Bad-Cavity Laser: Concepts & Applications

• Bad-cavity lasers are defined by deliberately broad cavity linewidths compared to the gain bandwidth, which shifts frequency determination from the cavity to the active medium.
• They suppress cavity pulling and technical noise by weakening the cavity’s role in frequency selection, thereby enhancing stability and precision in applications like optical clocks and spectroscopy.
• Experimental implementations across various platforms demonstrate significant linewidth reduction and robust performance even under low-feedback conditions.

A bad-cavity laser is a laser whose cavity linewidth is intentionally made much broader than the gain bandwidth or atomic coherence linewidth, so that the emitted frequency is determined primarily by the active medium rather than by the cavity resonance. In this regime, the cavity remains an essential emission channel and feedback element, but its role in frequency selection is weakened, which suppresses the transfer of cavity-length noise into the output. For that reason, bad-cavity lasers occupy a central position in active optical clocks, superradiant and Raman laser schemes, precision spectroscopy, and frequency metrology; recent work has pushed this logic to the limit of an “extremely bad-cavity laser” with finesse $2.01$, close to the mirrorless bound (Zhang et al., 2023).

1. Defining the bad-cavity regime

The defining hierarchy is that the cavity mode is broader and faster than the gain medium. In the language used for continuous-wave cesium lasing, the regime is specified by

$$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$

with $\Gamma_{\mathrm{cavity}}$ the cavity-mode linewidth and $\Gamma_{\mathrm{gain}}$ the atomic gain bandwidth. In the dual-wavelength active-optical-clock literature, the same idea is expressed through the bad-cavity coefficient

$a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$

so that larger $a$ corresponds to stronger bad-cavity behavior. In superradiant Raman implementations the equivalent statement is $\kappa \gg \gamma_\perp$, where $\kappa$ is the cavity power decay rate and $\gamma_\perp$ the atomic polarization dephasing rate; in that limit the cavity field can often be adiabatically eliminated, and the collective atomic Bloch vector becomes the slow dynamical object (Zhang et al., 2023, Shi et al., 2019, Bohnet et al., 2012).

The essential physical consequence is that the phase memory of the laser is stored predominantly in the atoms rather than in the cavity. This distinguishes bad-cavity lasers from conventional good-cavity lasers, where the narrow cavity resonance is the primary spectral filter. The contrast is explicit in the literature: ultrastable passive lasers based on Pound–Drever–Hall locking seek high finesse and narrow cavity resonances, whereas bad-cavity lasers seek broad, lossy resonances because broadening the cavity suppresses cavity pulling and makes the output more atomic-like (Zhang et al., 2023).

A bad cavity is therefore not merely a poor resonator in an engineering sense. It is a deliberate operating point in which the cavity still provides optical feedback, directional emission, and mode selection, but does so weakly enough that cavity-length fluctuations, thermal expansion, vibration, and drift are not efficiently transferred to the lasing frequency. This operating principle underlies the appeal of bad-cavity lasers as active optical frequency references (Zhang et al., 2023, Kazakov et al., 2016).

2. Frequency setting, cavity pulling, and linewidth

The standard description of frequency selection in a bad-cavity laser is the cavity-pulling relation

$\nu_0 - \nu_a = P(\nu_c - \nu_a),$

where $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$0 is the laser frequency, $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$1 the atomic transition frequency, $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$2 the cavity resonance, and $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$3 the cavity-pulling coefficient. Small $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$4 means that the laser frequency follows the cavity only weakly. In the experimentally realized extremely bad-cavity cesium laser, the cavity reflectance was reduced to $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$5, corresponding to finesse $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$6, with cavity length $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$7 cm and $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$8 MHz. The gain bandwidth at the operating point was estimated as $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$9. Even with such weak feedback, the device produced continuous-wave output powers of tens of $\Gamma_{\mathrm{cavity}}$0W, a linewidth of about $\Gamma_{\mathrm{cavity}}$1 kHz, and—after common-mode technical-noise rejection by beating two lasing transitions sharing the same cavity—an inferred intrinsic linewidth of $\Gamma_{\mathrm{cavity}}$2 Hz per mode from a measured beat linewidth of $\Gamma_{\mathrm{cavity}}$3 Hz. The corresponding mirrorless superradiant or ASE-like emission linewidth was about $\Gamma_{\mathrm{cavity}}$4 kHz. The cavity-pulling coefficient was measured as $\Gamma_{\mathrm{cavity}}$5, described there as the lowest value achieved for a continuous-wave laser (Zhang et al., 2023).

That result clarifies a recurrent misunderstanding: vanishing feedback is not synonymous with narrow output. In the cited cesium experiment, the mirrorless configuration was substantially broader than the extremely bad cavity. Weak feedback can enhance coherence while still leaving the frequency substantially immune to cavity fluctuations. A plausible implication is that the practically useful regime is often not the absence of a cavity, but a cavity near the limit where it remains spectrally helpful without strongly imposing its own resonance (Zhang et al., 2023).

Linewidth theory in the bad-cavity regime is correspondingly nontrivial. In a general $\Gamma_{\mathrm{cavity}}$6-matrix treatment of open lasers, material dispersion produces a bad-cavity linewidth reduction that can yield sub-Schawlow–Townes linewidths. For a uniform Fabry–Perot cavity, the familiar factorized form

$\Gamma_{\mathrm{cavity}}$7

separates a bad-cavity factor $\Gamma_{\mathrm{cavity}}$8, a Petermann factor $\Gamma_{\mathrm{cavity}}$9, and a passive-cavity decay scale $\Gamma_{\mathrm{gain}}$0; however, when line pulling is strong or spatial hole burning is appreciable, that factorization breaks down and bad-cavity and Petermann corrections are no longer independent (Pillay et al., 2014).

The prospect of linewidth reduction far below the natural atomic linewidth appears also in the proposed Ramsey laser with bad cavity. There the cavity is broad, while atoms experience two separated coherent interaction regions. The predicted stimulated-emission linewidth is below $\Gamma_{\mathrm{gain}}$1 Hz for a $\Gamma_{\mathrm{gain}}$2 example with natural linewidth about $\Gamma_{\mathrm{gain}}$3 Hz, that is, more than two orders of magnitude narrower than the atomic natural linewidth. In that proposal, the narrowing arises from the combination of Ramsey interference, stimulated emission, and suppression of cavity-related noise to a cavity-pulling effect (Li et al., 2010).

3. Experimental realizations and platform diversity

Bad-cavity laser physics spans continuous-wave, pulsed, Raman, and active-clock architectures, and the cavity need not be high finesse in the conventional metrological sense. The systems explicitly documented in the literature illustrate a wide range of operating points.

Platform	Regime metric	Reported result
Thermal Cs vapor (Zhang et al., 2023)	$\Gamma_{\mathrm{gain}}$4, $\Gamma_{\mathrm{gain}}$5	Tens of $\Gamma_{\mathrm{gain}}$6W, $\Gamma_{\mathrm{gain}}$7 kHz linewidth, $\Gamma_{\mathrm{gain}}$8
1064/1470 nm dual-wavelength Cs active optical clock (Shi et al., 2019)	$\Gamma_{\mathrm{gain}}$9 MHz, $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$0 MHz	1470 nm linewidth reduced to $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$1 Hz after phase locking
$a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$2Rb superradiant Raman laser (Bohnet et al., 2012)	$a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$3, $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$4	Direct observation of relaxation oscillations and cavity feedback
Proposed $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$5 Coulomb-crystal laser (Kazakov et al., 2017)	Forbidden transition in bad cavity	Output near $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$6 pW, linewidth $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$7 mHz
Ramsey laser proposal (Li et al., 2010)	$a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$8	Predicted linewidth below $a=\frac{\Gamma_{\rm cavity}}{\Gamma_{\rm gain}},$9 Hz

The cavity need not be conventionally “good” even when it enhances atom–light coupling. In low-finesse strontium cavity-QED stabilization experiments, $a$0 with $a$1 MHz still satisfied the bad-cavity condition relative to the $a$2 kHz intercombination linewidth, while a higher-finesse but still bad-cavity Sr system used $a$3 and $a$4 kHz against the same $a$5 kHz natural linewidth. These systems were not themselves lasing demonstrations, but they show how bad-cavity references can produce steep dispersive signals and projected stabilized linewidths at the $a$6 mHz and $a$7 mHz levels, respectively (Christensen et al., 2015, Schäffer et al., 2017).

At the opposite scale, the distinction between bad-cavity and bad-emitter regimes has been made explicit in a single-emitter SnV$a$8 microcavity system. That work did not demonstrate lasing, but it showed a transition from $a$9 at around $\kappa \gg \gamma_\perp$0 K to $\kappa \gg \gamma_\perp$1 at $\kappa \gg \gamma_\perp$2 K, together with a measured Purcell factor exceeding $\kappa \gg \gamma_\perp$3. This is relevant because it separates the condition for cavity-dominated emission from the separate requirement of a gain medium capable of lasing (Sachero et al., 9 Jul 2025).

4. Collective dynamics, stability, and many-body structure

Bad-cavity lasers differ dynamically from ordinary good-cavity lasers because the slow oscillator is often atomic rather than photonic. In the $\kappa \gg \gamma_\perp$4Rb superradiant Raman laser, the cavity field is adiabatically eliminated and the dynamics are described by a collective atomic Bloch vector with inversion $\kappa \gg \gamma_\perp$5 and coherence $\kappa \gg \gamma_\perp$6. The measured emitted intensity exhibits spiking and relaxation oscillations, and small-signal modulation of the repumping rate reveals transfer functions for the optical amplitude and phase. In the idealized three-level model at zero cavity detuning, the damping and natural frequency are

$\kappa \gg \gamma_\perp$7

The same study shows that intermediate repumping states reduce damping, lower the relaxation frequency, and can destabilize the oscillator, while dispersive cavity tuning can provide either negative feedback that suppresses oscillations or positive feedback that enhances them (Bohnet et al., 2012).

The role of gain inhomogeneity is a second major theme. For inhomogeneously broadened spin-$\kappa \gg \gamma_\perp$8 bad-cavity lasers relevant to active optical frequency standards, steady-state centered solutions, detuned solutions, and zero-field solutions can all occur. The cited stability analysis introduced an efficient method based on the block-arrowhead structure of the linearized equations and an argument-principle treatment of a rational characteristic function, rather than direct diagonalization of a very large matrix. The resulting phase diagrams show that modest inhomogeneous broadening and Zeeman splitting have little effect if they remain well below the pumping rate, but larger broadening reduces output power, shifts thresholds, and can produce unstable regimes with pulsation or irregular field dynamics; in that model, increasing the repumping rate can restore stability because it also increases the homogeneous decoherence rate (Kazakov et al., 2016).

Bad-cavity lasers also constitute driven-dissipative many-body systems with nontrivial phase structure. A model of $\kappa \gg \gamma_\perp$9 identical two-level atoms with incoherent repumping, spontaneous decay, and a collective cavity-mediated decay channel predicts that the onset of steady-state superradiance is preceded by a dissipative transition between two distinct subradiant phases at the critical repump rate $\kappa$0. Below the critical point the cavity output power is strongly suppressed and does not increase with atom number at leading order; above it, the output scales linearly with atom number. The same study reports a discontinuity in the normalized variance of the collective inversion and a sharp drop of the generalized spin-squeezing witness $\kappa$1, implying a macroscopically entangled steady state near criticality (Shankar et al., 2021).

In the extreme bad-cavity or spaser limit, the many-emitter cavity problem has also been solved with a numerically exact Lindblad treatment that uses emitter permutation symmetry and scales as the third power in emitter number. For realistic spaser parameters the average output curve is thresholdless, so coherence must be identified through full number statistics and $\kappa$2 rather than through a sharp intensity threshold. In that treatment, coherent emission corresponds to a transition from thermal-like to Poisson-like plasmon statistics, while the cavity or plasmon loss rate remains so large that the system sits in a very bad-cavity regime (Richter et al., 2014).

5. Metrological architectures and stabilization strategies

A major branch of bad-cavity-laser research concerns active optical clocks. In the 1064/1470 nm dual-wavelength cesium active optical clock, the $\kappa$3 nm field is the bad-cavity clock-transition laser with $\kappa$4 MHz and $\kappa$5 MHz, implying that cavity noise is reduced by about a factor of $\kappa$6 relative to a good-cavity laser. The practical limitation there was residual cavity pulling caused by asynchronous cavity-length variation between two independent systems. By phase-locking the two independent $\kappa$7 nm good-cavity lasers with an optical phase-locked loop, the cavity lengths were made to evolve synchronously. The reported tracking accuracy between the two main cavities was better than $\kappa$8 at $\kappa$9 s and about $\gamma_\perp$0 at $\gamma_\perp$1 s, while the most probable linewidth of each $\gamma_\perp$2 nm bad-cavity laser decreased from $\gamma_\perp$3 Hz to $\gamma_\perp$4 Hz (Shi et al., 2019).

Cold-atom strontium systems have developed the complementary strategy of using a bad-cavity atom–cavity resonance as a discriminator rather than as a laser oscillator. In one case, NICE-OHMS-style heterodyne detection on the $\gamma_\perp$5 line produced a steep dispersive phase response in a cavity with $\gamma_\perp$6 kHz and suppression of cavity fluctuations by approximately $\gamma_\perp$7; the projected minimum shot-noise-limited linewidth was about $\gamma_\perp$8 mHz (Schäffer et al., 2017). In a related low-finesse Sr system, simultaneous measurement of phase and absorption through the complex cavity transmission coefficient showed a strongly nonlinear power dependence of the phase-dispersion slope and an estimated shot-noise-limited linewidth of about $\gamma_\perp$9 mHz at optimal operating conditions (Christensen et al., 2015).

Pulsed lasing in the superradiant crossover regime adds a further nuance. On the $\nu_0 - \nu_a = P(\nu_c - \nu_a),$0Sr $\nu_0 - \nu_a = P(\nu_c - \nu_a),$1 line in a thermal ensemble at about $\nu_0 - \nu_a = P(\nu_c - \nu_a),$2 mK, the Doppler-broadened linewidth $\nu_0 - \nu_a = P(\nu_c - \nu_a),$3 MHz exceeded the cavity linewidth $\nu_0 - \nu_a = P(\nu_c - \nu_a),$4 kHz. In that regime, different atomic velocity classes alternately emit into and reabsorb from the cavity, and the spectral peak can become first-order insensitive to cavity detuning near $\nu_0 - \nu_a = P(\nu_c - \nu_a),$5 kHz even though the system is not deeply in the bad-cavity limit. The paper distinguishes carefully between the peak frequency and the spectrum’s center of mass: the former can be cavity-immune while the latter remains cavity-dependent (Tang et al., 2021).

6. Applications, distinctions, and recurrent misconceptions

The most direct applications identified in the literature are compact active optical clocks, ultra-stable portable lasers that do not require ultrahigh-finesse cavities, improved cavity-QED platforms, continuous-wave superradiant lasers, and explorations of quantum many-body physics. The extremely bad-cavity cesium experiment presents these as consequences of operating near the $\nu_0 - \nu_a = P(\nu_c - \nu_a),$6 limit, where the resonator supplies only minimal optical feedback while still supporting narrow continuous-wave emission (Zhang et al., 2023).

Large trapped-ion ensembles define a distinct route to active optical frequency standards. A proposed bad-cavity laser based on a $\nu_0 - \nu_a = P(\nu_c - \nu_a),$7 Coulomb crystal uses the forbidden $\nu_0 - \nu_a = P(\nu_c - \nu_a),$8 transition with spontaneous rate $\nu_0 - \nu_a = P(\nu_c - \nu_a),$9. In that proposal, micromotion-induced broadening can be suppressed by operating the Paul trap at a magic frequency $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$00, provided $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$01. For a spherical crystal with $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$02 ions and cavity finesse $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$03, the estimated output power is near $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$04 pW and the linewidth is $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$05 mHz, with output power scaling approximately as $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$06 at fixed optimum waist ratio (Kazakov et al., 2017).

Several distinctions are important. First, bad-cavity operation does not mean absence of a cavity; the cavity remains necessary for optical feedback, mode definition, and directed emission. Second, bad-cavity operation does not eliminate all cavity effects. Residual cavity pulling, cavity-feedback-induced relaxation oscillations, and inhomogeneous-broadening-driven instabilities remain central design constraints (Shi et al., 2019, Bohnet et al., 2012, Kazakov et al., 2016). Third, bad-cavity and bad-emitter are different limits. The SnV$$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$07 microcavity experiment shows that $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$08 describes a bad emitter, whereas $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$09 describes a bad cavity; neither condition by itself implies lasing, but both are relevant to the broader cavity-QED landscape in which bad-cavity lasers sit (Sachero et al., 9 Jul 2025).

A final misconception concerns depth of the regime. Deep bad-cavity operation is not the only route to cavity-immune behavior: the pulsed thermal Sr system shows first-order suppression of cavity sensitivity through velocity-class dynamics even outside the deep limit (Tang et al., 2021). Conversely, extremely weak feedback is not automatically beneficial unless coherence can still be supported; the mirrorless cesium case is broader than the extremely bad-cavity laser with $$\Gamma_{\mathrm{cavity}} \gg \Gamma_{\mathrm{gain}},$$10 (Zhang et al., 2023). The field therefore spans not one regime but a continuum between conventional cavity-stabilized lasers, superradiant and Raman oscillators, and near-mirrorless active optical references, unified by the principle that atomic coherence, rather than cavity storage time, sets the frequency reference.