• Tidak ada hasil yang ditemukan

Proof of Proposition 9.3

9.3 Preparations for the Proofs of Main Estimates for the Eikonal Function Quatities

9.3.1 Proof of Proposition 9.3

kξkLQg/(St,u)= Z

S2

|ξ|Qg/g/

Q1

Z

S2

er2|ξ|Qg/e/ Q1

= erQ2ξ

LQω(St,u). (9.47)

Proof of (9.30).

r12ξ

2

= Z

r−1|ξ|2

≤ r−1

Z

|ξ|412

≈ kξk2 . (9.48)

We now provekξkL4

g/(St,u).kξkH1t). For fixedtandu, letϕbe a cut-off function ofuverifying 0≤ϕ≤1, ϕ(u) =1 and supp(ϕ)⊂[u−t−u2 ,u+t−u2 ]. Integrating on the region S

u≤u0t+u2

St,u0, by the Fundamental Theorem of Calcules, we have:

kξk4L4 g

/(St,u)=−

Z u u−t−u2

Z

St,u0

D

/N|ϕξ|4+trg/θ|ϕξ|4

bdϖg/du0 (9.49)

=− Z u

u−t−u2 Z

St,u0

ϕ4|ξ|3·/DNξ+ϕ3|ξ|4·/DNϕ+trg/θ|ϕξ|4

bdϖg/du0.

For the first term on the RHS of (9.49), by bootstrap assumption (9.5b), the H¨older’s inequality and Sobolev embeddingL6,→H1, we have:

Z u u−t−u2

Z

St,u0

ϕ4|ξ|3·/DNξ

bdϖg/du0

.kD/NξkL2t)

|ξ|3

L2t).kξk4H1t), (9.50) For the second term on the RHS of (9.49), by bootstrap assumption (9.5b), and noticing that|D/Nϕ|.t−u1 , we have

Z u u−t−u2

Z

St,u0

ϕ3|ξ|4·/DNϕ

bdϖg/du0

.

1 t−u

Z u u−t−u2

Z

St,u0

|ξ|4g/du0

. (9.51)

By Lemma 9.4, Z

St,u0

|ξ|4g/. er12ξ

L2g/(St,u0)

|ξ|3

L2g/(St,u0)er12 .kξkH1t)

|ξ|3

L2g/(St,u0)er12. (9.52) Then by (9.29), (9.51), (9.52), the H¨older’s inequality, the Sobolev embeddingL6,→H1, we have:

1 t−u

Z u u−t−u2

Z

St,u0

|ξ|4g/du0

. 1

t−ukξkH1t)

|ξ|3

L2t)

Z u

u−t−u2 er(t,u0)du0 1/2

(9.53) .kξk4H1t).

Now we consider the third term on the RHS of (9.49), that is, Ru−u t−u 2

R St,u0

trg/θ|ϕξ|4

bdϖg/du0. By bootstrap assumptions (6.13), (9.4) and (9.5b) and the H¨older’s inequality, we have:

Z u u−t−u2

Z

St,u0

trg/θ|ϕξ|4

bdϖg/du0. Z u

u−t−u2 Z

St,u0

er−1|ϕξ|4

g/du0 (9.54)

+ Z u

u−t−u2 Z

St,u0

trg/θ−2 er

|ϕξ|4

g/du0 .

Z u u−t−u2

Z

St,u0

er−1|ξ|4

g/du0+

trg/θ−2 er L3t)

kξk4L6t).

By bootstrap assumptions (9.5c) where we have trg/θ−2

er

L3t)≤1, and Sobolev embeddingL6,→H1for kξk4L6t), we have:

Z u u−t−u

2

Z

St,u0

trg/θ|ϕξ|4

bdϖg/du0. Z u

u−t−u

2

Z

St,u0

er−1|ξ|4

g/du0+kξk4H1t). (9.55)

Using Lemma 9.4, the H´older inequality and the Sobolev embeddingL6,→H1, we have:

Z u u−t−u2

Z

St,u0

er−1|ξ|4

g/du0.kξkH1t)

|ξ|3

L2t)

Z u u−t−u2

Z

St,u0er−1(t,u0)du0

!1/2

.kξk4H1t). (9.56)

Proof of (9.31). Replaceξbyξ1/2in (9.49). By the same approach, we derive the following:

kξk2

L2g/(St,u).kξkH1t)kξkL2t). (9.57)

Integrating (9.57) with respect touonΣt, we obtain (9.31).

Proof of (9.32). We defineCu,ω:=CuT

A=ω}. For anyt∈[0,T∗;(λ)], letφbe a smooth function verify- ing 0≤φ≤1,φ(t) =1 andsuppφ⊂[12t,32t]. Since|L(φ)|.er−1, by the Fundamental Theorem of Calcules, we have:

erp|ξ|2pg/ φ2p

(t) = Z t

0

L

erp|ξ|2pg/

φ2p+erp|ξ|2pg/ L φ2p

dτ (9.58)

. Z t

0erp−1|ξ|2pg/ +erp|ξ|2p−2g/ /DLξ·ξdτ.

Hence, er12ξ

2p

L2pωLt(Cu). Z

S2 Z t

0erp−1|ξ|2pg/ +erp|ξ|2p−2g/ D/Lξ·ξdτdϖ/e (9.59) .

Z

S2

kξkL2

t(Cu,ω)kξkL2

t(Cu,ω)+kξkL2

t(Cu,ω)kerD/LξkL2 t(Cu,ω)

erp−1ξ2p−2

Lt(Cu,ω)/e .kξkL

ωL2t(Cu)

r12ξ

2p−2 L2pωLt (Cu)

kerD/LξkLp

ωL2t(Cu)+kξkLp

ωL2t(Cu)

.

This yields the desired result.

Proof of (9.33b). We first consider the Euclidean sphere case, applying the Morrey’s inequality and the

Gagliardo–Nirenberg interpolation inequality with dimensionn=2, ans using Young’s inequality, we have:

kξk

C

0,1−2 ω Q(St,u)

.

e/ξ

LQω(St,u)+kξk

LQω(St,u) (9.60)

.

e/ξ

LQω(St,u)+

e/ξ

1−Q2 L2ω(St,u)kξk

2 Q

L2ω(St,u)+kξkL2 ω(St,u)

.

e/ξ

LQω(St,u)+

e/ξ

1−2

Q

LQω(St,u)kξk

2 Q

L2ω(St,u)+kξkL2 ω(St,u)

.

e/ξ

LQω(St,u)+kξkL2 ω(St,u).

Noticing that for mn

tensorfieldξ, we have:

|ξ|e/≈erm−n|ξ|g/. (9.61)

Thus, for any 0<p<∞, there holds:

e/ξ e/

Lωp(St,u)≈ er∇e/ξ

Lωp(St,u). (9.62)

Now we consider the difference between round metric/eand the geometric sphere metricg/. We have:

/∇ξ=∇/eξ+ (ΓΓΓ−ΓΓΓ(e/))ξ, (9.63)

whereΓΓΓ(e/) are the Christoffel symbols ofe/relative to Euclidean sphere coordinatesωA. By the bootstrap assumptions (9.3a)-(9.3b), we have:

erm−n

(ΓΓΓ−ΓΓΓ(e/)/e

LQω(St,u)−ε0kξkL

ω(St,u). (9.64)

The RHS of (9.64) can be absorbed in the left hand side of (9.60).

Proof of (9.33a). We use the similar approach as in the proof of (9.33b), for the Euclidean sphere case, we have:

|ξ|/e

LQω(St,u).

/eξ e/

1−Q2 L2ω(St,u)

|ξ|e/

2 Q

L2ω(St,u)+ |ξ|/e

L2ω(St,u). (9.65) The proof of (9.65) can be reduced, by a partition of unity, to the case whereξhas compact support in a local chartU⊂St,u. Then we can apply the Gagliardo–Nirenberg interpolation inequality with dimensionn=2.

Now we consider the difference between Euclidean round metric/eand the geometric sphere metricg/. We

have:

/∇ξ=∇/eξ+ (ΓΓΓ−ΓΓΓ(e/))ξ. (9.66)

whereΓΓΓ(e/) are the Christoffel symbols ofe/relative to Euclidean sphere coordinatesωA. By the bootstrap assumptions (9.3a)-(9.3b), by H¨older’s inequality, we have

erm−n

(ΓΓΓ−ΓΓΓ(e/)e/

L2

ω(St,u).

ΓΓΓ−ΓΓΓ(e/)

Lωp(St,u)kξk

L

2p p−2 ω (St,u)

−4ε0kξkL ω(St,u)er

p−2

p . (9.67)

Note thater.λ1−8ε0. Hence, forpsufficiently close to 2, we have:

erm−n

(ΓΓΓ−ΓΓΓ(e/)e/

L2

ω(St,u).kξkL

ω(St,u). (9.68)

(9.33a) is then obtained by applying (9.33b).

Proof of (9.34). The proof is similar to the proof of (9.32). We use a cut-off function ofu with support in u∈[−45T∗;(λ),t]instead oft∈[0,T∗;(λ)], and we consider that the derivative vectorfield

u alongΣt instead ofLalongCu. Also noticing that

/D

∂u

ξ

g/.|(D/N,/∇)ξ|g/and:

er12Q1ξ

2

L2Qg/ Lut)= er12ξ

2

L2QωLut), (9.69)

we obtain the desired estimate.

Proof of (9.35a). We decompose f as follows:

f =

ν>1

Pνf+P≤1f. (9.70)

Using (9.33a), we have

erkPνfk

LQω(St,u).

/ePνf

1−Q2

L2g/(St,u)kPνfk

2 Q

L2g/(St,u)+kPνfkL2 g

/(St,u). (9.71) Integrating (9.71) with respect toualongΣt, using the finite band property and bootstrap assumption (9.3a)- (9.3b), and summing overν>1, when 0<1−Q2 <N−2, we have:

1−2

For the low frequency terms, we use Bernstein’s inequality:

kerP≤1fk

L2uLQωt)≈ kP≤1fk

L2uLgQ/t).kfkL2t). (9.73)

Lemma 9.5(Transport lemma). Let m be a constant, and letξandFbe St,u-tangent tensorfields such that the following transport equation holds along the null cone portionCu⊂M:

D

/Lξ+mtrg/χξ=F. (9.74)

Then we have the following identities, where[u]+:=max{u,0}:

mξ)(t,u,ω) = lim

τ↓[u]+

mξ)(τ,u,ω) + Z t

[u]+

mF)(τ,u,ω)dτ, (9.75) (er2mξ)(t,u,ω) = lim

τ↓[u]+(er2mξ)(τ,u,ω) (9.76)

+ Z t

[u]+

(er2mF)(τ,u,ω) +m

er2m 2

er−trg/χ

ξ

(τ,u,ω)dτ.

Similarly, ifξ,FandGare St,u-tangent tensorfields such that the following transport equation holds:

D /Lξ+2m

er ξ=G·ξ+F, (9.77)

and if

kGkL

ωL1t(Cu)≤C, (9.78)

then under the bootstrap assumptions, the following estimates holds (where the implicit constants in the estimates below depend on the constant C on (9.78)):

er2mξ

g/(t,u,ω). lim

τ↓[u]+

er2mξ

g/(τ,u,ω) + Z t

[u]+

er2mF

g/(τ,u,ω)dτ. (9.79) Proof of (9.75). By (9.74), we have:

L(υmξ) =υm/DLξ+mυm−1(Lυ)ξ=υm /DLξ+mtrg/χ

mF. (9.80)

Integratingυmalong null geodesics, we get the desired equation. Proof of (9.76) is by the same process.

Proof of (9.79). By (9.77), we have:

L er2mξ

g/

=er2m/DLξ· ξ

|ξ|g/+2mer2m−1|ξ|g/ (9.81)

=er2m

−2m

er ξ+G·ξ+F

· ξ

|ξ|g/+2mer2m−1|ξ|g/

=er2m G|ξ|g/+F· ξ

|ξ|g/

! .

Integrating er2mξ

g/along null geodesics and then using the Grownwall’s inequality, we get the desired esti- mate.

Proposition 9.6(Estimates for the fluid variables). ~Ψ, ~ω,~S, ~C,Dare rescaled variables defined in Definition 6.3. Under the bootstrap assumptions, for any2≤Q≤p, where0<δ0<1−2p<N−2, the following estimates hold onM:

∂∂∂(~Ψ, ~ω,~S), ~C,D L2

uLωpt), er1/2

∂(~Ψ, ~ω,~S), ~C,D L

uL2ωpt)−1/2, (9.82a)

er1−Q2

2(~Ψ, ~ω,~S),∂∂∂(C~,D) L2

uLQg/t)−1/2, (9.82b)

∂∂∂(~Ψ, ~ω,~S), ~C,D L2

tLω(Cu)−1/2−4ε0, (9.82c)

∂∂∂(~Ψ, ~ω,~S), ~C,D

L2tLpω(Cu)−1/2−4ε0, (9.82d)

er

∂(~Ψ, ~ω,~S), ~C,D

L2tLω(Cu)1/2−12ε0, (9.82e)

(∇/ ,/DL)∂∂∂~Ψ L2(Cu),

∂∂∂(C~,D) L2(Cu),

er1−2p(∇/ ,/DL)

∂∂~Ψ, ~C,D

L2tLpg/(Cu)−1/2, (9.82f)

er1/2∂∂∂

~Ψ, ~ω,~S

L2uLtL2pω(M)−4ε0. (9.82g) Moreover, for any smooth functionf, we have:

er(∇/ ,/DL) f

~Ψ, ~ω,~S,~L ∂∂∂~Ψ, ~C,D L2

tLQω(Cu)−1/2, (9.83a)

er∂∂∂

f

~Ψ, ~ω,~S,~L ∂∂∂(~Ψ, ~ω,~S), ~C,D

L2uLQωt)−1/2, (9.83b)

er1/2f

~Ψ,~L

∂∂

~Ψ, ~ω,~S

L2uLtL2pω(M)−4ε0. (9.83c)

Proof of (9.82b). By using (9.35a), and rescaling the top order energy estimates (5.1), we have:

er1−Q2

∂∂∂2(~Ψ, ~ω,~S),∂∂∂(C~,D)

Lu2LQg/t)≈ er

∂∂∂2(~Ψ, ~ω,~S),∂∂∂(C~,D)

L2uLQωt) (9.84) .

∂∂∂2(~Ψ, ~ω,~S),∂∂∂(C~,D)

HN−2t)−1/2.

Proof of (9.82a). By (9.31), (9.33a), (9.35a), (9.82b) and the energy estimates (5.1), we have:

∂∂∂(~Ψ, ~ω,~S), ~C,D L2

uLωpt). er∇/

∂∂(~Ψ, ~ω,~S), ~C,D

1−2p Lu2L2ωt)

∂∂∂(~Ψ, ~ω,~S), ~C,D

2 p

L2uL2ωt) (9.85) +

∂∂∂(~Ψ, ~ω,~S), ~C,D L2

uL2ωt)

.

∂∂∂(~Ψ, ~ω,~S), ~C,D H1

t)−1/2.

For estimate of the second norm, by (9.34), (9.33b), (9.35a), (9.31), (9.82b) and the energy estimates (5.1), we have:

er1/2

∂∂

∂(~Ψ, ~ω,~S), ~C,D

2

L2pωLut).

∂∂∂(~Ψ, ~ω,~S), ~C,D

L2uLωt)· (9.86)

er(D/N,/∇)

∂∂(~Ψ, ~ω,~S), ~C,D

LQωL2ut)+

∂∂∂(~Ψ, ~ω,~S), ~C,D LQωL2ut)

.

er∇/

∂∂(~Ψ, ~ω,~S), ~C,D L2

uLQωt)+

∂(~Ψ, ~ω,~S), ~C,D L2

uLω2t)

·

er(D/N,/∇)

∂∂(~Ψ, ~ω,~S), ~C,D LQ

ωL2ut)+

∂∂∂(~Ψ, ~ω,~S), ~C,D LQ

ωL2ut)

.

∂∂∂(~Ψ, ~ω,~S), ~C,D

HN−1t)−1.

Proof of (9.82g). Using the fact thatu.λ1−8ε0 inM and (9.82a), we deduce:

er1/2f

~Ψ,~L

∂∂∂

~Ψ, ~ω,~S

L2uLtL2ωp(M)1/2−4ε0 er1/2f

~Ψ,~L

∂∂∂

~Ψ, ~ω,~S

LuLtL2ωp(M) (9.87) .λ−4ε0.

Proof of other estimates in Proposition 9.6. (9.82c)-(9.82e) are direct results of rescaled bootstrap assump- tions (9.2).

The first and second estimates of (9.82f) are direct results of energy estimates along acoustic null hypersur- faces (8.3)-(8.4). The last estimates of (9.82f) is [36, Lemma 5.5], we use the Sobolev inequality (9.33a) and the following inequality in [34, Proposition 2.7]:

l>1

lN−2/Pl(D/L,/∇)f

2

L2tL2g/(Cu).

ν>1

ν2(N−2)

F(wave)[Pνf;Cu] +kPνfkH1t)

(9.88) +F(wave)[Pνf;Cu] +kfk2H1t),

where/Pl is the Littlewood-Paley projection operator onSt,u. We obtain the results by the energy estimates along constant-time hypersurfaces (5.1) and along acoustic null hypersurfaces (8.3)-(8.4).

(9.83a)-(9.83c) are the results of (9.82f), (9.82b) and (9.82g) respectively.

Proposition 9.7(Needed estimates to recover and improve the bootstrap assumption). Under the bootstrap assumptions, the following estimates hold whenever q>2is sufficiently close to 2, where p is the same as in the previous proposition.

Estimates for time-integrated terms

λ−1 er−1

Z t [u]+

er(C~,D)

g/

L2tLx(M)

−1/2−12ε0, (9.89a) λ−1

er−2

Z t [u]+

er2(C~,D) g/

L2tLx(M)

−1/2−12ε0, (9.89b) λ−1

er−2

Z t [u]+

er2(C~,D) g/

L

q t2Lx(M)

2 q−1−4ε0

4 q+2

, (9.89c)

λ−1 er−1/2

Z t [u]+

er(C~,D)

g/

LtLuLωp(M)

−1/2−12ε0, (9.89d) λ−1

er−3/2

Z t [u]+

er2(C~,D) g/

LtLuLωp(M)

−1/2−12ε0, (9.89e) λ−1

er−1

Z t [u]+

er2(C~,D) g/

L(M)

−16ε0, (9.89f)

λ−1 er32

Z t [u]+

er2(C~,D) g/

L(M)

−1/2−12ε0, (9.89g) λ−1

er−2

Z t [u]+

er2(C~,D) g/

L(M)

−1−8ε0. (9.89h)

λ−1 er−3/2

Z t [u]+

er3/∇(C~,D) g/

LtLuLωp(M)

−1/2−8ε0, (9.90a) t

λ−1 er−3/2

Z t [u]+

er3(∂∂∂~Ψ,∂∂∂~ω,∂∂∂~S)·

∂∂∂~Ψ,tr

ge

/χ˜(Small),χ,ˆ ζ,er−1 g/

LtLuLpω(M)

−1/2−12ε0, (9.91a) λ−1

er−2

Z t [u]+

er3(∂∂∂~Ψ,∂∂∂~ω,∂∂∂~S)·

∂∂∂~Ψ,tr

eg

/χ˜(Small),χ,ˆ ζ,er−1

g/

L2tLuLpω(M)

−1/2−12ε0, (9.91b) λ−1

er−1

Z t [u]+

er2(∂∂∂~Ψ,∂∂∂~ω,∂∂∂~S)·

∂∂∂~Ψ,tr

eg

/χ˜(Small),χ,ˆ ζ,er−1 g/

L2uL2tLpω(M)

−16ε0, (9.91c) λ−1

er−1/2

Z t [u]+

er2(∂∂∂~Ψ,∂∂∂~ω,∂∂∂~S)·

∂∂∂~Ψ,tr

eg

/χ˜(Small),χ,ˆ ζ,er−1 g/

L2uLt Lpω(M)

−16ε0. (9.91d)

λ−1 er−1

Z t [u]+

er2(∂∂∂C~,∂∂∂D) g/

L2uL2tLωp(M)

−12ε0, (9.92a)

λ−1 er−1/2

Z t [u]+

er2(∂∂∂C~,∂∂∂D) g/

L2uLtLωp(M)

−12ε0. (9.92b)

space-time norm estimates

λ−1

er(C~,D)

Lt2LuLωp(M).−1/2−8ε0, (9.93a) λ−1

er(C~,D) L

q

t2LuLωp(M).

2 q−1−4ε0

4 q+1

, (9.93b)

λ−1

er(C~,D)

L2uL2tLωp(M).−12ε0, (9.93c) λ−1

er∂∂∂(C~,D)

L2uL1tLωp(M).12−8ε0, (9.93d) λ−1

er

∂∂∂~Ψ,∂∂∂~ω,∂∂∂~S

·

∂∂∂~Ψ,tr

eg

/χ˜(Small),χ,ˆ ζ,er−1

L2uL1tLωp(M).12−10ε0 . (9.93e)

Dokumen terkait