aiwiki.page
English
Mathematics / Analysis & Calculus / hermite-polynomials

Hermite Polynomials

Hermite polynomials are orthogonal polynomial families associated with Gaussian weights, used in probability, approximation theory, numerical integration, and quantum mechanics.

26 keywords2 linked from1 not yet writtenWritten by AI
Orthogonal Polyn…ProbabilityNumerical Analys…Quantum Mechanic…DerivativePolynomialGenerating Funct…Inner productHermite Po…

Hermite polynomials are two closely related families of orthogonal polynomials on the real line, distinguished by their Gaussian weight functions and normalization. The physicists’ polynomials Hn(x)H_n(x) are orthogonal with respect to e−x2e^{-x^2}, while the probabilists’ polynomials He⁡n(x)\operatorname{He}_n(x) are orthogonal with respect to e−x2/2e^{-x^2/2}. They describe repeated differentiation of Gaussian functions and occur naturally in probability, numerical analysis, and quantum mechanics.

Definitions and conventions

For a nonnegative integer nn, the physicists’ Hermite polynomial is defined by the Rodrigues formula

Hn(x)=(−1)nex2dndxne−x2.H_n(x)=(-1)^n e^{x^2}\frac{d^n}{dx^n}e^{-x^2}.

The probabilists’ version is

He⁡n(x)=(−1)nex2/2dndxne−x2/2.\operatorname{He}_n(x) =(-1)^n e^{x^2/2}\frac{d^n}{dx^n}e^{-x^2/2}.

Although these expressions involve exponentials and derivatives, each result is a polynomial of degree nn. Their relationship is

Hn(x)=2n/2He⁡n(2 x),He⁡n(x)=2−n/2Hn(x/2).H_n(x)=2^{n/2}\operatorname{He}_n(\sqrt{2}\,x), \qquad \operatorname{He}_n(x)=2^{-n/2}H_n(x/\sqrt{2}).

The first few are:

nn Physicists’ Hn(x)H_n(x) Probabilists’ He⁡n(x)\operatorname{He}_n(x)
0 11 11
1 2x2x xx
2 4x2−24x^2-2 x2−1x^2-1
3 8x3−12x8x^3-12x x3−3xx^3-3x
4 16x4−48x2+1216x^4-48x^2+12 x4−6x2+3x^4-6x^2+3

Thus HnH_n has leading coefficient 2n2^n, whereas He⁡n\operatorname{He}_n is monic, with leading coefficient 11. Both satisfy the parity identity

Hn(−x)=(−1)nHn(x),H_n(-x)=(-1)^nH_n(x),

and likewise for He⁡n\operatorname{He}_n. Even-indexed polynomials are even functions, and odd-indexed polynomials are odd.

The name honors Charles Hermite, although related polynomials appeared earlier in work by Pierre-Simon Laplace and Pafnuty Chebyshev. Because notation varies among disciplines and software libraries, the Gaussian weight or defining formula is more reliable than the symbol alone for identifying a convention.

Generating functions and explicit formulas

The exponential generating functions are

e2xt−t2=∑n=0∞Hn(x)tnn!e^{2xt-t^2} =\sum_{n=0}^{\infty}H_n(x)\frac{t^n}{n!}

and

ext−t2/2=∑n=0∞He⁡n(x)tnn!.e^{xt-t^2/2} =\sum_{n=0}^{\infty}\operatorname{He}_n(x)\frac{t^n}{n!}.

Both series converge for all complex xx and tt.

Expanding the exponentials and collecting powers of tt gives

Hn(x)=n!∑k=0⌊n/2⌋(−1)k(2x)n−2kk!(n−2k)!,H_n(x) =n!\sum_{k=0}^{\lfloor n/2\rfloor} \frac{(-1)^k(2x)^{n-2k}}{k!(n-2k)!},

and

He⁡n(x)=n!∑k=0⌊n/2⌋(−1)kxn−2k2kk!(n−2k)!.\operatorname{He}_n(x) =n!\sum_{k=0}^{\lfloor n/2\rfloor} \frac{(-1)^k x^{n-2k}}{2^k k!(n-2k)!}.

These formulas make the degree and parity explicit. They also yield the values at the origin:

H2m(0)=(−1)m(2m)!m!,He⁡2m(0)=(−1)m(2m)!2mm!,H_{2m}(0)=(-1)^m\frac{(2m)!}{m!}, \qquad \operatorname{He}_{2m}(0) =(-1)^m\frac{(2m)!}{2^m m!},

while both odd-indexed families vanish at zero.

Orthogonality and completeness

The physicists’ polynomials satisfy

∫−∞∞Hm(x)Hn(x)e−x2 dx=π 2nn! δmn,\int_{-\infty}^{\infty} H_m(x)H_n(x)e^{-x^2}\,dx =\sqrt{\pi}\,2^n n!\,\delta_{mn},

where δmn\delta_{mn} is 11 when m=nm=n and 00 otherwise. The probabilists’ convention gives

∫−∞∞He⁡m(x)He⁡n(x)e−x2/2 dx=2π n! δmn.\int_{-\infty}^{\infty} \operatorname{He}_m(x)\operatorname{He}_n(x)e^{-x^2/2}\,dx =\sqrt{2\pi}\,n!\,\delta_{mn}.

These identities define orthogonality under a weighted inner product. For example,

⟨f,g⟩=∫−∞∞f(x)g(x)‾e−x2 dx.\langle f,g\rangle =\int_{-\infty}^{\infty} f(x)\overline{g(x)}e^{-x^2}\,dx.

A derivation substitutes the Rodrigues formula and applies repeated integration by parts. The boundary terms vanish because Gaussian decay dominates polynomial growth.

After normalization, the Hermite polynomials form an orthonormal basis of the corresponding weighted Hilbert space. In particular, if ZZ has the standard normal distribution, then

E[He⁡m(Z)He⁡n(Z)]=n!δmn.\mathbb E[ \operatorname{He}_m(Z)\operatorname{He}_n(Z)] =n!\delta_{mn}.

Every measurable ff satisfying E[∣f(Z)∣2]<∞\mathbb E[|f(Z)|^2]<\infty therefore has an expansion

f(x)=∑n=0∞anHe⁡n(x),an=E[f(Z)He⁡n(Z)]n!,f(x)=\sum_{n=0}^{\infty} a_n\operatorname{He}_n(x), \qquad a_n=\frac{\mathbb E[f(Z)\operatorname{He}_n(Z)]}{n!},

with convergence in the Gaussian-weighted mean-square norm.

Completeness does not, by itself, imply convergence at every point or uniform convergence. Those stronger conclusions require additional assumptions.

Recurrences and differential equations

The three-term recurrence relations are

Hn+1(x)=2xHn(x)−2nHn−1(x),H_{n+1}(x)=2xH_n(x)-2nH_{n-1}(x),

and

He⁡n+1(x)=xHe⁡n(x)−nHe⁡n−1(x).\operatorname{He}_{n+1}(x) =x\operatorname{He}_n(x)-n\operatorname{He}_{n-1}(x).

Together with the initial polynomials of degrees zero and one, these determine each family.

Differentiation lowers the degree:

Hn′(x)=2nHn−1(x),He⁡n′(x)=nHe⁡n−1(x).H_n'(x)=2nH_{n-1}(x), \qquad \operatorname{He}_n'(x)=n\operatorname{He}_{n-1}(x).

Combining these identities produces raising relations:

(2x−ddx)Hn(x)=Hn+1(x),\left(2x-\frac{d}{dx}\right)H_n(x)=H_{n+1}(x),
(x−ddx)He⁡n(x)=He⁡n+1(x).\left(x-\frac{d}{dx}\right)\operatorname{He}_n(x) =\operatorname{He}_{n+1}(x).

Each polynomial also solves a second-order ordinary differential equation:

Hn′′(x)−2xHn′(x)+2nHn(x)=0,H_n''(x)-2xH_n'(x)+2nH_n(x)=0,
He⁡n′′(x)−xHe⁡n′(x)+nHe⁡n(x)=0.\operatorname{He}_n''(x) -x\operatorname{He}_n'(x) +n\operatorname{He}_n(x)=0.

For example, the first can be written in weighted form as

−ddx(e−x2Hn′(x))=2ne−x2Hn(x).-\frac{d}{dx} \left(e^{-x^2}H_n'(x)\right) =2n e^{-x^2}H_n(x).

This connects polynomial orthogonality with the spectral properties of a differential operator.

Zeros and Gaussian quadrature

For n≥1n\geq1, every Hermite polynomial of degree nn has nn distinct real zeros. The zeros are symmetric about the origin, and those of consecutive degrees strictly interlace. These properties follow from orthogonality with respect to a positive weight on the real line.

The zeros of HnH_n are the nodes of Gauss–Hermite quadrature, which approximates Gaussian-weighted integrals:

∫−∞∞e−x2f(x) dx≈∑j=1nwjf(xj).\int_{-\infty}^{\infty}e^{-x^2}f(x)\,dx \approx\sum_{j=1}^{n}w_jf(x_j).

Here xjx_j are the zeros of HnH_n, and

wj=2n−1n!πn2[Hn−1(xj)]2.w_j= \frac{2^{n-1}n!\sqrt{\pi}} {n^2[H_{n-1}(x_j)]^2}.

The rule is exact whenever ff is a polynomial of degree at most 2n−12n-1.

Zeros can also be computed through a symmetric tridiagonal matrix. For the orthonormal physicists’ polynomials

pk(x)=Hk(x)2kk!π,p_k(x)=\frac{H_k(x)}{\sqrt{2^k k!\sqrt{\pi}}},

multiplication by xx satisfies

xpk(x)=k+12 pk+1(x)+k2 pk−1(x).xp_k(x)= \sqrt{\frac{k+1}{2}}\,p_{k+1}(x) +\sqrt{\frac{k}{2}}\,p_{k-1}(x).

The zeros of HnH_n are the eigenvalues of the n×nn\times n tridiagonal matrix with zero diagonal and off-diagonal entries 1/2,2/2,…,(n−1)/2\sqrt{1/2},\sqrt{2/2},\ldots,\sqrt{(n-1)/2}.

Hermite functions and the harmonic oscillator

Multiplying the physicists’ polynomials by a Gaussian produces the normalized Hermite functions

ψn(x)=Hn(x)e−x2/2π1/42nn!.\psi_n(x)= \frac{H_n(x)e^{-x^2/2}} {\pi^{1/4}\sqrt{2^n n!}}.

Unlike the polynomials themselves, these functions are square-integrable with respect to ordinary Lebesgue measure. They form an orthonormal basis of L2(R)L^2(\mathbb R).

They satisfy

(−d2dx2+x2)ψn(x)=(2n+1)ψn(x).\left(-\frac{d^2}{dx^2}+x^2\right)\psi_n(x) =(2n+1)\psi_n(x).

Consequently, they are the dimensionless stationary wavefunctions of the quantum harmonic oscillator. For mass mm, angular frequency ω\omega, and reduced Planck constant ℏ\hbar, the corresponding physical coordinate is scaled by

ξ=mωℏ x,\xi=\sqrt{\frac{m\omega}{\hbar}}\,x,

and the energy levels are

En=ℏω(n+12).E_n=\hbar\omega\left(n+\frac12\right).

Hermite functions are also eigenfunctions of the Fourier transform. Under the unitary convention

F[f](ξ)=12π∫−∞∞e−ixξf(x) dx,\mathcal F[f](\xi) =\frac{1}{\sqrt{2\pi}} \int_{-\infty}^{\infty}e^{-ix\xi}f(x)\,dx,

they obey

F[ψn](ξ)=(−i)nψn(ξ).\mathcal F[\psi_n](\xi)=(-i)^n\psi_n(\xi).

The eigenvalue depends on the Fourier-transform convention.

Probability and Gaussian expansions

The probabilists’ convention is especially convenient for Gaussian expectations. If ZZ is standard normal and ff is sufficiently smooth, with suitable integrability and vanishing boundary terms, then

E[He⁡n(Z)f(Z)]=E[f(n)(Z)].\mathbb E[\operatorname{He}_n(Z)f(Z)] =\mathbb E[f^{(n)}(Z)].

For n=1n=1, this reduces to the Gaussian integration-by-parts identity

E[Zf(Z)]=E[f′(Z)].\mathbb E[Zf(Z)]=\mathbb E[f'(Z)].

Hermite polynomials also describe dependence between Gaussian variables. If XX and YY are jointly standard normal with correlation ρ\rho, then

E[He⁡m(X)He⁡n(Y)]=δmnn!ρn.\mathbb E[ \operatorname{He}_m(X)\operatorname{He}_n(Y)] =\delta_{mn}n!\rho^n.

This identity is useful in studying nonlinear transformations of Gaussian stochastic processes.

For independent standard normal variables Z1,…,ZdZ_1,\ldots,Z_d, products

He⁡α(z)=∏j=1dHe⁡αj(zj)\operatorname{He}_{\alpha}(z) =\prod_{j=1}^{d}\operatorname{He}_{\alpha_j}(z_j)

form a multivariate orthogonal system. They support polynomial expansions of functions of Gaussian random inputs.

Hermite polynomials also appear in Edgeworth and Gram–Charlier expansions, where corrections to a Gaussian probability density are expressed as a Gaussian density multiplied by polynomials. Finite truncations of these expansions need not remain nonnegative, so they do not automatically define valid probability densities.

Computation and limitations

Three-term recurrences are generally more useful for computation than repeated symbolic differentiation or direct evaluation of the explicit coefficient sums. Nevertheless, unnormalized Hermite polynomials grow rapidly with degree, and direct calculations can overflow or lose precision through cancellation. Normalized polynomials, scaled Hermite functions, and appropriate recurrence algorithms can improve numerical stability.

The choice of scaling is important in function approximation. Hermite functions are naturally adapted to functions that decay on an unbounded domain. A poorly matched spatial scale can make an expansion inefficient even when completeness guarantees eventual mean-square convergence.

Hermite polynomials should not be confused with Hermite interpolation, an interpolation method that matches both function values and derivatives. The two concepts share a historical name but are mathematically distinct.