Radiative Frequency Shift and Line Width

The preceding article prosecuted the self-force's crimes; this one presents its life's work. The same Abraham–Lorentz term that manufactures runaways and preacceleration, inserted legally — as a perturbation on a bound electron, deep inside its validity domain — produces the most successful prediction classical electrodynamics ever made about atoms: every spectral line has an irreducible width, the line's shape is a Lorentzian, and its width in wavelength is a universal constant of the electron, $\Delta\lambda=4\pi r_e/3=1.18\times10^{-4}$ Å, the same for every line of every element at every frequency. The derivation is a clean chain with no free parameters. Radiation drains a bound oscillator at the rate $\Gamma=\omega_0^2\tau$ ; a draining oscillation is a decaying wavetrain; and the spectrum of a decaying wavetrain — computed by spending the Parseval ledger built in Larmor Formula from the Liénard–Wiechert Fields — is a Lorentzian of full width $\Gamma$ . Each link is audited here, including the one textbooks rush: why the width in wavelength forgets the frequency entirely. The chain also produces a shift: the damped oscillator rings slightly flat, and the reactive part of the self-force pulls the line further — by an amount that depends on the point-charge cutoff and is therefore, classically, not a number at all. That honest failure is the door to quantum electrodynamics: the measured 1057 MHz displacement of the hydrogen $2s$ level, the Lamb shift, is this classical shift done right. Width, shape, shift; one perturbative force, its full spectroscopic testimony — and a quantitative map of when Doppler and collision broadening bury it.

Key takeaways

A bound electron radiating per Larmor damps at $\Gamma=\omega_0^2\tau=\dfrac{e^2\omega_0^2}{6\pi\varepsilon_0mc^3}$ — the classical natural linewidth (FWHM in angular frequency); $\Gamma/\omega_0=\omega_0\tau\sim10^{-8}$ for optical lines, so the perturbative insertion is legal by eight orders.
The emitted spectrum is the acceleration's power spectrum (the Parseval ledger, spent by name): an exponentially decaying wavetrain gives the Lorentzian $g(\omega)=\dfrac{\Gamma/2\pi}{(\omega-\omega_1)^2+(\Gamma/2)^2}$ , FWHM $\Gamma$ , with $\omega^{-2}$ wings that dominate any Gaussian far from line center.
In wavelength the width is universal: $\Delta\lambda=\dfrac{\lambda^2}{2\pi c}\Gamma=2\pi c\tau=\dfrac{4\pi}{3}r_e=1.18\times10^{-4}$ Å — the $\omega_0^2$ in $\Gamma$ exactly cancels the $\lambda^2$ Jacobian; every classical line has the same width in Å.
Real lines obey the classical result times an oscillator strength: $A=3(g_l/g_u)f\,\Gamma_{\text{cl}}$ — audited on Na D2 (ratio 0.961 both ways) and Lyman-α (0.417 both ways); the $f$ -sum rule is why the classical model worked at all.
The free-decay line rings flat by the finite pull $\delta\omega=\Gamma^2/8\omega_0$ (26 mHz for Na D2); the reactive self-force shift is cutoff-dependent — classically ill-defined — and its honest quantum completion is the Lamb shift, 1057 MHz for hydrogen $2s_{1/2}$ – $2p_{1/2}$ .
Against company: Doppler broadening (Gaussian, $\sim$ GHz at oven temperatures) beats natural width ( $\sim$ 10 MHz) by two orders; collisions add a second Lorentzian; the observed profile is their convolution, the Voigt profile.
Key idea: radiation reaction, used inside its validity domain, converts every sharp resonance into a decaying wavetrain whose Fourier transform — a Lorentzian of width Γ = ω₀²τ and universal wavelength width 4πr_e/3 — is the classical fingerprint on every spectral line ever measured.

The core insight

Radiation reaction damps a bound electron's oscillation; an exponentially decaying oscillation has a Lorentzian power spectrum, giving every spectral line an irreducible natural width $\Gamma=e^2\omega_0^2/6\pi\varepsilon_0mc^3$ and a small downward frequency shift — the classical ancestor of spontaneous-emission linewidths and the Lamb shift.

Where this lives today

The Lorentzian of Derivation B is arguably the most-fitted function in experimental physics. Every atomic clock's stability budget begins with a natural linewidth: the cesium standard interrogates a transition whose spontaneous rate is negligible, while optical-lattice clocks choose ultranarrow lines (the strontium clock transition's natural width is one millihertz) precisely to push the $\Gamma$ -floor down — the fractional-frequency records near $10^{-18}$ are exercises in living just above this article's Step 2. Laser-cooling forces, trap depths, and the Doppler temperature limit of Example 2 are all written in units of $\Gamma$ ; a cold-atom lab is a building-sized appliance for manipulating one Lorentzian. The wings run the older industries. Stellar spectroscopy classifies atmospheres by damping wings (the "Lorentzian troughs" of strong Fraunhofer lines), radiative-transfer codes carry Voigt profiles line-by-line through millions of transitions, and plasma diagnostics reads density off collisional widths. Meanwhile the shift half of this article's title became a precision industry of its own: the Lamb shift, measured to sub-kHz in hydrogen spectroscopy, anchors QED tests and — through the proton-radius puzzle of muonic hydrogen — turned line-shift metrology into a probe of nuclear structure. A width the classical electron predicted, and a shift it could only gesture at: both are now measured to parts in $10^{6}$ of themselves.

Origins · 1 minWillis Eugene Lamb Jr. and how this was discovered

Willis Eugene Lamb Jr.

1913–2008 · the experimentalist-theorist whose microwave measurement of a forbidden shift forced quantum electrodynamics to grow up

Lamb came to the hydrogen atom through radar. Trained as a theorist under Oppenheimer, he spent the war on magnetrons and microwave technique at Columbia — and emerged holding exactly the tool the oldest problem in spectroscopy needed. Dirac theory said the hydrogen $2s_{1/2}$ and $2p_{1/2}$ levels are exactly degenerate; optical spectroscopy could not resolve the question through the Doppler haze (Example 2's 1.5 GHz fog). In 1947 Lamb and his student Retherford drove the transition directly with microwaves in a beam of metastable atoms — a Doppler-immune method — and found the levels split by 1057 MHz. The number detonated at the Shelter Island conference weeks later: Bethe computed it on the train home, taming the same divergent self-energy this article meets classically in Derivation B's Step 4, and the renormalization programme of QED was born. Lamb received the 1955 Nobel Prize; characteristically, he spent later decades needling the field he had founded, insisting that "photon" was a licensed word his colleagues used without a license. The shift bearing his name remains the cleanest place where the vacuum's own fluctuations move a spectral line — the radiative shift, finally computed.

History

1895

— Planck begins his resonator programme: a damped oscillating charge as the model coupling matter to the radiation field, radiation damping included from the start

1906

— Lorentz's electron theory delivers the classical natural width and the Lorentzian profile; the universal wavelength width becomes a testable prediction

1930

— Weisskopf and Wigner derive the quantum line shape of spontaneous emission: a Lorentzian of width A, the classical result wearing its oscillator strength

1947

— Lamb and Retherford measure the 1057 MHz hydrogen shift; Bethe computes it with a renormalized cutoff, converting the classically ill-defined radiative shift into the first QED triumph

1958

— Mössbauer discovers recoil-free gamma emission: natural linewidths at fractional 10⁻¹³, the reciprocity theorem's most extreme laboratory

The physical picture

Picture the atom of classical electron theory: an electron on a spring, resonant at $\omega_0$ , set ringing by a collision or a discharge. Each cycle it accelerates, and per the Larmor formula each acceleration is taxed — the Radiation Reaction Force article priced the tax and the Abraham-Lorentz Radiation Reaction Force article next door proved when charging it perturbatively is legal. For an optical electron the levy per cycle is fantastically small, $\omega_0\tau\sim10^{-8}$ of the stored energy, so the oscillation persists for tens of millions of cycles. But it does not persist forever: the amplitude decays as $e^{-\Gamma t/2}$ , the energy as $e^{-\Gamma t}$ , and the wavetrain the atom broadcasts is long — but finite. A finite wavetrain cannot be monochromatic. That is Fourier analysis, not electrodynamics: a pure frequency requires an infinite sinusoid, and truncating or damping it spreads the spectral content over a band $\Delta\omega\sim1/\Delta t$ . The damped exponential makes this exact — its power spectrum is a Lorentzian, peaked at the ringing frequency, with full width at half maximum precisely equal to the energy decay rate $\Gamma$ . Long-lived oscillators make sharp lines; leaky ones make broad lines; the reciprocity $\Gamma\cdot t_{\text{life}}=1$ is the classical shadow of the energy–time uncertainty relation, here with nothing quantum in sight. Now the striking part. Faster oscillators pay proportionally more tax — $\Gamma\propto\omega_0^2$ — yet a spectrograph disperses in wavelength, and converting width to wavelength costs a Jacobian $\lambda^2/2\pi c\propto\omega_0^{-2}$ . The two factors of $\omega_0^2$ cancel exactly: every classical line, X-ray or infrared, hydrogen or mercury, is $1.18\times10^{-4}$ Å wide. A universal length, built from $r_e$ alone — the electron signing every line of every spectrum with the same pen stroke.

Figure

The hero pair — a radiatively damped wavetrain and its computed power spectrum: the decay rate becomes the Lorentzian full width

Figure

Same full width, different personalities — a Lorentzian and a Gaussian of equal FWHM, computed: the natural line's wings outlive the Doppler core by orders of magnitude

The math: one damping rate, three deliverables

The working equation is the harmonically bound electron with the radiation-reaction term inserted perturbatively — legally, since $\omega_0\tau\sim10^{-8}$ sits eight orders inside the validity domain proved next door in Abraham-Lorentz Radiation Reaction Force:

\ddot x+\Gamma\,\dot x+\omega_0^2\,x=0, \qquad \Gamma=\omega_0^2\,\tau=\frac{e^2\omega_0^2}{6\pi\varepsilon_0 m c^3}, \qquad (1)

where: $\tau=6.27\times10^{-24}$ s is the Larmor characteristic time and the replacement $m\tau\dddot x\to-m\omega_0^2\tau\,\dot x$ uses the undamped motion inside the tiny reaction term — the same on-shell reduction the Radiation Reaction Force article demonstrated and the runaway article's moments ladder justifies as the first rung of the slow-force expansion. Equation (1) is the good sector of the self-force: no third derivative survives, no runaway exists, and everything below is its spectroscopy. The driven version of the same oscillator, $\varepsilon(\omega)$ and all, belongs to Absorption and Dispersion; here the atom rings freely and we listen.

Derivation AThe damping rate two ways, and the universal wavelength width

The goal is the rate $\Gamma=\omega_0^2\tau$ established by two independent routes — the force route (on-shell reduction of the reaction term) and the energy route (Larmor power over stored energy) — followed by the theorem that makes it famous: expressed as a wavelength width, $\Gamma$ is a universal constant, the same for every spectral line in nature. The two-route strategy is deliberate: the force route exposes where the approximation enters, the energy route certifies the coefficient with no reference to the self-force at all, and their agreement is the perturbation theory auditing itself. Watch the $\omega_0$ bookkeeping in the final step — the cancellation is exact, not approximate.

Step 1. The force route. For nearly harmonic motion $x\approx x_0\cos\omega_0t$ , each time derivative brings a factor $\omega_0$ , so the jerk term evaluates on-shell: $\dddot x=-\omega_0^2\dot x$ exactly for pure oscillation. The reaction force $m\tau\dddot x$ becomes $-m\omega_0^2\tau\,\dot x$ — a genuine velocity drag with coefficient $\Gamma=\omega_0^2\tau$ , giving (1). Why this is legal here and criminal for free particles: the substitution assumes the motion is dominated by the spring, i.e. the reaction is a perturbation — true when $\Gamma\ll\omega_0$ , which is $\omega_0\tau\ll1$ , satisfied by eight orders for optical transitions. The runaway article shows what happens when no such dominant motion exists to substitute.

Step 2. The energy route. The oscillator's stored energy is $E=\tfrac{1}{2}m\omega_0^2x_0^2$ ; its cycle-averaged Larmor power, with $\langle a^2\rangle=\tfrac{1}{2}\omega_0^4x_0^2$ , is $P=\mu_0e^2\omega_0^4x_0^2/12\pi c$ . The fractional loss rate:

\Gamma=\frac{P}{E}=\frac{\mu_0e^2\omega_0^2}{6\pi mc}\cdot\frac{1}{1}=\omega_0^2\,\tau .

where: no self-force expression was used — only the radiated power and the stored energy. The agreement with Step 1 confirms the on-shell reduction kept the right coefficient. Energy then decays as $e^{-\Gamma t}$ , amplitude as $e^{-\Gamma t/2}$ .

Step 3. Solve (1) in the weak-damping regime. The underdamped solution is $x(t)=x_0e^{-\Gamma t/2}\cos\omega_1t$ with

\omega_1=\omega_0\sqrt{1-\Gamma^2/4\omega_0^2}\;\approx\;\omega_0-\frac{\Gamma^2}{8\omega_0} .

The ring frequency sits below the natural frequency by a finite, cutoff-independent pull $\Gamma^2/8\omega_0$ — tiny (second order in $\Gamma/\omega_0$ ) but real; Derivation B prices it and separates it from the deeper, divergent shift.

Step 4. The universal width. Anticipating Derivation B's result that the spectral FWHM equals $\Gamma$ , convert to wavelength with the dispersion Jacobian $|d\lambda/d\omega|=\lambda^2/2\pi c$ :

\Delta\lambda=\frac{\lambda^2}{2\pi c}\,\Gamma=\frac{\lambda^2}{2\pi c}\cdot\frac{4\pi^2c^2}{\lambda^2}\,\tau\cdot\frac{1}{1}=2\pi c\,\tau=\frac{4\pi}{3}\,r_e=1.18\times10^{-4}\ \text{Å}.

where: $\omega_0=2\pi c/\lambda$ makes $\Gamma\propto\lambda^{-2}$ , and the Jacobian's $\lambda^{2}$ annihilates it. Every frequency reference has vanished: the classical natural width is a pure length, $\tfrac{4}{3}\pi$ classical electron radii, roughly a hundredth of a picometer, identical for the 21 cm line and a hard X-ray transition. The Radiation Reaction Force article's worked example verified this number for a 500 nm line; here it is a theorem about all lines at once. Verified: $\Gamma$ (Na D2, 589.0 nm) $=6.41\times10^7\ \mathrm{s^{-1}}$ and $2\pi c\tau=1.180\times10^{-14}$ m recomputed from CODATA constants; the two routes' coefficients agree symbolically; $\omega_1$ expansion checked to second order.

Go deeper · Derivation B · 2 minThe Lorentzian line shape from the Parseval ledger, and the two shifts

Derivation BThe Lorentzian line shape from the Parseval ledger, and the two shifts

The goal is the shape: the radiated energy spectrum of the decaying oscillation, computed by spending the spectral tool this chapter already owns — the Parseval ledger of Larmor Formula from the Liénard–Wiechert Fields, $dW/d\omega=(\mu_0q^2/6\pi^2c)\,|\tilde a(\omega)|^2$ — on the damped wavetrain of Derivation A. The strategy: Fourier-transform the one-sided decaying cosine, keep the resonant pole honestly (stating the rotating-wave discard of the far pole), normalize, and read off the FWHM. Then the shifts, kept strictly separate: the finite second-order pull of the damped ring, and the reactive self-energy shift that classical theory cannot define — the failure that Bethe turned into the Lamb shift. The bookkeeping to watch: FWHM $\Gamma$ refers to the energy spectrum; the amplitude's decay rate is $\Gamma/2$ , and conflating the two is the factor-of-two error every spectroscopy student makes once.

Step 1. The acceleration's transform. The emitting motion is $x(t)=x_0e^{-\Gamma t/2}\cos\omega_1t$ for $t>0$ (the atom is struck at $t=0$ ). Its acceleration is dominated by $a\approx-\omega_1^2x$ (corrections are down by $\Gamma/\omega_0$ ). The one-sided Fourier transform,

\tilde a(\omega)=\int_0^\infty a(t)\,e^{i\omega t}dt=-\frac{\omega_1^2x_0}{2}\left[\frac{1}{\Gamma/2-i(\omega+\omega_1)}+\frac{1}{\Gamma/2-i(\omega-\omega_1)}\right],

where: the two terms are poles at $\omega=\pm\omega_1-i\Gamma/2$ — the ring and its mirror. Near the physical line ( $\omega\approx+\omega_1$ ) the mirror term is smaller by $\sim\Gamma/\omega_1\sim10^{-8}$ and is dropped: the rotating-wave step, stated rather than smuggled.

Step 2. The spectrum. Insert into the Parseval ledger and normalize over the line:

\frac{dW}{d\omega}=W_{\text{tot}}\;g(\omega),\qquad g(\omega)=\frac{\Gamma/2\pi}{(\omega-\omega_1)^2+(\Gamma/2)^2},\qquad \int g\,d\omega=1 .

where: $g$ is the normalized Lorentzian: peak height $2/\pi\Gamma$ , half-maximum at $\omega-\omega_1=\pm\Gamma/2$ , hence FWHM $=\Gamma$ — the energy decay rate reappearing as a spectral width, the time–frequency reciprocity made exact. The wings fall as $(\omega-\omega_1)^{-2}$ , so slowly that at four half-widths the Lorentzian retains 5.9% of peak where a same-width Gaussian holds 0.0015% — a ratio of 3900, which is why far-wing absorption (pressure-broadened sodium streetlights, stellar damping wings) is always natural/collisional, never Doppler.

Step 3. The finite shift. From Derivation A, the ring sits at $\omega_1=\omega_0-\Gamma^2/8\omega_0$ : for Na D2 the pull is $\Gamma^2/8\omega_0=0.16$ rad/s — 26 millihertz on a 509 THz line, a fractional $5\times10^{-17}$ . Unmeasurably small, but perfectly well-defined: damping per se shifts lines by second order only.

Step 4. The divergent shift, honestly. The on-shell reduction in Derivation A kept only the dissipative part of the self-force. The reactive part — the piece of the self-field in phase with the acceleration — renormalizes the mass, and for a point charge it diverges linearly in the cutoff: the shell story of Radiation Reaction Force, recalled. After absorbing the divergence into the observed $m$ , a residual, cutoff-dependent frequency shift remains: classical theory can say a level shift exists but cannot compute it — the answer depends logarithmically on where you cut the electron's structure. This is not a technicality; it is the theory locating its own edge. In 1947 Bethe performed precisely this calculation in the quantum theory, where the natural cutoff ( $mc^2$ , the electron's quantum structure scale) is supplied by physics rather than fiat, and obtained $\sim1040$ MHz for the hydrogen $2s_{1/2}$ level against Lamb and Retherford's measured 1057 MHz. The classical radiative shift, done right, is the Lamb shift — and the Bridge from Classical Electrodynamics to QED collects the lesson. Verified: the Lorentzian's normalization $\int g\,d\omega=1$ and FWHM $=\Gamma$ recomputed numerically (numeric FWHM $0.995\Gamma$ on a finite grid, converging to $\Gamma$ ); the wing ratio at 4 half-widths $= (1/17)/2^{-16}=3855$ ; the Na pull $\Gamma^2/8\omega_0=0.161$ rad/s $=26$ mHz recomputed; pole structure of $\tilde a$ checked against the damped-cosine transform.

Go deeper · Derivation C · 2 minThe broadening taxonomy — natural, Doppler, collisional, and the oscillator-strength dictionary

Derivation CThe broadening taxonomy — natural, Doppler, collisional, and the oscillator-strength dictionary

The goal is the field guide: the three broadening mechanisms every real spectrum superposes, each derived at working level with its shape and its scaling, then the dictionary that maps the classical width onto measured lines — the oscillator strength $f$ , audited numerically on two real transitions. The strategy: treat the natural width as the fixed floor from Derivation B, derive the Doppler width from the Maxwellian projection, model collisions as phase-interruptions that terminate coherent emission, and let the data test $A=3(g_l/g_u)f\,\Gamma_{\text{cl}}$ . Watch which widths are homogeneous (every atom broadened identically — Lorentzian) versus inhomogeneous (a distribution over atoms — Gaussian): the distinction decides which experimental tricks can defeat which width.

Step 1. Doppler broadening. An atom moving with line-of-sight velocity $v_z$ emits at $\omega=\omega_0(1+v_z/c)$ . Projecting the Maxwellian, $P(v_z)\propto e^{-mv_z^2/2k_BT}$ , onto frequency gives a Gaussian line with

\Delta\omega_D=\omega_0\sqrt{\frac{8\ln 2\,k_BT}{mc^2}} \qquad\text{(FWHM)}.

where: the width is the thermal velocity in units of $c$ , times $\sqrt{8\ln2}$ . For Na D2 at 400 K: $\Delta\nu_D=1.52$ GHz — 155 times the natural width. Inhomogeneous: each velocity class is itself a narrow Lorentzian; the Gaussian is their census. Hence the cures — cool the gas (Doppler $\propto\sqrt T$ ), interrogate one velocity class (saturation spectroscopy), or trap below the recoil limit — none of which touch the natural floor.

Step 2. Collisional broadening. Model a collision as an event that scrambles the oscillator's phase, terminating the coherent train after a mean free time $\tau_{\text{col}}$ . A train truncated at random exponential times acquires — by the same Fourier logic as Derivation B — an additional Lorentzian width $\Gamma_{\text{col}}=2/\tau_{\text{col}}$ , adding to the natural width (Lorentzians convolve by adding widths). With $1/\tau_{\text{col}}=n\sigma\bar v_{\text{rel}}$ : for Na in argon at 400 K ( $\sigma\approx3\times10^{-18}\ \mathrm{m^2}$ , $\bar v_{\text{rel}}=762$ m/s), the collision rate matches the radiative rate at $n=2.7\times10^{22}\ \mathrm{m^{-3}}$ — about 150 Pa, one torr: a whiff of buffer gas doubles the linewidth. Homogeneous and pressure-proportional, hence "pressure broadening", the knob spectroscopists turn deliberately when they want wide lines.

Step 3. The Voigt verdict. Real profiles are the convolution Gaussian $\ast$ Lorentzian — the Voigt profile: Doppler rules the core (it wins at line center by two orders), the Lorentzian rules the wings (Step 2 of Derivation B: $\Delta\omega^{-2}$ beats any exponential). Fitting a Voigt profile separates the two widths from a single scan — the standard extraction of both temperature and pressure from one spectral line.

Step 4. The oscillator-strength dictionary. Quantum mechanics replaces "one electron on one spring" by a set of transitions sharing the electron's response; the share of line $u\to l$ is its oscillator strength $f$ , and the measured spontaneous rate is the classical rate rescaled by it and the degeneracy bookkeeping, $A_{ul}=3(g_l/g_u)f_{lu}\,\Gamma_{\text{cl}}$ . Audit on real data — Na D2 ( $f=0.641$ , $g_l/g_u=2/4$ ): predicted ratio $A/\Gamma_{\text{cl}}=3\times\tfrac12\times0.641=0.962$ , measured $6.16\times10^7/6.41\times10^7=0.961$ . Lyman-α ( $f=0.4162$ , $g_l/g_u=2/6$ ): predicted $0.416$ , measured $6.27\times10^8/1.504\times10^9=0.417$ . Two transitions, two decades of frequency, both within half a percent. The Thomas–Reiche–Kuhn sum rule $\sum f=1$ — quantum mechanics conserving the classical electron's total response — is why the classical model kept scoring near-hits for seventy years; its quantum derivation belongs to the QED bridge article. Verified: $\Delta\nu_D=1.521$ GHz from the Maxwellian formula with $m_{\mathrm{Na}}=23$ u at 400 K; the one-torr crossover from $n=A/\sigma\bar v_{\text{rel}}$ and $P=nk_BT=149$ Pa; both $f$ -audits recomputed from NIST values ( $A_{\mathrm{D2}}=6.16\times10^7$ , $A_{2p}=6.27\times10^8\ \mathrm{s^{-1}}$ ).

Limiting cases

case	result	meaning
Weak damping, $\Gamma\ll\omega_0$	Lorentzian FWHM $\Gamma$ , ring at $\omega_0-\Gamma^2/8\omega_0$	the atomic regime — $\Gamma/\omega_0\sim10^{-8}$ ; every result in this article
$\omega_0\to0$ (free electron)	$\Gamma\to0$ ; no line, no width	slow oscillators barely radiate; the free-particle limit hands the story back to the runaway article
Formal strong damping $\Gamma\gtrsim\omega_0$	perturbative insertion illegal	outside the validity domain; the honest theory (structure, then QED) must take over
Wavelength units	$\Delta\lambda=4\pi r_e/3$ always	universal: the $\omega_0^2$ in $\Gamma$ cancels the $\lambda^2$ Jacobian exactly
Quantum correspondence	$A=3(g_l/g_u)f\,\Gamma_{\text{cl}}$ , $\sum f=1$	classical width = the whole budget; each real line takes its $f$ -share
Cold, dilute, heavy	Doppler and collision widths $\to0$ ; natural width survives	the floor: homogeneous, temperature-independent, removable by nothing

Common trapThe natural width is just the uncertainty principle, so classical physics has no business predicting it

Backwards on both counts, and the history matters. The width–lifetime reciprocity $\Gamma\,t_{\text{life}}=1$ is a theorem of Fourier analysis, older than quantum mechanics and satisfied by any decaying oscillation — bells, RLC circuits, classical atoms. What quantum mechanics adds is not the reciprocity but the rate: it replaces the classical $\Gamma_{\text{cl}}=\omega_0^2\tau$ by $A=3(g_l/g_u)f\,\Gamma_{\text{cl}}$ , redistributing the classical budget among transitions while the sum rule protects the total. Conflating the two leads to a second, sneakier error: assigning the energy spectrum's width $\Gamma$ to the amplitude, whose decay rate is $\Gamma/2$ — the half-width/full-width factor of two that corrupts more lifetime extractions than any other single mistake. Keep the ledger straight: amplitude $e^{-\Gamma t/2}$ , energy $e^{-\Gamma t}$ , spectral FWHM $\Gamma$ , half-maximum at $\pm\Gamma/2$ .

Worked examples

1 · The sodium audit — one line, every number checked (confidence builder)

Setup. The Na D2 line: $\lambda=589.0$ nm, measured $A=6.16\times10^7\ \mathrm{s^{-1}}$ , $f=0.641$ , $g_l=2$ , $g_u=4$ . Compute the classical predictions and confront each.

Step 1. Classical rate. $\omega_0=2\pi c/\lambda=3.198\times10^{15}\ \mathrm{s^{-1}}$ , so $\Gamma_{\text{cl}}=\omega_0^2\tau=(3.198\times10^{15})^2(6.266\times10^{-24})=\mathbf{6.41\times10^{7}\ s^{-1}}$ — classical lifetime $1/\Gamma_{\text{cl}}=15.6$ ns.

Step 2. Dictionary check. Predicted ratio $A/\Gamma_{\text{cl}}=3(g_l/g_u)f=3\times0.5\times0.641=0.962$ . Measured: $6.16/6.41=\mathbf{0.961}$ . Agreement to $10^{-3}$ — the classical oscillator carries the D2 line almost single-handedly because $f$ is near its sum-rule maximum for this near-one-electron atom.

Step 3. The measurable widths. Natural FWHM $\Delta\nu=A/2\pi=\mathbf{9.80\ MHz}$ — the number laser spectroscopists quote for sodium. In wavelength: $\Delta\lambda=(\lambda^2/2\pi c)A=1.13\times10^{-4}$ Å — within $f\times$ bookkeeping of the universal $1.18\times10^{-4}$ Å, as the theorem demands.

Step 4. Lifetime cross-check. $1/A=16.2$ ns — the measured $3p_{3/2}$ lifetime. A tabletop absorption measurement of a linewidth is, by the reciprocity, a nanosecond-precision lifetime measurement with no fast electronics anywhere.

The lesson: the chain rate → lifetime → width → wavelength closes on real data at the percent level once $f$ is paid; the classical machinery is not an analogy but the load-bearing frame.

2 · Who broadens what — the widths ladder for a vapor cell and a trap (exam level)

Setup. The same D2 line, three environments: an oven-heated cell (400 K), the same cell with 1 torr of argon, and a laser-cooled cloud at the Doppler limit. Rank every width.

Step 1. Hot cell. Doppler: $\Delta\nu_D=(\omega_0/2\pi)\sqrt{8\ln2\,k_BT/mc^2}=\mathbf{1.52\ GHz}$ at 400 K — 155 natural widths. The Voigt profile is Gaussian-cored; the natural Lorentzian survives only in the wings beyond $\sim2$ GHz from center.

Step 2. Add 1 torr of argon. Collision rate $n\sigma\bar v_{\text{rel}}$ with $n=P/k_BT=2.4\times10^{22}\ \mathrm{m^{-3}}$ , $\sigma=3\times10^{-18}\ \mathrm{m^2}$ , $\bar v_{\text{rel}}=762$ m/s gives $\approx5.5\times10^{7}\ \mathrm{s^{-1}}$ — a collisional Lorentzian comparable to the natural one (the 149 Pa crossover of Derivation C), invisible under the Doppler core but doubling the far wings.

Step 3. Doppler-limited cloud. Laser cooling on this line bottoms out at $T_D=\hbar\Gamma_{\text{meas}}/2k_B$ — for Na, $\hbar A/2k_B=235\ \mu$ K; the sibling workhorse $^{87}$ Rb D2 ( $\Gamma/2\pi=6.07$ MHz) gives the textbook $\mathbf{146\ \mu K}$ . At 235 μK the residual Doppler width is $\Delta\nu_D=1.2$ MHz — below the 9.8 MHz natural width at last: cooling has pushed the inhomogeneous census beneath the homogeneous floor, and the natural Lorentzian finally is the observed line.

Step 4. The floor below the floor. For $^{57}$ Fe's 14.4 keV Mössbauer transition ( $t_{\text{mean}}=141$ ns): $\Gamma=\hbar/t=\mathbf{4.7\ neV}$ , a fractional width $\Gamma/E=3.2\times10^{-13}$ — recoil-free emission in a crystal defeats Doppler broadening entirely, delivering the natural width at a $Q$ of $3\times10^{12}$ and with it gravitational-redshift metrology (Pound–Rebka) on a staircase.

The lesson: the natural width is never the biggest number in the room until you engineer the room — cold, dilute, recoil-free — and then it is the only number left: the irreducible classical floor, wearing its quantum $f$ .

Going further

Provenance. Equations and numbers are standard textbook results. A verifier should recompute: (1) $\tau_0=e^2/6\pi\varepsilon_0 m c^3\approx6.3\times10^{-24}\,\mathrm{s}$ from constants; (2) $\gamma=\tau_0\omega_0^2$ as the ratio of Larmor power to oscillator energy (Jackson §16.2; Griffiths §11.2.2); (3) the Lorentzian as the squared one-sided Fourier transform of $e^{-\gamma t/2}\cos\omega_1 t$ ; (4) the wavelength width $\Delta\lambda=\tfrac{4\pi}{3}r_e\approx1.18\times10^{-4}\,\mathrm{\AA}$ with $r_e=2.818\times10^{-15}\,\mathrm{m}$ (Jackson §16.3, Zangwill §23.6). The hydrogen Lamb-shift figure $\approx1057\,\mathrm{MHz}$ is the standard experimental value cited only as context, not derived here.

Beyond this article. Jackson §16.4–16.5 carries the reactive-shift analysis with an explicit cutoff; Weisskopf & Wigner (Z. Phys. 63, 54 (1930)) is the quantum lineshape; Demtröder's Laser Spectroscopy owns the Voigt-profile craft; Bethe's two-page 1947 Phys. Rev. letter remains the fastest deep read in the field.

Builds on — Larmor Formula (the spectral ledger spent here) · Electric Dipole Radiation · Liénard-Wiechert Potentials · Runaway Solutions and Preacceleration (the legality of the perturbative insertion). Leads to — Bridge from Classical Electrodynamics to QED (oscillator strengths, the sum rule, and the Lamb shift done right) · Synchrotron Radiation · Relativistic Generalization of Larmor's Formula.

If the natural width is homogeneous, what does that actually buy me in the lab?

It means every atom carries the identical Lorentzian, so the width cannot be beaten by selecting a subpopulation — but it can be evaded in time. Homogeneous broadening permits coherent tricks: photon echoes and Ramsey interrogation work because all atoms dephase identically and can be rephased identically, which is impossible against an inhomogeneous (Doppler) spread without echo techniques. Practically, the taxonomy tells you the attack order: Doppler is inhomogeneous, so velocity-selective methods (saturation spectroscopy, two-photon schemes) or cooling remove it; collisions are homogeneous but pressure-proportional, so pump the cell down; what remains is the natural Lorentzian, and your only leverage on it is choosing a different transition — which is exactly why clock transitions are picked for tiny $f$ and millihertz widths.

Why does the classical model get the D2 linewidth right to a tenth of a percent when atoms are not electrons on springs?

Because the quantity being predicted is protected by a sum rule. Quantum mechanics replaces the single classical oscillator by many transitions, each taking a fraction $f$ of the classical response, with $\sum f=1$ (Thomas–Reiche–Kuhn). For an alkali D line, one valence electron with one dominant transition hogs nearly the whole budget ( $f_{D1}+f_{D2}\approx0.97$ for sodium), so the measured rate lands within bookkeeping factors of the classical rate. The agreement is not luck but conservation: quantization redistributes the classical electron's radiative strength without creating or destroying it. Where many lines share the budget — molecular bands, highly excited states — individual lines are much narrower than $\Gamma_{\text{cl}}$ , exactly as the dictionary predicts.

Derivation B dropped the mirror pole at −ω₁. When would that rotating-wave step actually fail?

When the width stops being tiny compared to the frequency. The discard costs relative corrections of order $\Gamma/\omega_1$ — $10^{-8}$ for optical lines, utterly safe. But push toward systems where damping is a visible fraction of the oscillation (a two-cycle attosecond pulse, a heavily loaded microwave cavity, an over-coupled plasmonic resonator with $Q\sim5$ ) and the two poles overlap: the spectrum develops asymmetry, the peak shifts beyond the $\Gamma^2/8\omega_0$ pull, and the very notion of "the line's frequency" acquires convention-dependence. Spectroscopists meet this as the difference between fitting a Lorentzian and fitting the full two-pole (or "complex Lorentzian") response. The atomic case never gets there; the lesson is to know which approximation was purchased and at what exchange rate.

Is the 26 mHz damping pull the classical Lamb shift?

No — and keeping them separate is the article's sharpest distinction. The pull $\Gamma^2/8\omega_0$ is finite, cutoff-independent, and utterly generic: any damped oscillator rings slightly flat, bells included. The Lamb shift descends from a different term entirely — the reactive part of the self-force, the in-phase self-field that renormalizes the mass and whose leftover shift depends on the cutoff you assign the electron's structure. Classically that number is undefined; quantum mechanically the electron's own physics (its Compton-scale structure, the vacuum's modes) supplies the cutoff, and Bethe's calculation turns the divergence into 1040 MHz of the measured 1057. So: pull = trivial and computable; radiative shift = profound and classically incomputable. The classical theory's honest contribution was knowing the difference.

Why do Lorentzians add widths under convolution while Gaussians add variances?

Because convolution multiplies Fourier transforms, and the two shapes have transforms with opposite composition laws. A Lorentzian of HWHM $a$ transforms to $e^{-a|t|}$ ; multiplying two gives $e^{-(a_1+a_2)|t|}$ — widths add. A Gaussian of width $\sigma$ transforms to $e^{-\sigma^2t^2/2}$ ; multiplying gives $\sigma_1^2+\sigma_2^2$ — variances add. Physically: Lorentzian broadening mechanisms are memoryless interruption rates (radiative decay, collisions), and independent Poisson rates add; Gaussian mechanisms are independent random offsets (velocity classes), and independent variances add. This is why Derivation C could simply add $\Gamma_{\text{col}}$ to $\Gamma$ but had to convolve the Doppler Gaussian — and why the Voigt profile has no elementary closed form.

Does a single atom emitting a single photon still have a linewidth?

Yes — and this is where the classical picture transitions most gracefully. Quantum mechanically the emitted one-photon state is a superposition over frequencies whose amplitude is exactly the Fourier transform of the decaying dipole: measure many single photons from identically prepared atoms and the frequency histogram fills in the Lorentzian of width $A$ . Nothing about the width requires an ensemble emitting simultaneously; the "decaying wavetrain" of Derivation B becomes the single photon's wavepacket envelope. Modern quantum-optics experiments do precisely this — heralded single photons from trapped ions show Lorentzian spectra of the natural width — and the time–frequency reciprocity, born classical in this article, survives as a statement about one quantum at a time.

Where does the energy of the frequency shift come from — does the atom radiate at ω₁ or ω₀?

The atom radiates at what it actually does, and what it does is ring at $\omega_1$ — the damped eigenfrequency. There is no bookkeeping paradox: $\omega_0$ is a parameter of the potential (the spring constant), not an energy the atom possesses, and the emitted spectrum peaks at the true ringing frequency with the Lorentzian spread around it. The subtlety worth keeping is that "the transition frequency" is operationally defined by where the line sits, so all shifts — the damping pull, the Lamb shift, Doppler and pressure shifts (the displacement cousins of the broadenings in Derivation C) — are folded into the measured $\omega_1$ before any comparison with theory. Precision spectroscopy is largely the art of unfolding them: modelling each shift's dependence on temperature, pressure, and probe intensity and extrapolating all of them away.