Introduction

The determination of the Earth’s external gravity field is usually formulated in terms of various types of Geodetic Boundary Value Problems (GBVPs) for the Laplace equation. Most investigations on GBVPs have been motivated by the need to approximate closer and closer the physical reality and to finding more accurate and reliable procedures to handle a variety of available gravity field related data. During the last thirty years, three kinds of the AltimetryGravimetry Boundary Value Problem (AGBVP) have been defined according to the type of input data over the sea part and the land part of the Earth’s surface. Nowadays, with the help of the satellite altimetry data we can estimate the Mean Sea Surface (MSS), where we are able to evaluate the disturbing potential. In land areas, we can have gravimetric data at points with precisely determined 3D positions provided by the Global Navigation Satellite Systems (GNSS) which yield surface gravity disturbances. Therefore the surface of the Earth can be considered as a fixed boundary. This situation can be formulated as a linear, fixed mixed boundary value problem with a Dirichlet condition imposed in the sea areas and an oblique derivative condition on land, as treated e.g. by Keller (1996) and is also known as AGBVP-III. In this paper, we analyze the problem in an unbounded Lipschitz domain representing the exterior of the Earth. The Stampacchia theorem enables us to decide upon the existence and uniqueness of the weak solution of the problem in a weighted Sobolev space. Finally, we briefly discuss the regularity of the solution of the problem.

T W

Formulation of the problem
Let Ω ⊂ R3 be the exterior of the Earth whose boundary ∂Ω is the surface of the Earth. The boundary of Ω is decomposed into two parts, sea and land, as ∂Ω = ∂ΩS ∪ ∂ΩL, where ¯ ∂ΩS ∩ ∂ΩL = Ø. Putting Ω = R3 − Ω, we will suppose that Ω is the so-called domain with a ¯ Lipschitz boundary which has an outward unit normal vector n almost everywhere (Ω = Ω ∪ ∂Ω). The problem is to find the function u (disturbing potential T ) in Ω such that ∆u = 0 in Ω u = fS on ∂ΩS (sea) (h · u) = −δg on ∂ΩL (land ) u(x) → 0 as |x| → ∞ (1) (2) (3) (4)

In (2), fS represents the disturbing potential on MSS evaluated by satellite altimetry data. In (3), the gravity disturbance δg is the ellipsoidal normal component h of the gravity disturbance vector δg = T (the deflection of the vertical is neglected). This equation represents an oblique derivative boundary condition because in general the normal n to the Earth’s surface does not coincide with the direction of the unit vector h. We will continue making the quite reasonable assumption that (h · n) ≥ c > 0 on ∂ΩL. This just means that the vector field h is non tangential to ∂ΩL for almost all x ∈ ∂ΩL. Let h, n and a be continuous vector fields on ∂ΩL. Then one has h = n + a. (h · n) This leads to the following equivalent formulation of the boundary condition (3) (n · u) + (a ·
∂ΩL u)

= −fL on ∂ΩL
∂ΩL

(5)
δg (h·n) .

where a is tangent to ∂ΩL, i.e. (a · n) = 0 on ∂ΩL,

is the gradient on ∂ΩL and fL =

G. Panou, N. Yiannakakis and D. Delikaraoglou "An analysis of the linear fixed altimetry-gravimetry boundary value problem". Poster presentation at the European Geophysical Union General Assembly, Vienna, 3-8 April 2011.

Basic tools
Lipschitz domain • A function ϕ : R2 → R is Lipschitz if there exists a constant m > 0 such that for all x, y in R2, |ϕ(x) − ϕ(y)| ≤ m |x − y|. • Ω is a Lipschitz domain if ∂Ω locally is given by the graph of a Lipschitz function ϕ. The exterior of the Earth is an example of a Lipschitz domain and also is an unbounded domain. ¯ In the sequel, Ω = R3 − Ω is a star-shaped domain at the origin with the Lipschitz boundary. Function spaces ∞ • C ∞(Ω) is the space of functions with continuous partial derivatives of any order and C0 (Ω) is the space of C ∞(Ω) functions with compact support in Ω. • L2(∂Ω) is the space of square integrable functions on ∂Ω. Following Holota (1997), we work with the weighted Sobolev space the inner product given by uv (u, v )1 ≡ ( u · v )d x. 2d x + Ω Ω |x|
1/2 (u, u)1 (1) W2

that is equipped with

This product induces the norm ≡ u 1. (1) W2 is a space produced by functions which are square integrable on Ω under the weight |x|−2 and have derivatives of the first order in a certain generalized sense. It is obvious that the condition (1) (4) follows automatically from the properties of the weighted space W2 . Let H be a Hilbert space with norm · and inner product (·, ·). Let H ∗ be the dual of H, the pairing between H ∗ and H is denoted by ·, · . Theorem 1. (Stampacchia, [3]). Let A : H × H → R be a bilinear continuous and coercive form (not necessarily symmetric), i.e. there exist positive constants c1 and c2 such that |A(u, v )| ≤ c1 u and A(v , v ) ≥ c2 v ∀ v ∈ H. Let K be a non empty, closed and convex subset of H and F ∈ H ∗. Then there exists a unique u ∈ K such that A(u, v − u) ≥ F , v − u ∀ v ∈ K. (1) Note that when K = H, (6) is equivalent to A(u, v ) = F , v for all v ∈ H. This leads to the well-known Lax-Milgram theorem. Theorem 2. (trace theorem, [1]). Let Ω be our unbounded domain. Then there exists a (1) continuous linear operator Z : W2 → L2(∂Ω) such that ¯ ∩ W (1) Z (u) = u|∂Ω ∀ u ∈ C (Ω)
2 2

v

∀ u, v ∈ H

and Z (u) L2(∂Ω) = u Z (u) is called the trace of u on ∂Ω.
L2(∂Ω) ≤ c3 u 1

∀ u ∈ W2 .
(1) W2

(1)

Theorem 3. (equivalent norms, [1]). Let Ω be our unbounded domain. Then for u ∈ norm u 1 and
1/2

the

u =

| u| d x
1

2

are equivalent, i.e. there exist positive constants c4 and C4 such that c4 u

≤ u ≤ C4 u 1 .

Solvability
Let K = u ∈
(1) W2 (Ω)

: Z (u) = fS

¯ on ∂ΩS , V = v ∈ C ∞(Ω) : v = 0 on ∂ΩS and ( u· v )d x −
∂ΩL (1) W2 (Ω).

A(u, v ) =

(a ·

∂ΩL u)vd σ.

K is non empty, closed and convex subset of Lemma 1. For all v ∈ V we have

A(u, v ) = F , v =
∂ΩL

fLvd σ.

Proof. If w = u ± v ⇒ Z (w ) = Z (u) ± Z (v ) = Z (u) = fS ⇒ w ∈ K . Setting v = u + v in the inequality (6), results in A(u, v ) ≥ F , v and setting v = u − v , it follows that A(u, v ) ≤ F , v . Consequently, we have the claimed equality. (1) Definition 1. The function u ∈ W2 (Ω) is called a weak solution of the problem if u ∈ K and A(u, v ) = F , v ( u·

∀ v ∈V =
∂ΩL

⇔ fLvd σ ∀ v ∈ V . (1)

v )d x −
∂ΩL

(a ·

∂ΩL u)vd σ

¯ Lemma 2. If u ∈ C 2(Ω) is a weak solution, then it is a classical solution. ¯ Proof. Since u ∈ C 2(Ω) is a weak solution we have that u ∈ K , Z (u) = fS on ∂ΩS ⇒ u = fS on ∂ΩS , [validity of the boundary conditions (2) and (4)]. One can use Green’s theorem to transform equation (7) to ∆uvd x +
Ω ∂Ω

(n ·

u)vd σ +
∂ΩL

[(a ·

∂ΩL u)

+ fL]vd σ = 0 ∀ v ∈ V .

(2)

∞ Since u has to fulfill this for all v ∈ V in particular the equality is valid for all v ∈ C0 (Ω). ∞ Hence, ∆u = 0 in Ω [C0 (Ω) is dense in L2(Ω)]. Now the result in (8) is

[(n ·
∂ΩL

u) + (a ·

∂ΩL u)

+ fL]vd σ = 0 ∀ v ∈ V .

Consequently, u has to fulfill also the boundary condition (3) and therefore, u is a classical solution [V is dense in L2(∂Ω)]. From H¨lder’s inequality and (trace) theorem 2 it follows that o | F , v | ≤ fL
L2(∂Ω)

v

L2(∂Ω)

≤ c 3 fL

L2(∂Ω) (1)

v

1

∀ v∈

(1) W2

and we see that F , v is a continuous functional on W2 . One can show that |A(u, v )| ≤ c1 u Hence, A(u, v ) is continuous on
(1) W2 1 (1) W2 .

v

1

∀ u, v ∈ W2 .
(1) W2 .

(1)

×

In order to apply Stampacchia’s theorem, it is only left to show that A(u, v ) is coercive on First of all we have 1 1 2 (a · ∂ΩL v )vd σ = a · ∂ΩL (v )d σ = − ( ∂ΩL · a)v 2d σ 2 ∂ΩL 2 ∂ΩL ∂ΩL and using theorem 3 1 (1) 2 2 2 2 A(v , v ) = | v | d x + ( ∂ΩL · a)v d σ ≥ c4 v 1 ∀ v ∈ W2 2 ∂ΩL Ω if (3) ∂ΩL · a = div∂ΩL (a) > 0. As last comment, the condition (9) has an interpretation from the geometric point of view
∂ΩL

· a ≈ 2(ke − k∂ΩL )

where ke is the mean curvature of the ellipsoid and k∂ΩL is the mean curvature of ∂ΩL.

Regularity
Solutions of mixed boundary value problems in Lipschitz domains may have singularities on the boundary at such points where the boundary condition is changing or where ∂Ω is not smooth. It is also known that the solutions of the mixed problems are not smooth in general, regardless of how regular the data may be. In order to recover more regular solutions, one has to impose some compatibility conditions between the data. On the other hand, it is interesting to know what is the maximal regularity of the solution that is allowed by the data without assuming any particular compatibility on them. Since we are considering a harmonic function u and sufficiently regular data, our solution is smooth in Ω. Our interest is in obtaining regularity on the Lipschitz boundary. In particular, we examine the problems arising in a neighborhood of the boundary between ∂ΩS and ∂ΩL. In order to provide an understanding of what is really happening we may consider an analogous problem in R2. The typical counterexample is given by the harmonic function u(x, y ) = Im(x + iy )1/2, or in polar coordinates U(r , θ) = r 1/2 sin(θ/2) which satisfies a homogeneous Dirichlet (Neumann) condition on the positive (negative) real line. In this example, one can easily show that when the inner angle between ∂ΩS and ∂ΩL is strictly less than π, the situation is better, while the gain of regularity vanishes as the angle approaches π. x ∈ R, y > 0

Conclusions
The Stampacchia theorem gave us a possibility to clarify the existence and uniqueness of the weak solution of our linear fixed mixed boundary value problem. The condition of validity for such theorem has an interpretation from the geometric point of view. Typically h is the opposite direction of the normal gravity and directed fairly close to the normal n to the Earth’s surface. In turn the magnitude of a is small in the average (apart from some extreme cases in mountainous areas). In the case that h = n we can put a = 0 and thus, clearly, the resulting Dirichlet-Neumann problem has a unique solution. Finally, it can happen that the solution is more regular when we have a Lipschitz boundary rather than a smooth boundary. This is of crucial importance for our problem, since the Lipschitz boundary is already general enough to represent, with a small degree of idealization, the Earth’s surface.

Acknowledgements
This work is supported by the Basic Research Grant No. 65/1846 (ΠEBE 2010) of the National Technical University of Athens (NTUA). The authors wish to thank Prof. V.-G. Papanicolaou, Department of Mathematics, NTUA and Dr. G. Manoussakis, Department of Surveying Engineering, NTUA for many useful discussions and their comments on the various aspects of this research.

References
[1] P. Holota. Coerciveness of the linear gravimetric boundary-value problem and a geometrical interpretation. J. Geod., 71(10): 640–651, 1997. [2] W. Keller. On a scalar fixed altimetry-gravimetry boundary value problem. J. Geod., 70(8): 459–469, 1996. [3] J. L. Lions and G. Stampacchia. Variational inequalities. Comm. Pure Appl. Math., 20(3): 493–519, 1967. [4] G. Savar´. Regularity and perturbation results for mixed second order elliptic problems. Comm. Partial Diff. Eq., e 22(5&6): 869–899, 1997.