Rodrigues' formula
In mathematics, Rodrigues' formula (formerly called the Ivory–Jacobi formula) generates the Legendre polynomials.
≈ 8 min read
Names in other languages
No names in other languages are held for this entry yet.
Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.
In mathematics, Rodrigues' formula (formerly called the Ivory–Jacobi formula) generates the Legendre polynomials. It was independently introduced by , and . The name "Rodrigues formula" was introduced by Heine in 1878, after Hermite pointed out in 1865 that Rodrigues was the first to discover it. The term is also used to describe similar formulas for other orthogonal polynomials. describes the history of the Rodrigues formula in detail.
Statement
Let (P_n(x))_{n=0}^\infty be a sequence of orthogonal polynomials on the interval [a, b] with respect to weight function w(x) . That is, they have degrees \deg(P_n) = n , satisfy the orthogonality condition \int_a^b P_m(x) P_n(x) w(x) \, dx = K_n \delta_{m,n} where K_n are nonzero constants depending on n , and \delta_{m,n} is the Kronecker delta. The interval [a, b] may be infinite in one or both ends. {{Math proof|title=Proof |proof= When n = 0 , it is trivial.
When n = 1 , it simplifies to AP_1' = A'P_1 , which is true since P_1 = \frac{c_1}{w}(Bw)' = c_1A . So assume n \geq 2 . Define I_n(x) = \frac{d^n}{dx^n}(B^n(x) w(x)) , then by direct computation and simplification, the equation to be proven is equivalent to \frac{d^2}{dx^2} (B(x) I_n(x)) - \frac{d}{dx} (A(x) I_n(x)) + \lambda_n I_n(x) = 0 By Leibniz differentiation rule, we have B(x) \frac{d^n}{dx^n} y = \frac{d^n}{dx^n} (B(x) y) - n \frac{d^{n-1}}{dx^{n-1}} (B'(x) y) + \frac{n(n-1)}{2} \frac{d^{n-2}}{dx^{n-2}} (B y) A(x) \frac{d^n}{dx^n} y = \frac{d^n}{dx^n} (A(x) y) - n \frac{d^{n-1}}{dx^{n-1}} (A' y) for arbitrary y . This allows us to move A(x), B(x) to the other side of the n -th derivative.
Set y = B^n(x) w(x) , and define J(x) = \frac{d^2}{dx^2} (B(x) y(x)) - n \frac{d}{dx} (B'(x) y(x)) + \frac{n(n-1)}{2} B y(x) K(x) = -\frac{d}{dx} (A(x) y(x)) + n A' y(x) L(x) = \lambda_n y(x) Then the equation simplifies to \frac{d^n}{dx^n} (J+K + L) = 0 J(x) has three terms, call them in order J_1(x), J_2(x), J_3(x) . K(x) has two terms, call them in order K_1(x), K_2(x) . J_3(x) + K_2(x) + L(x) = (\lambda_n + \frac{n(n-1)}{2} B + n A')y=0 . That J_1(x) + J_2(x) + K_1(x) = 0 . follows from first writing J_1(x) as J_1(x) = \frac{d^2}{dx^2} \left(B^n(x) \int \exp\left(\frac{A(x)}{B(x)}\right)dx \right) and then taking the innermost first derivative to obtain J_1(x) = \frac{d}{dx}\left[\bigg(nB'(x)B^{n-1}(x) + A(x)B^{n-1}(x)\bigg)\int \exp\left(\frac{A(x)}{B(x)}\right)dx\right] and then rewriting this as J_1(x) = \frac{d}{dx}\Big(nB'(x)B^{n}(x)w(x)+ A(x)B^{n}(x)w(x)\Big) The first term is the negative of J_2(x) and the second term is the negative of K_1(x) . }} More abstractly, this can be viewed through Sturm–Liouville theory.
Define an operator Lf := - \frac{1}{w} (Wf')' , then the differential equation is equivalent to LP_n = \lambda_n P_n . Define the functional space X = L^2([a,b], w(x)dx) as the Hilbert space of functions over [a, b] , such that \langle f, g\rangle := \int_a^b fgw . Then the operator L is self-adjoint on functions satisfying certain boundary conditions, allowing us to apply the spectral theorem.
G(x,u)=\sum_{n=0}^\infty u^nP_n(x) The P_n(x) functions here may not have the standard normalizations. But we can write this equivalently as G(x,u)=\sum_{n=0}^\infty \frac{u^n}{N_n}N_nP_n(x) where the N_n are chosen according to the application so as to give the desired normalizations.
The variable u may be replaced by a constant multiple of u so that G(x,\alpha u)=\sum_{n=0}^\infty \frac{\alpha^n u^n}{N_n}N_nP_n(x) This gives an alternate form of the generating function. By Cauchy's integral formula, Rodrigues’ formula is equivalent to P_n(x)=\frac{n!}{2\pi i}\frac{c_n}{w(x)}\oint_C \frac{B^n(t) w(t)}{(t-x)^{n+1}}\,dt where the integral is along a counterclockwise closed loop around x . Let u=\frac{t-x}{B(t)} Then the complex path integral takes the form P_n(x)=\frac{n!}{2\pi i}c_n\oint_C \frac{G(x,u)}{u^{n+1}}\,du G(x,u)=\frac{w(t)\frac{dt}{du}}{w(x)B(t)} where now the closed path C encircles the origin.
In the equation for G(x,u) , t is an implicit function of u . Expanding G(x,u) in the power series given earlier gives \frac{1}{2\pi i}\oint_C \frac{G(x,u)}{u^{n+1}}\,du=\frac{1}{2\pi i}\oint_C \frac{\sum_{m=0}^\infty u^mP_m(x)}{u^{n+1}}\,du=P_n(x) Only the m=n term has a nonzero residue, which is P_n(x) . The n!\,c_n coefficient was dropped since normalizations are conventions which can be inserted afterwards as discussed earlier. By expressing t in terms of u in the general formula just given for G(x,u) , explicit formulas for G(x,u) may be found.
As a simple example, let B(x)=1 and A(x)=-x (Hermite polynomials) so that w(x)=\exp\left(-\frac{x^2}{2}\right) , t=u+x , w(t)=\exp\left(-\frac{(u+x)^2}{2}\right) and so G(x,u)=\exp\left(-xu-\frac{u^2}{2}\right) . P_n(x) = \frac{1}{2^n n!} \frac{d^n}{dx^n} \!\left[ (x^2 -1)^n \right]\!.
(1 - x^2) P_n(x) - 2 x P_n'(x) + n (n + 1) P_n(x) = 0 For Legendre polynomials, the generating function is defined as G(x,u)=\sum_{n=0}^\infty u^nP_n(x) . The contour integral gives the Schläfli integral for Legendre polynomials: P_n(x) = \frac{1}{2\pi i 2^n} \oint_C \frac{(t^2-1)^n}{(t-x)^{n+1}} dt Summing up the integrand G(x,u) = \frac{1}{\sqrt{1 - 2ux + u^2}} \frac{1}{2\pi i} \oint_C \left(\frac{1}{t - t_-} - \frac{1}{t - t_+}\right) dt where t_\pm = \frac{1}{u} (1 \pm \sqrt{1 - 2ux + u^2}) . For small u , we have t_- \approx x, t_+ \to \infty , which heuristically suggests that the integral should be the residue around t_- , thus giving G(x,u) = \frac{1}{\sqrt{1 - 2ux + u^2}} H_n(x)=(-1)^n e^{x^2} \frac{d^n}{dx^n} \!\left[e^{-x^2}\right] = \left(2x-\frac{d}{dx} \right)^n\cdot 1.
H_n - 2xH_n' + 2nH_n = 0 The generating function is defined as G(x,u)=\sum_{n=0}^\infty \frac{H_n(x)}{n!}\, u^n. The contour integral gives H_n(x)=(-1)^n e^{x^2}\frac{n!}{2\pi i}\oint_C \frac{e^{-t^2}}{(t-x)^{n+1}}\,dt. \begin{aligned} G(x,u) &= \sum_{n=0}^\infty \frac{(-1)^n e^{x^2}}{n!}\frac{n!}{2\pi i}\, u^n \oint_C \frac{e^{-t^2}}{(t-x)^{n+1}}\,dt \\ &= e^{x^2}\frac{1}{2\pi i}\oint_C e^{-t^2}\left(\sum_{n=0}^\infty \frac{(-1)^n u^n}{(t-x)^{n+1}}\right)dt \\ &= e^{x^2}\frac{1}{2\pi i}\oint_C e^{-t^2} \frac{1}{t-x+u}\\ &= e^{x^2}\, e^{-(x-u)^2} \\ & = e^{2xu- u^2} \end{aligned} L_n^{(\alpha)}(x) = {x^{-\alpha} e^x \over n!}{d^n \over dx^n} \left(e^{-x} x^{n+\alpha}\right) = \frac{x^{-\alpha}}{n!}\left( \frac{d}{dx}-1\right)^nx^{n+\alpha}. xL^{(\alpha)}_n(x) + (\alpha + 1 - x)L^{(\alpha)}_n(x)' + nL^{(\alpha)}_n(x) = 0~. The generating function is defined as G(x,u) := \sum_{n=0}^\infty u^n L^{(\alpha)}_n(x) By the same method, we have G(x,u) = \frac{1}{(1-u)^{\alpha+1}} e^{-\frac{ux}{1-u}} .
Jacobi
Source: P_n^{(\alpha,\beta)}(x) = \frac{(-1)^n}{2^n n!} (1-x)^{-\alpha} (1+x)^{-\beta} \frac{d^n}{dx^n} \left\{ (1-x)^\alpha (1+x)^\beta \left (1 - x^2 \right )^n \right\}. \left (1-x^2 \right)P_n^{(\alpha,\beta)}{} + ( \beta-\alpha - (\alpha + \beta + 2)x )P_n^{(\alpha,\beta)}{}' + n(n+\alpha+\beta+1) P_n^{(\alpha,\beta)} = 0. \sum_{n=0}^\infty P_n^{(\alpha,\beta)}(x) u^n = 2^{\alpha + \beta} R^{-1} (1 - u + R)^{-\alpha} (1 + u + R)^{-\beta}, where R = \sqrt{1 - 2ux + u^2} , and the branch of square root is chosen so that R(x, 0) = 1 .
See also
Classical orthogonal polynomials
Text derived from https://en.wikipedia.org/w/index.php?oldid=1360733094 under CC BY-SA 4.0; modifications have been made.