Laser Linewidth and the Schawlow–Townes Limit

Above threshold a single-mode laser is, to superb approximation, the coherent state that quantum optics hands it: a nearly classical field with a well-defined amplitude and a phase that no equation of motion pins down. This article takes that field, together with the spontaneous-emission rate carried by Einstein's A-coefficient and the photon lifetime $\tau_p$ set by the cavity Q, and asks the question Schawlow and Townes asked in 1958 when the optical maser was still a proposal on paper: how narrow can the light be? The amplitude is not free — gain saturation clamps it. Push the intensity up and the gain drops, pushing it back down; that restoring force damps amplitude fluctuations. The phase has no such anchor: the gain medium does not care about the absolute optical phase, so nothing resists its drift. Every spontaneous-emission event drops one photon of random phase into the mode, nudging the giant coherent phasor by a tiny random angle, and millions per second accumulate into an unbiased random walk — phase diffusion. It is this slow, relentless wander, not any classical imperfection, that gives the line its finite width. The payoff is one compact formula, $\Delta\nu_{\mathrm{ST}} \propto (\Delta\nu_c)^2/P_{\mathrm{out}}$ , tying the quantum-limited linewidth to the cold-cavity bandwidth and the output power: a high-Q resonator delivering high circulating power produces a spectacularly pure tone, the floor for a milliwatt gas laser sitting below a millihertz. Everything downstream, from optical clocks to gravitational-wave interferometers, is an effort to reach it.

Key takeaways

The Schawlow–Townes limit is the fundamental quantum lower bound on the spectral width of a single-mode laser, set by spontaneous-emission photons of random phase kicking the coherent field.
Gain saturation clamps the amplitude but leaves the phase with no restoring force, so it random-walks — phase diffusion — which Fourier-transforms into a Lorentzian line of width $\Delta\nu_{\mathrm{ST}}$ .
The modified formula $\Delta\nu_{\mathrm{ST}} = \dfrac{\pi h\nu\,(\Delta\nu_c)^2\,n_{\mathrm{sp}}}{P_{\mathrm{out}}}$ falls as the inverse output power and the square of the small cold-cavity bandwidth.
The factor $n_{\mathrm{sp}}=N_2/(N_2-N_1)\ge 1$ raises the floor when inversion is incomplete; in semiconductors the Henry factor multiplies it by $1+\alpha^2$ , typically 5–50.
The floor is tiny: a 1 mW HeNe with a 1 MHz cavity bandwidth gives $\Delta\nu_{\mathrm{ST}}\approx 1$ mHz — far narrower than any real laser, which is pinned by technical noise.
Key idea: gain saturation clamps the laser amplitude but leaves the phase unconstrained; spontaneous-emission photons of random phase drive a phase random walk whose diffusion rate sets a Lorentzian spectral floor $\Delta\nu_{\mathrm{ST}}\propto (\Delta\nu_c)^2/P_{\mathrm{out}}$ .

Where this lives today

The Schawlow–Townes floor is the target of the world's most stable oscillators. Optical atomic clocks interrogate an ion or a lattice of atoms for seconds at a time, demanding a probe laser whose coherence time exceeds the interrogation window — so it is locked to a rigid, high-finesse reference cavity in a cryogenic, vibration-isolated vault; single-crystal silicon cavities at 124 K have reached thermal-noise-limited linewidths near 10 mHz, closing in on the floor the 1958 paper predicted. At the opposite extreme of size and cost, diode lasers for coherent communications live or die by the $1/P_{\mathrm{out}}$ and $(1+\alpha^2)$ scalings: phase-shift-keyed links need phase noise well below the symbol rate, so engineers push power up, wrap the chip in an external cavity to shrink $\Delta\nu_c$ , and hunt for the smallest Henry factor. The same equation explains the front end of a gravitational-wave detector: LIGO and Virgo run high-power, exquisitely narrow-linewidth lasers so that frequency and phase noise cannot masquerade as a spacetime strain — which is why both high circulating power and long-photon-lifetime reference cavities are non-negotiable. And across high-resolution laser spectroscopy, it is the Schawlow–Townes assurance that a good single-mode laser can be far narrower than the transitions it probes that makes the enterprise thinkable.

The physical picture

Picture the coherent laser field as a single long phasor rotating at the optical frequency $\nu$ , of length $\sqrt{n_0}$ where $n_0$ is the mean photon number in the mode — a huge number, $10^8$ to $10^{15}$ . Now let one spontaneously emitted photon join: it adds a unit phasor at a completely random angle. Its component along the field is an amplitude fluctuation the saturated gain springs straight back; its component perpendicular to the field tilts the phasor by roughly $1/\sqrt{n_0}$ radians, and nothing pulls that back — a permanent, tiny change of phase. Because each kick points in a random direction, the tilts have random sign and the phase executes an unbiased random walk — diffusion, whose variance grows linearly in time, $\langle\Delta\phi^2\rangle\propto t$ . A diffusing phase gives an exponentially decaying coherence, and the Fourier transform of an exponential is a Lorentzian. So the shape of the line and its width both follow from one act of bookkeeping — counting spontaneous photons entering the mode per second against the coherent photons already there.

Figure

The coherent laser field is a phasor of length √n₀; a spontaneous photon adds a unit vector at random angle whose ⟂ component tilts the phase by δφ ≈ 1/√n₀ ≈ 4.4×10⁻⁵ rad for n₀ ≈ 5×10⁸, so the phase random-walks with variance ⟨Δφ²⟩ ∝ t while gain clamps the amplitude

Why is the line so narrow? Because $n_0$ is enormous, each kick is minuscule and it takes an astronomical number to randomise the phase by one radian. Raise the power and $n_0$ grows, each fractional kick $1/\sqrt{n_0}$ shrinks, and the diffusion slows — the $1/P_{\mathrm{out}}$ scaling. The cold-cavity bandwidth enters twice: a narrower cavity means a longer photon lifetime, diluting each stored photon among more coherent ones, and the spontaneous feeding rate also scales with the cavity linewidth. The two effects multiply into the $(\Delta\nu_c)^2$ dependence, and the upshot is a line vastly narrower than the passive resonance it lives inside.

Figure

Gain plus stimulated emission narrow the passive cold-cavity resonance of width Δν_c = 1 MHz down to a lasing line of width Δν_ST ≈ 1 mHz — about 10⁹× narrower, drawn here as a needle because a true-scale line would be invisible

The core equations

The result everyone quotes is the modified Schawlow–Townes linewidth, which packages the phase-diffusion story into cavity parameters an engineer can measure: cold-cavity bandwidth, degree of inversion, output power. The word modified signals the amplitude-clamped (well-above-threshold) form; the 1958 paper carried an extra factor of 2 we revisit under limiting cases.

\Delta\nu_{\mathrm{ST}} = \dfrac{\pi h\nu\,(\Delta\nu_c)^2\,n_{\mathrm{sp}}}{P_{\mathrm{out}}} \qquad (1)

where: $\Delta\nu_{\mathrm{ST}}$ — full width at half maximum of the Lorentzian laser line (Hz); $h\nu$ — photon energy at the lasing frequency $\nu$ ; $\Delta\nu_c$ — cold-cavity (passive) resonance FWHM, equal to $1/(2\pi\tau_p)$ ; $n_{\mathrm{sp}}$ — spontaneous-emission (population-inversion) factor; $P_{\mathrm{out}}$ — power coupled out through the mirrors. Read it as a competition: photon energy and spontaneous seeding widen the line; stored coherent power squeezes it; the bandwidth enters squared because it governs both storage time and spontaneous feeding. The cavity bandwidth here is the passive resonance width measured with the gain off, fixed by how long a photon survives in the resonator — the photon lifetime of the cavity-Q chapter.

\Delta\nu_c = \dfrac{1}{2\pi\tau_p} = \dfrac{\nu}{Q} \qquad (2)

where: $\tau_p$ — cavity photon lifetime; $Q$ — loaded quality factor; $\nu$ — resonance frequency. A high-Q cavity has a long $\tau_p$ and narrow $\Delta\nu_c$ ; because $\Delta\nu_{\mathrm{ST}}\propto(\Delta\nu_c)^2$ , doubling the finesse quarters the fundamental linewidth before any change in power. To connect the photon number to a power meter, note that every stored photon leaks out: the energy $n_0 h\nu$ drains on the timescale $\tau_p$ , so the output power fixes the mode occupation.

P_{\mathrm{out}} = \dfrac{n_0\,h\nu}{\tau_p} = n_0\,h\nu\,(2\pi\Delta\nu_c) \qquad (3)

where: $n_0$ — mean photons stored in the lasing mode above threshold. Inverting, $n_0 = P_{\mathrm{out}}\tau_p/(h\nu)$ : a milliwatt of red light in a 160 ns cavity holds about $5\times10^8$ photons — a large occupation that makes each spontaneous kick negligible. Beneath the engineering formula lies the microscopic statement the picture built — the linewidth is the phase-diffusion rate: the spontaneous rate into the mode over the coherent photon number.

\Delta\nu_{\mathrm{ST}} = \dfrac{R_{\mathrm{sp}}}{4\pi\,n_0},\qquad \langle\Delta\phi^2(\tau)\rangle = \dfrac{R_{\mathrm{sp}}}{2 n_0}\,|\tau| \qquad (4)

where: $R_{\mathrm{sp}}$ — spontaneous-emission rate into the mode (per second); $n_0$ — mean photon number; $\tau$ — time delay. Each of the $R_{\mathrm{sp}}\tau$ kicks adds a variance $1/(2n_0)$ ; the total grows linearly (diffusion), and matching $2\pi\Delta\nu_{\mathrm{ST}}\tau$ delivers the $4\pi$ . The inversion enters through the spontaneous-emission factor — exactly one only for a perfectly inverted medium, growing without bound as the lower level fills.

n_{\mathrm{sp}} = \dfrac{N_2}{N_2 - N_1}\;\ge\;1 \qquad (5)

where: $N_2$ , $N_1$ — upper and lower laser-level population densities. It counts spontaneous photons injected per stimulated photon at clamped gain: a four-level laser with $N_1\to 0$ reaches the ideal $n_{\mathrm{sp}}=1$ , while a barely-inverted system pays a heavy penalty. A diffusing phase leaves a clean fingerprint: the normalised first-order coherence is $g^{(1)}(\tau)=e^{i2\pi\nu\tau}e^{-\pi\Delta\nu_{\mathrm{ST}}|\tau|}$ , an oscillation inside an exponentially decaying envelope (Gaussian statistics give $\langle e^{i\Delta\phi}\rangle=e^{-\langle\Delta\phi^2\rangle/2}$ ). By Wiener–Khinchin the power spectrum is its Fourier transform, and the transform of a decaying exponential is a Lorentzian of width exactly $\Delta\nu_{\mathrm{ST}}$ .

S(\nu') = \dfrac{2}{\pi\,\Delta\nu_{\mathrm{ST}}}\;\dfrac{1}{1+\left(\dfrac{2(\nu'-\nu)}{\Delta\nu_{\mathrm{ST}}}\right)^2} \qquad (6)

where: $S(\nu')$ — normalised spectral density; $\nu'$ — observation frequency. The lineshape is Lorentzian, not Gaussian — its wings fall only as $1/(\nu'-\nu)^2$ , so far-detuned phase noise is stubborn. Finally, semiconductor lasers are broader than equation (1) predicts, because amplitude and phase couple through the carrier-dependent refractive index. Henry captured the penalty in one multiplicative factor.

\Delta\nu = \Delta\nu_{\mathrm{ST}}\,\left(1+\alpha^2\right) \qquad (7)

where: $\alpha$ — linewidth-enhancement (Henry) factor, the ratio of the real to the imaginary part of the differential susceptibility. With $\alpha\approx 2$ – $7$ in typical diodes the line broadens 5–50 times; $\alpha=0$ recovers the pure value of gas lasers.

Derivation APhase diffusion and the Lorentzian linewidth

The goal is to build the whole result from one microscopic event — a single spontaneous photon added at random phase — and show that summing such events produces an exponentially decaying coherence and, by Fourier transform, a Lorentzian line whose width is the diffusion rate over $\pi$ . We treat the field as a coherent phasor of essentially fixed length $\sqrt{n_0}$ disturbed only by spontaneous emission, carrying the phase variance from a single kick to the measured FWHM.

Step 1. A unit photon added at a random angle $\theta$ to a phasor of length $\sqrt{n_0}$ tilts it by $\delta\phi\approx\sin\theta/\sqrt{n_0}$ ; averaging $\sin^2\theta=1/2$ gives a mean-square kick $\langle\delta\phi^2\rangle=1/(2n_0)$ . Why: only the perpendicular component changes the phase; the parallel component — an amplitude change — is erased by the saturated gain, so the phase kick is the whole story.

Step 2. In a time $\tau$ there are $R_{\mathrm{sp}}\tau$ uncorrelated events, and because the kicks are independent their variances add: $\langle\Delta\phi^2(\tau)\rangle=R_{\mathrm{sp}}\tau/(2n_0)$ . Why: variances of independent increments sum linearly — this linear growth in time is diffusion, and it is what gives the line a finite width rather than a jitter that saturates.

Step 3. The phase is a sum of many small independent kicks, hence Gaussian, and the characteristic-function identity gives $g^{(1)}(\tau)=\langle e^{i\Delta\phi}\rangle=e^{-\langle\Delta\phi^2\rangle/2}=e^{-R_{\mathrm{sp}}|\tau|/(4n_0)}$ . Why: the exponential envelope is not assumed — it falls out of Gaussian statistics acting on a linearly growing variance.

Step 4. Matching the envelope to the standard form $e^{-\pi\Delta\nu_{\mathrm{ST}}|\tau|}$ and Fourier transforming (an exponential coherence gives a Lorentzian) yields $\Delta\nu_{\mathrm{ST}}=R_{\mathrm{sp}}/(4\pi n_0)$ . Why: the Lorentzian FWHM equals the exponential decay rate over $\pi$ , fixing the factor and identifying diffusion rate with linewidth. *Note: the Wiener–Khinchin transform of $e^{-\pi\Delta\nu|\tau|}$ is a Lorentzian with FWHM $\Delta\nu$ , confirming that the exponential coherence decay and the Lorentzian linewidth are two faces of the same diffusion rate.*

Go deeper · Derivation B · 1 minRecasting into the Schawlow–Townes power form

Derivation BRecasting into the Schawlow–Townes power form

The goal is to convert the microscopic result $\Delta\nu_{\mathrm{ST}}=R_{\mathrm{sp}}/(4\pi n_0)$ into the engineering form in bench quantities — cold-cavity bandwidth, inversion factor, output power. We express the spontaneous rate and the coherent photon number in cavity language, substitute, and collect factors of the photon lifetime until the passive bandwidth appears squared.

Step 1. Above threshold the round-trip gain is clamped to equal the cavity loss $1/\tau_p$ ; the ratio of spontaneous to stimulated emission into the mode is $n_{\mathrm{sp}}$ , so the spontaneous injection rate is $R_{\mathrm{sp}}=n_{\mathrm{sp}}/\tau_p$ . Why: gain clamping freezes the emission rates just above threshold, letting us trade the microscopic $R_{\mathrm{sp}}$ for the macroscopic loss rate scaled by the inversion penalty.

Step 2. The output power fixes the mode occupation through $P_{\mathrm{out}}=n_0 h\nu/\tau_p$ , so $n_0=P_{\mathrm{out}}\tau_p/(h\nu)$ . Why: this ties the photon number to a power meter, and is where the $1/P_{\mathrm{out}}$ scaling is born.

Step 3. Substituting both gives $\Delta\nu_{\mathrm{ST}}=\dfrac{n_{\mathrm{sp}}/\tau_p}{4\pi\,P_{\mathrm{out}}\tau_p/(h\nu)}=\dfrac{h\nu\,n_{\mathrm{sp}}}{4\pi\,\tau_p^{2}\,P_{\mathrm{out}}}$ . Why: the photon lifetime now appears squared, foreshadowing the quadratic cavity-bandwidth dependence.

Step 4. Insert $\tau_p=1/(2\pi\Delta\nu_c)$ , so $1/(4\pi\tau_p^2)=\pi(\Delta\nu_c)^2$ , and the linewidth becomes $\Delta\nu_{\mathrm{ST}}=\pi h\nu\,(\Delta\nu_c)^2 n_{\mathrm{sp}}/P_{\mathrm{out}}$ . Why: the whole $(\Delta\nu_c)^2$ dependence traces to the two lifetime factors — one from storage, one from spontaneous feeding. *Dimensional check: $[h\nu]/[P]\cdot[(\Delta\nu_c)^2]=\mathrm{s}\cdot\mathrm{s}^{-2}=\mathrm{Hz}$ , and the worked example reaches the same 1 mHz by both the microscopic and engineering routes.*

Limiting cases

Case	Result	Meaning
Far above threshold	$\Delta\nu\to\Delta\nu_{\mathrm{ST}}=\dfrac{\pi h\nu(\Delta\nu_c)^2 n_{\mathrm{sp}}}{P_{\mathrm{out}}}$	Gain saturation damps amplitude fluctuations; only phase diffusion survives, so the prefactor is 1, not 2.
Near threshold	$\Delta\nu\approx 2\,\Delta\nu_{\mathrm{ST}}$	Amplitude fluctuations are undamped and contribute equally — the classic factor of 2 of the 1958 expression.
Ideal complete inversion	$n_{\mathrm{sp}}\to 1$	With $N_1\to 0$ the medium is perfectly inverted and the spontaneous penalty is minimal; four-level lasers approach this.
Poorly inverted medium	$n_{\mathrm{sp}}\gg 1$	As $N_1\to N_2$ the factor diverges and the line inflates — three-level and near-threshold lasers pay dearly.
Semiconductor laser	$\Delta\nu=\Delta\nu_{\mathrm{ST}}(1+\alpha^2)$	Carrier coupling of gain and index adds correlated phase noise, broadening the line 5–50 times.

Common trapA laser's linewidth is just its cavity or gain bandwidth

This is the most common confusion, and it is wrong by many orders of magnitude. The gain bandwidth (tens of GHz to THz) and the free spectral range select which mode oscillates; the cold-cavity width $\Delta\nu_c$ (kHz to MHz) is the passive filter that mode sits in. The actual laser line is far narrower than any of these — the 1 mHz HeNe floor is a billion times below its own 1 MHz cavity width. What narrows it is the active process: stimulated emission pours coherent photons into one mode until spontaneous emission is a vanishing perturbation. So never quote $\Delta\nu_c$ , the FSR, or the gain bandwidth as the linewidth. Beware too the factor-of-2 ambiguity — the 1958 formula, valid near threshold, is twice the amplitude-clamped result above it — and differing $n_{\mathrm{sp}}$ conventions.

Worked examples

1 · The fundamental floor of a HeNe laser (confidence builder)

A 1 mW helium–neon laser oscillates at 632.8 nm on a single mode in a cavity of cold bandwidth $\Delta\nu_c=1$ MHz, with $n_{\mathrm{sp}}=1$ . Find the Schawlow–Townes linewidth and the resulting coherence time.

Step 1. The photon energy: $\nu=c/\lambda=(2.998\times10^8)/(632.8\times10^{-9})=4.738\times10^{14}$ Hz, so $h\nu=(6.626\times10^{-34})(4.738\times10^{14})=3.139\times10^{-19}$ J.

Step 2. Apply equation (1): $\Delta\nu_{\mathrm{ST}}=\pi h\nu(\Delta\nu_c)^2 n_{\mathrm{sp}}/P_{\mathrm{out}}=\pi(3.139\times10^{-19})(10^{12})(1)/(10^{-3})=\pi(3.139\times10^{-4})=\mathbf{9.9\times10^{-4}\ Hz\approx 1\ mHz}$ .

Step 3. The coherence time is $\tau_c=1/(\pi\Delta\nu_{\mathrm{ST}})=1/(\pi\cdot 9.9\times10^{-4})=\mathbf{3.2\times10^{2}\ s}$ , roughly five minutes, for a coherence length $c\tau_c\approx 1\times10^{11}$ m. The microscopic route agrees: $\tau_p=1/(2\pi\Delta\nu_c)=159$ ns gives $n_0=P_{\mathrm{out}}\tau_p/(h\nu)=5.1\times10^8$ (so $\delta\phi\approx 1/\sqrt{n_0}\approx 4.4\times10^{-5}$ rad, the number on the first figure) and $R_{\mathrm{sp}}=n_{\mathrm{sp}}/\tau_p=6.3\times10^6$ s $^{-1}$ , whence $\Delta\nu_{\mathrm{ST}}=R_{\mathrm{sp}}/(4\pi n_0)=9.9\times10^{-4}$ Hz — identical. The floor is ≈ 1 mHz; no real HeNe comes within nine orders of this, because mirror vibration and thermal drift dominate.

2 · A semiconductor diode laser with the Henry penalty

A Fabry–Pérot diode laser at 1550 nm has cavity length 300 μm, group index 3.5, facet reflectivity 0.30, output power 5 mW, and $n_{\mathrm{sp}}=2$ . Find the bare Schawlow–Townes floor and the observed linewidth for a Henry factor $\alpha=5$ .

Step 1. Mirror loss is $\alpha_m=(1/L)\ln(1/R)=(1/300\times10^{-6})\ln(1/0.30)=4.0\times10^{3}$ m $^{-1}$ , so $\tau_p=n_g/(c\,\alpha_m)=3.5/[(2.998\times10^8)(4.0\times10^3)]=2.9\times10^{-12}$ s and $\Delta\nu_c=1/(2\pi\tau_p)=\mathbf{54.7\ GHz}$ — enormous, because the tiny lossy chip barely stores light.

Step 2. With $h\nu=(6.626\times10^{-34})(1.934\times10^{14})=1.28\times10^{-19}$ J, equation (1) gives $\Delta\nu_{\mathrm{ST}}=\pi(1.28\times10^{-19})(5.47\times10^{10})^{2}(2)/(5\times10^{-3})=\mathbf{0.48\ MHz}$ — the large $(\Delta\nu_c)^2$ lifts the floor from mHz to hundreds of kHz.

Step 3. Apply the Henry penalty: $1+\alpha^2=26$ , so $\Delta\nu=26\times0.48\ \mathrm{MHz}=\mathbf{12.5\ MHz}$ . This is why a bare diode is a poor spectroscopic source, and why narrowing it demands external cavities that shrink $\Delta\nu_c$ and low- $\alpha$ material.

Physics of the portrait

The linewidth limit is where the quantum theory of light meets the engineering of resonators. It was written down before a single laser existed, by two people arguing that spontaneous emission — the same process Einstein had quantified in 1917 — sets an irreducible floor on the purity of stimulated light.

Origins · 1 minArthur Schawlow and how this was discovered

Arthur Schawlow

1921–1999 · co-author of the 1958 optical-maser paper and a founder of laser spectroscopy

Arthur Leonard Schawlow was a Canadian-American physicist whose name is fixed to the quantum linewidth limit through the 1958 paper "Infrared and Optical Masers", written with his brother-in-law Charles Townes. The maser already existed — Townes, Gordon and Zeiger had run the ammonia maser in 1954 — and the open question was whether the trick could be pushed from microwaves to visible light. Schawlow and Townes worked out the resonator that would make it possible (an open Fabry–Pérot rather than a closed box) and, crucially here, estimated how narrow the emission could be, recognising that spontaneous emission into the mode imposes a fundamental floor scaling as $(\Delta\nu_c)^2/P$ . They did not build the laser — Maiman's ruby device came in 1960 — but they framed the physics everyone chased. Schawlow spent his career at Stanford pioneering Doppler-free saturation spectroscopy, sharing the 1981 Nobel Prize; Townes had taken the 1964 Nobel for the maser–laser principle.

History

1917

— Einstein introduces the A and B coefficients, quantifying spontaneous and stimulated emission

1954

— Gordon, Zeiger and Townes operate the first ammonia maser

1958

— Schawlow and Townes publish "Infrared and Optical Masers", predicting the quantum linewidth floor

1960

— Maiman demonstrates the first working laser, in ruby

1964

— Townes shares the Nobel Prize for the maser–laser principle

1982

— Henry derives the linewidth enhancement factor α for semiconductor lasers

Going further

Texts. Schawlow and Townes, "Infrared and Optical Masers", Phys. Rev. 112, 1940 (1958) · Loudon, The Quantum Theory of Light, 3rd ed. (laser fluctuations) · Saleh and Teich, Fundamentals of Photonics, 2nd ed., §16 (Laser Coherence) · Foot, Atomic Physics · Demtröder, Laser Spectroscopy, Vol. 1 · Henry, IEEE J. Quantum Electron. QE-18, 259 (1982). The key derivation chains are: the phase-diffusion result $\Delta\nu_{\mathrm{ST}}=R_{\mathrm{sp}}/(4\pi n_0)$ follows from the single-kick variance $\langle\delta\phi^2\rangle=1/(2n_0)$ and the Wiener–Khinchin transform of $e^{-\pi\Delta\nu|\tau|}$ ; the engineering form uses $R_{\mathrm{sp}}=n_{\mathrm{sp}}/\tau_p$ , $n_0=P_{\mathrm{out}}\tau_p/(h\nu)$ , and $\tau_p=1/(2\pi\Delta\nu_c)$ ; and the Henry broadening factor is $(1+\alpha^2)$ . The sub-mHz HeNe value follows from $P_{\mathrm{out}}=1$ mW, $\Delta\nu_c=1$ MHz, $\nu=4.74\times10^{14}$ Hz, $n_{\mathrm{sp}}=1$ ; the semiconductor value from the parameters in worked example 2. The factor-of-2 distinction between the near-threshold and well-above-threshold forms is noted in the limiting-cases table.

Prerequisites. Cavity Q Factor and Photon Lifetime · Stimulated Emission and the Gain Coefficient · Laser Rate Equations · Threshold Condition · Longitudinal Modes and Free Spectral Range. Leads to. Optical Resonators and Cavity Modes · Population Inversion and Optical Gain.

Ask the tutor: Give me your laser's wavelength, output power, and cavity finesse or bandwidth, and I will walk through $n_0$ , $R_{\mathrm{sp}}$ , and the Schawlow–Townes linewidth, then add the Henry penalty if it is a diode.

Is the Schawlow–Townes linewidth what a spectrum analyser would actually show?

Almost never directly. The fundamental floor for a decent laser is millihertz to kilohertz — far below the resolution of an optical spectrum analyser and below what mirror vibration, air currents, and current or temperature drift let you reach. What you measure is dominated by that technical noise, hundreds of kHz to MHz even in a good lab laser. The quantum floor becomes visible only after heroic stabilisation, and even then you infer it from a delayed self-heterodyne beat rather than read it off an analyser. The formula tells you the best you could ever do, not what comes out of the box.

Why is the lineshape Lorentzian rather than Gaussian?

Because the phase undergoes pure diffusion — a random walk with independent increments and a variance growing linearly in time. That makes the coherence decay as a simple exponential in the delay, and the Fourier transform of an exponential is a Lorentzian. A Gaussian line arises instead if the frequency wanders slowly and continuously (as with 1/f technical noise), which is why real lasers show a Lorentzian core from quantum phase diffusion inside a broader Gaussian pedestal from technical noise. The Lorentzian wings, falling only as the inverse square of detuning, make far-off phase noise stubborn.

Does more output power always narrow the line?

For the fundamental floor, yes — the width scales as $1/P_{\mathrm{out}}$ , because higher power means more coherent photons $n_0$ , each kick tilts the phasor by a smaller fraction $1/\sqrt{n_0}$ , and the phase diffuses more slowly; doubling the power halves the width. In practice, though, the floor is usually so far below the technical-noise level that adding power buys nothing measurable until vibration and drift are suppressed. In semiconductors there is even a ceiling: at high power, gain compression and heating cause rebroadening, so the clean $1/P$ law holds only over a middle range.

What is the Henry factor physically, and why do gas lasers not have it?

The Henry factor $\alpha$ measures how strongly a change in gain drags the refractive index along, because in a semiconductor both are set by the same carrier density. When a spontaneous photon perturbs the amplitude, the carriers relax to restore steady state, and that shifts the index and hence the phase — so amplitude noise, which gain saturation would otherwise erase, leaks into the phase and broadens the line by $1+\alpha^2$ . Gas and solid-state lasers work near the peak of a symmetric gain line, where the index change with gain is nearly zero, so $\alpha\approx 0$ . Diode lasers work on the side of an asymmetric gain curve, giving $\alpha\approx 2$ – $7$ and a line 5–50 times broader.

Where does the factor of 2 between the original and modified formulas come from?

The 1958 derivation counted amplitude and phase fluctuations from spontaneous emission as contributing equally. Well above threshold, however, gain saturation strongly damps the amplitude fluctuations — they see a restoring force — while the phase does not, so only the phase half survives and the width drops by a factor of 2. So near threshold you use the original expression with the 2; well above threshold you use the modified one without it. Textbooks differ on which they quote and on the definition of $n_{\mathrm{sp}}$ , so always check which form a formula is before comparing numbers.

Is the Schawlow–Townes limit really a quantum limit, or just classical noise?

It is genuinely quantum. The phase kicks come from spontaneous emission, which has no classical description — the field's coupling to the vacuum modes, the same physics as Einstein's A-coefficient and an atom's natural linewidth. You can dress the derivation in semiclassical phasor language, as we did, but the rate $R_{\mathrm{sp}}$ and the discreteness of the added photon are quantum facts. This is why the limit cannot be engineered away: suppress every classical noise source, cool the mirrors, lock to a perfect cavity, and the vacuum still delivers one random-phase photon at a time into the mode — the optical analogue of shot noise, setting the ultimate coherence of any laser.