Error Estimates for Gauss–Christoffel Quadrature under Reduced Regularity Conditions

2025

Read Error Estimates for Gauss–Christoffel Quadrature under Reduced Regularity Conditions (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-23540.txt.

Introduction

Mehdi Hamzehnejad and Abbas Salemi work on a key tool of applied math: rules for how to work out an integral to high accuracy. Such rules are judged by their error bounds. The bounds hinge on how fast a function's Chebyshev series fades out.

Older bounds ask a lot of the function near the ends of its range. Many real functions fail that test. They use a new result on the higher derivatives of Chebyshev polynomials. With it, they prove a sharper fade-out bound under a much weaker test. The error bound for the rule improves with it. The same ideas work for Gauss-Gegenbauer rules too. Tests with numbers back up the theory. More kinds of functions now come with error bounds a user can trust.

Catalog record

YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-23540

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Error Estimates for Gauss–Christoffel Quadrature under Reduced Regularity Conditions Abstract Gauss–Christoffel quadrature is a fundamental method for numerical integration, and its convergence analysis is closely related to the decay of Chebyshev expansion coefficients. Classical estimates, including those due to Trefethen, are based on weighted bounded variation assumptions involving the singular weight ( 1 − x 2 ) − 1 / 2 (1-x^{2})^{-1/2} , which may be too restrictive for functions with limited regularity at the endpoints. In this paper, we establish a new error bound for Gauss–Christoffel quadrature under weakened regularity assumptions. The analysis relies on a new identity for higher-order derivatives of Chebyshev polynomials. As a consequence, we obtain an improved decay estimate for Chebyshev coefficients, where the classical weighted condition V r = ∫ − 1 1 | f ( r + 1 ) ​ ( x ) | 1 − x 2 ​ 𝑑 x V_{r}=\int_{-1}^{1}\frac{|f^{(r+1)}(x)|}{\sqrt{1-x^{2}}}\,dx is replaced by the weaker condition U r = ∫ − 1 1 | f ( r + 1 ) ​ ( x ) | ​ 𝑑 x . U_{r}=\int_{-1}^{1}|f^{(r+1)}(x)|\,dx. This result leads to a corresponding error estimate for the Gauss–Christoffel quadrature rule, which is less restrictive than previous bounds. The approach is also extended to the Gauss–Gegenbauer case. Numerical experiments are provided to illustrate the theoretical results. Keywords: Gauss-Christoffel quadrature, Chebyshev coefficient, Error estimate, Reduced regularity, Numerical integration Mathematics Subject Classification (2020): 65D32, 65D30, 41A55, 42C10 Mehdi Hamzehnejad 1 and Abbas Salemi 2 1 Department of Mathematics, Graduate University of Advanced Technology, Kerman, Iran [email protected] 2 Department of Applied Mathematics, Shahid Bahonar University of Kerman, Kerman, Iran Received: date / Accepted: date arXiv preprint. Submitted for publication. 1 Introduction Numerical integration based on Gaussian quadrature plays a central role in scientific computing. The Gauss–Christoffel quadrature rule, associated with a weight function w(x) on [-1,1] , is given by \int_{-1}^{1}w(x)f(x)\,dx\simeq\sum_{i=1}^{N}w_{i}f(x_{i})=Q_{N}^{GC}[f], (1) and is well known for its high accuracy when applied to smooth integrands. A fundamental problem in its analysis is the derivation of reliable a priori error estimates that reflect the regularity of the integrand. A standard approach relates the quadrature error to the decay of the coefficients in orthogonal polynomial expansions, in particular the Chebyshev expansion [ 3 , 4 ] . If f(x)=\sum_{n=0}^{\infty}a_{n}T_{n}(x), then the quadrature error can be controlled in terms of the tail behaviour of the coefficients a_{n} . Consequently, sharp bounds for these coefficients play a central role in the convergence analysis of Gauss-type quadrature rules. Existing results and motivation A classical result due to Trefethen [ 3 , Theorem 4.2] states that if f,f^{\prime},\ldots,f^{(r-1)} are absolutely continuous and f^{(r)} is of bounded variation, and if V_{r}=\int_{-1}^{1}\frac{|f^{(r+1)}(x)|}{\sqrt{1-x^{2}}}\,dx<\infty, (2) then the Chebyshev coefficients satisfy |a_{n}|\leq\frac{2V_{r}}{\pi\prod_{j=0}^{r}(n-j)},\qquad n\geq r+1. (3) This estimate has served as the basis of several error bounds for Gauss-type quadrature rules. In particular, Xiang [ 1 ] derived the following estimate: \left|I[f]-Q_{N}^{GC}[f]\right|\leq\frac{4V_{r}\int_{-1}^{1}w(x)\,dx}{\pi r(2N+1)2N\cdots(2N-r+2)}. (4) The weighted condition ( 2 ) imposes control near the endpoints x=\pm 1 due to the singular factor (1-x^{2})^{-1/2} . This restriction may be overly strong for functions whose higher derivatives are integrable in the usual sense, but exhibit reduced regularity near the boundary. Several works [ 5 , 6 , 11 , 9 ] have refined constants and summation techniques, while largely retaining this structural assumption. Main contributions The objective of this work is to establish error bounds for Gauss–Christoffel quadrature under weakened regularity assumptions. In particular, we replace the weighted condition ( 2 ) by the absolute integrability condition U_{r}=\int_{-1}^{1}|f^{(r+1)}(x)|\,dx<\infty. (5) Our analysis is based on a new explicit identity for higher-order derivatives of Chebyshev polynomials, which allows a refined integration-by-parts argument in the estimation of the coefficients a_{n} . The main results of this paper can be summarized as follows: • We derive an improved bound for Chebyshev coefficients in terms of U_{r} (Theorem 2.3 ). • We obtain a corresponding error estimate for Gauss–Christoffel quadrature under these weakened assumptions (Theorem 3.1 ). • The analysis is extended to the Gauss–Gegenbauer case (Theorem 3.3 ). • Numerical experiments are presented to illustrate the theoretical results. Organization of the paper The remainder of the paper is organized as follows. Section 2 recalls basic properties of Chebyshev polynomials and establishes the improved coefficient bound. Section 3 develops the corresponding quadrature error estimates. Numerical illustrations are presented in Section 4 , and concluding remarks are given in Section 5 . 2 Improved upper bound for Chebyshev coefficients The Chebyshev polynomials of the first kind, T_{n}(x)=\cos\bigl(n\cos^{-1}x\bigr),\qquad n\geq 0, form an orthogonal system on [-1,1] with respect to the weight (1-x^{2})^{-1/2} , namely \int_{-1}^{1}\frac{T_{n}(x)T_{m}(x)}{\sqrt{1-x^{2}}}\,dx=\frac{\pi}{2}\,\delta_{mn}, where \delta_{mn} denotes the Kronecker delta. Moreover, T_{n} is an eigenfunction of the associated Sturm–Liouville problem [ 10 ] \sqrt{1-x^{2}}\bigl(\sqrt{1-x^{2}}T_{n}^{\prime}(x)\bigr)^{\prime}+n^{2}T_{n}(x)=0. (6) It is well known that Chebyshev polynomials satisfy the uniform bound |T_{n}(x)|\leq 1,\qquad x\in[-1,1],\;n\geq 0. (7) In addition, their derivatives obey the three-term relation [ 10 ] (1-x^{2})T_{n}^{\prime}(x)=\frac{n}{2}\bigl(T_{n-1}(x)-T_{n+1}(x)\bigr). (8) For a function f\colon[-1,1]\to\mathbb{R} , its Chebyshev expansion reads f(x)=\sum_{n=0}^{\infty}a_{n}T_{n}(x), with coefficients a_{0}=\frac{1}{\pi}\int_{-1}^{1}\frac{f(x)}{\sqrt{1-x^{2}}}\,dx,\qquad a_{n}=\frac{2}{\pi}\int_{-1}^{1}\frac{f(x)T_{n}(x)}{\sqrt{1-x^{2}}}\,dx,\quad n\geq 1. Obtaining sharp upper bounds for a_{n} hinges on a precise control of derivatives of Chebyshev polynomials in suitable weighted norms. Classical approaches, based on repeated integration by parts combined with coarse inequalities, typically lead to suboptimal constants. In contrast, we exploit the intrinsic Sturm–Liouville structure ( 6 ) together with the recurrence ( 8 ), which enables a finer analysis of derivative growth and ultimately sharper coefficient estimates. To normalize the dependence on the polynomial degree, we introduce the scaled quantity \mathcal{T}_{n}(x):=\frac{\sqrt{1-x^{2}}\,T_{n}^{\prime}(x)}{n^{2}}. (9) The next lemma constitutes the main technical ingredient of this section. It provides a recursive representation of \mathcal{T}_{n} in terms of derivatives of neighboring Chebyshev modes. Unlike classical two-term identities, the resulting expansion involves 2^{r} terms whose coefficients possess a precise monotonicity structure. Lemma 2.1 . Let \mathcal{T}_{n}(x) be defined by ( 9 ). For 1\leq r\leq n-1 , \mathcal{T}_{n} admits the representation \mathcal{T}_{n}(x)=\left(\frac{(-1)^{r}\mathcal{T}_{n-r}(x)}{\beta^{1}_{n,r}}+\frac{(-1)^{r-1}\mathcal{T}_{n-r+2}(x)}{\beta^{2}_{n,r}}+\cdots-\frac{\mathcal{T}_{n+r-2}(x)}{\beta^{2^{r-1}}_{n,r}}+\frac{\mathcal{T}_{n+r}(x)}{\beta^{2^{r}}_{n,r}}\right)^{(r)}, where \beta^{j}_{n,r}=\begin{cases}\displaystyle\prod_{i=1}^{r}(2n-2i+2),&j=1,2,\\[4.0pt] \displaystyle\prod_{i=1}^{r}(2n+2i-2),&j=2^{r}-1,2^{r}.\end{cases} Moreover, for 1\leq j\leq 2^{r-1} , \beta^{1}_{n,r}=\beta^{2}_{n,r}\leq\beta^{2j-1}_{n,r}=\beta^{2j}_{n,r}\leq\beta^{2^{r}-1}_{n,r}=\beta^{2^{r}}_{n,r}. Proof. Using ( 9 ), the Sturm–Liouville equation ( 6 ) can be rewritten as T_{n}(x)=-\sqrt{1-x^{2}}\,\mathcal{T}_{n}^{\prime}(x). (10) Similarly, ( 8 ) becomes \sqrt{1-x^{2}}\,\mathcal{T}_{n}(x)=\frac{1}{2n}\bigl(T_{n-1}(x)-T_{n+1}(x)\bigr). (11) Substituting ( 10 ) into ( 11 ) yields \sqrt{1-x^{2}}\,\mathcal{T}_{n}(x)=\frac{1}{2n}\bigl(-\sqrt{1-x^{2}}\,\mathcal{T}_{n-1}^{\prime}(x)+\sqrt{1-x^{2}}\,\mathcal{T}_{n+1}^{\prime}(x)\bigr). (12) For r=1 , \mathcal{T}_{n}(x)=\frac{1}{2n}\bigl(-\mathcal{T}_{n-1}^{\prime}(x)+\mathcal{T}_{n+1}^{\prime}(x)\bigr)=\left(\frac{-\mathcal{T}_{n-1}(x)}{2n}+\frac{\mathcal{T}_{n+1}(x)}{2n}\right)^{(1)}, (13) where \beta^{1}_{n,1}=\beta^{2}_{n,1}. Using ( 12 ) in two numerators of ( 13 ), we get the following representation for \mathcal{T}_{n}(x) \displaystyle\mathcal{T}_{n}(x) \displaystyle=\left(\frac{\mathcal{T}_{n-2}(x)}{\beta^{1}_{n,1}(2n-2)}-\frac{\mathcal{T}_{n}(x)}{\beta^{1}_{n,1}(2n-2)}-\frac{\mathcal{T}_{n}(x)}{\beta^{2}_{n,1}(2n+2)}+\frac{\mathcal{T}_{n+2}(x)}{\beta^{2}_{n,1}(2n+2)}\right)^{(2)} \displaystyle=\left(\frac{\mathcal{T}_{n-2}(x)}{\beta^{1}_{n,2}}-\frac{\mathcal{T}_{n}(x)}{\beta^{2}_{n,2}}-\frac{\mathcal{T}_{n}(x)}{\beta^{3}_{n,2}}+\frac{\mathcal{T}_{n+2}(x)}{\beta^{4}_{n,2}}\right)^{(2)}, (14) where \beta^{1}_{n,2}=\beta^{2}_{n,2}\leq\beta^{3}_{n,2}=\beta^{4}_{n,2}. The general case follows by induction by repeatedly inserting this identity into higher-order derivatives, which produces 2^{r} terms with the stated coefficient ordering. ∎ Lemma 2.1 reveals a hierarchical structure in the derivatives of Chebyshev polynomials, whereby higher-order derivatives are expressed as linear combinations of neighboring modes with explicitly controlled coefficients. Lemma 2.2 . Under the assumptions of Lemma 2.1 , \frac{1}{\beta^{1}_{n,r}(n-r)}+\frac{1}{\beta^{2}_{n,r}(n-r+2)}+\cdots+\frac{1}{\beta^{2^{r}}_{n,r}(n+r)}=\frac{1}{\prod_{j=0}^{r}(n-r+2j)}. Lemma 2.2 , the proof of which can be found in [ 12 ] , leads to a remarkable simplification of the expressions arising from Lemma 2.1 . Together, these results allow us to derive sharp upper bounds for Chebyshev coefficients using only interior information, thereby avoiding the loss of accuracy typically associated with crude norm estimates. We are now in a position to combine Lemma 2.1 and Lemma 2.2 to derive an improved upper bound for the Chebyshev coefficients. Theorem 2.3 . Suppose f,f^{\prime},\cdots,f^{(r-1)} are absolutely

Read the complete text