{"id":"arxiv/2609.30046","url":"https://arxiv.org/abs/2609.30046","tags":["stat.ME","math.ST","stat.CO"],"score":0,"title":"Truly optimal low rank thin plate spline smoothing using a truncated Demmler-Reinsch basis","domain":"stat.ME","source":"arxiv","authors":["Paul Bach"],"reviews":0,"subject":"stat.ME","summary":"Thin plate splines are highly attractive smoothers. However, they have cubic computational cost, which severely limits their use in practice. As a remedy, Wood (2003) suggested thin plate regression splines (TPRS), which provide a low rank approximation. The key step of the TPRS approximation is a truncated eigendecomposition of the radial basis function (RBF) design matrix. However, as Wood (2003) writes, the optimality of the TPRS approximation is a slightly weak one. This is because the RBF coefficients are subject to orthogonality constraints and the TPRS approximation is only optimal if…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-09-24","reactions":{}}