Associated Legendre Functions

Every numerical library — scipy's `lpmv` and `sph_harm`, geodesy packages, quantum chemistry codes — evaluates $P_\ell^m$ by the upward recurrence (4): Rodrigues is for proofs, the recurrence is for silicon. Earth-gravity models carry the expansion to $\ell\approx2190$ (EGM2008) — degree-two-thousand associated Legendre functions evaluated stably, millions of times. And the addition theorem runs cosmology's angular statistics: the CMB correlation function is $C(\gamma)=\sum_\ell\tfrac{2\ell+1}{4\pi}C_\ell P_\ell(\cos\gamma)$ — equation (6), summed across the whole sky.

Think latitude. The sphere's standing waves factor into a longitude part — the winding phases of the previous article — and a latitude part: $P_\ell^m(\cos\theta)$, a standing wave stretched pole to pole. Its interior zeros are the harmonic's nodal latitudes: exactly $\ell-|m|$ of them, which is why $Y_\ell^m$ carries $|m|$ nodal meridians plus $\ell-|m|$ circles — the drumhead bookkeeping, now with the circles' positions computable.

The factor $(1-x^2)^{m/2}=\sin^{|m|}\theta$ is the picture's physics. For $m\ne0$ the function is strangled at both poles: a state circulating about the $z$-axis cannot pile probability on the axis. Each unit of $|m|$ multiplies in another $\sin\theta$, squeezing the wave toward the equator, until the stretched case $|m|=\ell$ is a pure equatorial band $\sin^\ell\theta$ with no latitude nodes — the top rung whose one-line ODE seeded the ladder construction. And the quantization story has a genuinely new character here. On the circle, integer $m$ came from closure; on the algebra, integer-or-half-integer $j$ came from positivity. The polar equation quantizes by finiteness: its two singular points sit at the poles $x=\pm1$, and for generic $\ell$ every solution diverges there. Demanding a finite wavefunction at the poles — points of physical space like any other — is what forces the series to terminate, and it terminates precisely at integer $\ell\ge|m|$. Three different guardians, one spectrum.

Strip the settled azimuthal factor from the eigenproblem: write $Y=\Theta(\theta)\,e^{im\varphi}$ in $\hat{\mathbf L}^2Y=\ell(\ell+1)\hbar^2Y$ (the operator's spherical form is Orbital Angular Momentum Operators' result, consumed whole); $\partial_\varphi^2$ produces $-m^2$ and leaves a one-variable ODE. In the standard variable $x=\cos\theta$ (poles at $x=\pm1$):

where: $\ell(\ell+1)$ is the $\hat{\mathbf L}^2$ eigenvalue (units $\hbar^2$) and $m^2$ the azimuthal legacy — only $m^2$, so the equation cannot tell $+m$ from $-m$; the circulation sense lives in $e^{im\varphi}$. The prefactor $(1-x^2)$ vanishes at both poles: the equation's singular points, where everything interesting happens.

Termination in hand, the survivors have a closed generating formula — from the Legendre polynomials, the $m=0$ backbone, general $m$ follows by differentiation:

where: the $\ell$-fold derivative alone gives $P_\ell(x)$ (Rodrigues' 1816 formula), the extra $m$ derivatives and the $(1-x^2)^{m/2}$ build the associated function, and the $(-1)^m$ is the Condon–Shortley phase. The formula is legal for all $-\ell\le m\le\ell$ (the differentiation order $\ell+m$ stays non-negative), and it annihilates anything with $|m|>\ell$: differentiating the degree-$2\ell$ polynomial $(x^2-1)^\ell$ more than $2\ell$ times gives zero — the spectrum bound enforced by calculus. The low family, worth memorizing to $\ell=2$ ($x=\cos\theta$):

Note what odd $m$ does: $P_2^1$ contains a bare $\sqrt{1-x^2}$, so it is not a polynomial in $x$ — but it is perfectly smooth as a function on the sphere, the only citizenship that matters.

The Condon–Shortley doctrine. The phase must appear in $Y_\ell^m$ exactly once; the literature splits on where:

Both routes land on the same standard $Y_\ell^m$ — the one whose ladder matrix elements come out real and non-negative, the convention fixed in Ladder Operators and the Angular Momentum Spectrum and threaded through every table downstream. The failure mode is mixing: take $P$ from a CS source and the $Y$ prefactor from a non-CS source and you have zero phases or two — every odd-$m$ harmonic silently flips sign. Fingerprint your sources: $P_1^1(0)=-1$ in the CS convention (scipy's `lpmv(1,1,0)` returns exactly $-1$). Two structural relations complete the toolkit. The sign of $m$ is bookkeeping — the ODE knows $m^2$ alone:

where: the factorial ratio keeps normalization consistent across the multiplet — shapes depend on $|m|$ only; $e^{\pm im\varphi}$ carries the actual difference. And the three-term recurrence at fixed $m$,

where: two known neighbors generate the third — iterated upward from the seeds $P_m^m$ and $P_{m+1}^m$, this single line is how every numerical library actually evaluates the functions (upward recurrence in $\ell$ is stable; the Rodrigues derivative is not, once $\ell$ grows). The ladder operators step $m$ at fixed $\ell$ the same way — the $\ell=1$ chain $\hat L_-Y_1^1=\hbar\sqrt2\,Y_1^0$ quoted in Spherical Harmonics — and none of that stepping is re-derived here (see the FAQ for the differential form).

The assembly itself — the previous article's equation, re-quoted for the record and not re-derived:

where: with the CS phase already inside $P_\ell^m$ (eq. 2), the prefactor is a bare square root; catalogue, sphere orthonormality, parity, and Unsöld all live in Spherical Harmonics, consumed here by name only.

Spherical Harmonics stated it and promised the proof here. For any two directions $\hat n,\hat n'$ separated by angle $\gamma$:

where: the left side knows only the relative angle; the right side is built from coordinates of both directions separately. That mismatch — coordinate-full sum, coordinate-free answer — is the theorem's content, and rotational symmetry is the entire proof.

Why care: the theorem is the hinge between two-point and one-direction physics. Multipole expansions of $1/|\mathbf r-\mathbf r'|$ (the generating-function face belongs to the E&M track's Legendre article), angular correlations, and partial-wave sums all use (6) to trade a relative angle for factorized coordinates — and to swap one Legendre evaluation for $2\ell+1$ harmonic evaluations, whichever is cheaper.

Example 1 — building and auditing $P_2^1$ (confidence builder). Construct $P_2^1$ from the Rodrigues formula, locate its nodal latitude, and check its norm against the factorial formula.

Step 1. Rodrigues with $\ell=2,m=1$: $\mathrm d^3/\mathrm dx^3(x^4-2x^2+1)=24x$, so

Step 2. Nodes. Interior zero at $x=0$ only — one nodal latitude, at the equator, matching $\ell-|m|=1$. The pole zeros from $\sqrt{1-x^2}$ are the $m\ne0$ strangling, not counted.

Step 3. Norm. $\int_{-1}^1 9x^2(1-x^2)\,\mathrm dx=9\left(\tfrac23-\tfrac25\right)=\tfrac{12}{5}$. The formula agrees: $\tfrac{2}{5}\cdot\tfrac{(2+1)!}{(2-1)!}=\tfrac25\cdot6=\boldsymbol{2.4}$. Check passed — and the $Y_2^1$ prefactor $\sqrt{\tfrac{5}{4\pi}\cdot\tfrac{1}{6}}$ is precisely what beats $12/5\cdot2\pi$ down to 1 on the sphere.

Example 2 — the addition theorem on real angles (exam level). Detector A points along $(\theta,\varphi)=(45^\circ,0)$; detector B along $(90^\circ,60^\circ)$. Evaluate both sides of the $\ell=1$ addition theorem.

Step 1. The relative angle, from the dot product: $\cos\gamma=\sin\theta\sin\theta'\cos(\varphi-\varphi')+\cos\theta\cos\theta'=\sin45^\circ\cos60^\circ=\tfrac{\sqrt2}{2}\cdot\tfrac12=\boldsymbol{\tfrac{\sqrt2}{4}\approx0.35355}$, so $P_1(\cos\gamma)=\cos\gamma=0.35355$ and $\gamma\approx69.3^\circ$.

Step 2. The harmonic side needs all three $m$: with $Y_1^0=\sqrt{\tfrac{3}{4\pi}}\cos\theta$ and $Y_1^{\pm1}=\mp\sqrt{\tfrac{3}{8\pi}}\sin\theta\,e^{\pm i\varphi}$, the $m=0$ term contributes $\tfrac{3}{4\pi}\cos45^\circ\cos90^\circ=0$; the $m=\pm1$ pair contributes $2\cdot\tfrac{3}{8\pi}\sin45^\circ\sin90^\circ\cos(0-60^\circ)$.

Step 3. Sum and scale: $\sum_m{Y_1^m}^*(\hat n')Y_1^m(\hat n)=\tfrac{3}{4\pi}\cdot0.35355$; multiplying by $\tfrac{4\pi}{3}$ gives $\boldsymbol{0.35355}$ — agreement to machine precision ($10^{-12}$ in the verification script).

Step 4. Read the physics: the $m$-sum knows each detector's coordinates separately yet remembers only the $69.3^\circ$ between them — the rotational-invariance conspiracy that makes angular-correlation measurements coordinate-free.

References. Griffiths, Introduction to Quantum Mechanics, 3rd ed., §4.1 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §3.6 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 12 · Cohen-Tannoudji, Quantum Mechanics, Vol. 1, Ch. VI · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 15.

Provenance. Equations cross-checked against Griffiths (3rd ed.) §4.1, Sakurai & Napolitano (2nd ed.) §3.6, Shankar (2nd ed.) Ch. 12, Cohen-Tannoudji Vol. 1 Ch. VI, and Arfken (7th ed.) Ch. 15. A verifier should: (1) confirm the polar equation by separating $\mathbf{L}^2 Y=l(l+1)Y$; (2) check the Rodrigues formula and Condon–Shortley phase against Arfken; (3) verify the orthogonality norm $\frac{2}{2l+1}\frac{(l+m)!}{(l-m)!}$; (4) reproduce $P_1^0=\cos\theta$, $P_1^1=-\sin\theta$, $P_2^0=\tfrac12(3\cos^2\theta-1)$; (5) verify $Y_l^l\propto\sin^l\theta\,e^{il\phi}$ from $L_+Y_l^l=0$. All numbers stated are standard textbook values; no nonstandard constants introduced.

Upstream: Orbital Angular Momentum Operators (the $\hat{\mathbf L}^2$ spherical form) · Eigenfunctions of L_z and the Azimuthal Quantization (the $e^{im\varphi}$ factor and integer $m$) · Ladder Operators and the Angular Momentum Spectrum (the algebraic twin of the Frobenius eviction, and the stepping machinery). Downstream: Spherical Harmonics (the assembled family — its catalogue, parity, and Unsöld's theorem consumed the polar theory built here) · Wigner D-Matrices (where $d^\ell_{m0}$ reduces to these functions and the addition theorem gets its one-line second proof) · the hydrogen chapter (where the radial equation joins) · one cross-track pointer: the generating-function face of the Legendre polynomials and their multipole service live in the E&M track's Legendre article.