Spherical Bessel Functions

Spherical Bessel functions run inside some of the most active computations in physics. In cosmology they are the kernel that maps primordial density ripples onto the cosmic microwave background: the CMB angular power spectrum is built from line-of-sight integrals of the form $\Delta_\ell(k)\sim\int j_\ell\big(k(\eta_0-\eta)\big)\,S(k,\eta)\,\mathrm d\eta$, and codes such as CAMB and CLASS spend much of their runtime evaluating $j_\ell$ for $\ell$ up to several thousand — a hard problem because the functions are wildly oscillatory in exactly the regime that matters. In nanophotonics, Mie theory expands the field scattered by a sphere in $j_\ell$ (inside) and $h_\ell^{(1)}$ (outside), and the resonance where a denominator built from these functions vanishes produces the sharp color of gold colloids and the engineered response of dielectric nanoantennas. In nuclear physics the spherical wells of Example 1 are the zeroth-order shell model. The same functions drive T-matrix and partial-wave codes in electron-atom and nucleon-nucleon scattering, antenna multipole patterns, and the free-oscillation modes of the whole Earth measured after great earthquakes. And the highly oscillatory spherical-Bessel integral remains an open computational front: FFTLog and double-exponential quadrature methods are still being refined for next-generation galaxy surveys.

A plane wave keeps its amplitude forever, but a wave radiating from a point spreads its energy over a sphere of area $4\pi r^2$, so energy conservation forces the amplitude to thin as $1/r$. The spherical Bessel functions are the standing-wave shapes that do this: $j_0(x)=\sin x/x$ is a sine damped by the geometric $1/x$ factor, and every higher order is a phase-shifted oscillation inside the same $\pm 1/x$ envelope. The index $\ell$ adds a centrifugal barrier $\ell(\ell+1)/x^2$ that pushes the wave outward, which is why the regular solution is suppressed near the center as $j_\ell\sim x^\ell$. The first figure shows the three lowest orders and the number that anchors everything downstream: the first zero of $j_0$.

Choosing which solutions to keep is the entire art of a spherical wave problem, and it happens at two boundaries. At the origin, only $j_\ell$ survives if $r=0$ is in the domain, since $y_\ell$ diverges there. At infinity, a scattering or radiation problem needs a definite direction of energy flow, so it selects the Hankel combination $h_\ell^{(1)}\sim e^{ix}/x$ (outgoing) or $h_\ell^{(2)}\sim e^{-ix}/x$ (incoming). And when the domain is a hard-walled sphere, $j_\ell(ka)=0$ can only be met at discrete wavenumbers — the zeros of $j_\ell$ — so the continuous problem becomes a discrete spectrum. That last mechanism is the deepest payoff, and the second figure shows it cleanest: the bound-state ladder of a particle trapped in a spherical box.

Everything begins with one ordinary differential equation. Separating the Helmholtz equation in spherical coordinates and rescaling to the dimensionless variable $x=kr$ leaves the radial factor obeying the spherical Bessel equation. Note the coefficient $2x$ on the first-derivative term — the fingerprint of three-dimensional space, and the feature that most cleanly distinguishes this equation from its cylindrical cousin.

where: $x$ — dimensionless radial variable, $x=kr$; $R(x)$ — radial factor, any of $j_\ell,y_\ell,h_\ell^{(1,2)}$; $\ell$ — non-negative integer angular-momentum index; $k$ — wavenumber from $\nabla^2\psi+k^2\psi=0$. Read it as: a restoring term $x^2R$, a centrifugal barrier $-\ell(\ell+1)R$, and a $1/r$-spreading first-derivative term $2xR'$. The solutions are elementary because of a change of order: each spherical Bessel function is a half-integer-order cylindrical Bessel function with a $\sqrt{\pi/2x}$ dressing, and half-integer-order Bessel functions terminate.

where: $J_{\ell+1/2}$ — cylindrical Bessel function of the first kind at half-integer order; $Y_{\ell+1/2}$ — second kind. The factor $\sqrt{\pi/2x}$ is precisely what turns the $1/\sqrt{x}$ tail of a cylindrical function into the $1/x$ tail of a spherical wave. Written out, the two lowest orders are the whole idea in miniature: a trigonometric function divided by a power of $x$, and nothing worse. These are the seeds every recurrence and every derivation returns to.

where: $x=kr$. Read (3) as "a sine wave that thins as $1/x$"; read (4) as the same idea one unit of angular momentum higher, with an extra $1/x$ pushing amplitude away from the origin. Rayleigh's formula is the engine that makes the family. The operator $(1/x)\,\mathrm d/\mathrm dx$ raises the order by one and supplies one more inverse power of $x$ each time it acts, so applying it $\ell$ times to the seed generates $j_\ell$ in closed form.

where: $\big(\tfrac1x\tfrac{\mathrm d}{\mathrm dx}\big)^\ell$ — the ladder operator applied $\ell$ times. Read it as a manufacturing rule: start from $\sin x/x$, differentiate-and-divide $\ell$ times, and you have the $\ell$-th standing wave with no series left over. For numerical work you rarely differentiate; you climb the ladder with the three-term recurrence, shared by all four kinds of spherical Bessel function.

where: $f_\ell$ — any of $j_\ell,y_\ell,h_\ell^{(1)},h_\ell^{(2)}$. Note a numerical caution: (6) is stable downward in $\ell$ for $j_\ell$ and upward for $y_\ell$ — run it the wrong way and rounding error explodes. The outgoing and incoming radiation basis is built by the same trick that turns $\sin$ and $\cos$ into $e^{\pm ix}$. These spherical Hankel functions carry a single complex exponential and a clean $1/x$ tail.

where: $h_\ell^{(1)}\sim(-i)^{\ell+1}e^{ix}/x$ is an outgoing spherical wave; $h_\ell^{(2)}\sim i^{\ell+1}e^{-ix}/x$ is incoming. With the physics time convention $e^{-i\omega t}$, radiated and scattered fields always ride $h_\ell^{(1)}$. Two structural facts close the toolkit. The Wronskian certifies that $j_\ell$ and $y_\ell$ are independent and fixes the normalization of any Green's function built from them; the closure relation states their completeness over the continuous radial spectrum.

where: $W$ — Wronskian, nonzero everywhere so the pair spans the full solution space; $\delta(k-k')$ — Dirac delta over the continuum of radial wavenumbers. Equation (10) is the spherical analogue of Fourier completeness — it lets you expand any radial function in $j_\ell(kr)$.

The small- and large-argument behavior governs every boundary-matching argument: the small-$x$ forms decide what is allowed at the origin, the large-$x$ forms what radiates at infinity.

1 · A particle in an infinite spherical well (fully worked)

An electron is confined to a hard-walled sphere of radius $a=1\ \text{Å}=1.0\times10^{-10}\ \text{m}$. Find the ground-state ($1s$) energy, the first excited ($1p$) energy, and the ordering of the lowest five levels.

Step 1. Inside the well the radial function is $u_\ell(r)=r\,j_\ell(kr)$ with $k=\sqrt{2mE}/\hbar$, and the hard wall demands $j_\ell(ka)=0$. So $k_{n\ell}=\alpha_{n\ell}/a$ with $\alpha_{n\ell}$ the $n$-th positive zero of $j_\ell$, and $E_{n\ell}=\hbar^2\alpha_{n\ell}^2/(2ma^2)$.

Step 2. The lowest zeros are $\alpha_{1s}=\pi=3.14159$, $\alpha_{1p}=4.49341$, $\alpha_{1d}=5.76346$, $\alpha_{2s}=2\pi=6.28319$, $\alpha_{1f}=6.98793$. Ordering by $\alpha^2$ gives $1s<1p<1d<2s<1f$ — the $1d$ falls below the $2s$, so this is not the hydrogenic ordering.

Step 3. Evaluate the ground state. With $\hbar^2/(2m_e)=6.104\times10^{-39}\ \text{J}\,\cdot\,\text{m}^2$ and $a^2=10^{-20}\ \text{m}^2$,

Step 4. Excited levels scale as $\alpha^2$: $E_{1p}=E_{1s}(4.49341/\pi)^2=37.6\times2.046=\mathbf{76.9\ eV}$, and likewise $E_{1d}=126\ \text{eV}$, $E_{2s}=150\ \text{eV}$, $E_{1f}=186\ \text{eV}$. The spectrum is literally a table of squared spherical-Bessel zeros, and the $1s$–$1p$ gap is $39.3\ \text{eV}$.

2 · Low-energy scattering from a hard sphere (fully worked)

A wave scatters from an impenetrable sphere of radius $a$. Find the $s$-wave phase shift $\delta_0$ and the total cross section in the low-energy limit $ka\ll1$.

Step 1. Outside the sphere the $\ell=0$ field is $j_0(kr)\cos\delta_0-y_0(kr)\sin\delta_0$, and the hard wall forces it to vanish at $r=a$, giving $\tan\delta_0=j_0(ka)/y_0(ka)$.

Step 2. Insert the closed forms $j_0(ka)=\sin(ka)/(ka)$ and $y_0(ka)=-\cos(ka)/(ka)$: the $ka$ factors cancel and $\tan\delta_0=-\tan(ka)$, so $\delta_0=-ka$ exactly for all energies. The negative sign says a hard wall pushes the wave out — a repulsive phase shift.

Step 3. The $s$-wave cross section is $\sigma_0=(4\pi/k^2)\sin^2(ka)$; for $ka\ll1$, $\sin(ka)\approx ka$ gives $\sigma_0\to 4\pi a^2$. Higher partial waves, suppressed as $\delta_\ell\sim(ka)^{2\ell+1}$, contribute negligibly at low energy.

Step 4. The total low-energy cross section is therefore $\sigma\to4\pi a^2$, four times the geometric shadow $\pi a^2$. This factor-of-four enhancement is purely wave-mechanical — a classical particle sees only $\pi a^2$ — and it is the canonical illustration of how spherical Bessel functions turn a boundary condition into an observable cross section.

3 · Building $j_2$ from Rayleigh's formula

Use Rayleigh's formula to obtain $j_2(x)$ in closed form and check its small-$x$ limit against $x^\ell/(2\ell+1)!!$.

Step 1. Rayleigh's formula gives $j_2(x)=(-x)^2\big(\tfrac1x\tfrac{\mathrm d}{\mathrm dx}\big)^2\tfrac{\sin x}{x}$. Apply the operator once: $\tfrac1x\tfrac{\mathrm d}{\mathrm dx}\tfrac{\sin x}{x}=\tfrac{\cos x}{x^2}-\tfrac{\sin x}{x^3}$.

Step 2. Apply it a second time and multiply by $x^2$. Carrying the derivatives through gives

Step 3. Expand for small $x$ using $\sin x=x-x^3/6+x^5/120$ and $\cos x=1-x^2/2+x^4/24$. The $1/x^2$ and constant terms cancel identically, leaving $j_2(x)\to x^2/15$. Since $(2\cdot2+1)!!=5!!=15$, this matches $j_\ell\sim x^\ell/(2\ell+1)!!$ exactly.

Step 4. A numerical spot check confirms it: $j_2(0.1)=6.662\times10^{-4}$ against $x^2/15=6.667\times10^{-4}$, agreeing to the expected $O(x^4)$ correction. The infinite Bessel series has collapsed to three elementary terms, precisely because the order $\ell+\tfrac12=5/2$ is half-integer.

The spherical Bessel functions bear the name of the man who saw them not as an abstract special function but as the natural language of sound and light spreading from a source — and who found the differentiation trick that makes them elementary.

Texts. Arfken & Weber, Mathematical Methods for Physicists, 7th ed., Ch. 14 (Bessel and spherical Bessel functions) · Hassani, Mathematical Physics, Ch. 13–14 (separation of variables and special functions) · Stone & Goldbart, Mathematics for Physics, Ch. 8 (special functions and the Sturm–Liouville view) · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 6 · Nakahara, Geometry, Topology and Physics (Laplacian on spheres, for the geometric backdrop) · Abramowitz & Stegun, Handbook of Mathematical Functions, Ch. 10 · Jackson, Classical Electrodynamics, Ch. 9 (multipole radiation). The closed forms, Rayleigh's formula, the recurrences, the Wronskian $W[j_\ell,y_\ell]=1/x^2$, the asymptotics with their double factorials, and the Rayleigh plane-wave expansion are all standard results. Key numerical anchors: $j_2\to x^2/15$ for small $x$; $W[j_1,y_1](2.3)=0.18904=1/2.3^2$; spherical-well zeros $3.14159, 4.49341, 5.76346, 6.28319, 6.98793$ giving $E_{1s}=37.6\ \text{eV}$ and $E_{1p}=76.9\ \text{eV}$ for a $1\ \text{Å}$ electron box; hard-sphere results $\delta_0=-ka$ and $\sigma\to4\pi a^2$. No nonstandard constants are asserted.

Builds on: Bessel Functions of the First and Second Kind · Spherical Harmonics · Gamma Function · Hankel Functions and Asymptotic Behavior of Bessel Functions. Leads to: Recurrence Relations for Special Functions · Generating Functions for Special Functions · Legendre Polynomials.

Ask the tutor: Ask me to walk you through any spherical wave problem — a quantum well, a Mie-scattering resonance, or a CMB transfer integral — and we will pick the right combination of $j_\ell$, $y_\ell$, and $h_\ell^{(1)}$ together and match it at the boundaries.