Active and Passive Mode-Locking Mechanisms

A free-running laser oscillating on hundreds of longitudinal modes puts out a noisy, flickering beam: the modes are all present, but their phases wander independently, so their sum is random speckle in time. Mode-locking turns that noise into the shortest controllable events humans can make. It borrows two results from earlier in this chapter: from the resonator articles, that a cavity supports an evenly spaced comb of longitudinal modes separated by the free spectral range $\Delta\nu=c/2L$ ; from the gain articles, a saturable amplifier that holds many of those modes above threshold at once. Mode-locking then adds the one ingredient the free-running laser lacks — a fixed phase relationship between the modes. Once the phases are locked, the algebra is that of an $N+1$ -slit diffraction grating: equally spaced modes with a common phase add tip-to-tail into a sharp spike that recurs once per round trip. The output collapses from a continuous glow into a train of ultrashort pulses spaced by $T=2L/c$ , each a factor $N+1$ taller than the average power and $N+1$ times narrower than the round-trip time. The two families differ only in how the locking is imposed: active mode-locking drives an intracavity modulator with an external clock tuned to $\Delta\nu$ , while passive mode-locking lets the pulse switch a saturable absorber for itself. A single differential equation — Haus's master equation — governs both as a steady-state balance of gain, loss, dispersion and shaping. It is the engine behind femtosecond science, optical clocks, and the frequency comb.

Key takeaways

Mode-locking fixes a common phase across many longitudinal cavity modes, so their coherent sum is a train of ultrashort pulses spaced by the round-trip time $T=2L/c$ .
Active mode-locking uses an externally clocked modulator at the mode spacing $\Delta\nu=c/2L$ , opening a low-loss window once per round trip to carve out a pulse.
Passive mode-locking uses a saturable absorber whose loss falls at high intensity, so the pulse bleaches its own loss and shortens itself each pass — reaching far shorter pulses.
Pulse duration scales as the inverse of the phase-locked bandwidth, $\tau_p\sim 1/\Delta\nu_{\mathrm{osc}}$ , so broad-gain media like Ti:sapphire reach few-femtosecond, near-single-cycle pulses.
Haus's master equation casts both regimes as a fixed-point balance of gain, loss, dispersion and shaping — a Gaussian pulse for active AM, a sech soliton for the fast absorber.
Key idea: a pulse train is nothing but equally spaced cavity modes with locked phases; whether an external modulator (active) or the pulse's own intensity through a saturable absorber (passive) enforces the locking, the steady state is a self-consistent balance of gain, loss/modulation, and dispersion.

The core insight

A periodic train of ultrashort pulses is nothing but a coherent superposition of equally spaced cavity modes with locked phases; whether the locking is enforced by an external modulator (active) or by the pulse's own intensity through a saturable absorber (passive), the steady state is a self-consistent balance of gain, loss/modulation, and dispersion.

Where this lives today

The workhorse of ultrafast science is the Kerr-lens mode-locked Ti:sapphire oscillator. Its enormous gain bandwidth (roughly 700–1050 nm) supports sub-10-fs pulses, and its saturable-absorber action is a beautiful cheat: the intense pulse self-focuses through the nonlinear index $n_2$ , and a well-placed aperture makes the tightly focused peak see lower loss than the weak wings — an effectively instantaneous absorber. Prism pairs or chirped mirrors supply the anomalous dispersion that balances self-phase modulation, so the pulse circulates as a cavity soliton. Stabilise its repetition rate and carrier-envelope offset and the same laser becomes a self-referenced optical frequency comb — the invention that earned Hall and Hänsch a share of the 2005 Nobel Prize and now underpins optical atomic clocks, distance metrology, and broadband molecular fingerprinting. Where robustness or an electronic timing reference matters more than the shortest pulse, the other branches dominate. Semiconductor saturable absorber mirrors (SESAMs) build an engineered absorber into a Bragg stack, giving compact, self-starting mode-locking of fibre and thin-disk lasers used in micromachining and eye surgery. Actively mode-locked fibre and diode lasers, clocked at a harmonic of $c/2L$ , deliver picosecond trains phase-locked to an RF standard for optical time-division multiplexing and clock distribution. From attoscience drivers to telecom sources, every one is a steady-state solution of Haus's balance.

The physical picture

Picture $N+1$ phasors of equal length, one per mode, stepped in frequency by the mode spacing $\Delta\omega$ . With random phases they point every which way and their sum stays small and jittery, of order $\sqrt{N}$ — the noisy free-running beam. Mode-locking sets all the phases equal, so once per period they all point the same way and add tip-to-tail into one vector of length $N+1$ : a bright pulse. An instant later they fan out and cancel to near-darkness, realigning only after a full period $T=2\pi/\Delta\omega=2L/c$ — exactly one round trip. That rhythm is the pulse train, the same array-factor mathematics that makes a grating throw sharp orders.

Figure

Time and frequency are two views of one mode-locked laser: the evenly spaced longitudinal modes on the left, separated by Δν = c/2L = 100 MHz for a 1.5 m cavity, Fourier-transform into the pulse train on the right, spaced by T = 1/Δν = 10 ns with each pulse a factor N+1 taller than the average power

In the time domain the two families are two kinds of shutter. The active modulator is clocked externally: it opens once per round trip just as the pulse arrives, transmits the peak, and clips light that drifts early or late, so the pulse is re-timed and its wings trimmed each pass. The passive absorber is a shutter the pulse opens for itself — the intense peak bleaches it and sees low loss, while the dim edges see high loss and are shaved from both sides. The difference is speed: a clocked modulator opens on its nanosecond drive period, so its shaping force is feeble near the tip of an already-short pulse, whereas a fast absorber responds on the pulse's own timescale — the shorter the pulse, the sharper its window. That is why passive reaches femtoseconds and active stalls at picoseconds.

Figure

Why passive beats active — the net-gain window: active AM (top) dips the loss parabolically once per round trip, leaving a wide window that shapes only to τ_p ≈ 100 ps, while the fast saturable absorber (bottom) is bleached only at the intensity peak, so its window is as short as the pulse and reaches τ_p ≈ 10 fs

Whatever the shutter, steady state is a fixed point. Each round trip the finite gain bandwidth filters the spectral wings, broadening the pulse in time, while the modulator or absorber sharpens it in time, broadening its spectrum. When these tendencies cancel — and, for the shortest pulses, when dispersion and self-phase modulation also cancel — the pulse reproduces itself pass after pass and the laser has "locked."

The core equations

The total field is a sum over the oscillating longitudinal modes, each a phasor at its own frequency and phase $\phi_n$ ; the free-running laser has random $\phi_n$ , and mode-locking makes them equal (or a linear ramp).

E(t)=\sum_{n=-N/2}^{N/2} E_0\, e^{i[(\omega_0+n\,\Delta\omega)t+\phi_n]} \qquad (1)

where: $\omega_0$ — central optical angular frequency; $\Delta\omega=2\pi c/2L$ — angular mode spacing (the free spectral range in rad/s); $\phi_n$ — phase of mode $n$ , which mode-locking fixes to a common value; $E_0$ — per-mode field amplitude, taken equal here; $N+1$ — number of locked longitudinal modes. Set all $\phi_n=0$ , sum the geometric series, and square: the result is the array factor of an $N+1$ -slit grating — a periodic set of sharp intensity peaks, the pulse train in closed form.

I(t)=E_0^2\,\frac{\sin^2\!\left[(N+1)\,\Delta\omega\,t/2\right]}{\sin^2\!\left[\Delta\omega\,t/2\right]} \qquad (2)

where: $I(t)$ — instantaneous intensity of the coherent sum; the peak at $t=mT$ is $(N+1)^2E_0^2$ , growing as $N^2$ ; $T=2\pi/\Delta\omega=2L/c$ — round-trip time = pulse period. The main lobe has width $\sim T/(N+1)$ , so more modes mean a shorter pulse. The width of that main lobe gives the single most useful scaling in the field: the pulse duration is the inverse of the total phase-locked bandwidth, a floor no mechanism can beat — pure Fourier bookkeeping.

\tau_p \approx \frac{T}{N+1}=\frac{1}{(N+1)\,\Delta\nu}=\frac{1}{\Delta\nu_{\mathrm{osc}}} \qquad (3)

where: $\tau_p$ — pulse duration (peak to first null); $\Delta\nu=c/2L$ — mode spacing in Hz; $\Delta\nu_{\mathrm{osc}}=(N+1)\Delta\nu$ — total phase-locked bandwidth. The broader the locked spectrum, the shorter the pulse. The rigorous form of " $\tau_p\sim1/\Delta\nu$ " is the time–bandwidth product, an uncertainty-type inequality with a shape-dependent constant, with equality only for an unchirped, transform-limited pulse.

\Delta t\,\Delta\nu \ge K,\qquad K=\begin{cases}0.441 & \text{Gaussian}\\[2pt] 0.315 & \text{sech}^2\end{cases} \qquad (4)

where: $\Delta t$ — intensity FWHM pulse duration; $\Delta\nu$ — spectral FWHM; $K$ — time–bandwidth constant set by the pulse shape. A sech² pulse (the soliton shape of passive mode-locking) packs more spectrum into less time than a Gaussian. Both mechanisms are captured by one Haus master equation for the envelope $A$ , evolving over many round trips $T$ while structured on the fast time $t$ within a pulse: $T_R\,\partial A/\partial T=[(g-l)+(1/\Omega_g^2+iD'')\,\partial_t^2+(\text{shaping})]\,A$ . It balances the per-pass gain $g$ and loss $l$ , a finite-bandwidth filter $\partial_t^2/\Omega_g^2$ set by the gain bandwidth $\Omega_g$ , group-delay dispersion $D''$ , and a shaping term. For active amplitude modulation the shaping is the modulator's parabolic loss $-M\omega_m^2 t^2$ near its transmission minimum (drive $\omega_m=2\pi\Delta\nu$ ); a Gaussian ansatz at steady state balances the bandwidth curvature against the modulator curvature, and the width shrinks only as the fourth root of the shaping parameters.

A(t)=A_0\,e^{-t^2/2\tau_p^2},\qquad \tau_p=\left(\frac{g}{M\,\Omega_g^2\,\omega_m^2}\right)^{1/4} \qquad (5)

where: $A_0$ — peak amplitude fixed by gain saturation; $\tau_p$ — the $1/e$ half-width. The fourth-root dependence is why active mode-locking is stuck at picoseconds: quadrupling the modulation depth only halves the pulse, so you cannot brute-force your way to femtoseconds. Passive mode-locking keeps the same master equation but replaces the parabolic modulator loss with a nonlinear self-amplitude modulation $\gamma|A|^2$ (the fast absorber) and, for the shortest pulses, Kerr self-phase modulation $\delta|A|^2$ and anomalous dispersion $D''$ — the balance built explicitly in Derivation 3. Because these terms grow with intensity the shaping strengthens as the pulse sharpens, and the steady-state solution is a hyperbolic secant, the same envelope as an optical soliton, carrying a small chirp $\beta$ once the gain and absorber curvatures act.

A(t)=A_0\,\mathrm{sech}\!\left(\frac{t}{\tau_p}\right)^{1+i\beta} \qquad (6)

where: $\mathrm{sech}$ — the soliton envelope; $\tau_p$ — soliton width, set by the dispersion–nonlinearity balance $\tau_p\approx 2|D''|/(\delta E_p)$ ; $\beta$ — chirp, zero for a pure soliton. Because $\tau_p\propto 1/E_p$ , pumping harder shortens the pulse — the opposite of the active fourth-root sluggishness. In the frequency domain the same pulse train is an optical frequency comb: a rigid ladder of lines at the repetition rate, offset bodily by the carrier-envelope frequency.

f_{\mathrm{rep}}=\frac{1}{T_R}=\frac{c}{2L},\qquad \nu_n=n\,f_{\mathrm{rep}}+f_{\mathrm{ceo}} \qquad (7)

where: $f_{\mathrm{rep}}$ — pulse repetition rate, equal to the mode spacing; $L$ — one-way optical cavity length; $f_{\mathrm{ceo}}$ — carrier-envelope offset frequency, which appears because the pulse's carrier and envelope travel at different (phase vs group) velocities. Stabilise $f_{\mathrm{rep}}$ and $f_{\mathrm{ceo}}$ and every optical line $\nu_n$ is known to microwave precision.

Derivation AFrom locked modes to a pulse train

The goal is to turn the statement "lock the phases" into the concrete pulse train of equation (2) and read off its two headline numbers — the peak is $N+1$ times the average power, and the pulse is $N+1$ times shorter than the round-trip time. We factor out the common carrier, sum the finite geometric series, evaluate its peak by a limit, and locate its first null. The calculation is identical to the diffraction pattern of an $N+1$ -slit grating, which is the cleanest way to see why locked modes make sharp, evenly spaced spikes.

Step 1. Set every phase $\phi_n=0$ and pull the carrier out of equation (1): $E(t)=E_0e^{i\omega_0 t}\sum_{n} e^{in\Delta\omega t}$ . Why: the carrier sets only the colour; all pulse structure lives in the envelope, a finite sum of equally spaced phasors.

Step 2. Evaluate the geometric series with $x=e^{i\Delta\omega t}$ : $\sum_{n=-N/2}^{N/2}x^n=\sin[(N+1)\Delta\omega t/2]/\sin[\Delta\omega t/2]$ , the Dirichlet kernel; squaring gives $I(t)$ . Why: a sum of equal, equally spaced phasors is the grating array factor, so a mode-locked laser is a grating in time.

Step 3. Take the peak. At $t=mT$ both sines vanish; L'Hôpital gives $\lim I=(N+1)^2E_0^2$ , while the period-averaged power is $\bar I=(N+1)E_0^2$ , so $I_{\mathrm{peak}}/\bar I=N+1$ . Why: locking a thousand modes turns a watt of average power into kilowatt peaks — coherence, not extra energy, does the work.

Step 4. Locate the first null: the numerator sine returns to zero when $(N+1)\Delta\omega\,\tau_p/2=\pi$ , i.e. $\tau_p=2\pi/[(N+1)\Delta\omega]=T/(N+1)$ . Why: the main-lobe half-width is the round-trip time over the number of modes — equation (3), the whole compression story in one line.

Go deeper · Derivation B · 1 minThe sideband picture of active mode-locking

Derivation BThe sideband picture of active mode-locking

The aim is to show, in the frequency domain, how an externally clocked modulator forces the phases together — not by magic but by coherent injection. A modulator driven at the mode spacing puts sidebands on every mode that land precisely on its neighbours, seeding each with phase-coherent light from next door; pass after pass this pulls the whole comb into a common phase. We follow one mode through the modulator, apply a product-to-sum identity, and watch the sidebands fall on the adjacent resonances.

Step 1. Send mode $n$ , written $E_n\cos\omega_n t$ with $\omega_n=\omega_0+n\omega_m$ , through a modulator $T(t)=1-M[1-\cos\omega_m t]$ driven at $\omega_m=\Delta\omega$ ; the key piece is the product $M E_n\cos\omega_n t\cos\omega_m t$ . Why: the constant part just attenuates uniformly; only the time-varying tone at the mode spacing can move energy between modes.

Step 2. Apply the product-to-sum identity: $\cos\omega_n t\cos\omega_m t=\tfrac12\cos\omega_{n+1}t+\tfrac12\cos\omega_{n-1}t$ , generating sidebands at $\omega_n\pm\omega_m$ . Why: since $\omega_m$ equals the mode spacing, those sidebands land exactly on the neighbouring cavity modes $n\pm1$ — nowhere else would survive.

Step 3. Read this as injection: each mode receives $E_{n\pm1}\to E_{n\pm1}+\tfrac{M}{2}E_n$ , phase-coherent light from its neighbour. Why: coherent injection synchronises coupled oscillators; pass after pass it drags every phase into a fixed relationship — the modes lock.

Step 4. Note the time-domain dual: the same modulator imposes a periodic loss window, and only light crossing at the transmission maximum ( $\omega_m t=2\pi m$ ) survives, so the field self-organises into a pulse synchronised to the drive. Why: the sideband and shutter pictures are Fourier transforms of each other — one process, two views.

Go deeper · Derivation C · 1 minThe fast saturable absorber and the Haus balance

Derivation CThe fast saturable absorber and the Haus balance

The goal is to see why the passive scheme keeps shortening the pulse where the active scheme gives up, and why the steady state is a hyperbolic secant. A fast absorber removes loss in proportion to instantaneous intensity, so it sharpens the pulse and broadens its spectrum; the finite gain bandwidth does the reverse, filtering the spectral wings and broadening the pulse in time. Steady state balances the two — and because the absorber term grows with intensity, that balance keeps tightening as the pulse intensifies, down to the transform limit.

Step 1. Expand the fast absorber loss $q(I)=q_0/(1+I/I_s)\approx q_0(1-I/I_s)$ for $I\ll I_s$ , so the envelope is reshaped as $A\to A(1-q_0+\gamma|A|^2)$ with $\gamma=q_0/I_s$ . Why: the intense centre is favoured over the dim wings — self-amplitude modulation, a term that shaves the pulse and, by Fourier duality, adds spectrum.

Step 2. Add the gain-bandwidth filter $A\to A+(g/\Omega_g^2)\,\partial_t^2 A$ , which attenuates the temporal curvature and broadens the pulse. Why: no amplifier has infinite bandwidth, so the spectral edges are trimmed — a diffusion-like broadening that opposes the absorber.

Step 3. Set the round-trip change to zero: $\gamma|A|^2A+(g/\Omega_g^2)\partial_t^2A=(l-g)A$ . The fixed point is $A_0\,\mathrm{sech}(t/\tau_p)$ ; adding anomalous $D''$ and Kerr SPM $\delta|A|^2$ promotes it to the cavity soliton of equation (6). Why: sharpening against filtering has a self-similar sech solution, a soliton once nonlinear phase and dispersion balance — a dissipative Ginzburg–Landau equation.

Step 4. Observe the scaling: since the shaping term is $\propto|A|^2$ , it strengthens as the pulse shortens, staying effective at femtoseconds. Why: this is why passive beats active — the modulator's curvature is fixed by a slow clock (the fourth-root law), whereas the absorber's is set by the pulse itself.

Limiting cases

Case	Result	Meaning
Active AM, weak modulation	$\tau_p \propto \left(\dfrac{g}{M\,\Delta\nu_g^2\,\omega_m^2}\right)^{1/4}$	Gaussian pulse; width scales only as the fourth root of gain bandwidth and modulation depth, so active mode-locking yields picoseconds, not femtoseconds.
Passive fast absorber, with dispersion	$\tau_p \approx \dfrac{2\lvert D''\rvert}{\delta E_p}$	Soliton-like sech pulse; width set by dispersion over self-amplitude-modulation coefficient times pulse energy, $\propto 1/E_p$ — pump harder, get shorter.
Bandwidth (Fourier) limit	$\tau_p\,\Delta\nu_{\mathrm{osc}} \gtrsim K$	Absolute floor: no pulse beats the inverse of its phase-locked bandwidth. $K=0.441$ (Gaussian) or $0.315$ (sech²).
Slow vs fast absorber	$\tau_{\mathrm{abs}} \gg \tau_p$ or $\tau_{\mathrm{abs}} \lesssim \tau_p$	A slow absorber still locks if gain saturation closes the trailing window (dye, SESAM lasers); a fast absorber shapes both edges directly.
Random phases (no locking)	peak $\sim\!\sqrt{N}$ , noisy	Many modes without a phase relationship give speckle, not a pulse — locking, not multimode operation, is what matters.

Common trapHaving many modes oscillate already makes a laser emit short pulses

It does not, and this is the most common misconception. A laser oscillating on hundreds of modes with random, drifting phases emits a noisy, quasi-continuous beam wandering at the $\sqrt{N}$ level — not a clean pulse train. Only when the phases are locked do the modes add coherently into sharp pulses of peak power $\sim N$ times the average; multimode operation is necessary but nowhere near sufficient. A related trap is confusing mode-locking with Q-switching: Q-switching dumps the stored inversion into one giant nanosecond pulse per switching event, whereas mode-locking gives a continuous train of picosecond-to-femtosecond pulses at the round-trip rate with the gain only lightly saturated. The two can even run together, but they are different physics.

Worked examples

1 · A Ti:sapphire femtosecond oscillator (confidence builder)

A Kerr-lens mode-locked Ti:sapphire laser has a linear cavity of optical length $L=1.8$ m and gain bandwidth $\Delta\nu_g\approx100$ THz at 800 nm. Find (a) the repetition rate and spacing, (b) how many longitudinal modes lock, and (c) the shortest transform-limited sech² pulse, compared to one optical cycle.

Step 1. Repetition rate from equation (7): $f_{\mathrm{rep}}=c/2L=(2.998\times10^8)/(3.6)=83.3$ MHz, so pulses are spaced by the round-trip time $T=1/f_{\mathrm{rep}}=12.0$ ns — also the cavity free spectral range.

Step 2. Locked modes = gain bandwidth over mode spacing: $N+1\approx\Delta\nu_g/f_{\mathrm{rep}}=(1.0\times10^{14})/(8.33\times10^7)\approx\mathbf{1.2\times10^6}$ modes — over a million lines beat together per pulse.

Step 3. Shortest sech² pulse from equation (4): $\Delta t=K/\Delta\nu_g=0.315/(1.0\times10^{14})=\mathbf{3.15\ fs}$ . One optical cycle at 800 nm lasts $\lambda/c=2.67$ fs, so this is only about 1.2 optical cycles — essentially the single-cycle limit, which is why real Ti:sapphire oscillators bottom out near 5 fs with careful dispersion control. f_rep = 83.3 MHz, T = 12 ns, N ≈ 1.2 million, τ_min ≈ 3.15 fs.

2 · Peak-power enhancement in a mode-locked Nd:YAG

A cw Nd:YAG laser (gain bandwidth $\Delta\nu_g\approx120$ GHz at 1064 nm) is mode-locked in an $L=1.5$ m cavity, emitting 1 W average. Find the repetition rate, number of locked modes, transform-limited duration, and peak power.

Step 1. Repetition rate: $f_{\mathrm{rep}}=c/2L=(2.998\times10^8)/(3.0)=1.00\times10^8$ Hz $=100$ MHz, spacing $T=10$ ns.

Step 2. Locked modes: $N+1\approx\Delta\nu_g/f_{\mathrm{rep}}=(1.2\times10^{11})/(1.0\times10^8)=\mathbf{1200}$ modes. The narrow Nd:YAG gain, a thousand times narrower than Ti:sapphire, keeps the pulses long.

Step 3. Transform-limited duration: $\tau_p\approx1/\Delta\nu_{\mathrm{osc}}=1/(1.2\times10^{11})=\mathbf{8.3\ ps}$ — picosecond-class, as expected for a narrow-band medium.

Step 4. Peak power. Mode-locking leaves the average power unchanged, redistributing it into pulses of duty cycle $\tau_p/T$ : $P_{\mathrm{peak}}=P_{\mathrm{avg}}\,T/\tau_p=P_{\mathrm{avg}}(N+1)=1\ \text{W}\times1200=\mathbf{1.2\ kW}$ . Coherence alone turns a 1 W beam into kilowatt spikes. f_rep = 100 MHz, N ≈ 1200, τ_p ≈ 8.3 ps, P_peak ≈ 1.2 kW.

3 · Why active mode-locking cannot reach the bandwidth limit

Mode-lock the same Nd:YAG ( $\Delta\nu_g=120$ GHz) actively at $f_m=100$ MHz. Use the Kuizenga–Siegman result $\tau_p\approx0.45\,(g/M)^{1/4}/\sqrt{f_m\,\Delta\nu_g}$ (unity gain-to-modulation ratio) and compare with the 8.3 ps transform limit above.

Step 1. With $\sqrt{f_m\,\Delta\nu_g}=\sqrt{(1.0\times10^8)(1.2\times10^{11})}=3.46\times10^{9}\ \mathrm{s}^{-1}$ and $(g/M)^{1/4}\approx1$ , $\tau_p\approx0.45/(3.46\times10^{9})=\mathbf{130\ ps}$ — about 16× above the 8.3 ps bandwidth floor.

Step 2. That gap is the fourth-root penalty of equation (5): the shaping curvature is set by the slow 100 MHz drive, not the pulse, so active mode-locking cannot squeeze to the Fourier limit, though a fast absorber in the same medium would. Active AM gives ≈130 ps versus an 8.3 ps limit — a 16× shortfall.

Physics of the portrait

Mode-locking has many parents — the experimentalists who first coaxed pulse trains from He–Ne, ruby, and dye lasers — but one theorist reduced the whole zoo to a single, solvable equation.

Origins · 1 minHermann A. Haus and how this was discovered

Hermann A. Haus

1925–2003 · master-equation theory of laser mode-locking at MIT

Hermann Anton Haus was born in Ljubljana in 1925, took his PhD at MIT in 1954, and remained there as an Institute Professor until his death in 2003. He began in microwave electronics and the quantum theory of noise, work that sharpened his instinct for the ultimate limits of a device. In the mid-1970s he wrote the papers that define this subject — a trio on mode-locking with fast and slow saturable absorbers and with an active modulator — showing that the round-trip evolution of the pulse envelope obeys a single master equation, a driven, dissipative relative of the nonlinear Schrödinger equation. From it fall the Gaussian pulse of active AM, the sech soliton of the fast absorber, and the scaling laws every ultrafast laser obeys. With Erich Ippen and others he went on to additive-pulse and Kerr-lens mode-locking theory, the soliton laser, and fibre squeezing. Generations of femtosecond and comb lasers were designed with equations bearing his name.

History

1964

— Hargrove, Fork, and Pollack mode-lock a He–Ne laser with an intracavity acousto-optic modulator (active)

1965

— Mocker and Collins passively mode-lock a ruby laser with a saturable absorber dye

1972

— Ippen, Shank, and Dienes reach continuous-wave passive mode-locking of a dye laser, breaking the picosecond barrier

1975

— Haus publishes the master-equation theory unifying active and saturable-absorber mode-locking

1991

— Spence, Kean, and Sibbett discover self-starting Kerr-lens mode-locking in Ti:sapphire, sub-100 fs

Going further

Texts. Saleh and Teich, Fundamentals of Photonics, 2nd ed., Ch. 15–16 · Demtröder, Laser Spectroscopy, Vol. 1, Ch. 6 · Siegman, Lasers, Ch. 27–28 (Kuizenga–Siegman theory) · Loudon, The Quantum Theory of Light, 3rd ed. · Hecht, Optics, 5th ed., Ch. 13 · Foot, Atomic Physics, Ch. 7.

Provenance. The mode superposition, Dirichlet-kernel pulse train, peak-to-average enhancement, and time–bandwidth products are standard (Saleh and Teich Ch. 15–16; Demtröder Ch. 6; Hecht Ch. 13). The master equations — the active-AM Gaussian and the fast-absorber sech soliton — follow H. A. Haus, "Mode-locking of lasers," IEEE J. Sel. Top. Quantum Electron. 6, 1173 (2000), and his 1975 papers; the practical active-AM pulse width is the Kuizenga–Siegman result (IEEE J. Quantum Electron. 6, 694 (1970)). A verifier should confirm the $(N+1)^2$ peak and $\tau_p=T/(N+1)$ limits; $f_{\mathrm{rep}}=c/2L$ (100 MHz at 1.5 m, 83.3 MHz at 1.8 m); the constants $0.441$ / $0.315$ ; the Ti:sapphire estimate $3.15$ fs versus a $2.67$ fs cycle; the Nd:YAG values $N+1=1200$ , $\tau_p=8.3$ ps, $P_{\mathrm{peak}}=1.2$ kW; and the Kuizenga–Siegman $\tau_p\approx130$ ps. No non-standard numbers are asserted.

Prerequisites. Longitudinal Modes and Free Spectral Range · Optical Resonators and Cavity Modes · Population Inversion and Optical Gain · Cavity Q Factor and Photon Lifetime. Leads to. Optical Frequency Combs · Kerr-Lens Mode-Locking · Optical Solitons and Dispersion Management · Gaussian Beams.

Ask the tutor: Ask me to walk through how the Haus master equation collapses to a Gaussian for an active modulator but a sech soliton for a fast absorber — and what that means for your favourite laser.

Isn't a laser with lots of modes already emitting pulses?

No — this is the heart of the subject. A free-running multimode laser has many modes oscillating at once, but their phases drift independently, so at any instant they add to a random value; the output is a noisy, quasi-continuous beam flickering at the $\sqrt{N}$ level, not a clean train. Mode-locking adds the fixed phase relationship: once every mode shares a common phase, they add coherently into a sharp spike that recurs once per round trip. Multimode operation is necessary but not sufficient — the pulses come from phase coherence, not from the mere number of modes.

Why do the pulses always come out spaced by the round-trip time?

Because the pulse is a real object bouncing between the mirrors: it hits the output coupler, sends out a copy, travels down and back, and hits again one round trip later, so the spacing is $T=2L/c$ . In the frequency domain the same fact is the mode spacing: the locked modes sit a free spectral range $\Delta\nu=c/2L$ apart, and a sum of tones spaced by $\Delta\nu$ is periodic with period $1/\Delta\nu=2L/c$ . The circulating pulse and the FSR-spaced comb are Fourier duals — which is why an active modulator must be driven at exactly $c/2L$ or a harmonic.

What ultimately sets the shortest pulse a laser can make?

Three things in a hierarchy. The absolute floor is the gain bandwidth: through $\Delta t\,\Delta\nu\ge K$ a pulse cannot beat the inverse of its spectrum, so only broad-gain media (Ti:sapphire, dyes, fibres) reach femtoseconds while narrow-band Nd:YAG stalls at picoseconds. Next is the shaping mechanism: an active modulator only reaches the fourth-root-limited picosecond regime, whereas a fast absorber pushes toward the bandwidth floor. Finally, at the few-femtosecond level dispersion decides everything — unless the group-delay dispersion is compensated so all colours recirculate in step, the pulse re-broadens and never reaches its transform limit.

What is the difference between a fast and a slow saturable absorber?

It is the ratio of recovery time to pulse duration. A fast absorber recovers in a time short compared with the pulse ( $\tau_{\mathrm{abs}}\lesssim\tau_p$ ), so it shaves both wings directly — the Kerr-lens "absorber" is the extreme, effectively instantaneous case. A slow absorber recovers much more slowly, opening the leading edge but unable to re-close behind the pulse on its own. Mode-locking still works because dynamic gain saturation closes the trailing window: the pulse depletes the gain, so light arriving just after the peak sees net loss. This cooperation is how many dye and SESAM lasers lock.

Why does dispersion matter so much for the shortest pulses?

A femtosecond pulse spans a huge optical bandwidth, and in any real medium different colours travel at different speeds — group-delay dispersion. Left uncorrected, the red and blue parts drift apart each round trip, the pulse chirps and stretches, and never reaches its transform limit. The fix is to add negative (anomalous) dispersion with prism pairs or chirped mirrors, tuned to cancel the positive dispersion of the gain crystal and the Kerr self-phase modulation. When they balance, the pulse circulates as a soliton — a self-consistent sech shape that reproduces itself every pass, exactly what stabilises the fast-absorber sech solution.

What is the carrier-envelope offset frequency, and why does it appear?

Inside the pulse the field carrier oscillates at the optical frequency while the envelope bounces at the round-trip rate. The carrier moves at the phase velocity and the envelope at the group velocity, and in a dispersive cavity these differ, so each successive pulse has its carrier shifted slightly under its envelope. In the frequency domain that slippage bodily offsets the whole comb: the lines sit at $\nu_n=n\,f_{\mathrm{rep}}+f_{\mathrm{ceo}}$ rather than at exact multiples of $f_{\mathrm{rep}}$ . Measuring and stabilising $f_{\mathrm{ceo}}$ (usually by $f$ – $2f$ self-referencing) is the final step that turns the laser into a metrology-grade comb, pinning every optical line to a countable frequency.