An Exact Penalty Matrix for Cubic Smoothing Splines on Arbitary Knot Vectors

These notes derive the penalty matrix Ω of a cubic smoothing spline from first principles, for an arbitrary knot vector. Starting from the definition of the B-spline basis, we express differentiation as a matrix, integrate the squared second derivative in closed form, and arrive at an exact factorization Ω = \(C^{\mathsf{T}} R C\) built from the knot vector alone. Later sections derive the generalized cross-validation criterion for selecting the smoothing parameter, analyze the spectrum of the influence matrix, examine conditioning, and survey related software. These notes accompany an implementation in SciPy and are written so that any library can implement the same construction directly from them.