Homogeneous versus Inhomogeneous Broadening

Every spectral line this article takes apart arrives already shaped by results owned earlier in the chapter: the Lorentzian natural line shape that a finite radiative lifetime stamps on a single atom, and the resonant absorption and stimulated emission through which that atom couples to a field. Take those single-atom facts as given and ask a sharper question — what happens when you no longer watch one atom but a whole ensemble? Two conceptually different things can widen the observed line, and telling them apart is one of the most useful distinctions in spectroscopy. In homogeneous broadening every atom responds with the same resonant frequency and the same line shape; the width is intrinsic to the single-atom dynamics, and no trick of atom-selection can evade it. In inhomogeneous broadening each atom is individually sharp but sits in a slightly different environment — a different velocity, local field, or lattice strain — so their resonances spread across a band, and the broad line is a bookkeeping of many narrow ones. The distinction is not about the shape as such but about whether sub-ensembles of atoms can be spectrally distinguished. That single yes-or-no decides whether a strong laser bleaches the transition as a whole or burns a narrow hole into it, and whether hyperfine structure buried under a gigahertz-wide Doppler envelope stays hidden or can be recovered at the megahertz level. When both act, the observed line is the convolution of the two — a Voigt profile.

Key takeaways

Homogeneous broadening affects every atom identically: one line shape describes each emitter, so the width cannot be evaded by atom-selection. Its archetype is the Lorentzian natural and collisional width.
Inhomogeneous broadening comes from a static distribution of atomic environments or velocities: each atom is narrow, but the ensemble spread makes a broad envelope — Gaussian for Doppler broadening.
The operational test is saturation: a homogeneous line bleaches and power-broadens as a whole, while an inhomogeneous line permits spectral hole burning — a narrow dip whose width reveals the hidden homogeneous width.
Natural (lifetime) and collisional (pressure) broadening are homogeneous; Doppler and crystal-strain broadening are inhomogeneous.
The observed line is the convolution of the two — a Voigt profile — whose FWHM interpolates smoothly between the Lorentzian and Gaussian limits.
Key idea: homogeneous broadening gives every atom the same line shape and cannot be evaded by selecting atoms; inhomogeneous broadening comes from a distribution of atomic environments, so a narrow probe can address a single sub-ensemble (hole burning). The observed line is their convolution — a Voigt profile.

Where this lives today

The distinction is quietly load-bearing in precision optics. Doppler-free saturated-absorption spectroscopy — Lamb's dip taken into an external cell — is how a helium-neon laser is stabilised to a hyperfine component of molecular iodine at $633\ \mathrm{nm}$ , one of the recommended radiations that realised the metre; it works only because a low-pressure gas line is inhomogeneous, so counter-propagating pump and probe single out the zero-velocity atoms and expose the near-homogeneous width. The same physics runs in reverse in the most stable lasers ever built: rare-earth ions such as $\mathrm{Eu^{3+}}$ in cryogenic $\mathrm{Y_2SiO_5}$ have a gigahertz-wide inhomogeneous line from lattice strain but kilohertz-narrow homogeneous lines, and a hole burnt persistently into that band gives a reference against which sub-hertz laser linewidths are locked. That enormous ratio is also a resource: in rare-earth quantum memories and frequency-multiplexed optical storage, thousands of channels are addressed inside one inhomogeneous band by burning holes at different frequencies. And in plasma and stellar diagnostics the inverse problem is routine — a Voigt line is split into its Gaussian part, giving thermal or turbulent velocity through $\Delta\nu_D\propto\sqrt{T/M}$ , and its Lorentzian part, giving perturber density via pressure broadening.

The physical picture

Picture a choir. Homogeneous broadening is every singer holding the same note with an identical tremolo — whichever singer you lean toward, you hear the same wobble, and turning up one voice just makes the same broad note louder. Inhomogeneous broadening is a choir in which each singer holds a steady but slightly different pitch: the ensemble sounds broad, yet you could pick out one singer at one pitch, and silencing exactly those singers opens a notch — hole burning. Can you address a sub-group, or does everyone move together? That is the whole subject.

Figure

Homogeneous versus inhomogeneous broadening. On the left every atom carries the same Lorentzian centred on ν₀ so the transition saturates as one block. On the right the atoms are spread over a band of resonant frequencies ν_a and their narrow lines sum to a broad Gaussian envelope that a probe can address one class at a time

The Doppler effect is the cleanest inhomogeneous mechanism. An atom moving with line-of-sight velocity $v_z$ is shifted by $\nu_0 v_z/c$ ; the Maxwell–Boltzmann distribution of $v_z$ is Gaussian, so the frequency distribution is Gaussian, while each atom still radiates a narrow natural Lorentzian in its own frame — the two layer into the Voigt line. Collisions, by contrast, strike all atoms with the same statistics and add a homogeneous Lorentzian width. How a line responds to temperature (Doppler grows as $\sqrt{T/M}$ ), pressure (collisional grows with density), or a heavier isotope (Doppler narrows) often names the dominant family before any saturation experiment. Saturation is the sharpest discriminator — the second figure is what a saturation-spectroscopy lab produces: a broad Doppler envelope with a narrow hole burnt through its peak.

Figure

Spectral hole burning in the Doppler-broadened sodium D line at T = 500 K. A strong monochromatic pump bleaches only the resonant velocity class, leaving a narrow hole of the homogeneous width Δν_nat ≈ 10 MHz punched into the broad envelope Δν_D ≈ 1.70 GHz — about ×170 narrower, drawn exaggerated so it is visible

The core equations

Start with the homogeneous archetype. A single atom radiates for a finite time set by its lifetime $\tau$ , and a wavetrain truncated in time is a Lorentzian in frequency whose width is fixed by the transition.

\Delta\nu_{\mathrm{nat}}=\frac{1}{2\pi\tau}=\frac{A_{21}}{2\pi} \qquad (1)

where: $\tau$ — radiative lifetime of the upper level; $A_{21}$ — Einstein A coefficient (spontaneous-emission rate); $\Delta\nu_{\mathrm{nat}}$ — FWHM of the natural Lorentzian. Because $\tau$ and $A_{21}$ belong to the transition, this width is identical for every atom — the signature of a homogeneous mechanism. Collisions add a second homogeneous width: each phase-interrupting collision cuts the coherent wavetrain short, exactly as decay does, so the contribution is again Lorentzian and grows with the collision rate.

\Delta\nu_{\mathrm{coll}}=\frac{1}{\pi\,\tau_{\mathrm{coll}}}=\frac{n\,\sigma\,\bar{v}_{\mathrm{rel}}}{\pi} \qquad (2)

where: $\tau_{\mathrm{coll}}$ — mean time between phase-interrupting collisions; $n$ — perturber number density; $\sigma$ — collision cross-section; $\bar{v}_{\mathrm{rel}}$ — mean relative speed. Both homogeneous widths are Lorentzian, so they add directly: $\Delta\nu_H=\Delta\nu_{\mathrm{nat}}+\Delta\nu_{\mathrm{coll}}$ . Now the inhomogeneous archetype: thermal atoms carry a Maxwell–Boltzmann spread of line-of-sight velocities, and the linear Doppler shift maps that onto a Gaussian frequency distribution belonging to the ensemble, not to any one atom.

\Delta\nu_D=\nu_0\sqrt{\frac{8\ln 2\,k_B T}{M c^2}} \qquad (3)

where: $\nu_0$ — line-centre frequency; $k_B$ — Boltzmann constant; $T$ — absolute temperature; $M$ — atomic mass; $c$ — speed of light. The $\sqrt{T/M}$ scaling is the fingerprint of Doppler broadening — heat the gas and the line widens, use a heavier isotope and it narrows — behaviour no single-atom width can imitate. When both act, each atom's Lorentzian sits on its own shifted frequency, and the observed line is the sum over the Gaussian distribution of centres — a convolution.

g_V(\nu)=\int_{-\infty}^{\infty} g_L(\nu-\nu')\,g_G(\nu')\,\mathrm{d}\nu' \qquad (4)

where: $g_L$ — homogeneous Lorentzian component; $g_G$ — inhomogeneous Gaussian component; $g_V$ — observed Voigt profile. It has no closed elementary form, but its FWHM is captured to better than $0.02\%$ by the Olivero–Longbothum combination. The two widths do not add linearly, because a Lorentzian and a Gaussian convolve differently; the empirical FWHM interpolates smoothly between the pure limits.

\Delta\nu_V\approx 0.5346\,\Delta\nu_H+\sqrt{0.2166\,\Delta\nu_H^{2}+\Delta\nu_D^{2}} \qquad (5)

where: $\Delta\nu_H$ — total homogeneous (Lorentzian) FWHM; $\Delta\nu_D$ — inhomogeneous (Gaussian) FWHM; $\Delta\nu_V$ — Voigt FWHM. It returns $\Delta\nu_H$ when $\Delta\nu_D\to0$ and $\Delta\nu_D$ when $\Delta\nu_H\to0$ . The discriminator between the families is how the line responds to an intense monochromatic field. A homogeneous line has one resonance, so a strong beam depletes the whole ensemble at once, power-broadening it rather than notching it.

S=\frac{I/I_s}{1+I/I_s},\qquad \Delta\nu_{\mathrm{sat}}=\Delta\nu_H\sqrt{1+I/I_s} \qquad (6)

where: $I$ — probe intensity; $I_s$ — saturation intensity; $\Delta\nu_H$ — unsaturated homogeneous width; $S$ — fractional saturation. Because every atom feels the same field and shares the same resonance, no frequency-selective dip appears — the line just dims and broadens as a whole. In an inhomogeneous line the same beam saturates only the class in resonance with it, punching a hole whose width reports the homogeneous width of the addressed sub-ensemble; and because the integrated line strength is fixed, spreading a line wider lowers its peak.

\Delta\nu_{\mathrm{hole}}=2\,\Delta\nu_H\sqrt{1+I/I_s},\qquad \sigma_0=\frac{\lambda^{2}}{4\pi^{2}}\frac{A_{21}}{\Delta\nu} \qquad (7)

where: $\Delta\nu_{\mathrm{hole}}$ — FWHM of the saturation hole; $\sigma_0$ — peak absorption cross-section; $\lambda$ — transition wavelength; $\Delta\nu$ — total FWHM. The hole's narrowness ( $\sim\Delta\nu_H\ll\Delta\nu_D$ ) is what makes Doppler-free spectroscopy possible; the $\sigma_0\propto1/\Delta\nu$ scaling is why inhomogeneous broadening weakens peak absorption without touching total line strength.

Derivation ADoppler line shape from the velocity distribution

The goal is to build the canonical inhomogeneous profile from the ground up: treat each atom as radiating an infinitely sharp line in its own rest frame, so the whole observed width comes from the spread of Doppler shifts across the Maxwell–Boltzmann velocity distribution, then convert the Gaussian standard deviation into the FWHM that experiments quote. This isolates the inhomogeneous mechanism with the single-atom width switched off, and shows where the temperature and mass dependence of $\Delta\nu_D$ come from.

Step 1. An atom with line-of-sight velocity $v_z$ absorbs at $\nu=\nu_0(1+v_z/c)$ , which inverts to $v_z=c(\nu-\nu_0)/\nu_0$ . Why: this linear Doppler map ties a mechanical quantity, velocity, to the spectroscopic quantity, frequency detuning.

Step 2. The velocity component along the observation axis follows $f(v_z)\propto\exp(-Mv_z^{2}/2k_BT)$ , a Gaussian of variance $k_BT/M$ . Why: only the line-of-sight component shifts the frequency, so the relevant distribution is the 1D projection, Gaussian in equilibrium.

Step 3. Change variables with the Jacobian $\mathrm{d}v_z/\mathrm{d}\nu=c/\nu_0$ to get $g(\nu)\propto\exp[-Mc^{2}(\nu-\nu_0)^{2}/2k_BT\nu_0^{2}]$ , a Gaussian of variance $\sigma_\nu^{2}=k_BT\nu_0^{2}/(Mc^{2})$ . Why: a linear map sends a Gaussian in $v_z$ to a Gaussian in $\nu$ , rescaling only the width.

Step 4. Convert to FWHM with $\Delta\nu_D=2\sqrt{2\ln 2}\,\sigma_\nu=\nu_0\sqrt{8\ln 2\,k_BT/Mc^{2}}$ . Why: a Gaussian's half-maximum points sit at $\pm\sqrt{2\ln 2}\,\sigma_\nu$ , so the FWHM carries the universal factor $2\sqrt{2\ln 2}\approx2.355$ . For the sodium D line ( $\lambda=589\ \mathrm{nm}$ , $\nu_0=5.09\times10^{14}\ \mathrm{Hz}$ , $M\approx23\,$ u) at $T=500\ \mathrm{K}$ this gives $\Delta\nu_D\approx1.70\ \mathrm{GHz}$ , against a natural width of only $\sim10\ \mathrm{MHz}$ — Doppler broadening wins by two orders of magnitude, the textbook inhomogeneous regime.

Go deeper · Derivation B · 1 minSpectral hole burning distinguishes the two regimes

Derivation BSpectral hole burning distinguishes the two regimes

The purpose is to make the saturation test rigorous: to show step by step why a strong pump bleaches a homogeneous line uniformly but carves a narrow, frequency-selective hole into an inhomogeneous one, and to pin down the width of that hole. Consider a two-level medium driven by a strong pump at frequency $\nu_p$ while a weak beam scans the transition to read the absorption, and follow which atoms have their populations disturbed in each case.

Step 1. Homogeneous case: every atom shares resonance $\nu_0$ and one Lorentzian response, so the pump reduces the whole ensemble's population difference by $1/(1+I/I_s)$ ; absorption drops everywhere and the line power-broadens to $\Delta\nu_H\sqrt{1+I/I_s}$ , with no dip. Why: with one shared resonance, saturation is a global operation — there is no sub-group to leave behind.

Step 2. Inhomogeneous case: atoms spread over resonances $\nu_a$ , and the pump at $\nu_p$ saturates only those whose homogeneous Lorentzian overlaps $\nu_p$ — a class of width $\sim\Delta\nu_H$ around $\nu_p$ . Why: an atom detuned by much more than its homogeneous width barely feels the field, so only a thin slice is disturbed.

Step 3. That slice obeys $N_1-N_2\to(N_1-N_2)/[1+(I/I_s)\mathcal{L}(\nu_a-\nu_p)]$ ; atoms far from $\nu_p$ are untouched. Why: the saturation factor is weighted by each atom's own line shape, so it acts only where $\nu_a\approx\nu_p$ .

Step 4. The probe therefore sees reduced absorption over a band set by the pump–probe homogeneous widths — a hole of FWHM $\Delta\nu_{\mathrm{hole}}\approx2\,\Delta\nu_H\sqrt{1+I/I_s}$ inside the broad Gaussian. Why: the hole width tracks the homogeneous width, not the envelope, so it can be orders of magnitude narrower than the observed line.

Step 5. Because the hole reports $\Delta\nu_H$ directly, hole burning and the Lamb-dip technique recover Doppler-free resolution. Why: in sodium a $\sim1.70\ \mathrm{GHz}$ envelope yields a hole revealing the $\sim10\ \mathrm{MHz}$ natural width — a resolution gain near $\times170$ .

Limiting cases

The observed line interpolates between two extremes depending on which family of widths dominates:

Case	Result	Meaning
Pure homogeneous (Lorentzian)	$g(\nu)=\dfrac{\Delta\nu_H/2\pi}{(\nu-\nu_0)^{2}+(\Delta\nu_H/2)^{2}}$	When the homogeneous width (natural + collisional) far exceeds the inhomogeneous spread, the Voigt profile collapses to a Lorentzian; all atoms saturate together, no hole burning.
Pure inhomogeneous (Gaussian)	$g(\nu)=\dfrac{2}{\Delta\nu_D}\sqrt{\dfrac{\ln 2}{\pi}}\,\exp\!\left[-4\ln 2\,\dfrac{(\nu-\nu_0)^{2}}{\Delta\nu_D^{2}}\right]$	When Doppler or strain width dominates, the profile is Gaussian; distinct classes exist and a narrow probe burns a hole of the residual homogeneous width.
Comparable widths (Voigt)	$\Delta\nu_V\approx 0.5346\,\Delta\nu_H+\sqrt{0.2166\,\Delta\nu_H^{2}+\Delta\nu_D^{2}}$	General case: the Lorentzian–Gaussian convolution, with FWHM given by the Olivero–Longbothum approximation.
Strong saturation, $I\gg I_s$	$\Delta\nu_{\mathrm{sat}}\to\Delta\nu_H\sqrt{I/I_s}$	Power broadening dominates; the burnt hole widens the same way, degrading resolution at high pump power.

Common trapLorentzian means homogeneous and Gaussian means inhomogeneous, taken as a definition

Do not treat the shapes as the definition. The Lorentzian and Gaussian shapes are consequences of the two mechanisms in their idealised limits, not the criterion that distinguishes them. The real test is physical: do all atoms share one line shape (homogeneous), or do distinct atoms occupy distinct frequencies so that a narrow probe can address one sub-ensemble at a time (inhomogeneous, revealed by hole burning)? A Gaussian envelope can in principle be assembled from homogeneous contributions, and a genuinely inhomogeneous distribution need not be exactly Gaussian. Anchor the decision on the distinguishability of sub-ensembles — the saturation and hole-burning behaviour — not on curve-fitting the line shape.

Worked examples

1 · Doppler versus natural width of the sodium D line (the flagship number)

Sodium vapour at $T=500\ \mathrm{K}$ radiates the D line at $\lambda=589\ \mathrm{nm}$ ; the upper $3p$ level has lifetime $\tau=16.2\ \mathrm{ns}$ and mass $M\approx23\,$ u. Find the Doppler and natural widths and their ratio.

Step 1. Line-centre frequency: $\nu_0=c/\lambda=(2.998\times10^{8})/(589\times10^{-9})=\mathbf{5.09\times10^{14}\ Hz}$ .

Step 2. Doppler width from (3): with $8\ln2=5.545$ , $k_BT=6.90\times10^{-21}\ \mathrm{J}$ , and $Mc^{2}=(23\times1.66\times10^{-27})(2.998\times10^{8})^{2}=3.44\times10^{-9}\ \mathrm{J}$ , the radicand is $5.545\times6.90\times10^{-21}/3.44\times10^{-9}=1.11\times10^{-11}$ , whose root is $3.34\times10^{-6}$ . Then $\Delta\nu_D=(5.09\times10^{14})(3.34\times10^{-6})=\mathbf{1.70\times10^{9}\ Hz}=1.70\ \mathrm{GHz}$ .

Step 3. Natural width from (1): $\Delta\nu_{\mathrm{nat}}=1/(2\pi\times16.2\times10^{-9})=\mathbf{9.8\times10^{6}\ Hz}\approx10\ \mathrm{MHz}$ . The ratio $\Delta\nu_D/\Delta\nu_{\mathrm{nat}}\approx170$ : the inhomogeneous width buries the homogeneous one by two orders of magnitude, so a hole burnt in this line can be $\sim170$ times narrower than the visible envelope. Δν_D ≈ 1.70 GHz, Δν_nat ≈ 10 MHz, ratio ≈ 170 — strongly inhomogeneous.

2 · The pressure at which collisions match the natural width

For the same line, estimate the perturber pressure at which the collisional (homogeneous) width equals the $\sim10\ \mathrm{MHz}$ natural width. Take $\sigma\approx4\times10^{-19}\ \mathrm{m^{2}}$ and $\bar v_{\mathrm{rel}}\approx900\ \mathrm{m\,s^{-1}}$ at $500\ \mathrm{K}$ .

Step 1. Set $\Delta\nu_{\mathrm{coll}}=n\sigma\bar v_{\mathrm{rel}}/\pi=10^{7}\ \mathrm{Hz}$ and solve: $n=\pi\times10^{7}/(4\times10^{-19}\times900)=3.14\times10^{7}/3.6\times10^{-16}=\mathbf{8.7\times10^{22}\ m^{-3}}$ .

Step 2. Convert with the ideal-gas law: $P=nk_BT=(8.7\times10^{22})(1.38\times10^{-23})(500)=\mathbf{6.0\times10^{2}\ Pa}\approx6\ \mathrm{mbar}$ .

Step 3. Below a few millibar the homogeneous width is essentially the natural lifetime; above it, collisions grow the Lorentzian linearly with pressure, eating into the resolution a burnt hole can deliver — which is why Doppler-free cells run at low pressure. Collisions rival the natural width near P ≈ 600 Pa (a few mbar); with order-of-magnitude σ and speed, this is an estimate.

3 · Combining a Lorentzian and a Gaussian into a Voigt width

A line has homogeneous $\Delta\nu_H=200\ \mathrm{MHz}$ (Lorentzian) and inhomogeneous $\Delta\nu_D=800\ \mathrm{MHz}$ (Gaussian). Estimate the observed FWHM.

Step 1. Use Olivero–Longbothum, (5). Linear term: $0.5346\times200=106.9\ \mathrm{MHz}$ .

Step 2. Radical: $0.2166\times200^{2}=8664$ , plus $800^{2}=640\,000$ , gives $648\,664$ ; its root is $805.4\ \mathrm{MHz}$ .

Step 3. Add: $\Delta\nu_V\approx106.9+805.4=\mathbf{912\ MHz}$ . The Gaussian dominates, and the total exceeds the larger single width — as convolution must. The widths add neither in quadrature ( $824\ \mathrm{MHz}$ ) nor linearly ( $1000\ \mathrm{MHz}$ ); the Voigt value sits between. Δν_V ≈ 912 MHz.

Physics of the portrait

The distinction became a working tool the moment someone built a light source narrow enough to burn a hole in a Doppler line — first inside the gas laser, in the theory of Willis Lamb.

Origins · 1 minWillis Lamb and how this was discovered

Willis Lamb

1913–2008 · saturation spectroscopy and the line-centre Lamb dip

Willis Eugene Lamb Jr. is best remembered for the measurement that forced quantum electrodynamics into being: in 1947, with Robert Retherford, he resolved the tiny splitting between the $2s_{1/2}$ and $2p_{1/2}$ levels of hydrogen — the Lamb shift — that Dirac theory said should not exist, work honoured with the 1955 Nobel Prize in Physics. His second act is the one that owns this article. When the gas laser arrived, Lamb produced in 1964 a complete semiclassical theory of the optical maser, and inside it he predicted a feature that only a Doppler-broadened — that is, inhomogeneously broadened — gain medium can show. Tune a single-mode laser toward line centre and its two counter-propagating standing-wave components, which normally saturate two velocity classes at $\pm v_z$ , converge on the same zero-velocity atoms; those atoms are saturated twice as hard, the gain sags, and the output power dips at line centre. This Lamb dip is a spectral hole the laser burns on itself, its narrow width reporting the homogeneous linewidth hidden beneath the Doppler envelope — the direct ancestor of saturated-absorption Doppler-free spectroscopy and of every frequency standard locked to a sub-Doppler feature.

History

1842

— Doppler predicts the frequency shift of waves from a moving source

1912

— Voigt convolves Gaussian and Lorentzian shapes into the profile that bears his name

1962

— Bennett identifies spectral hole burning in the gain of the helium-neon laser

1964

— Lamb's semiclassical laser theory predicts the line-centre dip that bears his name

1971

— Hänsch and Schawlow demonstrate Doppler-free saturated-absorption spectroscopy

Going further

Texts. Foot, Atomic Physics, ch. 8 (Doppler and natural broadening; saturation spectroscopy) · Loudon, The Quantum Theory of Light, 3rd ed., ch. 2 and 8 (line shapes, homogeneous versus inhomogeneous) · Demtröder, Laser Spectroscopy, vol. 1, ch. 3 (widths and profiles; the Voigt profile) · Saleh and Teich, Fundamentals of Photonics, ch. 13 (laser amplifiers; broadening and saturation) · Hecht, Optics, 5th ed. (line-broadening overview). Key numerical checks: (1) $\Delta\nu_{\mathrm{nat}}=1/(2\pi\tau)$ for Na with $\tau\approx16.2\ \mathrm{ns}$ gives $9.8\ \mathrm{MHz}$ ; (2) $\Delta\nu_D$ for Na at $500\ \mathrm{K}$ gives $1.70\ \mathrm{GHz}$ , ratio $\approx170$ ; (3) the collisional-crossover pressure $\approx600\ \mathrm{Pa}$ follows from $n\sigma\bar v_{\mathrm{rel}}/\pi=10^{7}\ \mathrm{Hz}$ ; (4) the Olivero–Longbothum FWHM for $\Delta\nu_H=200\ \mathrm{MHz}$ and $\Delta\nu_D=800\ \mathrm{MHz}$ gives $912\ \mathrm{MHz}$ .

Prerequisites. Lorentzian Line Shape · Radiative Lifetime and Natural Linewidth · Resonant Absorption · Stimulated Emission. Leads to. Doppler-free Laser Spectroscopy · The Voigt Profile · Laser Gain Saturation and Mode Competition.

Ask the tutor: Ask me to walk through why a counter-propagating pump and probe carve a Lamb dip out of a Doppler line, and how narrow it can get.

Is it correct to just call Lorentzian lines homogeneous and Gaussian lines inhomogeneous?

It is a useful rule of thumb, not the definition, and leaning on it will mislead you. The Lorentzian follows from a single-atom coherence that decays exponentially (radiative or collisional), the Gaussian from a static distribution of centre frequencies, so in the clean limits the association holds. But the shape is a consequence, not the criterion. The real definition is whether all atoms share one response (homogeneous) or occupy distinct frequencies so a narrow probe can address one sub-group (inhomogeneous). Trust the saturation and hole-burning behaviour over the curve fit — a non-Gaussian inhomogeneous distribution or a composite homogeneous line breaks the shape rule.

Why is Doppler broadening called inhomogeneous when every sodium atom is identical?

Because "inhomogeneous" refers to the atoms' environments, not the atoms themselves. Every sodium atom has the same intrinsic transition and the same narrow natural line in its own rest frame. What differs is velocity: an atom moving toward you is blue-shifted, one moving away red-shifted, and the thermal spread of line-of-sight velocities smears their identical rest-frame lines across a gigahertz-wide band. The ensemble is inhomogeneous because different atoms sit at different observed frequencies, so you can pick out the ones at zero velocity with a resonant laser. Identical atoms in different motional states behave, spectroscopically, like a distribution of different emitters.

What is the difference between power broadening and hole burning?

Both happen when you drive a transition hard, but they are opposite symptoms. Power broadening is what a strong field does to a homogeneous line: every atom shares one resonance, so saturation flattens the peak and the whole line widens as $\Delta\nu_H\sqrt{1+I/I_s}$ , with no frequency-selective feature. Hole burning is what it does to an inhomogeneous line: the pump saturates only the resonant sub-ensemble, so a narrow dip of roughly the homogeneous width appears inside the broad envelope. The first says all atoms move together; the second says sub-groups are distinguishable. In practice the burnt hole itself power-broadens as $2\Delta\nu_H\sqrt{1+I/I_s}$ .

Can a single line be homogeneously and inhomogeneously broadened at once?

Almost always, yes — that is the normal situation, and the observed line is their convolution, a Voigt profile. Every atom carries a homogeneous Lorentzian (natural plus collisional) centred on its own frequency, and those frequencies are spread by an inhomogeneous distribution (Doppler or strain). Folding the Lorentzian over the Gaussian gives the Voigt line, which reduces to a pure Lorentzian or Gaussian when one width dwarfs the other. Fitting a measured line to a Voigt profile and reading off the two component widths is how spectroscopists separate temperature and turbulence (Gaussian part) from pressure and lifetime (Lorentzian part).

How does burning a hole let you beat the Doppler limit, and what sets the hole's width?

The Doppler envelope hides the interesting structure — hyperfine splittings, the true natural width — because it is hundreds of times wider than any single atom's line. A saturating pump talks only to atoms in resonance with it, a velocity class about one homogeneous width wide, so the hole it burns has a width set by $\Delta\nu_H$ , not the Doppler width. Reading it with a weak probe recovers resolution at the homogeneous level, roughly $\times170$ better in sodium. The hole is not infinitely narrow: it broadens as $2\Delta\nu_H\sqrt{1+I/I_s}$ , so precision work uses the lowest usable pump intensity and extrapolates to zero power.

Why does inhomogeneous broadening lower peak absorption but not the total line strength?

Because the frequency-integrated cross-section $\int\sigma\,\mathrm{d}\nu$ is fixed by the transition's oscillator strength, which does not care how the line is broadened; broadening only redistributes that fixed area. Spread it over a wider band and the peak must fall, so $\sigma_0\propto1/\Delta\nu$ — a strongly Doppler-broadened line has a low, shallow peak even though it absorbs the same total photons as a narrow line. This is why a cold, narrow sample can be optically thick at line centre while a hot one of the same column density is comparatively transparent, and why cooling a gas both narrows and deepens its peak.