---
title: Coherent Feedback Cooling in Optomechanics
url: https://www.emergentmind.com/topics/coherent-feedback-cooling-cfc
type: topic
---

# Coherent Feedback Cooling in Optomechanics

Coherent Feedback Cooling (CFC) is a class of feedback protocols in which the field carrying information about a target degree of freedom is coherently processed and fed back without any photodetection, so the loop remains measurement-free and the correlations between the system motion and the feedback signal are preserved [2210.07674, 2506.21341]. In optomechanics and related hybrid platforms, the feedback phase, delay, and transmission determine whether the returned signal acts primarily as a stiffness shift, a viscous damping channel, or a more general engineered reservoir. Recent realizations span levitated nanoparticles, cavity membranes, ultracoherent phononic-crystal membranes, and spin-mechanical hybrids, with reported cooling to \(n\approx344\) phonons at \(3\times10^{-7}\,\mathrm{mbar}\) without cryogenics, \(n_f=166\pm7\) at room temperature, and \(\bar n_m=4.89\pm0.14\) phonons in a \(20\,\mathrm{K}\) environment [2506.21341, 2605.20902, 2210.07674].

## 1. Conceptual basis and historical placement

CFC differs from standard feedback in that it does not first convert the system state into a classical measurement record. Instead, the output field is routed through a coherent controller or delay network and then reinjected, so both non-commuting quadratures can participate in the loop without the irreversible state modification associated with measurement [2210.07674, 1206.0829]. In the Linear Quadratic Gaussian setting, early analyses showed that coherent control schemes, in which the resonator is embedded in an interferometer to achieve all-optical feedback, can outperform optimal measurement-based feedback control schemes in the quantum regime of low steady-state excitation number [1206.2688, 1206.0829].

Within quantum optomechanics, resolved-sideband cooling was explicitly identified as an example of coherent feedback control, and analytical comparisons with optimal linear measurement-based feedback were carried out on an equal footing in the ground-state-cooling regime [1407.4883]. Subsequent work broadened the notion of CFC from static sideband-cooling architectures to traveling-wave loops with explicit phase and delay control, two-pass cavity interactions, ancilla-assisted collision models, and transfer-function-based controller synthesis [2210.07674, 2204.00479, 2604.01891].

Experimentally, the field has progressed from theoretical demonstrations of performance gains in the quantum regime to direct implementations in several platforms. These implementations share a common logic: information about motion is encoded into an optical or ancilla mode, rotated in phase space by a coherent transformation or propagation delay, and returned so as to preferentially enhance anti-Stokes-like processes or generate a momentum-proportional force [2111.09802, 2506.21341, 2605.20902].

## 2. Dynamical mechanism and canonical equations

In the levitated-nanoparticle realization, a single silica nanoparticle is held in an optical tweezer and its center-of-mass motion \(z_p(t)\) along the beam axis is imprinted as a small phase modulation \(\phi_p(t)=B z_p(t)\) on the field it scatters [2506.21341]. After a fixed loop delay \(\tau\), the scattered field returns with phase \(\phi(t)=\phi_p(t-\tau)+\phi_0\), and its interference with the trapping field shifts the trap equilibrium according to
\[
z_{\rm eq}(t)=\beta z_p(t-\tau),
\]
with \(\beta\propto\sqrt{\eta}\) set by the collection and re-injection efficiency, field amplitudes, and spatial mode overlap. For a harmonic oscillator of natural frequency \(\Omega\), choosing \(\tau\approx \pi/(2\Omega)\) makes the equilibrium shift track the velocity, producing an effective viscous force
\[
F_{\rm fb}\approx -m\Gamma_c v_p(t), \qquad \Gamma_c=\beta\Omega\sin(\Omega\tau),
\]
so the damping is generated without ever measuring \(z_p\) [2506.21341].

The same mechanism can be written as a delayed self-interaction. For the nanoparticle center-of-mass mode,
\[
H=\frac{p^2}{2m}+\frac12 m\Omega^2\,[z_p(t)-z_{\rm eq}(t)]^2,
\qquad z_{\rm eq}(t)=\beta z_p(t-\tau),
\]
and, to first order in \(\beta\),
\[
H_{\rm int}\approx -m\Omega^2\beta\, z_p(t)\,z_p(t-\tau).
\]
Using
\[
z_p(t-\tau)\approx z_p(t)\cos\Omega\tau-\frac{\dot z_p(t)}{\Omega}\sin\Omega\tau,
\]
one obtains both a stiffness modification proportional to \(\beta\Omega^2\cos\Omega\tau\) and a damping term proportional to \(\beta\Omega\sin\Omega\tau\) [2506.21341].

In cavity optomechanical implementations, the same physics is often expressed through a closed-loop susceptibility. For a linearized cavity mode \(\hat a\) and mechanical mode \(\hat b\),
\[
\hat H=\hbar\omega_c\,\hat a^\dagger\hat a+\hbar\omega_m\,\hat b^\dagger\hat b-\hbar g_0\,\hat a^\dagger\hat a\,(\hat b+\hat b^\dagger),
\]
and the coherent-feedback loop modifies the mechanical response as
\[
\chi_{\rm cf}^{-1}(\omega)=\chi_m^{-1}(\omega)-G_{\rm fb}(\omega),
\]
with
\[
G_{\rm fb}(\omega)=|G|^2\,\kappa_{\rm in}\,\eta\,e^{i(\omega\tau+\gamma)}\,\chi_c(\omega)\,\chi_c(-\omega),
\qquad
\Gamma_{\rm CFC}(\omega)=2\,\mathrm{Im}\{G_{\rm fb}(\omega)\}.
\]
The loop phase \(\gamma\), delay \(\tau\), and cavity susceptibility \(\chi_c\) therefore enter the feedback as a complex self-energy whose imaginary part produces damping and whose real part produces reactive shifts [2605.20902].

A closely related two-pass cavity treatment yields explicit damping and spring-shift expressions. In the unresolved-sideband membrane experiment,
\[
\Gamma_m
=16\,\sqrt\eta\,\frac{g_1g_2}{\kappa}\,\sin\varphi\,\sin(\Omega_m\tau),\qquad
\delta\Omega_m
=-8\,\sqrt\eta\,\frac{g_1g_2}{\kappa}\,\sin\varphi\,\cos(\Omega_m\tau),
\]
so \(\varphi\) and \(\tau\) determine whether the feedback is dominantly dissipative or dispersive. The optimal-cooling choice is
\[
\varphi=\frac{\pi}{2},\qquad \Omega_m\tau=\frac{\pi}{2},
\]
for which \(\Gamma_m\) is maximized and \(\delta\Omega_m=0\) [2210.07674].

## 3. Experimental realizations across platforms

The experimental literature now contains several distinct CFC architectures. Some use the same optical field interacting twice with the same mechanical mode through different cavity modes; others use a traveling-wave delay line and direct reinjection; still others use a remote ancilla oscillator as the coherent controller. In all cases, no classical estimate of the mechanical state is used inside the feedback loop [2210.07674, 2506.21341, 2111.09802].

| Platform | Coherent loop architecture | Reported performance |
|---|---|---|
| Levitated nanoparticle [2506.21341] | Back-scattered light delayed in a \(\approx1.3\,\mathrm{km}\) single-mode fiber and refocused onto the particle; no photodetection | \(T_{\rm eff}\approx705\,\mu\mathrm{K}\), \(n\approx344\), \(\Gamma_c\) up to \(\approx2\pi\times250\,\mathrm{Hz}\) |
| Cavity membrane [2210.07674] | Same laser interacts twice through two orthogonal-polarization cavity modes with auxiliary LO, tunable phase \(\varphi\), and delay \(\tau\) | \(\bar n_m=4.89\pm0.14\) phonons at \(20\,\mathrm{K}\) |
| Density phononic-crystal membrane [2605.20902] | Reflected field displaced, delayed, and re-injected into the cavity input in orthogonal polarization; combined with strong DBC | \(n_f=166\pm7\), cooling factor \(\approx3.3\times10^4\), \(T_{\rm eff}\approx9.1\,\mathrm{mK}\) |
| Membrane with atomic spins [2111.09802] | Light-mediated spin-mechanics loop with state swaps and stroboscopic spin pumping | \(T=216\,\mathrm{mK}\), \(\bar n_m=2.3\times10^3\), \(200\,\mu\mathrm{s}\) |

These realizations occupy different operating regimes. The cavity membrane experiment was performed in the unresolved-sideband regime \(\kappa\gg\Omega_m\) and cooled below the theoretical limit of cavity dynamical backaction cooling in that regime [2210.07674]. The ultracoherent phononic-crystal membrane combined CFC with strong dynamical backaction cooling in a relatively narrow cavity and achieved a cooling factor of \(3.3\times10^4\) at room temperature [2605.20902]. The levitated-nanoparticle experiment emphasized a purely all-optical, measurement-free loop that preserves correlations between motion and the feedback field [2506.21341]. The spin-membrane platform used coherent swaps with an ancilla that could be tuned from strong coupling to an overdamped regime, thereby interpolating between state-transfer and reservoir-cooling behavior [2111.09802].

## 4. Optimization, limiting noise, and stability

For the levitated nanoparticle, the center-of-mass spectrum \(S_{zz}(\omega)\) yields an effective mode temperature through equipartition, and the resulting steady-state expression contains both ordinary cooling and a feedback-noise contribution [2506.21341]. In the weak-gain, low-noise regime one recovers the cold-damping form
\[
T_{\rm eff}=T_0\,\frac{\Gamma_0}{\Gamma_0+\Gamma_c},
\qquad T_0=k_B T_{\rm env},
\]
but for stronger \(\beta\) the phase-noise term grows as \(\beta^2\) and eventually limits cooling. Minimizing the full expression with respect to \(\beta\) gives
\[
\beta_{\rm opt}=\frac{B}{m\Omega^2}\sqrt{\frac{\sigma_m^2+\sigma_r^2}{\sigma_\phi^2}},
\qquad
T_{\min}
=
\frac{\sigma_\phi\,\Omega\,\sqrt{\sigma_m^2+\sigma_r^2}}
{k_B\,B\,\sin\Omega\tau}.
\]
This minimum scales linearly in \(\sigma_\phi\) and \(\Omega\), and inversely in the interference contrast \(B\) and \(\sin\Omega\tau\). In the \(1.3\,\mathrm{km}\) fiber loop, the dominant limit is phase noise, with \(\sigma_\phi^2\propto L\), and the onset of noise squashing in the in-loop spectrum is the tell-tale signature that phase noise, not particle motion, dominates the loop [2506.21341].

For the room-temperature membrane, the final occupancy is written as
\[
n_f=
\frac{\gamma_m n_{\rm th}
+\Gamma_{\rm DBC}n'_{\rm add}
+\Gamma_{\rm CFC}n_{\rm add}}
{\Gamma_{\rm eff}},
\qquad
\Gamma_{\rm eff}=\gamma_m+\Gamma_{\rm DBC}+\Gamma_{\rm CFC}(\omega_m),
\]
so the role of CFC is separated from thermal loading, dynamical-backaction added noise, and loop-added noise [2605.20902]. The full model, which includes cavity dynamics, loop delay and loss, and measured technical noise, quantitatively reproduces the measured spectra and the extracted \(n_f(\gamma,\tau,\Delta_h)\). The principal experimental limitations are finite loop delay, optical loss, imperfect mode matching, and technical noise, with cavity-mirror vibrations and laser noise contributing \(n_{\rm add}\sim10^2\) to the final occupation [2605.20902].

Delay is a recurring control parameter rather than a small correction. In the room-temperature membrane experiment, finite loop delay sets \(\tau\le \pi/2\omega_m\) for optimal cooling [2605.20902]. In the spin-membrane system, the propagation delay of about \(15\,\mathrm{ns}\) shifts the optimum damping condition, makes the steady-state \(\bar n_m(\delta)\) asymmetric in detuning, and produces self-oscillation regions for negative detuning [2111.09802]. In a passive interferometric model of cavity optomechanics, moderate delays of order \(1/\omega_m\) degrade the cooling only slightly, whereas very long delays effectively remove the benefit of feedback [2011.11585]. This suggests that delay is simultaneously a resource for quadrature rotation and a constraint through added phase and stability sensitivity.

## 5. Relation to measurement-based feedback and standard sideband cooling

Measurement-based feedback relies on detecting the mechanical motion, processing the resulting classical signal, and then applying an electrical or optical force. In the levitated-particle comparison, this was stated to incur fundamental back action linked to detection efficiency, electronic noise in amplifiers and delays, and limited loop bandwidth with extra delay [2506.21341]. CFC avoids those particular penalties because no photodetection feeds the loop, and in all-optical implementations it preserves the motion-feedback correlations that measurement would otherwise discard [2506.21341, 2210.07674].

The comparison is especially sharp in the sideband-unresolved regime. Dynamical backaction cooling alone requires \(\kappa<\omega_m\) for strong cooling, but coherent feedback does not rely on resolving motional sidebands and can still produce a large \(\Gamma_{\rm CFC}\) when \(\tau\) and \(\gamma\) are chosen to yield negative-momentum feedback [2605.20902]. In the unresolved-sideband membrane experiment, standard dynamical backaction cooling had a lower bound of about \(7\) phonons for the stated parameters, whereas coherent feedback reached \(\bar n_m=4.89\pm0.14\) phonons and did so with roughly \(1\%\) of the optical power needed for cavity cooling to the same \(\bar n\) in the bad-cavity regime [2210.07674].

Analytical comparisons sharpen the origin of this advantage. In a perturbative treatment of resolved-sideband cooling and optimal linear measurement-based feedback, the maximal coherent-feedback cooling was found to scale as
\[
\bar n_{\min}^{\rm CFC}\approx 1.67\left(\frac{\gamma n_T}{\omega}\right)^{2/3},
\]
whereas the optimal measurement-based scaling was
\[
\bar n_{\min}^{\rm MBF}\approx
\left(\sqrt2+\frac12\right)\sqrt{\frac{\gamma n_T}{\omega}}.
\]
The performance gap was attributed not to the back-action noise of the measurement but to projection noise, with coherent feedback making better use of the same linear interaction [1407.4883].

A common oversimplification is that coherent feedback is always superior. A unified collision-model treatment gives a more conditional result: measurement-based feedback is typically superior in cooling, whilst coherent feedback is better at assisting quantum operations [2204.00479]. The same framework also shows that when a pure ancilla is available, coherent feedback can purify the system to the pure state \(|0\rangle\) for any nontrivial coupling angle, whereas if the ancilla is maximally mixed, coherent feedback alone cannot reduce the system entropy at all and a projective measurement-based loop yields a lower-entropy steady state [2204.00479]. This suggests that the comparative ranking depends on the controller resource and on what is meant by “cooling.”

## 6. Generalizations, controller synthesis, and prospects

Several recent works recast CFC as a systematic controller-design problem. In the loop-shaping approach, the controller is specified by a transfer function \(K(\omega)\), the closed-loop optical susceptibility becomes
\[
\chi_{\rm cl}(\omega)=\frac{\chi(\omega)K(\omega)}{1-L(\omega)},
\qquad
L(\omega)=\sqrt\kappa\,\chi(\omega)\,K(\omega),
\]
and the controller is designed so that the Stokes process is suppressed while the anti-Stokes process is preserved or enhanced [2604.01891]. For a notch-filter configuration, the shaped rates satisfy
\[
\Gamma_+^{(n)}=0,\qquad \Gamma_-^{(n)}=\frac{4g^2}{\kappa},
\]
so ground-state cooling becomes possible even in the unresolved-sideband regime. The same framework emphasizes standard loop-stability criteria, including Nyquist and gain/phase-margin conditions, and extends naturally to dissipative entanglement generation and stabilization of nonclassical states [2604.01891].

Passive-feedback proposals further show that CFC can make a broad-band cavity behave as though it were sideband resolved. One proposal uses a narrow-linewidth passive element so that an optomechanical system in the deeply sideband-unresolved regime exhibits dynamics similar to a sideband-resolved system; for representative chip-scale parameters, the optimized final occupation was reported as \(n_f=0.07\)–\(0.2\) for \(\eta_s=0.7\)–\(0.9\) [2206.13917]. A related Fano-mirror analysis found that a single-sided coherent-feedback loop can achieve ground-state cooling even when the total optical loss is more than seven orders of magnitude larger than the mechanical frequency and the feedback efficiency is relatively low, whereas a more standard double-sided feedback scheme is not appropriate for the highly asymmetric two-mirror Fano cavity [2405.13624].

The outlook toward the motional ground state is now platform specific but technically concrete. For levitated nanoparticles, the stated route is to shorten the delay line, reduce phase noise by better cancellation methods such as Pound–Drever–Hall feed-forward and fiber noise cancellation, work at \(p\sim10^{-8}\,\mathrm{mbar}\), reduce photon recoil with smaller particles, and improve collection and mode matching; a realistic extrapolation combining these was reported to yield \(T_{\min}\) corresponding to \(n\approx1\) phonon in an all-optical, measurement-free loop [2506.21341]. For the room-temperature phononic-crystal membrane, increasing \(g_0\) by \(\times10\), suppressing cavity-frequency noise by more than \(10\,\mathrm{dB}\), and improving loop efficiency toward \(\eta\to1\) leads in the model to \(n_f\approx8\), and to \(n_f\approx2\) in the quantum-noise-limited limit [2605.20902].

Beyond cooling, the same coherent loops have been proposed or demonstrated as tools for non-reciprocal interactions, reservoir-engineered entanglement between remote particles, spin-mechanics interfaces, photon-phonon entanglement generation, steady-state squeezing, and transiently assisted state transfer [2506.21341, 2206.13917, 2011.11585]. CFC is therefore not merely a cooling technique but a broader architecture for quantum control in which dissipation, delay, and interference are engineered coherently rather than through measurement and classical actuation.

Source: https://www.emergentmind.com/topics/coherent-feedback-cooling-cfc