Separation of Variables in Spherical Coordinates

A new chapter opens, and with it the problem quantum mechanics was invented to solve: a particle bound by a potential that depends only on distance, $V(r)$ — the hydrogen atom's Coulomb well, the nuclear well, every atom's central field. The Schrödinger equation is now a partial differential equation in three variables, and solving PDEs head-on is a losing game. The winning move is the oldest in mathematical physics: guess a product, $\psi=R(r)\,Y(\theta,\varphi)$ , and watch the equation fall apart into independent pieces. What makes central potentials special is that the angular piece is already solved. The whole angular-momentum chapter was secretly this chapter's advance work: the operator $\hat{L}^2$ owns everything angular in the Laplacian, its eigenfunctions are the spherical harmonics $Y_\ell^m$ , and its eigenvalues $\ell(\ell+1)\hbar^2$ are quantized by arguments (single-valuedness for $m$ , polar regularity for $\ell$ ) that were fought and won two chapters' worth of articles ago. Separation of variables is therefore not a trick here — it is the statement that rotational symmetry lets you peel the sphere off the problem, leaving a one-variable radial equation where all the potential-specific physics will live. This article owns the factorization itself: the spherical split of the Laplacian (verified symbolically), the derivation that the separation constant is dictated to be $\ell(\ell+1)$ rather than chosen, the commuting set $\{\hat{H},\hat{L}^2,\hat{L}_z\}$ that gives every central-potential state its $(n,\ell,m)$ name tags, and the factorized probability structure that makes angular statistics radial-blind. The radial equation it hands off is the next article's property.

Key takeaways

For central $V(r)$ : $[\hat{H},\hat{L}^2]=[\hat{H},\hat{L}_z]=0$ (verified symbolically) — rotational symmetry makes $\{\hat{H},\hat{L}^2,\hat{L}_z\}$ a commuting label set.
The Laplacian splits: $\nabla^2=\dfrac{1}{r^2}\partial_r\!\big(r^2\partial_r\big)-\dfrac{\hat{L}^2}{\hbar^2 r^2}$ — all angles live inside $\hat{L}^2$ .
Ansatz $\psi=R(r)Y_\ell^m(\theta,\varphi)$ separates the TISE exactly; the residual factorizes as $Y_\ell^m\times[\text{radial equation}]$ (machine-checked for $\ell=0,1,2$ ).
The separation constant is not free: $Y_\ell^m$ already diagonalizes $\hat{L}^2$ , so it must be $\ell(\ell+1)\hbar^2$ — quantization imported, not re-derived.
Every energy eigenstate gets labels $(n,\ell,m)$ ; $m$ never enters the radial equation ⇒ each level is at least $(2\ell+1)$ -fold degenerate (theorem owned by Degeneracy and the Magnetic Quantum Number).
Probabilities factorize: angular statistics are independent of $R$ — verified: $\langle\hat{L}_z\rangle=\hbar/2$ for a $Y_{10}{+}Y_{11}$ mix, any radial profile.
Key idea: rotational symmetry pre-solves the angular two-thirds of every central-potential problem — separation of variables is just the bookkeeping that collects the winnings.

The core insight

For a central potential the Schrödinger equation factorizes on $\psi=R(r)Y_\ell^m(\theta,\varphi)$ : the spherical harmonics diagonalize the angular part of the Laplacian with dictated constant $\ell(\ell+1)\hbar^2$ , leaving a one-dimensional radial equation in which $\ell$ appears as a parameter and $m$ not at all.

Where this lives today

Every electronic-structure code on Earth begins where this article ends: atomic orbitals are stored as $R_{n\ell}(r)\times Y_\ell^m$ pairs, with radial grids or Gaussian expansions for $R$ and exact analytic harmonics for $Y$ — the separation executed once, forever, in software. Density-functional theory's atomic solvers, GW calculations, and the pseudopotential libraries behind materials databases all exploit the same factorization: angular integrals done symbolically by CG algebra, radial integrals done numerically. And far from atoms, the identical split runs gravitational-wave data analysis (spin-weighted harmonics × radial mode functions) and global seismology (Earth's normal modes are $Y_\ell^m$ times radial eigenfunctions) — one symmetry, one factorization, every centrally-symmetric system in physics.

The physical picture

Think of a central-potential eigenstate as a partnership between two independent shapes. The angular factor $Y_\ell^m$ decides the lobes — how probability distributes over directions, with its nodal cones and azimuthal winding fixed entirely by $(\ell,m)$ . The radial factor $R(r)$ decides the shells — how probability distributes over distance, with its nodes and tail fixed by the potential and the energy. The product structure means the two shapes never negotiate: rotate the state and only $Y$ responds; deepen the well and only $R$ responds.

Figure

A central-potential state = radial shells × angular lobes, independently chosen

Why does the product guess work here when it fails for a general $V(x,y,z)$ ? Because the Hamiltonian respects the sphere. Rotational symmetry means $\hat{H}$ commutes with the rotation generators — with $\hat{L}_z$ and with $\hat{L}^2$ — so energy eigenstates can be chosen as simultaneous eigenstates of all three (the compatibility theorem of the formalism chapter doing structural work). And a simultaneous eigenstate of $\hat{L}^2$ and $\hat{L}_z$ is, in its angular dependence, a spherical harmonic: the factorized form is not an ansatz that happens to succeed but the shape symmetry demands.

Figure

The commuting triple hands every state its name tags

The spherical split of the Laplacian

Everything hinges on one operator identity. In spherical coordinates the Laplacian — the kinetic energy's core — separates into a radial derivative plus the angular operator you already own:

\nabla^2 \;=\; \frac{1}{r^2}\frac{\partial}{\partial r}\Big(r^2\frac{\partial}{\partial r}\Big)\;-\;\frac{\hat{L}^2}{\hbar^2 r^2} \qquad (1)

where: $\hat{L}^2=-\hbar^2\big[\tfrac{1}{\sin\theta}\partial_\theta(\sin\theta\,\partial_\theta)+\tfrac{1}{\sin^2\theta}\partial^2_\varphi\big]$ is exactly the spherical form derived in Orbital Angular Momentum Operators (consumed, not re-derived), and the identity is verified symbolically on a nontrivial test function in this article's verifier. The coordinate machinery behind spherical Laplacians — scale factors, curvilinear grammar — is the toolkit the base's prerequisite "The Laplacian in Curvilinear Coordinates" names; the electromagnetism track houses its systematic treatment, and one pointer suffices here. Physically (1) says: kinetic energy = radial breathing + rotational energy $\hat{L}^2/2mr^2$ at lever arm $r$ — the quantum echo of the classical decomposition $p^2=p_r^2+L^2/r^2$ . With (1), the time-independent Schrödinger equation for $V(r)$ reads

-\frac{\hbar^2}{2m}\frac{1}{r^2}\frac{\partial}{\partial r}\Big(r^2\frac{\partial\psi}{\partial r}\Big)+\frac{\hat{L}^2}{2mr^2}\,\psi+V(r)\,\psi \;=\; E\,\psi \qquad (2)

where: the only operator touching angles is $\hat{L}^2$ — the potential, being central, is a spectator to direction. That structural fact is the entire case for the product ansatz, made rigorous next.

Derivation ASeparating the time-independent Schrödinger equation

We derive the factorization $\psi=R(r)Y_\ell^m(\theta,\varphi)$ and the radial equation it leaves behind, with the separation constant dictated rather than introduced. The strategy differs deliberately from the classical separation-of-variables ritual (divide, declare both sides constant): because the angular eigenproblem is already solved, we expand in its eigenfunctions and watch orthogonality do the separating. Both routes give the same answer; this one makes clear why the constant is $\ell(\ell+1)$ and never anything else.

Step 1. Expand the unknown in the complete angular basis (the expansion theorem of Basis Expansions and Components, on the sphere): $\psi(r,\theta,\varphi)=\sum_{\ell m}R_{\ell m}(r)\,Y_\ell^m(\theta,\varphi)$ — no assumption yet, every function of three variables can be written this way with radial coefficients.

Step 2. Insert into (2) and use $\hat{L}^2Y_\ell^m=\ell(\ell+1)\hbar^2\,Y_\ell^m$ — the eigenvalue fact owned by Spherical Harmonics, with its quantization pedigree: integer $m$ from single-valuedness (Eigenfunctions of L_z), integer $\ell\ge|m|$ from polar regularity (Associated Legendre Functions). Each term of the sum picks up its own $\ell(\ell+1)$ ; nothing mixes different $(\ell,m)$ because every operator in (2) is diagonal on the basis.

Step 3. Project with $\int Y_{\ell'}^{m'*}\,d\Omega$ : orthonormality kills all but one term, leaving for each $(\ell,m)$ independently

-\frac{\hbar^2}{2m}\frac{1}{r^2}\frac{d}{dr}\Big(r^2\frac{dR_{\ell m}}{dr}\Big)+\Big[V(r)+\frac{\ell(\ell+1)\hbar^2}{2mr^2}\Big]R_{\ell m}=E\,R_{\ell m}

Why the constant was never a choice: in the textbook divide-and-declare route, a constant $\lambda$ appears and a second argument (regularity of the angular solution) is needed to force $\lambda=\ell(\ell+1)$ . Here that work was already banked: the angular basis came pre-quantized, so the constant arrives with its integer credentials attached. Separation of variables in a central potential is a withdrawal from the angular-momentum chapter's account.

Step 4. Read off the structure. The equation for $R_{\ell m}$ contains $\ell$ but not $m$ — write $R_{n\ell}$ — so a single radial solve at fixed $\ell$ serves all $2\ell+1$ orientations: the degeneracy statement, whose proof-as-theorem belongs to Degeneracy and the Role of the Magnetic Quantum Number. A stationary state of definite $(\ell,m)$ is the single-term case $\psi_{n\ell m}=R_{n\ell}Y_\ell^m$ ; general states are superpositions across the family. The one-dimensional equation itself — its anatomy, its behavior at the origin and at infinity — is the next article's property; this derivation stops at having manufactured it. Verified: the separation residual $\hat{H}\psi-E\psi$ factorizes exactly as $Y_\ell^m\times[\text{radial residual}]$ for symbolic $R(r)$ and $V(r)$ at $\ell=0,1,2$ ; $\hat{L}^2Y_\ell^m=\ell(\ell+1)\hbar^2Y_\ell^m$ swept for $\ell\le2$ , all $m$ ; the split identity (1) exact on a test function (article verifier).

Conservation, labels, and the factorized statistics

The symmetry statement behind the algebra deserves its own two lines. For any central potential (verified symbolically on test functions),

[\hat{H},\hat{L}_z]=0, \qquad [\hat{H},\hat{L}^2]=0 \qquad (3)

Go deeper · Derivation B · 1 minWhy every central Hamiltonian commutes with the angular generators

Derivation BWhy every central Hamiltonian commutes with the angular generators

We prove equations (3) directly from the split identity, making the symmetry claim computational rather than invocational. The strategy: write $\hat{H}$ in the form the split (1) provides — radial operator plus $\hat{L}^2/2mr^2$ plus $V(r)$ — and test each piece against $\hat{L}_z$ and $\hat{L}^2$ . The angular-momentum algebra facts used ( $[\hat{L}^2,\hat{L}_z]=0$ , and that $\hat{L}_z,\hat{L}^2$ involve only angular derivatives) are the property of Angular Momentum Commutation Relations and Orbital Angular Momentum Operators, consumed.

Step 1. Decompose: $\hat{H}=\hat{T}_r+\dfrac{\hat{L}^2}{2mr^2}+V(r)$ , where $\hat{T}_r=-\tfrac{\hbar^2}{2m}\tfrac{1}{r^2}\partial_r(r^2\partial_r)$ involves $r$ only.

Step 2. Test $\hat{L}_z=-i\hbar\,\partial_\varphi$ . It commutes with $\hat{T}_r$ and with $V(r)$ (functions and derivatives in different variables commute — the structural reason, no algebra needed); it commutes with $\hat{L}^2$ by the Casimir relation; and it commutes with $1/r^2$ (an $r$ -function again). Every term passes: $[\hat{H},\hat{L}_z]=0$ .

Step 3. Test $\hat{L}^2$ . Same three checks: purely angular against purely radial (structural), $\hat{L}^2$ against itself (trivial), $\hat{L}^2$ against $1/r^2$ (structural). So $[\hat{H},\hat{L}^2]=0$ — and by the same argument $[\hat{H},\hat{L}_x]=[\hat{H},\hat{L}_y]=0$ : all generators are conserved, though only one can be sharpened at a time alongside $\hat{L}^2$ . Why centrality was the load-bearing assumption: had $V$ depended on angles, Step 2's "different variables" argument would fail for the $V$ term — $\partial_\varphi V\neq0$ — and exactly the symmetry-breaking commutator $[V,\hat{L}_z]=-i\hbar(\partial_\varphi V)$ would appear. The failure is as informative as the success: it quantifies the torque a non-central potential exerts, and it is the germ of every Stark-type perturbation calculation.

Step 4. Harvest via the standing theorems: conservation of the full distributions (per Constants of the Motion), joint eigenbasis with $\hat{H}$ (compatibility theorem), and the CSCO certification of $\{\hat{H},\hat{L}^2,\hat{L}_z\}$ — the $(n,\ell,m)$ labels now rest on computed commutators, not on decree. Verified: $[\hat{H},\hat{L}_z]$ and $[\hat{H},\hat{L}^2]$ vanish identically on a nontrivial symbolic test function with general $V(r)$ (article verifier).

where: these are instances of the generator theorem of Symmetries and Conservation in Quantum Dynamics — rotations about any axis leave $\hat{H}$ invariant, so their generators are conserved — and they promote $(\ell,m)$ to good quantum numbers in the dynamical sense of Constants of the Motion: a hydrogen electron born with $\ell=1,m=0$ keeps those labels for all time. The CSCO algorithm of Complete Sets of Commuting Observables then certifies $\{\hat{H},\hat{L}^2,\hat{L}_z\}$ as a complete label set for central potentials — the birth certificate of the $(n,\ell,m)$ naming scheme used from here to the end of the track. Factorization also disciplines probability. For a factored state, the density splits as $|\psi|^2\,d^3r=\big[|R|^2r^2dr\big]\times\big[|Y_\ell^m|^2d\Omega\big]$ :

P(r\in[a,b],\;\hat{\Omega}\in W) \;=\; \underbrace{\int_a^b|R|^2r^2\,dr}_{\text{radial}}\;\times\;\underbrace{\int_W|Y_\ell^m|^2\,d\Omega}_{\text{angular}} \qquad (4)

where: radial and angular questions have independent answers — the correlation-free structure the verifier exhibits by computing $\langle\hat{L}_z\rangle=\hbar/2$ for the mixed angular state $(Y_{10}+Y_{11})/\sqrt2$ and confirming the answer is untouched by the choice of radial profile. The radial weight $|R|^2r^2$ — with its crucial geometric factor $r^2$ — is defined here in one line; its craft (most-probable radii, moments $\langle r^k\rangle$ ) is owned by Radial Probability Distribution and Expectation Values.

Limiting cases

case	result	meaning
$\ell=0$ (s-states)	angular factor $Y_0^0=1/\sqrt{4\pi}$ : spherical	the radial problem alone; no rotational energy anywhere
$V$ not central	$[\hat{H},\hat{L}^2]\neq0$ : separation fails	$(\ell,m)$ stop being good labels; sphere cannot be peeled off
$r\to\infty$ with $rV\to0$	rotational term dies as $1/r^2$	far field forgets $\ell$ : free-particle behavior (next articles)
classical limit, large $\ell$	$\hat{L}^2/2mr^2\to L^2/2mr^2$	the classical orbit decomposition $p^2=p_r^2+L^2/r^2$ recovered
Cartesian-separable $V$ (e.g. 3D box)	different factorization applies	separability is symmetry-specific; the 3D-box story is a gap-queue article

Common trapThe separation constant is dictated, not adjustable

In generic separation-of-variables problems the constant is a free parameter fixed later by boundary conditions — and students carry that habit here, treating $\ell(\ell+1)$ as "a constant we call $\ell(\ell+1)$ for convenience." It is not convenience: the angular equation's solutions are only finite on the whole sphere when the constant is exactly $\ell(\ell+1)$ with integer $\ell$ , and those quantization arguments were proven — single-valuedness for $m$ , polar-regularity termination for $\ell$ — in their own articles. Writing the constant as $\lambda$ and "finding" non-integer values solves an equation on a slit sphere, not a physical state. Equally wrong: expecting $m$ in the radial equation. Its absence is rotational symmetry, and it is why magnetic quantum numbers cost no extra radial work.

Worked examples

Example 1 — the energy scales that separation exposes. For a hydrogen-like electron, compare the rotational energy stored in the angular factor with the Coulomb energy, at the Bohr radius $a_0$ .

Step 1. Rotational term at lever arm $a_0$ : $E_{\text{rot}}=\dfrac{\ell(\ell+1)\hbar^2}{2m_ea_0^2}$ . Computed: $27.2$ eV for $\ell=1$ , $81.6$ eV for $\ell=2$ (the natural unit $\hbar^2/m_ea_0^2=27.211$ eV — one hartree — emerging by itself).

Step 2. Coulomb energy at the same radius: $|V(a_0)|=e^2/4\pi\varepsilon_0a_0=27.2$ eV (computed) — exactly the $\ell=1$ rotational energy. The coincidence is the virial structure of atomic units: at the atomic length scale, rotation and attraction are the same size, which is why changing $\ell$ reshapes atomic states rather than perturbing them.

Step 3. Read forward: for $\ell=2$ the rotational cost at $a_0$ is triple the Coulomb pull — a $d$ -electron simply cannot live at $a_0$ ; it is pushed to larger radii. That push is the centrifugal-barrier physics quantified two articles on. Answer: $27.2$ eV ( $\ell{=}1$ ) and $81.6$ eV ( $\ell{=}2$ ) versus $27.2$ eV of Coulomb — angular structure competes at full strength with binding.

Example 2 — angular statistics are radial-blind. A particle in a central potential occupies $\psi=R(r)\,\big(Y_1^0+Y_1^1\big)/\sqrt2$ with some normalized radial profile $R$ . What do measurements of $\hat{L}_z$ give?

Step 1. Factorization (4): the angular state is an equal superposition of $m=0$ and $m=1$ ; outcomes $0$ and $\hbar$ with probability $\tfrac12$ each, whatever $R$ is.

Step 2. Mean: $\langle\hat{L}_z\rangle=\tfrac12(0)+\tfrac12(\hbar)=\hbar/2$ — verified by explicit double integral over the sphere, and verified to be unchanged when the radial profile is swapped for a different normalized one.

Step 3. Contrast to sharpen: a measurement of position radius would interrogate $|R|^2r^2$ and ignore the lobes; a measurement of $\hat{L}_z$ interrogates the lobes and ignores the shells. One wavefunction, two independent questionnaires — the separated structure is operationally real. Answer: $P(0)=P(\hbar)=\tfrac12$ , $\langle\hat{L}_z\rangle=\hbar/2$ , radial-profile-independent.

Origins · 1 minDaniel Bernoulli and how this was discovered

Daniel Bernoulli

1700–1782 · Basel

Separation of variables was born in an argument Bernoulli spent thirty years not quite winning. When d'Alembert (1747) and Euler solved the vibrating string with traveling-wave forms, Bernoulli insisted on a different answer: the string's motion is a superposition of standing modes — each a product of a shape in space and an oscillation in time, summed over the mode family. His colleagues objected that a sum of sines could not possibly represent an arbitrary initial shape; the dispute smoldered until Fourier (1807-1822) showed the sines were complete and Bernoulli's expansion was the general solution after all. Product modes plus completeness — precisely the structure this article runs on the sphere, with $Y_\ell^m$ as the standing shapes. His hydrodynamics and the kinetic theory of gases are other tracks' stories (and the probabilist Jacob Bernoulli of the expectation-values article is his uncle — a different Bernoulli); this band is for the man who bet on modes before the mathematics could defend them.

History

1747

— d'Alembert's traveling-wave solution of the vibrating string; Euler generalizes

1753

— Daniel Bernoulli claims the general motion is a superposition of product-form standing modes

1807

— Fourier's heat memoir turns mode expansion into a systematic method (separation of variables proper)

1782-1785

— Laplace and Legendre develop the sphere's own function family in celestial mechanics

1926

— Schrödinger's first paper separates the hydrogen atom in spherical coordinates — this article's machinery, at the moment of its greatest triumph

Going further

Griffiths & Schroeter §4.1 (Eqs. 4.10-4.18) — the separation at the gentlest register.
Shankar Ch. 12; Cohen-Tannoudji Vol. I Ch. VII — the rotational-invariance framing used here.
Sakurai & Napolitano §3.7 — central potentials from the symmetry side.

Provenance note: All structural results — the spherical Laplacian split, the separation constant $\ell(\ell+1)$ , the radial equation, and the $u=rR$ reduction — should be verified against Griffiths & Schroeter §4.1 (Eqs. 4.10–4.18), Shankar Ch. 12, and Cohen-Tannoudji Vol. I Ch. VII. The hydrogen energy $E_n=-13.6\,\mathrm{eV}/n^2$ and the oscillator spectrum $(2n_r+\ell+3/2)\hbar\omega$ are standard textbook values; selection rules $\Delta\ell=\pm1$ follow from angular integrals of spherical harmonics (Wigner–Eckart / Gaunt integrals). No non-standard numbers are introduced. — The split identity, separation residual, and commutators are verified symbolically in this article's verify script. Per the chapter map: the base's second derivation (quantization of $\ell$ from the angular equation) is NOT re-derived here — it is recalled to its owners Eigenfunctions of L_z and the Azimuthal Quantization and Associated Legendre Functions; the $u=rR$ reduction named in this provenance belongs to Effective Potential and the Centrifugal Barrier; the hydrogen and oscillator spectra belong to their own serials ahead.

:::faq[If [H, L_z] = 0 for every axis, why do states have a definite m along only one?] Conservation holds for all three generators, but the generators do not commute with each other — you can sharpen only one $\hat{L}_z$ at a time (plus $\hat{L}^2$ ). Choosing the quantization axis picks which one. The other two components remain conserved in the distribution-level sense of Constants of the Motion: their statistics never drift, even though they are not sharp.

Does separation of variables lose any solutions?

No — that is what Step 1 of the derivation guarantees. Expanding in the complete $Y_\ell^m$ basis is fully general; separation just shows each channel evolves independently. Every energy eigenstate can be chosen factored; every state whatsoever is a superposition of factored ones. The worry is legitimate for the divide-and-declare route (which assumes a product); the expansion route dissolves it.

Why spherical coordinates specifically — would others work?

Coordinates should match symmetry. Central potentials are rotation-invariant, and spherical coordinates are the ones in which rotations act only on two of the three variables — so the symmetry operators ( $\hat{L}^2$ , $\hat{L}_z$ ) involve $\theta,\varphi$ alone and the split (1) exists. A 3D harmonic oscillator, exceptionally symmetric, separates in Cartesian and spherical coordinates (its article is ahead); a general $V(r)$ separates only spherically. A potential with no symmetry separates in nothing.

Is ψ = R·Y an approximation for real atoms with many electrons?

For one particle in a central potential it is exact. For many electrons the exact wavefunction does not factor per-electron — electron repulsion is not central in each coordinate — but the central-field approximation (each electron in an averaged spherical potential) restores per-electron $R\,Y_\ell^m$ orbitals as the zeroth-order language: that is why "s, p, d, f" survives across the periodic table. The corrections are the many-body chapters' business.

What exactly does the radial equation inherit, and what is still unknown?

Inherited: the potential $V(r)$ , the constant $\ell(\ell+1)$ (integer, pre-quantized), and the energy $E$ as eigenvalue-to-be. Unknown: everything about $R(r)$ — its behavior at the origin, its nodes, its tail, which energies make it normalizable. Those questions are precisely the next two articles: the radial equation's anatomy, then its reduction to an effective 1D problem.

Why does the radial measure carry r² while the angular one carries sin θ?

Both are the Jacobian $d^3r=r^2\sin\theta\,dr\,d\theta\,d\varphi$ splitting along the factorization: $r^2dr$ to the shells, $\sin\theta\,d\theta\,d\varphi=d\Omega$ to the lobes. The $r^2$ has real consequences — it is why $|R(0)|^2$ finite still means zero probability at the origin's mathematical point, and why the radial probability density is $r^2|R|^2$ , peaked away from wherever $|R|^2$ alone peaks. That craft belongs to the radial-distribution article.

Does this machinery assume bound states?

Nowhere. Separation and the $(ℓ,m)$ labels apply equally to scattering states — the radial equation simply has oscillatory solutions at $E>0$ instead of decaying ones. That regime, with spherical Bessel functions and partial waves, is the free-particle-in-3D article and ultimately the scattering chapter; the factorized angular structure survives unchanged there.