Analysis of an inexact Schwarz domain decomposition method in 3D

1 Introduction

• In COSMO, the Laplace problem on a union of overlapping atomic balls is used to compute electrostatic solvation energy; spherical-harmonic local solves make the ddCOSMO approach efficient for large molecules.Each local boundary problem is approximated using spherical harmonics up to a fixed degree N [1].Source: §1 · p. 1
• The work addresses the gap left by analyses that assume exact subdomain solves, extending the inexact-solver maximum-principle approach from 2D disks to 3D overlapping spheres.In ddCOSMO, boundary data on each sphere are projected onto spherical harmonics through degree N; the 3D truncated Legendre kernel oscillates and becomes negative near the boundary, unlike the exact Poisson kernel.Source: §1 · p. 2
• The inexact local solution operator is an integral operator whose kernel is the Poisson kernel projected onto spherical harmonics of degree at most N.This representation is established in Lemma 5.1.Source: §1 · p. 2
• For every odd N≥5, the truncated kernel K_N is nonnegative for all points with radius r≤r_N.The proof relates the kernel minimum to its value at the antipodal point and estimates zeros of the resulting closed form using the Lambert W function.Source: §1 · p. 2
• The inexact maximum principle bounds the local solution between the boundary-data minimum and maximum inside radius r_N, leaving a boundary layer of width O(ln N/N).The bound applies to piecewise continuous boundary data.Source: §1 · p. 2
• For odd N≤100, δ_th(N) exceeds δ_num(N) by less than a factor of two, and δ_num(N)/δ_th(N) increases with N, indicating O(ln N/N) is sharp.The experiments compute the largest radius r_num(N) for which K_N remains nonnegative.Source: §1 · p. 3
• The new maximum principle bounds the inexact Dirichlet-to-Dirichlet map in the maximum norm, and that map determines Schwarz-spherical convergence.The Schwarz-spherical method is formulated as a fixed-point iteration governed by inexact maps; the total error separates into discretization and iteration errors, and the paper studies the iteration error.Source: §1 · pp. 2–3

• The problem is to find a bounded solution of the Laplace equation in the unit ball with piecewise continuous boundary data.The data are assumed to have finitely many discontinuities, with left- and right-hand limits at those points; the discontinuity-free case is allowed.Source: §2 · p. 3
• Separation of variables yields a bounded solution represented through Legendre polynomials and spherical harmonics.The spherical harmonics form a complete orthonormal basis of L2 on the sphere.Source: §2 · pp. 3–4
• The Fourier-Laplace partial sums of the boundary data converge in L2 and pointwise away from the discontinuities.The section also gives the limiting behavior at discontinuities separately.Source: §2 · p. 5
• The solution can also be expressed using the 3D Poisson kernel, whose boundary limit recovers the prescribed data.The text identifies this representation as another solution of the same Laplace problem.Source: §2 · pp. 5–6
• The Legendre polynomials used in the derivation are obtained by expanding their generating function as a Taylor series.Source: §2 · p. 5

• The discussion invokes the maximum principle and a relation satisfied by the Poisson kernel.Source: §3 · p. 6
• The section introduces a comparison principle for solutions corresponding to piecewise continuous boundary data.Source: §3 · p. 6
• The comparison principle is framed for two boundary data functions and their respective Laplace-problem solutions.Source: §3 · p. 6

• The domain consists of two overlapping balls, with the shared boundary portions denoted Γ_i.Each ball has radius and center specified, and the boundary data are continuous on the domain boundary.Source: §4 · pp. 6–7
• The method uses an additive Schwarz iteration to approximately solve the Laplace problem on the two balls.Source: §4 · p. 7
• Dirichlet-to-Dirichlet maps transfer local solutions between the shared boundary portions of the balls.The maps are defined by restricting each ball's local solution to the other ball's interior boundary.Source: §4 · p. 7

5 Schwarz-spherical domain decomposition method

• The method projects boundary data on each sphere L2-orthogonally onto spherical harmonics of degree at most N, using the same fixed N for both balls and every iteration.The local approximate boundary data are finite sums of spherical harmonics.Source: §5 · p. 7
• The analysis separates total error into discretization and iteration components, but studies only the iteration error.It identifies accuracy of the discrete solution as a separate issue from convergence of the iterations.Source: §5 · p. 8

5.1 A variant of the maximum principle

• The truncated kernel can change sign near the boundary because weak damping fails to suppress oscillations of the Legendre polynomials.The passage describes this sign change as an instance of the Gibbs phenomenon.Source: §5.1 · p. 9
• For odd N ≥4, the projected Poisson kernel is nonnegative up to the radius r ≤ 1 − 2 ln(2(N+1))/(N+1).The theorem’s bound is sufficient for kernel positivity throughout the stated interior region.Source: §5.1 · pp. 9, 11, 12
• The paper derives a new maximum-principle variant for boundary data projected into the finite-dimensional spherical-harmonic space.The derivation uses the projected-kernel positivity result.Source: §5.1 · p. 13