Forbidden Transitions: Magnetic Dipole and Electric Quadrupole

In 1864 William Huggins pointed a spectroscope at a planetary nebula and found bright green lines belonging to no element on Earth. "Nebulium" haunted chemistry for sixty years — until Ira Bowen showed in 1927 that the lines are doubly-ionized oxygen doing what no laboratory gas is allowed to finish: radiating through transitions whose electric-dipole matrix element is exactly zero. Electric Dipole Selection Rules built the ban list; this article is about what happens on the banned routes. "Forbidden" turns out to be a misnomer for exponentially patient — the atom still couples to the field through the spatial variation the dipole approximation discarded, and the next term of $e^{i\mathbf{k}\cdot\mathbf{r}}$ splits cleanly into a magnetic-dipole (M1) operator built from $\hat{\mathbf L}+g_s\hat{\mathbf S}$ and an electric-quadrupole (E2) operator built from $r_ir_j$ . We derive both operators from the field gradient — repairing two slips in our source's algebra along the way, one sign convention and one stray $\hbar$ — then prove the suppression-scaling theorem that sets their price: M1 amplitudes are exactly $\alpha/2$ of an E1 amplitude's natural scale (a factor $1.3\times10^{-5}$ in rate), and E2 costs $(ka_0)^2\sim7\times10^{-6}$ . Each multipole carries its own selection-rule table, complementary to E1's: parity now must not change. And we prove the theorem behind hydrogen's strangest state: 2s is doubly dead — E1 killed by the angular integral, M1 killed by radial orthogonality — leaving only two-photon decay, 0.122 s of patience. The payoff spans twenty-three orders of magnitude of lifetime, from 2p's 1.6 ns to the 21 cm line's eleven million years.

Key takeaways

"Forbidden" means the E1 matrix element $\langle f|\mathbf r|i\rangle$ vanishes — not that the transition is impossible. The next term of $e^{i\mathbf k\cdot\mathbf r}=1+i\mathbf k\cdot\mathbf r+\cdots$ opens two slower doors.
The $i\mathbf k\cdot\mathbf r$ term splits by symmetric/antisymmetric decomposition: the antisymmetric piece is the magnetic dipole M1, $-\hat{\boldsymbol\mu}\cdot\mathbf B$ with $\hat{\boldsymbol\mu}=-\frac{\mu_B}{\hbar}(\hat{\mathbf L}+g_s\hat{\mathbf S})$ ; the symmetric piece is the electric quadrupole E2, built from the traceless tensor $r_ir_j-\frac13r^2\delta_{ij}$ .
Price of admission: M1 rates are suppressed by $(\alpha/2)^2\approx1.3\times10^{-5}$ (the amplitude ratio $\mu_B/(ea_0c)=\alpha/2$ exactly), E2 rates by $(ka_0)^2\approx7.5\times10^{-6}$ at optical frequencies.
New rule tables, complementary to E1: parity must NOT change. M1: $\Delta\ell=0$ , $\Delta j=0,\pm1$ (no $0\to0$ ), $\Delta m=0,\pm1$ — and no radial change at all. E2: $\Delta\ell=0,\pm2$ , $\Delta j$ up to $\pm2$ , $\Delta m$ up to $\pm2$ , with $0\to0$ , $0\leftrightarrow1$ , $\tfrac12\to\tfrac12$ banned.
Hydrogen 2s is doubly dead — E1 by parity/triangle, M1 by radial orthogonality $\langle R_{10}|R_{20}\rangle=0$ — so it decays by two photons in 0.122 s (Breit & Teller 1940).
Forbidden lines rule wherever collisions are rare: nebular [O III] 4959/5007 Å, the coronal green line, the 21 cm sky, and the deliberately narrow clock transitions of Al⁺ and Sr.
Key idea: Forbidden does not mean impossible: dropping the dipole approximation and keeping the next term of $e^{i\mathbf{k}\cdot\mathbf{r}}$ generates magnetic-dipole and electric-quadrupole operators whose selection rules and $(\alpha Z)^2$ -suppressed rates govern transitions where E1 vanishes.

Where this lives today

The most precise measuring devices humans have built are forbidden transitions in boxes. Optical clocks choose their ticking line because E1 is dead: the Al⁺ clock runs on a $^1S_0\to{}^3P_0$ transition — doubly forbidden, proceeding only through hyperfine mixing — with a natural linewidth near 8 mHz; strontium lattice clocks pull the same trick at 698 nm and hold fractional uncertainties near $10^{-18}$ . Astronomical patience and metrological perfection are the same property, priced in Lifetime, Linewidth, and Natural Broadening. Astronomy still runs on Bowen's inversion: [O III] ratios thermometer nebulae and JWST-era galaxies, coronal iron lines monitor solar storms, and 21 cm intensity maps are cosmology's next great survey. Slow doors are the ones worth listening at.

The physical picture

An optical photon is a wave five thousand times larger than the atom it drives. The Dipole Approximation made that mismatch into a tool: treat the field as uniform across the atom and you capture the dominant coupling. But across the atom's width the field does vary — by a part in a thousand — and that remainder grips the atom in two new ways: the wave's oscillating $\mathbf B$ torques the atom's magnetic moment (M1), and the gradient of $\mathbf E$ pulls unevenly on the two ends of the charge cloud (E2). Think of E1, M1, E2 as three doors out of an excited state, each five orders narrower than the last. When the E1 door is open, nobody notices the others. When symmetry welds it shut, the slow doors are the only doors — and whether the atom gets to use them is a race: at laboratory densities a collision steals the energy within microseconds (why nebulium never appeared on Earth); in a nebula at a few thousand particles per cm³, collisions take days, the patient doors win, and the "forbidden" green lines outshine everything else.

Figure

Three doors out of an excited state — and the outliers that made the slow doors famous

The magnetic door has a personality of its own. The M1 operator is built from $\hat{\mathbf L}+g_s\hat{\mathbf S}$ — pure angular momentum, no position operator anywhere. It cannot change the orbital's shape ( $\Delta\ell=0$ ) and, more sharply, cannot change the radial state at all: same- $\ell$ , different- $n$ radial functions are orthogonal and M1 has no $r$ to bridge them. M1 transitions are reorientations — fine-structure and hyperfine flips like the 21 cm spin-flip — never orbital jumps. That observation, proved below, is why 2s cannot even use the magnetic door.

Figure

One term, two operators: the ik·r correction splits into a torque on μ and a stretch on the charge cloud

The next term in the phase

The Dipole Approximation established the expansion and its small parameter; we now cash the deferral it made. The interaction of a bound electron (charge $q=-e$ , $e>0$ ) with the classical wave $\mathbf A=A_0\hat{\boldsymbol\epsilon}\,e^{i(\mathbf k\cdot\mathbf r-\omega t)}$ in Coulomb gauge is, from minimal coupling plus the Pauli spin term,

\hat H'=-\frac{q}{m_ec}\,\mathbf A\cdot\hat{\mathbf p}-\frac{g_s q}{2m_ec}\,\hat{\mathbf S}\cdot(\nabla\times\mathbf A)=+\frac{e}{m_ec}\,\mathbf A\cdot\hat{\mathbf p}+\frac{g_se}{2m_ec}\,\hat{\mathbf S}\cdot\mathbf B

where: the first term is the orbital coupling whose $e^{i\mathbf k\cdot\mathbf r}=1$ piece already gave E1, and the second is the spin–field coupling that will complete the magnetic dipole. A sign worth pausing on: our source states "electron charge $-e$ " yet writes the orbital term as $-\frac{e}{m_ec}\mathbf A\cdot\hat{\mathbf p}$ ; minimal coupling gives $-\frac{q}{m_ec}\mathbf A\cdot\hat{\mathbf p}=+\frac{e}{m_ec}\mathbf A\cdot\hat{\mathbf p}$ for $q=-e$ . Harmless in rates (they carry $|H'_{fi}|^2$ ), fatal the moment amplitudes interfere — repaired here, flagged in the cheat sheet. (Gaussian units this section; the full gauge story is a future article's.) Keeping the next order of the phase inside the orbital term produces the object this article is about:

\hat H'^{(1)}\;\propto\;i\,(\mathbf k\cdot\mathbf r)\,(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})

where: the product is neither symmetric nor antisymmetric under swapping $\mathbf k\leftrightarrow\hat{\boldsymbol\epsilon}$ — and that swap-symmetry is the right knife. The antisymmetric half will be a magnetic moment reading the wave's $\mathbf B$ ; the symmetric half, a quadrupole reading the gradient of $\mathbf E$ . Every "forbidden" line in the sky comes out of this one line of algebra.

Derivation AFrom the Field Gradient to the M1 and E2 Operators

We derive both next-order operators by splitting $i(\mathbf k\cdot\mathbf r)(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})$ into antisymmetric and symmetric parts under $\mathbf k\leftrightarrow\hat{\boldsymbol\epsilon}$ exchange: the antisymmetric part assembles into pure angular momentum (M1); the symmetric part converts, via $\hat{\mathbf p}=\frac{im_e}{\hbar}[\hat H_0,\mathbf r]$ , into a position-quadratic tensor (E2). Watch two things: operator ordering (the classical split identity picks up a quantum remainder $\propto(\mathbf k\cdot\hat{\boldsymbol\epsilon})[\hat r,\hat p]$ from the same-component terms — and transversality $\mathbf k\cdot\hat{\boldsymbol\epsilon}=0$ kills it exactly, a rescue worth seeing once), and the factors of $\hbar$ (our source slipped one in, and dimensions catch it).

Step 1. Split identically: $i(\mathbf k\cdot\mathbf r)(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})=\frac{i}{2}\big[(\mathbf k\cdot\mathbf r)(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})+(\mathbf k\cdot\hat{\mathbf p})(\hat{\boldsymbol\epsilon}\cdot\mathbf r)\big]+\frac{i}{2}\big[(\mathbf k\cdot\mathbf r)(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})-(\mathbf k\cdot\hat{\mathbf p})(\hat{\boldsymbol\epsilon}\cdot\mathbf r)\big]$ . Nothing is assumed; add and subtract.

Step 2. The antisymmetric bracket is a double cross product: $(\mathbf k\cdot\mathbf r)(\hat{\boldsymbol\epsilon}\cdot\hat{\mathbf p})-(\mathbf k\cdot\hat{\mathbf p})(\hat{\boldsymbol\epsilon}\cdot\mathbf r)=(\mathbf k\times\hat{\boldsymbol\epsilon})\cdot(\mathbf r\times\hat{\mathbf p})+(\mathbf k\cdot\hat{\boldsymbol\epsilon})\,i\hbar$ -terms — verified symbolically with noncommuting $\hat r_i,\hat p_j$ : the cross-product part uses only $i\neq j$ pairs (which commute), while the same-component commutators assemble into a remainder proportional to $\mathbf k\cdot\hat{\boldsymbol\epsilon}$ , which transversality sets to zero. Light's own geometry makes the operator identity exact. Recognize both factors: $\mathbf r\times\hat{\mathbf p}=\hat{\mathbf L}$ , and for our wave $\nabla\times\mathbf A$ points along $\mathbf k\times\hat{\boldsymbol\epsilon}$ — the direction of $\mathbf B$ . So the antisymmetric half reads $\frac{i}{2}(\mathbf k\times\hat{\boldsymbol\epsilon})\cdot\hat{\mathbf L}$ . Why the ħ-audit matters: our source writes this same line as $\frac{i}{2\hbar}(\mathbf k\times\hat{\boldsymbol\epsilon})\cdot\hat{\mathbf L}$ while using dimensionful $\hat{\mathbf L}$ elsewhere — the extra $1/\hbar$ fails a one-line dimension check ( $\mathbf r\times\hat{\mathbf p}$ IS $\hat{\mathbf L}$ , no conversion factor). Stray $\hbar$ 's are how factor-of- $10^3$ rate errors are born; audit them like signs.

Step 3. Restore prefactors, add spin. The orbital antisymmetric piece couples $\hat{\mathbf L}$ to the wave's $\mathbf B$ with coefficient $\frac{e}{2m_ec}=\frac{\mu_B}{\hbar}$ ; the Pauli term contributes $g_s\hat{\mathbf S}$ with the same structure (its $\nabla\times\mathbf A$ is $\mathbf B$ directly). Together:

\hat H'_{M1}=-\hat{\boldsymbol\mu}\cdot\mathbf B,\qquad \hat{\boldsymbol\mu}=-\frac{\mu_B}{\hbar}\big(\hat{\mathbf L}+g_s\hat{\mathbf S}\big)

the moment dictionary of Magnetic Moment and the Gyromagnetic Ratio — its loop-derived orbital moment and $g_s=2$ imported wholesale, the exactness of $g_s$ "owned by the relativistic track" in its words.

Step 4. The symmetric bracket becomes position-quadratic. Insert $\hat{\mathbf p}=\frac{im_e}{\hbar}[\hat H_0,\mathbf r]$ : between eigenstates the commutator collapses onto $\omega_{fi}$ , $\langle f|\{r_i,\hat p_j\}|i\rangle=im_e\omega_{fi}\langle f|r_ir_j|i\rangle$ , so the matrix element couples the field gradient to $r_ir_j$ . Subtracting the trace — which dies against $k_i\epsilon_i=0$ , transversality's second free lunch — leaves the traceless $Q_{ij}=e\,(r_ir_j-\frac13r^2\delta_{ij})$ : rank-2, parity-EVEN. Verified: the split identity checked with noncommuting components (verify script check 1); the ħ-audit (check 2); the sign of the minimal-coupling cross term under the stated charge convention (check 3); operator ranks and parities feed the sweeps of checks 6–7.

What the new doors allow

Both operators are parity-EVEN — $\hat{\mathbf L}$ is axial, $r_ir_j$ quadratic — so M1 and E2 connect same-parity states: precisely the pairs E1 can never touch, complementary by construction. Their rule tables follow from the Wigner–Eckart theorem as in Electric Dipole Selection Rules, with ranks 1 and 2:

	E1 (rank 1, odd)	M1 (rank 1, even)	E2 (rank 2, even)
parity	must flip	must NOT flip	must NOT flip
$\Delta\ell$	$\pm1$	$0$ (and no radial change)	$0,\pm2$ (not $0\leftrightarrow0$ , needs $\ell+\ell'\geq2$ )
$\Delta j$	$0,\pm1$ , no $0\to0$	$0,\pm1$ , no $0\to0$	$0,\pm1,\pm2$ ; banned: $0\to0$ , $0\leftrightarrow1$ , $\tfrac12\to\tfrac12$
$\Delta m$	$0,\pm1$	$0,\pm1$	$0,\pm1,\pm2$
natural habitat	everything bright	fine/hyperfine flips, 21 cm	nebular lines, d-state decays

where: M1's sharpest entry is the radial statement behind $\Delta\ell=0$ : with no $r$ in the operator, the radial factor is $\langle R_{n'\ell}|R_{n\ell}\rangle=\delta_{n'n}$ — M1 cannot change $n$ at all. The machine sweeps in the verify script confirm both patterns wholesale: every nonzero $\hat L_q$ element sits inside $\{\Delta\ell=0,\ m'=m+q,\ \ell\neq0\}$ and every nonzero $Y_2^q$ element inside $\{\Delta\ell=0,\pm2,\ m'=m+q,\ \ell+\ell'\geq2\}$ — with a lesson hiding in the sweep's fine print: a few elements inside the allowed set also vanish ( $\langle\ell m|\hat L_z|\ell m\rangle$ at $m=0$ ; the $q=0$ diagonals of $Y_2$ at $3m^2=\ell(\ell+1)$ , e.g. $\ell=3,m=\pm2$ ). Those are sporadic Clebsch–Gordan zeros, not new selection rules — rules forbid classes, CG zeros pick off individuals.

\frac{\Gamma_{M1}}{\Gamma_{E1}}\sim\Big(\frac{\mu_B}{e a_0 c}\Big)^2=\Big(\frac{\alpha}{2}\Big)^2\approx1.3\times10^{-5},\qquad \frac{\Gamma_{E2}}{\Gamma_{E1}}\sim(ka_0)^2\approx7.5\times10^{-6}

where: both ratios compare a forbidden amplitude to the E1 amplitude of the same frequency — the honest caveat being that real transitions never offer that choice: a forbidden line exists because its E1 twin does not. The numbers are the natural price tags of the operators, exhibited below, with actual rates always transition-specific.

Go deeper · Derivation B · 2 minThe Suppression-Scaling Theorem

Derivation BThe Suppression-Scaling Theorem

We prove the two price tags by amplitude bookkeeping: write each operator's matrix element as (its natural atomic scale) × (an order-unity angular factor), then take ratios — the scales are exact identities in $\alpha$ , the theorem's charm. The standing caveat: these are scales, not universal rate ratios; angular factors and $\omega^3$ phase space are transition-specific, and we exhibit both the clean scaling and a real case where frequency mismatch dominates.

Step 1. E1's natural amplitude scale: $\langle f|\hat H'_{E1}|i\rangle\sim eE_0\,a_0$ — field times dipole arm. Every E1 matrix element is this times an order-unity angular number ( $0.7449\,a_0/a_0$ for Lyman-α, from the sweep in Electric Dipole Selection Rules).

Step 2. M1's natural amplitude scale: $\langle f|\hat H'_{M1}|i\rangle\sim\mu_B B_0=\mu_B E_0/c$ for a plane wave ( $|\mathbf B|=|\mathbf E|/c$ ). The ratio is pure fundamental constants: $\frac{\mu_B B_0}{e a_0 E_0}=\frac{\mu_B}{e a_0 c}=\frac{\hbar}{2m_ec\,a_0}=\frac{\alpha}{2}$ (SI, where $\mu_B=e\hbar/2m_e$ ; in Gaussian units the same $\alpha/2$ emerges with the $c$ living inside $\mu_B$ ), using $a_0=\hbar/(m_ec\alpha)$ — an exact identity, verified symbolically. Rates square it: $(\alpha/2)^2=1.33\times10^{-5}$ . For hydrogenic $Z$ : $a_0\to a_0/Z$ makes the ratio $\alpha Z/2$ — heavier ions pay less, which is why forbidden lines brighten along isoelectronic sequences into the solar corona.

Step 3. E2's natural amplitude scale: the operator carries one extra power of position against one power of the wave's gradient, $\langle f|\hat H'_{E2}|i\rangle\sim eE_0a_0\times(ka_0)$ . The rate ratio is $(ka_0)^2$ ; at Lyman-α, $ka_0=2\pi a_0/\lambda=2.7\times10^{-3}$ , squared $7.5\times10^{-6}$ — and since $ka_0=\omega a_0/c\sim\alpha Z/2$ for transitions at the atomic-unit frequency scale, both doors cost the same $(\alpha Z)^2$ class. That coincidence is not numerology: both suppressions are the atom's size against the photon's size, expressed once through the field's magnetic component and once through its gradient.

Step 4. The honesty exhibit: the 21 cm line pays both taxes — the M1 factor and $\omega^3$ phase space at $5.8\times10^{-7}$ of Lyman-α's frequency. The ledger $(\alpha/2)^2\times(\omega_{21}/\omega_{Ly\alpha})^3=2.5\times10^{-24}$ lands within a factor of two of the true ratio $A_{10}/A_{2p}=4.5\times10^{-24}$ : scaling theorems predict exponents, specifics fill in order-unity dust. The end-to-end 21 cm estimate is owned by 21-cm Hydrogen Line ( $A_{est}=2.87\times10^{-15}\,\mathrm{s^{-1}}$ , "within a factor of a few of the accepted $2.85\times10^{-15}$ "); this article owns the operator and its price tag. Verified: $\mu_B/(ea_0c)=\alpha/2$ symbolic; $(\alpha/2)^2=1.33\times10^{-5}$ ; Lyman-α $ka_0=2.735\times10^{-3}$ , squared $7.48\times10^{-6}$ ; the 21 cm ledger within one decade of $A_{10}/A_{2p}$ — verify script checks 4, 8.

Go deeper · Derivation C · 1 minThe Doubly-Dead 2s State

Derivation CThe Doubly-Dead 2s State

We prove that hydrogen 2s cannot reach 1s through either one-photon door, forcing the two-photon channel. The goal is two exact zeros — one angular, one radial — arising from different operators failing for different reasons: the cleanest illustration in atomic physics that selection rules are structural, not accidental. Watch how short each kill is once the first derivation's operators are in hand.

Step 1. E1 kill (angular): $\langle 1s|r_q|2s\rangle$ carries $\int Y_0^{0*}Y_1^qY_0^0\,d\Omega=0$ for every $q$ — a $Y_1$ cannot bridge two $s$ -states (equivalently: both states even, $\mathbf r$ odd). Computed exactly: zero, all three components.

Step 2. M1 kill (radial). $\langle 1s|(\hat L_q+g_s\hat S_q)|2s\rangle$ : the operator touches no radial coordinate, so the matrix element factorizes with the radial overlap $\int_0^\infty R_{10}R_{20}\,r^2dr$ , which is exactly zero — same- $\ell$ eigenfunctions of the same Hamiltonian are orthogonal (Hydrogenic Wavefunctions and Quantum Numbers catalogues them). The kill is doubled on the angular side: $\hat{\mathbf L}Y_0^0=0$ — an $s$ -state has no orbital moment to flip — and the spin piece, blind to space, still rides the zero radial overlap.

Step 3. E2 kill, for completeness: rank 2 cannot connect $\ell=0$ to $\ell=0$ (triangle needs $\ell+\ell'\geq2$ ). Every one-photon multipole of every rank $k\geq1$ fails the same way — a photon carries at least one unit of angular momentum and $0\to0$ has none to give (the ban proved in Electric Dipole Selection Rules).

Step 4. What remains is simultaneous two-photon emission — 2E1 through virtual $p$ -states — computed by Breit and Teller in 1940: $\tau_{2s}=0.122$ s, $7.6\times10^7$ 2p-lifetimes. Symmetry does not merely slow the state; it changes the kind of light — a smooth continuum of photon pairs sharing 10.2 eV, visible in nebular spectra between the lines. Verified: $\langle R_{10}|R_{20}\rangle=0$ exact; $\int Y_0^{0*}Y_1^qY_0^0\,d\Omega=0$ all $q$ ; $\hat L\,Y_0^0=0$ ; $\tau_{2s}/\tau_{2p}=7.6\times10^7$ — verify script checks 5, 10.

Limiting cases

case	result	meaning
$ka_0\to0$ (long wavelength)	only E1 survives	the dipole approximation is exact in the limit; multipoles are the finite-size correction
E1 open	forbidden doors invisible	branching ratio $\sim10^{-5}$ ; nobody detects M1 against an open E1 channel
high $Z$ (isoelectronic sequence)	suppression $\propto(\alpha Z)^2$ weakens	coronal Fe XIV green line: forbidden yet prominent at $Z_{\rm eff}\sim14$
high density (lab, $\sim10^{19}$ cm⁻³)	collisions win	metastables quenched in μs; forbidden lines unobservable — the nebulium mystery
low density ( $\lesssim n_{\rm crit}\sim7\times10^5$ cm⁻³)	radiation wins	[O III] lines dominate nebular spectra; density diagnostics from line ratios
$j=0\to j'=0$	dead at every one-photon multipole	two-photon or internal conversion only — no rank can carry zero

Worked examples

Example 1 — price the doors, then read the ladder. Compute both suppression factors for hydrogen-class transitions and place the cast of this article on one lifetime ladder.

Step 1. M1 price: $\mu_B/(ea_0c)=\alpha/2=3.65\times10^{-3}$ in amplitude, $(\alpha/2)^2=1.33\times10^{-5}$ in rate.

Step 2. E2 price at Lyman-α: $ka_0=2\pi a_0/\lambda=2\pi(0.0529\,\mathrm{nm})/(121.57\,\mathrm{nm})=2.73\times10^{-3}$ ; squared, $7.5\times10^{-6}$ . Both doors: the $(\alpha Z)^2$ class, five-plus orders below E1.

Step 3. The ladder: a pure multipole tax on 2p's 1.6 ns predicts $\approx0.1$ ms metastables — real forbidden lines are slower still because $\omega^3$ bites: [O III] 41 s, 2s 0.122 s, 21 cm $3.5\times10^{14}$ s. Span: $\log_{10}(3.5\times10^{14}/1.6\times10^{-9})\approx23.3$ orders — from operators one order apart in $ka_0$ .

Example 2 — when does a nebula beat the lab? The [O III] critical density. The $^1D_2$ level of O²⁺ decays radiatively at $A_{\rm tot}=A_{5007}+A_{4959}=1.81\times10^{-2}+6.21\times10^{-3}=2.43\times10^{-2}\,\mathrm{s^{-1}}$ ( $\tau=41$ s). Collisions de-excite it at rate $n_e q$ with $q\sim3\times10^{-8}\,\mathrm{cm^3\,s^{-1}}$ at $10^4$ K.

Step 1. The race: radiation wins below the critical density $n_{\rm crit}=A_{\rm tot}/q$ .

Step 2. Evaluate: $n_{\rm crit}=2.43\times10^{-2}/3\times10^{-8}\approx8\times10^{5}\,\mathrm{cm^{-3}}$ — bracketing the accepted $\approx7\times10^5$ (our $q$ is an order-of-magnitude coefficient).

Step 3. Read it: nebulae at $10^2$ – $10^4\,\mathrm{cm^{-3}}$ sit three orders below critical — every excitation radiates, [O III] blazes. Air sits fourteen orders above — quenched utterly; Earth never saw nebulium. The forbidden-line sky is a density filter. (Bonus: $A_{5007}/A_{4959}=2.91$ is fixed by atomic physics — same upper level — the observers' built-in 3:1 calibration check.)

The detective story

Origins · 1 minForbidden Transition and how this was discovered

Forbidden Transition

1927–1942 · Pasadena, Uppsala

Two ghost elements died to make this concept. "Nebulium" was born in 1864 when Huggins found green nebular lines matching nothing terrestrial; "coronium" followed from the 1869 eclipse. Both survived six decades because no laboratory could replicate what a near-vacuum does effortlessly. In 1927 Ira Bowen at Caltech computed where the metastable levels of abundant ions would radiate if left alone: the [O III] arithmetic landed exactly on the nebular green pair — nebulium was oxygen, waiting out a forbidden transition in gas too thin for collisions. Bengt Edlén finished the job in Uppsala, pinning the coronal lines through 1939–1942: the 5303 Å green line fell to Fe XIV, iron stripped of thirteen electrons, announcing a million-degree corona nobody had believed. The brackets in [O III] now mark not ignorance but the most delicate density probes the sky provides.

History

1864

— Huggins resolves green emission lines in a planetary nebula; "nebulium" enters the lexicon

1869

— total-eclipse observers record the coronal green line; "coronium" follows

1927

— Bowen identifies the nebular lines as forbidden [O III] and [N II] transitions from metastable levels

1940

— Breit and Teller compute the two-photon decay of hydrogen 2s: 0.12 s, the escape route when every one-photon door is dead

1942

— Edlén (with Grotrian's 1939 lead) identifies the coronal lines with forbidden transitions of highly ionized iron: the corona is at a million kelvin

1951

— the 21 cm hyperfine line — the sky's grandest forbidden transition — is detected at Harvard, then Leiden and Sydney

Going further

References: Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §5.7–5.8 · Griffiths, Introduction to Quantum Mechanics, 3rd ed., Ch. 11 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. II, Complement A_XIII/D_XIII · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 18 · Bransden & Joachain, Physics of Atoms and Molecules, 2nd ed., Ch. 4 · Townsend, A Modern Approach to Quantum Mechanics, 2nd ed.

Provenance note (verbatim from the source corpus): All selection rules verified against Sakurai & Napolitano §5.8, Cohen-Tannoudji Vol. II Complement D_XIII, and Bransden–Joachain Ch. 4. A verifier should recompute: (i) the symmetric/antisymmetric decomposition $(\mathbf{k}\cdot\mathbf{r})(\hat\epsilon\cdot\mathbf{p})$ yielding $\mathbf{L}$ and $r_ir_j$ ; (ii) the Wigner–Eckart triangle rules for $k=1$ (M1) and $k=2$ (E2); (iii) the suppression scaling $(ka_0)^2=(\alpha Z)^2$ . Numbers stated: $g_s\approx2.0023$ (CODATA), $\alpha\approx1/137$ , [O III] lines 4959/5007 Å, 21 cm at 1420.4 MHz with $A\approx2.9\times10^{-15}\,\mathrm{s^{-1}}$ (Wild 1952; standard astrophysics texts), coronal green line 5303 Å [Fe XIV]. Omit any rate requiring a non-standard numerical coefficient. All recomputations are in this article's verify script — plus two repairs to the source's own algebra: the minimal-coupling sign under its stated charge convention, and a spurious $1/\hbar$ in its M1 identification line (both machine-checked).

Prerequisites: The Dipole Approximation · Fermi's Golden Rule · Semiclassical Theory of Atom–Radiation Interaction · First-Order Transition Amplitudes and Probabilities. Leads to: Absorption and Stimulated Emission · Density of Final States — and onward to Lifetime, Linewidth, and Natural Broadening, where patience becomes precision.

If the dipole approximation is so good, why do these tiny corrections ever matter

Because when E1's matrix element is exactly zero, "dominant" and "only" part ways: the multipole terms are invisible corrections to an allowed line but the entire signal on a forbidden one. Symmetry zeros, not smallness, decide when the fine print becomes the contract.

Is the M1 operator really independent of the radial coordinate — where did the r go

The antisymmetric piece assembles $\mathbf r$ and $\hat{\mathbf p}$ into $\hat{\mathbf L}=\mathbf r\times\hat{\mathbf p}$ — position enters only through the angular momentum it builds, and $\hat{\mathbf L}$ acts purely on angles. That is the physics, not a technicality: an M1 transition is the moment precessing in the wave's $\mathbf B$ , a reorientation leaving the orbital's shape and size untouched. Hence $\Delta\ell=0$ , $\Delta n=0$ , and the 21 cm line as a pure spin flip.

The suppression is only 10⁻⁵ — why is the 21 cm lifetime eleven MILLION years rather than milliseconds

Two separate taxes. The multipole tax is indeed $\sim10^{-5}$ ; but spontaneous rates also pay $\omega^3$ , and 21 cm sits at $5.8\times10^{-7}$ of Lyman-α's frequency — cube it and lose nineteen more orders. The full ledger matches the measured $A_{10}/A_{2p}=4.5\times10^{-24}$ to order-unity dust. Low frequency, not forbiddenness, is the senior partner in that lifetime.

What finally happens to a state where E1, M1 and E2 are all dead

Hydrogen 2s is the textbook case: E1 dead (parity/triangle), M1 dead (radial orthogonality), E2 dead ( $0\to0$ in ℓ). The atom emits two photons at once, sharing 10.2 eV in a continuous spectrum, $\tau=0.122$ s (Breit & Teller 1940). In dense plasmas collisions win instead; in nebulae the two-photon continuum is directly observed between the lines. Absolute stability belongs only to ground states.

Why do forbidden lines get relatively stronger in highly charged ions like Fe XIV

The suppression scales as $(\alpha Z)^2$ , so at $Z_{\rm eff}\sim14$ the M1 price drops from $10^{-5}$ to $\sim10^{-2}$ . Add coronal temperatures (collisions feeding metastable levels) and coronal thinness (nothing quenches), and the "forbidden" 5303 Å green line becomes the corona's visual signature — Edlén's 1942 identification, first proof of a million-kelvin corona.

Are these selection rules exact, or do they leak like Delta-S = 0 does

The rules for a given multipole operator are exact — Wigner–Eckart zeros. What leaks is the labeling of states: real eigenstates are slightly mixed by spin–orbit, hyperfine, or external fields, so a nominally forbidden line borrows a small allowed amplitude — precisely how the Al⁺ clock's $J=0\to J=0$ line acquires its 8 mHz width. Operator theorems never break; state labels do, by calculable amounts.