EXISTENCE, UNIQUENESS, AND REGULARITY RESULTS FOR ...
SIAM J. MATH. ANAL. Vol. 37, No. 2, pp. 651–672
c 2005 Society for Industrial and Applied Mathematics
EXISTENCE, UNIQUENESS, AND REGULARITY RESULTS FOR PIEZOELECTRIC SYSTEMS∗ D. MERCIER† AND S. NICAISE† Abstract. We investigate the time-harmonic piezoelectric system (a system coupling the elasticity system with the full Maxwell’s equations) in polyhedral domains of the space. Existence and uniqueness results of weak solutions are proved in different cases. We describe the corner and edge singularities of that system and deduce some regularity results. Key words. elasticity system, Maxwell’s system, singularities AMS subject classifications. 35J25, 35Q60 DOI. 10.1137/040617728
1. Introduction. Smart structures made of piezoelectric and/or piezomagnetic materials are gaining attention in applications since they are able to transform the energy from one type to another (magnetic, electric, and mechanical), allowing them to be used as sensors and/or actuators. Commonly used piezoelectric materials are ceramics and quartz. The mathematical model of this system starts to be well established [2, 8, 14, 24, 26] and corresponds to a coupling between the elasticity system and Maxwell’s equations (see below). A full mathematical analysis is not yet done, except in some particular cases [13, 19]. Namely, in these two works the electric field E is assumed to be curl free, i.e., E = ∇ϕ, where ϕ is an electric potential and a twodimensional reduction is made. In [13], existence and uniqueness results in smooth domains are obtained using integral equations, while in [19] a variational formulation in polygonal domains is given and two-dimensional singularities are briefly described. On the other hand, there exists an extensive list of papers from mechanics literature describing singularities of some particular piezoelectric materials with a plane crack [25, 27, 30] or along wedges [29]. But to our knowledge, an exact description of corner/edge singularities of the general piezoelectric system in three-dimensional polyhedral domains is not yet obtained. Such a description is very important since piezoelectric ceramics are very brittle, and therefore their fracture behavior must be understood. The knowledge of such singularities also has numerical implications, such as convergence speed. This paper has, therefore, the following goals: We present a general piezoelectric system, which includes standard models of ceramics like the P ZT or the BaT iO3 . We further develop some variational formulations which are the natural ones because they lead to solutions in the energy spaces (here called weak solutions). We prove existence and uniqueness results of weak solutions of the time-harmonic system in two different cases: the case when the magnetic permeability matrix is positive definite (BaT iO3 ) and the case when the magnetic permeability matrix is zero (P ZT ). In that second case, we even give two different formulations and show that generically they give rise to the same solutions. Moreover, we describe the corner and edge singularities of our ∗ Received
by the editors October 27, 2004; accepted for publication (in revised form) February 22, 2005; published electronically November 9, 2005. http://www.siam.org/journals/sima/37-2/61772.html † Universit´ e de Valenciennes et du Hainaut Cambr´esis, MACS, Institut des Sciences et Techniques de Valenciennes, F-59313 Valenciennes Cedex 9, France ([email protected], Serge. [email protected]). 651
652
D. MERCIER AND S. NICAISE
general system and deduce some regularity results. Some edge singular exponents are briefly described; more examples will be given in a forthcoming paper [20], where a two-dimensional model and fracture criteria will be considered. The analysis of more sophisticated models, like coupling between piezoelectric and magnetostrictive materials [26] or piezoelectric and purely elastic materials [29, 9], will be investigated in the future. The paper is organized as follows. Section 2 introduces the nonstationary model problem and its time-harmonic version. Existence and uniqueness results are given in section 3 when the magnetic permeability matrix is positive definite. Section 4 is devoted to existence and uniqueness results in the case of a zero magnetic permeability matrix. There we consider two different formulations: for the first one (called the E-formulation) the magnetic field H is eliminated, while for the second one (the Hformulation) the electric field E is eliminated. For the latter formulation, like for the eddy current problem, Gauge conditions are necessary. We further show their generic equivalence. After a short description of corner and edge singularities of some useful elliptic systems in section 5, we obtain the corner and edge singularities of our system in section 6. Regularity results are deduced in section 7. Some edge singular exponents are finally presented in section 8. 2. Setting of the problem. Let Ω be a bounded domain of R3 with a Lipschitz boundary Γ. For the sake of simplicity, we suppose that Γ is piecewise plane and connected and that Ω is simply connected. In this domain, we consider the following nonstationary piezoelectric system of constitutive equations [2, 8, 14, 24]: (2.1) (2.2) (2.3)
σij = aijkl γkl (u) − ekij Ek ∀i, j = 1, 2, 3, Di = ij Ej + eikl γkl (u) ∀i = 1, 2, 3, Bi = μij Hj ∀i = 1, 2, 3.
The equations of equilibrium are (2.4)
∂t2 ui = ∂j σji + fi ∀i = 1, 2, 3
for the elastic displacement and ∂t D + J = curl H, ∂t B = − curl E for the electric/magnetic fields (Maxwell’s equations), where f is the body force density and J is the vector current density function. As usual curl H = (∂2 H3 − ∂3 H2 , ∂3 H1 − ∂1 H3 , ∂1 H2 − ∂2 H1 ) , when H = (H1 , H2 , H3 ) . This system models the coupling between Maxwell’s system and the elastic one [2, 8, 14, 24], in which E(x, t), H(x, t) are the electric and magnetic fields at the point x ∈ Ω at time t, u(x, t) is the displacement field at the point x ∈ Ω at time t, and (γij (u))3i,j=1 is the strain tensor given by 1 γij (u) = 2
∂ui ∂uj + ∂xj ∂xi
.
Here (σij )3i,j=1 , D = (D1 , D2 , D3 ) , and B = (B1 , B2 , B3 ) are the stress tensor, electric displacement, and magnetic induction, respectively. , μ are the electric permittivity and magnetic permeability, respectively, and are supposed to be real, symmetric 3 × 3 matrices. In what follows the matrix is supposed to be positive definite,
653
RESULTS FOR PIEZOELECTRIC SYSTEMS
while μ is only supposed to be nonnegative. The elasticity tensor (aijkl )i,j,k,l=1,2,3 is made of constant entries such that aijkl = ajikl = aklij , and satisfies the ellipticity condition (2.5)
aijkl γij γkl ≥ αγij γij ,
for every symmetric tensor (γij ) and some α > 0. The piezoelectric tensor ekij is also made of constant entries such that ekij = ekji . The system is completed with the Dirichlet boundary conditions for the displacement field, (2.6)
u = 0 on Γ,
and those of a perfect conductor, (2.7)
E × n = 0 on Γ.
As usual n is the exterior unit normal vector along Γ. In order to write the above problem in a more compact form, the strain and stress tensors are expressed as 6 × 1 vectors, namely
γ(u) = (γ11 (u), γ22 (u), γ33 (u), 2γ23 (u), 2γ31 (u), 2γ12 (u)) , σ = (σ11 , σ22 , σ33 , σ23 , σ31 , σ12 ) . With this notation, the constitutive equations (2.1)–(2.3) may be equivalently written as ⎛ ⎞ ⎛ ⎞ σ γ(u) ⎝ D ⎠ = M ⎝ E ⎠, (2.8) B H where M is a 12 × 12 matrix given by ⎛
C M =⎝ e 0
−e 0
⎞ 0 0 ⎠, μ
where C is a 6 × 6 symmetric matrix depending on the elasticity tensor, e is a 3 × 6 matrix depending on the piezoelectric tensor, and , μ are as described above. Note that the ellipticity assumption (2.5) is equivalent to the fact that C is a positive definite matrix. For a monoclinic material with poling direction in the x3 -axis, the material constant matrices are expressed by (see, for instance, [26]) ⎞ ⎛ c11 c12 c13 0 0 c16 ⎜ c12 c22 c23 0 ⎛ ⎞ 0 c26 ⎟ ⎟ ⎜ 0 0 0 e14 e15 0 ⎟ ⎜ c13 c23 c33 0 0 c36 ⎟ 0 0 e24 e25 0 ⎠ , , e=⎝ 0 C=⎜ ⎜ 0 0 0 c44 c45 0 ⎟ ⎟ ⎜ e e e 0 0 e36 31 32 33 ⎝ 0 0 0 c45 c55 0 ⎠ 0 c66 c16 c26 c36 0 ⎞ ⎛ ⎞ ⎛ μ11 μ12 0 11 12 0 0 ⎠. = ⎝ 12 22 0 ⎠ , μ = ⎝ μ12 μ22 0 0 33 0 0 μ33
654
D. MERCIER AND S. NICAISE
For the orthotropic piezoelectric P ZT -4, the material coefficients are given by (cij in 109 N/m2 , eij in C/m2 , ij in 10−9 C 2 /N m2 ): 12 c11 = c22 = 23.8 c33 = 10.6 c44 = 2.15 c55 = 4.4 c66 = c11 −c = 6.43 2 c12 = 3.98 c13 = 2.19 c23 = 1.92 11 = 22 = 0.110625 33 = 0.106023 e32 = −0.14 e33 = −0.28 e24 = e15 = −0.01. e31 = −0.13
Here and below, nongiven coefficients are equal to zero. For the piezoelectric BaT iO3 , the material coefficients are given by (cij in 109 N/m2 , eij in C/m2 , ij in 10−9 C 2 /N m2 , μij in 10−6 N s2 /C 2 ): c11 = c22 = 166 c33 = 162 c44 = c55 = 43 c12 = 77 c13 = c23 = 78 11 = 22 = 11.2 e33 = −18.6 e24 = e15 = 11.6 e31 = e32 = −4.4
12 c66 = c11 −c = 44.5 2 33 = 12.6 μ11 = μ22 = 5 μ33 = 10
With the above notation, the equation of equilibrium (2.4) is also equivalent to ∂t2 u = Divσ + f, where Div is the operator-valued matrix defined by ⎛
∂1 Div = ⎝ 0 0
0 ∂2 0
0 0 ∂3
0 ∂3 ∂2
∂3 0 ∂1
⎞ ∂2 ∂1 ⎠ . 0
Assuming that u, E, H are of the form u(x, t) = e−iωt u(x), E(x, t) = e−iωt E(x), H(x, t) = e−iωt H(x), for some real constant ω (the data being of the same form), the above system is reduced to the time-harmonic piezoelectric system in Ω consisting of the constitutive equation (2.8), the time-harmonic equilibrium equation (2.9)
Divσ + ω 2 u = −f in Ω,
and the time-harmonic Maxwell’s equations (2.10)
curl H + iωD = J in Ω,
(2.11)
curl E − iωμH = 0 in Ω.
In the whole paper we consider the nonstationary case; namely, we assume that ω > 0. In other words we assume that the variation in time of u, E, and H is periodic in time with a frequency equal to 2π/ω. The stationary case ω = 0 requires the use of Gauss’ law divD = ρ, where ρ is the charge density function. This case is treated as in section 4.1 below and is then left to the reader.
RESULTS FOR PIEZOELECTRIC SYSTEMS
655
3. Existence and uniqueness results when μ is positive definite. Replacing D by its expression (2.2) in the first Maxwell equation (2.10), we get curl H + iω(E + eγ(u)) = J in Ω. This identity is clearly equivalent to iωE = J − curl H − iωeγ(u) in Ω.
(3.1)
With this identity, the second Maxwell equation (2.11) is then equivalent to curl(−1 (curl H + iωeγ(u))) − ω 2 μH = curl(−1 J) in Ω.
(3.2)
In the same way using the constitutive equation (2.8) in the equation of motion (2.9), we obtain
Div Cγ(u) − e E + ω 2 u = −f in Ω. (3.3) Using the identity (3.1), we arrive at 1 −1 −1 (3.4) Div Cγ(u) + e eγ(u) + e (curl H) iω
T −1 1 2 +ω u = −f + Div e J in Ω. iω The two equations (3.2) and (3.4) constitute the system of partial differential equations that we will study and whose unknowns are u and H. Due to the boundary conditions (2.6) and (2.7), this system is completed with (2.6) and −1 (curl H + iωeγ(u)) × n = J × n, (μH) · n = 0 on Γ.
(3.5)
The weak formulation of the above problem is obtained in the following way: We introduce the space (see [4, 5]) XT (Ω, μ) = {v ∈ L2 (Ω)3 : div(μv) ∈ L2 (Ω), curl v ∈ L2 (Ω)3 and (μv) · n = 0 on Γ} equipped with its natural norm. Then we multiply the system (3.2) by a test function ¯ ∈ XT (Ω, μ), integrate the result in Ω, and integrate by parts to get (assuming that H u and H are regular enough)
−1 ¯ − ω 2 μH · H ¯ dx (curl H + iωeγ(u)) · curl H Ω −1 ¯ ¯ ds ∀H ∈ XT (Ω, μ). = J · curl H dx + (J × n) · H Ω
Γ
Since from the second Maxwell equation (2.11) div(μH) = 0 in Ω, the above identity is equivalent to
−1 ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ dx (3.6) (curl H + iωe · γ(u)) · curl H Ω ¯ dx + (J × n) · H ¯ ds ∀H ∈ XT (Ω, μ). = −1 J · curl H Ω
Γ
This last equation may be called the regularized formulation of the system (3.2).
656
D. MERCIER AND S. NICAISE
Similarly multiplying (3.4) by v¯ ∈ H01 (Ω)3 , integrating the result in Ω, and using Green’s formula (Divγ) · v¯ dx = − γ · γ(¯ v ) dx, Ω
Ω
obtained by componentwise integration by parts in Ω, we arrive at 1 −1 −1 2 Cγ(u) + e eγ(u) + e curl H · γ(¯ (3.7) v ) − ω u · v¯ dx iω Ω 1 = f · v¯ − eT −1 J · γ(¯ v ) dx ∀v ∈ H01 (Ω)3 . iω Ω As these two identities are coupled, multiplying the second one by iω and summing the result we arrive at the following problem: ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ −1 (curl H + iωeγ(u)) · curl H Ω 1 +iω Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − iω 3 u · v¯ dx = F (v, H ), iω where we have set (3.8)
F (v, H ) =
−1
¯ − eT −1 J · γ(¯ J · curl H v ) + iωf · v¯ dx +
Ω
¯ ds. (J × n) · H
Γ
In order to get a well-posed problem we set uω = iωu. Then we see that the above problem is equivalent to finding a solution (uω , H) ∈ V of (3.9)
a((uω , H), (v, H )) = F (v, H ) ∀(v, H ) ∈ V,
where we set V = H01 (Ω)3 × XT (Ω, μ),
−1 ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ a((u, H), (v, H )) = (curl H + eγ(u)) · curl H Ω
+ Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − ω 2 u · v¯ dx. In summary we have shown the following lemma. Lemma 3.1. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), then (iωu, H) is a solution of (3.9). We now remark that the bilinear form a may be equivalently written a ((u, H), (v, H )) ¯ curl H curl H ¯ − ω 2 u · v¯ dx, ¯ ) − ω 2 μH · H = A · + div(μH)div(μH γ(u) γ(¯ v) Ω where A is the 9 × 9 symmetric matrix defined by −1 −1 e A= . e −1 C + e −1 e
657
RESULTS FOR PIEZOELECTRIC SYSTEMS
One easily checks that this matrix is positive definite for the material coefficients of the P ZT -4 and BaT iO3 , for instance. In fact this is always true if and C are positive definite (always true in our setting), independently of the piezoelectric coefficients. Lemma 3.2. If and C are positive definite matrices, then the matrix A defined above is also positive definite. Proof. We only need to show that X X X A · ≥ 0, ∀ ∈ R9 , γ γ γ X X X A · =0⇒ = 0. γ γ γ By the definition of A we see that X X A · = X −1 X + 2X −1 eγ + γ Cγ + γ e −1 eγ. γ γ ˜ = −1/2 X and Y˜ = −1/2 eγ (note that both vectors are in R3 ), we see that Setting X X X ˜ X ˜ + 2X ˜ Y˜ + γ Cγ + Y˜ Y˜ A · =X γ γ ˜ + Y˜ 22 + γ Cγ, = X where · 2 clearly means the Euclidean norm of R3 . This identity directly implies the first assertion by the positive definitiveness of C. For the second assertion, if X X A · = 0, γ γ then the above identity and again the positive definitiveness of C imply that ˜ + Y˜ 2 = γ Cγ = 0. X 2 Therefore γ = 0 and, consequently, Y˜ = 0 in view of its definition (independently of ˜ = 0 and, by the positive definitiveness of , we conclude e). We then obtain that X that X = 0. This lemma allows us to show that problem (3.9) enters within the framework of the Fredholm alternative. Indeed, we shall prove the following lemma. Lemma 3.3. There exists a discrete set S such that for 1 + ω 2 ∈ S, the problem (3.9) has a unique solution for any F ∈ V . Proof. Introduce the sesquilinear form 2 ¯ μH · H dx + u · v¯ dx . b((u, H), (v, H )) := a((u, H), (v, H )) + (1 + ω ) Ω
Ω
Then the above considerations yield b ((u, H), (v, H )) ¯ curl H curl H ¯ ¯ = A · + div(μH)div(μH ) + μH · H + u · v¯ dx. γ(u) γ(¯ v) Ω
658
D. MERCIER AND S. NICAISE
It is coercive on V, since by Lemma 3.2 we have
b((u, H), (u, H)) | curl H|2 + |γ(u)|2 + |div(μH)|2 + |H|2 + |u|2 dx. Ω
Therefore by Korn’s inequality we get b((u, H), (u, H)) u21,Ω + H2XT (Ω,μ) . Since the space XT (Ω, μ) is compactly embedded into L2 (Ω)3 [28], and since by the Rellich–Kondrasov theorem H01 (Ω) is compactly embedded into L2 (Ω), we deduce that V is compactly embedded into H := L2 (Ω)3 × L2 (Ω)3 . These facts imply that the Friedrichs extension B from H into H induced by the triple (V, H, b) is invertible with a compact inverse. Now denote by A the Friedrichs extension of the triple (V, H, a). The relation between a and b implies that A = B − (1 + ω 2 )Iμ , where the operator Iμ is defined by Iμ (u, H) = (u, μH) and is clearly continuous from H into H. Using this operator we may write (1 + ω 2 )−1 B −1 A = (1 + ω 2 )−1 I − B −1 Iμ , where I is the identity operator from H into H. Since B −1 Iμ is a compact operator, it has a discrete spectrum with positive eigenvalues. If we denote by S the set of the inverses of these eigenvalues, then for (1 + ω 2 ) ∈ S, the operator (1 + ω 2 )−1 I − B −1 Iμ is invertible and consequently A is as well. Remark 3.4. Note that in the case μ = 0, problem (3.9) does not enter in the Fredholm framework since XT (Ω, 0) is reduced to H(curl, Ω) and is no more compactly embedded into L2 (Ω)3 . This drawback will be avoided by introducing Gauge conditions; see section 4.2. Let us now show that the converse of Lemma 3.1 holds under some conditions on ω (compare with Theorem 0.1 of [4]). eu Theorem 3.5. Assume that ω 2 is not an eigenvalue of −ΔN := −div(μ∇·), μ the (nonnegative) Laplace operator with Neumann boundary conditions in Ω. If (uω , H) ∈ V is a solution of (3.9) with F defined by (3.8), then div(μH) = 0 and u, E, H are solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (3.1), and σ, D, B are defined by (2.8). ¯ with Φ ∈ D(ΔN eu ). This yields Proof. In (3.9) we take v = 0 and H = ∇Φ, μ
div(μH)div(μ∇Φ) − ω 2 μH · ∇Φ dx = 0. Ω
By Green’s formula we obtain 2 N eu div(μH)(ΔDir ). μ Φ + ω Φ) dx = 0 ∀Φ ∈ D(Δμ Ω eu D(ΔN ) μ
Since is dense in L2 (Ω), we conclude that div(μH) = 0 in Ω. The equations (2.9), (2.10), and (2.11) are obtained using Green’s formula with test functions H ∈ D(Ω)3 or v ∈ D(Ω)3 . It then remains to show the boundary condition −1 (curl H + iωeγ(u)) × n = J × n on Γ,
659
RESULTS FOR PIEZOELECTRIC SYSTEMS
which then implies (2.7), by the definition of E. We first remark that the above arguments show that w = −1 (curl H + iωeγ(u)) − J belongs to H(curl, Ω). We now state the next Green’s formula, which is a variant of the “standard” one (see Theorem I.2.11 of [10] or the identity (2.5) from [1]). A face F of Ω being fixed we denote by VF = {ϕ ∈ H 1 (Ω) : ϕ = 0 on Γ \ F }. We recall that for ϕ ∈ VF , Theorem 2.2 of [21] implies that the trace of ϕ on F 1/2 belongs to H00 (F ). By standard Green’s formula (w · curl ϕ − ϕ · curl w) dx = (w × n) · ϕ, Ω
Γ
valid for any w, ϕ ∈ H 1 (Ω)3 ; from the fact that the above left-hand side is continuous on H(curl, Ω) × (VF )3 , we deduce that if w ∈ H(curl, Ω), then w × n belongs to 1/2 (H00 (F ) )3 , and that Green’s identity (3.10) (w · curl ϕ − ϕ · curl w) dx = w × n, ϕ F ∀ϕ ∈ (VF )3 Ω
holds (where ., . F means the duality pairing between (H00 (F ) )3 and H00 (F )3 ). Now for a fixed face F , we temporarily assume that the z-axis is perpendicular to F and that F is included in the plane z = 0. For ϕ1 , ϕ2 ∈ D(F ), we take ⎛ ⎞ ϕ1 H (x, y, z) = η(z)μ−1 · ⎝ ϕ2 ⎠ 0 1/2
1/2
with a cut-off function η ∈ D(R) such that η(0) = 1 and a sufficiently small support so that H is zero on Γ\F . By construction the function H belongs to XT (Ω, μ)∩(VF )3 . Therefore in (3.9), taking as a test function v = 0 and this function H , by Green’s formula (3.10) we get w × n, H F = 0. As the third component of w × n is zero and the two first components of ⎛ ⎞ ϕ1 μ−1 · ⎝ ϕ2 ⎠ 0 1/2
are arbitrary in H00 (F ), we deduce that w × n = 0 in (H00 (F ) )3 . 1/2
By the definition of w, the requested boundary condition is proved.
660
D. MERCIER AND S. NICAISE
4. Existence and uniqueness results when μ = 0. 4.1. The E-formulation. In the case μ = 0, the second Maxwell equation (2.11) and the boundary condition (2.7) imply that iωE = ∇ϕ,
(4.1)
for some ϕ ∈ H01 (Ω). Therefore, in order to eliminate the vector field H, we take the divergence of the first Maxwell equation (2.10) to get div(iωD) = div J, and by the constitutive equation (2.8) we obtain div(iωeγ(u) + iωE) = div J. As before setting uω = iωu, the above identity is equivalent to div(eγ(uω ) + ∇ϕ) = div J.
(4.2)
This equation is now coupled with (3.3) or, equivalently, with
Div Cγ(uω ) − e ∇ϕ + ω 2 uω = −iωf in Ω (4.3) to form our system of partial differential equations. Clearly, its weak formulation is in finding a solution (uω , ϕ) ∈ H01 (Ω)4 of (4.4)
a((uω , ϕ), (v, ψ)) = F (v, ψ), ∀(v, ψ) ∈ H01 (Ω)4 ,
where the bilinear form a is defined by
a((u, ϕ), (v, ψ)) = (eγ(u) + ∇ϕ) · ∇ψ + (Cγ(u) − e ∇ϕ) · γ(v) − ω 2 u · v dx, Ω
and the linear form F is given by (4.5)
(iωf · v − J · ∇ψ) dx.
F (v, ψ) = Ω
The above considerations show the following lemma. Lemma 4.1. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), then (iωu, ϕ), with ϕ given by (4.1), is a solution of (4.4). As in the previous section, problem (4.4) satisfies the Fredholm alternative; namely, we shall prove the following lemma. Lemma 4.2. There exists a discrete set S0 such that for ω 2 ∈ S0 , the problem (4.4) has a unique solution for any F ∈ H −1 (Ω)4 . Proof. Introduce the bilinear form
b((u, ϕ), (v, ψ)) = (eγ(u) + ∇ϕ) · ∇ψ + (Cγ(u) − e ∇ϕ) · γ(v) dx. Ω
This form is coercive on H01 (Ω)4 since {∇ϕ · ∇ϕ + Cγ(u) · γ(u)} dx ϕ21,Ω + u21,Ω , b((u, ϕ), (u, ϕ)) = Ω
thanks to the positive definiteness of and C and Korn’s inequality. We conclude as in Lemma 3.3.
661
RESULTS FOR PIEZOELECTRIC SYSTEMS
We end this subsection by the converse result of Lemma 4.1. Theorem 4.3. If (uω , ϕ) ∈ H01 (Ω)4 is the solution of (4.4) with F defined by (4.5), then u, E are solutions of (2.9) and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (4.1), and σ, D, B are defined by (2.8) (implying B = 0 since μ = 0). Moreover, there exists H ∈ H 1 (Ω)3 such that (2.10) holds. Proof. The first part of the lemma follows from Green’s formula. For the second part we simply remark that D defined by (2.8) satisfies div(iωD) = div J in Ω, or, equivalently, J − iωD is divergence free. Therefore the existence of H such that curl H = J − iωD follows from Theorem I.3.4 of [10]. In the above lemma, we may notice that the magnetic field H is not unique, but this is not important since, in the case μ = 0, only u and E are of practical interest. 4.2. The H-formulation. Following the arguments of section 3, we may eliminate the electric field E and keep as unknowns u and H. Unfortunately H is no more uniquely determined, since the condition div(μH) = 0 and the boundary condition μH · n = 0 on Γ are trivially satisfied. Therefore as for eddy current problems [3] we impose the following Gauge conditions: (4.6) (4.7)
div H = 0 in Ω, H · n = 0 on Γ.
These conditions may be justified by an asymptotic argument, namely by taking μ = ηI, with η > 0 and letting η tends to 0 (cf. [6]). These arguments imply the following lemma. Lemma 4.4. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), assume that H satisfies the Gauge conditions (4.6)–(4.7). Then (iωu, H) belongs to V = H01 (Ω)3 × XT (Ω, I) and is solution of (3.9), with the sesquilinear form here defined by
−1 ¯ + div(H)div(H ¯ ) a((u, H), (v, H )) = (curl H + eγ(u)) · curl H Ω
+ Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − ω 2 u · v¯ dx. Again, the above problem enters in the Fredholm alternative. Lemma 4.5. There exists a discrete set S1 such that for 1 + ω 2 ∈ S1 , the problem (3.9), with a and V defined in Lemma 4.4, has a unique solution for any F ∈ V . Proof. Introduce the sesquilinear form b((u, H), (v, H )) := a((u, H), (v, H )) + (1 + ω 2 ) u · v¯ dx. Ω
Using Lemma 3.2, Korn’s inequality, and the compact embedding of XT (Ω, I) into L2 (Ω)3 [28], the bilinear form b is coercive on V . Consequently, the Friedrichs extension B associated with the triple (V, H, b) (with H defined as in the proof of Lemma 3.3) is invertible with a compact inverse. Again denote by A the Friedrichs extension of the triple (V, H, a). Clearly A = B − (1 + ω 2 )K, where the continuous operator K is defined by K(u, H) = (u, 0) .
662
D. MERCIER AND S. NICAISE
Writing (1 + ω 2 )−1 B −1 A = (1 + ω 2 )−1 I − B −1 K, we conclude as in Lemma 3.3, since B −1 K is a compact operator. Similarly to Theorem 3.5, the converse of Lemma 4.4 holds under some conditions on ω. eu Theorem 4.6. Assume that ω 2 is not an eigenvalue of −ΔN . If (uω , H) ∈ V I is a solution of (3.9) with F defined by (3.8) and a, V defined in Lemma 4.4, then divH = 0 and u, E, H are solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (3.1), and σ, D, B = 0 are defined by (2.8). 4.3. Equivalence between the E-formulation and the H-formulation. The goal of this section is to show that the displacement and electric fields obtained by the E-formulation and the H-formulation are identical. Theorem 4.7. Let u1 , E1 , H1 (resp., u2 , E2 , H2 ) be the solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by the E-formulation (resp., H-formulation). If ω 2 ∈ S0 (cf. Lemma 4.2), then u1 = u2 and E1 = E2 . Proof. Denote u = u1 − u2 and E = E1 − E2 . Since u1 , E1 and u2 , E2 satisfy (2.9) and (2.11), u, E satisfy Divσ + ω 2 u = 0 in Ω, curl H + iωD = 0 in Ω, curl E = 0 in Ω, where σ = σ1 − σ2 , andD = D1 − D2 are given by (2.8). Therefore E = ∇χ, with χ ∈ H01 (Ω) and the pair (u, χ) ∈ H01 (Ω)4 is a solution of (4.4) with F = 0. By Lemma 4.2, we conclude that u = 0 and χ = 0. Remark 4.8. The magnetic fields obtained by the E-formulation and the Hformulation cannot be identical, since the one obtained by the E-formulation does not necessarily satisfy the Gauge conditions. 5. Singularities of some elliptic systems. It is well known that the singularities of elliptic systems in Ω are produced by the corners and edges of Ω. Here we eu and of the 4×4 briefly describe the corner and edge singularities of the operator ΔN μ system (4.2)–(4.3) with Dirichlet boundary conditions. For the sake of brevity we restrict ourselves to a minimal description and refer the reader to the pioneer work [15] or standard books [11, 7, 18] for more details. 5.1. Corner singularities. Fix a corner c ∈ C of Ω and denote by (ρc , ϑc ) the spherical coordinates centered at c. Denote by Γc the infinite polyhedral cone which coincides with Ω near c. Let Gc be the intersection of Γc with the unit sphere centered at c. For any λ ∈ C, let us set Q (log ρc )q ψq (ϑc ) : ψq ∈ H 1 (Gc ) . S λ (Γc ) = ψ(x) = ρλc q=0 eu (Γc ) ΛN μ
eu of corner singular exponents of the operator ΔN in Γc is the The set μ λ set of λ ∈ C such that there exists a nonpolynomial solution Ψ ∈ S (Γc ) of
(5.1) (5.2)
Δμ Ψ = div(μ∇Ψ) = 0 in Γc , μ∇Ψ · n = 0 on ∂Γc .
663
RESULTS FOR PIEZOELECTRIC SYSTEMS
λ We denote by ZN eu (Γc , μ) the space of these solutions. Similarly for the 4 × 4 system (4.2)–(4.3) with Dirichlet boundary conditions, its set ΛDir C,,e (Γc ) of corner singular exponents is the set of λ ∈ C such that there exists a nonpolynomial solution (u, χ) ∈ S λ (Γc )4 of
⎧ ⎨ (5.3)
div(∇χ + eγ(u)) = 0
Div Cγ(u) − e ∇χ = 0 χ = 0, u = 0
⎩
in Γc , in Γc , on ∂Γc .
λ The space of these solutions is denoted by ZDir (Γc , C, , e).
5.2. Edge singularities. Fix an edge a ∈ A of Ω and denote by Γa × R the infinite polyhedral cone which coincides with Ω near a (Γa is then a two-dimensional sector). Denote by (ra , θa , za ) the cylindrical coordinates along a. Let Ga be the intersection of Γa with the unit sphere ra = 1. As before, for any λ ∈ C let us set λ
S (Γa ) =
ψ(x) =
raλ
Q
(log ra ) ψq (θa ) : ψq ∈ H (Ga ) . q
1
q=0 eu eu (Γa ) of corner singular exponents of the operator ΔN in Γa is the The set ΛN μ μ set of λ ∈ C such that there exists a nonpolynomial solution Ψ ∈ S λ (Γa ) of
div2 (μ2 ∇2 Ψ) = 0 in Γa , μ2 ∇2 Ψ · n = 0 on ∂Γa , where div2 (resp., ∇2 ) means the two-dimensional divergence (resp., gradient) and μ2 is the 2 × 2 matrix obtained from μ by dropping the third line and the third column of μ. λ We denote by ZN eu (Γa , μ) the space of these solutions. In the same way, the set ΛDir C,,e (Γa ) of edge singular exponents of the 4 × 4 system (4.2)–(4.3) with Dirichlet boundary conditions is the set of λ ∈ C such that there exists a nonpolynomial solution (u, χ) ∈ S λ (Γa )4 of ⎧ ⎪ ⎨ (5.4)
⎪ ⎩
˜ ∇χ ˜ + e˜ div( γ (u)) = 0 ˜ ˜ Div C γ˜ (u) − e ∇χ = 0 χ = 0, u = 0
in Γa , in Γa , on ∂Γa ,
where the sign ˜ means that all derivatives in the za -variable are replaced by zero. λ (Γa , C, , e). The space of these solutions is denoted by ZDir 6. Corner and edge singularities when μ is positive definite. Our goal is to describe the corner and edge singularities of the (regularized) problem (3.9). These singularities are obtained using some ideas from [4, 5, 6]. Let us start with the corner singularities. 6.1. Corner singularities. Fix a corner c ∈ C of Ω. We drop the index c in the above considerations. As usual we are looking for solutions of the homogeneous
664
D. MERCIER AND S. NICAISE
problem in the space λ 1 1 (Γ)3 × XT,loc (Γ, μ) :div(μH) ∈ Hloc (Γ), ST (Γ, μ) = (u, H) ∈ H0,loc u(x) = ρλ
Q
(log ρ)q Uq (ϑ),
q=0
H(x) = ρλ
Q
(log ρ)q Hq (ϑ) ,
q=0
the index loc meaning that the properties hold in all bounded domains far from c. This means that we look for a nonpolynomial solution (u, H) ∈ SλT (Γ, μ) of ⎧
curl −1 (curl H + eγ(u)) − μ∇ div(μH) = 0 in Γ, ⎪ ⎪ ⎪ ⎪ Div Cγ(u) + e −1 eγ(u) + e −1 curl H = 0 in Γ, ⎪ ⎪ ⎨ u=0 on ∂Γ, (6.1) μH · n = 0 on ∂Γ, ⎪ ⎪ ⎪ ⎪ μ∇(div(μH)) · n = 0 on ∂Γ, ⎪ ⎪ ⎩ −1 (curl H + eγ(u)) × n = 0 on ∂Γ. If a nontrivial solution exists, then we say that λ is a corner exponent. Inspired by [4, 5, 6], this problem is split up into three subproblems by introducing the auxiliary unknowns ψ = −−1 (curl H + eγ(u)) and q = div(μH). With this notation, problem (6.1) is equivalent to looking for q, (ψ, u), H successive solutions of div(μ∇q) = 0 in Γ, (6.2) μ∇q · n = 0 on ∂Γ, ⎧ ⎪ ⎪ ⎨ (6.3)
⎪ ⎪ ⎩ ⎧ ⎨
(6.4)
⎩
curl ψ = −μ∇q div(ψ
+ eγ(u)) = 0 Div Cγ(u) − e ψ = 0 ψ × n = 0, u = 0
in Γ, in Γ, in Γ, on ∂Γ,
curl H = −ψ − eγ(u) div(μH) = q μH · n = 0
in Γ, in Γ, on ∂Γ.
This means that three types of singularities may appear: Type 1: q = 0, ψ = u = 0, and H is a general solution of (6.4). Type 2: q = 0, (ψ, u) is a general solution of (6.3), and H is a particular solution of (6.4).
RESULTS FOR PIEZOELECTRIC SYSTEMS
665
Type 3: q is a general solution of (6.2), (ψ, u) is a particular solution of (6.3), and H is a particular solution of (6.4). These three types of singularities may be described with the help of the corner eu and of the 4 × 4 elliptic system (4.2)–(4.3) with singularities of the operator ΔN μ Dirichlet boundary conditions described in subsection 4.1. Since for our problem (3.9) div(μH) = 0 and is then regular, we do not describe the singularities of Type 3 because they cannot occur for any solution of (3.9). The singularities of Type 1 are obtained exactly as in Lemma 5.1 of [5] since only the magnetic field H is involved. Lemma 6.1. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is a singularity of λ+1 eu type 1 if and only if λ + 1 belongs to ΛN (Γ) and H = ∇Φ, with Φ ∈ ZN μ eu (Γ, μ). The situation is different for singularity of Type 2 since the coupling between the elasticity system and Maxwell’s equations appear via problem (6.3). Lemma 6.2. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is a singularity of λ type 2 if and only if λ belongs to ΛDir C,,e (Γ), ψ = ∇χ, with (u, χ) ∈ ZDir (Γ, C, , e), and H is given by (6.5)
H=
1 (a × x + ∇r) , λ+1
where a = −(ψ + eγ(u)) and r ∈ S λ+1 (Γ) is solution of div(μ∇r) = − div(μ(a × x)) in Γ, (6.6) μ∇r · n = −μ(a × x) · n on ∂Γ. Proof. As curl ψ = 0 in Γ, there exists χ ∈ S λ (Γ) such that ψ = ∇χ in Γ. From (6.3) we deduce that ⎧ + eγ(u)) = 0 ⎨ div(∇χ
Div Cγ(u) − e ∇χ = 0 ⎩ χ = 0, u = 0
in Γ, in Γ, on ∂Γ.
λ (Γ, C, , e). In view of section 5, we deduce that (u, χ) ∈ ZDir Now we readily check that H in the form (6.5) is a solution of (6.4) if and only if r is a solution of (6.6), whose solution exists by Theorem 4.14 of [23] and because curl(a × x) = (λ + 1)a. Lemma 6.3. There is no corner singularity of Type 2 in the strip λ ∈ ]−1, 0]. Proof. By Theorem 1 of [16] (see Remark 2 of [16], or Theorem 2 of [17]), the set ΛDir C,,e (Γ) ∩ [−1, 0] is empty. We conclude by Lemma 6.2. Since the singularities of our problem (3.9) have to be locally in XT (Γ, μ), among the singular exponents described above we select the subset Λc of λ > − 32 such that there exists (u, H) ∈ SλT (Γ, μ) solution of (6.1) such that
χ(u, H) ∈ XT (Γ, μ),
666
D. MERCIER AND S. NICAISE
where χ is a cut-off function equal to 1 near c. This last condition implies the following constraints for our first two types of singularities (see [5, 6]): eu eu Type 1: λ + 1 ∈ ΛN (Γ) and since ΛN (Γ) ∩ [−1, 0] is empty, by Lemma 5.4 μ μ of [5] we get the condition λ > −1. 1 1 3 Type 2: λ ∈ ΛDir C,,e (Γ) and by the condition χu ∈ H (Γ) , we get λ > − 2 . By Lemma 6.3, we arrive at the condition λ > 0. In conclusion, the set of corner singular exponents is eu Λc = {λ > −1 : λ + 1 ∈ ΛN (Γc )} ∪ {λ > 0 : λ ∈ ΛDir μ C,,e (Γc )}.
6.2. Edge singularities. Fix an edge a ∈ A of Ω and drop the index a. As before, we are looking for solutions of the homogeneous piezoelectric system (6.1) in Γ × R. For the elasticity system or Maxwell’s equations, the system obtained in the Cartesian coordinates (x, y, z) (according to the notation introduced in section 5, the z-axis contains the edge a) is split up into two independent problems in Γa . Here the coupling between these systems prevents this splitting. Therefore we simply follow the approach of the previous subsection. Namely, we search nonpolynomial solutions (u, H) of (6.1) independent of the z-variable and in the space λ 1 1 ST (Γ, μ) = (u, H) ∈ H0,loc (Γ × R)3 × XT,loc (Γ × R, μ) : div(μH) ∈ Hloc (Γ × R), u(x, y, z) = rλ
Q
(log r)q Uq (θ),
q=0
H(x, y, z) = rλ
Q
(log r)q Hq (θ) ;
q=0
the index loc here means that the properties hold in all bounded domains far from a. Then we introduce the auxiliary unknowns ψ and q defined as before but here independent of the variable z. This leads to the successive problems (6.2), (6.3), and (6.4) (where all derivatives in z are equal to zero). As before, singularities of Types 1, 2, and 3 appear. The singularities of Type 3 are not studied for the same reason as before. We now describe the singularities of Type 1 (compare with Lemma 6.1). Lemma 6.4. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is an edge singularity eu of type 1 if and only if λ + 1 belongs to ΛN (Γ), u = 0 and μ H = (∇2 Φ, 0) , λ+1 with Φ ∈ ZN eu (Γ, μ). Proof. As
⎞ ∂2 H3 ⎠ = 0, −∂1 H3 curl H = ⎝ ∂1 H2 − ∂2 H1 ⎛
the third component H3 is constant and the field H1 h= H2
RESULTS FOR PIEZOELECTRIC SYSTEMS
667
has a two-dimensional zero curl. Therefore there exists Φ such that h = ∇2 Φ, and we conclude as in Lemma 6.1. Let us continue with singularities of Type 2. Lemma 6.5. Assume that λ = −1 and λ = 0. Then (u, H) ∈ SλT (Γ, μ) is an edge singularity of Type 2 if and only if λ belongs to ΛDir C,,e (Γ), ψ = (∇2 χ, 0) , with λ (Γ, C, , e) and H given by (u, χ) ∈ ZDir ⎞ ⎛ a3 x2 (6.7)
H=⎝
λ a3 x1 λ a1 x2 −a2 x1 λ+1
⎠ + (∇2 r, 0) ,
where a = −(ψ + eγ(u)) and r ∈ S λ+1 (Γ) is the solution of ⎧ a3 x2 ⎪ λ ⎪ in Γ, ⎨ div2 (μ∇2 r) = − div2 μ a3 x1 λ (6.8) a3 x2 ⎪ ⎪ λ ⎩ ·n on ∂Γ. μ∇r · n = −μ a3 x1 λ
Proof. As in the previous lemma, since curl ψ = 0 in Γ, there exists χ ∈ S λ (Γ) such that ψ = (∇2 χ, 0) in Γ. From (6.3) we deduce that ⎧ + eγ(u)) = 0 ⎨ div(∇χ
Div Cγ(u) − e ∇χ = 0 ⎩ χ = 0, u = 0
in Γ, in Γ, on ∂Γ.
λ This means that (u, χ) ∈ ZDir (Γ, C, , e) owing to section 5 (recalling that u and χ are independent of the z-variable). As in Lemma 6.2, we easily check that H in the form (6.7) is a solution of (6.4) if and only if r is a solution of (6.8), whose existence follows from Theorem 4.14 of [23]. In summary we may formulate the following corollary. Corollary 6.6. The set Λa of edge exponents associated with a is given by eu Λa = {λ > −1 : λ + 1 ∈ ΛN (Γa )} ∪ {λ > 0 : λ ∈ ΛDir μ C,,e (Γa )}.
7. Regularity results. In this section, we describe regularity results of a solution u, H, E of our time-harmonic piezoelectric system. These results use its weak formulation obtained above, are based on the knowledge of corner and edge singularities of these formulation described in the previous section, and rely on the application of Mellin’s techniques as in [4, 5]. For the sake of brevity, we do not describe singular decomposition of such a solution. Using the techniques from [4, 5], for sufficiently smooth data we may obtain a decomposition of u, H, E into a regular part and a singular one. 7.1. The case μ positive definite. Before stating our regularity results, let us introduce the following notation: For any corner c ∈ C introduce λc,u = min{λ > 0 : λ ∈ ΛDir C,,e (Γc )}, λc,H = min{λ : λ ∈ Λc }.
668
D. MERCIER AND S. NICAISE
Similarly, for any edge a ∈ A define λa,u = min{λ > 0 : λ ∈ ΛDir C,,e (Γa )}, λa,H = min{λ : λ ∈ Λa }. Finally, we set 1 τu := min min λa,u , + min λc,u , a∈A 2 c∈C 1 τH := min min λa,H , + min λc,H . a∈A 2 c∈C
We remark that τu does not depend on μ. Theorem 7.1. Let (uω , H) ∈ H01 (Ω)3 × XT (Ω, μ) be a solution of problem (3.9), with J and f smooth enough. Then we have (7.1) (7.2)
uω ∈ H 1+τ (Ω)3 ∀τ < τu , H ∈ H τ (Ω)3 ∀τ < τH .
Proof. Mellin’s techniques directly imply that (uω , H) ∈ H τ (Ω)6 ∀τ < τH . The extra regularity for uω follows from the description of the singularities of problem (3.9) made in section 6, where we see that the regularity of uω is only determined by the singularity of type 2. Corollary 7.2. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 3.5. Then u has the regularity (7.1), H has the regularity (7.2), and E satisfies (7.3)
E ∈ H τ (Ω)3 ∀τ < τu .
Proof. The previous lemma directly yields the regularity for u and H. By (3.1), E has the regularity of curl H + iωeγ(u). Since the singularities of type 1 for H are gradient, the regularity of E is only determined by the singularities of type 2. 7.2. The case μ = 0. Here, to solve our piezoelectric system we may use either the E-formulation or the H-formulation. Both formulations give the same regularity for u and E, which is quite natural since generically they are identical (see Theorem 4.7). Theorem 7.3. Let (uω , ϕ) ∈ H01 (Ω)4 be a solution of problem (4.4) with J and f smooth enough; then we have (7.4)
(uω , ϕ) ∈ H 1+τ (Ω)4 ∀τ < τu ,
where τu is defined as before. Proof. Since the system associated with (4.4) is the strongly elliptic system (4.2)–(4.3) with Dirichlet boundary conditions, the mentioned regularity result follows from Theorem 5.11 of [7]. Corollary 7.4. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 4.3. Then u has the regularity (7.1) and E satisfies (7.3).
RESULTS FOR PIEZOELECTRIC SYSTEMS
669
1 ∇ϕ. Proof. This is a direct consequence of the previous theorem since E = iω For the H-formulation, the arguments of the previous subsection directly give the next results. Theorem 7.5. Let (uω , H) ∈ H01 (Ω)3 × XT (Ω, I) be a solution of problem (3.9) in the sense of Lemma 4.4 with J and f smooth enough. Then the regularity results (7.1)–(7.2) hold, τH being defined as before but with μ = I. Corollary 7.6. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 4.6. Then u has the regularity (7.1), H has the regularity (7.2), and E satisfies (7.3).
8. Numerical examples of edge singular exponents. To illustrate our theoretical results, we give some edge singular exponents of our piezoelectric system corresponding to some illustrative materials when c45 = c16 = c26 = c36 = 0, c11 = c22 , 12 , 11 = 22 , and 12 = 0. We further consider a dihedral cone c44 = c55 , c66 = c11 −c 2 Γa × R such that the edge a of this cone (corresponding to the axis z = 0) is parallel to the poling direction x3 . In that case, the system (5.4) reduces to the following 4 × 4 system (see, for instance, [19]): 12 12 c11 ∂12 + c11 −c u1 0 ∂22 c11 +c ∂1 ∂2 2 2 = in Γa , c11 −c12 2 c11 +c12 0 u2 ∂1 ∂2 ∂1 + c11 ∂22 2 2 c44 Δ e15 Δ 0 u3 = in Γa , −e15 Δ 11 Δ χ 0 χ = 0, u = 0 on ∂Γa , where, as usual, Δ = ∂12 + ∂22 . Setting v = c44 u3 + e15 χ and w = −e15 u3 + 11 χ, this system is equivalent to 12 12 c11 ∂12 + c11 −c u1 0 ∂22 c11 +c ∂1 ∂2 2 2 = in Γa , c11 −c12 2 c11 +c12 0 u2 ∂1 ∂2 ∂1 + c11 ∂22 2 2 Δv = 0 in Γa , Δw = 0 in Γa , u1 = u2 = v = w = 0 on ∂Γa . This last system is decoupled into a 2 × 2 system of the isotropic elasticity for the 12 ) with Dirichlet pair (u1 , u2 ) (with Lam´e coefficients given by λ = c12 and μ = c11 −c 2 boundary conditions and two Dirichlet problems for v and w. For this last Dirichlet problem, it is well known that the singular exponents are given by lπ ω , with l ∈ N, when ω is the interior opening of Γa . On the other hand, the singular exponents of the isotropic elasticity are also well known and are the set of λ ∈ C such that (see, for instance, [12]) k 2 sin(λω)2 = λ2 sin(ω)2 , 11 −c12 with k = 3c c11 +c12 . The zeros of this equation can be easily computed using a Newton method. Figures 8.1 and 8.2 show the real part of the singular exponents in the strip λ ∈ (0, 2) for all values of ω ∈ (0, 2π] for the piezoelectric materials P ZT -4 and BaT iO3 in black lines. In these figures, the gray lines correspond to the curves lπ ω , for l = 1, 2, 3. From these figures, we may conclude that for these materials the strongest singular
670
D. MERCIER AND S. NICAISE 2
Re ( λ )
1.5
1
0.5
0 0
50
100
150
200
250
300
350
300
350
ω
Fig. 8.1. The edge exponents for the P ZT -4. 2
Re ( λ )
1.5
1
0.5
0 0
50
100
150
200
250
ω
Fig. 8.2. The edge exponents for the BaT iO3 .
exponent λ0 (strongest in the sense that λ0 is minimal) is always the one coming from the 2 × 2 system of elasticity and that λ0 > 1/2 (see Theorem 2.2 of [22]). This last property implies the H 3/2 -regularity (resp., H 1/2 -regularity) for u (resp., for E) along the edge. Note further that the curves obtained for the P ZT -4 are similar to the ones shown in Figure 4b of [29] (in that paper, the singular exponents are obtained using an extension of Lekhnitskii’s representation for elastic solids). Remark 8.1. If the edge is not parallel to the poling direction, then the above decoupling phenomenon does not appear. The calculation of the edge singular expo-
RESULTS FOR PIEZOELECTRIC SYSTEMS
671
nents is then more complicated and will be done using a finite element method. This will be done in a forthcoming paper [20]. 9. Conclusions. We have investigated a general time-harmonic piezoelectric system, which contains as special cases standard models like ceramics. We developed the appropriate formalism in order to get existence and uniqueness results of weak solutions in the case when the magnetic permeability matrix is positive definite (case of the BaT iO3 ) and the case when the magnetic permeability matrix is equal to zero (case of the P ZT ). In the latter case, we gave two different formulations: the E-formulation and the H-formulation. For this second one, Gauge conditions are introduced as for eddy current problem. We further show that generically these two formulations yield the same solutions. We described the corner and edge singularities of our general system and deduced some regularity results. Some edge singular exponents were given in order to illustrate our theoretical results. Acknowledgments. We thank C. Courtois and A. Leriche from the laboratory LMP (UVHC) for valuable discussions about these topics. REFERENCES [1] C. Amrouche, C. Bernardi, M. Dauge, and V. Girault, Vector potentials in threedimensional nonsmooth domains, Math. Methods Applied Sci., 21 (1998), pp. 823–864. [2] D. A. Berlincourt, D. R. Curran, and H. Jaffe, Piezoelectric and piezomagnetic materials and their function in transducers, Physical Acoustics, 1 (1964), pp. 169–270. [3] A. Bossavit, Electromagn´ etisme en vue de la mod´ elisation, Springer-Verlag, Paris, 1993. [4] M. Costabel and M. Dauge, Singularities of electromagnetic fields in polyhedral domains, Arch. Ration. Mech. Anal., 151 (2000), pp. 221–276. [5] M. Costabel, M. Dauge, and S. Nicaise, Singularities of Maxwell interface problems, RAIRO Mod´ el. Math. Anal. Num´er., 33 (1999), pp. 627–649. [6] M. Costabel, M. Dauge, and S. Nicaise, Singularities of eddy current problems, RAIRO Mod´ el. Math. Anal. Num´er., 37 (2003), pp. 807–831. [7] M. Dauge, Elliptic Boundary Value Problems in Corner Domains. Smoothness and Asymptotics of Solutions, L. N. in Math. 1341, Springer-Verlag, Berlin, 1988. [8] A. C. Eringen and G. A. Maugin, Electrodynamics of Continua, Vols. 1, 2, Springer-Verlag, New York, 1990. ¨ndig, and G. Mishuris, Piezoelectricity in Multi-layer Actuators: Modelling [9] W. Geis, A.-M. Sa and Analysis in Two and Three-Dimensions, Preprint IANS, 2003/23, Univ. Stuttgart, Germany, 2003. [10] V. Girault and P.-A. Raviart, Finite Element Methods for Navier-Stokes Equations, Springer Ser. Comput. Math. 5, Springer-Verlag, New York, 1986. [11] P. Grisvard, Elliptic Problems in Nonsmooth Domains, Monogr. Stud. Math. 24, Pitman, Boston, 1985. [12] P. Grisvard, Singularit´ e en elasticit´ e, Archiv. Ration. Mech. Anal., 107 (1989), pp. 157–180. [13] D. Iesan, Plane strain problems in piezoelectricity, Internat. J. Engrg. Sci., 25 (1987), pp. 1511– 1523. [14] T. Ikeda, Fundamentals of Piezoelectricity, Oxford University Press, Oxford, 1996. [15] V. A. Kondrat’ev, Boundary-value problems for elliptic equations in domains with conical or angular points, Trans. Moscow Math. Soc., 16 (1967), pp. 227–313. [16] V. A. Kozlov and V. G. Maz’ya, Spectral properties of the operator bundles generated by elliptic boundary value problems in a cone, Funct. Anal. Appl., 22 (1988), pp. 38–46. [17] V. A. Kozlov and V. G. Maz’ya, On the behaviour of the spectrum of parameter-depending operators under small variation of the domain and an application to operator pencils generated by boundary value problems in a cone, Math. Nachr., 153 (1991), pp. 123–129. [18] V. A. Kozlov, V. G. Maz’ya, and J. Rossman, Elliptic Boundary Value Problems in Domains with Point Singularities, Math. Surveys Monogr. 52, AMS, Providence, RI, 1997. [19] L. Marin, The Behaviour of Piezoelectric Materials under Mechanical and Electrical Loadings, Master Thesis, Univ. Kaiserslautern, Germany, 1998.
672
D. MERCIER AND S. NICAISE
[20] D. Mercier and S. Nicaise, Regularity Results for a Two-Dimensional Piezoelectric System and Crack Propagation Criteria, manuscript. [21] S. Nicaise, Exact controllability of a pluridimensional coupled problem, Rev. Mat. Complut., 5 (1992), pp. 91–135. [22] S. Nicaise, About the Lam´ e system in a polygonal or a polyhedral domain and a coupled problem between the Lam´ e system and the plate equation I: Regularity of the solutions, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 19 (1992), pp. 327–361. [23] S. Nicaise, Polygonal Interface Problems, Peter Lang, Berlin, 1993. [24] W. Novacki, Mathematical models of phenomenological piezoelectricity, in Mathematical Models and Methods in Mechanics, Banach Center Publ. 15, Polish Academy of Sciences, Warsaw, Poland, 1985, pp. 593–607. [25] Y. E. Pak, Linear electro-elastic fracture mechanics of piezoelectric materials, Internat. J. Fracture, 54 (1992), pp. 79–100. [26] E. Pan and P. R. Heyliger, Exact solutions for magneto-electro-elastic laminates in cylindrical bending, Internat. J. Solids Structures, 40 (2003), pp. 6859–6876. [27] H. A. Sosa and Y. E. Pak, Three-dimensional eigenfunction analysis of a crack in a piezoelectric material, Internat. J. Solids Structures, 26 (1990), pp. 1–15. [28] Ch. Weber, A local compactness theorem for Maxwell’s equations, Math. Methods Appl. Sci., 2 (1980), pp. 12–25. [29] X.-L. Xu and R. Rajapakse, On singularities in composite piezoelectric wedges and junctions, Internat. J. Solids Structures, 37 (2000), pp. 3253–3275. [30] X.-L. Xu and R. Rajapakse, On a plane crack in piezoelectric solids, Internat. J. Solids Structures, 38 (2001), pp. 7643–7658.
c 2005 Society for Industrial and Applied Mathematics
EXISTENCE, UNIQUENESS, AND REGULARITY RESULTS FOR PIEZOELECTRIC SYSTEMS∗ D. MERCIER† AND S. NICAISE† Abstract. We investigate the time-harmonic piezoelectric system (a system coupling the elasticity system with the full Maxwell’s equations) in polyhedral domains of the space. Existence and uniqueness results of weak solutions are proved in different cases. We describe the corner and edge singularities of that system and deduce some regularity results. Key words. elasticity system, Maxwell’s system, singularities AMS subject classifications. 35J25, 35Q60 DOI. 10.1137/040617728
1. Introduction. Smart structures made of piezoelectric and/or piezomagnetic materials are gaining attention in applications since they are able to transform the energy from one type to another (magnetic, electric, and mechanical), allowing them to be used as sensors and/or actuators. Commonly used piezoelectric materials are ceramics and quartz. The mathematical model of this system starts to be well established [2, 8, 14, 24, 26] and corresponds to a coupling between the elasticity system and Maxwell’s equations (see below). A full mathematical analysis is not yet done, except in some particular cases [13, 19]. Namely, in these two works the electric field E is assumed to be curl free, i.e., E = ∇ϕ, where ϕ is an electric potential and a twodimensional reduction is made. In [13], existence and uniqueness results in smooth domains are obtained using integral equations, while in [19] a variational formulation in polygonal domains is given and two-dimensional singularities are briefly described. On the other hand, there exists an extensive list of papers from mechanics literature describing singularities of some particular piezoelectric materials with a plane crack [25, 27, 30] or along wedges [29]. But to our knowledge, an exact description of corner/edge singularities of the general piezoelectric system in three-dimensional polyhedral domains is not yet obtained. Such a description is very important since piezoelectric ceramics are very brittle, and therefore their fracture behavior must be understood. The knowledge of such singularities also has numerical implications, such as convergence speed. This paper has, therefore, the following goals: We present a general piezoelectric system, which includes standard models of ceramics like the P ZT or the BaT iO3 . We further develop some variational formulations which are the natural ones because they lead to solutions in the energy spaces (here called weak solutions). We prove existence and uniqueness results of weak solutions of the time-harmonic system in two different cases: the case when the magnetic permeability matrix is positive definite (BaT iO3 ) and the case when the magnetic permeability matrix is zero (P ZT ). In that second case, we even give two different formulations and show that generically they give rise to the same solutions. Moreover, we describe the corner and edge singularities of our ∗ Received
by the editors October 27, 2004; accepted for publication (in revised form) February 22, 2005; published electronically November 9, 2005. http://www.siam.org/journals/sima/37-2/61772.html † Universit´ e de Valenciennes et du Hainaut Cambr´esis, MACS, Institut des Sciences et Techniques de Valenciennes, F-59313 Valenciennes Cedex 9, France ([email protected], Serge. [email protected]). 651
652
D. MERCIER AND S. NICAISE
general system and deduce some regularity results. Some edge singular exponents are briefly described; more examples will be given in a forthcoming paper [20], where a two-dimensional model and fracture criteria will be considered. The analysis of more sophisticated models, like coupling between piezoelectric and magnetostrictive materials [26] or piezoelectric and purely elastic materials [29, 9], will be investigated in the future. The paper is organized as follows. Section 2 introduces the nonstationary model problem and its time-harmonic version. Existence and uniqueness results are given in section 3 when the magnetic permeability matrix is positive definite. Section 4 is devoted to existence and uniqueness results in the case of a zero magnetic permeability matrix. There we consider two different formulations: for the first one (called the E-formulation) the magnetic field H is eliminated, while for the second one (the Hformulation) the electric field E is eliminated. For the latter formulation, like for the eddy current problem, Gauge conditions are necessary. We further show their generic equivalence. After a short description of corner and edge singularities of some useful elliptic systems in section 5, we obtain the corner and edge singularities of our system in section 6. Regularity results are deduced in section 7. Some edge singular exponents are finally presented in section 8. 2. Setting of the problem. Let Ω be a bounded domain of R3 with a Lipschitz boundary Γ. For the sake of simplicity, we suppose that Γ is piecewise plane and connected and that Ω is simply connected. In this domain, we consider the following nonstationary piezoelectric system of constitutive equations [2, 8, 14, 24]: (2.1) (2.2) (2.3)
σij = aijkl γkl (u) − ekij Ek ∀i, j = 1, 2, 3, Di = ij Ej + eikl γkl (u) ∀i = 1, 2, 3, Bi = μij Hj ∀i = 1, 2, 3.
The equations of equilibrium are (2.4)
∂t2 ui = ∂j σji + fi ∀i = 1, 2, 3
for the elastic displacement and ∂t D + J = curl H, ∂t B = − curl E for the electric/magnetic fields (Maxwell’s equations), where f is the body force density and J is the vector current density function. As usual curl H = (∂2 H3 − ∂3 H2 , ∂3 H1 − ∂1 H3 , ∂1 H2 − ∂2 H1 ) , when H = (H1 , H2 , H3 ) . This system models the coupling between Maxwell’s system and the elastic one [2, 8, 14, 24], in which E(x, t), H(x, t) are the electric and magnetic fields at the point x ∈ Ω at time t, u(x, t) is the displacement field at the point x ∈ Ω at time t, and (γij (u))3i,j=1 is the strain tensor given by 1 γij (u) = 2
∂ui ∂uj + ∂xj ∂xi
.
Here (σij )3i,j=1 , D = (D1 , D2 , D3 ) , and B = (B1 , B2 , B3 ) are the stress tensor, electric displacement, and magnetic induction, respectively. , μ are the electric permittivity and magnetic permeability, respectively, and are supposed to be real, symmetric 3 × 3 matrices. In what follows the matrix is supposed to be positive definite,
653
RESULTS FOR PIEZOELECTRIC SYSTEMS
while μ is only supposed to be nonnegative. The elasticity tensor (aijkl )i,j,k,l=1,2,3 is made of constant entries such that aijkl = ajikl = aklij , and satisfies the ellipticity condition (2.5)
aijkl γij γkl ≥ αγij γij ,
for every symmetric tensor (γij ) and some α > 0. The piezoelectric tensor ekij is also made of constant entries such that ekij = ekji . The system is completed with the Dirichlet boundary conditions for the displacement field, (2.6)
u = 0 on Γ,
and those of a perfect conductor, (2.7)
E × n = 0 on Γ.
As usual n is the exterior unit normal vector along Γ. In order to write the above problem in a more compact form, the strain and stress tensors are expressed as 6 × 1 vectors, namely
γ(u) = (γ11 (u), γ22 (u), γ33 (u), 2γ23 (u), 2γ31 (u), 2γ12 (u)) , σ = (σ11 , σ22 , σ33 , σ23 , σ31 , σ12 ) . With this notation, the constitutive equations (2.1)–(2.3) may be equivalently written as ⎛ ⎞ ⎛ ⎞ σ γ(u) ⎝ D ⎠ = M ⎝ E ⎠, (2.8) B H where M is a 12 × 12 matrix given by ⎛
C M =⎝ e 0
−e 0
⎞ 0 0 ⎠, μ
where C is a 6 × 6 symmetric matrix depending on the elasticity tensor, e is a 3 × 6 matrix depending on the piezoelectric tensor, and , μ are as described above. Note that the ellipticity assumption (2.5) is equivalent to the fact that C is a positive definite matrix. For a monoclinic material with poling direction in the x3 -axis, the material constant matrices are expressed by (see, for instance, [26]) ⎞ ⎛ c11 c12 c13 0 0 c16 ⎜ c12 c22 c23 0 ⎛ ⎞ 0 c26 ⎟ ⎟ ⎜ 0 0 0 e14 e15 0 ⎟ ⎜ c13 c23 c33 0 0 c36 ⎟ 0 0 e24 e25 0 ⎠ , , e=⎝ 0 C=⎜ ⎜ 0 0 0 c44 c45 0 ⎟ ⎟ ⎜ e e e 0 0 e36 31 32 33 ⎝ 0 0 0 c45 c55 0 ⎠ 0 c66 c16 c26 c36 0 ⎞ ⎛ ⎞ ⎛ μ11 μ12 0 11 12 0 0 ⎠. = ⎝ 12 22 0 ⎠ , μ = ⎝ μ12 μ22 0 0 33 0 0 μ33
654
D. MERCIER AND S. NICAISE
For the orthotropic piezoelectric P ZT -4, the material coefficients are given by (cij in 109 N/m2 , eij in C/m2 , ij in 10−9 C 2 /N m2 ): 12 c11 = c22 = 23.8 c33 = 10.6 c44 = 2.15 c55 = 4.4 c66 = c11 −c = 6.43 2 c12 = 3.98 c13 = 2.19 c23 = 1.92 11 = 22 = 0.110625 33 = 0.106023 e32 = −0.14 e33 = −0.28 e24 = e15 = −0.01. e31 = −0.13
Here and below, nongiven coefficients are equal to zero. For the piezoelectric BaT iO3 , the material coefficients are given by (cij in 109 N/m2 , eij in C/m2 , ij in 10−9 C 2 /N m2 , μij in 10−6 N s2 /C 2 ): c11 = c22 = 166 c33 = 162 c44 = c55 = 43 c12 = 77 c13 = c23 = 78 11 = 22 = 11.2 e33 = −18.6 e24 = e15 = 11.6 e31 = e32 = −4.4
12 c66 = c11 −c = 44.5 2 33 = 12.6 μ11 = μ22 = 5 μ33 = 10
With the above notation, the equation of equilibrium (2.4) is also equivalent to ∂t2 u = Divσ + f, where Div is the operator-valued matrix defined by ⎛
∂1 Div = ⎝ 0 0
0 ∂2 0
0 0 ∂3
0 ∂3 ∂2
∂3 0 ∂1
⎞ ∂2 ∂1 ⎠ . 0
Assuming that u, E, H are of the form u(x, t) = e−iωt u(x), E(x, t) = e−iωt E(x), H(x, t) = e−iωt H(x), for some real constant ω (the data being of the same form), the above system is reduced to the time-harmonic piezoelectric system in Ω consisting of the constitutive equation (2.8), the time-harmonic equilibrium equation (2.9)
Divσ + ω 2 u = −f in Ω,
and the time-harmonic Maxwell’s equations (2.10)
curl H + iωD = J in Ω,
(2.11)
curl E − iωμH = 0 in Ω.
In the whole paper we consider the nonstationary case; namely, we assume that ω > 0. In other words we assume that the variation in time of u, E, and H is periodic in time with a frequency equal to 2π/ω. The stationary case ω = 0 requires the use of Gauss’ law divD = ρ, where ρ is the charge density function. This case is treated as in section 4.1 below and is then left to the reader.
RESULTS FOR PIEZOELECTRIC SYSTEMS
655
3. Existence and uniqueness results when μ is positive definite. Replacing D by its expression (2.2) in the first Maxwell equation (2.10), we get curl H + iω(E + eγ(u)) = J in Ω. This identity is clearly equivalent to iωE = J − curl H − iωeγ(u) in Ω.
(3.1)
With this identity, the second Maxwell equation (2.11) is then equivalent to curl(−1 (curl H + iωeγ(u))) − ω 2 μH = curl(−1 J) in Ω.
(3.2)
In the same way using the constitutive equation (2.8) in the equation of motion (2.9), we obtain
Div Cγ(u) − e E + ω 2 u = −f in Ω. (3.3) Using the identity (3.1), we arrive at 1 −1 −1 (3.4) Div Cγ(u) + e eγ(u) + e (curl H) iω
T −1 1 2 +ω u = −f + Div e J in Ω. iω The two equations (3.2) and (3.4) constitute the system of partial differential equations that we will study and whose unknowns are u and H. Due to the boundary conditions (2.6) and (2.7), this system is completed with (2.6) and −1 (curl H + iωeγ(u)) × n = J × n, (μH) · n = 0 on Γ.
(3.5)
The weak formulation of the above problem is obtained in the following way: We introduce the space (see [4, 5]) XT (Ω, μ) = {v ∈ L2 (Ω)3 : div(μv) ∈ L2 (Ω), curl v ∈ L2 (Ω)3 and (μv) · n = 0 on Γ} equipped with its natural norm. Then we multiply the system (3.2) by a test function ¯ ∈ XT (Ω, μ), integrate the result in Ω, and integrate by parts to get (assuming that H u and H are regular enough)
−1 ¯ − ω 2 μH · H ¯ dx (curl H + iωeγ(u)) · curl H Ω −1 ¯ ¯ ds ∀H ∈ XT (Ω, μ). = J · curl H dx + (J × n) · H Ω
Γ
Since from the second Maxwell equation (2.11) div(μH) = 0 in Ω, the above identity is equivalent to
−1 ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ dx (3.6) (curl H + iωe · γ(u)) · curl H Ω ¯ dx + (J × n) · H ¯ ds ∀H ∈ XT (Ω, μ). = −1 J · curl H Ω
Γ
This last equation may be called the regularized formulation of the system (3.2).
656
D. MERCIER AND S. NICAISE
Similarly multiplying (3.4) by v¯ ∈ H01 (Ω)3 , integrating the result in Ω, and using Green’s formula (Divγ) · v¯ dx = − γ · γ(¯ v ) dx, Ω
Ω
obtained by componentwise integration by parts in Ω, we arrive at 1 −1 −1 2 Cγ(u) + e eγ(u) + e curl H · γ(¯ (3.7) v ) − ω u · v¯ dx iω Ω 1 = f · v¯ − eT −1 J · γ(¯ v ) dx ∀v ∈ H01 (Ω)3 . iω Ω As these two identities are coupled, multiplying the second one by iω and summing the result we arrive at the following problem: ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ −1 (curl H + iωeγ(u)) · curl H Ω 1 +iω Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − iω 3 u · v¯ dx = F (v, H ), iω where we have set (3.8)
F (v, H ) =
−1
¯ − eT −1 J · γ(¯ J · curl H v ) + iωf · v¯ dx +
Ω
¯ ds. (J × n) · H
Γ
In order to get a well-posed problem we set uω = iωu. Then we see that the above problem is equivalent to finding a solution (uω , H) ∈ V of (3.9)
a((uω , H), (v, H )) = F (v, H ) ∀(v, H ) ∈ V,
where we set V = H01 (Ω)3 × XT (Ω, μ),
−1 ¯ + div(μH)div(μH ¯ ) − ω 2 μH · H ¯ a((u, H), (v, H )) = (curl H + eγ(u)) · curl H Ω
+ Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − ω 2 u · v¯ dx. In summary we have shown the following lemma. Lemma 3.1. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), then (iωu, H) is a solution of (3.9). We now remark that the bilinear form a may be equivalently written a ((u, H), (v, H )) ¯ curl H curl H ¯ − ω 2 u · v¯ dx, ¯ ) − ω 2 μH · H = A · + div(μH)div(μH γ(u) γ(¯ v) Ω where A is the 9 × 9 symmetric matrix defined by −1 −1 e A= . e −1 C + e −1 e
657
RESULTS FOR PIEZOELECTRIC SYSTEMS
One easily checks that this matrix is positive definite for the material coefficients of the P ZT -4 and BaT iO3 , for instance. In fact this is always true if and C are positive definite (always true in our setting), independently of the piezoelectric coefficients. Lemma 3.2. If and C are positive definite matrices, then the matrix A defined above is also positive definite. Proof. We only need to show that X X X A · ≥ 0, ∀ ∈ R9 , γ γ γ X X X A · =0⇒ = 0. γ γ γ By the definition of A we see that X X A · = X −1 X + 2X −1 eγ + γ Cγ + γ e −1 eγ. γ γ ˜ = −1/2 X and Y˜ = −1/2 eγ (note that both vectors are in R3 ), we see that Setting X X X ˜ X ˜ + 2X ˜ Y˜ + γ Cγ + Y˜ Y˜ A · =X γ γ ˜ + Y˜ 22 + γ Cγ, = X where · 2 clearly means the Euclidean norm of R3 . This identity directly implies the first assertion by the positive definitiveness of C. For the second assertion, if X X A · = 0, γ γ then the above identity and again the positive definitiveness of C imply that ˜ + Y˜ 2 = γ Cγ = 0. X 2 Therefore γ = 0 and, consequently, Y˜ = 0 in view of its definition (independently of ˜ = 0 and, by the positive definitiveness of , we conclude e). We then obtain that X that X = 0. This lemma allows us to show that problem (3.9) enters within the framework of the Fredholm alternative. Indeed, we shall prove the following lemma. Lemma 3.3. There exists a discrete set S such that for 1 + ω 2 ∈ S, the problem (3.9) has a unique solution for any F ∈ V . Proof. Introduce the sesquilinear form 2 ¯ μH · H dx + u · v¯ dx . b((u, H), (v, H )) := a((u, H), (v, H )) + (1 + ω ) Ω
Ω
Then the above considerations yield b ((u, H), (v, H )) ¯ curl H curl H ¯ ¯ = A · + div(μH)div(μH ) + μH · H + u · v¯ dx. γ(u) γ(¯ v) Ω
658
D. MERCIER AND S. NICAISE
It is coercive on V, since by Lemma 3.2 we have
b((u, H), (u, H)) | curl H|2 + |γ(u)|2 + |div(μH)|2 + |H|2 + |u|2 dx. Ω
Therefore by Korn’s inequality we get b((u, H), (u, H)) u21,Ω + H2XT (Ω,μ) . Since the space XT (Ω, μ) is compactly embedded into L2 (Ω)3 [28], and since by the Rellich–Kondrasov theorem H01 (Ω) is compactly embedded into L2 (Ω), we deduce that V is compactly embedded into H := L2 (Ω)3 × L2 (Ω)3 . These facts imply that the Friedrichs extension B from H into H induced by the triple (V, H, b) is invertible with a compact inverse. Now denote by A the Friedrichs extension of the triple (V, H, a). The relation between a and b implies that A = B − (1 + ω 2 )Iμ , where the operator Iμ is defined by Iμ (u, H) = (u, μH) and is clearly continuous from H into H. Using this operator we may write (1 + ω 2 )−1 B −1 A = (1 + ω 2 )−1 I − B −1 Iμ , where I is the identity operator from H into H. Since B −1 Iμ is a compact operator, it has a discrete spectrum with positive eigenvalues. If we denote by S the set of the inverses of these eigenvalues, then for (1 + ω 2 ) ∈ S, the operator (1 + ω 2 )−1 I − B −1 Iμ is invertible and consequently A is as well. Remark 3.4. Note that in the case μ = 0, problem (3.9) does not enter in the Fredholm framework since XT (Ω, 0) is reduced to H(curl, Ω) and is no more compactly embedded into L2 (Ω)3 . This drawback will be avoided by introducing Gauge conditions; see section 4.2. Let us now show that the converse of Lemma 3.1 holds under some conditions on ω (compare with Theorem 0.1 of [4]). eu Theorem 3.5. Assume that ω 2 is not an eigenvalue of −ΔN := −div(μ∇·), μ the (nonnegative) Laplace operator with Neumann boundary conditions in Ω. If (uω , H) ∈ V is a solution of (3.9) with F defined by (3.8), then div(μH) = 0 and u, E, H are solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (3.1), and σ, D, B are defined by (2.8). ¯ with Φ ∈ D(ΔN eu ). This yields Proof. In (3.9) we take v = 0 and H = ∇Φ, μ
div(μH)div(μ∇Φ) − ω 2 μH · ∇Φ dx = 0. Ω
By Green’s formula we obtain 2 N eu div(μH)(ΔDir ). μ Φ + ω Φ) dx = 0 ∀Φ ∈ D(Δμ Ω eu D(ΔN ) μ
Since is dense in L2 (Ω), we conclude that div(μH) = 0 in Ω. The equations (2.9), (2.10), and (2.11) are obtained using Green’s formula with test functions H ∈ D(Ω)3 or v ∈ D(Ω)3 . It then remains to show the boundary condition −1 (curl H + iωeγ(u)) × n = J × n on Γ,
659
RESULTS FOR PIEZOELECTRIC SYSTEMS
which then implies (2.7), by the definition of E. We first remark that the above arguments show that w = −1 (curl H + iωeγ(u)) − J belongs to H(curl, Ω). We now state the next Green’s formula, which is a variant of the “standard” one (see Theorem I.2.11 of [10] or the identity (2.5) from [1]). A face F of Ω being fixed we denote by VF = {ϕ ∈ H 1 (Ω) : ϕ = 0 on Γ \ F }. We recall that for ϕ ∈ VF , Theorem 2.2 of [21] implies that the trace of ϕ on F 1/2 belongs to H00 (F ). By standard Green’s formula (w · curl ϕ − ϕ · curl w) dx = (w × n) · ϕ, Ω
Γ
valid for any w, ϕ ∈ H 1 (Ω)3 ; from the fact that the above left-hand side is continuous on H(curl, Ω) × (VF )3 , we deduce that if w ∈ H(curl, Ω), then w × n belongs to 1/2 (H00 (F ) )3 , and that Green’s identity (3.10) (w · curl ϕ − ϕ · curl w) dx = w × n, ϕ F ∀ϕ ∈ (VF )3 Ω
holds (where ., . F means the duality pairing between (H00 (F ) )3 and H00 (F )3 ). Now for a fixed face F , we temporarily assume that the z-axis is perpendicular to F and that F is included in the plane z = 0. For ϕ1 , ϕ2 ∈ D(F ), we take ⎛ ⎞ ϕ1 H (x, y, z) = η(z)μ−1 · ⎝ ϕ2 ⎠ 0 1/2
1/2
with a cut-off function η ∈ D(R) such that η(0) = 1 and a sufficiently small support so that H is zero on Γ\F . By construction the function H belongs to XT (Ω, μ)∩(VF )3 . Therefore in (3.9), taking as a test function v = 0 and this function H , by Green’s formula (3.10) we get w × n, H F = 0. As the third component of w × n is zero and the two first components of ⎛ ⎞ ϕ1 μ−1 · ⎝ ϕ2 ⎠ 0 1/2
are arbitrary in H00 (F ), we deduce that w × n = 0 in (H00 (F ) )3 . 1/2
By the definition of w, the requested boundary condition is proved.
660
D. MERCIER AND S. NICAISE
4. Existence and uniqueness results when μ = 0. 4.1. The E-formulation. In the case μ = 0, the second Maxwell equation (2.11) and the boundary condition (2.7) imply that iωE = ∇ϕ,
(4.1)
for some ϕ ∈ H01 (Ω). Therefore, in order to eliminate the vector field H, we take the divergence of the first Maxwell equation (2.10) to get div(iωD) = div J, and by the constitutive equation (2.8) we obtain div(iωeγ(u) + iωE) = div J. As before setting uω = iωu, the above identity is equivalent to div(eγ(uω ) + ∇ϕ) = div J.
(4.2)
This equation is now coupled with (3.3) or, equivalently, with
Div Cγ(uω ) − e ∇ϕ + ω 2 uω = −iωf in Ω (4.3) to form our system of partial differential equations. Clearly, its weak formulation is in finding a solution (uω , ϕ) ∈ H01 (Ω)4 of (4.4)
a((uω , ϕ), (v, ψ)) = F (v, ψ), ∀(v, ψ) ∈ H01 (Ω)4 ,
where the bilinear form a is defined by
a((u, ϕ), (v, ψ)) = (eγ(u) + ∇ϕ) · ∇ψ + (Cγ(u) − e ∇ϕ) · γ(v) − ω 2 u · v dx, Ω
and the linear form F is given by (4.5)
(iωf · v − J · ∇ψ) dx.
F (v, ψ) = Ω
The above considerations show the following lemma. Lemma 4.1. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), then (iωu, ϕ), with ϕ given by (4.1), is a solution of (4.4). As in the previous section, problem (4.4) satisfies the Fredholm alternative; namely, we shall prove the following lemma. Lemma 4.2. There exists a discrete set S0 such that for ω 2 ∈ S0 , the problem (4.4) has a unique solution for any F ∈ H −1 (Ω)4 . Proof. Introduce the bilinear form
b((u, ϕ), (v, ψ)) = (eγ(u) + ∇ϕ) · ∇ψ + (Cγ(u) − e ∇ϕ) · γ(v) dx. Ω
This form is coercive on H01 (Ω)4 since {∇ϕ · ∇ϕ + Cγ(u) · γ(u)} dx ϕ21,Ω + u21,Ω , b((u, ϕ), (u, ϕ)) = Ω
thanks to the positive definiteness of and C and Korn’s inequality. We conclude as in Lemma 3.3.
661
RESULTS FOR PIEZOELECTRIC SYSTEMS
We end this subsection by the converse result of Lemma 4.1. Theorem 4.3. If (uω , ϕ) ∈ H01 (Ω)4 is the solution of (4.4) with F defined by (4.5), then u, E are solutions of (2.9) and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (4.1), and σ, D, B are defined by (2.8) (implying B = 0 since μ = 0). Moreover, there exists H ∈ H 1 (Ω)3 such that (2.10) holds. Proof. The first part of the lemma follows from Green’s formula. For the second part we simply remark that D defined by (2.8) satisfies div(iωD) = div J in Ω, or, equivalently, J − iωD is divergence free. Therefore the existence of H such that curl H = J − iωD follows from Theorem I.3.4 of [10]. In the above lemma, we may notice that the magnetic field H is not unique, but this is not important since, in the case μ = 0, only u and E are of practical interest. 4.2. The H-formulation. Following the arguments of section 3, we may eliminate the electric field E and keep as unknowns u and H. Unfortunately H is no more uniquely determined, since the condition div(μH) = 0 and the boundary condition μH · n = 0 on Γ are trivially satisfied. Therefore as for eddy current problems [3] we impose the following Gauge conditions: (4.6) (4.7)
div H = 0 in Ω, H · n = 0 on Γ.
These conditions may be justified by an asymptotic argument, namely by taking μ = ηI, with η > 0 and letting η tends to 0 (cf. [6]). These arguments imply the following lemma. Lemma 4.4. If u, E, H are solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7), assume that H satisfies the Gauge conditions (4.6)–(4.7). Then (iωu, H) belongs to V = H01 (Ω)3 × XT (Ω, I) and is solution of (3.9), with the sesquilinear form here defined by
−1 ¯ + div(H)div(H ¯ ) a((u, H), (v, H )) = (curl H + eγ(u)) · curl H Ω
+ Cγ(u) + e −1 eγ(u) + e −1 curl H · γ(¯ v ) − ω 2 u · v¯ dx. Again, the above problem enters in the Fredholm alternative. Lemma 4.5. There exists a discrete set S1 such that for 1 + ω 2 ∈ S1 , the problem (3.9), with a and V defined in Lemma 4.4, has a unique solution for any F ∈ V . Proof. Introduce the sesquilinear form b((u, H), (v, H )) := a((u, H), (v, H )) + (1 + ω 2 ) u · v¯ dx. Ω
Using Lemma 3.2, Korn’s inequality, and the compact embedding of XT (Ω, I) into L2 (Ω)3 [28], the bilinear form b is coercive on V . Consequently, the Friedrichs extension B associated with the triple (V, H, b) (with H defined as in the proof of Lemma 3.3) is invertible with a compact inverse. Again denote by A the Friedrichs extension of the triple (V, H, a). Clearly A = B − (1 + ω 2 )K, where the continuous operator K is defined by K(u, H) = (u, 0) .
662
D. MERCIER AND S. NICAISE
Writing (1 + ω 2 )−1 B −1 A = (1 + ω 2 )−1 I − B −1 K, we conclude as in Lemma 3.3, since B −1 K is a compact operator. Similarly to Theorem 3.5, the converse of Lemma 4.4 holds under some conditions on ω. eu Theorem 4.6. Assume that ω 2 is not an eigenvalue of −ΔN . If (uω , H) ∈ V I is a solution of (3.9) with F defined by (3.8) and a, V defined in Lemma 4.4, then divH = 0 and u, E, H are solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) when u = uiωω , E is given by (3.1), and σ, D, B = 0 are defined by (2.8). 4.3. Equivalence between the E-formulation and the H-formulation. The goal of this section is to show that the displacement and electric fields obtained by the E-formulation and the H-formulation are identical. Theorem 4.7. Let u1 , E1 , H1 (resp., u2 , E2 , H2 ) be the solutions of (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by the E-formulation (resp., H-formulation). If ω 2 ∈ S0 (cf. Lemma 4.2), then u1 = u2 and E1 = E2 . Proof. Denote u = u1 − u2 and E = E1 − E2 . Since u1 , E1 and u2 , E2 satisfy (2.9) and (2.11), u, E satisfy Divσ + ω 2 u = 0 in Ω, curl H + iωD = 0 in Ω, curl E = 0 in Ω, where σ = σ1 − σ2 , andD = D1 − D2 are given by (2.8). Therefore E = ∇χ, with χ ∈ H01 (Ω) and the pair (u, χ) ∈ H01 (Ω)4 is a solution of (4.4) with F = 0. By Lemma 4.2, we conclude that u = 0 and χ = 0. Remark 4.8. The magnetic fields obtained by the E-formulation and the Hformulation cannot be identical, since the one obtained by the E-formulation does not necessarily satisfy the Gauge conditions. 5. Singularities of some elliptic systems. It is well known that the singularities of elliptic systems in Ω are produced by the corners and edges of Ω. Here we eu and of the 4×4 briefly describe the corner and edge singularities of the operator ΔN μ system (4.2)–(4.3) with Dirichlet boundary conditions. For the sake of brevity we restrict ourselves to a minimal description and refer the reader to the pioneer work [15] or standard books [11, 7, 18] for more details. 5.1. Corner singularities. Fix a corner c ∈ C of Ω and denote by (ρc , ϑc ) the spherical coordinates centered at c. Denote by Γc the infinite polyhedral cone which coincides with Ω near c. Let Gc be the intersection of Γc with the unit sphere centered at c. For any λ ∈ C, let us set Q (log ρc )q ψq (ϑc ) : ψq ∈ H 1 (Gc ) . S λ (Γc ) = ψ(x) = ρλc q=0 eu (Γc ) ΛN μ
eu of corner singular exponents of the operator ΔN in Γc is the The set μ λ set of λ ∈ C such that there exists a nonpolynomial solution Ψ ∈ S (Γc ) of
(5.1) (5.2)
Δμ Ψ = div(μ∇Ψ) = 0 in Γc , μ∇Ψ · n = 0 on ∂Γc .
663
RESULTS FOR PIEZOELECTRIC SYSTEMS
λ We denote by ZN eu (Γc , μ) the space of these solutions. Similarly for the 4 × 4 system (4.2)–(4.3) with Dirichlet boundary conditions, its set ΛDir C,,e (Γc ) of corner singular exponents is the set of λ ∈ C such that there exists a nonpolynomial solution (u, χ) ∈ S λ (Γc )4 of
⎧ ⎨ (5.3)
div(∇χ + eγ(u)) = 0
Div Cγ(u) − e ∇χ = 0 χ = 0, u = 0
⎩
in Γc , in Γc , on ∂Γc .
λ The space of these solutions is denoted by ZDir (Γc , C, , e).
5.2. Edge singularities. Fix an edge a ∈ A of Ω and denote by Γa × R the infinite polyhedral cone which coincides with Ω near a (Γa is then a two-dimensional sector). Denote by (ra , θa , za ) the cylindrical coordinates along a. Let Ga be the intersection of Γa with the unit sphere ra = 1. As before, for any λ ∈ C let us set λ
S (Γa ) =
ψ(x) =
raλ
Q
(log ra ) ψq (θa ) : ψq ∈ H (Ga ) . q
1
q=0 eu eu (Γa ) of corner singular exponents of the operator ΔN in Γa is the The set ΛN μ μ set of λ ∈ C such that there exists a nonpolynomial solution Ψ ∈ S λ (Γa ) of
div2 (μ2 ∇2 Ψ) = 0 in Γa , μ2 ∇2 Ψ · n = 0 on ∂Γa , where div2 (resp., ∇2 ) means the two-dimensional divergence (resp., gradient) and μ2 is the 2 × 2 matrix obtained from μ by dropping the third line and the third column of μ. λ We denote by ZN eu (Γa , μ) the space of these solutions. In the same way, the set ΛDir C,,e (Γa ) of edge singular exponents of the 4 × 4 system (4.2)–(4.3) with Dirichlet boundary conditions is the set of λ ∈ C such that there exists a nonpolynomial solution (u, χ) ∈ S λ (Γa )4 of ⎧ ⎪ ⎨ (5.4)
⎪ ⎩
˜ ∇χ ˜ + e˜ div( γ (u)) = 0 ˜ ˜ Div C γ˜ (u) − e ∇χ = 0 χ = 0, u = 0
in Γa , in Γa , on ∂Γa ,
where the sign ˜ means that all derivatives in the za -variable are replaced by zero. λ (Γa , C, , e). The space of these solutions is denoted by ZDir 6. Corner and edge singularities when μ is positive definite. Our goal is to describe the corner and edge singularities of the (regularized) problem (3.9). These singularities are obtained using some ideas from [4, 5, 6]. Let us start with the corner singularities. 6.1. Corner singularities. Fix a corner c ∈ C of Ω. We drop the index c in the above considerations. As usual we are looking for solutions of the homogeneous
664
D. MERCIER AND S. NICAISE
problem in the space λ 1 1 (Γ)3 × XT,loc (Γ, μ) :div(μH) ∈ Hloc (Γ), ST (Γ, μ) = (u, H) ∈ H0,loc u(x) = ρλ
Q
(log ρ)q Uq (ϑ),
q=0
H(x) = ρλ
Q
(log ρ)q Hq (ϑ) ,
q=0
the index loc meaning that the properties hold in all bounded domains far from c. This means that we look for a nonpolynomial solution (u, H) ∈ SλT (Γ, μ) of ⎧
curl −1 (curl H + eγ(u)) − μ∇ div(μH) = 0 in Γ, ⎪ ⎪ ⎪ ⎪ Div Cγ(u) + e −1 eγ(u) + e −1 curl H = 0 in Γ, ⎪ ⎪ ⎨ u=0 on ∂Γ, (6.1) μH · n = 0 on ∂Γ, ⎪ ⎪ ⎪ ⎪ μ∇(div(μH)) · n = 0 on ∂Γ, ⎪ ⎪ ⎩ −1 (curl H + eγ(u)) × n = 0 on ∂Γ. If a nontrivial solution exists, then we say that λ is a corner exponent. Inspired by [4, 5, 6], this problem is split up into three subproblems by introducing the auxiliary unknowns ψ = −−1 (curl H + eγ(u)) and q = div(μH). With this notation, problem (6.1) is equivalent to looking for q, (ψ, u), H successive solutions of div(μ∇q) = 0 in Γ, (6.2) μ∇q · n = 0 on ∂Γ, ⎧ ⎪ ⎪ ⎨ (6.3)
⎪ ⎪ ⎩ ⎧ ⎨
(6.4)
⎩
curl ψ = −μ∇q div(ψ
+ eγ(u)) = 0 Div Cγ(u) − e ψ = 0 ψ × n = 0, u = 0
in Γ, in Γ, in Γ, on ∂Γ,
curl H = −ψ − eγ(u) div(μH) = q μH · n = 0
in Γ, in Γ, on ∂Γ.
This means that three types of singularities may appear: Type 1: q = 0, ψ = u = 0, and H is a general solution of (6.4). Type 2: q = 0, (ψ, u) is a general solution of (6.3), and H is a particular solution of (6.4).
RESULTS FOR PIEZOELECTRIC SYSTEMS
665
Type 3: q is a general solution of (6.2), (ψ, u) is a particular solution of (6.3), and H is a particular solution of (6.4). These three types of singularities may be described with the help of the corner eu and of the 4 × 4 elliptic system (4.2)–(4.3) with singularities of the operator ΔN μ Dirichlet boundary conditions described in subsection 4.1. Since for our problem (3.9) div(μH) = 0 and is then regular, we do not describe the singularities of Type 3 because they cannot occur for any solution of (3.9). The singularities of Type 1 are obtained exactly as in Lemma 5.1 of [5] since only the magnetic field H is involved. Lemma 6.1. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is a singularity of λ+1 eu type 1 if and only if λ + 1 belongs to ΛN (Γ) and H = ∇Φ, with Φ ∈ ZN μ eu (Γ, μ). The situation is different for singularity of Type 2 since the coupling between the elasticity system and Maxwell’s equations appear via problem (6.3). Lemma 6.2. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is a singularity of λ type 2 if and only if λ belongs to ΛDir C,,e (Γ), ψ = ∇χ, with (u, χ) ∈ ZDir (Γ, C, , e), and H is given by (6.5)
H=
1 (a × x + ∇r) , λ+1
where a = −(ψ + eγ(u)) and r ∈ S λ+1 (Γ) is solution of div(μ∇r) = − div(μ(a × x)) in Γ, (6.6) μ∇r · n = −μ(a × x) · n on ∂Γ. Proof. As curl ψ = 0 in Γ, there exists χ ∈ S λ (Γ) such that ψ = ∇χ in Γ. From (6.3) we deduce that ⎧ + eγ(u)) = 0 ⎨ div(∇χ
Div Cγ(u) − e ∇χ = 0 ⎩ χ = 0, u = 0
in Γ, in Γ, on ∂Γ.
λ (Γ, C, , e). In view of section 5, we deduce that (u, χ) ∈ ZDir Now we readily check that H in the form (6.5) is a solution of (6.4) if and only if r is a solution of (6.6), whose solution exists by Theorem 4.14 of [23] and because curl(a × x) = (λ + 1)a. Lemma 6.3. There is no corner singularity of Type 2 in the strip λ ∈ ]−1, 0]. Proof. By Theorem 1 of [16] (see Remark 2 of [16], or Theorem 2 of [17]), the set ΛDir C,,e (Γ) ∩ [−1, 0] is empty. We conclude by Lemma 6.2. Since the singularities of our problem (3.9) have to be locally in XT (Γ, μ), among the singular exponents described above we select the subset Λc of λ > − 32 such that there exists (u, H) ∈ SλT (Γ, μ) solution of (6.1) such that
χ(u, H) ∈ XT (Γ, μ),
666
D. MERCIER AND S. NICAISE
where χ is a cut-off function equal to 1 near c. This last condition implies the following constraints for our first two types of singularities (see [5, 6]): eu eu Type 1: λ + 1 ∈ ΛN (Γ) and since ΛN (Γ) ∩ [−1, 0] is empty, by Lemma 5.4 μ μ of [5] we get the condition λ > −1. 1 1 3 Type 2: λ ∈ ΛDir C,,e (Γ) and by the condition χu ∈ H (Γ) , we get λ > − 2 . By Lemma 6.3, we arrive at the condition λ > 0. In conclusion, the set of corner singular exponents is eu Λc = {λ > −1 : λ + 1 ∈ ΛN (Γc )} ∪ {λ > 0 : λ ∈ ΛDir μ C,,e (Γc )}.
6.2. Edge singularities. Fix an edge a ∈ A of Ω and drop the index a. As before, we are looking for solutions of the homogeneous piezoelectric system (6.1) in Γ × R. For the elasticity system or Maxwell’s equations, the system obtained in the Cartesian coordinates (x, y, z) (according to the notation introduced in section 5, the z-axis contains the edge a) is split up into two independent problems in Γa . Here the coupling between these systems prevents this splitting. Therefore we simply follow the approach of the previous subsection. Namely, we search nonpolynomial solutions (u, H) of (6.1) independent of the z-variable and in the space λ 1 1 ST (Γ, μ) = (u, H) ∈ H0,loc (Γ × R)3 × XT,loc (Γ × R, μ) : div(μH) ∈ Hloc (Γ × R), u(x, y, z) = rλ
Q
(log r)q Uq (θ),
q=0
H(x, y, z) = rλ
Q
(log r)q Hq (θ) ;
q=0
the index loc here means that the properties hold in all bounded domains far from a. Then we introduce the auxiliary unknowns ψ and q defined as before but here independent of the variable z. This leads to the successive problems (6.2), (6.3), and (6.4) (where all derivatives in z are equal to zero). As before, singularities of Types 1, 2, and 3 appear. The singularities of Type 3 are not studied for the same reason as before. We now describe the singularities of Type 1 (compare with Lemma 6.1). Lemma 6.4. Assume that λ = −1. Then (u, H) ∈ SλT (Γ, μ) is an edge singularity eu of type 1 if and only if λ + 1 belongs to ΛN (Γ), u = 0 and μ H = (∇2 Φ, 0) , λ+1 with Φ ∈ ZN eu (Γ, μ). Proof. As
⎞ ∂2 H3 ⎠ = 0, −∂1 H3 curl H = ⎝ ∂1 H2 − ∂2 H1 ⎛
the third component H3 is constant and the field H1 h= H2
RESULTS FOR PIEZOELECTRIC SYSTEMS
667
has a two-dimensional zero curl. Therefore there exists Φ such that h = ∇2 Φ, and we conclude as in Lemma 6.1. Let us continue with singularities of Type 2. Lemma 6.5. Assume that λ = −1 and λ = 0. Then (u, H) ∈ SλT (Γ, μ) is an edge singularity of Type 2 if and only if λ belongs to ΛDir C,,e (Γ), ψ = (∇2 χ, 0) , with λ (Γ, C, , e) and H given by (u, χ) ∈ ZDir ⎞ ⎛ a3 x2 (6.7)
H=⎝
λ a3 x1 λ a1 x2 −a2 x1 λ+1
⎠ + (∇2 r, 0) ,
where a = −(ψ + eγ(u)) and r ∈ S λ+1 (Γ) is the solution of ⎧ a3 x2 ⎪ λ ⎪ in Γ, ⎨ div2 (μ∇2 r) = − div2 μ a3 x1 λ (6.8) a3 x2 ⎪ ⎪ λ ⎩ ·n on ∂Γ. μ∇r · n = −μ a3 x1 λ
Proof. As in the previous lemma, since curl ψ = 0 in Γ, there exists χ ∈ S λ (Γ) such that ψ = (∇2 χ, 0) in Γ. From (6.3) we deduce that ⎧ + eγ(u)) = 0 ⎨ div(∇χ
Div Cγ(u) − e ∇χ = 0 ⎩ χ = 0, u = 0
in Γ, in Γ, on ∂Γ.
λ This means that (u, χ) ∈ ZDir (Γ, C, , e) owing to section 5 (recalling that u and χ are independent of the z-variable). As in Lemma 6.2, we easily check that H in the form (6.7) is a solution of (6.4) if and only if r is a solution of (6.8), whose existence follows from Theorem 4.14 of [23]. In summary we may formulate the following corollary. Corollary 6.6. The set Λa of edge exponents associated with a is given by eu Λa = {λ > −1 : λ + 1 ∈ ΛN (Γa )} ∪ {λ > 0 : λ ∈ ΛDir μ C,,e (Γa )}.
7. Regularity results. In this section, we describe regularity results of a solution u, H, E of our time-harmonic piezoelectric system. These results use its weak formulation obtained above, are based on the knowledge of corner and edge singularities of these formulation described in the previous section, and rely on the application of Mellin’s techniques as in [4, 5]. For the sake of brevity, we do not describe singular decomposition of such a solution. Using the techniques from [4, 5], for sufficiently smooth data we may obtain a decomposition of u, H, E into a regular part and a singular one. 7.1. The case μ positive definite. Before stating our regularity results, let us introduce the following notation: For any corner c ∈ C introduce λc,u = min{λ > 0 : λ ∈ ΛDir C,,e (Γc )}, λc,H = min{λ : λ ∈ Λc }.
668
D. MERCIER AND S. NICAISE
Similarly, for any edge a ∈ A define λa,u = min{λ > 0 : λ ∈ ΛDir C,,e (Γa )}, λa,H = min{λ : λ ∈ Λa }. Finally, we set 1 τu := min min λa,u , + min λc,u , a∈A 2 c∈C 1 τH := min min λa,H , + min λc,H . a∈A 2 c∈C
We remark that τu does not depend on μ. Theorem 7.1. Let (uω , H) ∈ H01 (Ω)3 × XT (Ω, μ) be a solution of problem (3.9), with J and f smooth enough. Then we have (7.1) (7.2)
uω ∈ H 1+τ (Ω)3 ∀τ < τu , H ∈ H τ (Ω)3 ∀τ < τH .
Proof. Mellin’s techniques directly imply that (uω , H) ∈ H τ (Ω)6 ∀τ < τH . The extra regularity for uω follows from the description of the singularities of problem (3.9) made in section 6, where we see that the regularity of uω is only determined by the singularity of type 2. Corollary 7.2. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 3.5. Then u has the regularity (7.1), H has the regularity (7.2), and E satisfies (7.3)
E ∈ H τ (Ω)3 ∀τ < τu .
Proof. The previous lemma directly yields the regularity for u and H. By (3.1), E has the regularity of curl H + iωeγ(u). Since the singularities of type 1 for H are gradient, the regularity of E is only determined by the singularities of type 2. 7.2. The case μ = 0. Here, to solve our piezoelectric system we may use either the E-formulation or the H-formulation. Both formulations give the same regularity for u and E, which is quite natural since generically they are identical (see Theorem 4.7). Theorem 7.3. Let (uω , ϕ) ∈ H01 (Ω)4 be a solution of problem (4.4) with J and f smooth enough; then we have (7.4)
(uω , ϕ) ∈ H 1+τ (Ω)4 ∀τ < τu ,
where τu is defined as before. Proof. Since the system associated with (4.4) is the strongly elliptic system (4.2)–(4.3) with Dirichlet boundary conditions, the mentioned regularity result follows from Theorem 5.11 of [7]. Corollary 7.4. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 4.3. Then u has the regularity (7.1) and E satisfies (7.3).
RESULTS FOR PIEZOELECTRIC SYSTEMS
669
1 ∇ϕ. Proof. This is a direct consequence of the previous theorem since E = iω For the H-formulation, the arguments of the previous subsection directly give the next results. Theorem 7.5. Let (uω , H) ∈ H01 (Ω)3 × XT (Ω, I) be a solution of problem (3.9) in the sense of Lemma 4.4 with J and f smooth enough. Then the regularity results (7.1)–(7.2) hold, τH being defined as before but with μ = I. Corollary 7.6. Assume that J and f are smooth. Let u, E, H be solutions of (2.8), (2.9), (2.10), and (2.11) with the boundary conditions (2.6) and (2.7) obtained by Theorem 4.6. Then u has the regularity (7.1), H has the regularity (7.2), and E satisfies (7.3).
8. Numerical examples of edge singular exponents. To illustrate our theoretical results, we give some edge singular exponents of our piezoelectric system corresponding to some illustrative materials when c45 = c16 = c26 = c36 = 0, c11 = c22 , 12 , 11 = 22 , and 12 = 0. We further consider a dihedral cone c44 = c55 , c66 = c11 −c 2 Γa × R such that the edge a of this cone (corresponding to the axis z = 0) is parallel to the poling direction x3 . In that case, the system (5.4) reduces to the following 4 × 4 system (see, for instance, [19]): 12 12 c11 ∂12 + c11 −c u1 0 ∂22 c11 +c ∂1 ∂2 2 2 = in Γa , c11 −c12 2 c11 +c12 0 u2 ∂1 ∂2 ∂1 + c11 ∂22 2 2 c44 Δ e15 Δ 0 u3 = in Γa , −e15 Δ 11 Δ χ 0 χ = 0, u = 0 on ∂Γa , where, as usual, Δ = ∂12 + ∂22 . Setting v = c44 u3 + e15 χ and w = −e15 u3 + 11 χ, this system is equivalent to 12 12 c11 ∂12 + c11 −c u1 0 ∂22 c11 +c ∂1 ∂2 2 2 = in Γa , c11 −c12 2 c11 +c12 0 u2 ∂1 ∂2 ∂1 + c11 ∂22 2 2 Δv = 0 in Γa , Δw = 0 in Γa , u1 = u2 = v = w = 0 on ∂Γa . This last system is decoupled into a 2 × 2 system of the isotropic elasticity for the 12 ) with Dirichlet pair (u1 , u2 ) (with Lam´e coefficients given by λ = c12 and μ = c11 −c 2 boundary conditions and two Dirichlet problems for v and w. For this last Dirichlet problem, it is well known that the singular exponents are given by lπ ω , with l ∈ N, when ω is the interior opening of Γa . On the other hand, the singular exponents of the isotropic elasticity are also well known and are the set of λ ∈ C such that (see, for instance, [12]) k 2 sin(λω)2 = λ2 sin(ω)2 , 11 −c12 with k = 3c c11 +c12 . The zeros of this equation can be easily computed using a Newton method. Figures 8.1 and 8.2 show the real part of the singular exponents in the strip λ ∈ (0, 2) for all values of ω ∈ (0, 2π] for the piezoelectric materials P ZT -4 and BaT iO3 in black lines. In these figures, the gray lines correspond to the curves lπ ω , for l = 1, 2, 3. From these figures, we may conclude that for these materials the strongest singular
670
D. MERCIER AND S. NICAISE 2
Re ( λ )
1.5
1
0.5
0 0
50
100
150
200
250
300
350
300
350
ω
Fig. 8.1. The edge exponents for the P ZT -4. 2
Re ( λ )
1.5
1
0.5
0 0
50
100
150
200
250
ω
Fig. 8.2. The edge exponents for the BaT iO3 .
exponent λ0 (strongest in the sense that λ0 is minimal) is always the one coming from the 2 × 2 system of elasticity and that λ0 > 1/2 (see Theorem 2.2 of [22]). This last property implies the H 3/2 -regularity (resp., H 1/2 -regularity) for u (resp., for E) along the edge. Note further that the curves obtained for the P ZT -4 are similar to the ones shown in Figure 4b of [29] (in that paper, the singular exponents are obtained using an extension of Lekhnitskii’s representation for elastic solids). Remark 8.1. If the edge is not parallel to the poling direction, then the above decoupling phenomenon does not appear. The calculation of the edge singular expo-
RESULTS FOR PIEZOELECTRIC SYSTEMS
671
nents is then more complicated and will be done using a finite element method. This will be done in a forthcoming paper [20]. 9. Conclusions. We have investigated a general time-harmonic piezoelectric system, which contains as special cases standard models like ceramics. We developed the appropriate formalism in order to get existence and uniqueness results of weak solutions in the case when the magnetic permeability matrix is positive definite (case of the BaT iO3 ) and the case when the magnetic permeability matrix is equal to zero (case of the P ZT ). In the latter case, we gave two different formulations: the E-formulation and the H-formulation. For this second one, Gauge conditions are introduced as for eddy current problem. We further show that generically these two formulations yield the same solutions. We described the corner and edge singularities of our general system and deduced some regularity results. Some edge singular exponents were given in order to illustrate our theoretical results. Acknowledgments. We thank C. Courtois and A. Leriche from the laboratory LMP (UVHC) for valuable discussions about these topics. REFERENCES [1] C. Amrouche, C. Bernardi, M. Dauge, and V. Girault, Vector potentials in threedimensional nonsmooth domains, Math. Methods Applied Sci., 21 (1998), pp. 823–864. [2] D. A. Berlincourt, D. R. Curran, and H. Jaffe, Piezoelectric and piezomagnetic materials and their function in transducers, Physical Acoustics, 1 (1964), pp. 169–270. [3] A. Bossavit, Electromagn´ etisme en vue de la mod´ elisation, Springer-Verlag, Paris, 1993. [4] M. Costabel and M. Dauge, Singularities of electromagnetic fields in polyhedral domains, Arch. Ration. Mech. Anal., 151 (2000), pp. 221–276. [5] M. Costabel, M. Dauge, and S. Nicaise, Singularities of Maxwell interface problems, RAIRO Mod´ el. Math. Anal. Num´er., 33 (1999), pp. 627–649. [6] M. Costabel, M. Dauge, and S. Nicaise, Singularities of eddy current problems, RAIRO Mod´ el. Math. Anal. Num´er., 37 (2003), pp. 807–831. [7] M. Dauge, Elliptic Boundary Value Problems in Corner Domains. Smoothness and Asymptotics of Solutions, L. N. in Math. 1341, Springer-Verlag, Berlin, 1988. [8] A. C. Eringen and G. A. Maugin, Electrodynamics of Continua, Vols. 1, 2, Springer-Verlag, New York, 1990. ¨ndig, and G. Mishuris, Piezoelectricity in Multi-layer Actuators: Modelling [9] W. Geis, A.-M. Sa and Analysis in Two and Three-Dimensions, Preprint IANS, 2003/23, Univ. Stuttgart, Germany, 2003. [10] V. Girault and P.-A. Raviart, Finite Element Methods for Navier-Stokes Equations, Springer Ser. Comput. Math. 5, Springer-Verlag, New York, 1986. [11] P. Grisvard, Elliptic Problems in Nonsmooth Domains, Monogr. Stud. Math. 24, Pitman, Boston, 1985. [12] P. Grisvard, Singularit´ e en elasticit´ e, Archiv. Ration. Mech. Anal., 107 (1989), pp. 157–180. [13] D. Iesan, Plane strain problems in piezoelectricity, Internat. J. Engrg. Sci., 25 (1987), pp. 1511– 1523. [14] T. Ikeda, Fundamentals of Piezoelectricity, Oxford University Press, Oxford, 1996. [15] V. A. Kondrat’ev, Boundary-value problems for elliptic equations in domains with conical or angular points, Trans. Moscow Math. Soc., 16 (1967), pp. 227–313. [16] V. A. Kozlov and V. G. Maz’ya, Spectral properties of the operator bundles generated by elliptic boundary value problems in a cone, Funct. Anal. Appl., 22 (1988), pp. 38–46. [17] V. A. Kozlov and V. G. Maz’ya, On the behaviour of the spectrum of parameter-depending operators under small variation of the domain and an application to operator pencils generated by boundary value problems in a cone, Math. Nachr., 153 (1991), pp. 123–129. [18] V. A. Kozlov, V. G. Maz’ya, and J. Rossman, Elliptic Boundary Value Problems in Domains with Point Singularities, Math. Surveys Monogr. 52, AMS, Providence, RI, 1997. [19] L. Marin, The Behaviour of Piezoelectric Materials under Mechanical and Electrical Loadings, Master Thesis, Univ. Kaiserslautern, Germany, 1998.
672
D. MERCIER AND S. NICAISE
[20] D. Mercier and S. Nicaise, Regularity Results for a Two-Dimensional Piezoelectric System and Crack Propagation Criteria, manuscript. [21] S. Nicaise, Exact controllability of a pluridimensional coupled problem, Rev. Mat. Complut., 5 (1992), pp. 91–135. [22] S. Nicaise, About the Lam´ e system in a polygonal or a polyhedral domain and a coupled problem between the Lam´ e system and the plate equation I: Regularity of the solutions, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 19 (1992), pp. 327–361. [23] S. Nicaise, Polygonal Interface Problems, Peter Lang, Berlin, 1993. [24] W. Novacki, Mathematical models of phenomenological piezoelectricity, in Mathematical Models and Methods in Mechanics, Banach Center Publ. 15, Polish Academy of Sciences, Warsaw, Poland, 1985, pp. 593–607. [25] Y. E. Pak, Linear electro-elastic fracture mechanics of piezoelectric materials, Internat. J. Fracture, 54 (1992), pp. 79–100. [26] E. Pan and P. R. Heyliger, Exact solutions for magneto-electro-elastic laminates in cylindrical bending, Internat. J. Solids Structures, 40 (2003), pp. 6859–6876. [27] H. A. Sosa and Y. E. Pak, Three-dimensional eigenfunction analysis of a crack in a piezoelectric material, Internat. J. Solids Structures, 26 (1990), pp. 1–15. [28] Ch. Weber, A local compactness theorem for Maxwell’s equations, Math. Methods Appl. Sci., 2 (1980), pp. 12–25. [29] X.-L. Xu and R. Rajapakse, On singularities in composite piezoelectric wedges and junctions, Internat. J. Solids Structures, 37 (2000), pp. 3253–3275. [30] X.-L. Xu and R. Rajapakse, On a plane crack in piezoelectric solids, Internat. J. Solids Structures, 38 (2001), pp. 7643–7658.
