In this paper, we present a meshless Galerkin scheme of boundary integral equations (BIEs), known as the Galerkin boundary node method (GBNM), for two-dimensional exterior Neumann problems that combines the moving least-squares (MLS) approximations and a variational formulation of BIEs. In this approach, boundary conditions can be implemented directly despite the MLS approximations lack the delta function property. Besides, the GBNM keeps the symmetry and positive definiteness of the variational problems. A rigorous error analysis and convergence study of the method is presented in Sobolev spaces. Numerical examples are also given to illustrate the capability of the method.Mathematics subject classification: 65N12, 65N30, 65N38.i Key words: Meshless, Galerkin boundary node method. Boundary integral equations. Moving least-squares. Error estimate. / 244 X.L. LI AND J.L. ZHU the boundary node method (BNM) [8], the boundary cloud method [9], the hybrid boundary node method [10], the boundary point interpolation method [11], the boundary element-free method [12] and the Galerkin boundary node method (GBNM) [13]. In contrast with the domain type methods, they are superior in treating problems dealing with infinite or semi-infinite domains. However, most boundary type meshless methods found in the literature lack a rich mathematical foundation to justify their use.The BNM is formulated using the moving least-squares (MLS) approximations [1,14] and the technique of BIEs. This method exploits the dimensionality of BIEs and the meshless attribute of the MLS. Nevertheless, since the MLS approximations lack the delta function property, the BNM cannot accurately satisfy boundary conditions. The strategy employed in the BNM [8] involves a new definition of the discrete norm used for the construction of the MLS approximations, which doubles the number of system equations.The GBNM is a boundary type meshless Galerkin method which combines the MLS scheme and a variational formulation of BIEs. Compared with the BNM, boundary conditions in the GBNM can be satisfied directly and system matrices are symmetric. The main difiFerence between the GBNM and the traditional Galerkin BEM is the way in which the shape function is formulated. In the GBNM, the boundary variables are approximated by the MLS technique that only use the boundary nodes, but in the Galerkin BEM, interpolants of the boundary variables are related to the geometry of the elements. The GBNM has been used for Dirichlet problems of Laplace equation [13] and biharmonic equation [15], and for Stokes fiow [16]. In this paper, the GBNM is further developed for solving 2-D exterior Neumann problems.As in many other meshless methods such as the EFGM and the BNM, background cells are used in the GBNM for numerical integration over the boundary. Cells are used for integration only, and have no restriction on shape or compatibility. The topology of cells can be much simpler than that of elements in the BEM or the FEM, since cells can be divided into smaller ones without affecting...