J. Math. Pures Appl.

89 (2008) 2548
www.elsevier.com/locate/matpur

Elliptic model problems including mixed boundary conditions

and material heterogeneities
Robert Haller-Dintelmann , Hans-Christoph Kaiser, Joachim Rehberg
Weierstrass Institute for Applied Analysis and Stochastics, Mohrenstr. 39, 10117 Berlin, Germany
Received 5 July 2007
Available online 25 September 2007

Abstract
1,p

We present model problems in three dimensions, where the operator maps the Sobolev space W () isomorphically
1,p

() for a p > 3. The emphasis is here on the case where different boundary conditions meet material heterogeneities.
onto W
2007 Elsevier Masson SAS. All rights reserved.
Rsum
Cet article prsente des situations modles, en trois dimensions, dans lesquelles loprateur est un isomorphisme de
1,p
1,p
W () sur W
() pour un p > 3. On sintresse notamment au cas o des conditions au bord mixtes Dirichlet/Neumann
sont combines avec des sauts du coefficient .
2007 Elsevier Masson SAS. All rights reserved.
MSC: 35B65; 35J25; 35R05
Keywords: Elliptic transmission problems; Mixed boundary problems; W 1,p regularity

1. Introduction
Many elliptic problems originating from science, engineering, and technology exhibit mixed boundary conditions
and non-smooth material parameters, see [1,41] and the references cited there. For instance, in the simulation of
operation and fabrication of semiconductor devices one is regularly confronted with heterogeneous materials in the
volume and on the boundary (contacts), see [46], while dealing with elliptic and parabolic equations as mathematical
models, see [17]. However, not much is known concerning maximal regularity for elliptic operators which include
mixed boundary conditions. Moreover, most of this is restricted to Hilbert space scales, see e.g. [44,43,25,12,8,7].
Unfortunately, the Hilbert space H 3/2 is a principle threshold for mixed elliptic, second order problems at least in the
case when the Dirichlet and Neumann boundary part meet on smooth parts of the boundary, see [48] and also [44].
Thus, within this scale one cannot expect an embedding of the domains of these operators in L (or even in C ) in
* Corresponding author.

E-mail addresses: haller@mathematik.tu-darmstadt.de (R. Haller-Dintelmann), kaiser@wias-berlin.de (H.-C. Kaiser),

rehberg@wias-berlin.de (J. Rehberg).
0021-7824/$ see front matter 2007 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.matpur.2007.09.001

26

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

case of three or more space dimensions. But exactly this is desirable in view of nonlinear, in particular quasilinear
problems, see [42,26,39].
Concerning optimal regularity in non-Hilbert spaces there are the results of [47,48,5,22,11,20,21]; for the pure
Dirichlet or pure Neumann case see [30] and [55], respectively. Grger proved in [22] that under only L (and ellipticity) assumptions on the coefficient function , the Lipschitz property of the domain and very weak assumptions
on the Dirichlet boundary part \ the operator,
1,p

1,p

: W () W

(1)

(),

1,p

is a topological isomorphism for a certain p > 2 (W () denoting the subspace of W 1,p () including a trace
1,p 

1,p

zero condition on the Dirichlet boundary part \ , and W () the dual of W ()). This result has found
numerous applications within the treatment of applied problems. Nevertheless, it is well known that under these
general assumptions one can only expect that p exceeds 2 arbitrarily little. This is the reason why the applications
of [22] remained restricted to two-dimensional problems. Because the demand for three-dimensional modelling and
simulation steadily increases, the question arises under which assumptions the isomorphism property of (1) can be
obtained for a p > 3 and, in particular, whether this is true with mixed boundary conditions. Dauge proved in [11]
that if the domain is a convex polyhedron and the border between Dirichlet and Neumann boundary part consists
1,p
1,p
of (finitely many) line segments, then the Laplacian provides a topological isomorphism between W and W
for some p > 3. In this paper we generalise this to prototypical situations where mixed boundary conditions and
heterogeneous, anisotropic coefficient functions occur simultaneously. Thus, this calculus allows for jumps in the
conormal derivative of solutions across internal interfaces. This means, e.g. in electrostatics, that the jump in the
normal component of the displacement + across a prescribed interface equals the surface charge
density on the interface, and this surface charge density is represented by a distribution on the underlying domain .
In view of an adequate localisation principle, see [22], the geometric constellations we investigate may be viewed
as local constituents of rather complex global settings.
Since the knowledge of the singularity of solutions is crucial for the efficiency of numerical methods, there exist of
course several numerical approaches to determine singular exponents of concrete anisotropic problems, see [35,10,51]
and the references therein. For a more general numerical approach to heterogeneous elliptic problems see for instance
[2,27,9,53] and the references cited there.
In detail, our results are as follows:
Theorem 1. Let R2 be an open triangle, let further P be the center of one of its sides and the open leg
def

between P and one of its neighbouring vertices. Define = ]1, 1[ and the boundary part as ]1, 1[.
Suppose to be a plane within R3 that intersects {P }]1, 1[ in exactly one point. Assume that the elliptic coefficient
function takes its values in the set of real, symmetric, positive definite 3 3 matrices and is constant on both
components of \ . Then there is a p > 3 such that
1,p

1,p

: W () W

()

(2)

is a topological isomorphism.
Theorem 2. Let R2 be an open triangle, be one of its open sides or \ one of its closed sides. Define
def

= ]1, 1[ and the boundary part as ]1, 1[. Let further be a plane the intersection of which with
the boundary of consists of exactly two points. Assume that the elliptic coefficient function takes its values in the
set of real, symmetric, positive definite 3 3 matrices and is constant on both components of \ . Then there is a
p > 3 such that (2) is a topological isomorphism.
Corollary 3. Let , and be as in Theorem 2. Let be ]1, 1[ combined with the ground plate or/and the
upper plate. Let further be a plane as in Theorem 2 which does neither touch the upper/lower plate and let be as
in Theorem 2. Then the conclusion of Theorem 2 also holds.
Remark 4. The supposition that the plane has only a finite intersection with edges where the Dirichlet boundary part
meets with the complementing boundary part is crucial. If this is not the case, a bimaterial outer edge (see Definition 10

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

27

Fig. 1. The model domains under consideration in Theorem 1 (left) and Theorem 2 (right) with generating triangle , Neumann boundary
(hatched area) and material interface (shaded area) cutting the domain.

below) with mixed boundary conditions occurs, for which the appearing singularities may be arbitrarily large, see
Remark 28 in Appendix A.
Remark 5. Let us further mention that in Theorem 1 can be taken as Grgers third model set, see [22],
and thus Theorem 1 can be viewed as a regularity assertion for Grgers third model constellation if the coefficient
function has a discontinuity along a plane.
Operators of type (1)which may be seen as the principal part of the (Dirichlet)-homogenization of an elliptic operatorare of fundamental significance in many application areas. This is the case not only in mechanics
(see [35, Ch. IV/V]), thermodynamics (see [50]), and electrodynamics (see [49]) of heterogeneous media, but also in
mining, multiphase flow, mathematical biology (see [16,6]), and semiconductor device simulation (see [46,17,19]), in
particular quantum electronics (see [54,4,33,52,53,36]).
The nonhomogeneous coefficient function represents varying material properties as the context requires. It may
be thermal conductivity in a heat equation (see [50, 21]), or dielectric permittivity in a Poisson equation, or diffusivity
in a transport equation (see for instance [46, 2.2] for carrier continuity equations), or effective electron mass in a
Schrdinger equation (see [33]).
Let us emphasise that the matrices which constitute the coefficient function may be not diagonal and, in particular, not multiples of the identity, see [1] and [35, Ch. IV/V]. This is motivated by the applications, for instance
in heat conduction, see [50, 21.B]. On the other hand anisotropic coefficients are absolutely necessary in view of
(local) deformation and transformation of the domain in the localisation procedure, see Proposition 16. It should be
noted that in case of an essentially anisotropic coefficient matrix the generic properties of the elliptic operator differ
dramatically from the case of a scalar coefficient, see [13, Remark 5.1], [14, 4], and [45, Ch. 5].
The outline of the paper is as follows: in the next section we will introduce some notation. In Section 3 the strategy
of proof is explained. Section 4 contains some preliminaries which establish the connection between the regularity
of the solution and the edge singularities. In Section 5 we collect some auxiliary results which justify at the end the
transformation of the problem to a Dirichlet one. Section 6 is devoted to the core of the proof of Theorem 1, essentially
based on the discussion of the edge singularities. In Section 7 we give the proofs of Theorem 2 and of Corollary 3.
Some concluding remarks are given in Section 8. Appendix A finishes the paper by establishing the required estimates
for the occurring singularities for geometric edges and bimaterial outer edges.
2. Notation
Throughout this paper Rd always denotes a bounded Lipschitz domain (see [24] for the definition) and
is an open part of its boundary. W 1,p () denotes the (complex) Sobolev space on consisting of those

28

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

Lp () functions whose first order distributional derivatives also belong to Lp () (see [24] or [37]). Note that
enjoys the extension property for W 1,p () in view of being a bounded Lipschitz domain, see [18, Thm. 3.10] or [37,
Ch. 1.1.16]. Thus, W 1,p () is identical with the completion of the set {v| : v C (R3 )} with respect to the norm
def 
1,p

v
W 1,p = ( (|v|p + |v|p ) dx)1/p . We use the symbol W () for the closure of,


v| : v C (R3 ), supp v ( \ ) = ,
1,p

1,p

1,p 

in W 1,p (). If = we write as usual W0 () instead of W (). W

W

1,p 

1,p

() denotes the dual to W0 (), when

1,p
W ,

1,p
W0

1
p

1
p

1,p

() denotes the dual to W () and

= 1 holds. If is understood, then we sometimes abbreviate

W 1,p ,

and
respectively. ,X always indicates the duality between a Banach space X and its dual; in
case of X = Cd we mostly write ,. If is a Lebesgue measurable, essentially bounded function on taking its
values in the set of real, symmetric d d matrices, then we define : W1,2 () W1,2 () by:

def
v, wW 1,2 = v, w dx; v, w W1,2 ().
(3)

1,p

The maximal restriction of to any of the spaces W () (p > 2) we will denote by the same symbol.
Finally, we define for any two complex numbers , :


def
= exp log | | + i arg , arg ], ];
(4)
and for , ], ] with < we define the sector,

def 
K = (r cos , r sin ): r > 0, ], [ .
3. Strategy of proof
Because the core of the proof of Theorem 1 is very technical, we will give here an exposition of the ideas
behind it for the convenience of the reader. Clearly, the problem is a mixed boundary value problem with discontinuous coefficients on a convex polyhedron. One should expect that this couldin principlebe treated as in [11]
(see also [38]), where the following is shown for the mixed problem with Hlder continuous coefficients:
If for any edge point or vertex x a certain spectrum (x) satisfies,
p<

2
z (x), z ]0,1[ 1 z
inf

for any edge point x,

(5)

for any vertex x,

(6)

and, additionally,
p<

inf

z (x), z ]0,1[

3
1 z

then the associated differential operator with Hlder continuous coefficients provides a topological isomorphism
1,p
1,p
between W and W . Here, the spectrum (x) is the spectrum of an associated (generalized) SturmLiouville
operator, if x is a point from an edge, see the next section for details, and in the case of a vertex it is the spectrum of
an associated LaplaceBeltrami operator, see [11] for details. The problem is that it is already difficult to determine
the spectrum of this LaplaceBeltrami operator if the coefficient function is constant, and we have no idea how to do
this in the case of heterogeneous materials.
Fortunately, there is a way out of this dilemma: for Dirichlet problems a deep idea of Mazya [39] permits to restrict
the investigation to the edge singularities as far as the integrability of the gradient of the solution up to an index p > 3
is concerned, see Proposition 11 below. (This heavily rests on the a priori known Hlder continuity of the solution,
see [34, Ch. III.14].) So we may circumvent the analysis of the vertex singularities, if we can transform the problem
to an equivalent one with Dirichlet boundary conditions. The strategy of proof is thus the following: We first deform
the problem via a bi-Lipschitz (in fact: piecewise linear) mapping, such that in the resulting polyhedron the Neumann
boundary part is a complete side of it, see Fig. 2. When doing so, we have to show that under this deformation
1,p
1,p
are suitably mapped on spaces of the same quality and that, additionally,
the occurring spaces W and W

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

29

Fig. 2. Model domain, original (left) and transformed (right), with Neumann boundary (hatched area) and material interface (shaded area). N.B.
the material interface is flexed after transformation, and there is an additional material interface framed by the dashed lines after transformation.
(, , and keep their names after transformation.)

the differential operator goes over into a similar one (see Proposition 16). In a second step we reflect the problem
across the Neumann boundary part (see Proposition 17) and identify the detailed structure of the resulting (Dirichlet)
problem. One especially obtains additional edges in the interior of the polyhedron coming from the transform and the
reflection process (see also Fig. 2). The main part of the proof is then to show, that all edges fulfill the supposition of
Proposition 11, which stands in an obvious relation to (5). Regrettably, this latter is a touchy business, heavily resting
on an adequate reformulation of the transmission conditions for the SturmLiouville problem (see Section 6.2) and,
finally, on the sophisticated estimates for the geometrical and bimaterial outer edges, which we present in Appendix A.
4. Edge singularities
In this section we first recall the optimal regularity result from [39] for heterogeneous Dirichlet problems on
polyhedral domains and explain how to identify the occurring edge singularities.
Definition 6. Let numbers 0 < 1 < < n  0 + 2 be given and, additionally, real, positive definite 2 2
matrices 1 , . . . , n . We introduce on ]0 , n [ \ {1 , . . . , n1 } coefficient functions b0 , b1 , b2 the restrictions of which
to the interval ]j , j +1 [, j = 0, . . . , n 1, are given by:
j

b0 ( ) = 11 cos2 + 212 sin cos + 22 sin2 ,

j

b1 ( ) = (22 11 ) sin cos + 12 (cos2 sin2 ),

j

b2 ( ) = 11 sin2 212 sin cos + 22 cos2 .

(7)

If n = 0 + 2 , then we define the space H as W01,2 (]0 , n [), else as the periodic Sobolev space W 1,2 (]0 , n [)
{: (0 ) = (n )} (which clearly may be identified with the Sobolev space W 1,2 (S 1 ) on the unit circle S 1 ). For
every C we define the quadratic form t on H by:
def

n

t [] =

b2   + b1  b1  2 b0 d,

(8)

and A as the operator which is induced by t on L2 (]0 , n [).

Remark 7. It is easy to check that b2 

11 22 (12 )2
j
22

sectorial, what is also true for A , see [31, Ch. VI].

. From this it is straightforward to see that each form t is

30

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

Definition 8. Let R3 be a polyhedron which, additionally, is a Lipschitz domain and {k }k a (finite, disjoint)
polyhedral partition of . Let be a matrix function on which is constant on each k and takes real, symmetric,
positive definite 3 3 matrices as values. Take any edge E of any of the k s and consider an arbitrary inner point P
of this edge. Choose a new orthogonal coordinate system (x, y, z) with origin at the point P such that the direction
of E coincides with the z-axis. We denote by OE the corresponding orthogonal transformation matrix and by E,P
1
1
the piecewise constant matrix function which coincides in a neighbourhood of P with OE (OE
(x + P ))OE
and
which satisfies:
E,P (tx, ty, z) = E,P (x, y, 0),

for all (x, y, z) R3 , t > 0.

(9)

By E (,) we denote the upper left 2 2 block of E,P (,, 0).

Remark 9. There exist angles 0 < 1 < < n  0 + 2 , such that E is constant on each of the sectors Kjj +1
and takes real, symmetric, positive definite matrices as values. Note that n = 0 + 2 if E corresponds to an
interior edge E, otherwise E is given on an infinite sector K0n which coincides near P with the intersection of (the
transformed) with the xy-plane.
Definition 10. We call an edge E of a geometric edge if E and all inner points of E belong to the closure
of exactly one sub-polyhedron k . Further, we say that E is a bimaterial outer edge if E and the function E
takes exactly two different values.
We proceed by quoting the central linear regularity result [39, Thm. 2.3], by means of which our regularity results
will be deduced:
Proposition 11. Let , {k }k and as in Definition 8. For any edge E let E be the 2 2 matrix valued function
on K0n in the sense of Definition 8. If for every edge E the thus induced operators A on L2 (]0 , n [) have a trivial
kernel for all with  ]0, 1/3 + [ ( > 0 arbitrarily small), then there is a p > 3 such that
: W0 () W 1,p ()
1,p

(10)

is a topological isomorphism.
Remark 12. Unfortunately, there are some errors in the paper [39], cf. also [13, Remark 2.2]. First, the assertion of
[39, Thm. 2.3] that the exponent p can be taken from the interval [2, 2/(1 )[ is erroneous, since the assumptions
of [39, Thm. 2.4] have to be taken into account. The correct formulation of the linear regularity result proved in [39]
is given in Proposition 11 above. Furthermore, the signs in formulas for the coefficients of certain generalized Sturm
Liouville equations are not correct, in detail: in [39, p. 240] there is a wrong sign in the formula for the Mellin


which has to be replaced by r
Therefore the formulas [39, (3.33)] for the sesquilinear
transform r
r u = u,
r u = u.
form a(u, v; ) and [39, (3.32)] for the corresponding differential problem differ in sign from the correct formulas (8)
and (11), (12). The correctness of the other considerations given in [39] is not affected by this.
Thus the question arises how to find the parameters for which the operator A has only a trivial kernel. One proceeds
as follows: standard arguments show that any function u from the kernel of the operator A obeys the differential
equation,
(b2 u ) + (b1 u) + b1 u + 2 b0 u = 0,

(11)

on each of the intervals ]j , j +1 [. Additionally, in every point {1 , . . . , n1 } the transmission conditions,

[u] = 0,

[b2 u + b1 u] = 0,

(12)

have to be satisfied. (As usual, [w] stands for lim w() lim w().) In order to find the critical parameters
, one employs the elementary solutions of the differential equation (11) on each of the subintervals ]j , j +1 [,
 ei (e2i + 1) ,

 ei (e
2i + 1) ,

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

31

which were announced in the pioneering paper [10] (see also [39, Ch. 3.6] for further details). The complex number
= j is determined by the matrix:

j
j

m11 m12 def 11 12

m=
=
,
j
j
m12 m22
12 22
as
def

1/2

i(m22 Dm ) m12
1/2

i(m22 + Dm ) + m12

(13)

where Dm denotes the determinant of the matrix m.

Remark 13. Because m22 is positive, necessarily satisfies 0  || < 1. Moreover, if

12
m11 m12
m
11 m
=
,
m
12 m
22
m12 m22
then = .

Making on any interval ]j , j +1 [ an Ansatz

uj ( ) = cj,+ ei (j e2i + 1) + cj, ei ( j e2i + 1) ,
def

(14)

these functions automatically satisfy (11), while the boundary conditions together with the transmission conditions (12) for = j (j {1, . . . , n 1}) lead to a 2n 2n homogeneous linear system for the coefficients cj,+ , cj, .
The usual criterion for the (nontrivial) solvability of this system gives the characteristic equation of the problem (11), (12) and allows (in principle) to determine the critical values or at least to give estimates for the real part
of them. In the next sections we will do this for all edges resulting from our problems.
5. Auxiliary results
Lemma 14. In the terminology from above let with  ]0, 1[ be a number such that there exists a (nontrivial)
function v H from the kernel of A , see Definition 6. Let 13 , 23 and 33 be real valued, bounded, measurable
functions on K0n and define the coefficient function on K0n

j
j
11
12
def
j
j
(x, y, z) =
22
12
13 (x, y) 23 (x, y)

def

= K0n R by:

13 (x, y)
j +1
23 (x, y) , if (x, y) Kj .
33 (x, y)

(15)

Then there is a compactly supported element f W 1,6 (K0n ) such that thealso compactly supportedvariational
2
1, 1

solution W01,2 (K0n ) of = f on K0n does not belong to W0

(K0n ).

Proof. It is not hard to calculate that the function 0 given by,



0 (x, y) = (x 2 + y 2 )/2 v arg(x + iy) ,

1,p

1,

(16)

2
belongs to Wloc (K0n ) if p [2, 1
[ but not to Wloc1 (K0n ). (Recall that v does not vanish identically on
]0 , n [.) By construction of A , the function 0 satisfies,

0 = 0,

(17)
def

in the distributional sense, see [39]. We define now the function by = (x, y, z) = 0 (x, y) and notice that
1,p

2
1, 1

2
belongs to Wloc (K0n ) for p [2, 1
[, but not to Wloc

2 C0 (R), then

(K0n ). Suppose = 1 2 with 1 C0 (K0n ) and

32

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

 


, R3 dx dy dz =

0 , 1 R2 dx dy 2 (z) dz
R

Kn
0

Kn
0

13


0
0
2
1 dx dy
+ 23
dz.
x
y
z

(18)

Kn
0

The first addend vanishes by (17) and the second by 2 C0 (R). The set of s with the above tensor product
structure is total in C0 (K0n ), therefore (18) is also zero for any from this latter space. Let be a function from
C0 (R3 ) which equals 1 in a neighbourhood of 0 R3 and which vanishes outside a ball B. Then one calculates for
any C0 (K0n ):








(), dx =
, () dx.
(19)
,  dx +
,  dx +
Kn

Kn


Kn
0

Kn

Kn

, () dx vanishes because (18) always is zero if C0 (K0n ). On the other hand, it is not hard to

see that the other two addends on the right-hand side definein their dependence on continuous linear forms
1,6/5
1,2
(K0n ), namely: the property Wloc
(K0n ) and the compact support property of imply , 
on W0
1,6/5

L2 (K0n ). Combining this with the embedding W0

(K0n )  L2 (K0n ), the claim becomes clear for the first addend
from the right-hand side of (19). Concerning the second addend, one easily estimates:






,  dx  

L (Kn )

L6 (BKn )

W 1,6/5 (Kn )



0
0
0
0
Kn
0

L (Kn )

W 1,2 (BKn )

W 1,6/5 (Kn ) .
0

def

Thus, setting = , one obtains the assertion.

2
1,p

Remark 15. If n = 0 + 2 , then K0n = R2 , K0n = R3 ; hence, W0 (K0n ) = W 1,p (R3 ).

Proposition 16. Let Rd be a bounded Lipschitz domain and be an open subset of its boundary. Assume that
is a mapping from a neighbourhood of into Rd which is bi-Lipschitz. Let us denote () = and ( ) = .
Then
(i) For any p ]1, [ induces a linear, topological isomorphism,
1,p

1,p

p : W ( ) W (),
which is given by (p f )(x) = f ((x)) = (f )(x);
1,p
1,p
(ii) p is a linear, topological isomorphism between W () and W ( );
(iii) If is a bounded measurable function on , taking its values in the set of d d matrices, then
p p = ,

(20)

with


 


(y) = (D) 1 (y) 1 (y) (D)T 1 (y)

1
.
| det(D)( 1 y)|

(21)

(D denotes the Jacobian of and det(D) the corresponding determinant.)

1,p
1,p
1,p
If, in particular, : W () W () is a topological isomorphism, then : W ( )
1,p

( ) also is (and vice versa).

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

33

Proof. The proof of (i) is contained in [23, Thm. 2.10]. (ii) follows from (i) by duality. We prove (iii): For
1,p
1,p 
f W ( ), g W ( ) we get by the change of variables formula:






p (p f ), g W 1,p ( ) = (p f ), p g W 1,p () = (f ), g W 1,p ()










= (x)(f )(x), (g )(x) dx = (x)(D)T (x)(f ) (x) , (D)T (x)(g) (x) dx




 | det(D)(x)|
(D)(x)(x)(D)T (x)(f ) (x) , (g) (x)
dx
| det(D)(x)|


 
 1   1  (D)T ( 1 (y))
(D) (y) (y)
f (y), g(y) dy
=
| det(D)( 1 y)|



 
 (D)T ( 1 ())
= (D) 1 () 1 ()
f
,g
.
1,p
| det(D)( 1 ())|
W
( )

The essential point is that as a Lipschitz continuous functionis differentiable almost everywhere and its (weak)
derivative is essentially bounded (see [15, Ch. 4.2.3]). The last assertion follows from (i), (ii) and (20). 2
Proposition 17. Let R3 be a bounded, convex, polygonal domain and be an open subset of such that
{(x, 0, z): x, z R} = . Let for any x = (x, y, z) the symbol x denote the element (x, y, z) and define as
the interior of
{x: x } .
If is a bounded, measurable function on taking its values in the set of real, symmetric 3 3 matrices, then we
define:

(x),
if x ,

11 (x ) 12 (x ) 13 (x )
def
(x)

=
(22)
12 (x ) 22 (x ) 23 (x ) , if x .

13 (x ) 23 (x ) 33 (x )
=
(i) If W1,2 () satisfies the equation = f W1,2 (), then the equation
1,2
holds for with
()
f W

(x),
if x ,

(x)
=
(x ), if x ,
def 1
def
and f defined by f, W 1,2 ()
= 2 f, | + | W 1,2 () . The function is defined by (x) = (x ) for

W 1,1 ().
1,p
1,p
and if
is a topological
(ii) Moreover, if f W (), then f W 1,p ();
: W0 ()
W 1,p ()
1,p
1,p
isomorphism, then : W () W () also is.
1,p
Proof. (i) It is known that belongs to W0 (),
see [18, Lemma 3.4]. Thus, (i) is obtained by the definitions of

, f , ,
and straightforward calculations, based on Proposition 16 when applied to the transformation x  x .
1,p
(ii) The operator f  f is the adjoint to  12 (| + | ). The latter maps each W0 ()
continuously into
1,p
W () for any p ]1, [. The last statement is then implied by the preceding ones and the definition of . 2

Remark 18. The proposition is mutatis mutandis true for the reflection at other planes.

34

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

1,p

1,p

Remark 19. In fact it can be shown that : W () W () is a topological isomorphism, if and

1,p
only if
is a topological isomorphism between the symmetric part of W0 ()
and the symmetric part of
1,p
The point is that this is not of use to us here because the reduction procedure of [39] applies to the whole
W
().
space and not only to the symmetric parts of the spaces (see also Remark 23).
In the sequel we will transform our model problems which include mixed boundary conditions to the case of Dirichlet
conditionswhich are imposed in Proposition 11. In essence, this happens via a linear transformation leading to a
peculiar triangle, a bi-Lipschitz transformation and a reflection argument. All of this is carried out in the next section.
6. Proof of Theorem 1
6.1. Transformation of the problem
Proposition 16 allows us in a first step to reduce the case of an arbitrary triangle to that one where is the
triangle with the vertices (1, 1), (1, 1), (1/2, 1/2) and, additionally, is the line segment between (0, 0) and
(1, 1), see Figs. 1 and 3. Namely, first one shifts the triangle such that P becomes the origin. Let P1 denote the
vertex where (the shifted) ends and P2 the vertex which does not touch . We now transform R2 under the linear
mapping which assigns P1 to (1, 1) and P2 to (1/2, 1/2). Extending this mapping to R3 by letting the z-component
invariant, one obtains the special geometric constellation of Fig. 3 stated above. Clearly, the transformed plane
maintains the properties demanded in the suppositions of Theorem 1. In particular, we denote the point, where the
(transformed) plane intersects the z-axis, by P . In a natural sense we may speak of an upper half space Gu and a
lower half space Gl (each on one side of the intersecting plane ), where the coefficient function takes the values:

a11 a12 a13

b11 b12 b13
+ = a12 a22 a23 on Gu ,
= b12 b22 b23 on Gl .
(23)
a13 a23 a33
b13 b23 b33
We transform the problem via the bi-Lipschitz transformation (see Fig. 4)

1/ 2 1/ 2 0

2 0 on {(x, y, z): y > x},

0
0
1
def

2
0 0

on {(x, y, z): y  x}.

1/ 2 1/ 2 0
0
0
1

(24)

Fig. 3. Transformation of a model domain (see the left side of Fig. 1) to the generic domain, with Neumann boundary (hatched area) and material
interface (shaded area). (, , , and keep their names after transformation.)

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

35

Fig. 4. The piecewise linear transformation in the xy-plane. Triangle before (left) and after (right) transformation by .

(Please notice that the determinants

ofboth matrices in (24) equal 1.) () is again a triangledenoted by and
has
now the vertices (0, 0), (0, 2), ( 2, 0), while the new domain is = ]1, 1[. equals the subinterval
]0, 2[ of the x-axis. The image of consists of two triangles having one common edge E {(x, x, z):
x > 0, z R}. (Of course, if was orthogonal to the z-axis, then both triangles are also orthogonal
to the
2, 0, 1),
z-axis.)
Clearly,
the
Neumann
boundary
part
is
now
the
rectangle
with
the
vertices
(0,
0,
1),
(0,
0,
1),
(

( 2, 0, 1). The transformed matrix (see Proposition 16),

T

1/ 2 1/ 2 0
a11 a12 a13
1/ 2 1/ 2 0
0
(25)
2 0 a12 a22 a23 0
2 0 ,
a13 a23 a33
0
0
1
0
0
1
is calculated as

a11 +2a12 +a22

2

a12 + a22
+a23
a13
2

while the transformed matrix

a11
0
0
2
1/ 2 1/ 2 0 a12
a13
0
0
1
is calculated as

2a11
a12 + a11

2a13

a12 + a22
2a22
2a23

a12
a22
a23

a13
+a23
2

2a23 ,
a33

T
a13
0
0
2
a23 1/ 2 1/ 2 0 ,
a33
0
0
1

a12 + a11

a11 +2a12 +a22

2
a13
+a23
2

2a13

+a23
a13
2

(26)

(27)

(28)

a33

(and analogously for the matrix b). We reflect the problem at the xz-plane in the spirit of Proposition 17 and obtain

def
a new triangle with the vertices (0, 2 ), ( 2, 0), (0, 2 ), a new domain = ]1, 1[ and the coefficient

function on is defined as in (22). Thus, we end up with a Dirichlet problem on . By Proposition 17 it suffices
to show that
1,p

: W0 ( ) W 1,p ()

is a topological isomorphism for a p > 3. For this, however, we may apply Proposition 11: we are done if we are able
to show that for all edges E the induced operators A have a trivial kernel for all with  ]0, 1/3 + [ ( > 0
arbitrarily small). The occurring edges E are the following, see also Fig. 5:

geometric edges,
bimaterial outer edges,
the edges Ez+ and Ez lying between P and (0, 0, 1), or between P and (0, 0, 1), respectively,
the edge Exz , which is the intersection of the (transformed) with the xz-plane,
E and the reflected E .

36

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

Fig. 5. Generic model domain after transformation by with Neumann boundary (hatched area), material interface (shaded area), and the edges
Ez+ , Exz , E , as well as bimaterial outer edges. ( keeps its name after transformation with .)

6.2. Reformulation of the transmission conditions

The aim of this subsection is to express the transmission conditions for the Ansatz functions (see (12)) in a
condensed manner in terms of j , j +1 , j .
Lemma 20. Let be defined by (13), and

m=

m11
m12

m12
m22

def

11
j
12

12
,
j
22

u( ) = c+ ei (e2i + 1) + c ei (e
2i + 1) ,
def

where c+ , c are arbitrary complex constants. Further, let b1 , b2 be defined as in (7). Then

1/2 
b2 ( )u ( ) + b1 ( )u( ) = iDm c+ ei (e2i + 1) c ei (e
2i + 1) ,

(29)

(30)

where Dm again denotes the determinant of the matrix m.

Proof. First, one easily verifies:
u ( ) = c+ ei (e2i + 1) (i)

1 e2i
1 e
2i
+ c ei (e
2i + 1) i
.
2i
1 + e
1 + e
2i

(31)

Next we want to prove:

b2 ( )i

1 e2i
+ b1 ( ) = iD1/2 .
1 + e2i

(32)

For this we calculate:

i
1/2

1 e2i
e2i
,
= i 2i
2i
1 + e
e
+

and abbreviate the denominator i(m22 + Dm ) + m12 of by N . One has:





1/2
1/2
e2i = e2i i(m22 + Dm ) + m12 i(m22 Dm ) + m12 /N

(33)

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

and

37





1/2
1/2
e2i + = e2i i(m22 + Dm ) + m12 + i(m22 Dm ) m12 /N ,

what leads to
1 e2i
m12 (e2i + 1) + m22 i(e2i 1) + Dm i(e2i + 1)
i
=
i
.
1/2
1 + e2i
m12 (e2i 1) + m22 i(e2i + 1) + Dm i(e2i 1)
1/2

We augment the last fraction by

sin
;
e2i 1

(34)

exploiting the equation

ei (ei + ei )
cos
e2i + 1
= i cos ,
sin = i i
sin =
2i
i
e
1
e (e ei )
the right-hand side of (34) becomes:
1/2

im12 cos + im22 sin Dm cos

1/2

m12 sin m22 cos + iDm sin

1/2

m12 cos m22 sin iDm cos

1/2

m12 sin m22 cos + iDm sin

1/2

1/2

(m12 cos m22 sin iDm cos )(m12 sin m22 cos iDm sin )
(m12 sin m22 cos )2 + Dm sin2
1/2

(m22 m11 ) cos sin + m12 (cos2 sin2 ) + iDm

m11 sin2 2m12 cos sin + m22 cos2

1/2

b1 ( ) + iDm
.
b2 ( )

(35)

Thus, (32) holds true. By complex conjugation one obtains from (32):
1 e
2i
1/2
+ b1 ( ) = iDm ;
1 + e
2i
(32) and (36) together with (31) give the assertion (30). 2
b2 ( )i

(36)

Corollary 21. Let u be the function on [0 , n ] which coincides on ]j , j +1 [ with uj defined in (14).
(i) Assume first j {1, . . . , n 1} and let Dj and Dj +1 denote the determinants of the matrices:
j
j +1
j

j +1

12
11 12
11
and
j
j
j +1
j +1 ,
12 22
12
22
def

def

respectively. If we abbreviate = j and = j +1 , then the transmission conditions in the point = j ,

[u] = [b2 u + b1 u] = 0,

(37)

express as
2i + 1) (38)
2i + 1) = cj +1,+ ei (e2i + 1) + cj +1, ei (e
cj,+ ei (e2i + 1) + cj, ei (e
and
1/2 


cj,+ ei (e2i + 1) cj, ei (e
2i + 1)

1/2 
2i + 1) ,
= Dj +1 cj +1,+ ei (e2i + 1) cj +1, ei (e

Dj

(39)

respectively. Thus, in case of Dj = Dj +1 for (37) it is necessary and sufficient that

cj,+ (e2i + 1) = cj +1,+ (e2i + 1)

(40)

2i + 1)
2i + 1) = cj +1, (e
cj, (e

(41)

and

hold.

38

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

(ii) Assume now n = 0 + 2 . Then the corresponding transmission conditions in 0 express as

c1,+ ei0 (1 e2i0 + 1) + c1, ei0 ( 1 e2i0 + 1)
= cn,+ ein (n e2i0 + 1) + cn, ein ( n e2i0 + 1)

(42)


c1,+ ei0 (1 e2i0 + 1) c1, ei0 ( 1 e2i0 + 1)

1/2 
= Dn cn,+ ein (n e2i0 + 1) cn, ein ( n e2i0 + 1) ,

(43)

and
1/2 

D1
respectively.

6.3. Discussion of the edge singularities

For geometric edges and bimaterial outer edges we show in Appendix A that the operators A have a trivial kernel
if  ]0, 1/2].
Next we consider the edges Ez+ and Ez : starting with Ez+ , one has to deal with the coefficient matrices:

a +2a +a
11
12
22
def
a12 a22
2
if ]/2, /4[,
m
=
a12 a22
2a22

a12 a11
2a11
def
o =
if ]/4, 0[,
a12 a11 a11 +2a212 +a22

2a11
a12 + a11
def
o=
if ]0, /4[,
a12 + a11 a11 +2a212 +a22
a11 +2a12 +a22

a12 + a22
def
2
m=
if ]/4, /2[.
a12 + a22
2a22
Thus, one has to consider the ansatz functions (see Remark 13):

def

w = c+ ei (e
2i + 1) + c ei (e2i + 1)

def i 2i
def
v = d+ e (e + 1) + d ei (e2i + 1)
u=
def

2i + 1)
v = d+ ei (e2i + 1) + d ei (e

def
2i + 1)
w = c+ ei (e2i + 1) + c ei (e

on ]/2, /4[,
on ]/4, 0[,
on ]0, /4[,
on ]/4, /2[,

with defined by (13) (and analogously from the entries of the matrix o). Please notice that the determinants of the
matrices m, m,
o, o all equal the determinant of the matrix

a11 a12
,
a12 a22
the value of which we denote by D in this proof. Taking this into account, the transmission conditions in = /4
read in view of (40)/(41),
,
c+ (1 i)
= d+ (1 i)

(44)

c (1 + i) = d (1 + i) .

(45)

and

Analogously, the transmission conditions in 0 equivalently express as

= d+ (1 + )
d+ (1 + )

(46)

,
d (1 + ) = d (1 + )

(47)

and

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

39

while those in /4 can be written as

d+ (1 + i) = c+ (1 + i)

(48)

= c (1 i)
.
d (1 i)

(49)

and

The boundary condition u(/2) = w(/2) = 0 leads to

= 0,
c+ ei/2 (1 ) + c ei/2 (1 )

(50)

or, in other words,

c+ = c ei

(1 )

,
(1 )

(51)

while the boundary condition u(/2) = w(/2)

= 0 gives:
+ c ei/2 (1 ) = 0,
c+ ei/2 (1 )
or, alternatively,
c = c+ ei

(1 )

.
(1 )

(52)

Combining (51), (49), (47), (45), (52), (44), (46), (48), one ends up with the characteristic equation for :
(1 + ) (1 + i) (1 )
(1 + ) (1 + i)
(1 )
(1 i)
(1 i)
(1 + i) (1 ) (1 i)
(1 + i)
(1 ) (1 i)
(1 + )
(1 + )

2
2

(1 + ) (1 i)

(1 + i) (1 )
=
= e2i .

(1 ) (1 i)

(1 + i) (1 + )

(53)

Let us remark that c+ cannot vanish unless also the other coefficients vanish. Moreover, we notice that all the terms
1
1
1
1
1 + i, 1
, 1 ,
1i
, 1 + , 1+i , 1 i, 1+ have positive real part because ||, || < 1. Hence, we have:
(1 + i)
=
(1 )
as well as
(1 + )
=
(1 + i)

1 + i
1
1+
1 + i

if   1. Further, observing the relations,

1 + i
=
1

and

(1 )

=
(1 i)

and

(1 i)
=

(1 + )

1
1
1i

and

1+
1 + i

1
1 i
1 i
1 +

1
1i
1+

(54)

(55)

(56)

1+
and putting = arg 1+i
1 and = arg 1+i , this altogether enables us to rewrite (53) as

e2i(+2( +)) = 1.

(57)

It is obvious that all satisfying (57) must be real. Our claim is now: + equals /2 or 3/2. For this, we
mention that, by definition, , ], ]; thus the claim is true, if we can show
1 + i 1 +
= i.
1 1 + i
This we will do now: exploiting the definitions of , we get:

(58)

40

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

1 + i 1 D 1/2 + m12 m22 + i(D 1/2 + m22 m12 )

=
1
2
m12 + iD 1/2

1  1/2
D + m11 m12 + i(D 1/2 m11 + m12 )
=
2m11

1
a11 a22
a22 a11
1/2
1/2
+i D +
.
D +
=
2m11
2
2
Analogously, we calculate:
1+
2io22
=
= 1/2
1 + i
D + o12 o22 + i(D 1/2 + o22 o12 ) D 1/2 +

2io22
a11 a22
2

+ i(D 1/2 +

a22 a11 .
)
2

Taking into account o22 = m11 , this gives (58). Hence, the transcendental equation (57) for reads in any case as
e4i = 1. Trivially, the smallest positive possible is 0 = 1/2. Thus, the edge Ez+ meets the preconditions of
Proposition 11. The considerations for the edge Ez are the same, word by word.
Next we consider the edge Exz , lying in the xz-plane. The coefficient matrices belonging to its neighbouring
sectors are:

a +2a +a
11
12
22

a12 + a22 a13+a23

2
q11 q12 q13
2
def
(59)
Q = q21 q22 q23 =
a
+
a
2a
2a23
12
22
22
,

+a23
a13
q31 q32 q33
a33
2a23
2

b +2b +b
+b23
b
13
11
12
22

b
+
b
12
22
2
r11 r12 r13
2
def
(60)
R = r21 r22 r23 =
b
+
b
2b
2b23
12
22
22
,

b13
+b23
r31 r32 r33
2b
b
23

if y > 0 and their reflected counterparts,

q11 q12

q
q22
Q=
12
q13 q23

q13
q23
q33

r11

r
and R =
12
r13

33

r12
r22
r23

r13
r23 ,
r33

if y < 0 (see (25), (27) and Proposition 17). According to Proposition 11 one has to perform a rotation in the
xz-plane which moves the edge Exz to the z-axis. This means, one has to consider the matrices,

cos 0 sin
cos 0 sin
0
1
0 M 0
1
0 ,
sin 0 cos
sin 0 cos
R,
respectively and being the angle between the edge Exz and the z-axis. A straightforward
M taken as Q, R, Q,
calculation shows that the resulting upper 2 2 blocks look alike:

t11 t12
s11 s12
if ], [,
if ]0, [,
s12 s22
t12 t22

t11 t12
s11 s12
if ]0, [,
if ], [.
(61)
t12 t22
s12 s22
Hence (see Remark 13), the corresponding numbers 1 , 2 , 3 , 4 are related by 1 = 4 and 2 = 3 . In the sequel
we employ the numbers for 3 and for 4 . In this notation we show:
Lemma 22. Assume the existence of complex numbers c+ , c , d+ , d , c+ , c , d+ , d (at least one of them nonzero)
such that

v def
2i + 1) + d ei (e2i + 1) on ], [,

= d+ ei (e

def
def
2i + 1) + c ei (e2i + 1) on ], 0[,
w = c+ ei (e
u=
def

w = c+ ei (e2i + 1) + c ei (e
2i + 1) on ]0, [,

def
2i + 1) on ], [,
v = d+ ei (e2i + 1) + d ei (e

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

41

obeys the transmission conditions in , , 0, . Then 

/ ]0, 1/2].
Proof. The transmission condition [b2 u + b1 u]0 = 0 together with Corollary 21 (see in particular (39)) implies:
c+ ( + 1) c ( + 1) = c+ ( + 1) c ( + 1) .

(62)

On the other hand, the transmission condition for b2 u + b1 u in / (see (43)) gives:
d+ ei ( + 1) d ei ( + 1) = d+ ei ( + 1) d ei ( + 1) .

(63)

Let us first consider the case, where

c+ = c ,

c = c+ ,

d+ = d ,

d = d+ .

(64)

Inserting these relations in (62) and (63) one obtains that both sides of (62) and (63) in fact have to vanish. But this
means in view of Lemma 20 nothing else but
b2 ( )u ( ) + b1 ( )u( ) = 0 for = 0, .
Thus, the restriction of u to the interval ]0, [ leads to a bimaterial problem including a Neumann condition on both
interval ends. Then 
/ ]0, 1/2], see Theorem 25 below.
Assume now that (64) is not satisfied. Then we introduce the function:

def
def w
= c ei (e2i + 1) + c+ ei (e
2i + 1) on ]0, [,
u =
(65)
def i
2i

i
2i + 1) on ], [,
v = d e (e + 1) + d+ e (e
on [0, ] and consider the function:

def
2i + 1)
(c+ c )ei (e2i + 1) + (c c+ )ei (e
u! = u|[0,] u =
i
2i

i
2i + 1)
(d+ d )e (e + 1) + (d d+ )e (e

on ]0, [,
on ], [.

It is straightforward to verify that the condition [u]0 = 0 implies u! (0) = 0 and the periodicity condition in /
yields u! () = 0. Next we intend to show the transmission conditions [u! ] = [b2 ( )u! +b1 ( )u! ] = 0. Because we
= [b2 ( )u  + b1 ( )u]
= 0.
already know by supposition [u] = [b2 ( )u + b1 ( )u] = 0 it remains to show [u]
One easily verifies [u]
= [u] , and the latter is zero by supposition. Finally, by Lemma 20 we have:


b2 ( )u  + b1 ( )u = (b2 w  + b1 w)|
(b2 v  + b1 v)|



= iD c ei (e2i + 1) c+ ei (e
2i + 1)


2i + 1)
+ iD d ei (e2i + 1) d+ ei (e
= (b2 w  + b1 w)|
+ (b2 v  + b1 v)|
= [b2 u + b1 u] .
But the right-hand side of this equation is zero in view of the transmission condition [b2 u + b1 u] = 0. Thus, this
second case leads to a bimaterial Dirichlet problem, for which also Theorem 25 gives 
/ ]0, 1/2]. 2
Remark 23. The key point is here that we can invest estimates of  for bimaterial edges in both, the Neumann and
Dirichlet case (see Theorem 25 below), despite the fact that in the original problem the edge is situated on a Neumann
boundary plane.
It remains to consider the edge E (and its reflected counterpart). Let first t R be a number such that (0, 0, t) +E
has its endpoint in 0 R3 and O be a rotation of the plane {(x, x, z): x, z R} which transforms (0, 0, t) + E to
the z-axis. Suppose that for one with  ]0, 1/2] there is a (nontrivial) function v from the kernel of the resulting
operator A . If one takes the coefficient function defined in (15) as

j
j
j
11 12 13
j +1
def j
j
j
(x, y, z) = 12
(66)
22 23 if (x, y) Kj ,
j
j
j
13 23 33

42

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

then, by Lemma 14, there is a compactly supported element f W 1,6 (R3 ) such that thealso compactly
supportedvariational solution W 1,2 (R3 ) of = f does not belong to W 1,4 (R3 ). Because the support of is compact, it can then (the more) not belong to W 1,6 (R3 ). Now we revoke the transformations O, the shift
(0, 0, t) and . Applying Proposition 16, one obtains a f W 1,6 (R3 ) and a W 1,2 (R3 ) \ W 1,4 (R3 ) satisfying
= f , or, equivalently, + = f + W 1,6 (R3 ). It is not hard to see that the matrix valued function equals above the matrix + and below the matrix (see (23)). But a result of [14, Thm. 3.11],
see also [3, Ch. 4.5], says that
+ 1 : W 1,p (R3 ) W 1,p (R3 )
is a topological isomorphism for any p ]1, [. This contradicts the above supposition. The proof for the reflected E
runs along the same lines; thus the proof of Theorem 1 is complete.
7. Proof of Theorem 2 and of Corollary 3
First we consider the case where is one side of the triangle . Modulo an affine transformation in R2 we
may focus on the case where is identical with the interval ]0, 1[ on the x-axis, see Proposition 16. We reflect
symmetrically at the xz-plane and obtain a domain and a reflected coefficient function .
The resulting boundary
conditions are then homogeneous Dirichlet on all . By Proposition 17 it is sufficient to show that
1,p

: W0 ()
W 1,p ()

is a topological isomorphism for a p > 3. Of course, we will again apply Proposition 11 and have, hence, to discuss
the edge singularities. The occurring edges are:
(i) geometric edges,
(ii) bimaterial outer edges,
(iii) the intersection of the xz-plane with , in particular, the parts of the z-axis below and above the intersection
point with is a bimaterial outer edge.
For all these edges we already know that the corresponding operators A have a trivial kernel provided  ]0, 1/2];
namely: the claim for geometric edges and bimaterial outer edges is shown in the next section (see Theorem 24 and
Theorem 25) while the situation of (iii) is exactly the same as treated in Lemma 22.
Let us now regard the second case: modulo an affine transformation in R2 we may restrict ourself to the case
where is the union of the interval ]0, 1[ on the x-axis and the interval [0, 1[ on the y-axis. Again we reflect the
problem at the xz-plane, but afterwards a second time at the yz-plane. Thus, we end up with a Dirichlet problem
def
on = V ]1, 1[, where V R2 is the square with the vertices (0, 1), (1, 0), (0, 1), (1, 0). Denoting the new
coefficient function by ,
it suffices by Proposition 17 to show that
1,p

: W0 ()
W 1,p ()

is a topological isomorphism for a p > 3. According to Proposition 11, it remains to show that for every edge E
the kernels of the corresponding operators A are trivial if  ]0, 1/3 + [ ( arbitrarily small). If (0, 0, t) is the
intersection point of with the z-axis, then the occurring edges are:
(i)
(ii)
(iii)
(iv)
(v)
(vi)
(vii)
(viii)

geometric edges,
bimaterial outer edges,
{(0, 0, s): s ]1, t[},
{(0, 0, s): s ]t, 1[},
the intersection of the xz-plane with ,
the intersection of the xz-plane with the reflected ,
the intersection of the yz-plane with ,
the intersection of the yz-plane with the reflected .

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

43

The geometric and bimaterial outer edges are treated in Appendix A. (iii), (iv), (v), (vi) lead again to a constellation (61), which was treated in Lemma 22. This is also true for (vii) and (viii), but requires here an additional moments
thought: let us denote the value of the coefficient function above by + and below by . Concerning (vii),
the reflected matrices then equal
+

+

11 +
11
12 13
12 13
+
and
.
+
+

12
22
23
12
22
23
+
+
+

13 23
33
13 23
33
We perform now a rotation within the xy-plane which transforms the (positive) y-axis into the (positive) x-axis and
the (positive) x-axis into the negative y-axis; clearly the transformed edge lies then in the xz-plane. One obtains the
transformed coefficient matrices,
+

+
+
22 +
+
11 +
0 1 0
0 1 0
12 13
12
23
,
1 0 0 + + + 1 0 0 = +
+
+
12
22
23
12
11
13
+
+
+
+
+
+
0
0
1
0 0 1
13 23 33
23 13 33

22
11 12 13

0 1 0
0 1 0
12
23
,
1 0 0 1 0 0 =

12
22
23
12
11
13

0 0 1
0 0 1
13 23 33
23 13
33
while the reflected matrices transform as follows:

+
11 +
0 1 0
12
1 0 0 +
+
12
22
0 0 1
+
+
13
23

0 1 0
11
12

1 0 0

12
22
0 0 1

13
23

+
22
+
0 1 0
13
1 0 0 = +
+
23
12
0 0 1
+
+
33
23

0 1 0
13
22
1 0 0 =

23
12
0 0 1

33
23

+
12
+
11
+
13

12

11

13

+
23
,
+
13
+
33

23
.

13

33

Thus, from this point on we are in the same situation as in the discussion for the edge Exz (see page 40) and everything
runs completely the same way. (viii) is analogous to (vii).
We come to the proof of Corollary 3: because we demanded that the plane should not touch the upper plate nor
the ground plate it is possible to divide the problem by a suitable partition of unity into one which affects the upper
(lower) part and is separated from and one which contains but has only a Dirichlet condition on its upper
(ground) plate. The latter is already treated in Theorem 2. The first can be reflected at the upper (ground) plate and
one ends up again with the setting which is treated in Theorem 2.
8. Concluding remarks
The results of this paper easily carry over to problems with Robin boundary conditions. Indeed, one can prove that
1,p
1,p
if " is the surface measure on and # L (, d" ), then the linear map T : W () W () given by:

T , W 1,p = # d"

(and representing the Robin boundary condition) is infinitesimally small with respect to the operator . Thus,
the domains of both operators are the same by classical perturbation theory, see [31, Ch. IV.1].
The reader has possibly asked himself why the results are deduced from [39] and why the concept of that paper does
not work for boundary conditions which are not Dirichlet. One problem consists in finding an adequate energy space
in case of edges on Neumann boundary parts which, additionally, has to be in correspondence with the properties of
the Mellin transform. Our attempts to find such an energy space have failed up to now.
Alternatively, the question arises whether it is possible to discard vertices from the analysisby reflection
argumentsalso for the Neumann case. It turns out that this can be done in relevant cases, but seems to be delicate in general.

44

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

In principle it is possible to generalize our results to the case where not only one plane intersects the domain, but
severals do. In order to classify the singularities stemming from the additional inner edges (where the planes meet)
one can apply the result [13, Thm. 2.5]. We have not carried out this here only for technical simplicity, see also [32].
Acknowledgement
We gratefully acknowledge that some of the ideas from Appendix A are due to our colleagues J. Elschner and
G. Schmidt.
Appendix A. The transcendental equation for geometric edges and bimaterial outer edges
It is the aim of this section to discuss the edge singularities for geometric edges and bimaterial outer edges;
precisely, we intend to show the following two theorems:
Theorem 24. For any geometric edge E the kernels of the associated operators A are trivial in each of the following
two cases:
(a) the opening angle 1 0 is not larger than and  ]0, 1[,
(b) 1 0 ], 2[ and  ]0, 1/2].
Theorem 25. Let K01 , K12 be two neighbouring sectors in R2 with 1 0 , 2 1  and 2 0 < 2 . Let 1 , 2

be two real, positive definite 2 2 matrices corresponding to the sectors K01 , K12 . Let t be the form defined in (8)
either on H01 (]0 , 2 [) or on H 1 (]0 , 2 [). Then there is an  > 0 such that the kernel of the corresponding operator A
(see Definition 6) is trivial if  ]0, 1/2 + ].
We will prove the theorems in several steps, starting with the following:
Lemma 26. Let C with || < 1, and define for ], ] the number:
def

= arg

e2i + 1
], ].
+1

Then either , + ], 0[ or = = 0 or , + ]0, [, or = + = .

Proof. The cases = 0 and = are straightforward. In the remaining cases one has:
ei( + ) = ei

e2i + 1 | + 1|
(ei + ei )( + 1) ||2 ei + ei + 2(ei )
=
=
.
+ 1 |e2i + 1|
| + 1||e2i + 1|
| + 1||e2i + 1|

Thus, the imaginary part of ei( + ) equals

(1||2 ) sin
,
|1+||e2i +1|

and its sign depends in an obvious way only on .

It follows the proof of Theorem 24; without loss of generality we may assume 1 = . Again exploiting the Ansatz
functions (14), the Dirichlet conditions in 0 , 1 ], ] reads
2i0 + 1) = 0,
c+ ei0 (e2i0 + 1) + c ei0 (e
c+ e

( + 1) + c e

( + 1) = 0.

(A.1)
(A.2)

These equations are nontrivially solvable in c+ , c iff

(e
2i0 + 1) ( + 1)
( + 1) (e2i0 + 1)
2i0

+1
+1
2i 2i0 e
=e
e
,
+ 1
e2i0 + 1

1 = e2i e2i0

(A.3)

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

45
2i0

+1
compare the considerations in Section 6.3, in particular (54). Putting = arg e
, we may write (A.3)
+1

2i(
+)
0
as e
= 1. Obviously, must be real and ||
= || < 1. Hence, in case (a), where 0 [0, [, we
obtain 0 + [0, ] by Lemma 26, which excludes  ]0, 1[. If 0 ], 0[, then, by Lemma 26, we have
0 + ], 0[, which shows the assertion in case (b).
Concerning Theorem 25, we may apply a rotation (corresponding to a shift in the angle space) and thus reduce the
general case to that one where 0 = , 1 = 0 and 2 = . Again using the Ansatz functions (14) we are getting the
following equations expressing the transmission conditions in 0, see Corollary 21,
def

c+ ( + 1) + c ( + 1) = d+ ( + 1) + d ( + 1) ,
and

1/2 

Dm



1/2 
c+ ( + 1) c ( + 1) = Do d+ ( + 1) d ( + 1) .

We define:
def

 =

(A.4)

(A.5)

1 if Dirichlet in ,
1
if Neumann in ,

and analogously for . In this convention (see Lemma 20), the boundary condition in yields
2i + 1) = 0
c+ ei (e2i + 1)  c ei (e
or, what is the same,
(e2i + 1)
.
(e
2i + 1)
On the other hand, the corresponding boundary condition in implies:
c =  c+ e2i

(A.6)

2i + 1) = 0
d+ ei (e2i + 1)  d ei (e
or, equivalently,
d+ =  d e2i

2i + 1)
(e
.
(e2i + 1)

We insert (A.6) and (A.7) in (A.4) and (A.5) and obtain:





2i + 1)
2i + 1)

2i (e

2i (e

( + 1) d  e
( + 1) + ( + 1) = 0,
c ( + 1) +  e
(e
2i + 1)
(e2i + 1)
and



(e2i + 1)
1/2

Dm c ( + 1)  e2i
(

+
1)
(e
2i + 1)


2i + 1)
(e
1/2

= 0,
+ Do d ( + 1)  e2i
(
+
1)
(e2i + 1)

for c = c+ and d = d . (A.8), (A.9) are nontrivially solvable iff




2i + 1) ( + 1)
(e2i + 1) ( + 1)
1/2
2i (e
1


Do 1 +  e2i
e

( + 1) (e
2i + 1)
( + 1) (e2i + 1)



2i + 1) (
2i + 1) ( + 1)
+ 1)
1/2
2i (e
2i (e
1  e
= 0.
+ Dm 1 +  e
(1 + ) (e
2i + 1)
( + 1) (e2i + 1)
Putting
= arg

e2i + 1
,
+1

= arg

2i + 1
e
,
+ 1

and arguing as in (54)(56), this altogether enables us to rewrite (A.10) as

Do [1 +  e2i( + ) ][1  e2i(+) ] + Dm [1 +  e2i(+) ][1  e2i( + ) ] = 0,
1/2

1/2

(A.7)

(A.8)

(A.9)

(A.10)

46

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

or, what is the same,

Do [ei( + ) +  ei( + ) ][ei(+)  ei(+) ]
1/2

+ Dm [ei(+) +  ei(+) ][ei( + )  ei( + ) ] = 0.

1/2

(A.11)

This means that in the pure Dirichlet case (with  =  = 1) (A.10) can be written equivalently as
1/2

1/2

(A.12)

1/2

(A.13)

Do sin ( + ) cos ( + ) + Dm cos ( + ) sin ( + ) = 0

and in the pure Neumann case (with  =  = 1) as
1/2

Do cos ( + ) sin ( + ) + Dm sin ( + ) cos ( + ) = 0.

Because Dm and Do are arbitrary positive constants it suffices to focus the following discussion on (A.12).

Lemma 27. If ,  with + < 2 , then any solution of (A.12) satisfies 

/ ]0, 1/2 + ] for an  > 0.
Proof. Since, by Lemma 26, sin ( + ) = 0 and sin ( + ) = 0, if 0 <  < 1, we can rewrite (A.12) as
1/2

1/2

Do cot ( + ) + Dm cot ( + ) = 0.

(A.14)

Note that
 cot( + i) =

(cosh2 sinh2 ) sin cos

sin 2
=
,
2
2
2
(sin cosh ) + (cos sinh )
2(sin + sinh2 )

hence, with = + i, the real part of (A.14) satisfies:

1/2

Do sin 2( + )
sin2 ( + ) + sinh2 ( + )

1/2

Dm sin 2( + )
sin2 ( + ) + sinh2 ( + )

= 0.

(A.15)

If 0 <  1/2, then Lemma 26 shows that 0 < 2( + ), 2( + )  , and therefore both terms on the left-hand
side of (A.15) are nonnegative. Due to + < 2 , at most one of them may be zero. This proves the assertion. 2
Remark 28. If one is confronted with a bimaterial outer edge supplemented by a Dirichlet condition in and a
Neumann condition in (what means  =  = 1), then (A.11) reads as

 


 

1/2
1/2
Do sin ( + ) sin ( + ) + Dm cos ( + ) cos ( + ) = 0.
(A.16)
If we again suppose , ]0, [, then we may divide (A.16) by sin(( + )) sin(( + )) (provided  ]0, 1[)
and obtain the equivalent condition:

 
 Do1/2
cot ( + ) cot ( + ) = 1/2 .
Dm

(A.17)

It is not hard to see that there are parameter configurations , , , , Do , Dm such that (A.17) is fulfilled for with
arbitrarily small (positive) real part; see also [40], where the case of scalar multiples of the Laplacian already was
treated.
Remark 29. In fact, the results of Theorem 24 and Theorem 25 are already proved in [13] (see Lemmas 2.9 and 2.5)
by completely different methods and based on the results of Ilyin [28,29]. Our intention was here to give a proof
which is straightforward and self-contained.
References
[1] H. Amann, Nonhomogeneous linear and quasilinear elliptic and parabolic boundary value problems, in: H.-J. Schmeisser, et al. (Eds.), Function Spaces, Differential Operators and Nonlinear Analysis, in: Teubner-Texte Math., vol. 133, 1993, pp. 9126.
[2] L. Adams, Z. Li, The immersed interface/multigrid methods for interface problems, SIAM J. Sci. Comput. 24 (2002) 463479.
[3] P. Auscher, P. Tchamitchian, Square root problem for divergence operators and related topics, Astrisque 249 (1998).

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

47

[4] U. Bandelow, H.-C. Kaiser, T. Koprucki, J. Rehberg, Modeling and simulation of strained quantum wells in semiconductor lasers, in:
W. Jger, H.-J. Krebs (Eds.), MathematicsKey Technology for the Future. Joint Projects Between Universities and Industry, Springer,
Berlin/Heidelberg/New York, 2003, pp. 377390.
[5] H. Beirao da Veiga, On the W 2,p -regularity for solutions of mixed problems, J. Math. Pures Appl. 53 (1974) 279290.
[6] H. Berestycki, F. Hamel, L. Roques, Analysis of the periodically fragmented environment model: Ispecies persistence, J. Math. Biol. 51
(2005) 75113.
[7] R. Bey, J.-P. Loheac, M. Moussaoui, Singularities of the solution of a mixed boundary problem for a general second order elliptic equation
and boundary stabilization of the wave equation, J. Math. Pures Appl. 78 (1999) 10431067.
[8] R. Brown, The mixed problem for Laplaces equation in a class of Lipschitz domains, Commun. Partial Differential Equations 19 (1994)
12171223.
[9] Z. Chen, J. Zou, Finite element methods and their convergence for elliptic and parabolic interface problems, Numer. Math. 79 (1998) 175202.
[10] M. Costabel, M. Dauge, Y. Lafranche, Fast semi-analytic computation of elastic edge singularities, Comput. Meth. Appl. Mech. Engrg. 190
(2001) 21112134.
[11] M. Dauge, Neumann and mixed problems on curvilinear polyhedra, Integral Equations Oper. Theory 15 (1992) 227261.
[12] C. Ebmeyer, J. Frehse, Mixed boundary value problems for nonlinear elliptic equations in multidimensional non-smooth domains, Math.
Nachr. 203 (1999) 4774.
[13] J. Elschner, H.-C. Kaiser, J. Rehberg, G. Schmidt, W 1,q regularity results for elliptic transmission problems on heterogeneous polyhedra,
Math. Models Methods Appl. Sci. 17 (2007) 593615.
[14] J. Elschner, J. Rehberg, G. Schmidt, Optimal regularity for elliptic transmission problems including C 1 interfaces, Interfaces Free Bound. 9
(2007) 233252.
[15] L.C. Evans, R.F. Gariepy, Measure Theory and Fine Properties of Functions, Studies in Advanced Mathematics, CRC Press, Boca Raton, New
York, London, Tokyo, 1992.
[16] P.C. Franzone, L. Guerri, S. Rovida, Wavefront propagation in an activation model of the anisotropic cardiac tissue: asymptotic analysis and
numerical simulation, J. Math. Biol. 28 (1990) 121176.
[17] H. Gajewski, Analysis und Numerik von Ladungstransport in Halbleitern (Analysis and numerics of carrier transport in semiconductors), Mitt.
Ges. Angew. Math. Mech. 16 (1993) 3557.
[18] E. Giusti, Metodi diretti nel calcolo delle variazioni, Unione Matematica Italiana, Bologna, 1994.
[19] A. Glitzky, R. Hnlich, Global estimates and asymptotics for electroreactiondiffusion systems in heterostructures, Appl. Anal. 66 (1997)
205226.
[20] J.A. Griepentrog, L. Recke, Linear elliptic boundary value problems with non-smooth data: Normal solvability on SobolevCampanato spaces,
Math. Nachr. 225 (2001) 3974.
[21] J.A. Griepentrog, Linear elliptic boundary value problems with non-smooth data: Campanato spaces of functionals, Math. Nachr. 243 (2002)
1942.
[22] K. Grger, A W 1,p -estimate for solutions to mixed boundary value problems for second order elliptic differential equations, Math. Ann. 283
(1989) 679687.
[23] J.A. Griepentrog, K. Grger, H.-C. Kaiser, J. Rehberg, Interpolation for function spaces related to mixed boundary value problems, Math.
Nachr. 241 (2002) 110120.
[24] P. Grisvard, Elliptic Problems in Nonsmooth Domains, Pitman, Boston, 1985.
[25] G. Harutjunjan, B.-W. Schulze, Parametrices of mixed elliptic problems, Math. Nachr. 277 (2004) 5682.
[26] M. Hieber, J. Rehberg, Quasilinear parabolic systems with mixed boundary conditions, WIAS-Preprint 1194, 2006.
[27] T. Hwang, W. Lin, J. Liu, W. Wang, JacobiDavidson methods for cubic eigenvalue problems, Numer. Linear Algebra Appl. 12 (2005)
605624.
[28] E.M. Ilyin, Singularities for solutions of elliptic boundary value problems with discontinuous highest order coefficients, Zap. Nauchn. Semin.
LOMI 38 (1973) 3345 (in Russian).
[29] E.M. Ilyin, Singularities of the weak solutions of elliptic equations with discontinuous higher coefficients. II. Corner points of the lines of
discontinuity, Zap. Nauchn. Semin. LOMI 47 (1974) 166169 (in Russian).
[30] D. Jerison, C. Kenig, The inhomogeneous Dirichlet problem in Lipschitz domains, J. Funct. Anal. 130 (1995) 161219.
[31] T. Kato, Perturbation Theory for Linear Operators, Reprint of the corr. print of the second ed. Classics in Mathematics, Springer,
Berlin/Heidelberg/New York, 1980.
[32] D. Knees, On the regularity of weak solutions of quasi-linear elliptic transmission problems on polyhedral domains, Zeitschr. Anal. Anwend. 23 (2004) 509546.
[33] T. Koprucki, H.-C. Kaiser, J. Fuhrmann, Electronic states in semiconductor nanostructures and upscaling to semi-classical models, in:
A. Mielke (Ed.), Analysis, Modeling and Simulation of Multiscale Problems, Springer, Berlin/Heidelberg/New York, 2006, pp. 367396.
[34] O.A. Ladyzhenskaya, N.N. Uraltseva, Linear and Quasilinear Elliptic Equations, Mathematics in Science and Engineering, Academic Press,
New York/London, 1968.
[35] D. Leguillon, E. Sanchez-Palenzia, Computation of Singular Solutions in Elliptic Problems and Elasticity, Wiley, Chichester, 1987.
[36] Y. Li, J. Liu, O. Voskoboynikov, C. Lee, S. Sze, Electron energy level calculations for cylindrical narrow gap semiconductor quantum dot,
Comp. Phys. Comm. 140 (2001) 399404.
[37] V. Mazya, Sobolev Spaces, Springer, Berlin/Heidelberg/New York, 1985.
[38] V. Mazya, J. Rossmann, Weighted Lp estimates of solutions to boundary value problems for second order elliptic systems in polyhedral
domains, Z. Angew. Math. Mech. 83 (2003) 435467.
[39] V. Mazya, J. Elschner, J. Rehberg, G. Schmidt, Solutions for quasilinear evolution systems in Lp , Arch. Rat. Mech. Anal. 171 (2004)
219262.

48

R. Haller-Dintelmann et al. / J. Math. Pures Appl. 89 (2008) 2548

[40] D. Mercier, Minimal regularity of the solutions of some transmission problems, Math. Meth. Appl. Sci. 26 (2003) 321348.
[41] I. Mitrea, M. Mitrea, The Poisson problem with mixed boundary conditions in Sobolev and Besov spaces in non-smooth domains, Trans.
Amer. Math. Soc. 359 (2007) 41434182.
[42] J. Rehberg, Quasilinear parabolic equations in Lp , in: M. Chipot, et al. (Eds.), Nonlinear elliptic and parabolic problems. A special tribute
to the work of Herbert Amann, in: Progress in Nonlinear Differential Equations and their Applications, vol. 64, Birkhuser, Basel, 2005,
pp. 413419.
[43] S. Rempel, B.-W. Schulze, Asymptotics for Elliptic Mixed Boundary Problems (Pseudo-Differential and Mellin Operators in Spaces with
Conormal Singularity), Math. Res., vol. 50, Akademie-Verlag, Berlin, 1989.
[44] G. Savar, Regularity and perturbation results for mixed second order elliptic problems, Commun. Partial Differential Equations 22 (1997)
869899.
[45] G. Savar, Regularity results for elliptic equations in Lipschitz domains, J. Funct. Anal. 152 (1998) 176201.
[46] S. Selberherr, Analysis and Simulation of Semiconductors, Springer, Wien, 1984.
[47] E. Shamir, Mixed boundary value problems for elliptic equations in the plane. The Lp theory, Ann. Sc. Norm. Super. Pisa, Sci. Fis. Mat., III.
Ser. 17 (1963) 117139.
[48] E. Shamir, Regularization of mixed second-order elliptic problems, Isr. J. Math. 6 (1968) 150168.
[49] A. Sommerfeld, Electrodynamics, Lectures on Theoretical Physics, vol. III, Academic Press, New York, 1952.
[50] A. Sommerfeld, Thermodynamics and Statistical Mechanics, Lectures on Theoretical Physics, vol. V, Academic Press, New York, 1956.
[51] B.A. Szabo, Z. Yosibash, Numerical analysis of singularities in two dimensions. Part II: Computation of generalized flux/stress intensity
factors, Int. J. Numer. Meth. Engrg. 39 (1996) 409434.
[52] W. Wang, T. Hwang, W. Lin, J. Liu, Numerical methods for semiconductor heterostructures with band nonparabolicity, J. Comp. Phys. 190
(2003) 141158.
[53] W. Wang, A jump condition capturing finite difference scheme for elliptic interface problems, SIAM J. Sci. Comp. 25 (2004) 14791496.
[54] C. Weisbuch, B. Vinter, Quantum Semiconductor Structures: Fundamentals and Applications, Academic Press, Boston, 1991.
[55] D. Zanger, The inhomogeneous Neumann problem in Lipschitz domains, Commun. Partial Differential Equations 25 (2000) 17711808.