Data gaps in ®nite-dimensional boundary value problems for satellite gradiometry A. Albertella, F. Migliaccio, F. Sanso DIIAR, Politecnico di Milano, Piazza Leonardo da Vinci, 32, 20133 Milano, Italy e-mail: alberta@ipmtf4.topo.polimi.it; Tel.: +39-02-2399-6504; Fax: +39-02-2399-6530 Received: 14 August 2000 / Accepted: 12 April 2001 Abstract. In the framework of a boundary value prob- lem BVP), when areas on the boundary are void of data the solution of the problem becomes undetermined and clearly more dicult. Physically, this could be the situation in which a gradiometer on a satellite on a perfectly circular orbit covers a sphere with measured second radial derivatives: if the satellite orbit is not polar, there are caps at satellite altitude which are not covered by data. A solution is presented based on an iterative algorithm, under the hypothesis of using a ®nite-dimensional model as is usually done in the time- wise approach. The convergence of the iterative solution is proved and a numerical example is shown to con®rm the theoretical result. Key words: Polar gap ± Finite-dimensional model 1 De®nition of the problem A boundary value problem BVP) has from the abstract point of view the same outlook as the problem of intersecting two straight lines in a plane or, a little more generally, as intersecting two complementary linear manifolds in R n . Basically this has the form: ®nd x 2 R n such that Ax a Bx b 1 where A and B are matrices; a and b are vectors with the same number of rows as A and B respectively. If each of Eqs. 1) does have solutions, i.e. there do exist two vectors x and e x such that A x a; Be x b then, after setting x x n; x e x g Equations 1) can be transformed into: ®nd n; g such that n g e x x 2 An 0 Bg 0 3 Basically Eqs. 2) and 3) mean that we want to decompose the known vector e x xinto two compo- nents n; gsuch that n belongs to the null space of A and g to that of B, i.e. e x x n  gn 2 N A ; g 2 N B 4 As we know, such a decomposition always exists and is unique if we have simultaneously N A N B R n N A \ N B f0g 5 If we relax one of the constraints of Eqs. 4), e.g. if we cancel one of the rows of Bx b, then N B gets larger and of course the ®rst of Eqs. 5) is still valid i.e. we guarantee the existence of the solution) but the second no longer needs to be true i.e. we loose the uniqueness). The opposite happens if we increase the number of constraints. In the framework of a BVP, Eqs. 1) can be rein- terpreted as described below. The variable x is an un- known element belonging to some to make it simple) Hilbert space H of functions de®ned on a possibly regular) open set X, typically a relatively compact set or, as in the geodetic case, the complement of a com- pact set. The operator A is a dierential operator sending H onto another Hilbert space L, to which a belongs. The Correspondence to: A. Albertella Journal of Geodesy 2001) 75: 641±646