Abstract:This article develops a geodesic interpolation framework for data on the Grassmann manifold. The motivation is that many matrix-valued data sets represent subspaces rather than fixed bases: if the columns of two matrices differ only by a right orthogonal transformation, then they describe the same point on $Gr(r,m)$. Interpolation should therefore be invariant under this basis ambiguity. We first introduce $GLERP$, a Grassmann analogue of spherical linear interpolation defined by the Grassmann exponential and logarithm maps. The method follows the constant-speed geodesic joining two nearby subspaces and is second-order accurate for smooth Grassmann-valued curves under the usual normal-neighborhood condition. We then define $GIDER_n$, a recursive higher-order construction obtained by replacing each affine interpolation step in Neville's algorithm by $GLERP$. The resulting interpolant matches $n+1$ subspace data exactly, is basis-invariant, and is locally of order $n+1$ for sufficiently smooth curves. We compare the construction with tangent-space interpolation and projection-matrix interpolation, discuss intrinsic and extrinsic error measures, and present numerical tests. The results confirm the expected convergence orders and show that $GIDER_n$and tangent-space interpolation are nearly indistinguishable in a smooth local regime, while projection-matrix interpolation provides a useful extrinsic baseline. We also outline how the recursive construction can be used as the basis of a Grassmann ENO procedure.
From: Shingyu Leung [view email]
[v1]
Mon, 15 Jun 2026 09:32:23 UTC (23 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。