical short-pulse ansatz, a work [2] relaxed the upper bound assumption in [21].
Theorem (An-Luk [2]). Prescribe Minkowskian data along H 0 with −1 ≤ u ≤ 0. For a fixed
δ, for every B, there exist a0 = a0 (B) and b0 = b0 (B) sufficiently large such that the following
1 1
hold: for any a and b verifying a0 ≤ a ≤ δ −1 and b0 ≤ b ≤ a 2 ≤ δ − 2 , if prescribed χ̂ along H−1
with 0 ≤ u ≤ δ satisfies
X 1
δ j a− 2 k∇je4 ∇i χ̂kL∞ 2
u L (S−1,u )
≤ B, (1.9)
i≤5,j≤3
then the Einstein vacuum equations admit a unique regular solution in the spacetime region
1
−1 ≤ u ≤ −bδa 2 and 0 ≤ u ≤ δ. Moreover, if we further require
Z δ
inf2 |χ̂|2 (−1, u′ , ω)du′ ≥ 4δa,
ω∈S 0
then S−δa,δ is the formed 2-dimensional trapped surface.
Later, via employing only the signature for decay rates and a spacetime rescaling, a reproof
1
and an extension of Theorem 1.2 to the past null infinity was obtained in [4] for the case b = a 2
by the author.
Remark 8. The scale-critical bounds in [2] inspire the hyperbolic estimates of this article.
However, there are at least three major differences:
(1) In [2] the formed trapped surface is S−δa,δ and it is in the approximately self-similar
regime. The designed short-pulse hierarchy and the hyperbolic estimates are all con-
sistent with the considerations of scaling criticality. This is the underlying reason that
in [4] the author can give an alternative proof in the far field regime and very few bor-
derline terms appear. In this article, the trapped surface is located around S−(δa)1−µ∗ ,δ
with 0 < µ∗ < 1 and it is not in the approximately self-similar regime. Hence, designing
a suitable hierarchy and closing the bootstrap arguments are more delicate. We need
to divide the spacetime into two regions and deal with many new borderline terms.
Moreover, finding a proper order to conduct the estimates for recovering the bootstrap
constants is crucial. The newly designed hierarchy is given in Section 2.4 and the order
of estimates is explained in Section 2.6.
(2) To carry out the hyperbolic estimates in [2] and [4], in the spacetime region considered,
we apply the below inequality ubiquitously
1
ua 2 1
≤ 1 ≪ 1.
|u| a2
The above inequality hinges on the largeness of the additional parameter a. While in
this article our conclusions do not depend on the largeness of a and we can set a = 1,
which advances the short-pulse method.
(3) In this paper, we can even set the short-pulse to be small in scale-critical norms. To
fulfill the requirement (13.15) in Theorem 13.3, we can choose
1 1
|χ̂(−1, u, ω)| + |∇4 φ(−1, u, ω) − ∇4 φ(−1, 0, ω)| ≈ [ln(ln )]− 4 .
u
1
Note that the function [ln(ln 1/u)]− 4 → 0 as u → 0+ . Hence, we can let |χ̂(−1, u, ω)| and
∇4 φ(−1, u, ω) − ∇4 φ(−1, 0, ω) to be absolutely continuous functions. Their BV norms
in u variable (and other scale-critical norms) can be set small. While in all previous
trapped surface formation works outside of spherical symmetry, the BV perturbation of
|χ̂(−1, u, ω)| in u variable requires to be large.
1.3. Emergence of Apparent Horizon. Besides trapped surface formation, the rise of the
apparent horizon in the gravitational collapse of Einstein’s field equations was proved in [3] by
the author and in [6] by the author and Han. In [3], the following result was proved
Theorem (An [3]). Under the same initial-data assumption as in Theorem 1.2, the Einstein
1
vacuum equations admit a unique regular solution in the region 0 < u ≤ δ and −1 ≤ u ≤ −bua 2 .
In addition, assuming
Z u
|χ̂|2 (−1, u′ , ω)du′ = f (ω, u)ua for each 0 < u ≤ δ, (1.10)
0
with f (ω, u) being a smooth function in S2 × (0, δ] satisfying 20/21 ≤ f (ω, u) ≤ 22/21. Then
there exists a unique MOTS Mu along each H u (0 < u ≤ δ). Furthermore, requiring
1
a− 2 kuj ∇je4 ∇i χ̂kL∞ 2
u L (S−1,u )
≤ B for and i, j ∈ Z+ ,
then the MOTSs {Mu } together form a smooth apparent horizon.
8 XINLIANG AN
,δ
ũ
δa
,
ũa
−
−
S
S
O
Hδ
(u
H ũ
=
δ)
(u
=
H0
ũ)
1)
(u
−
=
=
0)
1(
u
−
H
Figure 4
The next work by An-Han generalized the above result to the fully anisotropic scenario.
Theorem (An-Han [6]). With the same prescribed initial data as in Theorem 1.2, for each
1
0 < u ≤ δ, if we require a− 2 kuj ∇je4 ∇i χ̂kL∞ 2
u L (S−1,u )
≤ B for any i, j ∈ Z+ and
Z u
|χ̂|2 (−1, u′ , ω)du′ = f (ω, u)ua, (1.11)
0
with f (ω, u) satisfying
0≤f ≤1 on S2 × (0, δ] and f (·, u) ≥ m on Bp (ǫ)
2
for a point p ∈ S and constants m ∈ (0, 1) and ǫ ∈ (0, π/2). Then in a large region of each
H u with 0 < u ≤ δ, there exists a unique MOTS Mu and the collection of MOTS {Mu } emerge
from the central singularity and form a smooth apparent horizon. Furthermore, if we require
1
|u∂u f (ω, u)| ≤ a− 3 , then the apparent horizon is a spacelike hypersurface, hence a dynamical
horizon.
In [6], this theorem was further extended to the multi-valley scenario and concrete examples
of dynamically formed apparent horizon with single and multi valleys are provided. In Figure
5, one can see the shape of MOTS along H u with two valleys, which are bounded by the upper
and lower barriers.
Hu
B
C
upper barrier
A
MOTS Mu (ω)
lower barrier
p2 p1
S2
Figure 5
Remark 9. The proof in [6] takes two steps:

- deriving apriori estimates via working through details of Moser’s iteration and Schauder’s
estimates in the order of
1 1
C 0 → C 0,α → C 1, 3 → C 2, 3 with 0 < α ≪ 1;
- conducting the method of continuity based on the explicit structure of the linearized
operator.
Compared with [6], in this article the geometry along the initial incoming cone is completely
different and the main terms in the elliptic equation for R(u, ω) (with u = R(u, ω) being the
MOTS along H u ) differentiate, which raises extra challenges. In [6] we employ the ansatz
e e In this
R = uae−φ to translate the elliptic equation for R(u, ω) into an elliptic equation for φ.
article, we design and utilize two forms of ansatz:
∗ ∗ e
(1) R = (ua)1−µ e−(1−µ )φ . We apply this form of R(u, ω) to perform the apriori estimates
via Moser’s iteration and Schauder’s estimates.
1−µ∗
1−µ∗
(2) R = (ua) φ̌ µ∗
. We employ this ansatz to eliminate the additional nonlinear bor-
derline term (rising from naked-singularity initial data) in the elliptic equation of R(u, ω)
and to conduct the method of continuity for the corresponding new linearized equation.
1.4. Difficulties and New Ingredients in the Proof. To prove Theorem 1.1, we divide the
considered spacetime region into two parts:
1
- the exterior {(u, u, ω)| − 1 ≤ u ≤ u1 and 0 ≤ ua 2 ≤ |u1 |Ω2−δ̃ (u1 , 0)},
1
- the interior {(u, u, ω)| u1 ≤ u ≤ 0 and 0 ≤ ua 2 ≤ |u|Ω2−δ̃ (u, 0)}.
Here |u1 | is a small constant to be fixed later. For notational simplicity, in the rest of this article,
we set constant δ̃ to obey 0 < δ̃ ≪ 1. In reality, if needed we can choose δ̃ to be any number
satisfying 0 < δ̃ < 1/2 and the below proof still works.
1.4.1. Hyperbolic Part. Compared with [2] by An-Luk, here more challenges arise. We list some
of them and state several strategies:
- For the position of MOTS along each H u , in contrast to the results in [2, 4, 6], where
∗
the MOTS lies around u = −ua, in this paper the MOTS is around u = −(ua)1−µ
with 0 < µ∗ < 1. The hyperbolic estimates in [2,4,6] are consistent with asymptotically
self-similar bounds and crucially rely on
1
ua 2 1
. 1 ≪1
|u| a2
with a being a large parameter. While here the MOTS is not located in the asymp-
totically self-similar regime, and to obtain a desired instability result we would allow
general (tiny) anisotropic perturbations and would set the parameter a to be of size 1.
These bring many new difficulties, especially on how to design and prove the short-pulse
hierarchy when |u| is small.
- In this paper, we generalize our constructed ansatz in [2,4] by inserting extra |u| weights
to the hierarchy. Employing the fact
1 µ∗
Ω(u, 0) ≤ Ω(u, u, θ1 , θ2 ) ≤ 2Ω(u, 0) and Ω2 (u, 0) = |u| 1−µ∗ ,
2
the suitable |u| weights are added via imposing the corresponding Ω weights. Espe-
cially, the Ω-weight top-order energy estimates and top-order elliptic estimates are
critical for our proof. In the energy estimates, because of the additional Ω weights,
basic integration-by-part and pairing arguments are revised. In comparison to [2],
the top-order elliptic estimatesPare more subtle. As shown in Section 4 and Sec-
tion 7, enhanced estimates for 1≤i≤4 kui ∇i (Ωtrχ)kL2 (Su,u ) , ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u ) ,
precise estimates involving borderline terms for ku5 ∇5 ηkL2u L2 (Su,u ) , ku6 ∇5 ηkL2u L2 (Su,u ) ,
10 XINLIANG AN
ku5 ∇5 (trχ, χ̂)kL2u L2 (Su,u ) are derived and are employed in an essential way. Subtle struc-
tures of the Einstein-scalar field system are explored.
- When bounding the nonlinear terms in the interior region where u1 ≤ u ≤ 0, we utilize
the inequality
1
ua 2 1 µ∗ 1 µ∗ 1
3 ≤ Ω 2 −δ̃ (u, 0) = |u| 1−µ∗ · 2 −δ̃ ≤ |u1 | 1−µ∗ · 2 −δ̃ ≪ 1, (1.12)
|u|Ω 2
where 0 < δ̃ ≪ 1 and |u1 | ≪ 1. Notice that the largeness of a is no longer required, but
the power of Ω has to be discreetly tracked.
- To conduct energy estimates for the scalar field, to derive energy estimates for curvature
components and to obtain elliptic estimates, we encounter terms, which only obey the
upper bound
1 1 1 1
u2 a4 ua 2 2
1 = .
|u| 2 Ω |u|Ω2
By contract to (1.12), this quotient is no longer . 1. The terms satisfying only this
bound serve as the new borderline terms to be estimated. We then add the weights
1 1
1 + u 1a
2 4
to the bootstrap assumptions. Finding the correct order to retrieve the
|u| 2 Ω
constants (which do not depend on other bootstrap assumptions) in these borderline es-
1 1
timates is decisive. Additional 1+ u 1a factors are incorporated in the top-order boot-

2 4
|u| 2 Ω
strap assumptions for Õ5,2 (η), Õ5,2 (Ωtrχ, trχ), Õ5,2 (Ωχ̂, χ̂), S(∇A φ), S(∇3 φ), R(K, σ̌)
and R(β) listed in Section 2.4. To improve the bootstrap assumption for each of these
terms, the correctly designed order of arguments and enhanced estimates are vitally
employed.
- Compared with An-Luk [2], there are extra borderline terms arising from Ωω and ∂u φ.
Along u = 0, they satisfy
r
µ∗ 1 √ µ∗ 1
Ωω(u, 0) = · and ∂u φ(u, 0) = 2 .
4(1 − µ∗ ) |u| 1 − µ∗ |u|
They behave similarly to Ωtrχ and cause multiple new borderline terms.
- To carry out energy estimates, besides σ̌ defined in [2], we also use renormalized quan-
tities βA − 21 ∇4 φ∇A φ (equivalent to βA − 12 R4A ) and β A + 21 ∇3 φ∇A φ (equivalent to
β A + 12 R3A ) to combine contributions of ∇3 βA , ∇3 R4A or ∇4 β A , ∇4 R3A in null Bianchi
equations (2.13) together.3 Furthermore, we also introduce and employ the renormal-
ized Gaussian curvature K − |u|1 2 − 41 ∇A φ∇A φ. This new renormalization comes from
the below identity for the Weyl curvature
1 1 1
∇A WA434 = − ∇4 R43 + ∇3 R44 − ∇A R(gA3 g44 − gA4 g43 ).
2 2 12
It gives
1 1 1 1
∇4 (ρ − R43 − R) = ∇A WA434 − ∇3 R44 + non-top-order terms
4 12 2 4
with ρ = W3434 /4. Further calculations via applying the constraint equation imply that
1 1 1 1 1
ρ − R43 − R = −(K − ∇A φ∇A φ) + χ̂ · χ̂ − trχtrχ.
4 12 4 2 4
For using vanishing initial data along H 0 , we then utilize the renormalized Gaussian
curvature K − |u|1 2 − 41 ∇A φ∇A φ to carry out the energy estimates. In our estimates,
3A similar idea was used in [5] by An-Athanasiou for the Einstein-Maxwell system.
some hidden structures of the Einstein-scalar field system are revealed.
1.4.2. Elliptic Part. Because of the naked-singularity initial data, here the key structure of the
main elliptic equation for MOTS is quite different from its counterpart in [6]. In this article, we
design and utilize two forms of ansatz:
∗ ∗ e e
- R = (ua)1−µ e−(1−µ )φ . This is a generalization of [6], where R = uae−φ . We use this
form to construct the anisotropic trapped surfaces and to conduct Moser’s iteration and
Schauder’s estimates as in [6].
∗ 1−µ∗
- R = (ua)1−µ φ̌ µ∗ . The ansatz here is new and is indispensable for this paper. We
employ this form to carry out the key continuity method for the new linearized equation
to solve for the MOTS.
In below, we will give a more comprehensive explanation on the utilization of these two forms.
To construct the trapped and untrapped barriers and to derive the regularity estimates, we
∗ ∗ e
employ the ansatz R = (ua)1−µ e−(1−µ )φ . We observe that
- Owning to Christodoulou’s initial data, the constructions of the multi- and single-valley
anisotropic trapped surfaces as in [6] and [28] still hold. One can see the details in
Section 9.
- For the apriori regularity estimates, we find that the approach developed in [6] by An-
Han is robust enough and can be adapted to this new setting. Consider the main elliptic
equation (10.1):
1 e e2
0 =∆γ φe + 1 − f (u, ω)eφ − 4(1 − µ∗ )RΩω|∇γ φ|
2
R e 2 + 2γ ij ηi ∂j φe
− (1 + Ωtrg χ)(1 − µ∗ )|∇γ φ|
2
1 1 e
+[ ∗
RΩ−1 trg χ − 1 + f (u, ω)eφ ]
2(1 − µ ) 2
∗ e e
− 2(1 − µ )RΩχ̂ · ∇γ φ · ∇γ φ.
e
On the right side, besides the main terms ∆γ φe + 1 − 21 f (u, ω)eφ we denote
F1 := − 4(1 − µ∗ )RΩω|∇γ φ|e 2,

F2 := − (1 + Ωtrg χ)(1 − µ∗ )|∇γ φ|
2
1 1 e
+[ RΩ−1 trg χ − 1 + f (u, ω)eφ ]
2(1 − µ∗ ) 2
e
− 2(1 − µ∗ )RΩχ̂ · ∇γ φe · ∇γ φ.
In [6] we deal with the elliptic equation in the same form as above. But in [6] we
1 1 1
get |RΩω| ≤ |u|ua 2 /|u|2 = ua 2 /|u| ≤ 1/a 2 ≪ 1. Hence the F1 and F2 terms obey
|F1 | + |F2 | ≪ |Dφ̃|2 + 1. While in this paper, we only have |RΩω| . |u| · 1/|u| = 1 and it
only holds |F1 | + |F2 | ≤ |Dφ̃|2 + 1. This −4(1 − µ∗ )RΩω|∇γ φ|e 2 later serves as the main
term in the linearization arguments. Luckily, as pointed out one line below (4.3) in [6],
the approach developed in [6] via Moser’s iteration is robust enough to cover the case
e 2 + 1)
|F1 | + |F2 | ≤ c(|Dφ| with c = 1.
Hence it can still be applied here to obtain the apriori regularity estimates.
∗ 1−µ∗
The new ansatz R = (ua)1−µ φ̌ µ∗ plays a vital role in linearization arguments.
12 XINLIANG AN
∗ ∗ e
- If we proceed with the ansatz R = (ua)1−µ e−(1−µ )φ . The main elliptic equation for
MOTS is then reduced to
1 e
0 = ∆γ φe + 1 − f (ω, u)eφ − 4(1 − µ∗ )RΩω|∇γ φ|
e 2 + small terms.
2
In [6], since RΩω is small, the main terms in above equation are the first three terms
on the right and the continuity argument via linearization can be carried out accord-
e 2 term). While in this paper, due to the naked-
ingly (without troubles from the |∇γ φ|
singularity initial data, we have
µ∗ 1
−4(1 − µ∗ )RΩω(u, u, ω) ≈ −4(1 − µ∗ )R · ∗
· = −4µ∗
4(1 − µ ) |u|
being of size 1. It is no longer negligible. Here, by introducing and employing the new
∗ 1−µ∗
ansatz R = (ua)1−µ φ̌ µ∗ , we eliminate this quadratic gradient borderline term and
the equation of MOTS becomes
µ∗ 1
0 = ∆γ φ̌ − µ∗ φ̌ + f (ω, u)φ̌1− µ∗ + small terms.
2
This enables us to complete the method of continuity and to construct the unique MOTS
along each H u in the considered spacetime region.
We then consider the position of the MOTS.

- By the elliptic arguments in Section 9, along each H u we have that the MOTS is
∗
bounded in between by an untrapped surface located at u = −(2ua/3)1−µ and by a
trapped surface lying within
∗ ∗
−(ua · mǫ2 )1−µ ≤ u ≤ −(ua · mǫτ +2 )1−µ
∗
with τ > 2.2. Note that, along the inner hypersurface u = −(ua · mǫτ +2 )1−µ , it holds
2 1
Ω(u, 0) µ∗ = |u|Ω2 (u, 0) = |u| 1−µ∗ = ua · mǫτ +2 . (1.13)
To guarantee that this hypersurface is located within the hyperbolic existence region,
we require u ≤ |u|Ω2−δ̃ (u, 0), which is equivalent to |u|Ω2 (u, 0) ≥ uΩδ̃ (u, 0). Thus, we
need
ua · mǫτ +2 = |u|Ω2 (u, 0) ≥ uΩδ̃ (u, 0). (1.14)
The above inequality is equivalent to mǫτ +2 ≥ a−1 Ωδ̃ (u, 0). Back to (1.14), together
with (1.13) we need
µ∗ δ̃
mǫτ +2 ≥ a−1 (ua · mǫτ +2 ) 2 . (1.15)
Hence, if we let u satisfy (1.15), then the apparent horizon constructed in this article
stays in the obtained hyperbolic existence region. Moreover, to fix the small parameter
|u1 |, we only need to require
∗ µ∗
|2ua/3|1−µ ≤ |u1 | and Ω(u1 , 0) = |u1 | 1−µ∗ ≪ 1. (1.16)
By setting u to be sufficiently small, clearly, we can choose |u1 | verifying (1.16). In
addition, if we set a = 1 as in the statement of Theorem 1.1, we would deduce all the
elliptic arguments in the interior region.
1.4.3. Instability Results. Our proven anisotropic trapped surface and apparent horizon forma-
tion criteria enable us to obtain instability theorems of high co-dimensions.
• Allowing perturbations with finite BV norms along u = −1 with u > 0, in Theorem
13.2 we consider the space of initial data for χ̂(−1, u, ω) and ∂u φ(−1, u, ω), that can
be prescribed along u = −1 with u > 0. Given Christodoulou’s naked-singularity data
in [18] along u = 0, then for any k ∈ Z+ , we can find smooth functions f1 , f2 , ..., fk and
g1 , g2 , ..., gk with variables u > 0 and ω, such that the below prescribed initial data for
|χ̂| and ∂u φ with u > 0

|χ̂|(−1, u, ω), ∂u φ(−1, u, ω)

′ ′ ′
= 0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ1 g1 + λ2 g2 + · · · + λk gk
would lead to the trapped surface and the apparent horizon formation, when
k
X k
X 2
λ2i + λ′i 6= 0 with λk , λ′k ∈ R.
1 1

0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ′1 g1 + λ′2 g2 +···+ λ′k gk

= 0 + λ̃1 f1 + λ̃2 f2 + · · · + λ̃k fk , ∂u φ(−1, 0, ω) + λ̃′1 g1 + λ̃′2 g2 +···+ λ̃′k gk ,
then it renders
We may therefore say that, for any k ∈ Z+ , Christodoulou’s naked-singularity solution
in [18] has at least co-dimensional 2k nonlinear instability subject to outgoing BV char-
acteristic perturbations at (−1, 0, ω).
• In Theorem 13.2, with mi and m′i being positive constants, the constructed fi and gi
obey the requirement: for u > 0
1 1
fi (ω, u) ≥ mi a 2 on Bpi (ǫi ) ⊂ Di , gi (ω, u) ≥ m′i a 2 on Bp′i (ǫ′i ) ⊂ Di′ ,
and
fi (ω, u) = 0, gi (ω, u) = 0 for u ≤ 0.
Hence, functions fi (ω, u) and gi (ω, u) are not continuous in u, when u → 0+ . In Section
13.2, by exploiting the extra Ω-weight gained in the hyperbolic estimates, we extend
this instability theorem by allowing to choose continuous functions fi (ω, u) and gi (ω, u)
for the u variable with u ≥ 0. In Theorem 1.3, we impose the below requirements for fi
and gi :
1 1 1
With 1 ≤ i ≤ k, C1 > 2, Λ > 0 and g(u) = (C1 Λ + C1 ) 2 · [ln(ln )]− 2 ,
u
we request the prescribed fi and gi to obey
1 1 1 1
0 ≤ fi ≤ g(u) 2 a 2 on S2 × (0, δ], fi (·, u) ≥ mi g(u) 2 a 2 on Bpi (ǫi ) ⊂ Di ,
1 1 1 1
0 ≤ gi ≤ g(u) 2 a 2 on S2 × (0, δ], gi (·, u) ≥ m′i g(u) 2 a 2 on Bp′i (ǫ′i ) ⊂ Di′
with points pi ∈ Di ⊂ S2 , p′i ∈ Di′ ⊂ S2 and constants mi , m′i ∈ (0, 1), ǫi , ǫ′i ∈ (0, π/2).
Note that we can set a = 1. And as u → 0+ , it holds
1
lim g(u) 2 = 0.
u→0+
Therefore, we can choose continuous-in-u and smooth-in-ω functions fi , gi , χ̂, ∂u φ along

u = −1 from u = 0 to prove the corresponding high co-dimensional instability results
stated in Theorem 1.3.
1.5. Related Results.

14 XINLIANG AN
1.5.1. Trapped Surfaces, MOTS and Short-Pulse Method. After Christodoulou’s monumental
work [21], a systematical approach by introducing signature for the short pulse was developed
by Klainerman-Rodnianski in [29] to simplify and to extend [21] in a finite region. Yu in [38]
extended Klainerman-Rodnianski’s approach to the Einstein-Maxwell system. The author in [1]
introduced a new signature, called the signature for decay rates, and extended Klainerman-
Rodnianski’s results to past null infinity. Based on the hyperbolic estimates in [29], Klainerman-
Luk-Rodnianski in [28] found a fully anisotropic mechanism for trapped surface formation.
In [2] the author and Luk designed a new set of short-pulse hierarchies and proved the first
scale-critical trapped surface formation criterion for the Einstein vacuum equations. With only
the signature for decay rates, in [4] the author gave a short proof of the scale-critical trapped
surface formation in the far-field regime and related the short-pulse hierarchies to the peeling
properties of the gravitational waves. In [5], the author and Athanasiou further extended the
new approach in [4] to the Einstein-Maxwell system. With spherically symmetric singular initial
data prescribed along the incoming null hypersurface, Li and Liu in [32] studied the Einstein-
scalar field system and proved an almost scale-critical trapped surface formation criterion. Their
result showed that the spherically symmetric singularities considered in [32] are unstable (as a
result of trapped surface formation) subject to close-to-isotropic gravitational perturbations
and the instability in [32] is in the sense of the first category, slightly weaker than co-dimension
1. Very recently, the methods in [4] by the author and in [5] by the author and Athanasiou
are also extended to the Einstein-Yang-Mills system and the Einstein-scalar field system, by
Athanasiou-Mondal-Yau [15] and Zhao-Hilditch-Kroon [39], respectively.
Based on the hyperbolic estimates in [2], in the short-pulse regime, the MOTS along each
incoming null hypersurface and the corresponding apparent horizon are constructed in [3] by the
author and in [6] by the author and Han for the general anisotropic scenarios. For solving the
MOTS and the apparent horizon based on spacelike foliations, interested readers are referred
to [11–14] by Andersson, Eichmair, Mars, Metzger, Simon and references therein.
1.5.2. Weak Cosmic Censorship within Spherical Symmetry and Naked Singularity Formation.
In a series of works [16]- [19], Christodoulou achieved the proof of weak cosmic censorship for the
spherically symmetric Einstein-(real) scalar field system. After constructing the first example of
the naked-singularity solution under spherical symmetry, he ingenuously obtained 4-integrated
theorems:
(1) BV extension principle in triangle and in rectangle regions;
(2) a sharp trapped surface formation criterion;
(3) the first instability theorem;
(4) the second instability theorem.
Christodoulou’s arguments in [16]- [19] rely on the special structures of the spherically symmet-
ric Einstein-(real) scalar field system and the associated monotonicity properties. His 4-step
argument is unified. Each step is exactly matched with the other steps. In recent years, each of
Christodoulou’s steps has been separately extended. In [8] the author and Lim extended [16] to
the spherically symmetric Einstein-Maxwell-charged (complex) scalar field system. See also the
extension in [23] by Costa to the spherically symmetric Einstein-(real) scalar field system with
positive cosmological constant. For the instability argument of the Einstein-(real) scalar field
system, Li-Liu in [31] gave a proof with apriori estimates. Their instability theorem is slightly
different from Christodoulou’s.
Extending the proof of the weak cosmic censorship to the other spherically symmetric Einstein-
matter field is still very challenging. The Einstein-Maxwell-charged (complex) scalar field system
is the next to be considered. However, the corresponding weak cosmic censorship has remained
open since the late 1990s. Several difficulties are easily seen:
• Due to the presence of non-constant charge Q, almost all previously employed mono-
tonicity formulas fail to hold. These monotonicity formulas are critically used in Christodoulou’s
BV extension/trapped surface formation/instability arguments when Q = 0. A new
strategy to incorporate non-constant charge Q remains to be developed.
• For the trapped surface formation criterion, with η being the initial mass input and
δ being the small deformation parameter, Christodoulou needs η & δ log δ1 . Following
Christodoulou’s method, for the charged case, in [8] the author and Lim obtained a
ω
result requiring η & δ 1− 2 (> δ log δ1 ) with 0 < ω ≪ 1. But Christodoulou’s “blue-shift”
ω
γ in the instability argument is around log δ1 rather than δ − 2 . How to reconcile?
• Christodoulou’s way of proving his instability theorems is very delicate. Can one obtain
the same sharp arguments as Christodoulou did?
In a new preprint [9], the author and Tan address these questions and we show that the weak
cosmic censorship still holds (in the sense of Christodoulou) for the gravitational collapse of the
spherically symmetric Einstein-Maxwell-charged (complex) scalar field system.
Recently, there is also the exciting development of constructing the naked-singularity solutions
for Einstein’s field equations. Under spherical symmetry and self-similar ansatz, in [10] the
author and Zhang showed naked-singularity formation for Einstein vacuum equations in higher
dimensions, and in [25] Guo-Hadzic-Jang extended [18] to the Einstein-Euler system. Outside
spherical symmetry, important progress has been made by Rodnianski-Shlapentokh-Rothman in
[35] and by Shlapentokh-Rothman in [36]. There they prove the first naked-singularity formation
result for the Einstein vacuum equations in 3 + 1 dimensions.
1.6. Organizations of the Paper. The paper is organized as follows. In Section 2 we present
the settings. In Section 3, we derive the preliminary estimates. In Section 4, we provide the
L2 (Su,u ) estimates for Ricci coefficients. In Section 5, we derive the L2 (Su,u ) estimates for the
scalar field. In Section 6, we prove energy estimates for the scalar field. In Section 7, we establish
the elliptic estimates. In Section 8, we conduct the energy estimates for curvature components.
In Section 9, we construct the sub and super barriers for the MOTS. In Section 10, we carry
out Moser’s iteration and the Schauder estimates. In Section 11, we demonstrate the crucial
linearization and continuity arguments. In Section 12, we obtain the existence and properties
for the apparent horizon. In Section 13, we prove the instability theorems.
1.7. Acknowledgement. XA would like to thank Yu Deng and Willie Wong for valuable cor-
respondences. XA is supported by MOE Tier 1 grants A-0004287-00-00, A-0008492-00-00 and
MOE Tier 2 grant A-8000977-00-00.
2. Setting
2.1. Double Null Coordinates. In this paper, we adopt the double null coordinates (u, u, θ1 , θ2 )
for (M, g). For A, B = 1, 2, the spacetime metric g takes the form
g = −2Ω2 (du ⊗ du + du ⊗ du) + g/AB (dθA − dA du) ⊗ (dθB − dB du). (2.1)

Here u and u are optical functions satisfying
g µν ∂µ u∂ν u = 0, g µν ∂µ u∂ν u = 0.
We require both u and u are increasing towards the future. The level sets of u and u are called
Hu and H u , respectively. They are outgoing and incoming null cones. The intersections of
Hu and H u are topologically 2-shpere and are denoted as Su,u . In later sections, along H u
we consider the 2-sphere u = −R(u, θ1 , θ2 ) with R(u, θ1 , θ2 ) being a given function, and for
simplicity we will write this 2-sphere as S−R,u .
2.2. Equations. The equations studied here are the Einstein-scalar field system:
1
Rµν − Rgµν = Tµν ,
2
1
Tµν = Dµ φDν φ − gµν Dλ φDλ φ.
2
16 XINLIANG AN
Utilizing the null frame e3 , e4 and an frame e1 , e2 tangent to the 2-spheres Su,u , with the indices
A, B ∈ {1, 2}, we define the Ricci coefficients:
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
1 1 (2.2)
ω̂ = − g(D4 e3 , e4 ), ω = − g(D3 e4 , e3 ),
4 4
1
ζA = g(DA e4 , e3 ),
2
where DA = DeA . Note that here we use ω̂ to represent the Ricci coefficient and this is because
we save ω to stand for the angular variable on S2 . We also denote ∇ to be the induced covariant
derivative operator on Su,u and let ∇3 , ∇4 be the projections to Su,u of the covariant derivatives
D3 , D4 . By above definition, it also holds
1 1
ω̂ = − ∇4 (log Ω), ω = − ∇3 (log Ω),
2 2 (2.3)
ηA = ζA + ∇A (log Ω), η A = −ζA + ∇A (log Ω).
In 3 + 1 dimensions, for i, k, l, m = 1, 2, 3, 4 we further decompose the below Weyl tensor
1 1
Wiklm := Riklm + (Rim gkl − Ril gkm + Rkl gim − Rkm gil ) + R(gil gkm − gim gkl )
2 6
and define the following null curvature components
αAB = W (eA , e4 , eB , e4 ), αAB = W (eA , e3 , eB , e3 ),

1 1
βA = W (eA , e4 , e3 , e4 ), β A = W (eA , e3 , e3 , e4 ), (2.4)
2 2
1 1
ρ = W (e4 , e3 , e4 , e3 ), σ = ∗ W (e4 , e3 , e4 , e3 ),
4 4
where ∗ W is the Hodge dual of W .
We further set φ(1) · φ(2) to stand for a contraction of the tensor product of φ(1) and φ(2) with
respect to the metric g/ on Su,u . We also denote
(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) := /ǫ AB (g/−1 )CD φAC φBD for symmetric two tensors φAB , φAB ,
where /ǫ is the volume form associated to the metric g/. For totally symmetric tensors, we also
employ the below div and curl operators
(div φ)A1 ...Ar := ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar := /ǫ BC ∇B φCA1 ...Ar .
In addition, we define the trace to be
(trφ)A1 ...Ar−1 := (g/−1 )BC φBCA1 ...Ar−1 .
Let χ̂ and χ̂ be the traceless parts of χ and χ respectively. Employing above notations,
rewriting Rµν − 12 Rgµν = Tµν according to the null frame {e3 , e4 , eA , eB } with A, B = 1, 2, we
have that χ and χ obey the following null structure equations:

1
∇4 trχ + (trχ)2 = −|χ̂|2 − 2ω̂trχ − D4 φD4 φ,
2
∇4 χ̂AB + trχχ̂AB = −2ω̂χ̂AB − αAB ,
1
∇3 trχ + (trχ)2 = −2ωtrχ − |χ̂|2 − D3 φD3 φ,
2
∇3 χ̂AB + trχ χ̂AB = −2ωχ̂AB − αAB ,
1 1 1
∇4 trχ + trχtrχ = 2ω̂trχ + 2ρ − χ̂ · χ̂ + 2div η + 2|η|2 − D3 φD4 φ + DA φDA φ,
2 3 3
1 1 1
∇4 χ̂AB + trχχ̂AB = (∇⊗η)AB + 2ω̂χ̂AB − trχχ̂AB + (η ⊗η)AB + DA φDB φ − gAB DC φDC φ,
b b
2 2 2
1 1 1
∇3 trχ + trχtrχ = 2ωtrχ + 2ρ − χ̂ · χ̂ + 2div η + 2|η| − D3 φD4 φ + DA φDA φ,
2
2 3 3
1 1 1
∇3 χ̂AB + trχχ̂AB = (∇⊗η)AB + 2ωχ̂AB − trχχ̂AB + (η ⊗η)AB + DA φDB φ − gAB DC φDC φ.
b b
2 2 2
(2.5)
And the Ricci coefficients ηA , η A , ω, ω̂ obey the remaining null structure equations:
1
∇4 ηA = −χAB · (η − η)B − βA − DA φD4 φ,
2
1
∇3 η A = −χAB · (η − η)B + β A − DA φD3 φ,
2 (2.6)
1 1 1 1
∇4 ω = 2ω̂ω − ηA · η A + |η|2 + ρ + D3 φD4 φ + DA φDA φ,
2 2 6 12
1 1 1 1
∇3 ω̂ = 2ω̂ω − ηA · η A + |η|2 + ρ + D3 φD4 φ + DA φDA φ.
2 2 6 12
On Su,u , with the null frame, the Gauss-Codazzi equations boil down to the constraint equations:
1 1 1 1
div χ̂A = ∇A trχ − (ηB − η B ) · (χ̂BA − trχgBA ) − βA + DA φD4 φ,
2 2 2 2
1 1 1 1
div χ̂A = ∇A trχ + (ηB − η B ) · (χ̂BA − trχgBA ) + β A + DA φD3 φ,
2 2 2 2
1
curl η = − curl η = σ + χ̂ ∧ χ̂, (2.7)
2
1 1 1 1
K = − (ρ + R) − trχtrχ + χ̂ · χ̂ + g AB DA φDB φ
6 4 2 2
1 1 1 1
= − ρ − trχtrχ + χ̂ · χ̂ + D3 φD4 φ + DA φDA φ.
4 2 6 3
Here K is the Gauss curvature of the spheres Su,u .
For the system (1.1), the scalar field φ obeys ✷g φ = 0. And we rewrite the wave equation as
a transport equation
0 = ✷g φ =g λµ Dλ Dµ φ = g λµ Dλ (Dµ φ) − g λµ DDλ eµ φ
=g 34 D3 (D4 φ) − g 34 DD3 e4 φ + g 43 D4 (D3 φ) − g 43 DD4 e3 φ + g AB DA (DB φ) − g AB DDA eB φ
1 1
= − D3 D4 φ − D4 D3 φ + g AB DA (DB φ) − g AB D∇A eB + 12 χAB e3 + 21 χ e4 φ
2 2 AB
1 1
= − D3 D4 φ + ∆g φ − trχe3 φ − trχe4 φ.
2 2
It hence holds
1 1
D3 D4 φ + trχD3 φ + trχD4 φ = g AB ∇A ∇B φ. (2.8)
2 2
18 XINLIANG AN
For convenience, we also employ the following two equivalent forms (see Section 6)
1 1
(Ωe3 )(Ωe4 φ) + ΩtrχΩe4 φ = Ω2 ∆g φ − ΩtrχΩe3 φ+2Ω2 η A eA φ, (2.9)
2 2
1 1
(Ωe4 )(Ωe3 φ) + ΩtrχΩe4 φ = Ω2 ∆g φ − ΩtrχΩe3 φ+2Ω2 η A eA φ. (2.10)
2 2
Using the property of covariant derivatives, the scalar field φ also satisfies
∇Ωe3 ∇A φ =eA (Ωe3 φ) − ΩχA B eB φ,

(2.11)
∇Ωe4 ∇A φ =eA (Ωe4 φ) − ΩχA B eB φ.
Recall that in 3 + 1 dimensions for i, k, l, m = 1, 2, 3, 4, the Weyl curvatures obey
1 1 1
∇i Wiklm = − ∇m Rlk + ∇l Rmk − (∇l Rgmk − ∇m Rglk ). (2.12)
2 2 12
Employing the null frame {e3 , e4 , eA , eB } with A, B = 1, 2, we rewrite the above (2.12) as the
below null Bianchi equations:
1 1 1
∇3 α + trχα = ∇⊗β b + 4ωα − 3(χ̂ρ +∗ χ̂σ) + (ζ + 4η)⊗β
b + (D3 R44 − D4 R43 )gAB + ∇4 RgAB ,
2 4 6
1
∇4 β + 2trχβ = div α − 2ω̂β + η · α − (DA R44 − D4 R4A ),
2
1 1 1
∇3 β + trχβ = ∇ρ + 2ωβ +∗ ∇σ + 2χ̂ · β + 3(ηρ +∗ ησ) + ∇A (g CD RCD ) − ∇B RBA + ∇A R + ∇3 R4A ,
2 6 2
3 1 1
∇4 σ + trχσ = −div ∗ β + χ̂ ·∗ α − ζ ·∗ β − 2η ·∗ β − (Dµ R4ν − Dν R4µ )ǫµν 34 ,
2 2 4
3 1 1
∇3 σ + trχσ = −div ∗ β + χ̂ ·∗ α − ζ ·∗ β − 2η ·∗ β + (Dµ R3ν − Dν R3µ )ǫµν 34 ,
2 2 4
3 1 1 1
∇4 ρ + trχρ = div β − χ̂ · α + ζ · β + 2η · β − (D3 R44 − D4 R43 ) + ∇4 R,
2 2 4 12
3 1 1 1
∇3 ρ + trχρ = −div β − χ̂ · α + ζ · β − 2η · β + (D3 R34 − D4 R33 ) + ∇3 R,
2 2 4 12
1 1 1
∇4 β + trχβ = −∇ρ +∗ ∇σ + 2ω̂β + 2χ̂ · β − 3(ηρ −∗ ησ)− ∇A (g CD RCD ) + ∇B RBA − ∇A R − ∇4 R3A ,
2 6 2
1
∇3 β + 2trχ β = −div α − 2ωβ + η · α + (DA R33 − D3 R3A ),
2
1 1 1
∇4 α + trχα = −∇⊗β b + 4ω̂α − 3(χ̂ρ −∗ χ̂σ) + (ζ − 4η)⊗β
b + (D4 R33 − D3 R34 )gAB + ∇3 RgAB .
2 4 6
(2.13)
Here, ∗ denotes the Hodge dual on Su,u .

In our proof, we utilize the null Bianchi equations to deduce energy estimates. We will employ
the renormalized quantities4
1 1 1
βA − ∇4 φ∇A φ, K − 2 − ∇A φ∇A φ,
2 |u| 4
1 1
σ̌ = σ + χ̂ ∧ χ̂, β A + ∇3 φ∇A φ.
2 2
These renormalized quantities satisfy the below renormalized null Bianchi equations:
4For more general incoming initial data, the renormalized quantity for K is K − K(u, 0) − 1 ∇A φ∇ φ.
4 A
1
∇3 (β· − ∇4 φ∇· φ)A + trχβA = − ∇K +∗ ∇σ̌ + 2ωβ + 2χ̂ · β − 3(ηK − ∗ ησ̌)
2
1 3
+ (∇(χ̂ · χ̂) −∗ ∇(χ̂ ∧ χ̂)) − ηtrχtrχ
2 4
3 ∗ 1 (2.14)
+ (η χ̂ · χ̂ − η χ̂ ∧ χ̂) − (∇trχtrχ + trχ∇trχ)
2 4
+ ∇A (g CD RCD ) − ∇B RBA
1
+ ηA ∇3 φ∇4 φ + ηA ∇C φ∇C φ.
2
3 1 1 1
∇4 σ̌ + trχσ̌ = − div ∗ (β − ∇4 φ∇φ) − ζ ·∗ β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)
b + χ̂ ·∗ (η ⊗η)
b
2 2 2 2 (2.15)
1 1 C AB
− ∇1 R42 + ∇2 R41 + (∇A φ∇B φ − gAB ∇C φ∇ φ) ∧ χ̂ ,
2 2
1 1
∇4 (K − ∇A φ∇A φ) + trχ(K − ∇A φ∇A φ)
4 4
1 1
b + χ̂ · (η ⊗η)
= − div β − ζ · β − 2η · β + χ̂ · ∇⊗η b
2 2 (2.16)
1 1 1 1
− trχdiv η − trχ|η|2 + η A ∇A φ∇4 φ + ∆g φ∇4 φ
2 2 2 2
1 1
+ (∇A φ∇B φ − gAB ∇C φ∇C φ) · χ̂AB ,
2 2
3 1 1 1
∇3 σ̌ + trχσ̌ = − div ∗ (β + ∇3 φ∇φ) − ζ ·∗ β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)
b + χ̂ ·∗ (η ⊗η)
b
2 2 2 2 (2.17)
1 1
+ ∇1 R32 − ∇2 R31 + χ̂AB ∧ (∇A φ∇B φ − gAB ∇C φ∇C φ),
2 2
1 1 3 1 1
∇3 (K − − ∇A φ∇A φ) + trχ(K − 2 − ∇A φ∇A φ)
|u|2 4 2 |u| 4
1 1 A
=div (β + ∇3 φ∇φ) − ∇ ∇3 φ∇A φ − ζ · β + 2η · β
2 2
1 1 1 1
+ χ̂ · ∇⊗η b − trχ|η|2 + η A ∇A φ∇3 φ
b + χ̂ · (η ⊗η) (2.18)
2 2 2 2
1 1 C AB 1
+ (∇A φ∇B φ − gAB ∇C φ∇ φ) · χ̂ − trχ∇A φ∇A φ
2 2 8
1 1 1 A Ω−1 2
+ trχ(−div η + K − 2 − ∇ φ∇A φ) − (Ωtrχ + ),
2 |u| 4 |u|2 |u|
1 1 1
∇4 (β · + ∇3 φ∇· φ)A + trχβ A =∇(K − 2 − ∇A φ∇A φ) +∗ ∇σ̌ + 2ωβ + 2χ̂ · β + 3(−ηK +∗ ησ̌)
2 |u| 4
1 A 1 1 1
+ ∇(− ∇ φ∇A φ + trχtrχ − χ̂ · χ̂) − ∗ ∇(χ̂ ∧ χ̂)
4 4 2 2
+ ∇A (g CD RCD ) + ∇B RBA
1
− η A ∇3 φ∇4 φ − η A ∇C φ∇C φ.
2
(2.19)
Remark 10. In below, if no further clarification, we employ the capital Latin letters A ∈ {1, 2}
as frame indices on the spheres Su,u , and use Greek letters µ ∈ {1, 2, 3, 4} as frame indices in
the whole 3+1 dimensional spacetime.
20 XINLIANG AN
2.3. Naked Singularity Initial Data. In [18], Christodoulou constructed a spherically-symmetric

naked-singularity spacetime. It is portrayed in below picture.
0 O Anisotropic Apparent Horizon

=
u
u
O =
Naked Singularity δ
u
=
0
r=0
u
Anisotropic Perturbation
=
0
r=0
1
−
=
u
1
−
=
u
Christodoulou’s solution solves the Einstein-scalar field system 1.1. In [18] a spacetime sin-
gularity forms in gravitational collapse and the whole spacetime is free of trapped surfaces and
the apparent horizon.
In this article, we consider a characteristic initial value problem with data prescribed along
u = 0 with −1 ≤ u ≤ 0 and u = −1 with 0 ≤ u ≤ δ. For u ≤ 0, we adopt the spacetime
(Mint , gint ) constructed in [18] with Mint standing for the interior spacetime region u ≤ 0.
1
Setting v = 4u, we let gint = − Ω2 (u, v)(du ⊗ dv + dv ⊗ du) + r2 (u, v)[(dθ1 )2 + sin2 θ1 (dθ2 )2 ].
2
Here r(u, v) is the radius function.
We will add general (anisotropic) perturbation along u = −1 starting from u = 0 and for
u ≥ 0 the spacetime metric g is of the form (2.1). In particular, along u = 0, we prescribe
Christodoulou’s spherically-symmetric naked-singularity initial data: via Christodoulou’s con-
struction in [18], for Hawking mass m = 2r (1 + 4Ω−2 ∂u r∂v r), it holds that
2m
(u, 0) = µ∗ with µ∗ being a positive constant less than 1.
r
We also set
1 − µ∗
r(−1, 0) = 1, ∂v r(−1, 0) = and ∂u r(u, 0) = −1.
4
2m
Since along u = 0 it holds µ∗ = r = 1 + 4Ω−2 ∂u r∂v r, we get
1 − µ∗
Ω−2 ∂v r(u, 0) = along u = 0.
4
Employing the equation5
2m
∂u r r
∂u [ln(∂v r)] = · ,
r 1 − 2m
r
with r(−1, 0) = 1 and ∂v r(−1, 0) = (1 − µ∗ )/4, we obtain
1 − µ∗ 1−µ
µ∗ µ∗
∂v r(u, 0) = r ∗ (u, 0) and Ω2 (u, 0) = r 1−µ∗ (u, 0).
4
Using v = 4u, these further imply
µ∗
∂u r(u, 0) = (1 − µ∗ ) · r 1−µ∗ (u, 0) and Ω−2 ∂u r(u, 0) = 1 − µ∗ .
5It is equivalent to r∂ ∂ r = −∂ r∂ r − 1 Ω2 .
u v u v 4
Recall trχ(u, 0) = g AB g(DA e4 , eB )(u, 0), trχ(u, 0) = g AB g(DA e3 , eB )(u, 0) and along u = 0 it
holds e3 = Ω−1 ∂u , e4 = Ω−1 ∂u . We hence derive
2 −2 2(1 − µ∗ )
Ω−1 trχ(u, 0) = Ω−1 g AB · ΓD
A4 · gBD = Ω ∂u r(u, 0) = ,
r r(u, 0)
2 2
Ωtrχ(u, 0) = Ωg AB · ΓD
A3 · gBD = ∂u r(u, 0) = − .
r r(u, 0)
Utilizing ω = − 12 ∇3 log Ω, we also have
1 1 µ∗ µ∗ 1
Ωω(u, 0) = − ∂u log Ω(u, 0) = − ∂u log[r 2(1−µ∗ ) (u, 0)] = · . (2.20)
2 2 4(1 − µ∗ ) r(u, 0)
We proceed to calculate the initial value of K along u = 0. Applying the below null structure
equation and the Gauss equation
1 1 1
∇3 trχ + trχtrχ =2ωtrχ + 2ρ − χ̂ · χ̂ + 2div η + 2|η|2 − D3 φD4 φ + DA φDA φ,
2 3 3
1 1 1 1 A
K = − ρ − trχtrχ + χ̂ · χ̂ + D3 φD4 φ + DA φD φ,
4 2 6 3
1 −1
further employing ω = − 2 Ω ∇3 Ω and restricting the equations to u = 0, we obtain
2Ω2 K(u, 0) + ∂u (Ω2 · Ω−1 trχ)(u, 0) + Ω2 · Ω−1 trχ · Ωtrχ(u, 0) = 0.
Together with the fact
2(1 − µ∗ ) 2 µ∗
Ω−1 trχ(u, 0) = , Ωtrχ(u, 0) = − , Ω2 (u, 0) = r 1−µ∗ (u, 0), ∂u r(u, 0) = −1,
r(u, 0) r(u, 0)
we deduce
µ∗ µ∗ 4(1 − µ∗ )
2K(u, 0) + r− 1−µ∗ · ∂u (r 1−µ∗ −1 ) · 2(1 − µ∗ )(u, 0) − = 0,
r(u, 0)2
which infers
1 1
K(u, 0) = −(2 − 4µ∗ − 4 + 4µ∗ ) · = .
2r(u, 0)2 r(u, 0)2
Quoting the results in [18], for the scalar field along u = 0 it holds
r
√ µ∗ 1
Ω∂3 φ(u, 0) =∂u φ(u, 0) = 2 ,
1 − µ∗ r(u, 0)
r (2.21)
√ 1 − µ∗ ∂u r(u, 0)
Ω∂4 φ(u, 0) =∂u φ(u, 0) = 2 · .
µ∗ r(u, 0)
Remark 11. Along u = 0, employing ∂u r(u, 0) = −1 and ∂u (Ω−2 ∂u r) = − 21 rΩ−2 (∂u φ)2 , via
(2.21) we also obtain
1 |r∂u φ|2 µ∗ 1
Ωω(u, 0) = − ∂u log Ω(u, 0) = (u, 0) = ∗
· ,
2 8r 4(1 − µ ) r(u, 0)
which is consistent with (2.20).
√
Remark 12. The 2 on the right of (2.21) is owing to in [18] Christodoulou’s use of
1
Ricµν − Rgµν = 2Tµν ,
2
1
Tµν = ∂µ φc ∂ν φc − gµν ∂ σ φc ∂σ φc .
2
q
µ∗
Here we denote φc to be the scalar field function employed in [18]. Denoting k := 1−µ∗ , on
page 644 of [18], we have that along s = s∗ (i.e. u = 0) it holds
∂u φc ∂u φc 1
r (u, 0) = −k and r (u, 0) = . (2.22)
∂u r ∂u r k
22 XINLIANG AN
Setting the scalar field function φ(u, u, θ1 , θ2 ) used in this paper to be

√
φ(u, u, θ1 , θ2 ) = 2φc (u, u, θ1 , θ2 ),
via ∂u r(u, 0) = −1 we then translate (2.22) to (2.21).
Remark 13. At the end of this subsection, we point out that, along u = 0 by employing
equations r∂u ∂v φ = −∂u r∂v φ − ∂v r∂u φ and r∂u ∂v r = −∂u r∂v r − 14 Ω2 , it holds
∂ ∂v φ Ω2 ∂v φ ∂u φ
− · + = 0. (2.23)
∂u ∂v r 4r∂v r ∂v r r
Treat ∂v φ(u, 0) as it is determined by other geometric quantities. Applying the fact
r
Ω2 −1 1 ∂u φ √ µ∗ 1
− (u, 0) = ∗
· and (u, 0) = 2 · ,
4r∂v r 1 − µ r(u, 0) r 1 − µ r(u, 0)2
∗
we can solve an ODE and we see that the below equality as stated in (2.21)
r
∂v φ(u, 0) ∂u φ(u, 0) √ 1 − µ∗ 1 1
= = 2 · ≈
∂v r(u, 0) ∂u r(u, 0) µ∗ r(u, 0) r(u, 0)
is debt to a special initial choice of ∂v φ/∂v r at S−1,0 . If we do not use this choice (by adding a
general perturbation), back to (2.23), we would have that ∂v φ(u, 0)/∂v r(u, 0) behaviors like
−1 −µ∗ µ∗
r(u, 0) 1−µ∗ = r(u, 0)−1 · r(u, 0) 1−µ∗ . With ∂v r(u, 0) ≈ r 1−µ∗ (u, 0), it yields ∂v φ(u, 0) ≈
r(u, 0)−1 . In below sections, we demonstrate an approach, which can deal with these general
scenarios.
2.4. Norms. For geometric quantities defined on Su,u , we now set norms that we will work
with. We use ψ and ψ to denote the below geometric quantities
ψ ∈ {Ωtrχ, Ωχ̂, Ωω̂, Ω∂4 φ}, (2.24)
ψ ∈ {η, η, Ωω − Ωω(u, 0), Ωχ̂, Ωtrχ − Ωtrχ(u, 0), ∇e1 φ, ∇e2 φ, Ω∂3 φ − Ω∂3 φ(u, 0)}. (2.25)
For conventional simplicity, we also write
Ωω(u, 0) =(Ωω)0 , Ω∂3 φ(u, 0) = (Ω∂3 φ)0 ,
−1 (2.26)
Ωtrχ(u, 0) = (Ωtrχ)0 , Ω trχ(u, 0) = (Ω−1 trχ)0 , Ω(u, 0) = Ω0 .
With constants a, b chosen as in Section 3, we then define

1 1
Oi,∞ (u, u) := 1 kui+1 ∇i ψ kL∞ (Su,u ) + 1 kui+2 ∇i ψ kL∞ (Su,u ) with 0 ≤ i ≤ 2, (2.27)
a 2 ua 2
1 1 i+3 i
Ri,∞ (u, u) := 1 kui+2 ∇i (Ωβ)kL∞ (Su,u ) +ku ∇ ρkL∞ (Su,u )
a 2 ua
1 i+3 i 1 i+3 i 1
+ 1 ku ∇ σ̌kL∞ (Su,u ) + 1 ku ∇ (K − 2 )kL∞ (Su,u ) (2.28)
ua 2 ua 2 |u|
1 i+3 i
+ 1 ku ∇ (Ωβ)kL∞ (Su,u ) with 0 ≤ i ≤ 1.
ua 2
1 1
Oi,2 (u, u) := 1 kui ∇i ψ kL2 (Su,u ) + 1 kui+1 ∇i ψ kL2 (Su,u ) with 0 ≤ i ≤ 4. (2.29)
a 2 ua 2
We also set
1 5 5
Õ5,2 (u, u) := 1 1 ku ∇ (Ωtrχ, Ωω̂, Ωχ̂)kL2 2
u L (Su,u )
u a2
2
1 1 −1
u2 a4 1 5 5 1
+ 1 1 1 ku ∇ ηkL2 2
u L (Su,u )
+ 3 3 ku6 ∇5 ηkL2u L2 (Su,u )
|u| 2 Ω u2 a4 u2 a4
(2.30)
Ω 1
+ 1 1 ku5 ∇5 ηkL2u L2 (Su,u ) + 1 1 kΩu5 ∇5 ηkL2u L2 (Su,u )
u2 a2 u2 a2
1 1 −1
u2 a4 1 5 5
+ 1+ 1 1 1 ku ∇ (Ωtrχ, Ωχ̂, trχ, χ̂)kL2 2
u L (Su,u )
,
|u| 2 Ω u2 a4
′ 1
Õ5,2 (u, u) := 3 ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . (2.31)
u2 a
For 1 ≤ i ≤ 4, we further let
1 1
Ri,1 (u, u) := 1 1 kui+1 ∇i (ΩβA − Ω∇4 φ∇A φ)kL2u L2 (Su,u )
u a
2 2 2
(2.32)
1 1 1
+ 1 1 kΩui+1 ∇i (K − 2 − ∇A φ∇A φ, σ̌)kL2u L2 (Su,u ) ,
u2 a2 |u| 4
1 1 −1
u2 a4 1 i+2 i 1 1
Ri,2 (u, u) := 1 + 1 3 3 ku ∇ (K − 2 − ∇A φ∇A φ, σ̌)kL2u L2 (Su,u )
|u| 2 Ω u2 a4 |u| 4
1 1 −1
(2.33)
u2 a4 1 i+2 i 1
+ 1+ 1 3 3 ku ∇ (β A + ∇3 φ∇A φ)kL2u L2 (Su,u ) .
|u| 2 Ω u2 a4 2
For 1 ≤ i ≤ 5, we then set
1 1
Si,1 (u, u) := 1 1 kui ∇i (Ωe4 φ)kL2u L2 (Su,u ) + 1 1 kΩui ∇i (eA φ)kL2u L2 (Su,u ) , (2.34)
u a
2 2 u a2
2
1 1 −1
u2 a4 1 i i
Si,2 (u, u) := 1 + 1 1 1 ku ∇ (eA φ)kL2 2
u L (Su,u )
|u| 2 Ω u2 a4
1 1 −1
(2.35)
u2 a4 1 −1 i i
+ 1+ 1 1 1 kΩ u ∇ (Ωe3 φ)kL2u L2 (Su,u ) .
|u| 2 Ω u2 a4
And we further define O(u, u), R(u, u), S(u, u) as
X X X
O(u, u) := Oi,∞ (u, u) + Oi,2 (u, u), R(u, u) := Ri,1 (u, u) + Ri,2 (u, u),
i≤2 i≤4 1≤i≤4
X
S(u, u) := Si,1 (u, u) + Si,2 (u, u).
1≤i≤4
′
Next we denote Oi,∞ , Oi,2 , Õ5,2 , Õ5,2 , Ri,∞ , Ri,1 , Ri,2 , Si,1 , Si,2 to be the supremum over u, u
′
in the considered spacetime region of Oi,∞ (u, u), Oi,2 (u, u), Õ5,2 (u, u), Õ5,2 (u, u), Ri,∞ (u, u), Ri,1 (u, u), Ri,2 (u, u), Si
respectively. Finally, we set O, R, S to be:
X X X X
O := Oi,∞ + Oi,2 , R := Ri,1 + Ri,2 , S := Si,1 + Si,2 .
i≤2 i≤4 1≤i≤4 1≤i≤4
(0) (0) ′(0) (0) (0)

We also denote O , Õ , Õ , R , S to be the corresponding norms along the initial
hypersurfaces u = −1 and u = 0. Lastly, we define an initial data quantity
I (0) := sup I (0) (u)
0≤u≤δ
with
X X 1
I (0) (u) := 1 + 1 k(∇)m (∇4 )k (χ̂, ∇4 φ)kL2 (S−1,u ) .
0≤m≤6, 0≤k≤5
a2
24 XINLIANG AN
Remark 14. Following the arguments carried out in detail in Chapter of [21], we have
X X X
Oi,∞ (−1, u) + Oi,2 (−1, u) + Ri,∞ (−1, u) . I (0) .
i≤2 i≤4 i≤2
In Section 7, we will employ the initial values of Ωκ, µ, Ω−1 trχ along H−1 , with
1 1 1 1
κ := ∇ω̂ + ∗ ∇ω † − (β − ∇4 φ∇φ), µ := −div η + K − 2
− ∇A φ∇A φ.
2 2 |u| 4
Note that we set ω † to be 0 and Ω to be 1 along H−1 . And it holds
1 1 1
ku4 ∇4 (Ωκ)kL2 (S−1,u ) . a 2 , ku5 ∇4 µkL2 (S−1,u ) . ua 2 , ku5 ∇5 (Ω−1 trχ)kL2 (S−1,u ) . ua 2 .
2.5. Exterior and Interior Regions. To establish the hyperbolic existence result, we first
choose a u1 < 0 satisfying 0 < |u1 | ≪ 1 and then divide the spacetime M into two regions:
1
· the exterior: −1 ≤ u ≤ u1 and 0 ≤ ua 2 ≤ |u1 |Ω2−δ̃ (u1 , 0),
1
· the interior: u1 ≤ u ≤ 0 and 0 ≤ ua 2 ≤ |u|Ω2−δ̃ (u, 0).
Remark 15. For notational simplicity, in the rest of this paper, we set constant δ̃ obeying
0 < δ̃ ≪ 1. In reality, if needed we can set δ̃ to be any number satisfying 0 < δ̃ < 1/2 and all
the below proof stays the same.
In this article, we will use the below inequality frequently
1 1 1
ua 2 ua 2 ua 2
≤ ≤ 3 ≪ 1. (2.36)
|u| |u|Ω |u|Ω 2
This inequality is correct because (with a = 1)
1 3
ua 2 |u1 |Ω2−δ̃ (u1 ,0) |u1 |Ω 2 (u1 ,0) 1
· in the exterior we have 3 ≤ 3 = 3 · Ω 2 −δ̃ (u1 , 0) ≪ 1,
|u|Ω 2 |u|Ω 2 |u|Ω 2
1
ua 2 1 1
· in the interior we have 3 ≤ Ω 2 −δ̃ (u, 0) . Ω 2 −δ̃ (u1 , 0) ≪ 1.
|u|Ω 2
With above discussion, we can further choose a constant B, such that in both regions
1 1 1
ua 2 ua 2 ua 2 1
≤ ≤ 3 ≤ ≪ 1. (2.37)
|u| |u|Ω |u|Ω 2 B
2.6. An Approach of Bootstrap. In this article, we carry out a bootstrap argument to prove
uniform upper bounds for O, R and S. For initial data along u = 0 and u = −1, we have
O(0) + R(0) + S (0) ≤ C0 I (0) ,
where C0 depends only on initial data. For notational simplicity, we rewrite the above line as
O(0) + R(0) + S (0) . I (0) .
For the rest of this article, we keep the same notation: if P ≤ C0 Q with constant C0 depending
only on the initial data, we could write it as
P . Q.
(0)
Under this notation, we have I . 1.
Our goal in below sections is to show that in the considered spacetime region, it holds
O + R + S ≤ I (0) + (I (0) )2 + 1.
To achieve this, we make bootstrap assumptions
O(u, u) ≤ O, (2.38)
R(u, u) ≤ R, (2.39)
S(u, u) ≤ S, (2.40)
′
Õ5,2 (u, u) ≤ Õ, Õ5,2 (u, u) ≤ Õ′ . (2.41)
Recall (2.37). For constants in bootstrap assumptions (2.38)-(2.41), for the rest of this paper,
we assume that they satisfy
1 1
1 1 ua 2 1 ua 2 − 32
1
1 ≤ O, R, S, Õ ≪ (Õ′ ) 8 ≪ B 32 ≤ ( 3 )− 32 ≤ ( ) . (2.42)
|u|Ω 2 |u|Ω
In below, we demonstrate the steps for improving the bootstrap assumptions:6

In Section 4 we prove
P
O (χ̂, trχ) . 1 with improved estimates for trχ,
Pi≤4 i,2 P
Oi,2 (ω̂) . 1 + i≤4 Oi,2 (∇4 φ),
Pi≤4 P 2
i≤4 Oi,2 (ψ ) . 1 + i≤4 Oi,2 (ω̂) + Õ5,2 (ω̂) + R(β) with improved bound for Ωtrχ + |u| .
In Section 5 we obtain
P P P
i≤4 Oi,2 (∇4 φ) . 1, i≤4 Oi,2 (∇A φ) . S, i≤4 Oi,2 (∇3 φ) . S.
In Section 6 we derive
S5,1 . 1, S5,2 . 1.
With the bounds obtained in Section 5 and Section 6, we revisit the estimates in Section 4.
Hence we now get
P P
i≤4 Oi,2 (ψ ) . 1, i≤4 Oi,2 (ψ ) . 1 + Õ5,2 (ω̂) + R(β).
In Section 7 we show
Õ5,2 (ω̂) . 1 + R(β), Õ5,2 (trχ, χ̂) . 1 + R(β),
′
Õ5,2 (trχ) . 1,
Õ5,2 (η) . 1, Õ5,2 (η) . 1 + R(K, σ̌) + R(β) + R(K, σ̌),
Õ5,2 (Ωtrχ, trχ) . 1, Õ5,2 (Ωχ̂) . 1, Õ5,2 (χ̂) . R(β).
In Section 8 we derive
R(β) + R(K, σ̌) . 1, R(K, σ̌) + R(β) . 1.
Remark 16. The proof strategy and the main theorems of this paper also hold for perturbed
Christodoulou’s initial data prescribed along u = 0. Denote
∂u r ∂u r
Ω̊(u, 0) = Ω(u, 0) and Γ̊(u, 0) ∈ { (u, 0), (u, 0), ∂u φ(u, 0), ∂u φ(u, 0)}.
r r
We further set Ω̊c (u, 0) and Γ̊c (u, 0) to be the corresponding values of Christodoulou’s naked-
singularity solution along u = 0. Our below proof strategy also extends to spherically-symmetric
initial data along u = 0 satisfying
Ω̊(u, 0) Ω̊c (u, 0)δ̃ Γ̊(u, 0) Ω̊(u, 0)δ̃

| |≤1+ 1 , | |≤1+ 1
Ω̊c (u, 0) a 3 Γ̊c (u, 0) a3
with constants 0 < δ̃ ≪ 1 and a ≥ 1.7 For simplicity of the proof and for more precise
statements of the main theorems, in the main body of this paper we stick to Christodoulou’s
naked-singularity initial data with Footnotes 4, 9, 13, 14, 15, 16, 17 and Remark 20 added to
explain how to deal with the perturbed scenario.8
6A more detailed explanation of the bootstrap method can be found in Section 3.1 of [4].
7The reader is also referred to a spherically-symmetric extension of Christodoulou’s naked-singularity solution
by Singh in [37].
8The methods developed in this paper also allow non-spherically-symmetric initial data prescribed along
u = 0, the author will provide the details in a separate paper.
26 XINLIANG AN
3. Preliminary Estimates
3.1. Estimates for metric components. We now start the hyperbolic estimates. We first
obtain a bound for Ω under the bootstrap assumptions:
Proposition 3.1. Under the assumptions of Theorem 1.1 and the bootstrap assumptions (2.38),
(2.39), (2.40), (2.41), we have
Ω 1
k − 1kL∞ (Su,u ) ≤ B − 2 .
Ω(u, 0)
Proof. We employ the equation
Ω 1 ∂ ∂
Ωω̂ = − ∇4 log Ω = − ( + dA A ) log Ω. (3.1)
2 2 ∂u ∂θ
Integrating equation (3.1) along e4 direction, we obtain
Z u 1
Ω ua 2 O
|| log ||L∞ (Su,u ) . ||Ωω̂||L∞ (Su,u′ ) du′ . .
Ω(u, 0) 0 |u|
Here we use the bootstrap assumption (2.38). This implies that
1
Ω ua 2 O 1
k − 1kL∞ (Su,u ) . ≤ B− 2 .
Ω(u, 0) |u|

We then control g/ under the bootstrap assumptions. Following the same argument as for
Proposition 5.2 in [2], we have
Proposition 3.2. Fix a point (u, ω) on the initial hypersurface H 0 . Along the outgoing charac-
teristic emanating from (u, ω), we define Λ(u) and λ(u) to be the larger and smaller eigenvalue
of g/−1 (u, u = 0, ω)g/(u, u, ω). Under the assumptions of Theorem 1.1 and the bootstrap assump-
tions (2.38), (2.39), (2.40) and (2.41), then it holds
1
|Λ(u) − 1| + |λ(u) − 1| ≤ B − 2
for every u ∈ [0, δ]. Consequently, for every u ∈ [0, δ], this also implies
1 dvolg/u
|ξ(u) − 1| ≤ B − 2 with ξ(u) = .
dvolg/0
As corollaries, we also obtain the below area estimate and the Sobolev embedding.
(2.39), (2.40), (2.41), it holds that
1
sup |Area(Su,u ) − Area(Su,0 )| ≤ |u|2 B − 2 .
u,u
Proposition 3.4. With the assumptions of Theorem 1.1 and the bootstrap assumptions (2.38),
(2.39), (2.40) and (2.41), it holds that
X
kφkL∞ (Su,u ) . kui−1 ∇i φkL2 (Su,u ) . (3.2)
i≤2
Note that the detailed proof of the Sobolev embedding is provided in Propositions 5.6-5.9 in [2].
3.2. Estimates for Transport Equations. In this article, we will employ the below two
propositions frequently:
(2.39), (2.40) ,(2.41), for an Su,u -tangent tensor field φ of arbitrary rank it holds
Z u
||φ||L2 (Su,u ) . ||φ||L2 (Su,u′ ) + ||Ω∇4 φ||L2 (Su,u′′ ) du′′ .
u′
Proposition 3.6. Set φ and F to be Su,u -tangent tensor fields of rank k obeying the below
transport equation:
Ω∇3 φA1 ...Ak + λ0 ΩtrχφA1 ...Ak = ΩFA1 ...Ak
and denote λ1 = 2(λ0 − 21 ). Under the assumptions of Theorem 1.1 and the bootstrap assump-
tions (2.38), (2.39), (2.40), (2.41), then it holds that
Z u
|u|λ1 kφkL2 (Su,u ) . kφkL2 (S−1,u ) + |u′ |λ1 kΩF kL2 (Su′ ,u ) du′ .
−1
The proof of the above two propositions are the same as in Proposition 5.4 and Proposition 5.5
of [2] and hence are omitted.9
3.3. Commutation Formula. We quote the following formula from [27]:
Proposition 3.7. The commutator [∇4 , ∇] acting on an (0, r) S-tensor satisfies
[∇4 , ∇B ]φA1 ...Ar
=[D4 , DB ]φA1 ...Ar + (∇B log Ω)∇4 φA1 ...Ar − (g/−1 )CD χBD ∇C φA1 ...Ar
r
X r
X
− (g/−1 )CD χBD η A φA1 ...Âi C...Ar + (g/−1 )CD χAi B η D φA1 ...Âi C...Ar .
i
i=1 i=1
Proposition 3.8. The commutator [∇3 , ∇] acting on an (0, r) S-tensor satisfies

[∇3 , ∇B ]φA1 ...Ar
=[D3 , DB ]φA1 ...Ar + (∇B log Ω)∇3 φA1 ...Ar − (g/−1 )CD χBD ∇C φA1 ...Ar
r
X r
X
− (g/−1 )CD χBD ηAi φA1 ...Âi C...Ar + (g/−1 )CD χA ηD φA1 ...Âi C...Ar .
iB
i=1 i=1
By induction, using commutation formula repeatedly we get the schematic form:

Proposition 3.9. Assume Ω∇4 φ = ΩF0 . Let Ω∇4 ∇i φ = ΩFi . Then it holds
X X
ΩFi = ∇i1 (η + η)i2 ∇i3 (ΩF0 ) + ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 φ
i1 +i2 +i3 =i i1 +i2 +i3 +i4 =i
X
+ i1
∇ (η + η) ∇ (Ωβ)∇i4 φ.
i2 i3
i1 +i2 +i3 +i4 =i−1

i1
Here by ∇ (η + η)i2 we mean the sum of all terms, which itself is a product of i2 factors, each
factor being ∇j (η + η) for some j ∈ Z≥0 and that the sum of all j being i1 , i.e., ∇i1 (η + η)i2 =
X
∇j1 (η + η)...∇ji2 (η + η). Similarly, assume Ω∇3 φ = ΩG0 . Suppose Ω∇3 ∇i φ = ΩGi .
j1 +...+ji2 =i1
Then we have
i X
ΩGi = − Ωtrχ∇i φ + ∇i1 (η + η)i2 ∇i3 (ΩG0 )
2
i1 +i2 +i3 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 φ
i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 φ.
i1 +i2 +i3 +i4 =i−1
By replacing Ωβ and Ωβ via the schematic Codazzi equations:

1
Ωβ = ∇ψ + ψ ψ , Ωβ = ∇ψ + ψ + ψ ψ,
|u|
rewriting Ωtrχ, Ωχ̂ as ψ and substituting η, η, Ωχ̂, Ωtrχ−(Ωtrχ)0 with ψ , we obtain a simplified
version
9With perturbed Christodoulou initial data, the quantity kΩtrχ(u, u, ω) + 2 k is still uniformly bounded
|u| L1 u
and hence the proof for Proposition 3.6 is the same.
28 XINLIANG AN
Proposition 3.10. Suppose ∇4 φ = F0 . Let ∇4 ∇i φ = Fi . Then we have

X X
ΩFi = ∇i1 ψ i2 ∇i3 (ΩF0 ) + ∇i1 ψ i2 ∇i3 ψ ∇i4 φ.
i1 +i2 +i3 =i i1 +i2 +i3 +i4 =i
Similarly, assume ∇3 φ = G0 . Let ∇3 ∇i φ = Gi . Then it holds

i X X
ΩGi + Ωtrχ∇i φ = ∇i1 ψ i2 ∇i3 (ΩG0 ) ∇i1 ψ i2 ∇i3 φ
2 i1 +i2 +i3 =i i1 +i2 +i3 =i
X
+ (Ωtrχ, Ω∇3 φ)∇i1 ψ i2 ∇i3 φ.
i1 +i2 +i3 =i,
i3 ≤i−1
Remark 17. For the rest of this paper, when there is no danger of confusion, the constants on
the left of the equations are kept precise. The uniform constants in front of coefficients on the
right are usually omitted.
3.4. Elliptic Estimates for Hodge Systems. We first recall the definition of the divergence
and curl of a symmetric covariant tensor with an arbitrary rank:
(div φ)A1 ...Ar := ∇B φBA1 ...Ar ,
(curl φ)A1 ...Ar := /ǫ BC ∇B φCA1 ...Ar .
Here /ǫ is the volume form associated to the metric g/ on Su,u . Recall that the trace is defined
via
(tr φ)A1 ...Ar−1 = (g/−1 )BC φBCA1 ...Ar−1 .
In this article, we will use the following L2 (Su,u ) elliptic estimates proved in Chapter 7 of [21].
Proposition 3.11. Under the assumptions of Theorem 1.1 and the bootstrap assumptions
(2.38), (2.39), (2.40), (2.41), denoting φ to be a totally symmetric r + 1 covariant tensorfield on
a 2-sphere (Su,u , g/) satisfying
div φ = f˜, curl φ = g̃, trφ = h̃,
then, for 1 ≤ i ≤ 5, it holds that
i−1
X
kui ∇i φ||L2 (Su,u ) . (||uj+1 ∇j (f˜, g̃)||L2 (Su,u ) + kuj ∇j h̃kL2 (Su,u ) + ||uj ∇j φ||L2 (Su,u ) ).
j=0
As a special case with φ being a symmetric traceless 2-tensor, it holds that curl φ =∗ div φ.
Using the above proposition, we then get
Proposition 3.12. Assume that φ is a symmetric traceless 2-tensor obeying
div φ = f˜.
Then, under the assumptions of Theorem 1.1 and the bootstrap assumptions (2.38), (2.39),
(2.40), (2.41), for 1 ≤ i ≤ 5, it holds
i−1
X
i i
||u ∇ φ|| L2 (S u,u )
. (||uj+1 ∇j f˜||L2 (Su,u ) + ||uj ∇j φ||L2 (Su,u ) ).
j=0
4. L2 (Su,u ) estimates for Ricci coefficients

We then proceed to derive the L2 (Su,u ) estimates for the Ricci coefficients and their first four
angular derivatives.
We first state a useful lemma, which can be verified directly.10
10A detailed proof is provided in Proposition 6.1 of [2].

Lemma 4.1. Under the assumptions of Theorem 1.1 and the bootstrap assumptions (2.38),
(2.39), (2.40), (2.41), it holds that
X X
kui1 +i2 +1 ∇i1 ψ i2 +1 kL2 (Su,u ) . kui1 +1 ∇i1 ψ kL2 (Su,u )
i1 +i2 ≤4 i1 ≤4
and
X X
kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ (Su,u ) . kui1 +2 ∇i1 ψ kL∞ (Su,u ) .
i1 +i2 ≤2 i1 ≤2
In particular, by (2.38), we have

X X 1
kui1 +i2 +1 ∇i1 ψ i2 +1 kL2 (Su,u ) + kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞ (Su,u ) . ua 2 O.
i1 +i2 ≤4 i1 +i2 ≤2
We now start from the L2 (Su,u ) estimates for Ωχ̂.
(2.39), (2.40), (2.41), it holds
X 1
kui ∇i (Ωχ̂)kL2 (Su,u ) . a 2 .
i≤4
Proof. We employ the null structure equation

1
Ω∇3 (Ωχ̂) + Ωtrχ(Ωχ̂) = Ω2 ∇η + ψ ψ + ψ ψ .
2
After commuting this equation with i angular derivatives, we obtain
i+1
Ω∇3 ∇i (Ωχ̂) + Ωtrχ∇i (Ωχ̂)
2
X X X 1 i1 i2 +1 i3
=Ω2 ∇i+1 η + ∇i1 ψ i2 +2 + ∇i1 ψ i2 +1 ∇i3 ψ + ∇ ψ ∇ ψ.
i1 +i2 =i i1 +i2 +i3 =i i1 +i2 +i3 =i−1
|u|
Note that the fourth term only presents when i ≥ 1. To bound kui ∇i (Ωχ̂)kL∞ ∞ 2
u Lu L (Su,u )
, we
then apply Proposition 3.6 with λ0 = i+1
2 and estimate each term on the right.
For the ∇i+1 η term, when i = 4 we have
1 1
Ω u2 a2 1
− 14
ku4 Ω2 ∇5 ηkL1u L2 (Su,u ) ≤ k kL2u kΩu5 ∇5 ηkL2u L2 (Su,u ) . 1 Õ5,2 ≤ a B
2 .
|u| |u| 2
The rest terms are lower order. For example, we control the last term as
X X uaO2 1 1
k ui−1 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2 (Su,u ) . ≤ a 2 B− 4 . (4.1)
i1 +i2 +i3 =i−1
|u|
i≤4
1 1
All the remaining terms on the
P right obey the same upper bound a 2 B − 4 .
1
i
For initial data, we have i≤4 k∇ (Ωχ̂)kL∞ 2
u L (S−1,u )
≤ a 2 . Gathering all above estimates,
we get
X 1
kui ∇i (Ωχ̂)kL∞ 2
u L (Su,u )
. a2 .
i≤4
We then derive the estimates for Ωtrχ and its derivatives. We will also point out that
∇i (Ωtrχ) obeys a better bound than the standard estimate for ∇i ψ .
30 XINLIANG AN
(2.39), (2.40), (2.41), it holds
X 1
kui+1 ∇i (Ωtrχ−Ω2 (Ω−1 trχ)0 )kL2 (Su,u ) . uaO2 ≪ |u|a 2 ,
i≤4
X uaO2 1
kui ∇i (Ωtrχ)kL2 (Su,u ) . ≪ a2 .
|u|
i≤4
Proof. We use the following null structure equation

1
Ω∇4 (Ω−1 trχ) = −|χ̂|2 − (∇4 φ)2 − (trχ)2 .
2
Note that Ω−1 trχ(u, 0) = 2(1 − µ∗ )/|u|. Utilizing ∇4 (Ω−1 trχ)0 = 0, we have
1
Ω∇4 (Ω−1 trχ − (Ω−1 trχ)0 ) = −|χ̂|2 − (∇4 φ)2 − (trχ)2 .
2
Commuting this equation with i angular derivatives, we get the below form
Ω∇4 ∇i (Ω−1 trχ − (Ω−1 trχ)0 )

X X
= ∇i1 ψ i2 ∇i3 (χ̂, ∂4 φ)∇i4 (χ̂, ∂4 φ) + Ω−1 ∇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|
We then employ Proposition 3.5 and control the right hand side in the kui+1 · kL1u L2 (Su,u ) norm
to bound kui+1 ∇i (Ω−1 trχ − (Ω−1 trχ)0 )kL∞ 2
u L (Su,u )
.
Remark 18. In below estimates and for the rest of this article, when there is no danger of
confusion, we use Ω−1 to stand for kΩ−1 kL∞ (Su,u ) .
Employing the bootstrap assumption (2.38) together with Sobolev embedding in Proposition
3.4, for the case i2 = 0, we first control
X X
k ui+1 ∇i3 (χ̂, ∂4 φ)∇i4 (χ̂, ∂4 φ)kL1u L2 (Su,u )
i≤4 i3 +i4 =i
X X
.u kui3 +1 ∇i3 (χ̂, ∂4 φ)kL∞
u Lu L (Su,u )
∞ ∞ kui4 ∇i4 (χ̂, ∂4 φ)kL∞ ∞ 2
u Lu L (Su,u )
. Ω−2 uaO2 .
i3 ≤2 i4 ≤4
P
Similarly, contribution from i1 +i2 +i3 +i4 =i ∇i1 ψ i2 ∇i3 ψ ∇i4 trχ with i2 = 0 obeys
X X
k Ω−1 ui+1 ∇i3 ψ ∇i4 trχkL1u L2 (Su,u )
i≤4 i3 +i4 =i
ua O X i+1 i −1
1
2
. ku ∇ (Ω trχ − (Ω−1 trχ)0 )kL∞ ∞ 2
u Lu L (Su,u )
+ Ω−2 uaO2 .
|u|
i≤4
P 1 i1 i2 i3
When i2 = 0, we can also control the third term i1 +i2 +i3 =i |u| ∇ ψ ∇ ψ by
X X 1 Ω2 O 1
kui ∇i ψ kL1u L2 (Su,u ) . u kui ∇i ψ kL∞ ∞ 2
u Lu L (Su,u )
.ua 2 O = Ω−2 ua · 1 ≤ Ω−2 uaB − 2 .
i≤4 i≤4
a 2
For the case i2 ≥ 1, using the bootstrap assumption (2.38) together with Proposition 4.1, for
the first two terms we also have
X X 3
u2 a 2 O 3 1
i+1 i1 i2 +1 i3 i4
k u ∇ ψ ∇ ψ ∇ ψ kL1u L2 (Su,u ) . ≤ Ω−2 uaB − 4 .
i +i +i +i =i−1
|u|
i≤4 1 2 3 4
The third term can be bounded similarly via

X X u2 aO2 1
k ui ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2 (Su,u ) . ≤ Ω−2 uaB − 4 .
i +i +i =i−1
|u|
i≤4 1 2 3
Gathering above estimates, we obtain

X
kui+1 ∇i (Ω−1 trχ − (Ω−1 trχ)0 )kL∞ ∞ 2
u Lu L (Su,u )
i≤4
ua 2 O X i+1 i −1
1
.Ω−2 uaO2 + ku ∇ (Ω trχ − (Ω−1 trχ)0 )kL∞ ∞ 2

u Lu L (Su,u )
.
|u|
i≤4
1
− 12
Since ua O/|u| ≤ B
2 ≪ 1, we hence prove
X
kui+1 ∇i (Ω−1 trχ − (Ω−1 trχ)0 )kL∞ ∞ 2
u Lu L (Su,u )
. Ω−2 uaO2 .
i≤4
Using ∇A Ω = Ω(ηA + η A )/2, this further implies

X
kui+1 ∇i (Ωtrχ−Ω2 (Ω−1 trχ)0 )kL∞ ∞ 2
u Lu L (Su,u )
. uaO2 ,
i≤4
X uaO2 1
kui ∇i (Ωtrχ)kL∞ ∞ 2
u Lu L (Su,u )
. ≪ a2 .
|u|
i≤4

We proceed to prove the estimates for ω̂ and its derivatives. And ω̂ also obeys a slightly
better bound than a general component ψ .
(2.39), (2.40), (2.41), it holds
X 1
X
kui ∇i (Ωω̂)kL2 (Su,u ) . a 2 + kui ∇i (Ω∇4 φ)kL2 (Su,u ) .
i≤4 i≤4
Proof. We employ the following schematic null structure equation for ω:

Ω∇3 (Ωω̂) = Ω2 K + ψ ψ + ψ ψ + ΩtrχΩtrχ + Ω∇4 φ · Ω∇3 φ.
By commuting it with angular derivative for i times, we obtain
i
Ω∇3 ∇i (Ωω̂) + Ωtrχ∇i (Ωω̂)
X 2 X X
= Ω2 ∇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 i
+ ∇ ψ ∇ ψ+ ∇ (Ωtrχ, Ω∇4 φ).
i1 +i2 +i3 =i−1
|u| |u|
Note that we only encounter the fourth term when i ≥ 1.
We then apply Proposition 3.6 with λ0 = 2i . Since ω̂(−1, u)=0, we control kui−1 ∇i (Ωω̂)kL∞ ∞ 2
u Lu L (Su,u )
i−1
by the ku · kL∞ 1 2
u Lu L (Su,u )
norm of the right hand side.
X X
Denote 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|
We first have
X X 1 1 1
u 2 a 2 O2
1
a 2 B− 4
1
i−1 i
ku Fi kL∞ 1 2
u Lu L (Su,u )
. ku Fi kL∞ 1 2
u Lu L (Su,u )
. 3 ≤ .
|u| |u| 2 |u|
i≤4 i≤4
32 XINLIANG AN
For the terms containing K, if needed, we also employ K . (K − |u|1 2 )+ |u|1 2 . Applying bootstrap
assumptions (2.40),(2.38) and Proposition 4.1, their contributions obey
1 1 1 1 1
u 2 a 2 (R + O) ua 2 O′ 1 1 a 2 B− 4
. 3 + 5 · u 2 a 2 (R + O) ≤ .
|u| 2 |u| 2 |u|
We proceed to control the contribution from |u|1 2 . Note that for this term, the case i2 = 0
happens only when i = 0. For this scenario, the term obeys
1 1
ku−1 2 kL1u L2 (Su,u ) . . (4.2)
|u| |u|
For i2 ≥ 1, it holds
X X 1
1 1
ua 2 O a 2 B − 4
1
i−1 i1 i2 +1
k u ∇ ψ k L 1 L2 (S ) . ≤ .
i1 +i2 =i−1
|u|2 u u,u
|u|2 |u|
i≤4
Here in the second inequality, we employ the bootstrap assumption (2.38). Collecting all the
above estimates, we prove
X X 1 1
1
a 2 B− 4
1
2 i−1 i1 i2 +1 i3
k Ω u ∇ ψ ∇ (K − 2 )kL1u L2 (Su,u ) . + . (4.3)
i +i +i =i−1
|u| |u| |u|
i≤4 1 2 3
For the remaining term, using the bound for ∇i (Ωtrχ) in Proposition 4.3, we have
X ua 1 1 X i i
kui−2 ∇i (Ωtrχ, Ω∇4 φ)kL1u L2 (Su,u ) . 2 + + ku ∇ (Ω∇4 φ)kL2 (Su,u )
|u| |u| |u|
i≤4 i≤4
(4.4)
1 X i i
1 3
u a 2 4
. 3 + ku ∇ (Ω∇4 φ)kL2 (Su,u ) .
|u| 2 |u|
i≤4
Combining all above estimates, we obtain

X 1
1
a 2 B− 4
1
1 X i i
i−1 i
ku ∇ (Ωω̂)kL2 (Su,u ) . + + ku ∇ (Ω∇4 φ)kL2 (Su,u ) .
|u| |u| |u|
i≤4 i≤4
Multiplying by |u| on both sides, we then derive

X 1 1
X
kui ∇i (Ωω̂)kL2 (Su,u ) . 1 + a 2 B − 4 + kui ∇i (Ω∇4 φ)kL2 (Su,u ) . (4.5)
i≤4 i≤4

Remark 19. Note that in Proposition 5.1, we will prove
X 1
kui ∇i (Ω∇4 φ)kL2 (Su,u ) . a 2 .
i≤4
Once this is obtained, back to (4.5) we will get

X 1
kui ∇i (Ωω̂)kL2 (Su,u ) . a 2 .
i≤4
We now estimate the L2 (Su,u ) norms of the remaining Ricci coefficients.

(2.39), (2.40), (2.41), we get
X 1
i+1 i
1 ku ∇ ψ kL2 (Su,u ) . 1+O(ω̂) + Õ5,2 (ω̂) + R(β).
i≤4
ua 2
In particular, it holds
X 1 1
i+1 i
1 ku ∇ ψ kL2 (Su,u ) . 1+O(ω̂) + 1 1 ku5 ∇5 (Ωω̂)kL∞ 2 2
u Lu L (Su,u )
+ R(β).
i≤4
ua 2 u a2
2
Proof. For ψ ∈ {η, ∇ log Ω, Ωtrχ − (Ωtrχ)0 , Ωχ̂, Ωω − (Ωω)0 }, the below schematic transport
equation holds
1 11
Ω∇4 ψ = Ωβ+∇(Ωω̂) + Ω2 K + Ω2 ∇η + ψ ψ + ψ ψ + ψ. (4.6)
|u|
By commuting the above equation with i angular derivatives, we have
X
Ω∇4 ∇i ψ =∇i (Ωβ) + ∇i+1 (Ωω̂, Ω2 η) + Ω2 ∇i1 ψ i2 ∇i3 K
i1 +i2 +i3 =i
X X 1 i1 i2 i3
+ ∇i1 ψ i2 +1 ∇i3 (ψ , ψ ) + ∇ ψ ∇ ψ.
i1 +i2 +i3 =i i1 +i2 +i3 =i
|u|
Notice ∇(Ωω)0 = 0. Due to the naked-singularity initial data along u = 0, all the quantities ψ
are initially 0. By Proposition 3.5, hence to estimate the quantity kui ∇i ψ kL∞ 2
u L (Su,u )
, we only
need to bound the kui · kL1u L2 (Su,u ) norm of the right hand side. We proceed to control each of
the terms in the equation. We bound the Ωβ term first
X u 2 X i+1 i
1 1
ua 2
i i
ku ∇ (Ωβ)k L1u L2 (Su,u ) ≤ ku ∇ (Ωβ)kLu L (Su,u ) .
2 2 R(β). (4.7)
|u| |u|
i≤4 i≤4
We then estimate the term ∇i+1 (Ωω̂) and get

X
kui ∇i+1 (Ωω̂)kL1u L2 (Su,u )
i≤4
u X i+1 i+1
1
u2 5 5
. ku ∇ (Ωω̂)kL∞ 2
u L (Su,u )
+ ku ∇ (Ωω̂)kL2u L2 (Su,u ) (4.8)
|u| |u|
i≤3
1
ua 2
. (Oi,2 (ω̂) + Õ5,2 (ω̂)).
|u|
For the ∇i+1 η term, using the bootstrap assumptions (2.38) and (2.41), it holds
X
kΩ2 ui ∇i+1 ηkL1u L2 (Su,u )
i≤4
u X i+2 i+1
1
u2
. 2 ku ∇ ψ kL∞ 2
u L (Su,u )
+ kΩ2 u5 ∇5 ηkL2u L2 (Su,u ) (4.9)
|u| |u|
i≤3
1 1 1 1 3 3 3 1
u2 a 2 O u 2 1 1 u2 a4 u2 a 4 O u 2 a 4 ua 2
. 2
+ · Ω2 · u 2 a 4 · 1 · Õ ≤ 2
+ 3 ≤ .
|u| |u| |u| 2 Ω |u| |u| 2 |u|
Employing the bootstrap assumptions (2.38), (2.40), we then bound the term containing the
Gauss curvature K and we write
1 1 1 1
K = (K − 2 − ∇A φ∇A φ) + 2 + ∇A φ∇A φ
|u| 4 |u| 4
to deduce the estimate. It holds
X X ua 2
1
k Ω2 ui ∇i1 ψ i2 ∇i3 KkL1u L2 (Su,u ) ≤ . (4.10)

i1 +i2 +i3 =i
|u|
i≤4
Applying (2.38) to control ψ and using Proposition 4.1 to estimate the product of ψ , as well
as using Sobolev embedding in Proposition 3.4, for i ≤ 4 we bound the fourth term and the
11On the right of the equation (4.6), we have that Ω2 K obeys at least the same and even better estimates
compared with K. When there is no danger of confusion, we sometimes also write K instead of Ω2 K for the
notational convenience and then the Ω2 could be viewed as a coefficient like 1 and is usually omitted in the
schematic writing.
34 XINLIANG AN
final term as follows:

X u2 aO2 ua 2 B − 4
1 1
k ui ∇i1 ψ i2 +1 ∇i3 (ψ , ψ )kL1u L2 (Su,u ) . 2

≤ , (4.11)
i1 +i2 +i3 =i
|u| |u|
X 1
ua 2 u2 aO2
k ui−1 ∇i1 ψ i2 ∇i3 ψ kL1u L2 (Su,u ) . (1 + Oi,2 (ω̂)) + . (4.12)
i1 +i2 +i3 =i
|u| |u|2
1
Notice that here kui ∇i ψ kL∞ 2
u L (Su,u )
≤ a 2 (1 + Oi,2 (ω̂)) was derived before.
Collecting all above estimates, we hence obtain
X 1
ua 2 X
kui ∇i ψ kL∞ 2
u L (Su,u )
. [1+ Oi,2 (ω̂) + Õ5,2 (ω̂) + R(β)].
|u|
i≤4 i≤4

For future use, we also prove an improved bound for ∇i (Ωtrχ − (Ωtrχ)0 ) for i ≥ 1:
(2.39), (2.40), (2.41), it holds that
X 3
u2 a4
3 1
ua 2 B − 2
1
i i
ku ∇ (Ωtrχ − (Ωtrχ)0 )kL2 (Su,u ) . 3 ≤ .
|u| 2 |u|
1≤i≤4
Proof. Note that Ωtrχ − (Ωtrχ)0 obeys the equation

1
Ω∇4 (Ωtrχ − (Ωtrχ)0 ) = Ω2 K + Ω2 ∇η + ψ ψ + ψ ψ + Ωtrχ.
u
By commuting the above equation with i angular derivatives, we have
Ω∇4 ∇i (Ωtrχ − (Ωtrχ)0 )
X X
=Ω2 ∇i+1 η + Ω2 ∇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 ∇ (Ωtrχ).
i1 +i2 +i3 =i
|u|
We revisit the proof of Proposition 4.5. Except for the last term, all other terms obey the
3 3 3
u 2 a 4 /|u| 2 upper bound. And for the last term, we have
X X 3
u2 a4
3 1
ua 2 B − 2
1
k ui−1 ∇i1 ψ i2 ∇i3 (Ωtrχ)kL1u L2 (Su,u ) . 3 ≤ .

i1 +i2 +i3 =i |u| 2 |u|
1≤i≤4
Here besides using Proposition 4.1, we also employ the improved bound for ∇i (Ωtrχ) derived in
Propositions 4.3. Collecting this with (4.9) and (4.11), we then obtain the desired conclusion.
We also notice that the ∇4 equation for K − |u|1 2 − 41 ∇A φ∇A φ contains exactly the same type
of terms as the ∇4 equation for ∇ψ . Hence, for ∇i (K − |u|1 2 − 41 ∇A φ∇A φ) we have the same
estimates as for ∇i+1 ψ for i ≤ 3.
(2.39), (2.40) and (2.41), it holds
X 1 1 1
X
kui+2 ∇i (K − 2 − ∇A φ∇A φ)kL∞ L ∞ L2 (S
u,u ) . ua 2 [1 + Oi,2 (ω̂) + Õ5,2 (ω̂) + R].
|u| 4 u u
i≤3 i≤4
In particular,
X 1 1 A
1
ua 2 X
i+1 i
ku ∇ (K − 2 − ∇ φ∇A φ)kL∞ L ∞ L2 (S
u,u ) . [1 + Oi,2 (ω̂) + Õ5,2 (ω̂) + R] ≪ 1.
|u| 4 u u
|u|
i≤3 i≤4
5. L2 (Su,u ) Estimates for Scalar Field

We now move to derive the L2 (Su,u ) estimates for the scalar field. We first prove
(2.39), (2.40), (2.41), we have
4
X 4
X 1
1 ua 2 S
kui ∇i (Ωe4 φ)kL2 (Su,u ) . a 2 , kui ∇i (eA φ)kL2 (Su,u ) . ,
i=0 i=0
|u|
4
X 1
ua 2 S
kui ∇i (Ωe3 φ − (Ωe3 φ)0 )kL2 (Su,u ) . .
i=0
|u|
Proof. To estimate Ωe4 φ, we first rewrite equation (2.8) as

1 1
(Ωe3 )(Ωe4 φ) + ΩtrχΩe4 φ = Ω2 ∆g φ − ΩtrχΩe3 φ+2Ω2 η A eA φ. (5.1)
2 2
Commuting the above equation with i angular derivatives, we then obtain
i+1
(Ω∇3 )∇i (Ωe4 φ) + Ωtrχ∇i (Ωe4 φ)
2
X 1
i1 i2 i3 2 2 A
= ∇ (η + η) ∇ Ω ∆g φ − ΩtrχΩe3 φ+2Ω η eA φ
i1 +i2 +i3 =i
2
X 2
+ ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ + )∇i4 (Ωe4 φ)
i1 +i2 +i3 +i4 =i
|u|
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (Ωe4 φ).
i1 +i2 +i3 +i4 =i−1
Note that the last term only appears when i ≥ 1.Using Proposition 3.6 with λ0 = i+1 2 , we have
i i i 1 2
that ku ∇ (Ωe4 φ)kL2 (Su,u ) is controlled by the ku · kLu L (Su,u ) norm of the right. We first
estimate
X
kui ∇i1 (η + η)i2 ∇i3 (Ω2 ∆g φ)kL1u L2 (Su,u )
i1 +i2 +i3 =i
X
≤kΩ u ∇i+1 (eA φ)kL1u L2 (Su,u ) +kΩ2 ui
2 i
∇i1 (η + η)i2 +1 ∇i3 ∆g φkL1u L2 (Su,u )
i1 +i2 +i3 =i−1
1 1
Ω ua 2 O ua 2 S 2
≤kΩui+1 ∇i+1 (eA φ)kL2u L2 (Su,u ) k kL2u L∞ (Su,u ) + · |u|
|u| |u| |u|3
1 1
u2 a2 S u2 aOS 1 1
≤ 1 + ≤ a 2 B− 2 .
|u| 2 |u|2
Here for the last line we use bootstrap assumption
1 1
ku5 ∇5 (Ωe4 φ)kL2u L2 (Su,u ) + kΩu5 ∇5 (eA φ)kL2u L2 (Su,u ) . u 2 a 2 S.
The terms including − 21 Ωtrχ · Ωe3 φ are lower order and they obey
X 1 ua 1
kui ∇i1 (η + η)i2 ∇i3 (− Ωtrχ · Ωe3 φ)kL1u L2 (Su,u ) ≤ ≤ a 2 B −1 ,
i +i +i =i
2 |u|
1 2 3
where to control ku ∇i (Ωtrχ) · Ωe3 φkL1u L2 (Su,u ) we employ improved estimates in Proposition
i
(4.3).
Using the schematic Codazzi equation, we can rewrite Ωβ = ∇ψ + ψ (Ωtrχ + ψ ). All the rest
1
terms are already estimated in Proposition 4.2 and they are bounded by a 2 . Hence we prove
1
kui ∇i (Ωe4 φ)kL∞ 2
u L (Su,u )
≤ a2 . (5.2)
To derive estimates for eA φ, we employ the equation
∇Ωe4 ∇A φ = eA (Ωe4 φ) − ΩχA B eB φ.
36 XINLIANG AN
Commuting this equation with i angular derivatives, it holds

X
ΩD4 (∇i ∇A φ) = ∇i1 (η + η)i2 ∇i3 ∇A (Ωe4 φ)
i1 +i2 +i3 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 ∇A φ
i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 ∇A φ.
i1 +i2 +i3 +i4 =i−1
Via Proposition 3.5, we then bound kui ∇i (eA φ)kL2 (Su,u ) by the kui ·kL1u L2 (Su,u ) norm of the right
hand side. With the aid of bootstrap assumption 2.40, the term with the top-order derivative
can be controlled by
1
kui ∇i+1 (Ωe4 φ)kL1u L2 (Su,u ) ≤kui+1 ∇i+1 (Ωe4 φ)kL2u L2 (Su,u ) k kL2u L∞ (Su,u )
|u|
1 1
1 1 u2 ua 2
≤u 2 a 2 S · = S.
|u| |u|
All the rest terms can be estimated as in Proposition 4.5. There is neither ∇i (Ωβ) nor ∇i+1 (Ωω̂).
1
And they obey the upper bound ua 2 /|u|. Thus, we arrive at
4
X 1
ua 2 S
kui ∇i (eA φ)kL∞ 2
u L (Su,u )
. .
i=0
|u|
To estimate (Ωe3 φ) − (Ωe3 φ)0 , using (2.9) and commutation formula, we get
(Ω∇4 )∇i (Ωe3 φ − (Ωe3 φ)0 )
X 1 1
= ∇i1 (η + η)i2 ∇i3 (Ω2 ∆g φ − ΩtrχΩe3 φ − ΩtrχΩe4 φ + 2Ω2 η A eA φ)
i1 +i2 +i3 =i
2 2
X X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 (Ωe3 φ) + ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (Ωe3 φ).
i1 +i2 +i3 +i4 =i i1 +i2 +i3 +i4 =i−1
i i
With Proposition 3.5, we control ku ∇ [(Ωe3 φ) − (Ωe3 φ)0 ]kL∞ 2
u L (Su,u )
by the kui · kL1u L2 (Su,u )
norm of the right. The term with the top-order derivative is bounded by
Ω2
kΩ2 ui ∇i+1 (eA φ)kL1u L2 (Su,u ) ≤kui+1 ∇i+1 (eA φ)kL2u L2 (Su,u ) k kL2 L∞ (Su,u )
|u| u
1 1 1 1
1 1 u2 a4 u2 2 ua 2 S
≤u 2 a 4 1 + 1 S · Ω ≤ .
|u| 2 Ω |u| |u|
Here for the second inequality, we use bootstrap assumption 2.40
1 1
u2 a4 1 1
ku5 ∇5 (eA φ)kL2u L2 (Su,u ) + kΩ−1 u5 ∇5 (Ωe3 φ)kL2u L2 (Su,u ) . 1 + 1 u 2 a 4 S.
|u| 2 Ω
Two borderline terms on the right are (Ωe3 φ)∇i (Ωχ) and (Ωtrχ)∇i (Ωe4 φ) with i = 4. And
they obey
1
i i ua 2
ku (Ωe3 φ)∇ (Ωχ)kL1u L2 (Su,u ) ≤Oi,2 (Ωχ) ,
|u|
1
i i ua 2
ku (Ωtrχ)∇ (Ωe4 φ)kL1u L2 (Su,u ) ≤Oi,2 (Ωe4 φ) .
|u|
Note that by Proposition 4.2, Proposition 4.3 and inequality (5.2), we have that Oi,2 (Ωχ),
Oi,2 (Ωe4 φ) depend only on initial data.
All the rest terms can be estimated as in Proposition 4.5. And they are
1 1 1 1
ua 2 ua 2 ua 2 B − 2
≤ · · (O + S) ≤ .
|u| |u| |u|
We hence arrive at
4
X 1
i i ua 2 S
ku ∇ (Ωe3 φ − (Ωe3 φ)0 )kL2 (Su,u ) . .
i=0
|u|

6. Energy Estimates for Scalar Field

We now move to derive the energy bounds for the top-order derivatives of the scalar field.
We first recall the below covariant decompositions12
1 1
DA eB = ∇A eB + χAB e3 + χAB e4 , (6.1)
2 2
D3 eA = ∇3 eA + ηA e3 , D4 eA = ∇4 eA + η A e4 , (6.2)
♯B ♯B
D A e 3 = χA eB + ζA e3 , DA e4 = χA eB − ζA e4 , (6.3)
♯A ♯A
D3 e4 = 2η eA + 2 ω e4 , D4 e3 = 2η eA + 2 ω̂ e3 , (6.4)
D3 e3 = −2ωe3 , D4 e4 = −2ω̂e4 . (6.5)
Next we start to use equation (2.8)
1 1
D3 D4 φ = D4 D3 φ = ∆g φ − trχe3 φ − trχe4 φ.
2 2
With the fact −2ω̂ = ∇4 log Ω = Ω−1 ∇4 Ω and the properties of covariant derivative
D4 D3 φ = D4 (D3 φ) − DD4 e3 φ = D4 (D3 φ) − 2η A eA (φ) + Ω−1 D4 Ω e3 φ,
we can write (2.8) as
1 1
D4 (ΩD3 φ) − 2Ωη A eA φ = ΩD4 D3 φ = ΩD3 D4 φ = Ω∆g φ − Ωtrχe3 φ − Ωtrχe4 φ.
2 2
This implies
1 1
ΩD4 (ΩD3 φ) = Ω2 ∇A ∇A φ − Ωtrχ · Ωe3 φ − Ωtrχ · Ωe4 φ + 2Ω2 η A eA φ. (6.6)
2 2
Similarly, via employing the fact −2ω = ∇3 log Ω = Ω−1 ∇3 Ω and using
D3 D4 φ =D3 (D4 φ) − DD3 e4 φ = D3 (D4 φ) − D2ηA eA +2ωe4 φ
=D3 (D4 φ) − 2η A eA φ − 2ωe4 φ = D3 (D4 φ) − 2η A eA φ + Ω−1 D3 Ωe4 φ,
we also obtain
1 1
ΩD3 (ΩD4 φ) = Ω2 ∇A ∇A φ − Ωtrχ · Ωe3 φ − Ωtrχ · Ωe4 φ + 2Ω2 η A eA φ. (6.7)
2 2
We then proceed to derive the desired form for the transport equation of ∇A φ. For f being a
/f to be the 1-form restricted to each surface Su,u of df (the differential
scalar function, we define d
of f ). Denote L, X being vector fields, we first have
/f )(X) = L((df
(LL d /f )[L, X] = L(Xf ) − [L, X]f = X(Lf ).
/ )(X)) − (d
∂
Letting f = φ, L = ∂u and X = eA , we hence obtain
∂
(L ∂ /dφ)A = d
/A ( φ). (6.8)
∂u ∂u
For (L ∂ /
dφ)A , we also have
∂u
∂ ∂ ∂ ∂
(L ∂ /dφ)A = (eA φ)−[ , eA ]φ = (eA φ)−(D ∂ eA ) φ+(DeA ) φ = D ∂ DeA φ+DeA (Ωe3 )φ.
∂u ∂u ∂u ∂u ∂u ∂u ∂u
(6.9)
12The decompositions are the same as in [21].

38 XINLIANG AN
For the last term, it holds

DeA (Ωe3 )φ =Ω−1 eA (Ω)(Ωe3 φ) + ΩDeA e3 φ = eA (log Ω)(Ωe3 φ) + Ω(χA B eB φ + ζA e3 φ)
(6.10)
=ηA (Ωe3 φ) + ΩχA B eB φ.
Back to (6.9), employing (6.8) and (6.10), we get
∂
D ∂ φ) − DeA (Ωe3 ) φ
DeA φ =eA (
∂u ∂u (6.11)
∂
=eA ( φ) − ηA (Ωe3 φ) − ΩχA B eB φ.
∂u
At the same time, employing properties of the covariant derivative and the fact D3 eA = ∇3 eA +
ηA e3 , we also have
∂ ∂
D ∂ DeA φ = (eA φ) − DD ∂ e φ = (eA φ) − D∇Ωe3 eA +ΩηA e3 φ
∂u ∂u ∂u A ∂u (6.12)
=∇Ωe3 eA φ − ΩηA e3 φ.
Comparing (6.11) with (6.12) and noting ∂/∂u = Ωe3 , we hence obtain
∇Ωe3 ∇A φ = eA (Ωe3 φ) − ΩχA B eB φ. (6.13)
In the same fashion, similar to (6.10), we get
(DeA Ωe4 )φ = eA (log Ω)Ωe4 φ + Ω(χA B eB − ζA e4 )φ = η A (Ωe4 φ) + ΩχA B eB φ.
This implies
DeA DΩe4 φ = eA (Ωe4 φ) − (DeA Ωe4 ) φ = eA (Ωe4 φ) − η A (Ωe4 φ) − ΩχA B eB φ. (6.14)
On the other side, we also have
DΩe4 DeA φ =Ωe4 (eA φ) − DDΩe4 eA φ = Ωe4 (eA φ) − D∇Ωe4 eA +Ωη e4 φ
A
(6.15)
=∇Ωe4 eA φ − Ωη A e4 φ.
Employing DeA DΩe4 φ = DΩe4 DeA φ and collection both (6.14) and (6.15), we then arrive at
∇Ωe4 ∇A φ = eA (Ωe4 φ) − ΩχA B eB φ. (6.16)
In the rest of this section, to derive energy estimates for the scalar field, we rearrange (6.6),
(6.7), (6.13), (6.16) as the below paired equations
1
∇Ωe4 (Ω∇A )φ = ΩeA (Ωe4 φ) − Ω2 χ̂A B ∇B φ − Ω2 trχ∇A φ + Ωe4 (Ω)∇A φ, (6.17)
2
1 1
Ωe3 (Ωe4 φ) = Ω2 ∇A ∇A φ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω2 η A · eA φ, (6.18)
2 2
and
1
∇Ωe3 ∇A φ + Ωtrχ∇A φ = eA (Ωe3 φ) − Ωχ̂A B ∇B φ, (6.19)
2
1 1
Ωe4 (Ωe3 φ) = Ω2 ∇A ∇A φ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω2 η A · eA φ. (6.20)
2 2
For the first pair, on both sides multiplying Ω∇A φ for the first equation and multiplying Ωe4 φ
for the second equation, we get
(Ω∇A φ)(Ωe4 )(Ω∇A φ) =Ω2 (∇A φ)eA (Ωe4 φ) + Ω∇A φ∇A φ(Ωe4 Ω)
1
+ (Ω∇A φ) · − Ω2 χ̂A B ∇B φ − Ω2 trχ∇A φ ,
2
2 A 1 1 2 A
(Ωe4 φ)Ωe3 (Ωe4 φ) = (Ωe4 φ)Ω ∇ ∇A φ + (Ωe4 φ) · − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω η · eA φ .
2 2
(6.21)
For the second pair, multiplying ∇A φ for the first equation and multiplying Ω−2 (Ωe3 φ) for the
second equation, it holds
1
∇A φ∇Ωe3 ∇A φ + Ωtrχ∇A φ∇A φ = ∇A φeA (Ωe3 φ) − Ωχ̂A B ∇B φ∇A φ,
2
Ω−2 (Ωe3 φ)Ωe4 (Ωe3 φ) =Ω−2 (Ωe3 φ)Ω2 ∇A ∇A φ

1 1
+ Ω−2 (Ωe3 φ) · − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω2 η A · eA φ .
2 2
(6.22)
If we add the two equations in (6.21) and integrate on Su,u , we will encounter the potential
top-order-derivative terms
Z
Ω2 (∇A φ)eA (Ωe4 φ) + (Ωe4 φ)Ω2 ∇A ∇A φ. (6.23)
Su,u
Repeating the same procedure, for (6.22) we will have

Z
∇A φ eA (Ωe3 φ) + Ω−2 (Ωe3 φ) Ω2 ∇A ∇A φ. (6.24)
Su,u
Conducting integration by parts with respect to ∇A for both (6.23) and (6.24), the potential
top-order-derivative terms will be canceled.
We then recall two integration by parts formulas in [2]. Define

Z Z u Z Z Z u Z
φ :=2 ( Ωφ)du′ , φ := 2 ( Ωφ)du′
Hu 0 Su,u′ Hu −1 Su′ ,u
Z Z u Z u Z (6.25)
2 ′ ′
φ := 2 ( Ω φ)du du .
Du,u 0 −1 Su′ ,u′
Applying φ1 = Ω−1 (1) φ and φ2 = (2) φ in Proposition 8.1 of [2], we get
Proposition 6.1. Suppose (1) φ and (2) φ are r tensorfields, then it holds
Z Z
Ω−1 (1) φ∇4 (2) φ + Ω−1 (2) φ∇4 (1) φ
Du,u Du,u
Z Z Z
= Ω−1 (1) φ(2) φ − Ω−1 (1) φ(2) φ − Ω−1 trχ(1) φ(2) φ.
H u (−1,u) H 0 (−1,u) Du,u
(1)
Set φ = (2) φ = φ in the above proposition. Together with (6.25) we then obtain
Proposition 6.2. Suppose φ is a r tensorfield, then it holds
Z uZ Z uZ Z uZ u Z Z u Z u Z
2 2
|φ| = |φ| + 2φ(Ω∇4 φ) + Ωtrχφ2 .
−1 Su′ ,u −1 Su′ ,0 0 −1 Su′ ,u′ 0 −1 Su′ ,u′
By Proposition 8.2 in [2], we also have
Proposition 6.3. For an r tensorfield (1) φ and an r − 1 tensorfield (2) φ, they satisfy
Z Z
(1) A1 A2 ...Ar
φ ∇Ar (2) 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
At the same time, it also holds

40 XINLIANG AN
Proposition 6.4. Let φ be an r tensorfield and λ1 = 2(λ0 − 21 ). Then we get

Z uZ Z uZ Z uZ u Z
2λ1 2 2
|u| |φ| = |φ| + 2|u′ |2λ1 < φ, Ω∇3 φ + λ0 Ωtrχφ >
0 Su,u′ 0 S−1,u′ 0 −1 Su′ ,u′
Z uZ u Z
2
+ |u′ |2λ1 (1 − 2λ0 )(Ωtrχ + )|φ|2 .
0 −1 Su′ ,u′ |u′ |
Proof. Using the first variation of area formula, it holds
Z Z
d
( |u|2λ1 |φ|2 ) = 2|u|2λ1 < φ, Ω∇3 φ + λ0 Ωtrχφ >
du Su,u Su,u
Z
2λ1 · (−Ωe3 u)
+ |u|2λ1 + (1 − 2λ0 )Ωtrχ |φ|2 .
Su,u |u|
We then choose λ1 = 2λ0 − 1, which satisfies
2λ1 (−Ωe3 u) 2
+ (1 − 2λ0 )Ωtrχ = (1 − 2λ0 )(Ωtrχ + ).
|u| |u|
By integrating with respect to du du and applying the fundamental theorem of calculus in u, we
then prove this proposition.
We now proceed to prove
(2.39), (2.40), (2.41), it holds
1 1
ku5 ∇5 (Ωe4 φ)kL2u L2 (Su,u ) + kΩu5 ∇5 (eA φ)kL2u L2 (Su,u ) . u 2 a 2 . (6.26)
Proof. Taking equation (6.18)
1 1
Ωe3 (Ωe4 φ) = Ω2 ∇A ∇A φ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω2 η A · eA φ,
2 2
and commuting it with i angular derivatives, we obtain
i+1
Ω∇3 ∇i (Ωe4 φ) + Ωtrχ∇i (Ωe4 φ)
2
X 1
= ∇i1 (η + η)i2 ∇i3 Ω2 ∆g φ − Ωtrχ · Ωe3 φ + 2Ω2 η A eA φ
i1 +i2 +i3 =i
2
X
+ ∇i1 (η + η)i2 ∇i3 (Ωtrχ)∇i4 (Ωe4 φ)
i1 +i2 +i3 +i4 =i, (6.27)
i4 <i
X 2
+ ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ + )∇i4 (Ωe4 φ)
i1 +i2 +i3 +i4 =i
|u|
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (Ωe4 φ).
i1 +i2 +i3 +i4 =i−1
Note that the term with the top-order derivative on the right is Ω2 ∇i ∆g φ.
Similarly, commuting (6.17) with i angular derivatives, we obtain
X
Ω∇4 ∇i (Ω∇A )φ = ∇i1 (η + η)i2 ∇i3 ΩeA (Ωe4 φ) − Ω2 χA B eB φ + Ωe4 (Ω)∇A φ
i1 +i2 +i3 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 (Ω∇A )φ
i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (Ω∇A )φ.
i1 +i2 +i3 +i4 =i−1
(6.28)
i
Notice that the term with the top-order derivative on the right is ∇ [ΩeA (Ωe4 φ)].
We now consider
Z uZ uZ
i+1
u2i ∇A1 ···Ai (Ωe4 φ)[Ω∇3 ∇A1 ···Ai (Ωe4 φ) + Ωtrχ∇A1 ···Ai (Ωe4 φ)]
0 u0 Su,u 2
Z uZ uZ (6.29)
+ u2i ∇A1 ···Ai (Ω∇A )φ[Ω∇4 ∇A1 ···Ai (Ω∇A )φ].
0 u0 Su,u
Employing (6.27) and (6.28), on the right we encounter

Z
Ω2 u2i ∇A1 ···Ai (Ωe4 φ)∇A1 ···Ai ∆g φ + Ω2 u2i ∇A1 ···Ai ∇A φ∇A1 ···Ai ∇A (Ωe4 φ).
Su,u
Employing integration by parts with respect to ∇A , we then cancel the top-order derivatives
∇i+1 (∇φ), ∇i+1 (Ωe4 φ). Applying Proposition 6.2 for (6.29), together with (6.27) and (6.28) we
arrive at
kui ∇i (Ωe4 φ)k2L2u L2 (Su,u ) + kΩui ∇i (eA φ)k2L2u L2 (Su,u )

X
≤kui ∇i (Ωe4 φ)k2L2u L2 (S−1,u ) + ku2i Ω2 ∇i1 (η + η)i2 +1 ∇i3 ∆g φ∇i (Ωe4 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 =i−1
X 1
+ ku2i ∇i1 (η + η)i2 ∇i3 − Ωtrχ · Ωe3 φ + 2Ω2 η A eA φ ∇i (Ωe4 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 =i
2
X
2i
+ ku ∇i1 (η + η)i2 ∇i3 (Ωtrχ)∇i4 (Ωe4 φ)∇i (Ωe4 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i,
i4 <i
X
+ ku2i ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 (Ωe4 φ)∇i (Ωe4 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i
X
2i
+ ku ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (Ωe4 φ)∇i (Ωe4 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i−1
X
+ kΩ2 u2i ∇i1 (η + η)i2 +1 ∇i3 eA (Ωe4 φ) ∇i ∇A φkL1u L1u L1 (Su,u )
i1 +i2 +i3 =i−1
X
+ kΩ2 u2i ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 ∇A φ∇i ∇A φkL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i
X
2 2i
+ kΩ u ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 ∇A φ∇i ∇A φkL1u L1u L1 (Su,u ) .
i1 +i2 +i3 +i4 =i−1
(6.30)
We then bound each of the terms on the right. For the initial data term, we have
1
a 2 12
kui ∇i (Ωe4 φ)k2L2u L2 (S−1,u ) ≤ ( u |u|)2 = ua.
|u|
The terms with top-order derivatives (when i = 5) can be controlled as follows: we first
bound the terms involving Ωe3 φ and get
ku10 (Ωtrχ)∇5 (Ωe3 φ)∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ(Ωtrχ)kL2u L2u L∞ (Su,u )
1 1 1 1 1 1 1
1 1 u2 a4 1 1 a2 1 1 u2 a4 u2 a4 2 1
≤u 2 a 4 1 + 1 S · u 2 a2 S · |u| 2 u2 O =
1 1 + 1 S O · ua ≤ uaB − 4 ,
|u| 2 Ω |u| |u| 2 |u| 2 Ω
ku10 (Ωe3 φ)∇5 (Ωtrχ)∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )

≤ku6 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
ku−1 (Ωe3 φ)kL∞ 1 ∞
u Lu L (Su,u )
1
3 1 1 1 ua 2 ÕS 1
≤u 2 aÕ · u 2 a 2 S · = ua ≤ uaB − 2 .
|u| |u|
42 XINLIANG AN
For the top-order terms with η, we have

kΩ2 u10 (η, η)∇5 (eA φ)∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ2 (η, η)kL∞ 1 ∞
u Lu L (Su,u )
1 1 1 1 1 1
1 1 u2 a4 1 1 ua 2 ua 4 u2 a4 2
≤u 2 a 4 1 + 1 S · u2 a2 S · 2
|u|O = ua 1 + 1 S O ≤ uaB −1 ,
|u| 2 Ω |u| |u| |u| 2 Ω
kΩ2 u10 eA φ∇5 η∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ2 eA φkL∞ 1 ∞
u Lu L (Su,u )
1 1
ua 2 ua 2
1 1 11 1
≤Ω−1 u 2 a 2 Õ · u 2 a 2 S ·
O= ÕSO · ua ≤ uaB − 2 .
|u| |u|Ω
For the top-order terms containing Ωχ̂, Ωtrχ − (Ωtrχ)0 , they can be controlled via
ku10 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇5 (Ωe4 φ)∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
k(Ωχ̂, Ωtrχ − (Ωtrχ)0 )kL∞ 1 ∞
u Lu L (Su,u )
1
1 1 ua 2 1 1 1
≤u a S · u a S ·
2 2 O ≤ uaB − 2 ,
2 2
|u|
ku10 (Ωe4 φ)∇5 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩe4 φkL2u L2u L∞ (Su,u )
1 1 1 1 1 1 1
1 u2 a4
1 1 1 a2 1 1 u 2 a 4 ÕSO u2 a4 1
≤u 2 a 4 1 + 1 Õ · u 2 a2 S · |u| 2 u2 O =
1 ua 1 + 1 ≤ uaB − 4 .
|u| 2 Ω |u| |u| 2 |u| 2 Ω
The top-order terms involving Ωχ obey
kΩ2 u10 (Ωχ)∇5 (eA φ)∇5 (eA φ)kL1u L1u L1 (Su,u )
≤kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩχkL∞ 1 ∞
u Lu L (Su,u )
1
1 1 1 1 a2 1
≤u 2 a 2 S · u 2 a 2 S · uO ≤ uaB − 2 ,
|u|
ku10 Ω2 eA φ∇5 (Ωχ)∇5 (eA φ)kL1u L1u L1 (Su,u )
≤ku10 Ω2 eA φ∇5 (Ωχ̂)∇5 (eA φ)kL1u L1u L1 (Su,u ) + ku10 Ω2 eA φ∇5 (Ωtrχ)∇5 (eA φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩeA φkL2u L2u L∞ (Su,u )
+ ku6 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩu−1 eA φkL2u L2u L∞ (Su,u )
1 1
1 1ua 2 1 1 1 3 1 1 1 ua 2 1 1
≤u a Õ · u a S ·
2 2 |u| 2 u 2 O + u 2 aÕ′ · u 2 a 2 S ·
2 2 |u| 2 u 2 O
|u|2 |u|3
3 1 5
u2 a2 u 2 aÕ′ SO
= 3 ua ≤ uaB −1 .
ÕSO · ua +
5
|u| 2 |u| 2
Finally, the top-order terms with β and β can be bounded via
ku10 (Ωe4 φ)∇4 (Ωβ)∇5 (Ωe4 φ)kL1u L1u L1 (Su,u )
≤kΩ−1 u6 ∇4 (Ωβ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩu−1 (Ωe4 φ)kL2u L2u L∞ (Su,u )
1 1 1 3 3 1 1
3 3 u2 a4 1 1 a2 1 1 u 2 a 4 RSO u2 a4
≤u a 2 4 1+ 1 R · u a S · 2 |u| u O =
2 2 2 2
3 1+ 1 ua ≤ uaB −1 ,
|u| Ω
2 |u| |u| 2 |u| Ω
2
kΩ2 u10 eA φ∇4 (Ωβ)∇5 (eA φ)kL1u L1u L1 (Su,u )

≤ku5 ∇4 (Ωβ)kL∞ 2 2
u Lu L (Su,u )
kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
1 3 1
1 ua 2
1 1 1 1 u 2 a 2 RSO
1
≤u a R · u a S ·
2 2 |u| 2 u 2 O =
2 2
3 ua ≤ uaB −1 .
|u|2 |u| 2
It is a straight-forward check that the non-top-order-derivative terms in (6.30) obey the same
(and even smaller) upper bounds, compared with the estimates above. There is no borderline
term and their upper bounds are much smaller than ua. Putting all the estimates together, we
then finish the proof of this current proposition.
Next, we move to establish the energy estimate for the other pair and we have
(2.39), (2.40), (2.41), it holds
1 1
5 5 −1 5 5 u2 a4 1 1
ku ∇ (eA φ)kL2u L2 (Su,u ) + kΩ u ∇ (Ωe3 φ)kL2u L2 (Su,u ) . 1+ 1 u2 a4 .
|u| 2 Ω
In addition, we also have the improved estimate
1 1 1
u2 a4 1 1 ua 2 1 1 1
ku5 ∇5 (eA φ)kL2u L2 (Su,u ) + kΩ−1 u5 ∇5 (Ωe3 φ)kL2u L2 (Su,u ) . 1 · u2 a4 + ( 3 )4 · u2 a4 .
|u| 2 Ω |u|Ω 2
Proof. Employing the paired equations (6.19), (6.20)
1
∇Ωe3 ∇A φ + Ωtrχ∇A φ =eA (Ωe3 φ) − Ωχ̂A B ∇B φ,
2
1 1
∇Ωe4 [(Ωe3 φ) − (Ωe3 φ)0 ] =Ω2 ∆φ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω2 η A · eA φ.
2 2
and commutating them with i angular derivatives, for i ≥ 1 we arrive at
i+1
(Ω∇3 )∇i (eA φ) + Ωtrχ∇i (eA φ)
2
X
= ∇i1 (η + η)i2 ∇i3 ∇A (Ω∇3 φ) − Ωχ̂A B ∇B φ
i1 +i2 +i3 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωtrχ)∇i4 (eA φ)
i1 +i2 +i3 +i4 =i,
i4 <i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 (eA φ)
i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (eA φ),
i1 +i2 +i3 +i4 =i−1
i
(Ω∇4 )∇ (Ωe3 φ)
=(Ω∇4 )∇i [(Ωe3 φ) − (Ωe3 φ)0 ]
X
i1 i2 i3 2 1 1 2 A
= ∇ (η + η) ∇ Ω ∆g φ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω η eA φ
i1 +i2 +i3 =i
2 2
X
+ ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 [(Ωe3 φ) − (Ωe3 φ)0 ]
i1 +i2 +i3 +i4 =i
X
+ ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 [(Ωe3 φ) − (Ωe3 φ)0 ].
i1 +i2 +i3 +i4 =i−1
With i ≥ 1, we then consider
Z uZ u Z
i+1
u2i ∇A1 ···Ai (eA φ)[Ω∇3 ∇A1 ···Ai (eA φ) + Ωtrχ∇A1 ···Ai (eA φ)]
0 −1 Su,u 2
Z uZ u Z
+ u2i ∇A1 ···Ai [(Ωe3 φ) − (Ωe3 φ)0 ]{Ω−2 · Ω∇4 ∇A1 ···Ai [(Ωe3 φ) − (Ωe3 φ)0 ]}.
0 −1 Su,u
Using (6.19) and (6.20) into the above expression, we see that the terms involving the top-order
derivatives are of the form
Z
u2i ∇A1 ···Ai (eA φ)∇A1 ···Ai [eA (Ωe3 φ)] + u2i Ω−2 Ω2 ∇A1 ···Ai (Ωe3 φ)∇A1 ···Ai (∆g φ).
Su,u
44 XINLIANG AN
Employing integration by parts with respect to ∇A , we notice that the top-order-derivative

terms ∇i+1 (Ωe3 φ) and ∇i+1 (eA φ) are cancelled. Carrying out the details of this approach, we
get
kui ∇i (eA φ)k2L2u L2 (Su,u ) + kΩ−1 ui ∇i (Ωe3 φ)k2L2u L2 (Su,u )

X
≤kui ∇i (eA φ)k2L2u L2 (S−1,u ) + ku2i ∇i1 (η + η)i2 +1 ∇i3 ∇A (Ωe3 φ)∇i (eA φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 =i−1
X
+ ku2i ∇i1 (η + η)i2 ∇i3 (Ωχ̂A B ∇B φ)∇i (eA φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 =i
X
2i
+ ku ∇i1 (η + η)i2 ∇i3 (Ωtrχ)∇i4 (eA φ)∇i (eA φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i, i4 <i
X
2i
+ ku ∇i1 (η + η)i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 (eA φ)∇i (eA φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i
X
2i
+ ku ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 (eA φ)∇i (eA φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i−1
X
−2
+ kΩ Ω2 u2i ∇i1 (η + η)i2 +1 ∇i3 ∆g φ∇i (Ωe3 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 =i−1
X
−2 2i
+ kΩ u ∇i1 (η + η)i2
i1 +i2 +i3 =i

i3 1 1
×∇ − Ωtrχ · Ωe4 φ − Ωtrχ · Ωe3 φ + 2Ω η eA φ ∇i (Ωe3 φ)kL1u L1u L1 (Su,u )
2 A
2 2
X
+ kΩ−2 u2i ∇i1 (η + η)i2 ∇i3 (Ωχ)∇i4 [(Ωe3 φ) − (Ωe3 φ)0 ]∇i (Ωe3 φ)kL1u L1u L1 (Su,u )
i1 +i2 +i3 +i4 =i
X
−2 2i
+ kΩ u ∇i1 (η + η)i2 ∇i3 (Ωβ)∇i4 [(Ωe3 φ) − (Ωe3 φ)0 ]∇i (Ωe3 φ)kL1u L1u L1 (Su,u ) .
i1 +i2 +i3 +i4 =i−1
(6.31)
We now derive bound for terms on the right. The initial-data term obeys
1 3 1 1 1
ua 2 1 u2 a2 2 ua 4 1 1 2 1 u2 a 2
kui ∇i (eA φ)k2L2u L2 (S−1,u ) ≤ ( 2
u 2 |u|)2 = ( ) =( u 2 a 4 ) ≤ ua 2 · .
|u| |u| |u| |u|2
The terms with top-order derivatives (when i = 5) are controlled as below: we first bound
the 4 terms involving ∇5 (eA φ) and they satisfy
ku10 (η, η)∇5 (Ωe3 φ)∇5 (eA φ))kL1u L1u L1 (Su,u )
≤kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩ(η, η)kL2u L2u L∞ (Su,u )
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u 2 a 4 · 1 + 1 S · u 2 a4 · 1 +
1 S · u 2 |u| 2 O ≤ ua 2 · ,
|u| 2 Ω |u| 2 Ω |u|2 |u|Ω
ku10 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇5 (eA φ)∇5 (eA φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩχ̂, Ωtrχ − (Ωtrχ)0 kL1u L∞
u L (Su,u )
∞
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 ua 2 1 ua 2 1
≤u 2 a 4 · 1 + 1 S · u2 a4 · 1 + 1 S· O ≤ ua 2 · ( 3 ) ,
2
|u| 2 Ω |u| 2 Ω |u| |u|Ω 2
ku10 (eA φ)∇5 (Ωtrχ, Ωχ̂)∇5 (eA φ)kL1u L1u L1 (Su,u )

≤ku5 ∇5 (Ωtrχ, Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
keA φkL2u L2u L∞ (Su,u )
1
1 11 1 1
1 1 u2 a4
u2 a4 ua 2 1 111 1 ua 2
≤u a · 1 + 2
1 Õ · u a · 1 +
4
1 S· |u| u O ≤ ua ·
2 242 2 ,
|u| 2 Ω |u| 2 Ω |u|2 |u|Ω
ku10 (eA φ)∇4 (Ωβ)∇5 (eA φ)kL1u L1u L1 (Su,u )

≤ku6 Ω−1 ∇4 (Ωβ)kL∞ 2 2
u Lu L (Su,u )
ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
ku−1 ΩeA φkL2u L2u L∞ (Su,u )
1 1 1 1 1 1
3 3 u2 a4 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u 2 a 4 1 + 1 R · u 2 a4 · 1 +
1 S · |u| 2 u 2 O ≤ ua 2 · .
|u| 2 Ω |u| 2 Ω |u|3 |u|Ω
For top-order-derivative terms containing ∇5 (Ωe3 φ) and Ωχ, we have
kΩ−2 u10 (Ωχ)∇5 (Ωe3 φ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩχkL1uL∞
u L (Su,u )
∞
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 a2 1 ua 2 1
≤u 2 a 4 1 + 1 S · u 2 a4 1 +
1 S · uO ≤ ua 2 · ( )2,
|u| 2 Ω |u| 2 Ω |u| |u|Ω
kΩ−2 u10 (Ωe3 φ)∇5 (Ωtrχ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )

≤ku6 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u−1 (Ωe3 φ)kL2u L2u L∞ (Su,u )
1 1 1
3 1 1 u2 a4 1 1 1 1 ua 2
≤u 2 aÕ′ · u 2 a 4 S · 1 + 1 · 2
|u| 2 u 2 ≤ ua 2 · .
|u| 2 Ω |u| Ω |u|Ω
Noting that in above inequality, we employ the crucial estimates via Õ′ .
To deal with the top-order-derivative terms involving both ∇5 (Ωe3 φ) and Ωtrχ, we encounter
the borderline estimate. Using the derived Proposition 6.5, we have
kΩ−2 u10 (Ωtrχ)∇5 (Ωe4 φ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )

≤ku5 ∇5 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 (Ωtrχ)kL2u L2u L∞ (Su,u )
1 1 1 1 1
≤u 2 a 2 · |u| 2 u 2 · kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
|u|Ω
1 1
1 u2 a4 2 1
≤c0 kΩ−1 u5 ∇5 (Ωe3 φ)k2L∞ 2 2
u Lu L (Su,u )
+ ( 1 ) · ua 2 .
c0 |u| 2 Ω
By choosing c0 to be suitably small, the term c0 kΩ−1 u5 ∇5 (Ωe3 φ)k2L∞ L2 L2 (Su,u ) can be absorbed
u u
to the left of (6.31).
The other top-order-derivative terms with Ωtrχ satisfies a better upper bound
kΩ−2 u10 (Ωe4 φ)∇5 (Ωtrχ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 (Ωe4 φ)kL1u L∞
u L (Su,u )
∞
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 a2 1 ua 2 1
≤u 2 a 4 1 + 1 Õ · u 2 a4 1 +
1 S · uO ≤ ua 2 · (
3 ) .
2
|u| 2 Ω |u| 2 Ω |u|Ω |u|Ω 2
For the top-order-derivative terms containing ∇5 (Ωe3 φ) and η, we have
ku10 η A ∇5 (eA φ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩη A kL2u L2u L∞ (Su,u )
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u 2 a 4 1 + 1 S · u2 a4 1 + 1 S· 2
|u| 2 u 2 O ≤ ua 2 · ,
|u| 2 Ω |u| 2 Ω |u| |u|Ω
ku10 (eA φ)∇5 η∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )

≤ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
1 1 1 1 1 1
1 1 u2 a4 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u a 2 4 1+ 1 Õ · u a 1 +
2 4
1 S· |u| u O ≤ ua ·
2 2 2 .
|u| 2 Ω |u| 2 Ω |u|2 |u|Ω
46 XINLIANG AN
The top-order-derivative terms involving ∇5 (Ωe3 φ) and Ωe3 φ − (Ωe3 φ)0 obey
kΩ−2 u10 [(Ωe3 φ) − (Ωe3 φ)0 ]∇5 (Ωtrχ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
× kΩ−1 [(Ωe3 φ) − (Ωe3 φ)0 ]kL2u L2u L∞ (Su,u )

1 1 1 1
1 1 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u 2 a 2 Õ · u 2 a 4 1 + 1 S · 2 |u| 2 u 2 O ≤ ua 2 · ,
|u| 2 Ω |u| Ω |u|Ω
kΩ−2 u10 [(Ωe3 φ) − (Ωe3 φ)0 ]∇5 (Ωχ̂)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇5 (Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 [(Ωe3 φ) − (Ωe3 φ)0 ]kL2u L2u L∞ (Su,u )
1 1 1 1
1 1 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u a Õ · u a
2 2 2 4 1+ 1 S · 2 |u| u O ≤ ua ·
2 2 2 .
|u| 2 Ω |u| Ω |u|Ω
And the top-order-derivative term with ∇5 (Ωe3 φ) and ∇4 (Ωβ) obey
kΩ−2 u10 [(Ωe3 φ) − (Ωe3 φ)0 ]∇4 (Ωβ)∇5 (Ωe3 φ)kL1u L1u L1 (Su,u )
≤ku5 ∇4 (Ωβ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
× kΩ−1 [(Ωe3 φ) − (Ωe3 φ)0 ]kL2u L2u L∞ (Su,u )

1 1 1 1
1 1 1 1 u2 a4 ua 2 1 1 1 ua 2
≤u 2 a 2 R · u 2 a 4 1 + 1 S · |u| 2 u 2 O ≤ ua 2 · .
|u| 2 Ω |u|2 Ω |u|Ω
We hence bound all the top-order-derivative terms. It is a straight-forward check that compared
with the upper bounds of the obtained estimates above, the rest terms obey the same (and even
smaller) upper bounds. There is no other borderline term and their upper bounds are much
1
smaller than ua 2 . Gathering all these estimates, we finish the proof of this current proposition.

7. Elliptic estimates for the fifth derivatives of the Ricci coefficients

We now move to bound the fifth angular derivatives for the Ricci coefficients. We start from
the estimates for ω̂.
(2.39), (2.40), (2.41), it holds
1 1
ku5 ∇5 (Ωω̂)kL∞ 2 2
u Lu L (Su,u )
. u 2 a 2 [1 + R(β)].
Proof. We first construct an auxiliary function ω † , which satisfies

1
∇3 ω † = σ̌
2
and has zero initial data on H−1 . We further let
1 1
κ := ∇ω̂ +∗ ∇ω † − (β − ∇4 φ∇φ).
2 2
Using Proposition 3.6 and bootstrap assumption (2.39), we have
X 1
u2 a2
1
i i † i+1 i −1
ku ∇ ω kL∞ ∞ 2
u Lu L (Su,u )
≤kΩu ∇ σ̌kL∞ 2 2
u Lu L (Su,u )
ku kL∞ 2 ∞
u Lu L (Su,u )
+ 1
i≤4
|u| 2
1 1 1 1
u2 a2 R u2 a2 1 1
≤ 1 + 1 ≤ a 2 B− 4 .
|u| 2 |u| 2
This also implies X 1 1 1
kui ∇i (Ωω † )kL∞ ∞ 2
u Lu L (Su,u )
≤ a 2 B− 4 ≤ a 2 .
i≤4
Note that in above inequality Ωω † obeys the same bounds as for ψ . Hence we regard Ωω † as
one of the ψ terms and we employ the notation ψ ∈ {Ωtrχ, Ωχ̂, Ωω̂, Ωω † }. Using this notation,
we write the transport equation for Ωκ as
1
Ω∇3 (Ωκ) + Ωtrχ(Ωκ)
2
X 2 X
= ψ i1 ∇i2 (Ωtrχ + , Ωχ̂) · ∇i3 ψ + Ω2 ψ i1 +1 ∇i2 (η, η, eA φ)
i +i +i =1
|u| i +i =1
1 2 3 1 2
1 Ω 1
+ ψψ + β + ψ K + ∇(Ωe3 φ)ψ + ∇(Ωe4 φ).
|u| |u| |u|
Commuting
P with angular derivatives for 4 times and employing the schematic Codazzi equation
Ωβ = i1 +i2 =1 ψ i1 ∇i2 ψ , we obtain
5
Ω∇3 ∇4 (Ωκ) + Ωtrχ∇4 (Ωκ)
2
X 2 X
i1 i2 i3
= ∇ ψ ∇ (Ωtrχ + , Ωχ̂) · ∇i4 ψ + Ω2 ∇i1 ψ i2 +1 ∇i3 (η, η, eA φ)
i +i +i +i =5
|u| i
1 2 3 4 1 +i2 +i3 =5
i1 ≤4
1 4 1 X X
+ ∇ (Ωβ) + ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 +1 ψ i2 +1 ∇i3 ψ
|u| |u| i
1 +i2 +i3 =4 i1 +i2 +i3 =4
i1 ≤3
X 1 5
+ ∇i1 ψ i2 ∇i3 ψ ∇i4 K+ψ ∇5 (Ωe3 φ) + ∇ (Ωe4 φ).
i1 +i2 +i3 +i4 =4
|u|
Employing Proposition 3.6 with λ0 = 52 , we then bound the ku4 ∇4 (Ωκ)kL2u L∞ 2

u L (Su,u )
norm by
4
the ku · kL2u L1u L2 (Su,u ) norm of the right.
In the first term, we encounter u4 ψ ∇5 (Ωtrχ, Ωχ̂) and it satisfies
ku4 ψ ∇5 (Ωtrχ, Ωχ̂)kL2u L1u L2 (Su,u ) ≤ku5 ∇5 (Ωtrχ, Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
· ku−1 ψ kL2u L2u L∞ (Su,u )
1 1 1 1 1 1 1
1 1 u2 a4 u2 a2 u 2 a 2 B− 4
≤u 2 a 4 1 + 1 R· 3 O ≤ .
|u| 2 Ω |u| 2 |u|
The rest of the first term obey
2 X
.ku4 (Ωtrχ + , Ωχ̂)∇5 ψ kL2u L1u L2 (Su,u ) + ku4 ∇i1 ψ i2 +1 ∇i3 ψ kL2u L1u L2 (Su,u )
|u| i 1 +i2 +i3 =5
i1 ,i3 ≤4
.ku2 ψ kL∞ u Lu L (Su,u )

∞ ∞ ku−3 kL1u
3
u 2 aO2
× ku5 ∇5 (Ωtrχ, Ωχ̂, Ωω̂)kL∞ 2 2
u Lu L (Su,u )
+ ku 5 5
∇ (Ωω †
)k 2 2
Lu Lu L (Su,u )
∞ +
|u|2
1 3
ua 2 O 1 1 u 2 aO2
. · u 2 a 2 Õ + ku5 ∇5 (Ωω † )k ∞ 2 2 +
L L L (S )
|u|2 u u u,u
|u|2
1 1 1 1
u 2 a 2 B− 2 B− 2 5 5
≤ + ku ∇ (Ωω † )kL∞ 2 2
u Lu L (Su,u )
.
|u| |u|
In the second term, there is Ω2 u4 ψ ∇5 (η, η, eA φ) and it obeys
kΩ2 u4 ψ ∇5 (η, η, eA φ)kL2u L1u L2 (Su,u )
≤ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
kΩ2 u−1 ψ kL1u L∞
u L (Su,u )
∞ + kΩu5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
kΩu−1 ψ kL2u L2u L∞ (Su,u )
+ kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
1 1 1 1 1 1
1 1 u2 a4 ua 2 1 1 ua 2 O 1 1 u 2 a 2 B −1
≤u a2 4 1+ 1 · Õ · O + u 2 a 2 (Õ + S) · u 2 |u| 2 ≤ .
|u| 2 Ω |u|2 |u|3 |u|
48 XINLIANG AN
The rest of the second term can be bounded by

X 3
u 2 aO2
1 1 1
u 2 a 2 B− 2
2 4 i1 i2 +1 i3
kΩ u ∇ ψ ∇ ψ kL2u L1u L2 (Su,u ) . ≤ .
i +i +i =5
|u|2 |u|
1 2 3
i1 ,i3 ≤4
The third term containing β satisfies

1 1
u2 a2
ku3 ∇4 (Ωβ)kL2u L1u L2 (Su,u ) .ku−2 kL1u ku5 ∇4 (Ωβ)kL∞ 2 2
u Lu L (Su,u )
. R(β).
|u|
For the next two terms, we have
X 3
u 2 aO2
1 1 1
u 2 a 2 B− 2
ku3 ∇i1 ψ i2 +1 ∇i3 ψ kL2u L1u L2 (Su,u ) . ≤ ,
i +i +i =4
|u|2 |u|
1 2 3
X 3
u 2 aO2
1
u 2 a 2 B− 2
1 1
ku4 ∇i1 +1 ψ i2 +1 ∇i3 ψ kL2u L1u L2 (Su,u ) . 2

≤ .
i1 +i2 +i3 =4
|u| |u|
i1 ≤3
The sixth term containing K obeys

X
ku4 ∇i1 ψ i2 +1 ∇i3 KkL2u L1u L2 (Su,u )
i1 +i2 +i3 =4
3 3 1 1 1 1 1 1
1 u2 a4 u2 a4 u2 u 2 a 2 B− 2
.ua 2 O 3
1+ 1 R+ 2 . ,
|u| |u| 2 Ω |u| |u|
where we use Proposition 4.1 and the bootstrap assumption (2.40).
For the last two terms, we have
ku4 ψ ∇5 (Ωe3 φ)kL2u L1u L2 (Su,u ) ≤kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
1 1 1 1 1 1
1 1 u2 a4 a2 1 1 u 2 a 2 B− 2
≤u a 2 4 1+ 1 S · 2 |u| u O ≤
2 2 .
|u| 2 Ω |u| |u|
And by Proposition (6.5), it holds
1 1
3 5 5 5 −2 u2 a2
ku ∇ (Ωe4 φ)kL2u L1u L2 (Su,u ) . ku ∇ (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
ku kL1u . .
|u|
Collecting all the above estimates, we arrive at

1 1 1
u2 a2 B− 2 5 5
ku4 ∇4 (Ωκ)kL2u L∞ 2
u L (Su,u )
. [1 + R(β)] + ku ∇ (Ωω † )kL∞ 2 2
u Lu L (Su,u )
.
|u| |u|
This also implies
1
1 1 B− 2
ku5 ∇4 (Ωκ)kL2u L∞ 2
u L (Su,u )
. u 2 a 2 [1 + R(β) +
1
5 5 †
1 ku ∇ (Ωω )kL∞ 2 2
u Lu L (Su,u )
]. (7.1)
u2 a2
We then appeal to the following div -curl system:
1 1
div ∇ω̂ = div κ + div (β − ∇4 φ∇φ), curl ∇ω̂ = 0,
2 2
1 1
†
curl ∇ω = curl κ + curl (β − ∇4 φ∇φ), div ∇ω † = 0.
2 2
Applying the elliptic estimates from Proposition 3.11, we get
ku5 ∇5 (Ωω̂, Ωω † )kL2 (Su,u )
X 1

. kuj+1 ∇j (Ωκ)kL2 (Su,u ) + kuj+1 ∇j (Ωβ − Ω∇4 φ∇φ)kL2 (Su,u ) + kuj ∇j (Ωω̂, Ωω † )kL2 (Su,u )
2
j≤4
X
5 4 j+1 j 1 j j †
.ku ∇ (Ωκ)kL2 (Su,u ) + ku ∇ (Ωβ − Ω∇4 φ∇φ)kL2 (Su,u ) + ku ∇ (Ωω̂, Ωω )kL2 (Su,u ) .
2
j≤4
After taking L2 in u, together with (7.1), the above inequality gives

ku5 ∇5 (Ωω̂, Ωω † )kL2u L2 (Su,u )
1
1 1 B− 2 5 5 † 1 1
.u a 1 + R(β) + 1 1 ku ∇ (Ωω )kL∞
2 2 2 2
u Lu L (Su,u )
+ u 2 a 2 [1 + R(β)].
u2 a2
With B sufficiently large, we absorb the u5 ∇5 (Ωω † ) term to the left and get
1 1
ku5 ∇5 (Ωω̂, Ωω † )kL2u L2 (Su,u ) .u 2 a 2 [1 + R(β)].
We then derive the estimates for trχ and χ̂.

(2.39), (2.40),(2.41), we have
3 1 1
ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . u 2 a[1 + R(β)], ku5 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . u 2 a 2 ,
1 1
ku5 ∇5 (Ωχ̂)kL2u L2 (Su,u ) . u 2 a 2 [1 + R(β)].
Proof. We use the following equation:

1
(Ω∇4 )(Ωtrχ) = − (Ωtrχ)2 − |Ωχ̂|2 − 4Ωω̂Ωtrχ − (Ω∇4 φ)2 .
2
Commuting it with i angular derivatives, we get
X
(Ω∇4 )∇5 (Ωtrχ) = ∇i1 ψ i2 ∇i3 ψ ∇i4 ψ .
i1 +i2 +i3 +i4 =5
Via applying Proposition 3.5, we can bound ku6 ∇5 (Ωtrχ)kL∞ ∞ 2

u Lu L (Su,u )
by the ku6 ·kL∞ 1 2
u Lu L (Su,u )
5
norm of the right. Together with ∇ (Ωtrχ)(u, 0) = 0, we get
X
ku6 ∇5 (Ωtrχ)kL2 (Su,u ) ≤ku6 ∇i1 ψ i2 ∇i3 ψ ∇i4 ψ kL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 +i4 =5
1
.u kuψ kL∞
2
u L (Su,u )
∞ ku5 ∇5 ψ kL∞ 2 2
u Lu L (Su,u )
X (7.2)
+ku6 ∇i1 ψ i2 ∇i3 ψ ∇i4 ψ kL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 +i4 =5
i3 ,i4 ≤4
.ua(OÕ + O2 ).
This gives
3 3
ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . u 2 a(OÕ + O2 ) ≪ u 2 aÕ′ . (7.3)
It also implies
1 1
ku5 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . u 2 a 2 . (7.4)
We then use the Codazzi equation with the schematic form

1
div (Ωχ̂) − ∇(Ωtrχ) + Ωβ = ψ ψ .
2
Applying the elliptic estimates in Proposition 3.12, we obtain
X X
ku5 ∇5 (Ωχ̂)kL2 (Su,u ) . kui ∇i (Ωtrχ)kL2 (Su,u ) + kui+1 ∇i (Ωβ)kL2 (Su,u )
i≤5 i≤4
X X X
+ kui+1 ∇i1 ψ ∇i2 ψ kL2 (Su,u ) + kui ∇i (Ωχ̂)kL2 (Su,u ) .
i≤4 i1 +i2 =i i≤4
50 XINLIANG AN
We further take the L2 norm in u. With the bound (7.3) for ∇5 trχ, we deduce
X
ku5 ∇5 (Ωχ̂)kL2u L2 (Su,u ) . kui ∇i (Ωtrχ)kL2u L2 (Su,u )
i≤5
X Ω X
+ kui+1 ∇i (Ωβ− ∇4 φ∇A φ)kL2u L2 (Su,u ) + kui+1 ∇i (Ω∇4 φ∇A φ)kL2u L2 (Su,u )
2
i≤4 i≤4
X X X
i+1 i1 i2
+ ku ∇ ψ ∇ ψ kL2u L2 (Su,u ) + kui ∇i (Ωχ̂)kL2u L2 (Su,u )
i≤4 i1 +i2 =i i≤4
3
2
1 1 1 1 u aO2 1 1
.u 2 a 2 + u 2 a 2 R(β) + . u 2 a 2 [1 + R(β)].
|u|
With the above estimate for u5 ∇5 (Ωχ̂), together with Proposition 4.3, Proposition 5.1, Propo-
sition 6.5, we then revisit (7.2) and obtain
1
ku6 ∇5 (Ωtrχ)kL2 (Su,u ) ≤ u 2 k(Ωtrχ, Ωω̂)kL∞
u L (Su,u )
∞ ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u )
1
+ u 2 kuΩtrχkL∞
u L (Su,u )
∞ ku5 ∇5 (Ωω̂)kL2u L2 (Su,u )
1
+ u 2 kuΩχ̂kL∞
u L (Su,u )
∞ ku5 ∇5 (Ωχ̂)kL2u L2 (Su,u )
1
+u 2 kuΩ∇4 φkL∞
u L (Su,u )
∞ ku5 ∇5 (Ω∇4 φ)kL2u L2 (Su,u )
X
+ ku6 ∇i1 ψ i2 ∇i3 ψ ∇i4 ψ kL∞ 1 2
u Lu L (Su,u )
i1 +i2 +i3 +i4 =5
i3 ,i4 ≤4
1 1
u2 a2 3 1 1 1 1 1 1 1 1 1 1 1
≤ · u 2 aÕ + u 2 Ω2 · u 2 a 4 [1 + R(β)] + u 2 a 2 · u 2 a 2 [1 + R(β)]+u 2 a 2 · u 2 a 2
|u|
≤ua[1 + R(β)].
This further implies
3
ku6 ∇5 (Ωtrχ)kL2u L2 (Su,u ) . u 2 a[1 + R(β)]. (7.5)
We then control the highest derivative of η.
(2.39), (2.40), (2.41), it holds
1 1
1 1 u2 a4
ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
. u2 a4 · 1 .
|u| Ω2
Proof. We define µ via

1 1
µ = −div η + K − − ∇A φ∇A φ.
|u|2 4
Thus η and µ satisfy the following Hodge system:
1 1
div η = −µ + K − 2
− ∇A φ∇A φ, curl η = −σ̌.
|u| 4
Recall ∇3 η A = −χAB · (η − η)B + β A − 12 ∇A φ · e3 φ and
1 1 3 1 1
∇3 (K − − ∇A φ∇A φ) + trχ(K − 2 − ∇A φ∇A φ)
|u|2 4 2 |u| 4
1 1
=div (β + ∇3 φ∇φ) − ∇A ∇3 φ∇A φ − ζ · β + 2η · β
2 2
1 1 1 1
b − trχ|η|2 + η A ∇A φ∇3 φ
b + χ̂ · (η ⊗η)
+ χ̂ · ∇⊗η
2 2 2 2
1 1 1
+ (∇A φ∇B φ − gAB ∇C φ∇C φ) · χ̂AB − trχ∇A φ∇A φ
2 2 8
1 1 1 Ω−1 2
+ trχ(−div η + K − 2 − ∇A φ∇A φ) − 2
(Ωtrχ + ).
2 |u| 4 |u| |u|
Hence, in the schematic form, we have that µ obeys
Ω
Ω∇3 µ + Ωtrχ µ =ψ ∇(η, η) + ψ ψ ψ + ψ ∇(Ωχ̂, Ωtrχ) + Ωtrχdiv η + ΩtrχK +
|u|3
+Ωtrχ ψ ψ + ∆g φ(Ωe3 φ) + eA φeA (Ωe3 φ) + ηeA φ(Ωe3 φ).
Commuting with 4 angular derivatives, we get
Ω∇3 ∇4 µ + 3Ωtrχ∇4 µ
Ω 1 X
= ∇5 η + Ωe3 φ · ∇5 (eA φ) + eA φ · ∇5 (Ωe3 φ) + ψ ∇5 (η, η, Ωtrχ, Ωχ̂) + ∇i1 ψ i2 +1 ∇i3 ψ
|u| |u| i +i +i =4
1 2 3
1 X X X
+ ∇i1 ψ i2 ∇i3 K + ∇i1 ψ i2 +2 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i3 K
|u| i +i +i =4 i1 +i2 +i3 =4 i1 +i2 +i3 =4
1 2 3
X X
i1 i2 i3 +1 i4
+ ∇ ψ ∇ eA φ∇ (Ωe3 φ) + i1 i2 i3
∇ ψ ∇ eA φ∇i4 +1 (Ωe3 φ).
i1 +i2 +i3 =4 i1 +i2 +i3 =4
i3 ≤3 i4 ≤3
We then employ Proposition 3.6 with λ0 = 3 and ku5 ∇4 µkL∞ 2

u L (Su,u )
can be controlled by
the ku5 · kL1u L2 (Su,u ) norm of the right hand side. We bound each of these terms. We begin
with terms containing the 5 angular derivatives on the Ricci coefficients. First note that by
Proposition 7.4, we have
3 3 1 1 1 1
u2 a4 u2 a4 1 1 u2 a4
kΩu4 ∇5 ηkL2u L1u L2 (Su,u ) ≤ kΩu−2 kL1u ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
. 1+ 1 R ≤ u2 a4 · 1 .
|u| |u| 2 Ω |u| 2 Ω
The next term is a borderline term. Via Proposition 6.5, we control it as
ku5 ∇5 (eA φ)(Ωe3 φ)kL2u L1u L2 (Su,u )

≤kΩu5 ∇5 (eA φ)kL∞ 2 2
u Lu L (Su,u )
kΩ−1 (Ωe3 φ)kL2u L2u L∞ (Su,u )
1 1
1 Ω−1
1 1 1 1 1 u2 a4
≤u a ·
2 2 · |u| 2 u 2 = u 2 a 4 · 1 .
|u| |u| 2 Ω
And the term containing ∇5 (Ωe3 φ) obeys
ku5 ∇5 (Ωe3 φ)(eA φ)kL2u L1u L2 (Su,u )

≤kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
1 1 1 3 1 1 1 1 1
1 1 u2 a4 ua 2 1 1 1 1 u2 a2 u2 a4 1 1 u2 a4
≤u 2 a 4 1 + 1 S · |u| 2 u2 O = u2 a4 ·
3 1 + 1 SO ≤ u 2 a4 ·
1 .
|u| 2 Ω |u|2 |u| 2 |u| 2 Ω |u| 2 Ω
52 XINLIANG AN
Then the top-order-derivative terms with ∇5 η or ∇5 η obey

ku5 ψ ∇5 (η, η)kL2u L1u L2 (Su,u )
≤ku−2 kL1u ku2 ψ kL∞
u Lu L (Su,u )
∞ ∞ ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
+ ku−1 ψ kL∞ 1 ∞
u Lu L (Su,u )
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
1 1 1 1 1 1
ua 2 O 1 1 u2 a4 ua 2 3 3 1 1 u2 a4
≤ · u2 a4 1 + 1 Õ + 2
u 2 a 4 OÕ ≤ u 2 a 4 · 1 .
|u| |u| 2 Ω |u| |u| 2 Ω
And the top-order-derivative terms ∇5 (Ωtrχ, Ωχ̂) satisfy

ku5 ψ ∇5 (Ωtrχ, Ωχ̂)kL2u L1u L2 (Su,u )
≤kψ kL2u L2u L∞ (Su,u ) ku5 ∇5 (Ωtrχ, Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
1 1 1 1 1
ua 2 1 1 1 1 u2 a4 1 1 u2 a4
≤ 2
u 2 |u| 2 O · u 2 a 4 1 + 1 Õ ≤ u 2 a 4 · 1 .
|u| |u| 2 Ω |u| 2 Ω
The next term is lower-order and we have

X 5
u 2 aO2
1 1
5 1 i1 i2 +1 i3 1 1 u2 a4
ku ∇ ψ ∇ ψ kL2u L1u L2 (Su,u ) ≤ ≤u a ·
2 4
1 .
|u| i +i +i =4 |u|2 |u| 2 Ω
1 2 3
For the term containing K, via using Proposition 3.4, we deduce

1 X 1 1 u2 a4
1 1
5 i1 i2 i3
ku ∇ ψ ∇ KkL2u L1u L2 (Su,u ) ≤ u a · 2 4
1 .
|u| i |u| 2 Ω
1 +i2 +i3 =4
1 1
1 1 u2 a4
The remaining lower-order terms are all bounded by u 2 a 4 · 1 . Hence, combining the
|u| 2 Ω
above estimates, we arrive at
1 1
1 1 u2 a4
ku5 ∇4 µkL2u L∞ 2
u L (Su,u )
≤ u2 a4 · 1 .
|u| 2 Ω
By using the div-curl system
1 1
div η = −µ + K − − ∇A φ∇A φ, curl η = −σ̌
|u|2 4
and employing elliptic estimates from Proposition 3.11, we further deduce
1 1
ku5 ∇5 ηkL2u L2 (Su,u ) .ku5 ∇4 µkL2u L2 (Su,u ) + ku5 ∇4 (K − 2 − ∇A φ∇A φ, σ̌)kL2u L2 (Su,u )
|u| 4
X 1 1
+ kui+1 ∇i (−µ + K − 2 − ∇A φ∇A φ, σ̌)kL2u L2 (Su,u )
|u| 4
i≤3
1 1 3 3 1 1 1 1
1 1 u2 a4
u2 a4 u2 a4 1 1 u2 a4
≤u a ·2
1
4 + 1+ 1 R ≤ u2 a4 · 1 .
|u| Ω
2 |u| |u| Ω
2 |u| 2 Ω

We proceed to obtain the highest order derivative estimates for η. In the proof we will derive
an improved bound for ∇i µ with 1 ≤ i ≤ 4.
(2.39), (2.40), (2.41) and the proved O5,2 (η) . 1, we have
1 1
3 3 u2 a4 3 3 1
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
. u2 a4 1 + 1 [1 + R(K, σ̌)] + u 2 a 4 · a− 4 R(β),
|u| 2 Ω
1 1
ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
. Ω−1 u 2 a 2 [1 + R(K, σ̌)],
1 1
ku5 Ω∇5 ηkL∞ 2 2
u Lu L (Su,u )
. u 2 a 2 [1 + R(K, σ̌)].
And ∇i µ obeys the below improved bound:

X 3
1
u2 a4
1
i+2 i
ku ∇ µkL∞ ∞ 2
u Lu L (Su,u )
. ua ·
4
1 .
1≤i≤4
|u| 2 Ω
Proof. We first define the mass aspect function µ

1 1
µ := −div η + K − 2
− ∇A φ∇A φ.
|u| 4
Recall that η obeys the following elliptic system
1 1
div η = −µ + K − 2 − ∇A φ∇A φ, curl η = σ̌.
|u| 4
And via employing the below renormalized null Bianchi equation
1 1 1 1
∇4 [Ω(K − 2 − ∇A φ∇A φ)] + trχ[Ω(K − 2 − ∇A φ∇A φ)]
|u| 4 |u| 4
1 1 1 1 1
= − Ωdiv (β − ∇4 ∇φ) − Ω∇A ∇4 φ · ∇A φ − 2Ωω̂(K − 2 − ∇A φ∇A φ) − Ωtrχ · 2
2 2 |u| 4 |u|
1 1 1 1
b − trχdiv η − trχ|η|2 ]
b + χ̂ · (η ⊗η)
+ Ω[−ζ · β − 2η · β + χ̂ · ∇⊗η
2 2 2 2
1 A 1 1
+ Ω[ η ∇A φ∇4 φ + (∇A φ∇B φ − gAB ∇C φ∇C φ) · χ̂AB ],
2 2 2
(7.6)
we deduce that µ satisfies the schematic equation
1 1 Ωtrχ
(Ω∇4 )µ =χ∇(η, η) + ψ ψ ψ + ψ ∇(Ωtrχ, Ωχ̂) + ψ (K − 2
− ∇A φ∇A φ) +
|u| 4 |u|2
+∆g φ(Ωe4 φ) + ∇A φ∇A (Ωe4 φ).
Commuting with i angular derivatives, we have
(Ω∇4 )∇i µ
1 4
=Ωe4 φ · ∇i+1 (eA φ) + eA φ∇i+1 (Ωe4 φ) + χ∇i+1 (η, η) + ψ ∇i+1 (Ωtrχ, Ωχ̂)+ ∇ (Ωtrχ)
|u|2
X X 1 1
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 ∇i3 ψ ∇i4 (K − 2 − ∇A φ∇A φ)
i1 +i2 +i3 =i+1 i1 +i2 +i3 +i4 =i
|u| 4
i1 ,i3 ≤i
X X
+ ∇i1 ψ i2 ∇i3 +1 (∇A φ)∇i4 (Ωe4 φ)+ ∇i1 ψ i2 ∇i3 +1 (Ωe4 φ)∇i4 (eA φ).
i1 +i2 +i3 +i4 =i i1 +i2 +i3 +i4 =i
i3 ≤i−1 i3 ≤i−1
Since ∇i µ vanishes on u = 0, by Proposition 3.5, to bound kui+2 ∇i µkL∞ ∞ 2

u Lu L (Su,u )
, we only
i+2
need to control the ku · kL∞ 1 2
u Lu L (Su,u )
norm of the right. We now estimate them.
Via employing the obtained estimates in Proposition (6.5), Proposition (6.6) and Proposition
(5.1), the terms involving the highest angular derivatives of the scalar field satisfy
kui+2 ∇i+1 (eA φ)(Ωe4 φ)kL1u L2 (Su,u )
≤kui+1 ∇i+1 (eA φ)kL2u L2 (Su,u ) ku(Ωe4 φ)kL2u L∞ (Su,u )
1 1 1 1 1
u2 a4 1 1 ua 2 1 1 1 1 1 3 u2 a4
≤ 1 · u 2 a4 + (
3 ) 4 · u2 a4 · u 2 a 2 ≤ ua 4 ·
1 ,
|u| 2 Ω |u|Ω 2 |u| 2 Ω
kui+2 ∇i+1 (Ωe4 φ)(eA φ)kL1u L2 (Su,u )

≤kui+1 ∇i+1 (Ωe4 φ)kL2u L2 (Su,u ) kueAφkL2u L∞ (Su,u )
1 1 1 1
1 1ua 2 1 1 ua 2 3 u2 a4
≤u a S ·
2 2 u 2 O = ua 2 SO ≤ ua 4 · 1 .
|u| |u| |u| 2 Ω
54 XINLIANG AN
The terms containing the highest angular derivatives of η and η obey

X
kui+2 χ∇i+1 (η, η)kL1u L2 (Su,u )
i≤4
1 1
≤u 2 kχkL∞
u L (Su,u )
∞ ku6 ∇5 ηkL2u L2 (Su,u ) + u 2 kuχkL∞
u L (Su,u )
∞ ku5 ∇5 ηkL2u L2 (Su,u )
1 1 1 1 1 1
u2 a2 O 3 3 1 1 1 1 u2 a4 3 u2 a4
≤ u 2 a 4 R + u 2 a 2 O(χ) · u 2 a 4 · 1 Õ(η) . ua 4 · 1 .
|u| |u| Ω
2 |u| 2 Ω
For the last line we utilize Proposition (4.2), Proposition (4.3) and Proposition (7.3).
The terms with the highest derivatives of Ωtrχ and Ωχ̂ satisfy
X
kui+2 ψ ∇i+1 (Ωtrχ, Ωχ̂)kL1u L2 (Su,u )
i≤4
1 X
u2
. ku2 ψ kL∞
u L (Su,u )
∞ · ku i+1 i+1
∇ 5
(Ωtrχ, Ωχ̂)kL2u L2 (Su,u ) + ku ∇ (Ωtrχ, Ωχ̂)kL2u L2 (Su,u )5
|u|
i≤3
1 1 1
u2 1 1 1 u2 aOÕ 3 u2 a4
. ua 2 O · u 2 a 2 Õ . . ua 4 · 1 .
|u| |u| |u| 2 Ω
By Proposition (4.3), we also have
1
ku4 ∇4 (Ωtrχ)kL1u L2 (Su,u ) ≤k kL1 L∞ (Su,u ) ku5 ∇4 (Ωtrχ − Ω2 (Ω−1 trχ)0 )kL∞ 2
u L (Su,u )
+ ku3 ∇4 Ω2 kL1u L2 (Su,u )
|u| u
1 1 1 1
u ua 2 3 ua 2 1 uO 3 u2 a4
≤ · ua + 2
· u|u|O ≤ ua 4 · + ua 2 · ≤ ua 4 · 1 .
|u| |u| |u| |u| |u| 2 Ω
Employing Proposition 4.1 together with the bootstrap assumption (2.39), we then bound the
1 1
3 u2 a4
other lower order terms and they obey . ua 4 · 1 .
|u| 2 Ω
Combining all the estimates above, we obtain
X 3
1
u2 a4
1
kui+2 ∇i µkL∞ 2
u L (Su,u )
.ua 4 · 1 . (7.7)
1≤i≤4
|u| Ω2
We then recall the div-curl system

1 1
div η = −µ + K − 2
− ∇A φ∇A φ, curl η = σ̌.
|u| 4
Via employing elliptic estimates from Proposition 3.11 and the fact Ω ≤ 1, we obtain
ku6 ∇5 ηkL2 (Su,u )
X 1 1

. kui+2 ∇i µkL2 (Su,u ) + kui+2 ∇i (K − 2 − ∇A φ∇A φ, σ̌)kL2 (Su,u ) + kui+1 ∇i ηkL2 (Su,u )
|u| 4
i≤4
3
1 1
u2 a4 X 1 1 1
.ua 4 1 + 1 · R(K) + kui+2 ∇i (K − 2
− ∇A φ∇A φ, σ̌)kL2 (Su,u ) + ua 2 R(β),
|u| Ω2 |u| 4
i≤4
ku5 Ω∇5 ηkL2 (Su,u )

X 1 1

. kΩui+1 ∇i µkL2 (Su,u ) + kΩui+1 ∇i (K − 2 − ∇A φ∇A φ, σ̌)kL2 (Su,u ) + kΩui ∇i ηkL2 (Su,u )
|u| 4
i≤4
X 1 1 ua 4 u2 a4
3
ua 2
1 1 1
i+1
. kΩu ∇ (K − 2 − ∇A φ∇A φ, σ̌)kL2 (Su,u ) +
i
·Ω· 1+ 1 · R(K) + · ΩO.
|u| 4 |u| |u| Ω
2 |u|
i≤4
We then take the L2 norm in u and u respectively. Via using (7.7), we further obtain
1 1
3 3 u2 a4 3 3 1
ku6 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
.u 2 a 4 1 + 1 · R(K, σ̌) + u 2 a 4 · a− 4 R(β),
|u| Ω 2
1 1 1 1 1 1
kΩu5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
. u 2 a 2 R(K, σ̌) + u 2 a 2 . u 2 a 2 [1 + R(K, σ̌)]. (7.8)
This also infers
1 1
ku5 ∇5 ηkL∞ 2 2
u Lu L (Su,u )
≤ Ω−1 u 2 a 2 · [1 + R(K, σ̌)].
We hence finish the proof of this proposition.
Finally, we move to bound the highest order estimates for trχ and χ̂.
(2.39), (2.40), (2.41), it holds
1 1
5 5 1 u2 a4
1
ku ∇ (Ωtrχ, Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
.u a 1 +2 4
1 ,
|u| 2 Ω
1 1
1 1 u2 a4
ku5 ∇5 trχkL∞ 2 2
u Lu L (Su,u )
.u a 1 +
2 4
1 ,
|u| 2 Ω
1 1
1 1 u2 a4
ku5 ∇5 χ̂kL∞ 2 2
u Lu L (Su,u )
.u 2 a 4 1 + 1 · R(β).
|u| 2 Ω
Proof. We utilize the following equation for trχ:
1
Ω∇3 (Ω−1 trχ) + Ωtrχ(Ω−1 trχ) = −|χ̂|2 − ∇3 φ∇3 φ.
2
Via commuting the above equation with 5 angular derivatives, we obtain
Ω∇3 ∇5 (Ω−1 trχ) + 3Ωtrχ∇5 (Ω−1 trχ)
X
=ψ ∇5 (Ω−1 trχ, Ω−1 χ̂) + Ω−1 e3 φ∇5 (Ωe3 φ) + Ω−2 ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 +i3 =5
i1 ,i3 ≤4 (7.9)
X Ω −2 X Ω −2
+ ∇i1 ψ i2 +1 ∇i3 ψ + 2
∇i1 ψ i2 +1 .
i1 +i2 +i3 =4
|u| i
|u|
1 +i2 =4
In the derivation of above equation, we employ

e3 φ∇5 (e3 φ) =e3 φ∇5 [e3 φ − (e3 φ)0 ] = e3 φ∇5 [Ω−1 Ω(e3 φ − (e3 φ)0 )]
X
=Ω−1 e3 φ ψ i1 ∇i2 [Ω(e3 φ − (e3 φ)0 )] + Ω−1 e3 φ∇5 [Ω(e3 φ − (e3 φ)0 )]
i1 +i2 =5
i2 ≤4
X
=Ω−1 e3 φ ψ i1 ∇i2 [Ω(e3 φ − (e3 φ)0 ] + Ω−1 e3 φ∇5 [(Ωe3 φ) − (Ωe3 φ)0 ]
i1 +i2 =5
i2 ≤4
− Ω−1 e3 φ∇5 [(Ω − Ω0 )(e3 φ)0 ].

Note that Ω−1 e3 φ∇5 [(Ωe3 φ) − (Ωe3 φ)0 ] is the same as the last term in (7.9) and the rest terms
are absorbed by the other same-type terms in (7.9). We also notice
1
5 5 2 5 5 −1 ua 2 O
ku ∇ (Ωtrχ)k L2 (S u,u )
.kΩ k L∞ (S u,u )
ku ∇ (Ω trχ)k L2 (S u,u )
+ ,
|u|
1
ua 2 O
ku5 ∇5 trχkL2 (Su,u ) .kΩkL∞ (Su,u ) ku5 ∇5 (Ω−1 trχ)kL2 (Su,u ) + .
|u|Ω
To bound ku5 ∇5 (Ω−1 trχ)kL2 (Su,u ) , we then apply Proposition 3.6 with λ0 = 3 and it can be
controlled by the ku5 · kL∞ 1 2
u Lu L (Su,u )
norm of the right hand side of (7.9).
56 XINLIANG AN
Together, to estimate ku5 ∇5 (Ωtrχ)kL2 (Su,u ) , we first bound
kΩ2 kL∞ (Su,u ) · ku5 ψ ∇5 (Ω−1 trχ, Ω−1 χ̂)kL∞ 1 2

u Lu L (Su,u )
≤ku5 ∇5 (Ωtrχ, Ωχ̂)kL∞ 2 2

u Lu L (Su,u )
· kψ kL∞ 2 ∞
u Lu L (Su,u )
1 1 1 1 1 1 1 1 1 1 1 1
1 1 u2 a4 ua 2 1 u 2 a 4 ua 2 u2 a4 u2 a4 u2 a4
≤u 2 a 4 1 + 1 Õ · 2
|u| 2 O = 1 · 1+ 1 ÕO ≤ 1 · 1 1 .
|u| 2 Ω |u| |u| 2 |u| |u| 2 Ω |u| 2 |u| 2 Ω 2
To bound ku5 ∇5 trχkL2 (Su,u ) , similarly, we get
kΩkL∞ (Su,u ) · ku5 ψ ∇5 (Ω−1 trχ, Ω−1 χ̂)kL∞ 1 2

u Lu L (Su,u )
≤ku5 ∇5 (trχ, χ̂)kL∞ 2 2

u Lu L (Su,u )
· kψ kL∞ 2 ∞
u Lu L (Su,u )
1 1 1 1 1 1 1 1 1 1 1 1
1 1 u2 a4 ua 2 1 u 2 a 4 ua 2 u2 a4 u2 a4 u2 a4
≤u 2 a 4 1 + 1 Õ · |u| 2O =
1 · 1 + 1 ÕO ≤ 1 · 1 1 .
|u| 2 Ω |u|2 |u| 2 |u| |u| 2 Ω |u| 2 |u| 2 Ω 2
We then move to control the remaining terms. We first bound
k(Ω2 , Ω)kL∞ (Su,u ) · kΩ−1 u5 e3 φ∇5 (Ωe3 φ)kL∞ 1 2
u Lu L (Su,u )
.k(Ω, 1)kL∞ (Su,u ) kΩe3 φkL∞ 2 ∞

u Lu L (Su,u )
kΩ−1 u5 ∇5 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
1 1 1
u2 a4 1 1 ua 2 1 1 1
.k(Ω, 1)kL∞ (Su,u ) · ku−1 kL2u · 1 · u 2 a4 + (
3 ) 4 · u2 a4
|u| 2 Ω |u|Ω 2
1 1 1
1 u2 a4 1 1 ua 2 1 1 1
. 1 · 1 · u2 a4 + ( 3 )
4 · u2 a4 .
|u| 2 |u| 2 Ω |u|Ω 2
Here for the third line, we employ Proposition 6.6. For the rest terms, we have
X 1
u2 a4
1 1
u2 a4
1
k(Ω2 , Ω)kL∞ (Su,u ) · kΩ−2 u5 ∇i1 ψ i2 +1 ∇i3 ψ kL∞ 1 2

u Lu L (Su,u )
≤ 1 · 1 1 ,
i1 +i2 +i3 =5 |u| 2 |u| Ω 2 2
i1 ,i3 ≤4
X Ω−2 i1 i2 +1 i3 u2 a4 u2 a4
1 1 1 1
k(Ω2 , Ω)kL∞ (Su,u ) · ku5 ∇ ψ ∇ ψ kL∞ 1 2

u Lu L (Su,u )
≤ 1 · 1 1 ,
i1 +i2 +i3 =4
|u| |u| 2 |u| 2 Ω 2
X Ω−2 i1 i2 +1 u2 a4 u2 a4 1
1 1 1 1
k(Ω2 , Ω)kL∞ (Su,u ) · ku5 2

∇ ψ kL∞ 1 2
u Lu L (Su,u )
≤ 1 · ( 1 1 )2.
i1 +i2 =4
|u| |u| 2 |u| 2 Ω 2
Collecting all the above estimates, we derive

1 1 1 1 1 1 1 1
u2 a4 u2 a4 u2 a4 u2 a4 1
ku5 ∇5 (Ωtrχ)kL2 (Su,u ) ≤ 1 · Ω · (1 + 1 )+ 1 ( 1 1 )2 ,
|u| 2 |u| Ω 2 |u| 2 |u| Ω
2 2
1 1
1 1 u a
2 4
ku5 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
≤ u 2 a 4 (1 + 1 ),
|u| 2 Ω
1 1 1 1 1 1 1 1 1 1 1
(7.10)
5 5 u2 a4 u2 a4 ua 2 1 u2 a4 u2 a4 u2 a4 1
ku ∇ trχkL2 (Su,u ) ≤ 1 · 1 +( 3 ) · 4
1 + 1 ( 1 1 ) , 2
|u| 2 Ω |u| 2 |u|Ω 2 |u| 2 |u| 2 |u| 2 Ω 2
1 1
1 1 u2 a4
ku5 ∇5 trχkL∞ 2 2
u Lu L (Su,u )
≤ u 2 a 4 (1 + 1 ).
|u| 2 Ω
To bound the fifth angular derivatives of χ̂, we now employ the Codazzi equation
1 1 1
div χ̂ = β + ∇trχ − (η − η) · (χ̂ − trχ).
2 2 2
Applying Proposition 3.12, we get

X 1
ua 2 O
ku5 ∇5 (Ωχ̂)kL2 (Su,u ) . kui+1 ∇i+1 (Ωtrχ)kL2 (Su,u ) + kui+1 ∇i (Ωβ)kL2 (Su,u ) +
|u|
i≤4
1 1 1 1 1 1 1 1 1
u2 a4 Ω u2 a4 u2 a4 u2 a4 1 ua 2 O
. 1 (1 + 1 )+ 1 ·( 1 1 ) 2 + kui+1 ∇i (Ωβ)kL2 (Su,u ) + ,
|u| 2 |u| Ω2 |u| 2 |u| Ω
2 2 |u|
X 1
ua 2 O(η, η)
5 5 i+1 i+1 i+1 i
ku ∇ χ̂kL2 (Su,u ) . ku ∇ trχkL2 (Su,u ) + ku ∇ βkL2 (Su,u ) + .
|u|Ω
i≤4
2
We then take the L norm in u and obtain
ku5 ∇5 (Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
1 1
1 1
u2 a4 X 1
ua 2 O
.u 2 a 4 1 + 1 + kui+2 ∇i βkL∞ 2 2
u Lu L (Su,u )
ku−1 ΩkL∞
u Lu L (Su,u )
∞ ∞ + 1
|u| 2 Ω i≤4
|u| 2
1 1 1 1 1 1 1
1 1 u2 a4 3 3 u2 a4 1 ua 2 O 1 1 u2 a4
.u 2 a 4 (1 + 1 ) + u2 a4 1 + 1 R· + 1 ≤ u 2 a 4 (1 + 1 ),
|u| Ω 2 |u| 2 Ω |u| |u| 2 |u| 2 Ω
ku5 ∇5 χ̂kL∞ 2 2
u Lu L (Su,u )
X 1
i+1 i+1 i+1 i
ua 2 O(η, η)
. ku ∇ trχkL∞ 2 2
u Lu L (Su,u )
+ ku ∇ βkL∞ 2 2
u Lu L (Su,u )
+ 1
i≤4
|u| 2 Ω
X X 1
ua 2 R(β)
i+1 i+1 i+2 i −1
. ku ∇ trχkL∞ 2 2
u Lu L (Su,u )
+ ku ∇ βkL∞ 2 2
u Lu L (Su,u )
· ku kL∞
u Lu L (Su,u )
∞ ∞ + 1
i≤4 i≤4
|u| 2 Ω
1 1 1 1
1 1 u2 a4 1 1 u2 a4
.u 2 a 4 (1 + 1 ) + u2 a4 · 1 · R(β),
|u| 2 Ω |u| 2 Ω
where we use Proposition 4.5 and Proposition 7.1.
Collecting all above estimates, we hence prove this proposition.

We conclude this section by summarizing all the proved estimates in this section with the aid
of the Õ5,2 norm:
(2.39), (2.40), (2.41), it holds
Õ5,2 . 1 + R.
8. Energy Estimates for Curvature Components

To carry out the energy estimates, we employ the below paired equations with suitable Ω
weights. They are from the renormalized null Bianchi equations. The first pair are
1 1
∇3 [Ω(β· − ∇4 φ∇· φ)]A + trχΩ(βA − ∇4 φ∇A φ)
2 2
1 1 C ∗ 1
= − Ω∇(K − 2 − ∇ φ∇C φ) + Ω ∇σ̌ − Ωtrχ∇4 φ∇A φ + Ωω∇4 φ∇A φ
|u| 4 2
∗ 1 ∗ 3 (8.1)
+ Ω[2χ̂ · β − 3(ηK − ησ̌) + (∇(χ̂ · χ̂) − ∇(χ̂ ∧ χ̂)) − ηtrχtrχ]
2 4
3 ∗ 1
+ Ω[ (η χ̂ · χ̂ − η χ̂ ∧ χ̂) − (∇trχtrχ + trχ∇trχ)]
2 4
3 1
+ Ω∇A (∇ φ∇C φ) − Ω∇ RBA + ηA ∇3 φ∇4 φ + ηA ∇C φ∇C φ,
C B
4 2
58 XINLIANG AN
3 1
∇4 (Ωσ̌) + trχΩσ̌ = − Ωdiv ∗ (β − ∇4 φ∇φ) + Ω(−∇1 R42 + ∇2 R41 ) − 2Ωω̂σ̌
2 2
1 1
+ Ω[−ζ · β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)
∗ b + χ̂ ·∗ (η ⊗η)]
b (8.2)
2 2
1 1
+ Ω(∇A φ∇B φ − gAB ∇C φ∇C φ) ∧ χ̂AB ,
2 2
1 1 1 1
∇4 [Ω(K − − ∇A φ∇A φ)] + trχΩ(K − 2 − ∇A φ∇A φ)
|u|2 4 |u| 4
1 1 1 1 1
= − Ωdiv (β − ∇4 ∇φ) − Ω∇A ∇4 φ∇A φ − 2Ωω̂(K − 2 − ∇A φ∇A φ) − Ωtrχ · 2
2 2 |u| 4 |u| (8.3)
1 1 1 1
b − trχdiv η − trχ|η|2 ]
b + χ̂ · (η ⊗η)
+ Ω[−ζ · β − 2η · β + χ̂ · ∇⊗η
2 2 2 2
1 A 1 1 C AB
+ Ω[ η ∇A φ∇4 φ + (∇A φ∇B φ − gAB ∇C φ∇ φ) · χ̂ ].
2 2 2
The second pair are
3 1 1 1
∇3 σ̌ + trχσ̌ = − div ∗ (β + ∇3 φ∇φ) − ζ ·∗ β − 2η ·∗ β + χ̂ ·∗ (∇⊗η)
b + χ̂ ·∗ (η ⊗η)
b
2 2 2 2 (8.4)
1 1
+ ∇1 R32 − ∇2 R31 + χ̂AB ∧ (∇A φ∇B φ − gAB ∇C φ∇C φ),
2 2
1 1 3 1 1
∇3 (K − 2
− ∇A φ∇A φ) + trχ(K − 2 − ∇A φ∇A φ)
|u| 4 2 |u| 4
1 1 A
=div (β + ∇3 φ∇φ) − ∇ ∇3 φ∇A φ − ζ · β + 2η · β
2 2
1 1 1 1
+ χ̂ · ∇⊗η b − trχ|η|2 + η A ∇A φ∇3 φ
b + χ̂ · (η ⊗η) (8.5)
2 2 2 2
1 1 1
+ (∇A φ∇B φ − gAB ∇C φ∇C φ) · χ̂AB − trχ∇A φ∇A φ
2 2 8
1 1 1 Ω−1 2
+ trχ(−div η + K − 2 − ∇A φ∇A φ) − 2
(Ωtrχ + ),
2 |u| 4 |u| |u|
1 1
∇4 (β · + ∇3 φ∇· φ)A + trχ(β A + ∇3 φ∇A φ)
2 2
1 1
=∇(K − 2 − ∇A φ∇A φ) +∗ ∇σ̌ + 2ω̂β + 2χ̂ · β + 3(−ηK +∗ ησ̌)
|u| 4
(8.6)
1 1 1 1 1
+ ∇(− ∇A φ∇A φ + trχtrχ − χ̂ · χ̂) − ∗ ∇(χ̂ ∧ χ̂)+ trχ∇3 φ∇A φ
4 4 2 2 2
CD B 1 C
+ ∇A (g RCD ) + ∇ RBA − η A ∇3 φ∇4 φ − η A ∇C φ∇ φ.
2
Via applying Propositions 6.3-6.4, for the first pair we have
(2.39), (2.40), (2.41), it holds
X 1 1 1 1 1
(kui+1 ∇i (Ωβ − Ω∇4 φ∇φ)kL∞ 2 2
u Lu L (Su,u )
+ kΩui+1 ∇i (K − 2 − ∇A φ∇A φ, σ̌)kL∞ 2 2
u Lu L (Su,u )
) . u2 a2 .
2 |u| 4
i≤4
By the obtained L2 (Su,u ) estimates in Section 5.1 for φ, it also holds

X 1 1 1
(kui+1 ∇i (Ωβ)kL∞ 2 2
u Lu L (Su,u )
+ kΩui+1 ∇i (K − 2 , σ̌)kL∞ 2 2
u Lu L (Su,u )
) . u2 a2 .
|u|
i≤4
Proof. Commuting equations for

1 1 1
∇3 [Ω(β − ∇4 φ∇φ)], ∇4 (Ωσ̌), ∇4 [Ω(K − 2
− ∇A φ∇A φ)]
2 |u| 4
with i angular derivatives, we get
1 1 1 2+i 1
∇3 ∇i [Ω(β· − ∇4 φ∇· φ]A +Ω∇∇i (K− 2 − ∇B φ∇B φ)−Ω∗ ∇∇i σ̌+ trχ∇i [Ω(β· − ∇4 φ∇· φ]A = G1,i
2 |u| 4 2 2
(8.7)
with G1,i being
1 i+1 X 1
G1,i =Ω−1 ψ ∇i+1 ψ + ∇ trχ + ψ ∇i+1 (χ̂, trχ) + ∇i1 ψ i2 +1 ∇i3 (K − , σ̌)
|u| i
|u|2
1 +i2 +i3 =i
X Ω−1 X X
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +2
i1 +i2 +i3 =i+1
|u| i1 +i2 +i3 =i i1 +i2 =i+1
i1 ,i3 ≤i i1 ≤i
1 X
+ ∇i1 ψ i2 +1 +Ω∇i+2 φ∇φ,
|u|2 i1 +i2 =i
1
∇4 ∇i (Ωσ̌) + Ωdiv ∗ ∇i (β − ∇4 φ∇φ) = G2,i , (8.8)
2
where G2,i is
X
G2,i =ψ ∇i+1 ψ + ψ ∇i+1 η + ∇i1 ψ i2 ∇i3 ψ ∇i4 σ̌
i1 +i2 +i3 +i4 =i
X
i1 i2 +1
+ ∇ ψ ∇i3 ψ +∇i+1 (Ωe4 φ)∇φ,
i1 +i2 +i3 =i+1
i1 ,i3 ≤i
1 1 1
∇4 ∇i [Ω(K − − ∇A φ∇A φ)] + Ωdiv ∇i (β − ∇4 φ∇φ) = G3,i (8.9)
|u|2 4 2
with G3,i satisfying
X 1
G3,i =ψ ∇i+1 ψ + ψ ∇i+1 η + ∇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 ∇i3 ψ + ∇i+1 (Ωe4 φ)∇φ.
i1 +i2 +i3 =i+1
|u|2 i1 +i2 +i3 =i
i1 ,i3 ≤i
Note that by the structure of the renormalized null Bianchi equation, we will have −∇1 R42 +
∇2 R41 , which enables us to avoid the troublesome term ∇i+2 φe4 φ in above two equations.
We now employ Proposition 6.2 for φ = ui+1 ∇i [Ω(K− |u|1 2 − 41 ∇A φ∇A φ)] and φ = ui+1 ∇i [Ωσ̌],
and apply Proposition 6.4 for φ = ∇i [Ω(β − 12 ∇4 φ∇φ)] with λ1 = i + 1. We then add these
identities together and obtain
X 1

kui+1 ∇i (Ωβ)kL2u L2 (Su,u ) + kΩui+1 ∇i (K − 2 )kL2 (H) + kΩui+1 ∇i σ̌kL2u L2 (Su,u )
|u|
i≤4
X X
. kui+1 ∇i (Ωβ)kL2u L2 (S0,u ) + kΩui+1 G1,i kL1u L2u L2 (Su,u ) + kΩui+1 G2,i kL1u L2u L2 (Su,u )
i≤4 i≤4
X
i+1
+ kΩu G3,i kL1u L2u L2 (Su,u ) .
i≤4
(8.10)
60 XINLIANG AN
Note that the highest derivatives of the scalar field in in G1,i satisfies
kΩui+1 ∇i+1 ∇φ∇φkL1u L2u L2 (Su,u ) ≤ kui+1 ∇i+1 ∇φkL∞ 2 2
u Lu L (Su,u )
kΩ∇φkL1u L∞
u L (Su,u )
∞
1 1 1
1 u2 a4
1 ua 2 O 1 1
≤u a · 1 +
2 2
1 S· ≤ u2 a2 .
|u| Ω
2 |u|
The corresponding terms in G2,i and G3,i obey
kΩui+1 ∇i+1 (Ωe4 φ)∇φkL1u L2u L2 (Su,u ) ≤ kui+1 ∇i+1 (Ωe4 φ)kL∞ 2 2
u Lu L (Su,u )
kΩ∇φkL2u L2u L∞ (Su,u )
1 3 1
1 ua 2 1 1 1 1 1 u2 a2 1 1
≤u a S ·
2 |u| 2 u 2 O = u 2 a 2
2
3 SO ≤ u a .
2 2
|u|2 |u| 2
We then proceed to bound other terms. Among the terms in Ωui+1 G1,i , we first control the
contributions from the fifth derivatives the Ricci coefficients. These terms are
1 5
ψ ∇5 (Ωtrχ, Ωχ̂), ∇ (Ωtrχ), ψ ∇5 (Ωχ̂, Ωtrχ) (8.11)
|u|
and in ku5 · kL1u L2u L2 (Su,u ) norm they satisfy
ku5 ψ ∇5 (Ωtrχ, Ωχ̂)kL1u L2u L2 (Su,u ) .ku5 ∇5 (Ωtrχ, Ωχ̂)kL∞ 2 2

u Lu L (Su,u )
kψ kL1u L∞
u L (Su,u )
∞
1
1 1 ua 2 O 1 1
.u 2 a 2 Õ · ≤ u2 a2 ,
|u|
3
4 5 6 5 1 u 2 aÕ 1 1
ku ∇ (Ωtrχ)kL1u L2u L2 (Su,u ) .ku ∇ (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
k 2 kL1u . ≤ u2 a2 .
|u| |u|
Here we use the improved bound for u6 ∇5 (Ωtrχ) in Proposition 7.2, which is better than other
∇5 (Ωψ ) components. For the terms containing (Ωχ̂, Ωtrχ), employing the bootstrap assumption
(2.40), we get
ku5 ψ ∇5 (Ωχ̂, Ωtrχ)kL1u L2u L2 (Su,u ) . ku5 ∇5 (Ωχ̂, Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
kψ kL2u L2u L∞ (Su,u )
1 1 1 1 1 1 1 1
1 1 u2 a4 u2 a2 O 1 1 u2 a4 u2 a4 1 1
.u 2 a 4 1 + 1 Õ · 1 ≤ u 2 a2
1 1 + 1 ÕO ≤ u 2 a 2 .
|u| 2 Ω |u| 2 |u| 2 |u| 2 Ω
For kΩui+1 G1,i kL1u L2u L2 (Su,u ) , the other terms obey
X 1 ua 2 O 12 12 1 1
1
1 1
kΩui+1 ∇i1 ψ i2 +1 ∇i3 (K − 2
, σ̌)kL1u L2u L2 (Su,u ) ≤ |u| u · u 2 a 2 ≤ u 2 a 2 ,
i1 +i2 +i3 =i
|u| |u|2
X 1
ua 2 O a 2 O 1
1
1 1
kΩui+1 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2u L2 (Su,u ) ≤ 2
· · |u|2 u 2 ≤ u 2 a 2 ,
i1 +i2 +i3 =i+1
|u| |u|
i1 ,i3 ≤i
X 1
ua 2 O a 2 O 1 1
1
1
i i1 i2 +1 i3
ku ∇ ψ ∇ ψ kL1u L2u L2 (Su,u ) ≤ · · |u|2 u 2 ≤ u 2 a 2 ,
i1 +i2 +i3 =i
|u|2 |u|
X 1
ua 2 · ua 2 O2 1
1
1 1
i+1 i1 i2 +2
kΩu ∇ ψ kL1u L2u L2 (Su,u ) ≤ · |u|2 u 2 ≤ u 2 a 2 ,
i1 +i2 =i+1
|u|4
i1 ≤i
X ua 2 O
1
1 1 1
i−1 i1 i2 +1
kΩu ∇ ψ k L1u L2u L2 (Su,u ) ≤ · |u|2 u 2 ≤ u 2 a 2 .
i1 +i2 =i
|u|3
We then proceed to derive the estimates for G2,i and G3,i . Since all the terms of G2,i are
contained in the expression for G3,i , hence we only need to bound the terms in G3,i . The terms
containing the 5 derivatives of the Ricci coefficients are
Ωu5 ψ ∇5 ψ , Ωu5 ψ ∇5 η,
and they obey

3 1
5 5 5 5 1 1 u2 a2 O 1 1
kΩu ψ ∇ ψ kL1u L2u L2 (Su,u ) .ku ∇ ψ kL∞ 2 2
u Lu L (Su,u )
kΩψ kL2u L2u L∞ (Su,u ) . u a Õ 2 2
3 ≤ u2 a2 ,
|u| 2
kΩu5 ψ ∇5 ηkL1u L2u L2 (Su,u ) .ku5 ∇5 ηkL∞ 2 2

u Lu L (Su,u )
kΩψ kL2u L2u L∞ (Su,u )
1 1 1 1 1 1 1 1
1 1 u2 a4 u2 a2 1 1 u2 a4 u2 a4 1 1
.u a
2 4 1+ 1 Õ · 1 ≤ u2 a2 1 1+ 1 Õ ≤ u 2 a 2 .
|u| Ω
2 |u| 2 |u| 2 |u| Ω
2
The rest terms in kΩui+1 (G2,i , G3,i )kL1u L2u L2 (Su,u ) are lower order and they satisfy
X 1
1
ua 2 O 1 1 1 1
kΩui+1 ∇i1 ψ i2 ∇i3 ψ ∇i4 (K − 2
, σ̌)kL1u L2u L2 (Su,u ) ≤ · u2 a2 ≤ u2 a2 ,
i1 +i2 +i3 +i4 =i
|u| |u|
X 1
ua 2 O a 2 O
1
3 1 1
kΩui+1 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2u L2 (Su,u ) ≤ 2
· · |u| 2 u ≤ u 2 a 2 ,
i1 +i2 +i3 =i+1
|u| |u|
i1 ,i3 ≤i
X a2
1
3 1 1
kΩui−1 ∇i1 ψ i2 ∇i3 ψ kL1u L2u L2 (Su,u ) ≤ |u| 2 uO ≤ u 2 a 2 .
i1 +i2 +i3 =i
|u|2
Back to (8.10), we hence prove this proposition.

We then utilize the other paired equations and proceed to prove
(2.39), (2.40), (2.41), it holds
X 1 1 A 1

i+2 i i+2 i
ku ∇ (K − 2 − ∇ φ∇A φ, σ̌)kL∞ 2 2
u Lu L (Su,u )
+ ku ∇ (β + ∇3 φ∇φ)kL∞ 2 2
u Lu L (Su,u )
|u| 4 2
1≤i≤4
1 1
3 3 u2 a4
.u 2 a 4 1 + 1 .
|u| 2 Ω
By further employing the L2 (Su,u ) estimates proved in Section 5.1 for φ, it also holds
X 1

3 3
1
u2 a4
1
i+2 i i+2 i
ku ∇ (K − 2 , σ̌)kL∞ 2 2
u Lu L (Su,u )
+ ku ∇ βkL∞ 2 2
u Lu L (Su,u )
.u a 1+
2 4
1 .
|u| |u| 2 Ω
1≤i≤4
Proof. We first commute the renormalized null Bianchi equations with i angular derivatives and
get
3+i 1
∇3 ∇i σ̌ + trχ∇i σ̌ + div ∗ ∇i (β + ∇3 φ∇φ) = F1,i , (8.12)
2 2
where F1,i is
X Ω−1 X
F1,i =∇i+1 (e3 φ)∇φ + Ω−1 ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 +1 ∇i3 σ̌,
i1 +i2 +i3 =i+1
|u| i1 +i2 +i3 =i−1
1 1 1 3+i 1 1
∇3 ∇i (K − − ∇A φ∇A φ)−div ∇i (β + ∇3 φ∇φ)+ trχ∇i (K − 2 − ∇A φ∇A φ) = F2,i
|u|2 4 2 2 |u| 4
(8.13)
62 XINLIANG AN
with F2,i being

Ω−1 i X
F2,i =∇i+1 (e3 φ)∇φ + ∇ µ + Ω−1 ∇i1 ψ i2 +1 ∇i3 ψ
|u| i 1 +i2 +i3 =i+1
X 1 1
+ Ω−1 ∇i1 ψ i2 +1 ∇i3 (K − 2
− ∇A φ∇A φ)
i1 +i2 +i3 =i
|u| 4
Ω−1 X Ω−1 X
+ ∇i1 ψ i2 +1 ∇i3 ψ + ∇i1 ψ i2 ∇i3 (Ωtrχ − (Ωtrχ)0 )
|u| i1 +i2 +i3 =i
|u|2 i1 +i2 +i3 =i
X 1 i1 i2 +1 i3 1 1
+ Ω−1 ∇ ψ ∇ (K − 2 − ∇A φ∇A φ),
i1 +i2 +i3 =i−1
|u| |u| 4
1 1 1
∇4 ∇i (β + ∇3 φ∇φ) − div ∇i (K − 2 − ∇A φ∇A φ) −∗ ∇∇i σ̌ = F3,i , (8.14)
2 |u| 4
where F3,i satisfies
F3,i =Ω−2 (Ωtrχ)∇i+1 (Ωtrχ) + Ω−2 · Ωχ̂∇i+1 (Ωχ̂, Ωtrχ) + Ω−2 · Ωχ̂∇i+1 (Ωχ̂)
+ Ω−2 (Ωtrχ)∇i+1 (Ωtrχ) + Ω−2 · Ωω̂∇i+1 (Ωχ̂, Ωtrχ − (Ωtrχ)0 ) + ∇i+1 (∇φ)∇φ
X
+ Ω−2 ∇i1 ψ i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 (Ωtrχ, Ωχ̂, Ωω̂)
i1 +i2 +i3 +i4 =i+1
i3 ,i4 ≤i (8.15)
X Ω−2 X
+ ∇i1 ψ i2 +1 ∇i3 (K, σ̌) + ∇i1 ψ i2 +1 ∇i3 ψ
i1 +i2 +i3 =i
|u| i1 +i2 +i3 =i
X X
i1 +1 i2 +1 −2
+ ∇ ψ∇ ψ +Ω ∇i1 ψ ∇i2 ψ ∇i3 ψ .
i1 +i2 =i−1 i1 +i2 +i3 =i
We then applying Proposition 6.2 for φ = ui+2 ∇i (β + 21 ∇3 φ∇φ) and Proposition 6.4 for
φ = ∇i (K − |u|1 2 − 41 ∇A φ∇A φ) and φ = ∇i σ̌ with λ1 = i + 2 and add these identities together.
We then obtain
1 1 1
kui+2 ∇i (K − − ∇A φ∇A φ, σ̌)k2L2u L2 (Su,u ) + kui+2 ∇i (β+ ∇3 φ∇φ)k2L2u L2 (Su,u )
|u|2 4 2
1 1
.k∇i (K − 2 − ∇A φ∇A φ, σ̌)k2L2u L2 (S0,u ) + ku2i+4 ∇i σ̌·ΩF1,i k2L1u L1u L1 (Su,u )
|u| 4
1 1 1
+ ku2i+4 ∇i (K − 2 − ∇A φ∇A φ)·ΩF2,i kL1u L1u L1 (Su,u ) + ku2i+4 ∇i (β+ ∇3 φ∇φ)·ΩF3,i kL1u L1u L1 (Su,u ) .
|u| 4 2
(8.16)
We proceed to control terms on the right. For the term involving F3,i we have
1
ku2i+4 ∇i (β+ ∇3 φ∇φ)·ΩF3,i kL1u L1u L1 (Su,u )
2
i+2 i 1
≤ku ∇ (β+ ∇3 φ∇φ)kL∞ 2 2
u Lu L (Su,u )
kΩui+2 F3,i kL1u L2u L2 (Su,u ) .
2
In kΩui+2 F3,i kL1u L2u L2 (Su,u ) , we note that the borderline terms are from the first two lines of
(8.15). These borderline terms obey
ku6 ∇5 (Ωtrχ)trχkL1u L2u L2 (Su,u ) .kΩ−1 u5 ∇5 (Ωtrχ)kL1u L2u L2 (Su,u )
1
≤ku6 ∇5 (Ωtrχ)kL∞ 2 2
u Lu L (Su,u )
·k kL2 L2 L2 (Su,u )
|u|Ω u u
1 1 1 1 1
3 u2 3 3 u2 a4 3 3 u2 a4
≤u aÕ(trχ) ·
2
1 ≤u a ·
2 4
1 R(β) . u a ·
2 4
1 ,
|u| 2 Ω |u| 2 Ω |u| 2 Ω
ku6 ∇5 (trχ, χ̂)(Ωtrχ, Ωχ̂, Ωω̂)kL1u L2u L2 (Su,u ) + ku6 ∇5 (Ωχ̂, Ωtrχ)χ̂kL1u L2u L2 (Su,u )
≤ku5 ∇5 (trχ, χ̂)kL∞ 2 2
u Lu L (Su,u )
· kuψ kL1u L∞
u L (Su,u )
∞
+ ku5 ∇5 (Ωχ̂, Ωtrχ)kL∞ 2 2

u Lu L (Su,u )
· kuΩ−1 (Ωχ̂)kL2u L2u L∞ (Su,u )
1 1 1
1 1 u2 a4 ua 2 1 1
1 1 1
≤u a (1 +2 4
1 )R(β) · ua + u a R(β) · 2u 2 |u| 2
2 2
|u| Ω
2 |u|Ω
1 1 1 1
3 3 u2 a4 3 3 u2 a4
≤u 2 a 4 1 + 1 R(β) . u 2 a4 1 + 1 .
|u| 2 Ω |u| 2 Ω
Here we employ Proposition (7.2), Proposition (7.5) and Proposition (8.1). And to control
kΩχ̂kL∞ (Su,u ) , we utilize
1 1
(Ω∇4 )(Ωχ̂)AB + Ωtrχ(Ωχ̂)AB =Ω2 (∇⊗η)
b AB − Ωtrχ(Ωχ̂)AB + Ω2 (η ⊗η)
b AB
2 2
1
+ Ω2 ∇A φ∇B φ − Ω2 gAB ∇C φ∇C φ.
2
Applying Proposition (4.2), it holds
1 1
u u ua 2 (O + O2 ) ua 2
kΩχ̂kL∞ (Su,u ) ≤ kΩχ̂kL∞ (Su,u ) + · 2
≤ .
|u| |u| |u| |u|
Similarly, we bound the rest terms in kΩui+2 F3,i kL1u L2u L2 (Su,u ) and they obey
kΩui+2 ∇i+1 ∇φ∇φkL1u L2u L2 (Su,u ) ≤ kui+1 ∇i+1 ∇φkL∞ 2 2

u Lu L (Su,u )
kuΩ∇φkL2u L2u L∞ (Su,u )
1 1 1 1 1 1
1 1 u2 a4 ua 2 1 1 3 3 u2 a4 u2 3 3
≤u 2 a 4 1 + 1 S · |u| 2 u2 O = u2 a4 1 +
1 1 SO ≤ u a ,
2 4
|u| 2 Ω |u| |u| 2 Ω |u| 2
X
kΩ−2 ui+2 ∇i1 ψ i2 ∇i3 (Ωχ̂, Ωtrχ − (Ωtrχ)0 )∇i4 (Ωtrχ, Ωχ̂, Ωω̂)kL1u L2u L2 (Su,u )
i1 +i2 +i3 +i4 =i+1
i3 ,i4 ≤i
1 1 1 1
3 3 u2 a4 3 3 u2 a4
≤u 2 a 4 1 + 1 R(β) ≤ u 2 a 4 1 + 1 ,
|u| 2 Ω |u| 2 Ω
X
kΩ−2 ui+1 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
1 1 1 3 3
ua 2 a 2 ua 2 2 3 3 3 u 2 a 4 O2 3 3
≤Ω−2 |u| · 2
· O · u|u| 2 ≤ u 2 a 4 · 3 ≤ u2 a4 ,
|u| |u| |u| |u| 2 Ω2
X
kui+2 ∇i1 ψ i2 +1 ∇i3 (K, σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
X 1 X
≤ kΩui3 +1 ∇i3 (K −
2
, σ̌)kL∞ 2 2
u Lu L (Su,u )
· kΩ−1 ui1 +i2 +1 ∇i1 ψ i2 +1 kL1u L∞
u L (Su,u )
∞
|u|
i3 ≤4 i2 ≤2
X X
+ kui3 +2 ∇i3 (K, σ̌)kL∞ u Lu L (Su,u )
∞ ∞ · kui1 +i2 ∇i1 ψ i2 +1 kL1u L2u L2 (Su,u )
i3 ≤2 i1 ≤4
1 1 1 1
1 1 ua O 2 ua O 32 3 3 ua 4 RO 3 1 u2 O 3 3
≤u 2 a 2 R · ·u+ 2
· u|u| 2 ≤ u 2 a 4 · + u2 a2 · 1 ≤ u a ,
2 4
|u|Ω |u| |u|Ω |u| 2
X u3 aO2 3 3
kΩui+2 ∇i1 +1 ψ ∇i2 +1 ψ kL1u L2u L2 (Su,u ) ≤ 3 ≤ u2 a4 ,
i1 +i2 =i−1 |u| 2
X 3 3
3 3
u 2 a 4 O3 3 3
−1 i+2 i1 i2 i3
kΩ u · ∇ ψ ∇ ψ ∇ ψ kL1u L2u L2 (Su,u ) ≤ u a · 2 4
3 ≤ u2 a4 .
i1 +i2 +i3 =i
|u| Ω 2
64 XINLIANG AN
In the similar manner, we also control terms in F2,i and F1,i . In kΩui+2 F2,i kL1u L2u L2 (Su,u ) and
kΩui+2 F1,i kL1u L2u L2 (Su,u ) , the top-order derivatives appear in the next four terms. The first term
satisfies
kui+2 ∇i+1 (Ωe3 φ)∇φkL1u L2u L2 (Su,u )
≤kΩ−1 ui+1 ∇i+1 (Ωe3 φ)kL∞ 2 2
u Lu L (Su,u )
kuΩ∇φkL2u L2u L∞ (Su,u )
1 1 1 1 1 1
1 1 u2 a4 ua 2 1 1 3 3 u2 a4 u2 3 3
≤u 2 a 4 1 + 1 S· |u| 2 u 2 ≤ u 2 a 4 1 + 1 1 S ≤ u a .
2 4
|u| 2 Ω |u| |u| 2 Ω |u| 2

The second term obeys
kui+1 ∇i µkL1u L2u L2 (Su,u )
1 1 1 1 1
5 3 3 u2 a4 u2 3 3 u2 a4
≤kui+ 2 ∇i µkL∞ ∞ 2
u Lu L (Su,u )
ku− 2 kL1u L2u L∞ (Su,u ) ≤ ua 4 · · 1 = u a
2 4 ·
1
Ω |u| 2 |u| 2 Ω
where Proposition (7.4) is used. For the third term, we have
X X
kui+2 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2u L2 (Su,u )
i≤4 i1 +i2 +i3 =i+1
X X
1
. kui1 +i2 +2 ∇i1 ψ i2 +1 kL∞
u Lu L (Su,u )
∞ ∞ · u2 kui3 +1 ∇i3 ψ kL∞ ∞ 2
u Lu L (Su,u )
ku−2 kL1u
i1 +i2 ≤5 i3 ≤4
i1 ≤2

1
+ u 2 kui+1 ∇i+1 (Ωtrχ, Ωχ̂)kL∞ 2 2
u Lu L (Su,u )
ku−1 kL2u + kui+2 ∇i+1 ηkL∞ 2 2
u Lu L (Su,u )
ku−2 kL1u
3 1 1 1 1 3 3
1 u2 a2 O u 2 12 41 u2 a4 1 u 2 a 4 Õ 3 3
.ua O · 2 + 1 u a 1+ 1 + ua O ·
2 ≤ u2 a4 ,
|u| |u| 2 |u| Ω
2 |u|
(8.17)
where for the first inequality of (8.17) we use the top-order terms are of form ψ ∇i+1 (Ωχ̂, Ωtrχ)
and ψ ∇i+1 η. And for the fourth term, it holds
X 1 1
kΩ−1 ui+2 ∇i1 ψ i2 +1 ∇i3 (K − 2
− ∇A φ∇A φ)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
|u| 4
1 1 1
ua O
2 3 3 u2 a4 3 3
≤ · u2 a4 · 1 + 1 R ≤ u2 a4 .
|u|Ω |u| 2 Ω
For the rest terms in kΩui+2 F2,i kL1u L2u L2 (Su,u ) and kΩui+2 F1,i kL1u L2u L2 (Su,u ) , we have
X 1 1
ua 2 ua 2 O2 2 12 3 3 ua 4 O2 3 3
1
kui+1 ∇i1 ψ i2 +1 ∇i3 ψ kL1u L2u L2 (Su,u ) ≤ |u| · 2

· 2
|u| u = u 2 a 4 · ≤ u2 a4 ,
i1 +i2 +i3 =i
|u| |u| |u|
X 1 1
kΩ−1 ui+1 ∇i1 ψ i2 ∇i3 (K − − ∇A φ∇A φ, σ̌)kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i−1
|u|2 4
1 1 1
ua 2 ua 2 O2 2 1 3 3 ua 4 3 3
≤Ω−1 |u|2 · 2
· 3
|u| u 2 ≤ u 2 a 4 · ≤ u2 a4 ,
|u| |u| |u|Ω
X
kui ∇i1 ψ i2 ∇i3 (Ωtrχ − (Ωtrχ)0 )kL1u L2u L2 (Su,u )
i1 +i2 +i3 =i
X 1
u2
i+ 23 i1 i2 i3 − 32 3 3 3 3
≤ku ∇ ψ ∇ (Ωtrχ − (Ωtrχ)0 )k L∞ ∞ 2
u Lu L (Su,u )
ku k L1u L2u L∞ (Su,u ) ≤u a ·
2 4
1 ≤ u2 a4 ,
i1 +i2 +i3 =i |u| 2
where Proposition (4.6) is employed.

Gathering the above estimates, together with Proposition (8.1) we then obtain
X 1

kui+2 ∇i (K − 2 , σ̌)kL2u L2 (Su,u ) + kui+2 ∇i βkL2u L2 (Su,u )
|u|
i≤4
1 1 1 1
3 3 u2 a4 3 3 u2 a4
.u 2 a 4 1 + 1 (1 + R[β]) . u 2 a 4 1 + 1 .
|u| 2 Ω |u| 2 Ω

Combining Propositions 8.2, 8.1, 8.3, we thus conclude
R . 1. (8.18)
Substituting this bound (8.18) into Proposition 7.6, we furthermore acquire
Õ5,2 . 1. (8.19)
This proves the hyperbolic part of Theorem 1.1.
1
We further derive an additional estimate for K − |u|2 − 14 ∇A φ∇A φ.
(2.39), (2.40),(2.41), it holds
1 1 3 3
ku2 (K − 2 − ∇A φ∇A φ)kL∞ 2 2
u Lu L (S)
.u 2 a 4 .
|u| 4
Proof. Applying Proposition 4.7 and the bootstrap assumption (2.40), we get
1 1 1 1 3 3
ku2 (K − 2 − ∇A φ∇A φ)kL∞ 2 2
u Lu L (S)
. u 2 · ua 2 = u 2 a 4 .
|u| 4

The above sections finish the hyperbolic estimates. Together with a standard local existence
and extension argument, we have obtained a desired spacetime existence region. We now move
to the elliptic part and construct the apparent horizon within this obtained existence regime.
9. Subsolutions and Supersolutions for the MOTS

Along each fixed H u , we consider a 2-sphere (u, u, ω) on it with ω ∈ S2 and u satisfying
u = −R(ω, u). Considering R = R(ω, u) as a function on S2 , we define
1 1
G(R, u) := ∆g R + Ωtrg χ|∇g R|2 − Ω−1 trg χ
2 2 (9.1)
2
+ 4Ωω|∇g R| + 2Ωχ̂(∇g R, ∇g R) + 2g(η, ∇g R),
where g is the induced metric on the 2-sphere M = {u = −R} along H u .
When there is no confusion, we suppress u and write R = R(ω). Following [6], along H u it
holds that 

MOTS if G(R, u) = 0 on S2 ,
u = −R(ω) is a trapped surface if G(R, u) > 0 on S2 ,


untrapped surface if G(R, u) < 0 on S2 .
A remaining main goal of this article is to solve R = R(ω, u), which satisfies G(R, u) = 0 for
each u. In later sections, we will also study the regularity of R = R(ω, u) in terms of u.
To solve for R = R(ω, u), we first construct the anisotropic sub- and supersolutions to
G(R, u) = 0. With u = −R(ω, u), on Su,u we define γ and φ via
∗
g = R2 γ and R = (ua)1−µ e−φ .
Remark 20. For perturbed Christodoulou’s initial data prescribed along u = 0, we can define
the corresponding φ via letting
φ
R · Ω(−R, 0)2 = uae− 1−µ∗ .
66 XINLIANG AN
Hu
u = −R(u, ω)
Mu (ω)
ω ∈ S2 u = −1
Figure 6
Remark 21. Compared with [6] by An-Han, there µ∗ = 0 and a ≫ 1. In this paper, we have
0 < µ∗ < 1 and we can set a = 1. Two major observations in below elliptic arguments are that
i) owing to Christodoulou’s initial data, a potentially dangersome term is with a favored sign
and hence the approach to construct the multi- and single-valley anisotropic trapped surface as
in [6] and [28] still works. ii) even though the borderline quadratic nonlinear term with gradient
is no longer negligible as 1/a fails to be the small parameter, the method of deriving apriori
estimates via Moser’s iteration developed in [6] by An-Han is robust enough to be extended to
this new setting. Besides these observations, a novel ingredient in the step of linearization is a
new ansatz for the to-be-solved solution and the use of it plays a vital role.
In below, we provide the details. Note that on Su,u it holds
∆g φ = R−2 ∆γ φ.
And equation G(R, u) = 0 can be rewritten as
0 = ∆γ φ − |∇γ φ|2 − 4ΩωR|∇γ φ|2 + 2γ ij ηi ∂j φ + 2R−1 Ωχ̂kl γ ik γ jl ∇i φ∇j φ
R R (9.2)
− Ωtrg χ|∇γ φ|2 + Ω−1 trg χ.
2 2
We point out that the main contribution in Ω−1 trg χ is
2 uaf (u, ω)
· (1 − µ∗ ) − 2 2 · (1 − µ∗ )
R R Ω0 (R)
µ∗
and Ω20 (R) := Ω2 (−R, 0) = R 1−µ∗ . We then write 12 RΩ−1 trg χ as
1 uaf (u, ω) · (1 − µ∗ ) 1 uaf (u, ω) · (1 − µ∗ )
RΩ−1 trg χ = 1 − µ∗ − 2 + [ RΩ−1 trg χ − (1 − µ∗ ) + ].
2 2RΩ0 (R) 2 2RΩ20 (R)
Together with the fact
µ∗ 1 ∗ 1 φ
RΩ20 (R) = R · R 1−µ∗ = R 1−µ∗ = [(ua)1−µ e−φ ] 1−µ∗ = uae− 1−µ∗ , (9.3)
we can rewrite G(R, u) = 0 as
1 φ
0 = ∆γ φ + (1 − µ∗ ) − f (u, ω)(1 − µ∗ )e 1−µ∗ − 4ΩωR|∇γ φ|2
2
R
− (1 + Ωtrg χ)|∇γ φ|2 + 2γ ij ηi ∂j φ + 2R−1 Ωχ̂kl γ ik γ jl ∇i φ∇j φ
2
1 1 φ
+ [ RΩ−1 trg χ − (1 − µ∗ ) + f (u, ω)(1 − µ∗ )e 1−µ∗ ].
2 2
Multiplying 1/(1 − µ∗ ) on both sides of the equation and setting
φ
φe := ,
1 − µ∗
we then transfer G(R, u) = 0 into
e u) =∆γ φe + 1 − 1 f (u, ω)eφe − 4(1 − µ∗ )ΩωR|∇γ φ|

0 = S(φ, e2
2
− (1 + Ωtrg χ)(1 − µ∗ )|∇γ φ|
2
e
e j φ+[ R 1 e
+ 2(1 − µ∗ )R−1 Ωχ̂kl γ ik γ jl ∇i φ∇ Ω−1 trg χ − 1 + f (u, ω)eφ ].
2(1 − µ∗ ) 2
(9.4)
And we denote
e := − (1 + R Ωtrg χ)(1 − µ∗ )|∇γ φ|
F (φ) e 2 + 2γ ij ηi ∂j φe
2
(9.5)
e
e j φ+[ R 1 e
+ 2(1 − µ∗ )R−1 Ωχ̂kl γ ik γ jl ∇i φ∇ Ω−1 trg χ − 1 + f (u, ω)eφ ].
2(1 − µ∗ ) 2
Applying g = R2 γ and the estimates in Section 3.1, we have that γ satisfies
1 p 3
≤ |γ|(u, ω) ≤ , Λ−1 I ≤ γij (u, ω) ≤ ΛI, (9.6)
2 2
|Dω γij (u, ω)|+R|∂u γij (u, ω)| ≤ Λ, (9.7)
|Dω2 γij (u, ω)| + R|Dω ∂u γij (u, ω)| + R 2
|∂u2 γij (u, ω)| ≤Λ (9.8)
with Λ being a uniform positive constant.
To proceed, we use the below lemma in [6]:
Lemma 9.1. (Lemma 3.5 in [6]) Let {γ(s)} be a family of metrics and {φ(s)} be a family of
functions, both parametrized by s. Then,
1
(∆γ φ)· = ∆γ φ̇ − γ ij γ kl γ̇jl ∇2ik φ − γ ij γ kl ∇i γ̇jl ∇k φ + γ ij γ kl ∇l γ̇ij ∇k φ, (9.9)
2
d
where we denote ˙ = ds .
As a corollary, via the same proof as stated in Lemma 3.6 of [6], it holds
Lemma 9.2. Set φe ∈ C 2 (S2 ) to be an arbitrary positive function. Then, we have
e ≤ C(1 + φ)(|∇
|∆γ φe − ∆γ0 φ| e 2e e e 2 φe − 13 Ωδ̃
γ φ| + |∇γ φ| + |∇γ φ| )e a (9.10)
e
with C a positive constant and δ̃ a small positive constant, independently of a and φ.
1
Remark 22. Note that the additional a− 3 Ωδ̃ is gained by applying
1 1 1
ua 2 ua 2 Ω2 ua 2 Ω2 e 1 e 1
= 2
. = eφ a− 2 Ω2 ≪ eφ a− 3 Ωδ̃
|u| |u|Ω uae−φe
in the proof. Here for the first inequality we use (9.3). And for the last inequality, if setting
a = 1, the whole elliptic arguments would be developed in the interior region, hence we would
employ the property 0 < Ω(u, 0) ≤ Ω(u1 , 0) ≪ 1.
∗ ∗ e
Recall that, with R(ω) = (ua)1−µ e−(1−µ )φ it holds

 e u) = 0 on S2 ,
MOTS if S(φ,
u = −R(ω) is a trapped surface e u) < 0 on S2 ,
if S(φ,


untrapped surface if S(φ,e u) > 0 on S2 .
e
Back to (9.4), to construct the subsolution (untrapped surface), we just set φ(ω) = ln( 32 ). Note
0 ≤ f ≤ 1. Using (9.5) and hyperbolic estimates, we have
e 1
e ≤ Ceφ a− 3 Ωδ̃ (|∇γ φ|
e 2 + 1).
|F (φ)|
For Ω being sufficiently small in the interior, it holds
3 f3 3 3 3
S(ln( ), u) = 1 − + F (ln( )) ≥ 1 − + F (ln( )) > 0. (9.11)
2 22 2 4 2
68 XINLIANG AN
∗ ∗ e ∗
Hence, the 2-sphere Su,u with u = −(ua)1−µ e−(1−µ )φ
= ( 23 ua)1−µ is an untrapped surface.
¯
To construct the supersolution (trapped surface), we need S(φ̃, u) < 0. Here we apply a key
property from the hyperbolic parts:
1 µ∗ 1
Ωω(u, u) ≥ Ωω(u, 0) = ∗
· > 0.13
2 8(1 − µ ) r(u, 0)
¯ it suffices to require
Thus, back to (9.4) to find the supersolution φ̃,
¯ 1 ¯ R ¯
∆γ φ̃ + 1 − f (u, ω)eφ̃ − (1 + Ωtrg χ)(1 − µ∗ )|∇γ φ̃|2
2 2
¯ ¯ ¯ R 1 ¯
+ 2γ ij ηi ∂j φ̃ + 2(1 − µ∗ )R−1 Ωχ̂kl γ ik γ jl ∇i φ̃∇j φ̃+[ Ω−1 trg χ − 1 + f (u, ω)eφ̃ ] < 0.
2(1 − µ∗ ) 2
Notice that the one-valley supersolution φ̃¯ given in Lemma 3.7 of [6] and the multi-valley superso-
¯ ¯
lution φ̃ constructed in Theorem 7.4 of [6] fulfill this requirement. Since −4(1−µ∗ )ΩωR|∇γ φ̃|2 ≤
¯ ¯
0, with these φ̃ we also have S(φ̃, u) < 0. As listed in Lemma 3.7 of [6], for any τ > 2.2, the
¯
constructed φ̃ obeys
φ̃¯min = − log(mǫ2 ) + C(τ ) and φ̃¯max = − log(mǫτ +2 ) + C(τ ). (9.12)

¯ we have that the 2-sphere S
Adopting this φ̃,
¯
1−µ∗ −(1−µ∗ )φ̃
u,u with u = −(ua) e is a trapped
surface and it holds
∗ ∗
−(ua · mǫ2 )1−µ ≤ u ≤ −(ua · mǫτ +2 )1−µ .
In below, for equation (9.4) we seek the solution φe bounded between φ̃ = ln( 23 ) and φ̃¯
constructed above as in [6]. We require that φe satisfies
3
0 < ln( ) ≤ φe ≤ κ (9.13)
2
¯ Here κ = κ(m, ǫ) depends on constants m
with κ being the maximum of the supersolution φ̃.
and ǫ stated in Theorem 1.1. We further define constant K via
ē C 2 (S2 ) }
K := max{C(κ), |φ| (9.14)
with
4Λκeκ
C(κ) = (Ceκ )Ce (9.15)
being explicitly given in Theorem 10.4.
Throughout this paper, we will use parameters a, function Ω and constant κ, which satisfy
4Λκ ¯ 1 1
K = max{(Ceκ )Ce , kφ̃kC 2 } ≤ a 4 · Ω−δ̃/2 ≪ a 3 · Ω−δ̃ . (9.16)
Here, C is a uniform positive constant defined in Theorem 10.4, and Λ is introduced in (9.6).
Note that for the interior region, we have a = 1 and 0 < Ω . Ω(u, 0) ≤ Ω(u1 , 0) ≪ 1. When
κ = κ(m, ǫ) is fixed, in both regions we choose either parameter a being sufficiently large or
0 < Ω ≤ Ω(u, 0) being sufficiently small, such that (9.16) holds.
Remark 23. We can set a = 1 and the above arguments hold. Moreover, the hypersurface
∗
u = −(ua · mǫτ +2 )1−µ locates within the hyperbolic existence region.
13Note that the key property Ωω(u, 0) > 0 also holds for perturbed initial data.
10. Regularity Estimates via Moser’s Iteration

∗ ∗ e
With u = −R and R = (ua)1−µ e−(1−µ )φ
, via (9.4), S−R,u being a MOTS requires
1 e e 2 − (1 + R Ωtrg χ)(1 − µ∗ )|∇γ φ|
0 =∆γ φe + 1 − f (u, ω)eφ − 4(1 − µ∗ )RΩω|∇γ φ| e 2 + 2γ ij ηi ∂j φe
2 2
1 1 e e
e j φ.
+[ RΩ−1 trg χ − 1 + f (u, ω)eφ ] + 2(1 − µ∗ )R−1 Ωχ̂kl γ ik γ jl ∇i φ∇
2(1 − µ∗ ) 2
(10.1)
And we denote
F1 = − 4(1 − µ∗ )RΩω|∇γ φ|e 2, (10.2)
F2 = − (1 + Ωtrg χ)(1 − µ∗ )|∇γ φ| (10.3)
2
1 1 e e
e j φ.
+[ RΩ−1 trg χ − 1 + f (u, ω)eφ ] + 2(1 − µ∗ )R−1 Ωχ̂kl γ ik γ jl ∇i φ∇
2(1 − µ∗ ) 2
Our aim in this section is to prove Theorem 10.4. We take three steps. In the first step, we
control the Hölder norms of the solutions via using the local boundedness and the weak Harnack
inequality due to Moser. In the second step, we estimate the Hölder norms of solutions’ first
derivatives. This step is established based on an integral characterization of Hölder continuous
functions. In the third step, we bound the Hölder norms of solutions second derivatives. During
the proof, we track the dependence on κ.
With (10.2) and (10.3), the above equation (10.1) can be rewrite as
1
−∆γ φ̃ = 1 − f eφ̃ + F1 + F2 (10.4)
2
with φe = φ(ω)
e being the solution and 0 ≤ f ≤ 1 on Su,u . Note that (10.4) is a quasilinear
e F contains the gradient of φ̃.
elliptic equation for φ̃; γ depends on φ;
Remark 24. For notational simplicity, we set Br (p) to be the standard ball centered at p with
radius r in R2 , which lies in a coordinate chart of S2 . For notational simplicity, when there is
no danger of confusion, we also write Br (p) ⊂ S2 . When the center is located at the origin, we
often write Br for short. When emphasizing the radius r, we also employ the notation Bp (r).
With coordinates, we can rewrite 10.4 as
p h 1 i
−∂j aij ∂i φ̃ = |γ| 1 − f eφ̃ + F1 + F2 . (10.5)
2
By the hyperbolic part, we have

Λ−1 I ≤ aij (ω, φ̃) ≤ ΛI,
p
for some positive constant Λ, and Λ−1 ≤ |γ|(ω, φ̃) ≤ Λ.
Allowing us to abuse notations slightly, for any (ω, φ̃, p) ∈ B1 ×R×Rn, applying the hyperbolic
estimates, we can also write
1
F2 (ω, φ̃, p) = a− 3 Ωδ̃ eφ̃ cij pi pj + ci pi + c0
with cij , ci , c0 being functions of (ω, φ̃) on B1 × R+ . Denoting pi := Di φ̃ and Dφ̃ := ∇φ̃, the
Cauchy-Schwarz inequality then implies
1
|F2 | ≤ Ca− 3 Ωδ̃ eφ̃ (|Dφ̃|2 + 1) ≤ c(|Dφ̃|2 + 1)
with c a small positive constant.
e 2 and RΩω ≥ µ∗
Since F1 = −4(1 − µ∗ )RΩω|∇γ φ| 8(1−µ∗ ) > 0, here we only have
|F1 | ≤ |Dφ̃|2 + 1.
While in [6], Ωω obeys the same upper bound as for Ωχ̂ and there |F1 | satisfies the same estimate
e 2 + 1) with c being a small positive constant as for |F2 | here. Hence it is negligible
|F1 | ≤ c(|Dφ|
in [6].
70 XINLIANG AN
In this paper, F1 is of size 1 and serves as the main term. A key observation of this paper
(also pointed out one line below (4.3) in [6]) is that the regularity estimates carried out in [6]
also work for the scenario
e 2 + 1) with c = 1.
|F1 | + |F2 | ≤ c(|Dφ| (10.6)
In below, we start to conduct these elliptic estimates. We then write (10.5) in divergence
form as Z Z p h i
1 e e Dφ)
e ψ̃ dω,
aij (ω, φ̃)Di φ̃Dj ψ̃ dω = |γ| 1 − f eφ + F (ω, φ, (10.7)
2
with ψ̃ being any function in H01 (B1 ) ∩ L∞ (B1 ). We also assume φ̃ = φ̃(ω) ∈ H 1 (B1 ) and
require
0 ≤ φ̃ ≤ κ on B1 (10.8)
with κ being some positive constant, which is related to the anisotropicity via the construction
of upper and lower barriers.
Our first goal is to bound the Hölder norm of φ̃ via Moser’s iteration. For the estimates, we
pay attention to their explicit dependence on κ. Following the steps in the proof of Theorem
4.2 in [6], we obtain
Proposition 10.1. Let φ̃ ∈ H 1 (B1 ) be a positive solution to (10.7) with (10.8) satisfied. Then,
it holds
[φ̃]C α (B1/2 ) ≤ Ceκ (10.9)
with C being a positive constant depending only on Λ. And α is given by
α = ǫ0 e−4Λκ ∈ (0, 1). (10.10)
Here ǫ0 > 0 is a small constant depending on Λ.
Proof. For s ∈ (0, 1), we set
M (s) = max φ̃, m(s) = min φ̃.
Bs Bs
Writing m = m(s) and M = M (s) for brevity, proceeding the same as in the reasoning for
Theorem 4.2 in [6], we get
s h s i
M − m(s) ≤ Ce4Λκ m − m(s) + Ce4Λκ+κ s2 , (10.11)
2 2
s h s i
M (s) − m ≤ Ce4Λκ M (s) − M + Ce4Λκ+κ s2 . (10.12)
2 2
For any s ∈ (0, 1), define
h(s) := M (s) − m(s). (10.13)
Adding (10.11) and (10.12), we then obtain
s h s i
h(s) + h ≤ Ce4Λκ h(s) − h + Ce4Λκ+κ s2 .
2 2
This implies
s Ce4Λκ − 1
h ≤ Cγ h(s) + eκ s2 with Cγ = < 1.
2 Ce4Λκ + 1
Using this Cγ , we pick up µ ∈ (0, 1) satisfying
α := (1 − µ) log Cγ / log(1/2) < 2µ.
Further applying Lemma 8.23 in [24] or Lemma 4.19 in [26], for any s ∈ (0, 1/2], we hence prove

h(s) ≤ Csα h(1) + eκ s2µ ≤ Ceκ sα , (10.14)
where we employ h(1) ≤ M (1) ≤ κ ≤ eκ for the second inequality. This concludes (10.9).

Next, we derive the gradient estimate for φ̃. By the hyperbolic part, we have
|Dω akl (ω, φ̃)| + |∂φ̃ akl (ω, φ̃)| ≤ Λ1 (10.15)
with Λ1 being some positive constant. The constant in the below theorem depends on Λ1 . And
we have
Proposition 10.2. Set φ̃ ∈ H 1 (B1 ) to be a positive solution to (10.7) with (10.8) satisfied.
Then, it holds
4Λκ
|φ̃|C 1,1/3 (B1/2 ) ≤ (Ceκ )Ce (10.16)
with C being a positive constant depending only on Λ and Λ1 .
Proof. The detailed steps for proving Theorem 4.3 in [6] can also be carried out here. For any
function ψ̃ belonging to H01 (B1 ) ∩ L∞ (B1 ) with ψ̃ ≥ 0, by (10.7) and (10.6), it holds
Z Z p h i Z
1
aij Di φ̃Dj ψ̃ dω ≤ |γ| 1 − f eφ̃ + |Dφ̃|2 + 1 ψ̃ dω ≤ [4 + 2|Dφ̃|2 ]ψ̃ dω.
2
Taking ξ(τ ) = 2−1 Λ−1 (e2Λτ − 1), we have
Z Z
aij Di (ξ(φ̃))Dj ψ̃ dω ≤ 4ξ ′ (φ̃)ψ̃ dω.
This is equivalent to
Z Z
aij e2Λφ̃ Di φ̃Dj ψ̃ dω ≤ 4e2Λφ̃ ψ̃ dω.
Taking ψ̃ = φ̃µ̃2 for some µ̃ ∈ C01 (B1 ) with µ̃ = 1 on B3/4 , via using the Cauchy-Schwarz
inequality, we get Z
|Dφ̃|2 dω ≤ Cκ2 e2Λκ (10.17)
B3/4
with C being a positive constant depending only on Λ.
To derive a bound for the Hölder norm of Dφ, e we need to employ the regularity of aij (ω0 , φ̃(ω0 ))−
aij (ω, φ̃(ω)), which was already obtained in Proposition 10.1. Following the procedures of prov-
ing Theorem 4.3 in [6], for any 0 < ρ ≤ s we then obtain
Z h ρ 2 iZ
|Dφ̃|2 dω ≤ C + h(s) + [h(s)]2 + s2 |Dφ̃|2 dω + Ce2κ s4 ,
Bρ (ω0 ) s Bs (ω0 )
where
e
h(s) = h(ω0 ; s) := sup φe − inf φ.
Bs (ω0 ) Bs (ω)
Take any δ ∈ (0, 1). For a small ǫδ > 0, we choose sδ small such that Bsδ (ω0 ) ⊂ B3/4 and
h(sδ ) + [h(sδ )]2 + s2δ < ǫδ . (10.18)
κ −C2 e4Λκ
By employing (10.14) and κ ≫ 1, if we set sδ = (C1 e ) , it satisfies the above requirement.
Applying Lemma 3.4 in [26], for any s ≤ sδ we then get
Z h Z i
|Dφ̃|2 dω ≤ Cs2δ s−2δ
δ |Dφ̃|2 dω + e2κ .
Bs (ω0 ) Bsδ (ω0 )
By (10.17), this further implies

Z

|Dφ̃|2 dω ≤ Cs2δ κ2 e2Λκ s−2δ
δ + e2κ . (10.19)
Bs (ω0 )
Proceeding the same as in [6], defining

Z
1
(Dφ̃)ω0 ,s := Dφ̃,
|Bs (ω0 )| Bs (ω0 )
it also holds
Z
′
|Dφ̃ − (Dφ̃)ω0 ,s |2 dω ≤ Cκ3 e3Λκ s−3δ
δ s2+2α
Bs (ω0 )
72 XINLIANG AN
if we further impose δ ∈ (2/3, 1) and α′ = 3δ/2 − 1.

Applying Theorem 3.1 in [26], together with (10.17), we deduce that, for any ω ∈ B1/2
3 3 − 3δ 3 3 − 3δ
|Dφ̃(ω)| ≤ C κ 2 e 2 Λκ sδ 2 + κeΛκ ≤ Cκ 2 e 2 Λκ sδ 2 , (10.20)
and for any ω1 , ω2 ∈ B1/2 satisfying |ω1 − ω2 | < sδ ,
3 3 − 3δ ′
|Dφ̃(ω1 ) − Dφ̃(ω2 )| ≤ Cκ 2 e 2 Λκ sδ 2
|ω1 − ω2 |α .
In addition, for any ω1 , ω2 ∈ B1/2 we also have
3 3 − 3δ ′ ′
|Dφ̃(ω1 ) − Dφ̃(ω2 )| ≤ Cκ 2 e 2 Λκ sδ 2
s−α
δ |ω1 − ω 2 |α . (10.21)
4Λκ
Recall that sδ = (C1 eκ )−C2 e and we have 0 < α′ < 1/2. We now choose α′ = 1/3. Utilizing
(10.20) and (10.21), we hence conclude
2Λκ
|Dφ̃|C 1/3 (B1/2 ) ≤ (Ceκ )Ce . (10.22)

We further prove the following Schauder estimates. We work under the assumption
|Dω2 akl (ω, φ̃)| + |Dω ∂φ̃ akl (ω, φ̃)| + |∂φ̃2 akl (ω, φ̃)| ≤ Λ2 (10.23)
with Λ2 being a positive constant.
Proposition 10.3. Suppose φ̃ ∈ H 1 (B1 ) is a positive solution to (10.7) with (10.8) satisfied,
and f ∈ C 1/3 (B1 ). Then, it holds
4Λκ
|φ̃|C 2,1/3 (B1/2 ) ≤ (Ceκ )Ce . (10.24)
p
Here C is a positive constant depending only on Λ, Λ1 , Λ2 , the C 1 -norms of |γ|, cij , ci , and
c0 on B1 × R+ , and the C 1/3 -norm of f on B1 .
Proof. We have already proved φe ∈ C 1,1/3 (B1/2 ) and we can write (10.7) in nondivergence form
aij (ω, φ̃)∂ij φ̃ = h(ω) (10.25)
with
p h 1 i
h(ω) = −∂j aij (ω, φ̃)∂i φ̃ − ∂φ̃ aij (ω, φ̃)∂i φ̃∂j φ̃ − e .
|γ| 1 − f eφ̃ + F (ω, φ̃, Dφ)
2
We now treat equation (10.25) as a linear equation of φ̃, and consider h as a given term defined
in B1 . Using the expression of F and Proposition 10.2, we get
4Λκ
|h|C 1/3 (B3/4 ) ≤ (Ceκ )Ce .
Employing the standard interior Schauder estimate, we then obtain
4Λκ
|φ̃|C 2,1/3 (B1/2 ) ≤ C∗ |φ̃|L∞ (B3/4 ) + |h|C 1/3 (B3/4 ) ≤ C∗ (Ceκ )Ce
with C∗ > 0 a constant depending only on Λ, and the C 1/3 -norm of aij on B3/4 .
Gathering the above three propositions just obtained, we are ready to prove
Theorem 10.4. With S(φ̃) being defined as in (9.4) and κ being a positive constant, if we
assume φ̃ is a solution to S(φ̃) = 0 satisfying 0 ≤ φ̃ ≤ κ on S2 , then it holds
4Λκ
|φ̃|C 2,1/3 (S2 ) ≤ (Ceκ )Ce ,
with Λ being the ellipticity constant for the metric component γij and C being a positive
constant depending only on
1 1 1 1 − µ∗ uaf (ω, u)
γ, χ̂, η, α′1 = −( Ωtrg χ + ), α2 = Ω−1 trg χ − + (1 − µ∗ )
2 R 2 R 2Ω20 R2
and their angular derivatives.
Proof. We first fix an arbitrary ball B on S2 . In local coordinates within B, we have (10.5) and
p
aij (ω, φ̃) = |γ(u, ω)|γ ij (u, ω).
∗ ∗
Here u = −R = −(ua)1−µ e−(1−µ )φ̃ . Recall that F = F1 + F2 with F1 = −4(1 − µ∗ )RΩω|∇γ φ̃|2
and F2 being a linear combination of ∇i φ̃∇j φ̃, ∇i φ̃, 1 with the below coefficients
γ ij , R−1 Ωχ̂kl , ηi , Rα′1 , Rα2 .

∗ ∗
The coefficients of F2 are functions of ω and u = −R = −(ua)1−µ e−(1−µ )φ̃ . To control the
C 1/3 -norm of F2 , we estimate these coefficients and their derivatives with respect to ω and φ̃.
With ∂φ̃ = R∂u , we hence need to bound
R−1 Ωχ̂kl , ηk , Rα′1 , Rα2 ,

R−1 ∇i (Ωχ̂kl ), ∇i ηk , R∇i α′1 , R∇i α2 ,
∂u (Ωχ̂kl ), R∂u ηk , R2 ∂u α′1 , R2 ∂u α2
with
1 1 1 −1 1 − µ∗ uaf (u, ω)
α′1 := −( Ωtrg χ + ), α2 := Ω trg χ − + (1 − µ∗ ). (10.26)
2 R 2 R 2Ω20 R2
1
By the hyperbolic estimates, all these quantities are bounded by eφ̃ a− 3 Ωδ̃ . The conditions
in Proposition 10.1, Proposition 10.2, and Proposition 10.3 all hold. Therefore, we derive the
desired C 2,1/3 estimates for solution φe to the equation of MOTS.
11. The Linearized Equation

To derive and to analyze the linearized equation, in contrast to [6] and previous sections,
here we recast the equation of MOTS with a different form and carry out the associated linear
analysis.
∗
For u = −R(ω, u) being the MOTS along H u , we set R = (ua)1−µ φ̌λ with λ to be determined
later. Then, it holds
∗
∇g R = λ(ua)1−µ φ̌λ−1 ∇g φ̌ = λRφ̌−1 ∇g φ̌,
which implies
div (∇g R) =div (λRφ̌−1 ∇g φ̌) = (λ2 − λ)Rφ̌−2 |∇g φ̌|2 + λRφ̌−1 ∆g φ̌.
Thus, with the aid of (9.1), the equation of MOTS 0 = G(R, u) is changed into
1 1
0 =∆g R + Ωtrg χ|∇g R|2 + 4Ωω|∇g R|2 − Ω−1 trg χ + 2g(η, ∇g R)+2Ωχ̂(∇g R, ∇g R)
2 2
µ∗ 1
=[−λ + λ2 ]Rφ̌−2 |∇g φ̌|2 + λRφ̌−1 ∆g φ̌ − Ω−1 trg χ + 2λRφ̌−1 g(η, ∇g φ̌)
1 − µ∗ 2
1 1 µ∗ 1
+( Ωtrg χ + )λ2 R2 φ̌−2 |∇g φ̌|2 + [4Ωω − · ]λ2 R2 φ̌−2 |∇g φ̌|2 + 2Ωχ̂(∇g R, ∇g R).
2 R 1 − µ∗ R
We now pick
1 − µ∗ µ∗
λ= ∗
, which satisfies − λ + λ2 = 0.
µ 1 − µ∗
This renders the above equation to be
1 1 1
0 =λRφ̌−1 ∆g φ̌ − Ω−1 trg χ + 2λRφ̌−1 g(η, ∇g φ̌) + ( Ωtrg χ + )λ2 R2 φ̌−2 |∇g φ̌|2
2 2 R
µ∗ 1 2 2 −2 2
+[4Ωω − · ]λ R φ̌ |∇g φ̌| + 2Ωχ̂(∇g R, ∇g R).
1 − µ∗ R
74 XINLIANG AN
µ∗ 1
With Christodoulou’s naked-singularity initial data, we note that 4Ωω(−R, 0) = 1−µ∗ ·R . Hence
we further define α1 as below and recall α2 given in (10.26)
1 1
α1 := −( Ωtrg χ + )−[4Ωω − (4Ωω)0 ],14
2 R
1 −1 1 − µ∗ uaf (ω, u)
α2 := Ω trg χ − + (1 − µ∗ ).
2 R 2Ω20 R2
∗ µ∗
Together with R = (ua)1−µ φ̌λ , γ = R−2 g, ∆g φ̌ = R−2 ∆γ φ̌ and Ω20 (R) = Ω2 (u, 0) = R 1−µ∗ ,
the above equation infers
f (ω, u) − 1−µ
λ
λφ̌−1 ∆γ φ̌ − (1 − µ∗ ) + (1 − µ∗ ) φ̌ ∗
+2λR2 φ̌−1 g(η, ∇g φ̌)
2
− λ2 R3 φ̌−2 |∇g φ̌|2 α1 − Rα2 +2RΩχ̂(∇g R, ∇g R) = 0.
With λ = (1 − µ∗ )/µ∗ , we now define
µ∗ 1 2µ∗
S(φ̌, ω) :=∆γ φ̌ − µ∗ φ̌ + f (ω, u)φ̌1− µ∗ + λR2 g(η, ∇g φ̌)
2 1 − µ∗
(11.1)
µ∗ µ∗ 2µ∗
− ∗
λ2 R3 φ̌−1 |∇g φ̌|2 α1 − ∗
φ̌Rα2 + φ̌RΩχ̂(∇g R, ∇g R).
1−µ 1−µ 1 − µ∗
And for 2-sphere S 1−µ∗ , we have that, if
−(ua)1−µ∗ φ̌ µ∗ ,u


< 0, it corresponds to an untrapped surface,
S(φ̌, ω) = 0, it corresponds to a MOTS,


> 0, it corresponds to a trapped surface.
For the operator S(φ̌, ω) defined in (11.1), it further holds
Proposition 11.1. Suppose φ̌ is a given function on S2 . The linearized operator ∂φ̌ S(φ̌) can
be expressed as
1 − µ∗ 1
∂φ̌ S(φ̌)[w] = ∆γ w − µ∗ w − f (ω)φ̌− µ∗ w + ci ∇i w + cw. (11.2)
2
Here the connection is with respect to the metric γ, and functions c = c(ω, φ̌, u) and ci =
ci (ω, φ̌, u) satisfy
1
|c| + |ci | ≤ Ca− 3 Ωδ̃ (|φ̌−1 ∇2γ φ̌| + |φ̌−1 ∇γ φ̌|2 + 1).
Proof. We first write
S(φ̌) = I1 + I2 + I3 + I4 + I5 , (11.3)
with
µ∗ 1− µ1∗
I1 = ∆γ φ̌ − µ∗ φ̌ + f φ̌ , (11.4)
2
2µ∗ 2 −1 −1 ik jl 2µ∗
I2 = λ φ̌ R Ωχ̂ γ γ ∇i φ̌∇j φ̌, I3 = λγ ij ηi ∇j φ̌,
1 − µ∗ kl 1 − µ∗
µ∗ µ∗
I4 = − ∗
λ2 φ̌−1 Rα1 |∇γ φ̌|2γ , I5 = − φ̌Rα2 . (11.5)
1−µ 1 − µ∗
∗ 1−µ∗
Then we take ∂φ̌ with respect to each Ii with i = 1, 2, 3, 4, 5. Recall R = (ua)1−µ φ̌ µ∗ and we
set
∗ 1−µ∗
R(ǫ) = (ua)1−µ (φ̌ + ǫw) µ∗ .
14For perturbed initial data, in α there is an additional term, i.e., 4Ωω(u, 0) − µ∗ · 1 . Its absolute value
1 1−µ∗ |u|
1 Ω(u,0)δ̃
is bounded by |u|
· 1 and in later arguments, this additional term obeys a desired control.
a3
d
Denoting ˙ = dǫ ǫ=0 , it then holds
1 − µ∗ 1 − µ∗
Ṙ = Rφ̌−1 w, u̇ = −Rφ̌−1 w, φ̌˙ = w.
µ∗ µ∗
We also have
1 − µ∗ −1
γ̇ ij = −γ ik γ jl γ̇kl and γ̇kl = u̇∂u γkl = − φ̌ w[(RΩtrg χ + 2)γkl + 2R−1 Ωχ̂kl ].
µ∗
µ∗ 1− µ1∗
For I1 = ∆γ φ̌ − µ∗ φ̌ + 2 f φ̌ , we further apply Lemma 9.1 to γ. Together with φ̇ = w, we
obtain
1
(∆γ φ̌)· =∆γ w − γ ik γ jl γ̇kl ∇2ij φ̌ − γ ij γ kl ∇i γ̇jl ∇k φ̌ + γ ij γ kl ∇l γ̇ij ∇k φ̌ (11.6)
2
and
1 − µ∗ 1
I˙1 = ∆γ w − µ∗ w − f (ω)φ̌− µ∗ w.
2
2µ∗ ∗
For I2 = 1−µ ∗ λ2 −1 −1
φ̌ R Ωχ̂ kl
γ ik jl
γ ∇i φ̌∇j φ̌, with λ = 1−µ
µ∗ it holds
I˙2 = − 2λφ̌−2 wR−1 Ωχ̂kl γ ik γ jl ∇i φ̌∇j φ̌ − 2λφ̌−1 R−2 Rφ̌−1 λwΩχ̂kl γ ik γ jl ∇i φ̌∇j φ̌
+ 2λφ̌−1 R−1 ∂u (Ωχ̂kl ) · (−Rφ̌−1 λw) · γ ik γ jl ∇i φ̌∇j φ̌ + 2λφ̌−1 R−1 Ωχ̂kl (γ̇ ik γ jl + γ ik γ̇ jl )∇i φ̌∇j φ̌
+ 2λφ̌−1 R−1 Ωχ̂kl γ ik γ jl (∇i w∇j φ̌ + ∇i φ̌∇j w),
where the coefficient c for w and the coefficient ci for ∇i w satisfying
Ωδ̃ −1 Ωδ̃
|c| ≤ 1 |φ̌ ∇j φ̌|2 , |ci | ≤ 1 |φ̌−1 ∇φ̌|.
a3 a3
∗
2µ ij
For I3 = − 1−µ ∗γ ηi ∇j φ̌, we get

˙ 2µ∗ ik jl 1 − µ∗ −1 −1
I3 = γ γ − φ̌ w[(RΩtrg χ + 2)γkl + 2R Ωχ̂kl ] ηi ∇i φ̌
1 − µ∗ µ∗
2µ∗ ij 1 − µ∗ 2µ∗ ij
− ∗
γ ∂u ηi · (−Rφ̌−1 ∗
w)∇j φ̌ − γ ηi ∇j w.
1−µ µ 1 − µ∗
And on the right the coefficient c for w and the coefficient ci for ∇i w satisfy
Ωδ̃ Ωδ̃
|c| ≤ 1 |φ̌−1 ∇j φ̌|, |ci | ≤ 1 .
a 3 a3
∗
µ 2 −1
For I4 = − 1−µ∗ λ φ̌ Rα1 |∇γ φ̌|2 , we have the expression
I˙4 =λφ̌−1 R∂u α1 Rφ̌−1 λwγ ij ∇i φ̌∇j φ̌ + λφ̌−2 wRα1 γ ij ∇i φ̌∇j φ̌

− λφ̌−1 Rφ̌−1 λwα1 γ ij ∇i φ̌∇j φ̌ − λφ̌−1 Rα1 γ̇ ij ∇i φ̌∇j φ̌
− λφ̌−1 Rα1 γ ij (∇i w∇j φ̌ + ∇i φ̌∇j w).
With precise calculations, we show that the size of the coefficients for ω and ∇i ω are small.
First recall
1 2
∂u α1 = − ∂u [Ωtrg χ + ] − 4∂u [Ωω − (Ωω)0 ].
2 R
For the term −∂u [ 12 Ωtrg χ + R
1
], there are cancellations. By the null structure equation and
−1 −2
∂u (2R ) − 2R = 0, it holds
2 1 2 2
∂u (Ωtrχ + ) + (Ωtrχ − )(Ωtrχ + )
R 2 R R
= − 4Ωω · Ωtrχ − Ω2 |χ̂|2g − (∂u φ)2
= − 4[Ωω − (Ωω)0 ]Ωtrχ − 4(Ωω)0 [Ωtrχ − (Ωtrχ)0 ] − Ω2 |χ̂|2g − [∂u φ − (∂φ)0 ]∂u φ
− (∂u φ)0 [∂u φ − (∂u φ)0 ] − 4(Ωω)0 (Ωtrχ)0 − (∂u φ)0 (∂u φ)0 .
76 XINLIANG AN
Using the hyperbolic part, we have improved estimates

1 1
ua 2 ua 2
|Ωω − (Ωω)0 | ≤ , |∂u φ − (∂u φ)0 | ≤ .
|u|2 |u|2
The remaining worrisome terms are
−4(Ωω)0 (Ωtrχ)0 − (∂u φ)0 (∂u φ)0 .
They could be of size R−2 and not small, which is dangerous. However, cancellations of
Christodoulou’s naked-singularity initial data help us. Along u = 0, we have
µ∗ 1 −2 µ∗ 1
−4(Ωω)0 · (Ωtrχ)0 − (∂u φ)20 = − 4 ∗
· · ( ) − 2 ∗
· 2
4(1 − µ ) R R 1−µ R
2µ∗ 1 2µ∗ 1
= · − · = 0.
1 − µ∗ R 2 1 − µ∗ R 2
Hence, the worrisome terms disappear.15
The term −4∂u [Ωω − (Ωω)0 ] in ∂u α1 also encodes this smallness. We have
Z u Z u
′ ′
∂u [Ωω − (Ωω)0 ] =∂u [ (Ω∇4 )(Ωω)(u, u , θ1 , θ2 )du ] = ∂u (Ω∇4 )(Ωω)(u, u′ , θ1 , θ2 )du′
0 0
Z u
1 1 1 1
= [Ω∇3 (−Ω2 η · η + Ω2 |η|2 + Ω2 ρ+ Ωe3 φ · Ωe4 φ + Ω2 eA φeA φ)](u, u′ , θ1 , θ2 )du′ .
0 2 2 6 12
In the last line of above equation, except Ωe3 (Ωe3 φ) and ∇3 η, all the other terms can be
explicitly expressed and obey desired estimates. To control Ω∇3 (Ω∇3 φ) term, for h being a
scalar function, we recall a fact
Ω∇4 (Ω∇3 h) = Ω∇3 (Ω∇4 h) − 4Ω2 ζ · ∇h.
Setting h = Ωe3 φ, we then have
Ωe4 (Ωe3 )(Ωe3 φ) =Ωe3 (Ωe4 )(Ωe3 φ) − 4Ω2 ζ∇(Ωe3 φ)
1 1
=Ωe3 [− Ωtrχ · Ωe4 φ − Ω2 ∆g φ − Ωtrχ · Ωe3 φ + 2Ω2 η A eA φ] − 4Ω2 ζ∇(Ωe3 φ).
2 2
Using naked-singularity initial data and integrating along e4 direction, we get
1
|Ωe3 (Ωe3 φ)| . ,
|u|2
which is a desired estimate for here. To bound ∇3 η term, we utilize
∇3 η = −∇3 η + 2∇3 ∇ log Ω.
There is the null structure equation for ∇3 η and we also have
1
∇3 ∇ log Ω = ∇∇3 log Ω + (η + η)∇3 log Ω − χ · ∇ log Ω = −2∇ω − (η + η)ω − χ · ∇ log Ω.
2
These imply
1
ua 2 1 Ωδ̃
≤
|∇3 η| ≤ · ,
|u|3 |u|2 a 31
which is a desired estimate for here as well.
With the derived estimate above, a direct check shows that, for I˙4 the coefficient c for w and
the coefficient ci for ∇i w obey
Ωδ̃ Ωδ̃
|c| ≤ 1 (φ̌−1 ∇j φ̌)2 , |ci | ≤ 1 |φ̌−1 ∇γ φ̌|.
a 3 a3
15For perturbed initial data, after the cancellation of the leading terms, these worrisome items are of size
1
Ω(−R, 0)δ̃ R−2 a− 3 , which are sufficiently small.
We proceed to study I˙5 . For I5 = −λ−1 φ̌Rα2 with

1 1 − µ∗ uaf (ω, u)
α2 =[ Ω−1 trg χ − + (1 − µ∗ )](u, u, ω)
2 R 2Ω20 R2
Z
1 u 1
=− ( trχtrχ + |χ̂|2g + ∂4 φ∂4 φ)(u, u′ , ω)du′
2 0 2
Z
1 u Ω−2 (u, u′ , ω)
+ |Ωχ̂|g (−1, u , ω) + [Ωe4 φ(−1, u , ω) − Ωe4 φ(−1, 0, ω)] du′
2 ′ ′ 2
2 0 |u|2
Z Z
1 u Ω−2 2 ′ ′ 1 u Ω−2
=− (Ωtrχ) (u, u , ω)du − [|uΩχ̂|2g (u, u′ , ω) − |Ωχ̂|2g (−1, u′ , ω)]du′
2 0 2 2 0 |u|2
Z
1 u Ω−2
− (uΩe4 φ) (u, u , ω) − [Ωe4 φ(−1, u, ω) − Ωe4 φ(−1, 0, ω)] du′ ,
2 ′ 2
2 0 |u|2
it holds
I˙5 = λ−1 φ̌R∂u α2 · Rφ̌−1 λw − λ−1 wRα2 − λ−1 φ̌Rφ̌−1 λwα2 .
To bound ∂u α2 , we first control
Z u
∂u trχ · trχ(u, u′ , ω)du′
0
Z u
1
= − Ωtrχtrχ + 2Ωωtrχ + 2Ωρ − Ωχ̂ · χ̂ trχ(u, u′ , ω)du′ (11.7)
0 2
Z u
1 1
+ 2Ωdiv η + 2Ω|η|2 − Ωe3 φe4 φ + ΩeA φeA φ trχ(u, u′ , ω)du′
0 3 3
Applying the improved estimate for trχ, we bound the potentially dangerous terms
Z u
(Ωtrχ, Ωω, Ωe3 φ) · (trχ, e4 φ) · trχ(u, u′ , ω)du′
0
1
uΩ−2 a 2 Ω ua
≤ · ·( + Ω−1 2 )
|u| |u| |u| |u|
1 1
ua a− 2 Ω ua 1 ua 2 1 Ωδ̃
= · + 2 2· · ≤ · .
2
|u| Ω 2 |u| |u| Ω |u| |u|Ω |u|2 a 31
All the other terms in (11.7) obey even smaller upper bound.
For the remaining terms in ∂u α2 , we need to show that16
1 1
||u|Ωe4 φ(u, u, ω) − [Ωe4 φ(−1, u, ω) − Ωe4 φ(−1, 0, ω)]| ≤ a 2 · a− 3 Ωδ̃ . (11.8)
A detailed discussion in Section 13.1, in particular (13.2), will give
ua µ∗ 1 1
|uΩe4 φ(u, u, ω) − [Ωe4 φ(−1, u, ω) − Ωe4 φ(−1, 0, ω)]| . + |u| 1−µ∗ ≤ a 2 · a− 3 Ωδ̃ .
|u|
With these estimates, we can check that ∂u α2 obeys
1 Ωδ̃
|∂u α2 | ≤ .
|u|2 a 13
Back to I˙5 , we conclude the coefficient c for w and the coefficient ci for ∇i w obey
Ωδ̃
|c| + |ci | ≤ 1 .
a3

16With perturbed initial data, we would get a similar bound.

78 XINLIANG AN
For the next step, we set

µ∗ − 1 1
L0 w := ∆γ w − µ∗ w + f (ω, u)φ̌− µ∗ w.
2
Integrating −wL0 w on S2 and applying integration by parts, we obtain
Z Z Z
∗ 2 2 ∗ 2 1 − µ∗ − µ1∗ 2
µ w dω ≤ |∇γ w| + µ w + f (ω, u)φ̌ w dω = − wL0 wdω.
S2 S2 2 S2
This implies Z Z
w2 dω ≤ C |L0 w|2 dω.
S2 S2
We further let
Z
∗ 1 − µ∗ − µ1∗
< w1 , w2 >:= ∇γ w1 · ∇γ w2 + µ w1 w2 + f (ω, u)φ̌ w1 w2 dω.
S2 2
Since 0 < µ∗ < 1 and f ≥ 0, we can use < w1 , w2 > as an inner-product on H 1 (S2 ), equivalent
to the standard H 1 (S2 ) inner product. The first eigenvalue of −L0 can be expressed as
Z Z
1 − µ∗ 1
µ1 = inf{ [|∇γ w|2 + µ∗ w2 + f (ω, u)φ̌− µ∗ w2 ]dω; w2 dω = 1}.
S2 2 S2
Immediately, we see that µ1 ≥ µ∗ . And via using Theorem 8.38 in [24], we also have
Lemma 11.2. Assume f and φ̌ to be given nonnegative functions on S2 with (1.4) satisfied.
Then the equation L0 w = 0 only admits the trivial solution. Moreover, the first eigenvalue µ1
of −L0 is positive and simple and it has a positive eigenfunction.
With this property, we obtain the crucial invertibility of the operator ∂φ̌ S(φ̌):
Lemma 11.3. Let β ∈ (0, 1) be a fixed constant. Set f and φ̌ to be given nonnegative functions
on S2 with (1.4) satisfied and |φ̌|C 2 (S2 ) ≤ K with K obeying (9.14). Then within the hyperbolic
region, it holds that ∂φ̌ S(φ̌) : C 2,β (S2 ) → C 0,β (S2 ) is invertible.
Proof. Applying Lemma 11.2, we first take a positive eigenfunction ψ1 associated with the first
eigenvalue µ1 > 0, i.e.,
1 − µ∗ − µ1∗
∆γ ψ1 − µ∗ ψ1 − f φ̌ ψ1 = −µ1 ψ1 .
2
Using Proposition 11.1, we have
1 − µ∗ − µ1∗
∂φ̌ S(φ̌)[ψ1 ] = ∆γ ψ1 − µ∗ ψ1 − f φ̌ ψ1 + ci ∇i ψ1 + cψ1
2
= −µ1 ψ1 + ci ∇i ψ1 + cψ1 .
From derived apriori estimates and constructed lower barrier for φ̌, we have |φ̌−1 ∇2γ φ̌|+|φ̌−1 ∇γ φ̌|2 +
1 is independent of u, u and it satisfies
1
|c| + |ci | ≤ Ca− 3 Ωδ̃ (|φ̌−1 ∇2γ φ̌| + |φ̌−1 ∇γ φ̌|2 + 1).
With Ω being sufficiently small depending on u, u in the interior region, we have |ci | + |c| ≪ 1.
This implies ∂φ̌ S(φ̌)[ψ1 ] < 0 on S2 . Employing the calculation in the proof of Theorem 2.11
in [26], for any w ∈ C 2 (S2 ), we consider an elliptic equation for ψw1 . For notional simplicity, let
us rewrite the elliptic operator ∂φ̌ S(φ̌) as
∂φ̌ S(φ̌)[w] = aij Dij w + bi Di w + cw
w
with aij , bi , c being functions. We then calculate and directly check that ψ1 satisfies
w 2 w ∂φ̌ S(φ̌)[ψ1 ] w ∂φ̌ S(φ̌)[w]

aij Dij + (bi + aij Dj ψ1 )Di + = . (11.9)
ψ1 ψ1 ψ1 ψ1 ψ1 ψ1
Since ∂φ̌ S(φ̌)[ψ1 ] < 0 and ψ1 > 0, the above elliptic equation is invertible. For fixed ψ1 > 0,
giving ∂φ̌ S(φ̌)[w]/ψ1 , we can solve for the unique solution ψw1 . This further implies that ∂φ̌ S(φ̌) :
C 2,β (S2 ) → C 0,β (S2 ) is invertible and it holds
w ∂φ̌ S(φ̌)[w]
| |C 2,β (S2 ) ≤ C| |C 0,β (S2 ) and |w|C 2,β (S2 ) ≤ C|∂φ̌ S(φ̌)[w]|C 0,β (S2 ) (11.10)
ψ1 ψ1
with some positive constant C independent of w and depending on function φ̌ and its associated
eigenfunction ψ1 .
In the next lemma, we eliminate such dependence on φ̌.
Lemma 11.4. Assume f and φ̌ to be given nonnegative functions on S2 with (1.4) satisfied and
|φ̌|C 2 (S2 ) ≤ K with K obeying (9.14). Then, within the hyperbolic region, for any w ∈ C 2 (S2 ),
we have that
max
2
|w| ≤ C max
2
|∂φ̌ S(φ̌)[w]|. (11.11)
S S
Here the positive constant C depends only on m and ǫ in (1.4) and K, independent of φ̌ and
u ∈ (0, δ].
Proof. We take a smooth nonnegative function f0 = f0 (ω) on S2 satisfying f0 ≥ m/2 on Bp (ǫ)
and f (·, u) ≥ f0 on S2 for all u ∈ (0, δ]. With the standard spherical metric γ0 on S2 , we first
consider
1 − µ∗
L∗ w = ∆γ0 w−µ∗ w − f0 w.
2
Employing Lemma 11.2 to L∗ , we can find a positive eigenfunction ψ1 associated with its first
eigenvalue µ1 > 0, i.e.,
1 − µ∗
∆γ0 ψ1 −µ∗ ψ1 − f ψ1 = −µ1 ψ1 .
2
Here µ1 and ψ1 depend only on m, ǫ and are independent of u ∈ (0, δ]. With the fact 0 < φ̌ ≤ 1,
f ≥ f0 , applying Lemma 9.2, and Proposition 11.1, we then get
1 − µ∗ − µ1∗
∂φ̌ S(φ̌)[ψ1 ] = ∆γ0 ψ1 − µ∗ ψ1 − f φ̌ ψ1 + ∆γ ψ1 − ∆γ0 ψ1 + ci ∇i ψ1 + cψ1
2
1 − µ∗
≤ ∆γ0 ψ1 − µ∗ ψ1 − f0 ψ1 + ∆γ ψ1 − ∆γ0 ψ1 + ci ∇i ψ1 + cψ1
2
= −µ1 ψ1 + ∆γ ψ1 − ∆γ0 ψ1 + ci ∇i ψ1 + cψ1 ≤ −c(m, ǫ, K) < 0.
For the last inequality, we choose a being suitably large in the exterior and Ω being sufficiently
small in the interior. We then repeat the proof of Lemma 11.3 with this ψ1 , which is independent
of φ̌. And we prove the current lemma.
In below we further prove that the equation S(φ̌) = 0 admits a unique solution φ̌ satisfying
|φ̌|C 2 (S2 ) ≤ K.
Proposition 11.5. Set f and φ̌1 , φ̌2 to be given nonnegative functions on S2 with (1.4) satisfied.
Require |φ̌1 |C 2 (S2 ) ≤ K and |φ̌2 |C 2 (S2 ) ≤ K with K obeying (9.14). If S(φ̌1 ) = S(φ̌2 ), then it
holds that φ̌1 ≡ φ̌2 for all ω ∈ S2 .
Proof. For the given φ̌1 and φ̌2 , we define the below L and view it as a linear operator of φ̌1 − φ̌2 .
Z 1
L(φ̌1 − φ̌2 ) := S(φ̌1 ) − S(φ̌2 ) = ∂φ̌ S(tφ̌1 + (1 − t)φ̌2 )[φ̌1 − φ̌2 ]dt.
0
The same reasoning as for Lemma 5.5 of [6] yields the desired result.
R1
Here operator 0 ∂φ̌ S(tφ̌1 +(1−t)φ̌2 )dt possesses a similar structure to the linearized operator
∂φ̌ S(φ̌) in Proposition 11.1. The corresponding Lemmas 11.2-11.4 can be obtained in the same
manner. By the assumption, we first use S(φ̌1 ) ≥ S(φ̌2 ) and it gives L(φ̌1 − φ̌2 ) ≥ 0. With
the same proof as in Lemma 11.3, for the operator L, applying Theorem 2.11 in [26], the strong
maximum principle holds here. Hence it holds either φ̌1 < φ̌2 or φ̌1 ≡ φ̌2 on S2 . We then
80 XINLIANG AN
use S(φ̌2 ) ≥ S(φ̌1 ). Repeating the argument, we now have either φ̌2 < φ̌1 or φ̌1 ≡ φ̌2 on S2 .
Combining these together, we have proved the desired uniqueness result.
As a corollary of the above lemma, we also prove that
Theorem 11.6. Take u to be a fixed constant in (0, δ]. Let f be a given smooth function on
S2 , satisfying (1.3) and (1.4), and φ̌ and φ̌ be constructed as in Section 9. Then, there exists a
unique smooth solution φ̌ = φ̌(ω, u) of S(φ̌, u) = 0, satisfying φ̌ < φ̌(·, u) < φ̌ on S2 .
Proof. Here, for any constant λ ∈ [0, 1] and any C 2 -function φ̌ on S2 , we set
µ∗ 1
Sλ (φ̌) = ∆γ φ̌ − µ∗ φ̌ + λf + (1 − λ)]φ̌1− µ∗ + λ(I2 + I3 + I4 + I5 ),
2
where Ii , with 2 ≤ i ≤ 5, are defined in (11.5). With the same line of reasoning as for the detailed
proof of Theorem 6.1 in [6], by applying the method of continuity for parameter λ ∈ [0, 1],
together with the utilization of the invertibility of ∂φ̌ Sλ (φ̌), we conclude the existence. The
uniqueness follows from Proposition 11.5.
12. The Anisotropic Apparent Horizon

Considering u as a parameter, in this section we show that our constructed MOTSs form a
smooth apparent horizon. Viewing the solution φ̌ in Theorem 11.6 as a function of ω and u, we
have
Theorem 12.1. With the assumptions in Theorem 11.6, for each u ∈ (0, δ], setting φ̌ = φ̌(ω, u)
to be the solution to S(φ̌, u) = 0 as in Theorem 11.6, then it holds that the function φ̌ = φ̌(ω, u)
is smooth with respect to both variables (ω, u) ∈ S2 × (0, δ].
Proof. We employ the method of continuity again as in Section 11. This time we take u as a
new parameter. Given φ̌(ω, u) being the constructed unique solution to S(φ̌, u) = 0, we then
apply the implicit function theorem to obtain a smooth solution φ̌(ω, u′ ) to S(φ̌, u′ ) = 0 with u′
close to u. The key is the invertibility of ∂φ̌ S(φ̌, u) : C 2,β (S2 ) → C 0,β (S2 ), which was proved in
Lemma 11.3. Hence, φ̌(ω, u) is smooth with respect to both ω and u variables.
Applying a null comparison principle and an upcoming theorem in [7] by the author and He,
with double null foliations we show that the apparent horizon formed in gravitational collapse
is always either null or spacelike.
In this section, we specify a condition of f to guarantee the spacelikeness of our constructed
∗ 1−µ∗
apparent horizon {u = −(ua)1−µ φ̌ µ∗ }. To do so, we derive the equation for ∂u φ̌ and to
obtain its estimate.
Proposition 12.2. Let φ̌ = φ̌(ω, u) be the solution to S(φ̌, u) = 0 as in Theorem 11.6. For
each u ∈ (0, δ], ∂u φ̌ obeys
1 − µ∗ 1 1 µ∗ 1
∆γ (∂u φ̌) − µ∗ ∂u φ̌ − f (ω, u)φ̌− µ∗ ∂u φ̌ + ci ∇i (∂u φ̌) + c∂u φ̌ = h + ∂u f φ̌1− µ∗ . (12.1)
2 u 2
Here c = c(ω, φ̌, u), ci = ci (ω, φ̌, u) and h = h(ω, φ̌, u) are functions satisfying
1 1
|c| + |ci | + |h| ≤ Ca− 3 · Ωδ̃ · φ̌− µ∗ (|∇2γ φ̌| + |∇γ φ̌|2 + 1). (12.2)
Proof. The main part of this proof is similar to the corresponding one in [6] and hence is omitted.
Only noting that, in the estimates for here we employ
1 1
α1 := −[ Ωtrg χ + ] − 4[Ωω − (Ωω)0 ],
2 R
Z
1 u 1
α2 := − [( trχtrχ + |χ̂|2g + ∂4 φ∂4 φ)(u, u′ , ω)]du′
2 0 2
Z
1 u Ω−2 (u, u′ , ω)
+ |Ωχ̂|g (−1, u , ω) + [Ωe4 φ(−1, u , ω) − Ωe4 φ(−1, 0, ω)] du′ .
2 ′ ′ 2
2 0 |u|2
The desired estimates for ∂u α1 and ∂u α2 were already obtained in Proposition 11.1. To bound
∂u α1 and ∂u α2 , we utilize the explicit formulas for ∂u [ 12 Ωtrg χ + R1 ] and ∂u (Ωω), the estimates
for ∂u α1 can be obtained straightforwardly. And for ∂u α2 , we have
1 1
∂u α2 (u, u, ω) = − × ( trχtrχ + |χ̂|2g + ∂4 φ∂4 φ)(u, u, ω)
2 2
1 Ω−2 (u, u, ω) 2 2
+ × |Ω χ̂| g (−1, u, ω) + [Ωe 4 φ(−1, u, ω) − Ωe 4 φ(−1, 0, ω)] .
2 |u|2
The desired estimate for above expression can be got in the same fashion as for (11.8).
Proceeding similarly to the argument for Corollary 6.4 in [6], we then obtain
Corollary 12.3. For each u ∈ (0, δ], let φ̌ = φ̌(ω, u) be the solution to S(φ̌, u) = 0. Under the
assumptions in Theorem 11.6, it holds
1 1
max
2
|∂u φ̌| ≤ C u−1 a− 3 Ωδ̃ max
2
(|φ̌| + |∇γ φ̌| + |∇2ij φ̌| + |φ̌−1 ∇γ φ̌∇γ φ̌|) + max
2
|∂u f φ̌1− µ∗ | ,
S S S
(12.3)
where C is a positive constant independent of a, u and φ̌.
This further implies
Proposition 12.4. For each u ∈ (0, δ], set φ̌ = φ̌(ω, u) to be the solution of S(φ̌, u) = 0. Under
the assumptions in Theorem 11.6, if we further require
1
|u∂u f (ω, u)| ≤ a− 3 Ωδ̃ (u1 ), (12.4)
∗
the apparent horizon {(ω, u, u)| u = −(ua)1−µ φ̌λ (ω, u)} is spacelike.
∗ 1−µ∗
Proof. Using R = (ua)1−µ φ̌λ (ω, u) and λ = µ∗ , we have ∂u R = u−1 R[(1 − µ∗ ) + λuφ̌−1 ∂u φ̌].
∗
− 31 δ̃
Applying Corollary 12.3, if |u∂u f (ω, u)| ≤ a Ω , then we get ∂u R ≥ 1−µ
2 u
−1
R. Assume g ′ is
2
the induced metric to the apparent horizon u = −R(ω, u). Rewrite ω ∈ S as ω = (θ1 , θ2 ). As
shown in [3], then it holds
gθ′ i θj = gθi θj , gθ′ i u = 2∂θi R, gu′ u = 4∂u R.
Suppose X = λ1 ∂θ1 + λ2 ∂θ2 + λ3 ∂u to be an arbitrary nonzero tangent vector along the
apparent horizon with constants λ1 , λ2 , λ3 . It holds
g ′ (X, X) = λi λj gij + 4λi λ3 ∂i R + 4λ23 ∂u R,
where i, j = 1, 2 are summed over. The case λ3 = 0 implies g ′ (X, X) = λi λj gij > 0. For the
case λ3 6= 0, using gij = R2 γij and ∂i R = λRφ̌−1 ∇i φ̌, we get
g ′ (X, X) =λi λj gij + 4λi λ3 ∂i R + 4λ23 ∂u R
1 − µ∗ −1
≥R2 λi λj γij + 4λi λ3 λRφ̌−1 ∇i φ̌ + 4λ23 · u R
2 (12.5)
≥2λ23 (1 − µ∗ )u−1 R − 4λ23 λ2 (φ̌−1 ∇i φ̌)2
∗ ∗
>2λ23 (1 − µ∗ )u−µ (amǫτ +2 )1−µ − 4λ3 λ2 (φ̌−1 ∇i φ̌)2 .
For the last inequality, we use φēmax = − log(mǫτ +2 ) + C(τ ) in (9.12) and hence
∗
R ≥ (ua · mǫτ +2 )1−µ with τ > 2.2 fixed.
∗
At the same time, by (9.11) we have u = −( 23 ua)1−µ is an untrapped surface. For the con-
∗ 1−µ∗ 1−µ∗ ∗
structed MOTS locating at u = −R = −(ua)1−µ φ̌ , it thus holds φ̌ µ∗ ≥ ( 23 )1−µ . That is
µ∗
2 µ∗
φ̌ ≥ ( 3 ) . Therefore, back to (12.5), when u is sufficiently small, we have g ′ (X, X) > 0. This
renders our constructed apparent horizon spacelike and hence a dynamical horizon.
82 XINLIANG AN
13. The Instability Theorems

In this section, we prove nonlinear instability theorems with Christodoulou’s naked-singularity
initial data prescribed along u = 0.17
13.1. Instability Subject to Perturbations with Finite BV Norms. We first recall equa-
tion (5.1)
1 1
(Ωe3 )(Ωe4 φ) + ΩtrχΩe4 φ = Ω2 ∆g φ − ΩtrχΩe3 φ+2Ω2 η A eA φ.
2 2
With e3 = Ω−1 ∂/∂u, this equation is equivalent to
∂ 1 2 1
(uΩe4 φ) + (Ωtrχ + )uΩe4 φ = uΩ2 ∆g φ − uΩtrχ · Ωe3 φ + 2uΩ2 η A eA φ.
∂u 2 |u| 2
Via integration, we hence obtain
Z u
1
uΩe4 φ(u, u, ω) =uΩe4 φ(−1, u, ω) + − u′ Ωtrχ · Ωe3 φ(u′ , u, ω)du′
−1 2
Z u (13.1)
1 2
+ [− (Ωtrχ + ) · uΩe4 φ + uΩ2 ∆g φ + 2uΩ2 η A eA φ](u′ , u, ω)du′ .
−1 2 |u|
For the first two terms on the right of (13.1), we can rewrite them as
Z u
1
uΩe4 φ(−1, u, ω) + − u′ Ωtrχ · Ωe3 φ(u′ , u, ω)du′
−1 2
Z
1 u ′
=uΩe4 φ(−1, u, ω) − uΩe4 φ(−1, 0, ω) + uΩe4 φ(−1, 0, ω) − u Ωtrχ · [Ωe3 φ − (Ωe3 φ)0 ](u′ , u, ω)du′
2 −1
Z Z
1 u ′ 1 u ′
− u [Ωtrχ − (Ωtrχ)0 ] · (Ωe3 φ)0 (u′ , u, ω)du′ − u (Ωtrχ)0 · (Ωe3 φ)0 (u′ , u, ω)du′ .
2 −1 2 −1
Employing (13.1), with Christodoulou’s initial data along u = 0 we have
Z u
1
uΩe4 φ(u, 0, ω) = uΩe4 φ(−1, 0, ω) + − u′ (Ωtrχ)0 · (Ωe3 φ)0 (u′ , 0, ω)du′ .
−1 2
Hence, the underlined two terms can be replaced by uΩe4 φ(u, 0, ω). Back to (13.1) again, we
now obtain
uΩe4 φ(u, u, ω) − [uΩe4 φ(−1, u, ω) − uΩe4 φ(−1, 0, ω)]
Z
1 u ′
=uΩe4 φ(u, 0, ω) − u Ωtrχ[Ωe3 φ − (Ωe3 φ)0 ](u′ , u, ω)du′
2 −1
Z
1 u ′
− u [Ωtrχ − (Ωtrχ)0 ] · (Ωe3 φ)0 (u′ , u, ω)du′
2 −1
Z u
1 2
+ [− (Ωtrχ + ) · uΩe4 φ + uΩ2 ∆g φ + 2uΩ2 η A eA φ](u′ , u, ω)du′ .
−1 2 |u|
Utilizing the fact
r
√ 1 − µ∗ 1 − µ∗ µ∗
uΩe4 φ(u, 0, ω) = − 2 · · |u| 1−µ∗
µ∗ 4
and the hyperbolic estimates derived in previous sections, we get
ua µ∗
|uΩe4 φ(u, u, ω) − [Ωe4 φ(−1, u, ω) − Ωe4 φ(−1, 0, ω)]| . + |u| 1−µ∗ . (13.2)
|u|
Now we consider the key equation that drives the trapped surface formation
1
Ω∇4 (Ω−1 trχ − (Ω−1 trχ)0 ) = −|χ̂|2 − (∇4 φ)2 − (trχ)2 .
2
17For perturbed initial data, analogous instability theorems can be proved correspondingly.
With this equation, we have

Ω−1 trχ(u, u, ω) − (Ω−1 trχ)(u, 0, ω)
Z u Z u Z u
1 (13.3)
= −|χ̂|2 (u, u′ , ω)du′ + −|∇4 φ|2 (u, u′ , ω)du′ + − (trχ)2 (u, u′ , ω)du′ .
0 0 0 2
We then find the main terms on the right. By hyperbolic estimates, we first have
Z u Z u 1
|Ωχ̂|2 (−1, u′ , ω) ′ ua ua 2
| −|χ̂|2 (u, u′ , ω)du′ − − du | ≤ · .
0 0 |u|2 Ω2 |u|2 Ω2 |u|
Using (13.2), we also get
Z u Z u
1
| −|∇4 φ|2 (u, u′ , ω)du′ + [Ωe4 φ(−1, u′ , ω) − Ωe4 φ(−1, 0, ω)]2 du′ |
0 |u|2 Ω2 0
Z u 3
1 1 u′ a µ∗ 1
1−µ∗ ) · a 2 du′ ≤
1 u2 a 2 µ∗ 1
≤ 2
· 2
· ( + |u| 2 2
· ( + |u| 1−µ∗ · ua 2 )
0 Ω |u| |u| |u| Ω |u|

1
ua ua 2 µ∗ 1
≤ 2 2
·( + |u| 1−µ∗ a− 2 ).
|u| Ω |u|
For the last term, via the obtained estimate in Proposition 4.3
2(1 − µ∗ )Ω2 ua
|Ωtrχ − |. ,
|u| |u|2
we deduce
Z u Z u
1 1
| − (trχ)2 (u, u′ , ω)du′ | . 2
· (Ωtrχ)2 (u, u′ , ω)du′
0 2 0 Ω
Z u
1 2(1 − µ∗ )Ω2 u′ a
≤ 2
( + 2 )2 (u, u′ , ω)du′
0 Ω |u| |u|
u ua u2 a
. 2 2 Ω4 (u, u, ω) + 2 2 · 2 (u, u, ω).
|u| Ω |u| Ω |u|
Note that the MOTS forms in the interior region, where Ω(u, u, ω) . Ω(u, 0) ≤ Ω(u1 , 0) ≪ 1.
With these estimates, we spot the main contribution on the right of (13.3), that is
Z u Z u
|Ωχ̂|2 (−1, u′ , ω) ′ 1
− du − [Ωe4 φ(−1, u′ , ω) − Ωe4 φ(−1, 0, ω)]2 du′ .
0 |u|2 Ω2 |u|2 Ω2 0
We then define f (ω, u) via requiring
uaf (ω, u)
· (1 − µ∗ )
|u|2 Ω2 (u, 0)
Z u (13.4)
|Ωχ̂|2 (−1, u′ , ω) + [Ωe4 φ(−1, u′ , ω) − Ωe4 φ(−1, 0, ω)]2 ′
:= du .
0 |u|2 Ω2 (u, 0)
With this defined f (ω, u), combining all the previous arguments above, we arrive at our first
main theorem
system (1.1). Assigning Christodoulou’s naked-singularity initial data in [18] along u = 0 with
−1 ≤ u ≤ 0 and prescribing perturbed initial data along u = −1 satisfying
X 1 1
uj a− 2 k(∂u )j ∇i χ̂kL∞ 2
u L (S−1,u )
+ uj a− 2 k(∂u )j ∇i ∇4 φkL∞ 2
u L (S−1,u )
≤ B, (13.5)
i≤5,j≤3
then for each B there exist a sufficient small δ = δ(B) and constant a = a(B)18, and the
Einstein-scalar field system admits a unique regular solution in the spacetime region (u, u, ω),
where 0 ≤ u ≤ δ and u ≤ |u|Ω2−δ̃ (u, 0) ≤ 1 with 0 < δ̃ ≪ 1.
18In particular, we can choose a = 1.

84 XINLIANG AN

Z u
|χ̂|2 (−1, u′ , ω) + [∇4 φ(−1, u′ , ω) − ∇4 φ(−1, 0, ω)]2 du′ = f (ω, u)ua (13.6)
0
with f (ω, u) obeying

0 ≤ f ≤ 1 on S2 × (0, δ] and f (·, u) ≥ m on Bp (ǫ) (13.7)
2
for a point p ∈ S and constants m ∈ (0, 1), ǫ ∈ (0, π/2), then within the solved hyperbolic
region, there exists a unique MOTS Mu on each H u with 0 < u ≤ δ, and the collection of
MOTS {Mu } emerges and censors the central singularity. And they form an achronal apparent
horizon.
Set a = 1, we then get Theorem 1.1.
Note that for u < 0 we require our spacetime to coincide with Christodoulou’s naked-
singularity solution in [18]. Along u = −1 we consider the possible extensions of χ̂(−1, u, ω), ∂u φ(−1, u, ω)
for u > 0. Via the next theorem, we can see that Christodoulou’s naked-singularity solution is
associated with nonlinear instability of high co-dimensions:
Theorem 13.2. Consider the space of prescribed initial data for χ̂(−1, u, ω) and ∂u φ(−1, u, ω)
√ q 1−µ∗
along u = −1 with u > 0. With |χ̂|(−1, 0, ω) = 0, ∂u φ(−1, 0, ω) = 2 · 1−µ
∗
µ∗ · 4 cor-
responding to their values of Christodoulou’s naked-singularity solution in [18], then for any
k ∈ Z+ , we can find smooth functions f1 , f2 , ..., fk and g1 , g2 , ..., gk with variables u and ω, such
that the below prescribed initial data for |χ̂| and ∂u φ with u > 0

|χ̂|(−1, u, ω), ∂u φ(−1, u, ω)

= 0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ′1 g1 + λ′2 g2 + · · · + λ′k gk
would lead to the trapped surface and the apparent horizon formation, when
k
X k
X 2
λ2i + λ′i 6= 0 with λk , λ′k ∈ R.
1 1

0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ′1 g1 + λ′2 g2 + · · · + λ′k gk

= 0 + λ̃1 f1 + λ̃2 f2 + · · · + λ̃k fk , ∂u φ(−1, 0, ω) + λ̃′1 g1 + λ̃′2 g2 + · · · + λ̃′k gk ,
then it renders
+
We may therefore say that, for any k ∈ Z , Christodoulou’s naked-singularity solution in
[18] has at least co-dimensional 2k nonlinear instability subject to outgoing BV characteristic
perturbations at (−1, 0, ω).
Proof. For any k ∈ Z+ and 0 ≤ u ≪ 1, on the 2-sphere S−1,u we first fix 2k disjoint discs:
D1 , D2 , ..., Dk and D1′ , D2′ , ..., Dk′ . We prescribe smooth functions fi being compactly supported
in Di and smooth functions gi being compactly supported in Di′ . For 1 ≤ i ≤ k, we further
require
1 1
0 ≤ fi ≤ a 2 on S2 × (0, δ], fi (·, u) ≥ mi a 2 on Bpi (ǫi ) ⊂ Di ,
1 1
0 ≤ gi ≤ a 2 on S2 × (0, δ], gi (·, u) ≥ m′i a 2 on Bp′i (ǫ′i ) ⊂ Di′
with points pi ∈ Di ⊂ S2 , p′i ∈ Di′ ⊂ S2 and constants mi , m′i ∈ (0, 1), ǫi , ǫ′i ∈ (0, π/2). With
these choices of initial data, conditions (13.6) and (13.7) in Theorem 13.1 are satisfied and the
corresponding conclusions for trapped surface and apparent horizon formation hold. And if we
have

′ ′ ′
0 + λ1 f1 + λ2 f2 + · · · + λk fk , ∂u φ(−1, 0, ω) + λ1 g1 + λ2 g2 + · · · + λk gk

′ ′ ′
= 0 + λ̃1 f1 + λ̃2 f2 + · · · + λ̃k fk , ∂u φ(−1, 0, ω) + λ̃1 g1 + λ̃2 g2 + · · · + λ̃k gk ,
since all the fi and gi with 1 ≤ i ≤ k are compacted supported in disjoint discs, it immediately
implies

13.2. Instability Subject to Perturbations with Finite C 0 Norms. The above two theo-
rems in Section 13.1 generalize both the results in [2] by An-Luk and in [6] by An-Han with the
requirement:
1 1
fi (ω, u) ≥ mi a 2 on Bpi (ǫi ) ⊂ Di , gi (ω, u) ≥ m′i a 2 on Bp′i (ǫ′i ) ⊂ Di′ for u > 0,
and
fi (ω, u) = 0, gi (ω, u) = 0 for u ≤ 0.
In this subsection, we extend the instability theorems in Section 13.1 further by allowing choosing
continuous functions fi (ω, u) and gi (ω, u) for the u variable.
We start from reconsidering
2(1 − µ∗ )
Ω−1 trχ(u, u, ω) −
|u|
Z u Z u Z u
1
= −|χ̂|2 (u, u′ , ω)du′ + −|∇4 φ|2 (u, u′ , ω)du′ + − (trχ)2 (u, u′ , ω)du′ .
0 0 0 2
By the estimates derived above, we have
Z u
−1 2(1 − µ∗ ) |Ωχ̂|2 (−1, u′ , ω) + [Ωe4 φ(−1, u′ , ω) − Ωe4 φ(−1, 0, ω)]2 ′
|Ω trχ(u, u, ω) − − du |
|u| 0 |u|2 Ω2 (u, 0, ω)
1 1
ua ua 2 ua ua 2 µ∗ 1 u ua u2 a
≤ 2 2
· + 2 2 ·( + |u| 1−µ∗ a− 2 ) + 2 2 Ω4 (u, u, ω) + 2 2 · 2 (u, u, ω).
|u| Ω |u| |u| Ω |u| |u| Ω |u| Ω |u|
µ∗
Together with Ω(u, 0) = |u| 2(1−µ∗ ) , this implies
Z u
2(1 − µ∗ ) |Ωχ̂|2 (−1, u′ , ω) + [Ωe4 φ(−1, u′ , ω) − Ωe4 φ(−1, 0, ω)]2 ′
|Ω−1 trχ(u, u, ω) − − du |
|u| 0 |u|2 Ω2 (u, 0, ω)
1 1 5
ua 3 ua 2 Ω2 Ω2 ua 3
≤ 2 2
· Ω2 · [ 3 + 1 + ]≪ 2 2
· Ω2 .
|u| Ω |u|Ω 2 a2 a |u| Ω
For the f (ω, u) introduced in (13.6), with any constant C1 > 2 and Λ defined in Proposition
3.2, we further set
1 1 1
f (ω, u) = g(u)f˜(ω, u) with g(u) = (C1 Λ + C1 ) 2 · [ln(ln )]− 2 (13.8)
u
and f˜(ω, u) to satisfy
0 ≤ f˜ ≤ 1 on S2 × (0, δ] and f˜(·, u)≥m on Bp (ǫ) (13.9)
2
for a point p ∈ S and constants m ∈ (0, 1), ǫ ∈ (0, π/2). With this choice of g(u), we let

3 3 1 1 1
κ := ln[ g(u)−1 ] = ln {ln[(ln ) C1 Λ+C1 ]} 2 .
2 2 u
1 1
Now we have eκ = 3{ln[(ln 1/u) C1 Λ+C1 ]} 2 /2. And when 0 ≤ u ≪ 1, it holds
1 1
8Λκeκ ≤ e2κ = ln[(ln ) C1 Λ+C1 ] ≤ ln[l ln(u−1 )] = ln ln(u−l ) with 0 < l ≪ 1.
u
86 XINLIANG AN
For the C(κ) defined in (9.15), we then obtain

κ κ
ln[C(κ)] = Ce4Λκe ln(Ceκ ) ≤ e8Λκe ≤ ln(u−l ).
Hence it holds that the C(κ) defined in (9.15) obeys
4Λκeκ
C(κ) := (Ceκ )Ce ≤ u−l with 0 < l ≪ 1.
If we require
−δ̃µ∗ δ̃µ∗
u−l ≤ |u| 2(1−µ∗ ) ⇐⇒ u ≥ |u| 2l(1−µ∗ ) , (13.10)
we immediately have
4Λκeκ −δ̃µ∗
Ceκ ≪ C(κ) = (Ceκ )Ce ≤ u−l ≤ |u| 2(1−µ∗ ) = Ω(0, u)−δ̃ . (13.11)
With new conditions (13.8) and (13.9) we then revisit our detailed constructions of the
anisotropic trapped and untrapped surfaces in Section 3 of [6]. We replace f (ω, u) by g(u)f˜(ω, u).
Consider the below main operator defined in (3.1) of [6]
1
S(φ) = ∆γ φ + 1 − f eφ + F1 (φ) + F2 (φ)
2
with F1 (φ) := − 4Ωω · uae−φ |∇γ φ|2 ,
1 1
F2 (φ) :=2R−1 Ωχ̂kl γ ik γ jl ∇i φ∇j φ + 2γ(η, ∇γ φ) + [− Ωtrg χ − ] · uae−φ |∇γ φ|2
2 uae−φ
1 1 uaf (ω, u)
+ uae−φ · [ Ω−1 trg χ − + ].
2 uae−φ 2(ua)2 e−2φ
Replacing f by g(u)f˜, by the hyperbolic estimates in this article, we have
1 1
S(φ) = ∆γ φ + 1 − g(u)f˜eφ + F1 (φ) + F2 (φ) with |F2 (φ)| ≤ Ceφ a− 3 Ωδ̃ (|∇γ φ|2 + 1). (13.12)
2
Recalling
3 3 1 1 1
κ = ln[ g(u)−1 ] = ln {ln[(ln ) C1 Λ+C1 ]} 2 ,
2 2 u
and using (13.12), we first check that
1 3 3 1 1
S(κ) = 1 − g(u) · f˜ · g(u)−1 + F1 (κ) + F2 (κ) ≥ 1 − − Ceκ a− 3 Ω(0, u)−δ̃ > ,
2 2 4 8
¯
where we use F1 (κ) = 0 and (13.11). We also revisit the construction of φ̃ in Section 3 of [6]
and with f replaced by g(u)f˜ we have that
¯ 0 2 ≤ − log(g(u)mǫτ +2 ), |∇ φ̃|
|φ̃| ¯ ≤ ǫ−1 , |∇2 φ̃| ¯ ≤ ǫ−2
C (S ) γ0 γ0
with τ being a fixed number larger than 2.2. Back to (9.16), if requiring (13.10), we then verify
∗
4Λκ ¯ 1 1 −δ̃µ
K = max{(Ceκ )Ce , kφ̃kC 2 } ≤ u−l ≤ a 3 · Ω(u, 0)−δ̃ = a 3 · |u| 2(1−µ∗ ) . (13.13)
Now we revisit the arguments in Section 10, Section 11 and Section 12 of this article with
f (ω, u) replaced by g(u)f˜(ω, u). The arguments in Section 10 stay the same. For the arguments
in Lemma 11.3 of Section 11, we now consider the positive eigenfunction ψ1 associated with the
corresponding first eigenvalue µ1 > 0, i.e.,
1 − µ∗ − µ1∗
∂φ̌ S(φ̌)[ψ1 ] = ∆γ ψ1 − µ∗ ψ1 − f φ̌ ψ1 + ci ∇i ψ1 + cψ1
2
= −µ1 ψ1 + ci ∇i ψ1 + cψ1 .
With Ω being sufficiently small in the interior region, we have |ci | + |c| ≪ 1. To make sure that
1
∂φ̌ S(φ̌)[ψ1 ] < 0 on S2 , here we check the size of f φ̌− µ∗ . Using the facts f (ω, u) = g(u)f˜(ω, u)
1−µ∗
1 1
with 0 ≤ f˜ ≤ 1 and R = (ua)1−µ φ̌ µ∗ being the MOTS, we have uaφ̌ µ∗ = R 1−µ∗ ≈ uag(u),
∗
which is from the construction of the barriers. Here ≈ is independent of u. Hence it holds
1 1
f φ̌− µ∗ = g(u)f˜(u, ω)φ̌− µ∗ ≈ 1, being independent of u. Thus, the rest arguments stay the
same as in Section 11. We proceed similarly as in Lemma 11.4. With replacing f and f0 by
g(u)f˜ and g(u)f˜0 , we get the corresponding desired conclusion. All the other proofs for the
elliptic part can be obtained in the same way as in Section 11 and Section 12. In summary,
allowing the continuous perturbation for χ̂(−1, u, ω) and ∇4 φ(−1, u, ω) with respect to the u
variable, proceeding the same as in Section 13.1 we then prove
system (1.1). Assigning Christodoulou’s naked-singularity initial data in [18] along u = 0 with
−1 ≤ u ≤ 0 and prescribing perturbed initial data along u = −1 satisfying
X 1 1
uj a− 2 k(∂u )j ∇i χ̂kL∞ 2
u L (S−1,u )
+ uj a− 2 k(∂u )j ∇i ∇4 φkL∞ 2
u L (S−1,u )
≤ B, (13.14)
i≤5,j≤3
then for each B there exist a sufficient small δ = δ(B) and constant a = a(B)19, and the
Einstein-scalar field system admits a unique regular solution in the spacetime region (u, u, ω)
δ̃
satisfying 0 ≤ u ≤ δ and u ≤ |u|Ω2− 2 (u, 0) ≤ 1 with 0 < δ̃ ≪ 1.
Z u
|χ̂|2 (−1, u′ , ω) + [∇4 φ(−1, u′ , ω) − ∇4 φ(−1, 0, ω)]2 du′ = g(u)f˜(ω, u)ua, (13.15)
0
with
1 1 1
g(u) = (C1 Λ + C1 ) 2 · [ln(ln )]− 2
u
where C1 > 2, Λ being a fixed positive constant, and f˜(ω, u) obeying
0 ≤ f˜ ≤ 1 on S2 × (0, δ] and f˜(·, u) ≥ m on Bp (ǫ) (13.16)
for a point p ∈ S2 and constants m ∈ (0, 1), ǫ ∈ (0, π/2), then within the solved hyperbolic
region, there exists a unique MOTS Mu on each H u with 0 < u ≤ δ and the collection of MOTS
{Mu } emerges and censors the central singularity. And they form an achronal apparent horizon.
We also have the corresponding nonlinear instability result with high co-dimensions. It is
summarized in Theorem 1.3. Here we prove it.
Proof. We proceed in the same way as in Theorem 13.2. Only we modify the requirement for
fi and gi as: For
1 1 1
1 ≤ i ≤ k, C1 > 2 and g(u) = (C1 Λ + C1 ) 2 · [ln(ln )]− 2 ,
u
the prescribed fi and gi obey
1 1 1 1
0 ≤ fi ≤ g(u) 2 a 2 on S2 × (0, δ], fi (·, u) ≥ mi g(u) 2 a 2 on Bpi (ǫi ) ⊂ Di ,
1 1 1 1
0 ≤ gi ≤ g(u) 2 a 2 on S2 × (0, δ], gi (·, u) ≥ m′i g(u) 2 a 2 on Bp′i (ǫ′i ) ⊂ Di′
with points pi ∈ Di ⊂ S2 , p′i ∈ Di′ ⊂ S2 and constants mi , m′i ∈ (0, 1), ǫi , ǫ′i ∈ (0, π/2). Note
that we can set a = 1. And as u → 0+ , we have
1
lim g(u) 2 = 0.
u→0+
Therefore, we can choose continuous-in-u and smooth-in-ω functions fi , gi , χ̂, ∂u φ along u = −1

to verify all the requirement of this theorem. The rest proof is the same as in Theorem 13.2.

At the end of this section, we check the size of the initial perturbation. We first prove
Proposition 13.4. For 0 ≤ u ≤ δ, it holds
1 1 C
k[ln(ln )]− 4 k 21 ≤ 1
u Ḣ ([0,δ]) | ln δ1 | 2
with C a uniform positive constant.
19In particular, we can choose a(B) = 1.
88 XINLIANG AN
Proof. Let fˆ(u) := [ln(ln u1 )]− 4 . Using the Sobolev-Slobodeckij space, we have
1
ZZ ZZ
|fˆ(x) − fˆ(y)|2 |fˆ(x) − fˆ(y)|2
kfˆk2 1 = 2
dxdy = 2 dxdy
Ḣ ([0,δ])
2 0≤x≤δ, |x − y| 0≤x≤y≤δ |x − y|2
0≤y≤δ
ZZ ZZ
|fˆ(x) − fˆ(y)|2 |fˆ(x) − fˆ(y)|2
=2 dxdy + 2 dxdy
0≤x≤y≤δ, |x − y|2 0≤x≤y≤δ, |x − y|2
x≤y≤2x y>2x
=Iˆ1 + Iˆ2 .
For Iˆ1 , applying the intermediate value theorem, with θ being a constant ∈ [0, 1], we get
ZZ 2
1 1 5 1 1
Iˆ =2 [ln(ln )]− 4 · 1 · dxdy
0≤x≤y≤δ, 4 x + θ(y − x) ln x+θ(y−x) x + θ(y − x)
x≤y≤2x
ZZ Z δ
1 1 1 1 1
.2 · dxdy . 1 2 · x dx . 1.
0≤x≤y≤δ, (ln 1 )2 y 2 0 (ln ) ln
x≤y≤2x x x δ
To bound Iˆ2 , we set x := eu and y := ev . The requirement 0 < 2x < y < δ implies
1 1
−∞ < u ≤ − ln − ln 2, −∞ < v ≤ − ln and v − u > ln 2 > 0.
δ δ
Utilizing (u, v) and (u, w) coordinates with w = v − u, we then obtain
ZZ 1
|[ln(ln y1 )]− 4 − [ln(ln x1 )]− 4 |2
1
Iˆ2 =2 dxdy
0≤x≤y≤δ, |y − x|2
y>2x
ZZ
1 1
.2 1
e−2v eu ev |[ln(−v)]− 4 − [ln(−u)]− 4 |2 dudv
−∞<u,v≤− ln δ,
v−u≥ln 2
Z +∞ Z − ln 1
δ −w 1 1
.2 e−w |[ln(−u − w)]− 4 − [ln(−u)]− 4 |2 dudw.
ln 2 −∞
The last line can be bounded as

Z +∞ Z − ln δ1 −w Z −u−w
1 5 1
Iˆ2 .2 e−w | − (ln r)− 4 · dr|2 dudw
ln 2 −∞ −u 4 r
Z +∞ Z − ln 1δ −w 2
Z +∞ Z − ln 1 −w
w δ 1
. e−w · du dw = e −w 2
w du dw
ln 2 −∞ |u + w|2 ln 2 −∞ |u + w|2
Z +∞ Z − ln 1
−w 2
δ 1 1
= e w dũ dw . .
ln 2 −∞ |ũ| 2 ln 1δ

[0,δ]
Back to Theorem 13.3, along H−1 we could set the perturbation
1 1 1 1
|χ̂|(−1, u, ω) + |∇4 φ(−1, u, ω) − ∇4 φ(−1, 0, ω)| = [ln(ln )]− 4 · f˜(ω) 2 a 2 (13.17)
u
with f˜(ω) being a smooth function of ω. Hence, to trigger the naked-singularity censoring, we
only need the perturbation of g in scale-critical k · k 23 [0,δ] norm, i.e.,
Ḣ (H−1 )
the k · k 1 [0,δ] = k·k 1 +k·k 1 norm

Ḣ 2 (H−1 ) L2ω Ḣu2 ([0,δ]) L2u Ḣω2 (S−1,u )
1 1
of the above perturbation in (13.17) to be of size (ln 1δ )− 2 a 2 ≪ 1. And as δ → 0+ , we allow the
size of the scale-critical perturbation shrink to 0.
Appendix A. Construction of Localized-In-Angle Initial Data

Along u = −1, in this appendix we construct the localized-in-angle initial data for ∂u φ(−1, u, ω)
and |χ̂|2g (−1, u, ω) with u ≥ 0. Since ∂u φ(−1, u, ω) is a scalar function, we can prescribe the de-
sired localized-in-angle initial data as we need. Notice that χ̂AB (−1, u, ω) is a traceless 2-form.
To prescribe its characteristic initial data, we follow calculations in Chapter 2 of [21] and in
Appendix C of [3]. With stereographic coordinates (θ1 , θ2 ), we have
g|S−1,u = (φ̂|S−1,u )2 ĝ|S−1,u (A.1)
1
with ĝAB (−1, u, θ1 , θ2 ) = 1 2 mAB (−1, u, θ1 , θ2 )
1 + 4 (θ1 + θ22 )
and mAB can be expressed as m(−1, u, θ1 , θ2 ) = exp Ψ(u, θ1 , θ2 ). Here Ψ is a symmetric trace-
free 2-dimensional matrix. Notice that as pointed out in Chapter 2 of [21], Ψ(u, θ1 , θ2 ) is the
free data we can prescribe along u = −1 and φ̂|S−1,u in (A.1) is determined by Ψ(u, θ1 , θ2 ).20
p
For θ12 + θ22 < 20 we set
√ 1
Ψ(u, θ1 , θ2 ) := 2x(u, θ1 , θ2 ) · a 2 · Ψ0 (u, θ1 , θ2 ).
Here ∂x(u, θ1 , θ2 )/∂u is a free scalar function to be prescribed and we also choose

1 0
Ψ0 (u, θ1 , θ2 ) = for u ≥ 0.
0 −1
With a similar calculation carried out in Appendix C of [3], we deduce
 
√ 1
√ 1 exp 2x(u, θ ,
1 2θ ) · a 2 0 
∂
m(−1, u, θ1 , θ2 ) = 2a 2 


√


∂u 1
0 − exp − 2x(u, θ1 , θ2 ) · a 2
∂x
×
(u, θ1 , θ2 )
∂u
From Chapter 2 in [21], we have
1 ∂ĝ 1 ∂
χ̂AB = φ2 = φ2 W 2 (θ1 , θ2 ) mAB (−1, u, θ1 , θ2 ) and
2 ∂u 2 ∂u
1 2 1 −1 CA ∂m ∂m
|χ̂| = (m ) ( )AB (m−1 )BD ( )DC .
2 g 8 ∂u ∂u
For our concrete construction, along u = −1 it hence holds
2 2
1 2 1 √ ∂x 1 1 0 1 ∂x
|χ̂| = · 2· (u, θ1 , θ2 ) · a 2 · tr = a· (u, θ1 , θ2 ) . (1.2)
2 g 8 ∂u 0 1 2 ∂u
With ∂x(u, θ1 , θ2 )/∂u as the freely prescribed scalar function on S−1,u , we therefore obtain the
desired localized-in-angle initial data for |χ̂|2g (−1, u, ω).
References
[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, Adv. Theor.
Math. Phys. Volume 21, Number 1, 1-120 (2017).
[3] X. An, Emergence of apparent horizon in gravitational collapse, Ann. PDE 6, 10 (2020), 89 pages.
[4] X. An, A scale-critical trapped surface formation criterion: a new proof via signature for decay rates, Ann.
PDE 8, 3 (2022), 89 pages.
[5] X. An, N. Athanasiou, A scale-critical trapped surface formation criterion for the Einstein-Maxwell system,
arXiv: 2005.12699, preprint 105 pages, to appear in J. Math. Pures Appl.
[6] X. An, Q. Han, Anisotropic dynamical horizons arising in gravitational collapse, preprint (2020),
arXiv:2010:12524.
[7] X. An, T. He, Dynamics of apparent horizon and a null comparison principle, in preparation.
20In the below, we will prescribe Ψ(u, θ , θ ) in the north polar chart. And Ψ′ (u, θ ′ , θ ′ ) in the south polar
1 2 1 2
chart can be obtained accordingly.
90 XINLIANG AN
[8] X. An, Z. Lim, Trapped surface formation for spherically symmetric Einstein-Maxwell-charged scalar field
system with double null foliation, 34 pages, to appear in Ann. Henri Poincaré.
[9] X. An, H. Tan, in preparation.
[10] X. An, X. Zhang, Examples of naked singularity formation in higher-dimensional Einstein-vacuum space-
times, Ann. Henri Poincaré 19, 619-651 (2018).
[11] L. Andersson, M. Eichmair, J. Metzger, Jang’s equation and its applications to marginally trapped surfaces,
Complex analysis and dynamical systems IV. Part 2, 13-45, Contemp. Math., 554, Israel Math. Conf. Proc.,
Amer. Math. Soc., Providence, RI, 2011.
[12] L. Andersson, M. Mars, W. Simon, Local existence of dynamical and trapping horizons, Phys. Rev. Lett. 95
(2005), 111102.
[13] L. Andersson, M. Mars, W. Simon, Stability of marginally outer trapped surfaces and existence of marginally
outer trapped tubes, Adv. Theor. Math. Phys. 12 (2008), 853-888.
[14] L. Andersson, J. Metzger, The area of horizons and the trapped region, Comm. Math. Phys. 290 (2009),
no.3, 941-972.
[15] N. Athanasiou, P. Mondal, S. T. Yau, Formation of trapped surface in the Einstein-Yang-Mills system,
preprint (2023), arXiv: 2302.06915.
[16] D. Christodoulou, The formation of black holes and singularities in spherically symmetric gravitational
collapse, Comm. Pure Appl. Math. 44 (1991), no. 3, 339–373.
[17] D. Christodoulou, Bounded variation solutions of the spherically symmetric Einstein-scalar field equations,
Comm. Pure Appl. Math. 46 (1993), no. 8, 1131–1220.
[18] 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.
[19] 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.
[20] D. Christodoulou, On the global initial value problem and the issue of singularities, Class. Quantum Grav.
16 (1999), A23–A35.
[21] D. Christodoulou, The formation of black holes in general relativity, Monographs in Mathematics, European
Mathematical Soc. (2009).
[22] D. Christodoulou, S. Klainerman, The global nonlinear stability of the Minkowski space, Princeton mathe-
matical series 41 (1993).
[23] J. L. Costa, The formation of trapped surfaces in the gravitational collapse of spherically symmetric scalar
fields with a positive cosmological constant, Class. Quantum Grav. 37 (2020), 195022 (16pp).
[24] D. Gilbarg, N. S. Trudinger, Elliptic Partial Differential Equations of Second Order Grundlehren, Vol. 224,
Springer-Verlag, Berlin, (1983).
[25] Y. Guo, M. Hadzic, J. Jang, Naked singularities in the Einstein-Euler system, arXiv:2112.10826.
[26] Q. Han, F-H. Lin, Elliptic partial differential equations (2nd ed.), Courant lecture notes, Vol. 1, (2011).
[27] S. Klainerman, F. Nicolo, The evolution problem in General Relativity, Progress in Mathematical Physics,
Birkhaüser (2003).
[28] S. Klainerman, J. Luk, I. Rodnianski, A fully anisotropic mechanism for formation of trapped surfaces in
vacuum, Invent. math. 198, 1–26 (2014).
[29] S. Klainerman, I. Rodnianski, On the the formation of trapped surfaces, Acta Math. 208 (2012), no.2,
211–333.
[30] P. Le, The intersection of a hyperplane with a lightcone in the Minkowski spacetime, J. Differential Geom,
Vol. 109 (2018), no.3, 497-507.
[31] J. Li, J. Liu, A robust proof of the instability of naked singularities of a scalar field in spherical symmetry,
Commun. Math. Phys, Vol. 363, 561–578 (2018).
[32] J. Li, J. Liu, Instability of spherical naked singularities of a scalar field under gravitational perturbations,
to appear in J. Differential Geom.
[33] R. Penrose, Gravitational collapse and space-time singularities, Phys. Rev. Lett. 14 (1965), 57–59.
[34] R. Penrose, Gravitational collapse: The role of general relativity, Riv. Nuovo Cim., 1:252–276, 1969.
[35] I. Rodnianski, Y. Shlapentokh-Rothman, Naked Singularities for the Einstein Vacuum Equations: The
Exterior Solution, to appear in Ann. of Math.
[36] Y. Shlapentokh-Rothman, Naked Singularities for the Einstein Vacuum Equations: The Interior Solution,
arXiv:2204.09891.
[37] J. Singh, A construction of approximately self-similar naked singularities for the spherically symmetric
Einstein-scalar field system, arXiv:2210.11325.
[38] P. Yu, Dynamical Formation of black holes due to the condensation of matter field, preprint (2011),
arXiv:1105.5898.
[39] P. Zhao, D. Hilditch, J. Kroon, Trapped surface formation for the Einstein-scalar system, preprint (2023),
arXiv:2304.01695.
Department of Mathematics, National University of Singapore, 10 Lower Kent Ridge

Road, Singapore, 119076
Email address: matax@nus.edu.sg

Naked Singularity Censoring With Anisotropic Apparent Horizon

Uploaded by

Copyright:

Available Formats

Naked Singularity Censoring With Anisotropic Apparent Horizon

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Naked Singularity Censoring With Anisotropic Apparent Horizon

Uploaded by

Copyright:

Available Formats

NAKED SINGULARITY CENSORING

WITH ANISOTROPIC APPARENT HORIZON

Date: January 5, 2024.

Remark 1. To set the above-mentioned C 0 perturbation, we can let

the k · k 1 [0,δ] =k·k 1 +k·k 1 norm

Setting a = a(B) = 1, we then retrieve assumptions (1.2), (1.3).

Remark 4. In below, we also give a stereoscopic perspective, which demonstrates naked-

Our next result further exhibits a high co-dimensional nonlinear instability:

initial data for |χ̂| and ∂u φ with u ≥ 0

then the 2-sphere Su∗ ,δ is a trapped surface.

Then, for sufficiently small δ, a trapped surface forms in the evolution.

Remark 9. The proof in [6] takes two steps:

timates is decisive. Additional 1+ u 1a factors are incorporated in the top-order boot-

some hidden structures of the Einstein-scalar field system are revealed.

F1 := − 4(1 − µ∗ )RΩω|∇γ φ|e 2,

We then consider the position of the MOTS.

Therefore, we can choose continuous-in-u and smooth-in-ω functions fi , gi , χ̂, ∂u φ along

1.5. Related Results.

g = −2Ω2 (du ⊗ du + du ⊗ du) + g/AB (dθA − dA du) ⊗ (dθB − dB du). (2.1)

χAB = g(DA e4 , eB ), χAB = g(DA e3 , eB ),

In 3 + 1 dimensions, for i, k, l, m = 1, 2, 3, 4 we further decompose the below Weyl tensor

αAB = W (eA , e4 , eB , e4 ), αAB = W (eA , e3 , eB , e3 ),

(1) (2) (1) (2) (1) (2)

(1) (2) (1) (2)

(div φ)A1 ...Ar := ∇B φBA1 ...Ar ,

(curl φ)A1 ...Ar := /ǫ BC ∇B φCA1 ...Ar .

In addition, we define the trace to be

(trφ)A1 ...Ar−1 := (g/−1 )BC φBCA1 ...Ar−1 .

have that χ and χ obey the following null structure equations:

∇Ωe3 ∇A φ =eA (Ωe3 φ) − ΩχA B eB φ,

Recall that in 3 + 1 dimensions for i, k, l, m = 1, 2, 3, 4, the Weyl curvatures obey

Here, ∗ denotes the Hodge dual on Su,u .

2.3. Naked Singularity Initial Data. In [18], Christodoulou constructed a spherically-symmetric

0 O Anisotropic Apparent Horizon

Setting the scalar field function φ(u, u, θ1 , θ2 ) used in this paper to be

via ∂u r(u, 0) = −1 we then translate (2.22) to (2.21).

ψ ∈ {Ωtrχ, Ωχ̂, Ωω̂, Ω∂4 φ}, (2.24)

With constants a, b chosen as in Section 3, we then define

(0) (0) ′(0) (0) (0)

In below, we demonstrate the steps for improving the bootstrap assumptions:6

Ω̊(u, 0) Ω̊c (u, 0)δ̃ Γ̊(u, 0) Ω̊(u, 0)δ̃

Proposition 3.8. The commutator [∇3 , ∇] acting on an (0, r) S-tensor satisfies

By induction, using commutation formula repeatedly we get the schematic form:

i1 +i2 +i3 +i4 =i−1

By replacing Ωβ and Ωβ via the schematic Codazzi equations:

Proposition 3.10. Suppose ∇4 φ = F0 . Let ∇4 ∇i φ = Fi . Then we have

Similarly, assume ∇3 φ = G0 . Let ∇3 ∇i φ = Gi . Then it holds

4. L2 (Su,u ) estimates for Ricci coefficients

10A detailed proof is provided in Proposition 6.1 of [2].

In particular, by (2.38), we have

We now start from the L2 (Su,u ) estimates for Ωχ̂.

Proof. We employ the null structure equation

Proof. We use the following null structure equation

Ω∇4 ∇i (Ω−1 trχ − (Ω−1 trχ)0 )

The third term can be bounded similarly via

Gathering above estimates, we obtain

.Ω−2 uaO2 + ku ∇ (Ω trχ − (Ω−1 trχ)0 )kL∞ ∞ 2

Using ∇A Ω = Ω(ηA + η A )/2, this further implies

Proof. We employ the following schematic null structure equation for ω: