News for hackers. Stuff that matters.
Re:Cubic Spline Interpolation by creata on Sunday October 22, 2023 @01:27PM (#37977322)
> in general perform much better than cubic splines

Source? (I couldn't find that in the paper you linked.)

Re:Cubic Spline Interpolation by rnburn on Sunday October 22, 2023 @03:11PM (#37978172)
The paper gives the rate of convergence you get with Polynomial interpolants in Chebyshev nodes:

> If f has v derivatives, with the vth derivative being of bounded variation V, then ||f - p_n|| = O(V n^{-v}) as n -> ∞

and

> If f is analytic, the convergence is geometric, with ||f - p_n|| = O(p^{-n}) for some p > 1

You will not get that good of a rate of convergence with cubic splines. See https://www.researchgate.net/publication/243095286_On_the_Or...

This is further explained in Trefethen's book https://www.amazon.com/Approximation-Theory-Practice-Applied...

Quoting from Ch 14

> In fact, polynomial interpolants in Chebyshev points are problem-free when evaluated by the barycentric interpolation formula. They have the same behavior as discrete Fourier series for period functions, whose reliability nobody worries about. The introduction of splines is a red herring: the true advantage of splines is not that they converge where polynomials fail to do so, but that they are more easility adapted to irregular point distributions and more localized.

You can see also the software package https://www.chebfun.org/ for Chebyshev interpolations with Matlab and https://github.com/rnburn/bbai for Chebyshev interpolation of arbitrary dimension functions with sparse grids for Python. And here is a quick notebook for an experiment you can run that will compare Chebyshev interpolants to cubic splines: https://github.com/rnburn/bbai/blob/master/example/13-sparse...

Re:Cubic Spline Interpolation by creata on Sunday October 22, 2023 @03:20PM (#37978245)
Thank you!
Re:Cubic Spline Interpolation by jacobolus on Sunday October 22, 2023 @07:34PM (#37980039)
However, in many circumstances, you have equispaced nodes (not Chebyshev nodes), for which Lagrange interpolation is terrible and it is literally impossible to make a global method which is without pitfalls (Trefethen has some other papers about this).

In some cases cubic splines are a fine tool.