Intra-Cavity Squeezing: Principles & Applications

• Intra-cavity squeezing (ICS) is the internal generation and transformation of squeezed bosonic modes within a resonant cavity, enabling improved quantum sensing and enhanced light-matter interactions.
• ICS employs methods such as degenerate parametric amplification, Kerr nonlinearities, and dissipative coupling to reshape cavity quadratures by de-amplifying one mode while amplifying its conjugate.
• Practical implementations across optical, microwave, and hybrid systems have demonstrated benefits including a 36% gain in sensitivity-bandwidth and prospects for achieving single-photon detection precision.

Intra-cavity squeezing (ICS) denotes the generation, stabilization, or functional use of squeezed bosonic modes inside a resonant cavity rather than solely by external injection. Across optical, microwave, optomechanical, magnonic, mesoscopic-conductor, and interferometric settings, the common operation is an internal transformation that reshapes the cavity quadratures so that one quadrature is de-amplified or squeezed while its conjugate is anti-squeezed or amplified. In the literature summarized here, ICS is implemented by degenerate parametric amplification in a cavity, by Kerr- or self-Kerr-induced effective quadratic interactions, by dissipative coupling to a driven conductor, and by dynamical back-action mechanisms; it is used both as a source of nonclassical light and as a method for improving sensing, quantum non-demolition (QND) readout, cooling, and effective light-matter coupling (Korobko et al., 2017, Korobko et al., 2023, Salykina et al., 2024, Mendes et al., 2015).

1. Conceptual scope and physical mechanisms

The minimal ICS paradigm is a single cavity mode $a$ subject to an internal quadratic interaction of the form

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$

or equivalent variants written with a real parametric amplitude $\epsilon$, $\nu$, or $G$, depending on platform and notation (Korobko et al., 2023, Stefszky et al., 2017, Peano et al., 2015). In quadrature language, the canonical amplitude and phase quadratures,

$$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$

transform under the squeeze operator as

$$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$

with the pump phase selecting which quadrature is squeezed and which is anti-squeezed (Korobko et al., 2023).

This basic parametric picture is realized in several distinct ways. In $\chi^{(2)}$ optical cavities, a nonlinear crystal pumped near $2\omega_c$ implements a degenerate optical parametric amplifier (DOPA) or optical parametric amplification (OPA) stage inside the resonator (Korobko et al., 2017, Stefszky et al., 2017). In Kerr systems, an internal parametric drive can be tuned to cancel the unwanted self-phase-modulation term that ordinarily limits a QND cross-phase-modulation readout (Salykina et al., 2024). In cavity optomechanics, a two-photon drive of strength $\Lambda$ or an intracavity $$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$0 medium reshapes the optical susceptibilities and the readout noise budget (Mahana et al., 26 Jun 2026, Peano et al., 2015). In cavity magnomechanics, a strong drive and magnon self-Kerr nonlinearity generate an effective squeezing term

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$1

which is then exploited to suppress Stokes processes (Asjad et al., 2022). A qualitatively different route appears in driven mesoscopic conductors, where the conductor functions as a dissipative squeezing bath for a cavity mode, and the relevant pair-creation term is generated by non-stationary current-noise harmonics (Mendes et al., 2015).

A recurrent feature of the field is that ICS is not restricted to producing squeezed output states. It also appears as an internal resource transformation that changes the relation between signal transfer, decoherence, and vacuum admixture. This is explicit in interferometric force sensing, where internal squeezing can mitigate readout loss (Korobko et al., 2023), in cavity-enhanced interferometers where it can beat the standard sensitivity-bandwidth limit at fixed intracavity power (Korobko et al., 2017), and in the Kerr-QND setting where it removes the probe SPM penalty by satisfying the cancellation condition

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$2

for the intracavity parametric pump (Salykina et al., 2024).

2. Core theoretical structure

Despite the diversity of implementations, the technical description of ICS is highly uniform. One starts from a cavity or hybrid-system Hamiltonian, linearizes about a large coherent amplitude when appropriate, and passes to either quadrature variables or a Bogoliubov-transformed squeezed mode. In the optical-parametric setting, the Heisenberg-Langevin equation typically takes the form

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$3

together with the usual input-output relation

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$4

where the internal loss channel is carried by $$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$5 (Stefszky et al., 2017).

In a squeezed-mode description, the cavity operator is written as

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$6

or equivalently $$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$7, with $$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$8 set by a parametric-amplifier relation such as

$$H_{\rm sqz}=\frac{i\hbar}{2}\,\chi\bigl(a^{\dagger 2}-a^2\bigr),$$9

for the magnon problem, or

$\epsilon$0

for a squeezed primary cavity in coupled-cavity QED (Asjad et al., 2022, Wang et al., 2019). This transformation diagonalizes the quadratic part and exposes how the effective couplings inherit $\epsilon$1, $\epsilon$2, or $\epsilon$3 factors. In one coupled-cavity proposal, the effective atom-auxiliary-cavity interaction becomes

$\epsilon$4

showing explicit enhancement with squeezing (Wang et al., 2019).

For output-state characterization, the relevant quantities are quadrature spectra or steady-state variances. In the driven-conductor cavity problem, the steady-state quadrature variances are

$\epsilon$5

where $\epsilon$6 is the anomalous photo-assisted current-noise component responsible for pair creation (Mendes et al., 2015). In waveguide OPO realizations, the normalized squeezing spectrum is written directly in terms of pump fraction and cavity linewidth (Stefszky et al., 2017). In several sensing papers, the final observable is a signal-referred spectral density $\epsilon$7 obtained by dividing an output noise spectrum by a displacement or force transfer function (Korobko et al., 2023, Korobko et al., 2023, Mahana et al., 26 Jun 2026).

The role of loss enters these formulations in a structurally similar way. External or readout loss mixes in vacuum after the internal interaction, while intra-cavity loss injects vacuum directly into the sensing or state-preparation dynamics. Consequently, the strongest claims for ICS are almost always claims about which loss channel remains fundamental. In the interferometric analyses, optimized internal squeezing removes the dependence on readout loss to first order, leaving intra-cavity loss as the sole bound (Korobko et al., 2023, Korobko et al., 2023). In the Kerr-QND proposal, once SPM is canceled, the sensitivity is limited only by available pump power and by input/output losses in the signal beam (Salykina et al., 2024).

3. ICS in precision measurement and QND protocols

A major modern use of ICS is in cavity-enhanced sensing, where it is employed to reshape the quantum noise budget rather than merely to produce a squeezed beam. In the 2017 cavity-enhanced interferometer experiment, internal squeezing was analyzed as a way to overcome the classical sensitivity-bandwidth trade-off at fixed intracavity power. The measured result was a 36% increase in the sensitivity-bandwidth product compared to the classical case, with enhancements of 26%, 31%, 33% and a maximum 36% across four runs (Korobko et al., 2017).

The subsequent force-sensing framework with a second-order nonlinear crystal placed inside the cavity made the same point more explicitly. In the single-mode approximation, and neglecting radiation-pressure back-action, the shot-noise power spectral density and signal transfer were written as

$\epsilon$8

$\epsilon$9

with the measured displacement noise $\nu$0 (Korobko et al., 2023). At $\nu$1, the optimized noise floor is

$\nu$2

and in the limit of infinite external squeezing the ultimate intra-cavity bound reduces to

$\nu$3

so that the only remaining limiter is the intra-cavity loss (Korobko et al., 2023).

The tabletop experiment accompanying that theory reported that, across 10%, 20%, and 30% emulated readout-loss settings, the peak SNR gain relative to the no-ICS case was a constant $\nu$4. The paper interpreted this plateau as demonstrating that ICS had effectively removed the dependence on $\nu$5 from the final sensitivity (Korobko et al., 2023). A related general analysis of lossy cavity-enhanced interferometers derived a closed-form optimal internal squeeze parameter $\nu$6 and showed that, at large optimized internal squeezing, one asymptotically cancels readout loss to first order, again leaving only the fundamental intra-cavity-loss limit (Korobko et al., 2023).

Optomechanical sensing adds a counterintuitive variant: the enhancement can come from de-amplifying the signal quadrature without losing quantum information. In the position-detection proposal, the phase quadrature carrying mechanical motion is de-amplified, yet the signal-to-noise ratio improves because the noise is suppressed even more strongly (Peano et al., 2015). The same paper emphasized that this mechanism is especially useful for weak optomechanical coupling and/or strong mechanical damping, and that it can be extended straightforwardly to QND qubit detection (Peano et al., 2015).

The Kerr-QND scheme of 2024 places ICS directly inside the measurement cavity. After linearization, the unwanted SPM back-action appears as the $\nu$7 term in the probe-phase equation. Choosing the internal parametric pump such that

$\nu$8

removes that term, yielding a probe phase quadrature determined only by the desired XPM coupling, probe noise terms, and cavity response (Salykina et al., 2024). For a Gaussian signal pulse with $\nu$9, the intracavity-number imprecision variance is

$G$0

and the paper states that, using the best optical microresonators currently available, single-photon sensitivity for the intracavity photon number can be achieved (Salykina et al., 2024).

4. Realizations across platforms

ICS has been realized or proposed in a wide range of experimental architectures. The implementations differ mainly in the origin of the squeezing interaction and in whether the primary objective is squeezed-state generation, sensing enhancement, cooling, or coupling enhancement.

Platform	Internal mechanism	Representative result
Waveguide resonator	$G$1 OPA in Ti:LiNbO$G$2 cavity	$G$3 dB measured, $G$4 dB produced (Stefszky et al., 2017)
Compact fibered source	Degenerate OPO in fiber-mirror cavity with PPKTP	Raw squeezing from $G$5 dB to $G$6 dB across three fiber types (Brieussel et al., 2018)
Levitated optomechanics	Coherent-scattering optomechanical back-action	About 17 dB single-mode squeezing at room temperature (Li et al., 2021)
Quantum conductor cavity	Dissipative squeezing bath from driven conductor noise	$G$7 for a tunnel junction; ideal Levitons approach perfect squeezing (Mendes et al., 2015)
Kerr-QND microresonator	Internal DOPA cancels probe SPM	Single-photon sensitivity for intracavity photon number under realistic parameters (Salykina et al., 2024)
GW signal-extraction cavity	Bidirectional internal OPA in SE cavity	Saturates the most stringent internal-loss CWB at all $G$8 (Vermeulen et al., 15 May 2026)

Integrated and cavity-enhanced $G$9 devices provide the most direct optical realization. The titanium-indiffused lithium-niobate waveguide resonator produced continuous-wave optical squeezing, with $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$0 dB of single-mode squeezing directly measured, corresponding to $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$1 dB after accounting for detection losses (Stefszky et al., 2017). The device parameters included $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$2, $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$3, $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$4, waveguide loss $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$5, and an approximate finesse $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$6 (Stefszky et al., 2017). The compact fibered sources used related OPO physics in near-hemispherical fiber-mirror cavities containing thin PPKTP crystals. There, the measured raw squeezing values were $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$7 dB, $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$8 dB, and $$X=\frac{a+a^\dagger}{\sqrt2},\qquad Y=\frac{a-a^\dagger}{i\sqrt2},$$9 dB for three fiber architectures, while the best estimated internal squeezing reached $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$0 dB under ideal detection in the tapered-fiber case (Brieussel et al., 2018).

At the opposite extreme, levitated optomechanics uses dynamical back-action rather than a nonlinear crystal as the effective squeezing resource. In the bichromatic cavity proposal with a levitated nanodiamond torsional mode, the reported realistic room-temperature parameters were $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$1, $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$2, $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$3, and $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$4, with a numerical minimum squeezing $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$5, described as about 17 dB below shot noise, over roughly $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$6 bandwidth (Li et al., 2021). The same framework also supports two-mode squeezing near the stability boundary (Li et al., 2021).

Hybrid microwave and magnonic systems use ICS differently. In cavity magnomechanics, magnon self-Kerr-induced squeezing modifies the magnomechanical interaction so that, at optimal detuning $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$7,

$$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$8

thereby exponentially enhancing anti-Stokes scattering and exponentially suppressing Stokes scattering (Asjad et al., 2022). The paper argued that this makes intracavity squeezing particularly useful in the unresolved-sideband regime and also found that coupling to the microwave cavity has only an adverse effect in mechanical cooling, so that the effectively two-mode magnomechanical system is preferred (Asjad et al., 2022).

In mesoscopic cavity QED with a driven conductor, ICS emerges as a dissipative steady state. For a tunnel junction, a train of quantized $$X_{\rm out}=X_{\rm in}e^{-r},\qquad Y_{\rm out}=Y_{\rm in}e^{+r},$$9-peaks in the bias voltage gives

$\chi^{(2)}$0

while in an asymmetric quantum dot a sharp Leviton pulse yields

$\chi^{(2)}$1

so that the ideal $\chi^{(2)}$2 limit approaches perfect cavity squeezing (Mendes et al., 2015).

Coupled-cavity and cavity-QED variants use ICS mainly as a coupling enhancer. In the high-dissipation coupled-cavity proposal, squeezing of a primary cavity that is only virtually excited enhances the effective atom-auxiliary-cavity coupling and allows observation of vacuum Rabi oscillations even when the primary cavity has $\chi^{(2)}$3, with reported simulations showing clear oscillations for $\chi^{(2)}$4 at $\chi^{(2)}$5 (Wang et al., 2019).

5. Performance bounds, trade-offs, and common misconceptions

A central misconception is that ICS is synonymous with simple internal amplification of a desired signal. The sensing literature repeatedly shows that the operative mechanism can instead be de-amplification of the signal quadrature. In the optomechanical position-detection proposal, this is stated explicitly as the counterintuitive reason the SNR improves: the quadrature sensitive to the mechanical motion is de-amplified without losing quantum information (Peano et al., 2015). In force sensing, the optimum internal gain $\chi^{(2)}$6 can be negative or positive depending on external squeezing and readout loss; with low external squeeze or low $\chi^{(2)}$7, the best SNR occurs at negative $\chi^{(2)}$8, whereas with high external squeeze or high $\chi^{(2)}$9, it occurs at positive $2\omega_c$0 (Korobko et al., 2023).

A second misconception is that ICS removes all decoherence limits. The recent interferometric analyses are explicit that it does not. Rather, optimized ICS can make the sensitivity independent of downstream readout loss over a broad regime, but the remaining fundamental limit is the intra-cavity loss (Korobko et al., 2023, Korobko et al., 2023). The 2026 gravitational-wave proposal sharpened this point by presenting bidirectional internal squeezing in the signal-extraction cavity and showing that the resulting signal-referred quantum-noise spectral density is independent of the arm-cavity input and signal-extraction transmissivities at high frequencies, while saturating the lowest known lower bounds on quantum noise from internal optical dissipation (Vermeulen et al., 15 May 2026).

The same pattern appears in the Kerr-QND problem. Previous analysis had identified the interplay of optical losses and probe self-phase modulation as the main sensitivity limiter. The 2024 ICS scheme removes the SPM term, but the remaining limits are available pump power and signal-beam input/output losses (Salykina et al., 2024). Using a CaF$2\omega_c$1 resonator with intrinsic $2\omega_c$2 and loaded $2\omega_c$3, corresponding to $2\omega_c$4, $2\omega_c$5, and $2\omega_c$6, the paper estimated

$2\omega_c$7

Reaching single-photon precision therefore requires $2\omega_c$8–$2\omega_c$9 circulating probe photons, and for the input and output errors also to approach unity at $\Lambda$0, the total signal-path loss must be $\Lambda$1 (Salykina et al., 2024).

Practical constraints recur across platforms. High internal squeezing or high circulating power can be limited by parametric threshold, pump depletion, thermal effects, photorefractive or photothermal noise, mode mismatch, and quantum efficiency of the final photodetector (Korobko et al., 2017, Brieussel et al., 2018, Salykina et al., 2024, Vermeulen et al., 15 May 2026). Several papers give concrete examples. The waveguide resonator identified waveguide propagation loss and detection efficiency as the dominant restrictions, and stated that lowering $\Lambda$2 toward $\Lambda$3 would permit more than 8 dB squeezing (Stefszky et al., 2017). The gravitational-wave proposal found that a bow-tie OPA inside the signal-extraction cavity introduces transverse-mode mismatch, but that “mode healing” in the signal-extraction cavity can suppress mismatch losses (Vermeulen et al., 15 May 2026).

6. Relations to adjacent fields and current directions

ICS now sits at the intersection of squeezed-light generation, quantum sensing, hybrid quantum systems, and QND measurement. In some subfields it functions primarily as a source technology; in others it is a control primitive that changes effective susceptibilities, couplings, or dissipation pathways. The same mathematical object—a cavity-local squeeze transformation—supports otherwise distinct goals: beating a sensitivity-bandwidth limit (Korobko et al., 2017), mitigating readout decoherence (Korobko et al., 2023), saturating internal-dissipation bounds in gravitational-wave detectors (Vermeulen et al., 15 May 2026), canceling Kerr SPM in optical QND readout (Salykina et al., 2024), suppressing Stokes heating in magnomechanics (Asjad et al., 2022), and enhancing effective strong coupling in dissipative cavity QED (Wang et al., 2019).

A plausible implication is that the most consequential distinction in current ICS research is no longer between “source” and “sensor,” but between schemes that merely redistribute external loss and schemes that move the fundamental bound to a deeper internal channel. The interferometric and gravitational-wave results are the clearest examples of the latter (Korobko et al., 2023, Vermeulen et al., 15 May 2026). By contrast, integrated waveguide and fibered OPO sources remain primarily constrained by escape efficiency, propagation loss, and detection efficiency, even when their internal squeezing dynamics are well understood (Stefszky et al., 2017, Brieussel et al., 2018).

The recent Kerr-QND work illustrates a complementary trend: ICS is being embedded into nonlinear quantum-information architectures not only to generate squeezing, but to neutralize a specific nonlinear back-action term. In that scheme, once the probe SPM is canceled by matching the internal parametric drive to the Kerr coefficient, the precision scales fundamentally with probe photon number and is limited in practice only by power and loss. The paper’s conclusion that single-photon precision of the intracavity signal number is within reach under currently realistic parameters makes ICS relevant to optical quantum information processing tasks in a direct operational sense (Salykina et al., 2024).

Taken together, the literature defines ICS as a family of internal cavity transformations whose significance depends on where the dominant vacuum admixture and dynamical back-action enter the problem. When that structure is favorable, ICS can generate strongly squeezed states, improve measurement precision in a loss-resilient manner, stabilize nonclassical steady states, or re-engineer hybrid-system interactions well beyond the naive unsqueezed limit (Mendes et al., 2015, Korobko et al., 2017, Korobko et al., 2023, Salykina et al., 2024).