English

Polynomial approximation of noisy functions

Numerical Analysis 2025-07-08 v3 Numerical Analysis Statistics Theory Computation Statistics Theory

Abstract

Approximating a univariate function on the interval [1,1][-1,1] with a polynomial is among the most classical problems in numerical analysis. When the function evaluations come with noise, a least-squares fit is known to reduce the effect of noise as more samples are taken. The generic algorithm for the least-squares problem requires O(Nn2)O(Nn^2) operations, where N+1N+1 is the number of sample points and nn is the degree of the polynomial approximant. This algorithm is unstable when nn is large, for example nNn\gg \sqrt{N} for equispaced sample points. In this study, we blend numerical analysis and statistics to introduce a stable and fast O(NlogN)O(N\log N) algorithm called NoisyChebtrunc based on the Chebyshev interpolation. It has the same error reduction effect as least-squares and the convergence is spectral until the error reaches O(σn/N)O(\sigma \sqrt{{n}/{N}}), where σ\sigma is the noise level, after which the error continues to decrease at the Monte-Carlo O(1/N)O(1/\sqrt{N}) rate. To determine the polynomial degree, NoisyChebtrunc employs a statistical criterion, namely Mallows' CpC_p. We analyze NoisyChebtrunc in terms of the variance and concentration in the infinity norm to the underlying noiseless function. These results show that with high probability the infinity-norm error is bounded by a small constant times σn/N\sigma \sqrt{{n}/{N}}, when the noise {is} independent and follows a subgaussian or subexponential distribution. We illustrate the performance of NoisyChebtrunc with numerical experiments.

Keywords

Cite

@article{arxiv.2410.02317,
  title  = {Polynomial approximation of noisy functions},
  author = {Takeru Matsuda and Yuji Nakatsukasa},
  journal= {arXiv preprint arXiv:2410.02317},
  year   = {2025}
}
R2 v1 2026-06-28T19:06:42.843Z