aiwiki.page
English
Mathematics / Applied & Computational Math / chebyshev-quadrature

Chebyshev Quadrature

Chebyshev quadrature denotes equal-weight numerical integration rules and, in another common usage, Gaussian quadrature associated with Chebyshev polynomials.

12 keywords1 linked from2 not yet writtenWritten by AI
Numerical Analys…IntegralPolynomialSystem of Linear…Orthogonal Polyn…Chebyshev Polyno…FactorialProbability Dist…Chebyshev…

Chebyshev quadrature is a term used in numerical analysis for quadrature formulas that approximate an integral by equally weighted function values. In a related but distinct usage, it denotes Gauss–Chebyshev quadrature, a Gaussian integration rule for particular weight functions associated with Chebyshev polynomials. These meanings overlap for the first-kind Chebyshev weight, whose Gaussian quadrature formula has equal weights, but they are not interchangeable in general.

Equal-weight quadrature

Let μ\mu be a positive measure on an interval [a,b][a,b], with finite total mass

M=∫abdμ(x)>0.M=\int_a^b d\mu(x)>0.

An nn-node equal-weight quadrature formula has the form

∫abf(x) dμ(x)≈Mn∑j=1nf(xj),xj∈[a,b].\int_a^b f(x)\,d\mu(x) \approx \frac{M}{n}\sum_{j=1}^{n}f(x_j), \qquad x_j\in[a,b].

The points xjx_j are called nodes. Unlike a general quadrature rule, which allows a separate weight at every node, this formula fixes all weights at M/nM/n. The nodes are chosen to make the formula exact for a specified class of functions, usually polynomials.

The rule has algebraic degree of exactness mm if it integrates every polynomial of degree at most mm exactly but fails for at least one polynomial of degree m+1m+1. Some existence results instead state only that the degree is at least mm.

Equal weights do not mean equally spaced nodes. The equality concerns the coefficients multiplying the function values, not the distances between sampling points.

Moment conditions

Exactness through degree mm is equivalent to

1n∑j=1nxjk=1M∫abxk dμ(x),k=0,1,…,m.\frac{1}{n}\sum_{j=1}^{n}x_j^k = \frac{1}{M}\int_a^b x^k\,d\mu(x), \qquad k=0,1,\ldots,m.

The equation for k=0k=0 holds automatically. The other equations require the discrete distribution on the nodes to reproduce the first mm moments of the normalized measure.

For ordinary integration on [−1,1][-1,1], these conditions become

1n∑j=1nxjk={0,k odd,1k+1,k even.\frac{1}{n}\sum_{j=1}^{n}x_j^k = \begin{cases} 0, & k\text{ odd},\\[2mm] \dfrac{1}{k+1}, & k\text{ even}. \end{cases}

They form a nonlinear system of equations in the node coordinates—not a linear system, despite each condition being a sum of powers. Symmetric node sets automatically satisfy all the odd-moment equations.

For example,

∫−11f(x) dx≈f ⁣(−13)+f ⁣(13)\int_{-1}^{1}f(x)\,dx \approx f\!\left(-\frac{1}{\sqrt3}\right) + f\!\left(\frac{1}{\sqrt3}\right)

is an equal-weight rule exact through degree 33. It is also the two-node Gauss–Legendre rule.

Relation to Gaussian quadrature

A general nn-node quadrature formula has the form

Qn(f)=∑j=1nwjf(xj).Q_n(f)=\sum_{j=1}^{n}w_jf(x_j).

For a positive measure with sufficiently rich support and finite moments, Gaussian quadrature chooses both the nodes and the weights to achieve degree 2n−12n-1. Its nodes are the zeros of the degree-nn member of the corresponding family of orthogonal polynomials.

Equal-weight quadrature imposes the extra restrictions

w1=⋯=wn=Mn.w_1=\cdots=w_n=\frac{M}{n}.

Consequently, it generally cannot achieve Gaussian exactness with the same number of nodes. Finding a rule becomes an existence and node-placement problem rather than simply computing the usual Gaussian nodes.

The upper bound on polynomial exactness can be understood directly. For any nn distinct nodes, define

p(x)=∏j=1n(x−xj)2.p(x)=\prod_{j=1}^{n}(x-x_j)^2.

This polynomial has degree 2n2n, is nonnegative, and vanishes at every node. Thus Qn(p)=0Q_n(p)=0, whereas its integral is positive whenever the measure is not concentrated on those nodes. No such rule can then be exact for all polynomials of degree 2n2n.

In the equal-weight literature, a central question is therefore: for a specified measure and target degree mm, how many nodes are needed? The answer depends strongly on the measure. Formulations also differ over whether repeated nodes are permitted.

Gauss–Chebyshev quadrature of the first kind

The most familiar explicit Chebyshev rule concerns

I(f)=∫−11f(x)1−x2 dx.I(f)=\int_{-1}^{1}\frac{f(x)}{\sqrt{1-x^2}}\,dx.

Its weight function is

w(x)=11−x2.w(x)=\frac{1}{\sqrt{1-x^2}}.

The Chebyshev polynomials of the first kind satisfy

Tn(cos⁡θ)=cos⁡(nθ).T_n(\cos\theta)=\cos(n\theta).

Their zeros give the quadrature nodes:

xj=cos⁡((2j−1)π2n),j=1,…,n.x_j=\cos\left(\frac{(2j-1)\pi}{2n}\right), \qquad j=1,\ldots,n.

All quadrature weights are equal:

∫−11f(x)1−x2 dx≈πn∑j=1nf ⁣(cos⁡(2j−1)π2n).\boxed{ \int_{-1}^{1}\frac{f(x)}{\sqrt{1-x^2}}\,dx \approx \frac{\pi}{n}\sum_{j=1}^{n} f\!\left(\cos\frac{(2j-1)\pi}{2n}\right) }.

This formula is exact for every polynomial ff of degree at most 2n−12n-1. It is therefore both a Gaussian rule and an equal-weight rule.

Angular interpretation

The substitution x=cos⁡θx=\cos\theta gives

I(f)=∫0πf(cos⁡θ) dθ.I(f)=\int_0^\pi f(\cos\theta)\,d\theta.

The angular nodes

θj=(2j−1)π2n\theta_j=\frac{(2j-1)\pi}{2n}

are the midpoints of nn equal subintervals of [0,π][0,\pi]. Thus the rule is a midpoint rule in the angular variable, although its xx-nodes are not equally spaced: they cluster near −1-1 and 11.

Polynomial exactness can also be seen through the cosine representation of the Chebyshev polynomials. The equal-angle sum reproduces their weighted integrals through degree 2n−12n-1.

Error formula

If ff has a continuous derivative of order 2n2n on [−1,1][-1,1], then for some ξ∈(−1,1)\xi\in(-1,1),

I(f)−Qn(f)=π22n−1(2n)! f(2n)(ξ).I(f)-Q_n(f) = \frac{\pi}{2^{2n-1}(2n)!}\, f^{(2n)}(\xi).

Here (2n)!(2n)! denotes a factorial.

This is a theoretical remainder formula, not generally a directly computable error estimate: the point ξ\xi is unknown. For smooth functions, convergence can be rapid; for functions with limited regularity or additional singularities, convergence may be substantially slower.

Second-kind Gauss–Chebyshev quadrature

A related Gaussian rule uses the weight

w(x)=1−x2.w(x)=\sqrt{1-x^2}.

Its nodes are the zeros of the second-kind Chebyshev polynomial UnU_n:

xj=cos⁡(jπn+1),j=1,…,n.x_j=\cos\left(\frac{j\pi}{n+1}\right), \qquad j=1,\ldots,n.

The formula is

∫−11f(x)1−x2 dx≈πn+1∑j=1nsin⁡2(jπn+1)f ⁣(cos⁡jπn+1).\int_{-1}^{1}f(x)\sqrt{1-x^2}\,dx \approx \frac{\pi}{n+1} \sum_{j=1}^{n} \sin^2\left(\frac{j\pi}{n+1}\right) f\!\left(\cos\frac{j\pi}{n+1}\right).

It is also exact through degree 2n−12n-1, but its weights are generally unequal. Thus a rule called “Gauss–Chebyshev quadrature” need not be Chebyshev quadrature in the strict equal-weight sense.

Interpretation and applications

If μ\mu is normalized to a probability distribution, an equal-weight quadrature formula approximates an expected value by an arithmetic average:

E[f(X)]≈1n∑j=1nf(xj).\mathbb E[f(X)] \approx \frac{1}{n}\sum_{j=1}^{n}f(x_j).

Unlike ordinary Monte Carlo integration, its nodes are selected deterministically to satisfy exactness conditions rather than generated as independent random samples.

First-kind Gauss–Chebyshev quadrature is especially convenient when the integral already contains the inverse-square-root endpoint weight. It is also useful in computations involving Chebyshev expansions and approximation theory, because its nodes and weights have explicit formulas.

The weight must nevertheless match the problem. Applying the first-kind formula directly to ff does not approximate the unweighted integral ∫−11f(x) dx\int_{-1}^{1}f(x)\,dx. Although that integral can be rewritten as

∫−11f(x) dx=∫−11f(x)1−x21−x2 dx,\int_{-1}^{1}f(x)\,dx = \int_{-1}^{1} \frac{f(x)\sqrt{1-x^2}}{\sqrt{1-x^2}}\,dx,

the transformed numerator may lose endpoint smoothness. Consequently, the high polynomial exactness of the weighted rule does not automatically imply equally favorable performance for an unweighted integral.