Orthogonal functions

In mathematics, orthogonal functions belong to a function space that is a vector space equipped with a bilinear form. When the function space has an interval as the domain, the bilinear form may be the integral of the product of functions over the interval:

${\displaystyle \langle f,g\rangle =\int {\overline {f(x)}}g(x),円dx.}${\displaystyle \langle f,g\rangle =\int {\overline {f(x)}}g(x),円dx.}

The functions ${\displaystyle f}${\displaystyle f} and ${\displaystyle g}${\displaystyle g} are orthogonal when this integral is zero, i.e. ${\displaystyle \langle f,,円g\rangle =0}${\displaystyle \langle f,,円g\rangle =0} whenever ${\displaystyle f\neq g}${\displaystyle f\neq g}. As with a basis of vectors in a finite-dimensional space, orthogonal functions can form an infinite basis for a function space. Conceptually, the above integral is the equivalent of a vector dot product; two vectors are mutually independent (orthogonal) if their dot-product is zero.

Suppose ${\displaystyle \{f_{0},f_{1},\ldots \}}${\displaystyle \{f_{0},f_{1},\ldots \}} is a sequence of orthogonal functions of nonzero L²-norms ${\textstyle \left\|f_{n}\right\|_{2}={\sqrt {\langle f_{n},f_{n}\rangle }}=\left(\int f_{n}^{2}\ dx\right)^{\frac {1}{2}}}${\textstyle \left\|f_{n}\right\|_{2}={\sqrt {\langle f_{n},f_{n}\rangle }}=\left(\int f_{n}^{2}\ dx\right)^{\frac {1}{2}}}. It follows that the sequence ${\displaystyle \left\{f_{n}/\left\|f_{n}\right\|_{2}\right\}}${\displaystyle \left\{f_{n}/\left\|f_{n}\right\|_{2}\right\}} is of functions of L²-norm one, forming an orthonormal sequence. To have a defined L²-norm, the integral must be bounded, which restricts the functions to being square-integrable.

Several sets of orthogonal functions have become standard bases for approximating functions. For example, the sine functions sin nx and sin mx are orthogonal on the interval ${\displaystyle x\in (-\pi ,\pi )}${\displaystyle x\in (-\pi ,\pi )} when ${\displaystyle m\neq n}${\displaystyle m\neq n} and n and m are positive integers. For then

${\displaystyle 2\sin \left(mx\right)\sin \left(nx\right)=\cos \left(\left(m-n\right)x\right)-\cos \left(\left(m+n\right)x\right),}${\displaystyle 2\sin \left(mx\right)\sin \left(nx\right)=\cos \left(\left(m-n\right)x\right)-\cos \left(\left(m+n\right)x\right),}

and the integral of the product of the two sine functions vanishes.^[1] Together with cosine functions, these orthogonal functions may be assembled into a trigonometric polynomial to approximate a given function on the interval with its Fourier series.

If one begins with the monomial sequence ${\displaystyle \left\{1,x,x^{2},\dots \right\}}${\displaystyle \left\{1,x,x^{2},\dots \right\}} on the interval ${\displaystyle [-1,1]}${\displaystyle [-1,1]} and applies the Gram–Schmidt process, then one obtains the Legendre polynomials. Another collection of orthogonal polynomials are the associated Legendre polynomials.

The study of orthogonal polynomials involves weight functions ${\displaystyle w(x)}${\displaystyle w(x)} that are inserted in the bilinear form:

${\displaystyle \langle f,g\rangle =\int w(x)f(x)g(x),円dx.}${\displaystyle \langle f,g\rangle =\int w(x)f(x)g(x),円dx.}

For Laguerre polynomials on ${\displaystyle (0,\infty )}${\displaystyle (0,\infty )} the weight function is ${\displaystyle w(x)=e^{-x}}${\displaystyle w(x)=e^{-x}}.

Both physicists and probability theorists use Hermite polynomials on ${\displaystyle (-\infty ,\infty )}${\displaystyle (-\infty ,\infty )}, where the weight function is ${\displaystyle w(x)=e^{-x^{2}}}${\displaystyle w(x)=e^{-x^{2}}} or ${\displaystyle w(x)=e^{-x^{2}/2}}${\displaystyle w(x)=e^{-x^{2}/2}}.

Chebyshev polynomials are defined on ${\displaystyle [-1,1]}${\displaystyle [-1,1]} and use weights ${\textstyle w(x)={\frac {1}{\sqrt {1-x^{2}}}}}${\textstyle w(x)={\frac {1}{\sqrt {1-x^{2}}}}} or ${\textstyle w(x)={\sqrt {1-x^{2}}}}${\textstyle w(x)={\sqrt {1-x^{2}}}}.

Zernike polynomials are defined on the unit disk and have orthogonality of both radial and angular parts.

Walsh functions and Haar wavelets are examples of orthogonal functions with discrete ranges.

Legendre and Chebyshev polynomials provide orthogonal families for the interval [−1, 1] while occasionally orthogonal families are required on [0, ∞). In this case it is convenient to apply the Cayley transform first, to bring the argument into [−1, 1]. This procedure results in families of rational orthogonal functions called Legendre rational functions and Chebyshev rational functions.

Solutions of linear differential equations with boundary conditions can often be written as a weighted sum of orthogonal solution functions (a.k.a. eigenfunctions), leading to generalized Fourier series.

• Eigenvalues and eigenvectors
• Hilbert space
• Karhunen–Loève theorem
• Lauricella's theorem
• Wannier function

• George B. Arfken & Hans J. Weber (2005) Mathematical Methods for Physicists, 6th edition, chapter 10: Sturm-Liouville Theory — Orthogonal Functions, Academic Press.
• Price, Justin J. (1975). "Topics in orthogonal functions". American Mathematical Monthly . 82: 594–609. doi:10.2307/2319690. Archived from the original on 2021年01月15日. Retrieved 2019年02月09日.
• Giovanni Sansone (translated by Ainsley H. Diamond) (1959) Orthogonal Functions, Interscience Publishers.

• Orthogonal Functions, on MathWorld.

• Functional analysis
• Types of functions

• Articles with short description
• Short description matches Wikidata