You are on page 1of 14

343

Math. Proc. Camb. Phil. Soc. (1997), 122, 343
Printed in the United Kingdom

On the limiting absorption and amplitude principles for transmission
problems with unbounded interfaces
By BO ZHANG
Department of Mathematics and Statistics, Brunel University,
Uxbridge UB8 3PH, UK
(Received 4 October 1995)
Abstract
We consider the transmission problem for second-order differential equations with
unbounded interfaces. The spaces B and B ∗ are used to prove the limiting absorption
principle for the steady-state problem and to establish the resolvent estimate at low
frequencies for the steady-state operator. These are then applied in showing the
validity of the limiting amplitude principle for such an operator.

1. Introduction
In this paper we mainly deal with the asymptotic behaviour (as t → +∞) of the
equation
(t, x) ∈ R+ × Ω

Wtt − C 2 ρ∇ . (ρ−1 a∇W ) = f (x)e−iωt ,

(1·1)

under the transmission conditions
αW1 (t, x) = W2 (t, x),
βρ−1
(1) (a(x)∇W1 (t, x)) . ν(x)

=

(t, x) ∈ R+ × Γ

ρ−1
(2) (a(x)∇W2 (t, x)) . ν(x),

(1·2)
+

(t, x) ∈ R × Γ (1·3)

and the initial conditions
W (0, x) = Wt (0, x) = 0,

x ∈ Ω.

(1·4)

Here α, β are two non-zero complex number, R+ = (0, ∞), Ω = Ω1 x Ω2 and Ωj (j =
1, 2) are two unbounded domains in Rn (n > 2) separated by an infinite surface
Γ ∈ C 2 , ν(x) = (ν1 (x), . . . , νn (x)) is the unit normal to Γ at x ∈ Γ drawn in the
direction from Ω1 to Ω2 , W = (W1 , W2 ), Wt = ∂W/∂t, ω is a positive constant, a(x) is
a real-valued symmetric and positively definite matrix which tends to the identity
matrix I as r = |x| → ∞, C and ρ are two positive functions tending to Cj > 0 and
ρj > 0, respectively, as r → ∞ in Ωj (j = 1, 2), f is an L2 − function and ρ(j) (x) is the
value of ρ(x) when x approaches the interface from Ωj with j = 1, 2.
Throughout this paper we assume that the following assumptions hold.
¯ > 0.
(A1) αβ
(A2) There is a d > 0 such that d < C(x) < 1/d, d < ρ(x) < 1/d, and
|ρ(x) − ρj | + |∇ρ| + |C(x) − Cj | = O(|x|−1−δ )
as |x| → ∞ in Ωj (j = 1, 2) for some 0 < δ < 1.

(a(x)x) = 0. 12] for a paraboloid-type interface in the case when a = I. and Lε2 = Lε2 (Ω) is the weighted L2 space defined in the next section. with Iz  0. 6. where u+ = (u1 .l |∂k ∂l aij | = O(|x| (A5) There is a constant R0 > 0 such that for all x ∈ Γ with |x| > R0 P P ν(x) . 14] in the case when a = I. Define q = q(x) = qj = Cj−2 ρ−1 j (> 0) for x in Ωj (j = 1. R(z). and there is a constant C0 > 0 such that ζ .k. the above work and [1. α = β = 1 and C(x) = Cj in Ωj . (ρ−1 a∇u+ ) − ω 2 µu+ = µf. chapter 1] for studying the asymptotic behavior. The limiting amplitude principle for the problem (1·1)–(1·4) is proved based on a general theory developed by Eidus [2. (A4) i. By this theory. of solutions to the Cauchy problem for the abstract wave equation.j. ν = u2 . ρ = 1 and Γ is a local perturbation of a hyperplane by introducing a radiation condition to the problems. αu1 −1 βρ(1) (a∇u1 ) . (1·8) which is given by LABP: u+ = R(ω 2 + i0)f = lim R(ω 2 + iτ )f τ →0+ (1·9) with σ > 21 .344 Bo Zhang (A3) a(x) = [akl ]n×n is a real-valued symmetric matrix with akl ∈ C 2 (Rn ). W2 ) of the problem (1·1)–(1·4) has the property that if f ∈ Lη2 . and the validity of the limiting amplitude principle (LAMP) for the problem (1·1)–(1·4). l=1 akl (x)ζk ζl > C0 |ζ|2 . 7. for large time. 5. Then we show that the solution uλ of (1·6)–(1·8) with any λ > 0 instead of ω 2 is Hoelder . 4. Then by (A2) |µ(x) − q| = O(|x|−1−δ ) (1·10) as |x| → ∞ in Ω. −1−δ ) as |x| → ∞ in Rn . then W = u+ exp (−iωt) + o(1) in the norm of L−η 2 (Ω) with η > the problem 3 2 (1·5) as t → +∞. k |∂k aij | + i. on Γ. The last statement means that the solution W = (W1 . j. Under the above assumptions we prove in this paper the validity of the limiting absorption principle (LABP) for the operator A = −C 2 ρ∇ . Here µ(x) = C −2 (x)ρ−1 (x). and in [9. There is a large literature about these two principles (see. is in the norm of L−σ 2 the resolvent of the self-adjoint realization of the operator A under the transmission conditions (1·7) and (1·8) in a weighted L2 space. ζ ∈ Rn . ν. 2). 2. for example. ρ = 1. 11. (1·6) on Γ (1·7) ρ−1 (2) (a∇u2 ) . 10. ∀x. (a(x)ζ) = n X k. we first have to prove the limiting absorption principle for the associated steady-state problem (1·6)–(1·8) with any λ > 0 instead of ω 2 . = in Ω. The limiting absorption and amplitude principles for A have been proved in [3. u2 ) is the solution of −∇ . 13] and the references there). (ρ−1 a∇) associated with the problem (1·1)–(1·4).

v)0 = ξµuv¯ dx. Preliminaries We use the following abbreviations Ω(R) = {x ∈ Ω||x| < R}. 2. We write Hε0 (V ) = Lε2 (V ). E(R) = Ω\Ω(R) Γ(R) = {x ∈ Γ||x| < R} Ω(R1 . u2 ) ∈ H 1 (Ω) w H0 |αu1 = u2 on Γ} . . ·) = Ω uv¯ dx by (A2). |σ| 6 k} with the weighted norm kuk2k. V ÷ X Z |σ|6k V < x >2ε |∂ σ u|2 dx. . When V = Ω. For an open set V ⊂ Rn and a non-negative integer k We define the weighted Sobolev space Hεk (V ) by Hεk (V ) ÷ {u ∈ H k (V w B(R))∀R > 0| < x >ε ∂ σ u ∈ L2 (V ). and H k (V ) is the standard Sobolev space with the usual inner product (·. ξµdx) with the inner product Z (u. Finally we study the asymptotic behaviour of uλ as λ → 0. space and integral. j > 1. These properties will be examined precisely in what follows. the inner product (·. inner product. ¯ 2−j for x in Ωj (j = 1. ∂ σ = ∂1σ1 · · · ∂nσn . ε.345 Transmission problems with unbounded interfaces continuous in λ. S(R) = {x ∈ Ω||x| = R}. R2 ) = {x ∈ Γ|R1 < |x| < R2 }. Rj = 2j−1 . where R0 = 0. In order to obtain a Hilbert space formulation of the problem (1·1)–(1·4) with a self-adjoint operator. When ε = 0 or k = 0. We also define the spaces B(V ) and B ∗ (V ) by   ! 12 Z ∞   X <∞ |v(x)|2 dx Rj B(V ) = v ∈ L2 (VR )∀R > 0|kvkB(V ) =   Vj j=1 and  Z 2 −1 B (V ) = v ∈ L2 (VR )∀R > 0|kvkB ∗ (V ) = sup R ∗ R>1 VR 2 |v(x)| dx   < +∞ . B(R) = {x ∈ Rn ||x| < R}. . ·)k. R2 ) = {x ∈ Ω|R1 < |x| < R2 } Γ(R1 . V will be also omitted from the norm. Let ε be a real number. Γe (R) = Γ\Γ(R). σj > 0 is an integer. 1 < x >= (1 + |x|2 ) 2 . ∂j = ∂/∂xj . where σ = (σ1 . Ω R Evidently.V . we introduce the following Hilbert space H = {u = (u1 . . We remark that for every ε > 12 the continuous inclusions Lε2 (V ) ⊂ B(V ) ⊂ L2 (V ) ⊂ B ∗ (V ) ⊂ L−ε 2 (V ) (2·1) hold [14]. Vj = {x ∈ V |Rj−1 < |x| < Rj } and VR = {x ∈ V ||x| < R}. σn ). 2) and to It is convenient to define ξ = ξ(x) = ξj = (αβ) introduce the weighted Hilbert space H0 = L2 (Ω. the subscript ε or k will be omitted from the norm and inner product. ·)0 is equivalent to the usual one (·.

A is symmetric and positive with C\[0. u2 ) ∈ H|βρ−1 (1) (a∇u1 ) . v) = αβ Z Ω1 −1 ρ ∇v¯ 1 (a∇u1 )dx + Z Ω2 ρ−1 ∇v¯ 2 (a∇u2 )dx for u. . ∞) ⊂ ρ(A). A ⊂ A∗ . the resolvent set of A. from the fact that D(A∗ ) = D(A) ⊕ R(A + i)⊥ ⊕ R(A − i)⊥ [8. x) can be thought of as dW (t)/dt. we also prove the absence of eigenvalues for A in this section. (ρ−1 a∇u) for u ∈ D(A). Remark 2·3. W (0) = dt t>0 (2·2) (2·3) where f ∈ H0 . D(A)) which is given by   Z ∞ √ √ iω 1 −iωt W (t) = + √ sin [ (λ)t] − cos [ (λ)t] dP (λ)f. u) > C1 j=1 kuj k20. The operator A is the unique self-adjoint realization of the operator −µ−1 ∇ . For ω > 0 and f ∈ H0 . D(A 2 )) w C(R+ . the problem (2·2)–(2·3) has a unique solution 1 W ∈ C 2 (R+ . By (A1) and (A3). P2 Remark 2·1. dt2 dW (0) = 0. p.346 Bo Zhang with the norm kukH by P2 1 = ( j=1 kuj k2j. e λ − ω2 λ 0 where P (λ) is the spectral family of A (see [2. ν = ρ(2) (a∇u2 ) . Indeed. v) = (h. and D(A) = H0 . we obtain that D(A) = D(A∗ ) and therefore. Thus −1 D(A) = {u = (u1 . Define the operator A by D(A) = {u ∈ H|∃h ∈ H0 . Hence. Define the function W (t) ∈ H0 on R+ by W (t)(x) = W (t. v)0 } ⊂ H0 and Au = h. we claim that A is closed and self-adjoint. and the bilinear functional on H ×H defined ¯ a(u. the derivative of the H0 -valued function W (t). that is. H0 ) w C 1 (R+ . Since A is symmetric. A uniqueness theorem In this section we establish a uniqueness theorem which plays an important role in proving LABP and LAMP. 3. Then Wt (t. a(u. By elliptic regularity estimates together with (A2) and (A3) it follows that D(A) ⊂ H 2 (Ω). Since A is self-adjoint and positive. ∀v ∈ H. (ρ−1 a∇·) with the transmission conditions (1·7)– (1·8). ν on Γ. v ∈ H. Ωj > 0 for u ∈ H and u  0. it easily follows that A is closed. 113]). a(u. the spectral theorem holds for A. As an application. Remark 2·2. A is self-adjoint. A = A∗ . Ωj ) 2 . and the problem (1·1)–(1·4) can be written in the following Hilbert space formulation: find W (t) ∈ H0 such that d2 W + AW = e−iωt f. 17]. x) for all x ∈ Ω. ∇ . Furthermore. p. (ρ−1 a∇u) ∈ L2 } and Au = −µ−1 ∇ .

(3·1)–(3·3) and the transmission conditions (1·7)–(1·8) together with the basic identity ¯ 2R{∇ .1 in [14] by making use of (A2)–(A5). l −1 −1 −2hρ |a∇u| − 2r∂r hρ |xˆ · (a∇u)|2 . k. A priori estimates Let z = λ + iτ with λ > 0 and τ  0. then u = 0 a. ν = u2 . we make the following assumptions: X |aij − δij | = O(|x|−1−δ ) (4·4) i. k. ν. (ρ−1 a∇u)rh[xˆ . lim inf R→∞ (3·2) S(R) If X i. αu1 −1 βρ(1) (a∇u1 ) . = in Ω (4·1) on Γ (4·2) ρ−1 (2) (a∇u2 ) . (3·1) satisfying the transmission conditions (1·7)–(1·8) and the condition Z [|∇u|2 + |u|2 ]ds = 0. where R(z) = (A − z)−1 is the resolvent of A in H0 . k |∂k aij | + |∇ρ| = (|x|−2−δ ) (4·5) . Corollary 3·1. and u = R(z)f. j. (4·3) In this section we shall establish some a priori estimates for u which is necessary in the proof of LABP and LAMP. 4. Let k > 0 and let u ∈ H (Ω(R)) for all R > 0 be a solution of the homogeneous equation −∇ · (ρ−1 a∇u) − k 2 µu = 0.e.Transmission problems with unbounded interfaces 347 2 Theorem 3·1. The theorem can be proved similarly as in the proof of theorem 3. ′ where a = 2 [∂i akl ]ni. l=1 . f ∈ L2 (Ω). on Γ. (3·4) (See [15] for details. To this end. in Ω. The operator A has no eigenvalues. u2 ) ∈ H 2 (Ω) and −∇ · (ρ−1 a∇u) − zµu = µf. l |∂k ∂l aij | + X i. k. Then u = (u1 . Proof. j |aij − δij | = O(|x|−δ ) (3·3) as |x| → ∞. j and |x| as |x| → ∞. (a∇u)]} −1 ¯ a∇u)} − hρ−1 ∇u¯ · (a∇u)(ax)] = ∇ · [2R{rh[xˆ · (a∇u)](ρ +{hρ−1 ∇ · (ax) + ∂r hρ−1 [xˆ · (ax)] + h∇ρ−1 · (ax)}∇u¯ · (a∇u) X alk xi ∂l aij ∂k u∂j u¯ +hρ−1 ∇u¯ · {[a′ (ax)]∇u} − 2hρ−1 R i. X i. From Theorem 3·1 the absence of the eigenvalues for A easily follows and we have the following result. j. where δij is the Kronecker delta.) The proof is complete. j.

and P (r) = r−1 . Lemma 4·2 and . Then √ k∇ukB ∗ + λkukB ∗ + kuk−θ 6 C(kf kθ + kuk1. Make use of (2·1). u and f. λ0 6 λ 6 λ1 and 0 < |τ | 6 τ1 for some positive numbers λ0 .−η 6 C(kf k0. we are able to obtain. and making use of the basic idendity (3·4) with h(r) = r−1 . ¯ ¯ integrating over E(r1 ) Proof. λ1 and τ1 . (4·7) Ω The following results play an important role in establishing the a priori estimate for u. Lemma 4·1.−η ).−η + kuk0. of both sides of the equation thus obtained and making use of (A3). taking the real and imaginary parts Multiplying (4·1) by ξ u. (a∇u)+[∇ . (i) Assume that (4·4) holds. Ω(R1 ) ) (4·8) where C is a constant independent of z. Q(r) = r−3 and N (r) = r−2 when both (4·4) and (4·5) hold. n > 2. u and f. Ω(R1 ) ) (4·9) where θ > 23 .348 Bo Zhang We begin with the following result which can be obtained by using Lax–Milgram theorem and elliptic regularity estimates. (ii) Assume that (4·4) and (4·5) hold. and R1 is a fixed number. u and f such that for any η > 0 kuk2. independent of z. 0 < λ 6 λ1 and 0 < |τ | 6 τ1 . such that Z |τ | (|u|2 + |∇u|2 )dx 6 C(|f |. Then k∇ukB ∗ + kukB ∗ 6 C(kf kB + kuk1. |u|)0 . (4·2) and (4·3). Theorem 4·1. there is a constant C independent of z. that for sufficiently large r1 (> R0 ) Z (n − 3)(n − 1) 2 ¯ ˆ x) ˆ − |x(a∇u)| ˆ 2 ξρ−1 r−1 [∇u(a∇u) x(a ]dx + 2 E(r1 ) Z Z 2 2 ˆ ˆ x)λρq|u| ˆ ξρ−1 r−3 |u|2 dx + ξρ−1 [2|x(a∇u)| + x(a × E(r1 ) S(r1 ) Z Z n−1 ¯ ˆ x)]ds ˆ ξρ−1 r−2 [|u|2 + r∂r (|u|2 )]ds 6 C −∇u(a∇u) x(a + 2 S(r1 ) E(r1 ) ×{r−1−δ [(|z| + P (r))|∇u|2 + (|z| + Q(r))|u|2 ] + |f |(|∇u| + |u|/r)}dx  Z Z −1−δ 2 2 ˆ r (N (r)|u| + |∇u| )ds + 2τ I ξµu¯ x(a∇u)dx. Multiplying (4·1) by ξr−1 {2x . + S(r1 ) (4·10) E(r1 ) where P (r) = Q(r) = N (r) = 1 when only (4·4) holds. Let f ∈ Lθ2 . For any f ∈ L2 (Ω). by integrating by parts together with (A1)–(A5) and (4·2)–(4·4). Lemma 4·2. There is a constant C. we can easily obtain the following lemma. (ax)−1]u}. (4·6) ¯ integrating over Ω. n > 3. Let f ∈ B(Ω).

λ1 ){t−δ/2 [k∇uk2B ∗ + kuk2B ∗ ] R Ω(R) +(|f |. Let f ∈ Lθ2 with θ > 32 . and m = 0 and M (r1 ) = r1−1 when both (4·4) and (4·5) hold. |∇u| + |u|) + kuk21. .R)  Z Z t + ξρ−1 r−3 |u|2 dx − ξρ−1 r−3 |u|2 dx 6 C(λ1 ){(|f |. Then by (2·1) and (4·13) we have for R > t > 1 Z 1 [|∇u|2 + |u|2 ]dx 6 C(λ0 . |∇u| + |u|) + R kuk1. Ω(4t) . where m = 1 and M (r1 ) = 0 when only (4·4) holds. Ω(4t) }. n > 3 and 0 < λ 6 λ1 .Transmission problems with unbounded interfaces 349 the fact that |z| < λ + |τ | to deduce Z (n − 3)(n − 1) 2 ¯ ˆ x) ˆ − |x(a∇u)| ˆ 2 ξρ−1 r−1 [∇u(a∇u) x(a ]dx + 2 E(r1 ) Z Z 2 2 ˆ ˆ x)λρq|u| ˆ ξρ−1 r−3 |u|2 dx + ξρ−1 [2|x(a∇u)| + x(a × E(r1 ) S(r1 ) Z n−1 ¯ ˆ x)]ds ˆ ξρ−1 r−2 [|u|2 + r∂r (|u|2 )]ds −∇u(a∇u) x(a + 2 S(r1 )  −δ/2 6 C(λ1 ) r1 [(λ + P (r1 ))k∇uk2B ∗ + (λ + m)kuk2B ∗ + M (r1 )k < x >−1 uk2B ∗ ]  Z −1−δ 2 2 + r (N (r)|u| + |∇u| )ds + (|f |. (4·12) By integrating by parts the second integral on the left-hand side of (4·12) and making use of (A1)–(A4) it is obtained that   Z Z (n − 3)(n − 1) 1 −2 2 2 2 r |u| dx + [|∇u| + λ|u| ]dx + C R Ω(R) 2R S(R)  Z × (R − t)ξρ−1 r−3 |u|2 dx 6 C(λ1 ) R−δ/2 [(λ + P (R))k∇uk2B ∗ E(R) +(λ + m)kuk2B ∗ + M (R)k < x >−1 uk2B ∗ ]  Z t −3 2 −1 2 r |u| dx + (|f |. Let f ∈ B. Integrating (4·11) multiplied by R−1 with respect to r1 from t to R gives    Z Z n−1 2t 1 −1 2 ¯ ˆ x) ˆ ∇u(a∇u) + λρq|u| dx + ξρ x(a 1− R Ω(t. + R Ω(t. (4·14) which together with Schwarz’s inequality implies the estimate (4·8) for sufficiently large t with R1 = 4t.R) (4·13) (i) Assume that (4·4) holds. 0 < λ0 6 λ 6 λ1 and n > 2.R) r 2R Ω(t. (ii) Assume that (4·4) and (4·5) hold. |∇u| + |u|) R E(t) E(R) +R−δ/2 [(λ + P (R))k∇uk2B ∗ + (λ + m)kuk2B ∗ + M (R)k < x >−1 uk2B ∗ ]}.R)  Z (n − 3)(n − 1) 1 ×ξρ−1 r−2 [|u|2 + r∂r (|u|2 )]dx + ξρ−1 r−2 |u|2 dx 2 R Ω(t. |∇u| + |u|) (4·11) S(r1 ) for r1 > t with sufficiently large t > 1.

(4·21) Proof. which combined with (2·1) implies the required estimate (4·9) with R1 = t1. 0 < λ 6 λ1 and n > 3. Multiply (4·15) by T −1 and integrate the inequality thus obtained with respect to R from t1 to T to yield Z 1 (1 + r2 )−1 |u|2 dx 6 C(λ1 ){t−δ/2 [k∇uk2B ∗ + λkuk2B ∗ + k < x >−1 uk2B ∗ ] T Ω(T ) +(|f |. Let f ∈ B(Ω). Ω(t1) } (4·17) for sufficiently large t. |∇u| + |u|) + kuk21.350 Bo Zhang Then it follows from (4·13) that Z Z 1 r−2 |u|2 ds 6 C(λ1 ) [|∇u|2 + λ|u|2 ]dx + R Ω(R) S(R) ×{t−δ/2 [k∇uk2B ∗ + λkuk2B ∗ + k < x >−1 uk2B ∗ ] + (|f |. Ω(R1 ) + R kf kθ + R Ω(1. R Ω(R) (4·18) if (4·4) holds. and Z Z 1 r−2 |u|2 ds [|∇u|2 + λ|u|2 ]dx + R Ω(R) S(R)   Z 1 −3 2 −δ/2 δ/2−1 2 δ/2 2 6C R r |u| dx (4·19) (λ + R )kuk1. (ii) Assume that (4·4) holds. where t1 = max (4t. Theorem 4·2. From (4·15) and (4·16) it follows on using Schwarz’s inequality that k∇uk2B ∗ + λkuk2B ∗ + k < x >−1 uk2B ∗ 6 C{k < x > f k2B + kuk21. There exists a constant r1 > 0 such that for all R > r1 Z 1 [|∇u|2 + |u|2 ]dx 6 C{R−δ/2 kuk21. t2 ). (i) To prove (4·20) it is enough to show that there is a constant C > 0 such that kuk1. such that √ (4·20) k∇ukB ∗ + λkukB ∗ + kuk−θ 6 Ckf kθ . Ω(t1) } (4·16) for T > t1. Ω(t1) } (4·15) for R > t2 . 0 < λ0 6 λ 6 λ1 and n > 2. f ∈ Lθ2 . Then k∇ukB ∗ + kukB ∗ 6 Ckf kB . Lemma 4·3. The theorem is thus proved. By using a compact argument combined with Lemma 4·3 and Theorem 3·1 the term kuk1. Ω(R1 ) + Rδ/2 kf k2B }. where θ and R1 are the same as in Theorem 4·1. From the inequalities (4·13) and (4·17) and Theorem 4·1 the following result easily follows. |∇u| + |u|) + kuk21. 0 < λ0 6 λ 6 λ1 and 0 < |τ | 6 τ1 . (i) Assume that (4·4) and (4·5) hold. Ω(R1 ) in Theorem 4·1 can be removed from the right-hand side of (4·8) and (4·9). 0 < λ 6 λ1 and 0 < |τ | 6 τ1 . u and f. independent of z. n > 2. Ω(R1 ) 6 Ckf kθ (4·22) . n > 3. Then there is a constant C > 0. Let f ∈ Lθ2 with θ > 23 . f ∈ B. R) if (4·4) and (4·5) hold.

(4·28) 2 (Ω) with η > 23 . R) (4·29) for R > r1. v satisfies (4·1)–(4·3) with f = 0. Case 1. that is. which combined with Rellich’s selection theorem and standard elliptic estimates asserts that there is a subsequence of {vm }. 1 which. (4·25) Next. In this case. and from (4·24) √ k∇vkB ∗ + λkvkB ∗ + kvk−θ 6 C.T ) For sufficiently large t we have  Z Z −3 2 −1 1+ r |u| dx 6 Ct Ω(t. τ1 ]. z = 0. There are three cases to be distinguished. such that vm 2 strongly converges to some function v in H−θ (Ω) and |∇vm | strongly converges to −η 1 |∇v| in L2 (Ω) with η > 2 as m → ∞. on letting T → ∞. This contradicts (4·23). by Theorem 3·1.e. and the required estimate (4·20) then follows. (ρ−1 a∇v) = 0. t) 3 r −3 2  |v| dx . T ) Ω(1. Thus v = 0 a. ν in Ω = v2 . and on passing the limit as m → ∞ in the inequality thus obtained it follows that v satisfies the condition (3·2) of Theorem 3·1. Ω(1. If (4·22) were false. vm satisfies (4·19) with λ replaced by λm and f by fm . Case 3. = (4·26) on Γ (4·27) ρ−1 (2) (a∇v2 ) . ∇ . Multiply (4·29) by R−1 and integrate the inequality thus obtained with respect to R from t to T with t > r1 to get   Z Z −3 2 −1 −3 2 1+ r |u| dx 6 Ct r |v| dx .Transmission problems with unbounded interfaces 351 for all λ ∈ (0.e. λ > 0 and τ = 0.e. as m → ∞. {vm } and {fm } such that kvm k1. where vm = R(zm )fm . implies < x >− 2 u ∈ L2 (Ω). In this case (4·7) can be used to yield that v = 0 a. λ1 ] and |τ | ∈ (0. where < x >= (1 + |x|2 ) 2 . then there would be sequences {zm } with zm = λm + iτm . denoted by {vm } again. ν. kfm kθ → 0 (4·23) and zm → z = λ + iτ with 0 6 λ 6 λ1 and 0 6 |τ | 6 τ1 . on Γ. we prove that v = 0 a. This together with (4·29) asserts that |∇v| ∈ L2 (Ω). T ) Ω(t. λ = τ = 0. v2 ) ∈ H−η   Z Z Z 1 |∇v|2 dx + r−2 |v|2 ds 6 CR−1 1 + r−3 |v|2 dx R Ω(R) S(R) Ω(1. Ω(R1 ) = 1. λ > 0 and |τ | > 0. Case 2. and by Lemma 4·3 we have where v = (v1 . By Lemma 4·3. By Theorem 4·1. √ (4·24) k∇vm kB ∗ + λm kvm kB ∗ + kvm k−θ 6 C(kfm kθ + 1) < ∞. αv1 −1 βρ(1) (a∇v1 ) . . Clearly.

Lemma 5·2. vj = Cj a. However (4·32) asserts that Cj = 0 (j = 1. by (4·11). in Ω. (A1)–(A3) and Schwarz’s inequality it can be seen that vm satisfies   Z Z −2 2 −δ/2 −1 2 δ/2 2 r |vm | ds 6 C R (λm + R ) + (4·30) |∇vm | ds + R kfm kθ S(R) S(R) for R > t with sufficiently large t > 1. in this case only Cases 1 and 2 above should be distinguished since z  0. S(R) (4·31) S(R) which in conjunction with the fact that |∇v| ∈ L2 (Ω) guarantees Z [r−1 |v|2 + r|∇v|2 ]ds = 0.352 Bo Zhang On the other hand. z = λ + iτ. Ω This combined with (A1)–(A3) implies ∇vj = 0 a. By combining Lemma 5·1 with Lemma 4·2 and Theorem 4·2 (ii) the following result can be obtained. and let n > 2. (4·32) and Multiply (4·26) by ξ v. Theorem 4·2 (ii) together with (2·1) gives kuk1. in Ωj . and 0 < |τ | 6 τ1 . (5·3) On the other hand. Then kuk1. α3 (5·1) where p = (α3 − α1 )/(α3 − α2 ) and q = (α3 − α1 )/(α2 − α1 ). and the estimate (4·20) is thus proved. (5·4) . which will be established in this section.−η 6 C|τ |−(1−η)/2 kf kB .e. Theorem 4·2 (ii) and Schwarz’s inequality we have kuk21. 2). lim inf R→∞ (4·32) S(R) ¯ integrate over Ω and make use of (4·27)–(4·28). in Ωj . |u|)0 6 C|τ |−1 kf k2B . Proof. Assume that (4·4) holds.e. Lemma 5·1. Letting m → ∞ in (4·30) leads to   Z Z −2 2 −1−δ/2 2 r |v| ds 6 C R + |∇v| ds . 2). (4·23). where Cj is a constant (j = 1.e. (ii) The estimate (4·21) can be proved similarly. α2 6 kvk1/p m. To this end. that is. α1 kvkm. 5. 0 6 |τ |−1 (|f |. (a∇v)dx = 0. Let u = R(z)f with f ∈ B(Ω). More a priori estimates In order to prove the limiting absorption and amplitude principles we need more a priori estimates for u = R(z)f with Iz  0. (2·1). (4·17). 0 < λ0 6 λ 6 λ1 . The proof is complete.−1 6 Ckf kB . Let α1 < α2 < α3 and let v ∈ Hαm3 (Ω) with m > 0 an integer. By Lemma 4·2. In fact. partial integration to obtain Z ξρ−1 ∇v¯ . Then 1/q kvkm. Hence v = 0 a. (5·2) where 0 6 η 6 21 . we introduce the following interpolation inequality for which a proof can be found in [3].

where j = 1. −ρ−1 (5·6) i. u(j) and f (j) . (a∇v) + ρ−1 ∇u . integrate it over E(R) with R > R0 and make use of (A5). Lemma 5·3. Let u(j) = R(z)f (j) with f (j) ∈ L12 (Ω). 0 < λ0 6 λ 6 λ1 . (1·10) and (4·4) it can be obtained that . l By using partial integration and Cauchy–Schwarz inequality together with (A2).353 Transmission problems with unbounded interfaces By using (5·3) and (5·4) in conjunction with Lemma 5·1 the required estimate (5·2) follows on taking α1 = −1. δ0 are positive constants independent of z. {[x . (a∇u(3−j) )]. (5·5) where C. Assume that (4·4) holds. {[a′ (ax)]∇v} X alk xi ∂l aij [∂k v∂j u + ∂k u∂j v] − 2ρ−1 (a∇u)(a∇v). 2. and let n > 2. It follows from the definition of R(z) that u(j) ∈ H 2 (Ω) satisfies (4·1)–(4·3) with u. 2 and 21 < δ0 < 1. (ax)}∇u . u¯ (2) )| 6 C|τ |−δ0 kf (1) k1 kf (2) k1 . Proof. Multiply (4·1) with u(j) and f (j) replacing u and f. Then |(σµu(1) . (A4). k. z = λ + iτ. by σ[x . (a∇v)(ax)} +{ρ−1 ∇ . and 0 < |τ | 6 τ1 . σ = (αβ)2−j in Ωj with j = 1. (ρ−1 a∇u)[x . (ax) + ∇ρ−1 . j. α2 = −η and α3 = 0. the transmission conditions (4·2)–(4·3) and the basic identity ∇ . f replaced by u(j) . (a∇u)](ρ−1 a∇v) − ρ−1 ∇u . (a∇v)] + ∇ . (ρ−1 a∇v)[x . f (j) respectively. (a∇u)] = ∇ . The proof is complete. respectively. (a∇v)](ρ−1 a∇u) + [x .

Z .

.

Z .

.

.

.

.

−1 (1) (2) (1) (2) (1) (2) .

.

.

2|z| .

σµu u dx.

6 (n − 2) .

σ[ρ ∇u · (a∇u ) − zµu u ]dx.

.

that . f replaced with u(j) . integrating by parts combined with (4·1)–(4·3) with u.−δ/2 ku k1. on using Cauchy– Schwarz inequality and applying standard elliptic estimates to the integrals over S(R) together with (1·10). (5·7) On the other hand.S(R) + 2 X j=1 (j) kf k1 k∇u (3−j) ) k . where j = 1. From (5·7) and (5·8) we obtain. implies Z Z Z −1 (1) (2) (1) (2) (5·8) σρ ∇u · (a∇u )dx = z σµu u dx + σµf (2) u(1) dx Ω Ω Ω for sufficiently large R > R0 . Theorem 4·2 (ii) and (2·1). f (j) respectively. 2. + C E(R) E(R) ( (1) (2) (1) (2) × ku k1.−δ/2 + Rku k1.S(R) ku k1.

Z .

.

.

|z| .

.

σµu(1) u(2) dx.

.

−δ/2 ku(2) k1. 6 CR Ω ) ( 2 X × ku(1) k1. This completes the proof. . −δ/2 + kf (1) k1 kf (2) k1 + kf (j) k1 k∇u(3−j) k . Now combining (5·9) with Lemma 5·2 and the fact that 0 < δ < 1 gives the required estimate (5·5) with δ0 = 1 − δ/2. (5·9) j=1 where CR is a positive constant depending on R.

and 0 < |τ | 6 τ1 . By Lemma 5·3 . where σ is as defined in Lemma 5·3. z = λ + iτ. −1 where 1 2 < δ0 < 1 is the same as in Lemma 5·3. Proof. 0 < λ0 6 λ 6 λ1 . and let n > 2. g¯ = (σµR(z)f. Then we have du (5·10) 6 C|τ |−δ0 kf k1 . By the first resolvent identity we have dR(z) = R2 (z). Assume that (4·4) holds.354 Bo Zhang Theorem 5·1. R(z)g) dz since A is self-adjoint in H0 . Let u = R(z)f with f ∈ L12 (Ω). dz (5·11) Hence for any g ∈ L12 (Ω) ⊂ H0 . dz 2.   dR(z) σµ f.

 .

.

.

.

σµ dR(z) f. g¯ .

6 C|τ |−δ0 kf k1 kgk1 . .

.

the limit R(λ ± i0) are local in the norm of H−1 Hoelder continuous in λ in the strong topology of L(L12 (Ω). Proof. kR(λ ± i0) − R(λ′ ± i0)k1→−1 6 C|λ − λ′ |ε (6·2) for any λ and λ′ in an arbitrarily fixed compact interval of (0. Theorem 6·1. Then for each λ > 0 there exist the operators R(λ ± i0) in L(L12 (Ω). dz Taking g¯ =< x >−2 (dR(z)/dz)f in the above inequality and using (A2) we obtain that dR(z) −δ0 (5·12) dz f 6 C|τ | kf k1 . and the theorem is thus proved. Theorem 4·2 (ii) and (5·12). ∞). L−1 2 (Ω)) such that lim R(λ ± iτ )f = R(λ ± i0)f τ →0+ (6·1) 2 (Ω) for all f ∈ L12 (Ω). The limiting absorption and amplitude principles Theorems 4·2 and 5·1 will be used in this section to justify the validity of the limiting absorption and amplitude principles for the operator A. L−1 2 (Ω)). Y ) is the Banach space of all bounded linear operators from X into Y. k · kη→θ denotes the operator norm when considered as an operator from Lη2 (Ω) into Lθ2 (Ω). ε do not depend on λ and λ′ . Let n > 2. 6. Therefore (5·10) follows from Lemma 4·1. where L(X. Theorem 5·1 implies that for . Furthermore. (i) The existence of the operators R(λ±i0) and the limits (6·1) easily follows from Theorem 5·1. that is. −1 Next. Then by (5·11) v = R(z)u. let v = du/dz. Assume that (4·4) holds. 0 < ε < 1 and C. (ii) It is sufficient to prove (6·2) for |λ − λ′ | small.

We prove the theorem only for the ‘+’ case. (6·7) which combined with Rellich’s selection theorem and standard elliptic estimates implies that there is a subsequence of {um }. On the other hand we find from (6·1) that the sequence {um } itself converges to R(λ + i0)f in L−1 2 (Ω). Assume that (4·4) holds. As a result we have u = R(λ + i0)f and by (6·7) k∇ukB ∗ + kukB ∗ + kR(λ + i0)f k2. (6·4) On taking τ = |λ−λ′ |. On the other hand. it is easy to see that (6·5) is true for f ∈ B(Ω). The extension will be denoted by R(λ + i0) again. L−1 2 ) obtained in Theorem 6·1. Theorem 4·2 (ii) and Lemma 4·1. denoted by {um } again.−η 6 Ckf kB . by the use of the denseness of L12 (Ω) in B(Ω). From Theorems 6·2 and 4·2 (i) the following asymptotic behavior easily follows for the resolvent R(λ ± i0) as λ → 0+. we can prove that the limits (6·1) also exist in 2 the norm of H−η (Ω) with η > 12 for all f ∈ B(Ω). (6·8). we conclude that (6·5) is true for f ∈ L12 (Ω). Then by Theorem 4·2 (ii) and Lemma 4·1 k∇um kB ∗ + kum kB ∗ + kum k2. It is enough to assume that 21 < η 6 1. Let {τm } be an arbitrary positive sequence such that τm → 0 as m → ∞. Moreover. . Theorem 6·2 (Limiting absorption principle). B ∗ (Ω)) such that lim R(λ ± iτ )f = R(λ ± i0)f τ →0+ 2 (Ω) with η > in the norm of H−η 1 2 (6·5) for all f ∈ B(Ω). we have k∇u± kB ∗ + ku± kB ∗ + ku± k2. which plays an important role in proving the limiting amplitude principle. from Theorem 5·1 we have for any 0 < τ 6 1 kR(λ ± iτ ) − R(λ′ ± iτ )k1→−1 6 Cτ −δ0 |λ − λ′ |. The ‘−’ case can be treated similarly. Making use of Theorems 6·1 and 4·2 together with the same argument as used by Saito in proving Theorem 4·3 in [12]. Let R(λ + i0) be the operator in L(L12 .Transmission problems with unbounded interfaces 355 kR(λ ± iτ ) − R(λ ± i0)k1→−1 6 Cτ 1−δ0 (6·3) any 0 < τ 6 1 uniformly in λ in the required region.−η 6 Ckf kB . and let um = R(λ + iτm )f. Proof. (6·8) Since {τm } is an arbitrary sequence.−η 6 Ckf kB . The estimate (6·6) follows from (6·8). and we first 1 prove (6·5) for f ∈ L2 (Ω). (6·6) where u± = R(λ ± i0)f. Let n > 2. The theorem is proved. (6·3) and (6·4) yield (6·2) with ε = 1−δ0 . such that um 2 strongly converges to a function u in H−η (Ω). The proof is complete. Then. Then for each λ > 0 there exist the operators R(λ ± i0) in L(B(Ω). From (6·8) and the denseness of L12 (Ω) in B(Ω) that R(λ + i0) can be uniquely extended 2 to a bounded operator from B(Ω) into H−η (Ω).

O. Roach and B. 149–161. Math. Soc. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] REFERENCES C. Acknowledgements. Edinburgh 121A (1992). 183–208. Roach and B. Let n > 3 and let η > 32 . 136 (1989). Camb. see also [2. Proc. J. A remark on the limiting absorption principle for the reduced wave equation with two unbounded media. The limiting amplitude principle for the wave propagation problem with two unbounded media. Zhang. Then kR(λ ± i0)kη→−η = O(1) (6·9) as λ → 0 + . Tamura. Soc. Initial boundary value problems in mathematical physics (John Wiley. Then W (t) behaves like W = exp (−iωt)R(ω 2 + i0)f + o(1) strongly in L−η 2 (Ω) (6·10) as t → +∞. Phil. 345–362. Differential Equations 93 (1991). The limiting absorption principle for the acoustic wave operators in two unbounded media. Proc. Differential Equations 19 (1975). Leis. A436 (1992). Math. Brunel University. 18 (1994). Let n > 3 and let W (t) = W (t. 545–558. G. 217–236. Assume that (4·4) and (4·5) hold. Math. F. Kadowaki. Eidus. Eidus. Bloom. Phil. M.356 Bo Zhang Theorem 6·3. A transmission problem for the reduced wave equation in inhomogeneous media with an infinite interface. M. Limiting amplitude principle for the acoustic propagator in perturbed stratified fluids. Zhang. Camb. Tsukuba J. 97–167. D. Roy. Proc. G. 121–140. K. Soc. H. On the limiting absorption and amplitude principles for transmission problems with unbounded interfaces. 17 (1993). 29–38. D. Math. Assume that (4·4) and (4·5) hold. J. This work was partly done when the author was in the Department of Mathematics at Keele University. Pacific J. F. 11. 1986). 115 (1994). 112 (1992). The principle of limiting amplitude. Phys. Eidus. Russian Math. A rate of approach to the steady state of solutions of second-order hyperbolic equations. Tamura. Kadowaki. Tsukuba J. . Japan 41 (1989). By means of the local Hoelder continuity (6·2) in λ(> 0) and the low frequency behaviour (6·9) for the resolvent R(λ ± i0) the limiting amplitude principle can be easily shown on following exactly the same argument as in the proof of theorem 1·1 in [7] (see [7] for details. On Sommerfeld radiation conditions for the diffraction problem with two unbounded media. Math. UK. Resolvent estimates at low frequencies and limiting amplitude principle for acoustic propagators. Commun. Saito. D. 107 (1986). 549–575. Y. B. Department of Mathematics and Statistics Technical Report (1995). The limiting amplitude principle for the acoustic wave operators in two unbounded media. The limiting absorption and amplitude principles for the diffraction problem with two unbounded media. x) be the solution to (1·1)–(1·4) with f ∈ Lη2 (Ω). Soc. Math. 207–223. F. Theorem 6·4. η > 32 . R. Kikuchi and H. 260–282. Proc. Math. B. Lond. 13]). On transmission problems for wave propagation in two locally perturbed halfspaces. Roy. Soc. 296–329. Asymptotically periodic solutions of the wave equation in the Hilbert space. Roach and B. Zhang. 175–191. G. Asymptotic Anal. 8 (1994). Zhang. Surveys 24 (1969). J. Zhang.