Legendre differential equation arises from the study of axisymmetric solutions of Laplace's equation in R3\mathbb R^3. Specifically, let the spherical coordinate be configured as

x=ρsinθcosϕ,y=ρsinθsinϕ,z=ρcosθ.x=\rho\sin\theta\cos\phi,\quad y=\rho\sin\theta\sin\phi,\quad z=\rho\cos\theta.

Then Laplace's equation Δu=0\Delta u=0 becomes

ρ2ρ(ρ2uρ)+ρ2csc2θuϕϕ+ρ2cscθθ(sinθuθ)=0.(1)\rho^{-2}\partial_\rho(\rho^2 u_\rho)+\rho^{-2}\csc^2\theta u_{\phi\phi}+\rho^{-2}\csc\theta\partial_\theta(\sin\theta u_\theta)=0.\tag1

Let uu be symmetric about the zz-axis, so it does not depend on ϕ\phi, which means we should investigate (1) by the separation of variables u=f(ρ)g(θ)u=f(\rho)g(\theta):

gρ2ρ(ρ2fρ)+fρ2cscθθ(sinθgθ)=0.g\rho^{-2}\partial_\rho(\rho^2 f_\rho)+f\rho^{-2}\csc\theta\partial_\theta(\sin\theta g_\theta)=0.

Rearranging the terms gives

ρ(ρ2fρ)f=cscθθ(sinθgθ)g.(2){\partial_\rho(\rho^2 f_\rho)\over f}=-{\csc\theta\partial_\theta(\sin\theta g_\theta)\over g}.\tag2

Because the left-hand side does not depend on θ\theta and the right-hand side does not depend on ρ\rho, both sides of (2) should be equal to some constant λ\lambda. As a result, we convert Laplace's equation into two ordinary differential equations:

ρ2fρρ+2ρfρ=λf,(3)\rho^2f_{\rho\rho}+2\rho f_\rho=\lambda f,\tag3 cscθθ(sinθgθ)=λg.(4)-\csc\theta\partial_\theta(\sin\theta g_\theta)=\lambda g.\tag4

The radial equation (3) is a Cauchy-Euler equation, whose general solution is given by f=Aρm+Bρm1f=A\rho^m+B\rho^{-m-1}, where m[0,+)m\in[0,+\infty) is chosen to satisfy λ=m(m+1)\lambda=m(m+1). Because the solution needs to be regular at the origin, we must have B=0B=0, so f=Aρmf=A\rho^m.

More interesting is the polar equation (4). By making the substitution μ=cosθ\mu=\cos\theta, we have θ=sinθμ\partial_\theta=-\sin\theta\partial_\mu, so (4) becomes

Lg:=ddμ[(1μ2)dgdμ]=λg,(5)Lg:=-{\mathrm d\over\mathrm d\mu}\left[(1-\mu^2){\mathrm dg\over\mathrm d\mu}\right]=\lambda g,\tag5

which is known as the Legendre differential equation. Because g(θ)g(\theta) needs to be 2π2\pi-periodic and continuous on [0,2π][0,2\pi], g(μ)g(\mu) needs to be continuous on [1,1][-1,1]. In this article, we prove that under these conditions, the admissible eigenvalues λ\lambda of the operator LL in (5) are of the form (+1)\ell(\ell+1) with Z\ell\in\mathbb Z.

Application of the Frobenius method

Let ana_n be a sequence of numbers defined on non-negative integers and assume g(μ)g(\mu) to be in the form of

g(μ)=n0anμn+α.(6)g(\mu)=\sum_{n\ge0}a_n\mu^{n+\alpha}.\tag6

Then we have

dgdμ=n0(n+α)anμn+α1=αa0μα1+(α+1)a1μα+n0(n+α+2)an+2μn+α+1.\begin{aligned} {\mathrm dg\over\mathrm d\mu} &=\sum_{n\ge0}(n+\alpha)a_n\mu^{n+\alpha-1} \\ &=\alpha a_0\mu^{\alpha-1}+(\alpha+1)a_1\mu^\alpha+\sum_{n\ge0}(n+\alpha+2)a_{n+2}\mu^{n+\alpha+1}. \end{aligned}

Consequently, (5) becomes

α(α1)a0μα2+α(α+1)a1μα1+n0[(n+α+1)(n+α+2)an+2(n+α+1)(n+α)an+λ]μn+α=0.\begin{aligned} &\alpha(\alpha-1)a_0\mu^{\alpha-2}+\alpha(\alpha+1)a_1\mu^{\alpha-1} \\ &+\sum_{n\ge0}[(n+\alpha+1)(n+\alpha+2)a_{n+2}-(n+\alpha+1)(n+\alpha)a_n+\lambda]\mu^{n+\alpha}=0. \end{aligned}

Comparing coefficients gives α=0\alpha=0, so we obtain a recursive relation for ana_n:

an+2=n(n+1)λ(n+1)(n+2)an.(7)a_{n+2}={n(n+1)-\lambda\over(n+1)(n+2)}a_n.\tag7

Asymptotic analysis of ana_n

Let ν{0,1}\nu\in\lbrace 0,1\rbrace, so substituting n=2k+ν2n=2k+\nu-2 into (7) gives

a2k+ν={122k+ν+O(1k2)}a2(k1)+ν={11k+O(1k2)}a2(k1)+ν.\begin{aligned} a_{2k+\nu} &=\left\lbrace 1-{2\over2k+\nu}+O\left(1\over k^2\right)\right\rbrace a_{2(k-1)+\nu} \\ &=\left\lbrace 1-\frac1k+O\left(1\over k^2\right)\right\rbrace a_{2(k-1)+\nu}. \end{aligned}

Because log(1+z)=z+O(z2)\log(1+z)=z+O(z^2) for small zz, there exists some k0k_0 such that whenever k>k0k>k_0,

11k+O(1k2)=exp{1k+O(1k2)}.1-\frac1k+O\left(1\over k^2\right)=\exp\left\lbrace-\frac1k+O\left(1\over k^2\right)\right\rbrace.

Consequently, we obtain

a2k+ν=exp{k0<rk1r+O(k0<rk1r2)}a2k0+ν=exp{logk+O(1)}a2k0+ν.\begin{aligned} a_{2k+\nu} &=\exp\left\lbrace-\sum_{k_0<r\le k}\frac1r+O\left(\sum_{k_0<r\le k}{1\over r^2}\right)\right\rbrace a_{2k_0+\nu} \\ &=\exp\lbrace-\log k+O(1)\rbrace a_{2k_0+\nu}. \end{aligned}

This indicates that if a2k0+ν0a_{2k_0+\nu\ne0}, then there exists some constant A>0A>0 such that

eAk<a2k+νa2k0+ν<eAk.(8){e^{-A}\over k}<{a_{2k+\nu}\over a_{2k_0+\nu}}<{e^A\over k}.\tag8

Setting α=0\alpha=0 in (6), we have

g(μ)=k0a2kμ2k+k0a2k+1μ2k+1.(9)g(\mu)=\sum_{k\ge0}a_{2k}\mu^{2k}+\sum_{k\ge0}a_{2k+1}\mu^{2k+1}.\tag9

Because a2k+νa_{2k+\nu} is a multiple of aνa_\nu, we conclude that the odd and even components in (9) are exactly the fundamental solutions of (5), and we can adjust the values of a0a_0 and a1a_1 to represent all general solutions.

The spectrum of the Legendre differential operator

According to (8) and the comparison test, g(μ)g(\mu) is regular when 1<μ<1-1<\mu<1. When μ=±1\mu=\pm1, we see that the fundamental solution

k0a2k+νμ2k+ν\sum_{k\ge0}a_{2k+\nu}\mu^{2k+\nu}

will diverge if a2k+νa_{2k+\nu} is nonzero for all sufficiently large kk. Consequently, for (5) to have meaningful solutions, at least of the sequences a2k,a2k+1a_{2k}, a_{2k+1} needs to vanish at all large kk.

From the properties of the recursive relation (7), this is possible if and only if λ=(+1)\lambda=\ell(\ell+1) for some integer \ell. In this situation, the meaningful solutions are exactly the constant multiples of one fundamental solution satisfying ν(mod2)\nu\equiv\ell\pmod2.

Legendre polynomials

Let λ=(+1)\lambda=\ell(\ell+1) for some non-negative integer \ell and ν{0,1}\nu\in\lbrace 0,1\rbrace be such that ν\ell-\nu is even, so the sequence a2k+νa_{2k+\nu} becomes zero after 2k+ν>2k+\nu>\ell, so it follows from (7) that for 0m(ν)/20\le m\le(\ell-\nu)/2, there is

a2m=(2m+1)(2m+2)(2m)(2m+2)(+1)a2m+2=(2m+1)(2m+2)(2m)(2m)+(2m)(+1)a2m+2=(2m+1)(2m+2)(2+12m)(2m)a2m+2=(1)m(2m)!!(2m+1)(2m+2)(1)a(2+12m)(2+32m)(21)=(1)m(2m)!!(22m1)!!(2m)!!a(21)!!=(1)m(2m)!!(22m)!(2m)!(22m)!!!a(21)!!=(1)m2m!(22m)!(m)!(2m)!!a(21)!!.\begin{aligned} a_{\ell-2m} &={(\ell-2m+1)(\ell-2m+2)\over(\ell-2m)(\ell-2m+2)-\ell(\ell+1)}a_{\ell-2m+2} \\ &={(\ell-2m+1)(\ell-2m+2)\over(\ell-2m)(-2m)+(-2m)(\ell+1)}a_{\ell-2m+2} \\ &={(\ell-2m+1)(\ell-2m+2)\over(2\ell+1-2m)(-2m)}a_{\ell-2m+2} \\ &={(-1)^m\over(2m)!!}\cdot{(\ell-2m+1)(\ell-2m+2)\cdots(\ell-1)\ell a_\ell\over(2\ell+1-2m)(2\ell+3-2m)\cdots(2\ell-1)} \\ &={(-1)^m\over(2m)!!}\cdot{(2\ell-2m-1)!!\over(\ell-2m)!}\cdot{\ell!a_\ell\over(2\ell-1)!!} \\ &={(-1)^m\over(2m)!!}\cdot{(2\ell-2m)!\over(\ell-2m)!(2\ell-2m)!!}\cdot{\ell!a_\ell\over(2\ell-1)!!} \\ &={(-1)^m\over2^\ell m!}\cdot{(2\ell-2m)!\over(\ell-m)!(\ell-2m)!}\cdot{\ell!a_\ell\over(2\ell-1)!!}. \end{aligned}

Setting a=(21)!!/!a_\ell=(2\ell-1)!!/\ell! gives the Legendre polynomial P(μ)P_\ell(\mu) of order \ell:

P(μ)=0m/2(1)m2m!(22m)!(m)!(2m)!μ2m.(10)P_\ell(\mu)=\sum_{0\le m\le\ell/2}{(-1)^m\over2^\ell m!}{(2\ell-2m)!\over(\ell-m)!(\ell-2m)!}\mu^{\ell-2m}.\tag{10}

Notice that

(22m)!(2m)!=ddμ(μ22m),{(2\ell-2m)!\over(\ell-2m)!}={\mathrm d^\ell\over\mathrm d\mu^\ell}(\mu^{2\ell-2m}),

so we have

P(μ)=ddμ0m/2(1)m2m!(m)!μ22m=12!ddμm=0(m)(1)m(μ2)m.\begin{aligned} P_\ell(\mu) &={\mathrm d^\ell\over\mathrm d\mu^\ell}\sum_{0\le m\le\ell/2}{(-1)^m\over 2^\ell m!(\ell-m)!}\mu^{2\ell-2m} \\ &={1\over2^\ell\ell!}{\mathrm d^\ell\over\mathrm d\mu^\ell}\sum_{m=0}^\ell\binom\ell m(-1)^m(\mu^2)^{\ell-m}. \end{aligned}

Finally, applying the binomial theorem, we obtain a neat expression for P(μ)P_\ell(\mu):

P(μ)=12!ddμ[(μ21)].(11)P_\ell(\mu)={1\over 2^\ell\ell!}{\mathrm d^\ell\over\mathrm d\mu^\ell}[(\mu^2-1)^\ell].\tag{11}

This is known as Rodrigues' formula for Legendre polynomials.

Conclusion

In this article, we began our discussion by seeking axisymmetric solutions from Laplace's equation. By the separation of variables, we introduced Legendre's differential equation. Subsequently, we used the Frobenius method and some asymptotic analysis to demonstrate that the eigenvalues λ\lambda of the Legendre differential operator LL are exactly in the form of (+1)\ell(\ell+1) for Z\ell\in\mathbb Z. Finally, we derived expressions for the solutions to Legendre's differential equation when λ=(+1)\lambda=\ell(\ell+1).