3.4 Proof of the Carleman estimate in a neighborhood of the interface
3.4.2 Estimate in the region G
Here, we place ourselves in the region G, and prove a Carleman estimate for uG,j, and conse- quently for ΞGv.
We introduce a microlocal cut-off functionχG F ∈C∞
c (M∗+), 0≤χG F ≤1, satisfying χG F = 1 on a neighborhood of supp(χG),
χG+χF = 1 on a neighborhood of supp(χG F). (3.75) We choose ζ2 ∈Cc∞(0, X0) such that 0≤ζ2≤1,ζ2 = 1 on a neighborhood of supp(ζ1) (withζ1 defined in (3.61)), and such that ˜ζj = 1 on supp( φ−j1∗
ζj2) where ζj2(x0, y) =ζ2(x0)ψj(y). As in (3.63) we set
χG F,j = ˜ζj φ−j1∗ χG F,
and we define the associated tangential pseudo-differential operator ΞG F by ΞG F = P
j∈J
ΞG F,j, with ΞG F,j =φ∗jOpT(χG F,j) φ−j1∗
ζj2, j∈J,
Note that the local symbol (see Proposition 3.45) of ΞG F in each chart is equal to one in the support of that of ΞG.
We recall that the functionζ=ζ(xn)∈C∞
c ([0,2ε)) satisfiesζ(0) = 1 on [0, ε).
Making use of the Calder´on projector technique forPϕ,jr and of the standard Carleman techniques forPϕ,jl , we obtain the following partial estimate.
Proposition 3.28. Suppose that the weight functionϕsatisfies the properties listed in Section 3.3.1.
Then, for all δ0 >0, there exist C >0 andh0 >0 such that, for all0 < δ ≤δ0 and0 < h≤h0, vr/l ∈Cc∞((0, X0)×S×[0,2ε))andvs∈Cc∞((0, X0)×S)satisfying (3.47), we have
kΞGvrk21+h|ΞGvr|xn=0+|21+h|DxnΞGvr|xn=0+|20≤C
kPϕrvrk20+h2kvrk21+h4|Dxnvr|xn=0+|20
, (3.76)
3.4. Proof of the Carleman estimate in a neighborhood of the interface 121 and
hkΞGvlk21+h|ΞGvl|xn=0+|21+h|DxnΞGvl|xn=0+|20
≤C 1 + δ2 h2
kζPϕrvrk20+h2kΞG Fvrk21+h4|Dxnvr|xn=0+|20+h4kvrk21+h3|vs|21
+C
kPϕlvlk20+h2kvlk21+h|θϕl|21+δ2
h|θrϕ|20+h|θϕr|21+h|Θsϕ|20
. (3.77)
Proof. The function uG,j, defined in (3.64), satisfies (TC•,j), with • = G. On the “r” side, the root configuration described in Lemma 3.24 (and represented in Figure 3.3) allows us to apply the Calder´on projector technique used in [LR97, LR10]. According to [LR10, Remark 2.5] and using Eqs. (2.59), (2.60), and (2.61) therein, applied withvd replaced here byvjr, we have
kurG,jk1+h12|γ0(urG,j)|1+h12|γ1(urG,j)|0.kPϕ,jr vrjk0+hkvjrk1+h2|Dxnvjr|xn=0+|0, (3.78) which is a local version of (3.76).
Let us now explain how such local estimates can be patched together to yield (3.76). Concerning the first term on the left hand-side of (3.78), and using the definition of Sobolev norms given in (3.20)–(3.22), we have
kΞGvrk1. P
j∈JkurG,jk1, |ΞGvr|xn=0+|1. P
j∈J|γ0(urG,j)|1 (3.79) by (3.65) and Lemma 3.53. Similarly we have DxnΞGvr|xn=0+ =P
jφ∗jγ1(urG,j) since φ∗j does not depend on the xn-variable. As a consequence, we obtain
|(DxnΞGvr)|xn=0+|0≤ P
j∈J|φ∗jγ1(urG,j)|0. P
j∈J|γ1(urG,j)|0, (3.80) by Lemma 3.9.
Now concerning the right hand-side of (3.78), we directly have kvjrk1=k φ−j1∗
ζjvrk1=kζ1 φ−j1∗
ψjvrk1.k φ−j1∗
ψjvrk1.kvrk1, (3.81) by the definition ofk.k1 onM+, as well as
|Dxnvrj|xn=0+|0.|Dxnvr|xn=0+|0. (3.82) Finally, we computePϕ,jr vrj = φ−j1∗
Pϕrφ∗j φ−j1∗
ζjvr= φ−j1∗
ζjPϕrvr+ φ−j1∗
[Pϕr, ζj]vr. We have k φ−j1∗
ζjPϕrvrk0=kζ1 φ−j1∗
ψjPϕrvrk0≤ kPϕrvrk0, (3.83) and, using Lemma 3.9,
k φ−j1∗
[Pϕr, ζj]vrk0.k[Pϕr, ζj]vrk0.hkvrk1, (3.84) since [Pϕr, ζj]∈hD1(M+). Finally combining all the estimates (3.79)-(3.84), together with the local inequalities (3.78) summed overj∈J, we obtain the sought global estimate (3.76) onM+.
To obtain Estimate (3.77) on the “l” side we first need a more precise estimate for the “r” side. For this, we introduce another microlocal cut-off function ˜χG F satisfying the same requirements (3.75) asχG F, and such thatχG F = 1 on a neighborhood of supp( ˜χG F). We chooseζ3∈C∞
c (0, X0) such that 0≤ζ3≤1,ζ3= 1 on a neighborhood of supp(ζ1), and such thatζ2= 1 on a neighborhood of supp(ζ3). As in (3.63) we set
˜
χG F,j = ˜ζj φ−j1∗
˜ χG F,
and we define the associated tangential pseudo-differential operator ˜ΞG F by Ξ˜G F = P
j∈J
Ξ˜G F,j, with Ξ˜G F,j =φ∗jOpT( ˜χG F,j) φ−j1∗
ζj3, ζj3=ζ3ψj, j∈J, According to [LR10, Remark 2.5] and using (2.60) and (2.61) therein, applied withvd replaced by ζ(xn) φ−j1∗
ζjΞ˜G Fvr, we have h12|γ0(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|1+h12|γ1(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|0 .kPϕ,jr ζ φ−j1∗
ζjΞ˜G Fvrk0+hkζ φ−j1∗
ζjΞ˜G Fvrk1+h2|γ1 φ−j1∗
ζjΞ˜G Fvr
|0. (3.85) We notice that the right hand-side of this inequality can directly be bounded by global quantities.
First, we have
kζ φ−j1∗
ζjΞ˜G Fvrk1.kΞ˜G Fvrk1 (3.86) Second, we estimate
γ1 φ−j1∗
ζjΞ˜G Fvr0≤ | DxnΞ˜G Fvr
|xn=0+|0, where
DxnΞ˜G Fvr
|xn=0+ = ˜ΞG FDxnvr
|xn=0++ [Dxn,Ξ˜G F]
| {z }
∈hΨ0T(M+)
vr
|xn=0+.
Using Proposition 3.50 and the trace formula (3.23), we have the estimate h2|γ1 φ−j1∗
ζjΞ˜G Fvr
|0.h2|Dxnvr|xn=0+|0+h52kvrk1. (3.87) Concerning the term with Pϕ,jr in the right hand-side of (3.85), we can proceed as in (3.83)-(3.84) to obtain
kPϕ,jr ζ φ−j1∗
ζjΞ˜G Fvrk0=k φ−j1∗
PϕrζζjΞ˜G Fvrk0.kPϕrζΞ˜G Fvrk0+hkΞ˜G Fvrk1 (3.88) Moreover, using Proposition 3.48, we have ˜ΞG F(1−ΞG F)∈h∞Ψ−∞T (M+), as their local symbols in every chart have disjoint supports by Proposition 3.52, because of the supports ofζ3 and ˜χG F. We then obtain with Proposition 3.50
hkΞ˜G Fvrk1.hkΞ˜G FΞG F vrk1+hkΞ˜G F(1−ΞG F)vrk1.hkΞG Fvrk1+h2kvrk1. (3.89) We also have
kPϕrζΞ˜G Fvrk0.kΞ˜G FζPϕrvrk0+k[Pϕr,Ξ˜G Fζ]vrk0. (3.90) Arguing as above with Propositions 3.48 and 3.52, and also Corollary 3.49, we have
[Pϕr,Ξ˜G Fζ] = [Pϕr,Ξ˜G Fζ]
| {z }
∈hΨ1(M+)
ΞG F + [Pϕr,Ξ˜G Fζ](1−ΞG F)
| {z }
∈h∞Ψ−∞(M+)
so that (3.90) now reads with Proposition 3.50
kPϕrζΞ˜G Fvrk0.kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1. (3.91) The three estimates (3.88), (3.89) and (3.91) give
kPϕ,jr ζ φ−j1∗
ζjΞ˜G Fvrk0.kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1. (3.92) Combining (3.85) together with (3.86)–(3.87), (3.89) and (3.92), we finally have
h12|γ0(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|1+h12|γ1(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|0
.kζPϕrvrk0+hkΞG Fvrk1+h2|Dxnvr|xn=0+|0+h2kvrk1. (3.93) Then, we need the following lemma to come back to the variableurG,j = OpT(χG,j) φ−j1∗
ζjvr on the left hand-side of (3.93).
3.4. Proof of the Carleman estimate in a neighborhood of the interface 123 Lemma 3.29. There existsR∈h∞Ψ−∞T (M+), such that
OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr=urG,j+ φ−j1∗ Rvr. This lemma is proven in Appendix 3.8.6. As a consequence we have
h12|γ0(urG,j)|1.h12|γ0(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|1+h12|γ0( φ−j1∗
Rvr)|0 .h12|γ0(OpT(χG,j) φ−j1∗
ζjΞ˜G Fvr)|1+h2kvrk1 with the trace formula (3.23). This, together with Estimate (3.93) give
h12|γ0(urG,j)|1.kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1+h2|Dxnvr|xn=0+|0. (3.94) Lemma 3.29 also yields
h12|γ1(urG,j)|0.h12|γ1(OpT(χG,j) φ−j1∗
ψjΞ˜G Fvr)|0+h12|γ1( φ−j1∗
Rvr)|0, .h12|γ1(OpT(χG,j) φ−j1∗
ψjΞ˜G Fvr)|0+h2kvrk1+h2|Dxnvr|xn=0+|0, (3.95) Combining (3.93) together with (3.95), we finally obtain
h12|γ1(urG,j)|0.kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1+h2|Dxnvr|xn=0+|0. (3.96)
On the “l” side, we apply the Carleman method. With the properties of the weight function of Section 3.3.1 and in particular by (3.56), and by Lemma 2 in [LR95], we then have
hkulG,jk21+Re
hBl(ulG,j) +h2 (DnulG,j+Ll1ulG,j)|xn=0+, Ll0ulG,j|xn=0+
0
.kPϕ,julG,jk20, (3.97) for 0< h≤h0, h0 sufficiently small, whereLl1∈D1
T,Ll0∈Ψ0T. The quadratic formBl is given by Bl(ψ) =
2∂xnϕlj|xn=0+ Bl1 B1l′ Bl2
! γ1(ψ) γ0(ψ)
, γ1(ψ)
γ0(ψ)
!
0
, supp(ψ)⊂(0, X0)×U˜j×[0,2ε), (3.98) whereB1l,B1l′ ∈D1
T, with principal symbolsσ(B1l) =σ(Bl1′) = 2 φ−j1∗
ql1|xn=0+andB2l ∈D2
T, with σ(Bl2) =−2∂xnϕlj φ−j1∗
q2l|xn=0+. Observe that we have
(DnulG,j+Ll1ulG,j)|xn=0+, Ll0ulG,j|xn=0+
0
.|γ1(ulG,j)|20+|γ0(ulG,j)|21. (3.99) and
|Bl(ulG,j)|.|γ0(ulG,j)|21+|γ1(ulG,j)|20. (3.100) Now, using (3.97), together with the estimates (3.99) and (3.100), we have,
hkulG,jk21.kPϕ,jl ulG,jk20+h|γ0(ulG,j)|21+h|γ1(ulG,j)|20. (3.101) It remains to estimate the traces on the “l” side by the traces on the “r” side, through the transmission conditions (TC•,j) :
γ0(ulG,j) =γ0(urG,j) +θGl,j−θrG,j
γ1(ulG,j) = δcsj
h i cljPϕ,js γ0(urG,j)−θrG,j
−βγ1(urG,j) +kγ0(urG,j)−G˜1, usG,j =γ0(urG,j)−θGr,j.
As a consequence,γ0(ulG,j) andγ1(ulG,j) can be estimated as follows
|γ0(ulG,j)|1≤ |γ0(urG,j)|1+|θlG,j|1+|θGr,j|1,
|γ1(ulG,j)|0.|γ1(urG,j)|0+δ
h|Pϕ,js γ0(urG,j)|0+ δ
h|Pϕ,js θrG,j|0+|γ0(urG,j)|0+|G˜1|0. (3.102) We now prove that, on the support of χG,j, the operator Pϕ,js is of order 0. For this, let ˜χ ∈ C∞
c (T∗(Rn)), be equal to one on a neighborhood of the supp(χG,j|xn=0+). We then have γ0(urG,j) = OpT(χG,j)vjr|xn=0+= OpT( ˜χ) OpT(χG,j)vrj|xn=0++ OpT(1−χ) Op˜ T(χG,j)
| {z }
∈h∞Ψ−∞T
vjr|xn=0+,
which yields
Pϕ,js γ0(urG,j) = Pϕ,js OpT( ˜χ)
| {z }
∈Ψ0T
γ0(urG,j) +Pϕ,js OpT(1−χ) Op˜ T(χG,j)
| {z }
∈h∞Ψ−∞T
vjr|xn=0+.
This, together with the trace formula (3.23) gives the estimate, δ
h|Pϕ,js γ0(urG,j)|0≤Cδ
h|γ0(urG,j)|0+CNδhNkvjrk1, N ∈N.
Similarly, we have the estimate δ
h|Pϕ,js θrG,j|0≤Cδ
h|θrG,j|0+CNδhN|θϕ,jr |0. δ h|θϕ,jr |0. The last two estimates and the second equation of (3.102) yield,
|γ1(ulG,j)|0.|γ1(urG,j)|0+ 1 + δ h
|γ0(urG,j)|0+δ
h|θϕ,jr |0+|G˜1|0+CNδhNkvrjk1, N∈N. Using estimates (3.94) and (3.96) to bound the traces on the “r” side, we obtain
h12|γ1(ulG,j)|0. 1 + δ h
kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1+h2|Dxnvr|xn=0+|0 + δ
h12|θϕ,jr |0+h12|G˜1|0,
for 0< h≤h0, and, using (3.74) to estimate the remainder, we have h12|γ1(ulG,j)|0. 1 + δ
h
kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1+h2|Dxnvr|xn=0+|0 +h32|vsj|1+h12|θrϕ,j|0
+h12|Θsϕ,j|0+h12|θlϕ,j|0, (3.103) We observe now that the first line of (3.102) together with (3.94) yields
h12|γ0(ulG,j)|1.kζPϕrvrk0+hkΞG Fvrk1+h2kvrk1+h2|Dxnvr|xn=0+|0
+h12|θlϕ,j|1+h12|θrϕ,j|1. (3.104) Combining (3.68), with (3.101), (3.103) and (3.104) we obtain
hkulG,jk21+h|γ0(ulG,j)|21+h|γ1(ulG,j)|20
. 1 + δ2 h2
kζPϕrvrk20+h2kΞG Fvrk21+h4kvrk21+h4|Dxnvr|xn=0+|20+h3|vjs|21
+h|θlϕ,j|21+δ2
h|θrϕ,j|20+h|θϕ,jr |21+h|Θsϕ,j|20+kPϕ,jl vjlk20+h2kvljk21. (3.105) This is a local version of (3.77). Patching together onM+ the local Carleman estimates (3.105) as we did in (3.79)-(3.84) yields (3.77). This concludes the proof of Proposition 3.28.
3.4. Proof of the Carleman estimate in a neighborhood of the interface 125
Reξn
Imξn
ρl,+
Pϕl
ρl,− Reξn
Imξn
ρr,− ρr,+
Pϕr
Figure 3.5 – Root configuration in the regionF.