When is numerical integration needed instead of finding an antiderivative?

Many integrands have no antiderivative expressible in elementary functions, and in practice the integrand may be available only as data or as the output of a simulation. In both cases a quadrature rule estimates the integral directly from function values.

Why is Gaussian quadrature so efficient?

By optimally placing both the nodes and the weights, an n-point Gaussian rule integrates polynomials up to degree 2n-1 exactly — twice the degree of a Newton-Cotes rule with the same number of points — so it achieves high accuracy with few function evaluations for smooth integrands.

Numerical Integration

Numerical integration, or quadrature, approximates definite integrals by weighted sums of function values, providing accurate values when an antiderivative is unavailable or the integrand is known only at sample points.

Definition

Numerical integration is the approximation of a definite integral by a finite weighted combination of integrand values, called a quadrature rule, together with the analysis of its accuracy.

Scope

This area covers interpolatory quadrature rules built by integrating polynomial interpolants (Newton-Cotes), optimal-degree Gaussian rules based on orthogonal polynomials, composite and adaptive schemes that control error automatically, and the error analysis that governs accuracy and convergence; multidimensional integration is treated as an extension of these one-dimensional foundations.

Sub-topics

Core questions

How are quadrature rules constructed from polynomial interpolation, and what determines their accuracy?
What is the degree of exactness of a rule, and how do Gaussian rules maximize it for a given number of points?
How do composite and adaptive strategies control error across an interval?
How does the smoothness of the integrand govern the convergence rate of a quadrature rule?

Key theories

Interpolatory quadrature: Integrating the polynomial that interpolates the integrand at chosen nodes yields a quadrature rule whose weights are integrals of the Lagrange basis functions; the rule is exact for all polynomials up to the interpolation degree.
Gaussian quadrature and orthogonal polynomials: Choosing the nodes as the roots of orthogonal polynomials produces an n-point rule exact for polynomials up to degree 2n-1, the maximum possible, linking optimal quadrature to the theory of orthogonal polynomials.
Adaptive error control: Comparing estimates from rules of different orders or from refined subdivisions yields an error estimate that drives automatic subdivision, concentrating effort where the integrand varies rapidly.

Clinical relevance

Quadrature is required wherever integrals cannot be evaluated in closed form: computing expectations and normalizing constants in probability and statistics, evaluating element integrals in finite-element methods, summing radiative and force contributions in physics simulations, and pricing instruments in computational finance; the choice of rule trades accuracy against the number of (often expensive) integrand evaluations.

History

Classical interpolatory rules go back to Newton and Cotes, while Gauss introduced his optimal-degree quadrature in 1814; the computer era added automatic adaptive algorithms and high-quality software libraries, and renewed attention to the conditioning and stability of quadrature for difficult integrands.

Key figures

Carl Friedrich Gauss
Isaac Newton
Roger Cotes
Philip J. Davis

Seminal works

davis1984
quarteroni2007

Frequently asked questions

When is numerical integration needed instead of finding an antiderivative?: Many integrands have no antiderivative expressible in elementary functions, and in practice the integrand may be available only as data or as the output of a simulation. In both cases a quadrature rule estimates the integral directly from function values.
Why is Gaussian quadrature so efficient?: By optimally placing both the nodes and the weights, an n-point Gaussian rule integrates polynomials up to degree 2n-1 exactly — twice the degree of a Newton-Cotes rule with the same number of points — so it achieves high accuracy with few function evaluations for smooth integrands.