---
title: Parametric Feedback Cooling
url: https://www.emergentmind.com/topics/parametric-feedback-cooling
type: topic
---

# Parametric Feedback Cooling

Searching arXiv for the provided ids to ground the article in the cited literature.
Parametric feedback cooling is an active cooling method for mechanical motion in which the stiffness of a trap or resonator is modulated at approximately twice the mechanical resonance frequency so that the modulation extracts energy rather than injects it. In its standard form, the method acts on the center-of-mass or librational motion of a trapped object by engineering an additional damping term through phase-controlled modulation of the restoring potential, with the effective temperature reduced according to the ratio of intrinsic to total damping [1202.6435]. The technique has been developed in levitated optomechanics, cavity QED, rotational nanomechanics, and low-frequency pendulum systems, with implementations ranging from analog sinusoidal modulation to square-wave digital control, phase-adaptive protocols, and measurement-based optimal quantum feedback [1603.02917].

## 1. Fundamental mechanism

For a harmonically trapped degree of freedom \(x(t)\), the uncooled dynamics are commonly written as
\[
m\ddot x + m\Gamma_0 \dot x + m\Omega_0^2 x = F_{\rm th}(t),
\]
with \(F_{\rm th}(t)\) a Langevin force satisfying \(\langle F_{\rm th}(t)F_{\rm th}(t')\rangle = 2m\Gamma_0 k_B T_0 \delta(t-t')\) [1202.6435]. Parametric feedback introduces a time-dependent spring constant \(k(t)=k_0[1+\epsilon(t)]\), so that the oscillator equation becomes
\[
m\ddot x + m\Gamma_0 \dot x + m\Omega_0^2[1+\epsilon(t)]x = F_{\rm th}(t).
\]
A convenient choice is \(\epsilon(t)=\epsilon_0\cos(2\Omega_0 t+\phi)\), namely stiffness modulation at \(2\Omega_0\) with a controllable phase \(\phi\) [1202.6435].

Under a rotating-wave or slow-amplitude analysis, the \(2\Omega_0\) modulation yields an effective friction term \(m\Gamma_{\rm fb}\dot x\), with
\[
\Gamma_{\rm fb}=\frac{\epsilon_0\Omega_0}{2}\sin\phi.
\]
If \(\sin\phi>0\), the modulation adds damping; if \(\sin\phi<0\), it amplifies the motion [1202.6435]. Closely related formulations appear in cavityless levitated optomechanics, where the control signal \(u(t)\) modulates the trap frequency in the Hamiltonian \(H(z,p,u)=p^2/(2m)+\tfrac12 m\omega^2[1+\beta u(t)]z^2\), and in librational systems, where the restoring potential is written as \(V(\theta,t)=\tfrac12 I\Omega^2[1+\epsilon\cos(2\Omega t+\phi)]\theta^2\) [1904.05274]; [2402.19245].

The physical picture is consistent across platforms. When the particle is near a turning point, the trap is stiffened; when it passes through the origin, the trap is softened. If the modulation is kept in the appropriate phase relation with the motion, the net effect is energy extraction over each cycle rather than parametric excitation [2102.01060]. In experimental practice, this phase relation can be implemented by using \(x(t)\dot x(t)\), by PLL-based phase tracking, or by continuous quadrature estimation [1603.02917]; [2102.01060].

## 2. Dynamical descriptions and temperature reduction

The basic thermal consequence of the added damping is expressed by
\[
\Gamma=\Gamma_0+\Gamma_{\rm fb},
\qquad
T_{\rm eff}=T_0\frac{\Gamma_0}{\Gamma_0+\Gamma_{\rm fb}},
\]
so that the center-of-mass temperature is reduced by increasing the parametric damping while the thermal force remains tied to the intrinsic damping \(\Gamma_0\) [1202.6435]. The same structure appears in several analyses. In the direct comparison study of levitated oscillators, the effective parametric damping rate is identified as \(\Gamma_p\equiv G\omega_0/2\), giving
\[
T_{\rm eff}=T_0\Bigl[1+\frac{G\omega_0}{2\gamma_0}\Bigr]^{-1}
=T_0\frac{\gamma_0}{\gamma_0+\Gamma_p},
\]
when detection noise and back-action are neglected in the analytic model [2102.01060]. In mechanical pendulum cooling, the corresponding extra damping is
\[
\mu=\frac{\Delta k}{4m\omega_0}\sin\varphi,
\qquad
\frac{T_r}{T}=\frac{\gamma}{\gamma+\mu},
\]
derived from a suspension-point modulation at \(2\omega_0\) [2012.12158].

The frequency-domain description is equally central. For translational motion of a nanoparticle, the displacement power spectral density can be written as
\[
S_x(\omega)=
\frac{2\Gamma_0 k_B T_0/m}
{(\Omega_0^2-\omega^2)^2+\omega^2(\Gamma_0+\Gamma_{\rm fb})^2},
\]
so that the spectral area is proportional to \(T_{\rm eff}\) [1202.6435]. In the parabolic-mirror-trap implementation, the spectral density is written as a shifted, broadened Lorentzian,
\[
S_x(\omega)=\frac{k_B T_{\rm env}}{\pi m}
\frac{\Gamma_0}
{\bigl[(\Omega_0+\delta\Omega)^2-\omega^2\bigr]^2+\omega^2(\Gamma_0+\Gamma_{\rm fb})^2},
\]
with \(\delta\Omega=(\epsilon\Omega_0/2)\cos\phi\) and \(\Gamma_{\rm fb}=(\epsilon\Omega_0/2)\sin\phi\), thereby making explicit that parametric actuation changes both damping and resonance frequency [1603.02917].

Several variants modify the feedback waveform while retaining the same principle. In digital parametric feedback, the trap stiffness is switched between two levels through
\[
k(t)=k_0[1+\delta\cdot {\rm sgn}(\sin 2\omega_0 t+\phi)],
\]
leading to an extra damping
\[
\delta\Gamma(\eta,\phi)=\frac{\eta\omega_0}{\pi}\cos(2\phi),
\]
and a temperature
\[
T_{\rm cm}=T_{\rm env}\frac{\Gamma_{\rm th}}{\Gamma_{\rm th}+\delta\Gamma(\eta,\phi)}.
\]
For the square-wave case, optimal cooling corresponds to \(\phi=0 \bmod \pi\), unlike the sinusoidal analog, where the optimum phase is tied to the sign of \(\sin\phi\) [1904.06410]. This difference is a frequent source of confusion when comparing implementations.

## 3. Experimental realizations across platforms

The earliest levitated-nanoparticle demonstrations emphasized that a laser-trapped nanoparticle is entirely isolated from the thermal bath and lacks a clamping mechanism, enabling robust decoupling from internal vibrations and cooling in all degrees of freedom by means of a single laser beam [1202.6435]. In one implementation, a single 1064 nm laser of approximately 100 mW was focused by an NA \(=0.8\) objective into a vacuum chamber to trap a fused-silica nanosphere of radius \(R\approx 70\) nm and mass \(m\approx 3\times 10^{-18}\) kg. Three balanced photodetectors monitored forward-scattered light in \(x\), \(y\), and \(z\), with a phase-sensitive interferometric noise floor of approximately \(1.2\) pm/\(\sqrt{\rm Hz}\) [1202.6435]. A related parabolic-mirror geometry used a single-mode 1550 nm fibre laser, a single photodiode, and a single beam for trapping, position detection, and cooling of all three dimensions, with backscattered self-homodyne detection reaching a position sensitivity of approximately \(200\) fm/\(\sqrt{\rm Hz}\) [1603.02917].

Feedback-loop architectures differ substantially. The standard analog loop in nanoparticle trapping differentiates each detector signal, multiplies \(x(t)\dot x(t)\) to generate a component at \(2\Omega_0\), applies a phase shift, sums the three axes’ \(2\Omega\) signals, and drives a Pockels cell or AOM so that the trapping-laser power becomes \(P(t)=P_0[1+\epsilon(t)]\) [1202.6435]. Digital parametric feedback replaces continuous phase-shifters by FPGA processing: the position signals are digitized, Kalman-filtered, delay-compensated, converted into synchronous square waves, and combined through a “majority voting” logic for tri-axial cooling [1904.06410]. In cavityless levitated optomechanics, the control signal is derived from an LQG formulation and applied by an AOM through \(P_L(t)=P_0[1+\beta u^*(t)]\), while the state estimator explicitly incorporates the time-dependent stiffness \(A(t)\) to avoid lock loss at large modulation depth [1904.05274].

Outside center-of-mass levitation, the same principle has been adapted to rotational and atomic degrees of freedom. In optically levitated libration, a backward-scattering heterodyne scheme with a 1550 nm trapping beam and an LO offset by \(\Delta f\approx 9\) Hz provided linear signals for all three libration modes, and the PLL phase was shifted by \(\pi/4\) to approximate \(-\dot\theta\) before driving an EOM at \(2\Omega\) [2402.19245]. For a single atom in an optical cavity, parametric feedback combined trap-depth modulation at \(2\omega\) with fast repetitive measurements of the atomic position derived from cavity transmission and processed on an FPGA through an IQ demodulator [1805.00241]. In pendulum-based gravity experiments, the restoring stiffness was modulated mechanically via vertical suspension-point motion \(z(t)=z_0\cos(\Omega t+\varphi)\), with a digital phase tracker maintaining the actuator at \(2\omega_0\) and near-optimal phase [2012.12158].

## 4. Demonstrated performance and limiting mechanisms

The 2012 nanoparticle experiment reported bare trap frequencies \(f_x=120\) kHz, \(f_y=134\) kHz, and \(f_z=37\) kHz, with \(\Gamma_0\propto P\) in the free-molecular regime and \(\Gamma_0\approx 2\pi\times 10\) mHz at \(10^{-5}\) mBar, corresponding to \(Q\approx 10^7\) [1202.6435]. With feedback engaged, all three axes cooled; in the lowest-noise run, the reported values were \(T_{\rm eff,z}\approx 50\) mK at \(P\approx 2.5\times 10^{-6}\) mBar, \(T_{\rm eff,x}\approx 0.6\) K at \(P\approx 3.4\times 10^{-4}\) mBar, and \(T_{\rm eff,y}\approx 1.6\) K at the same pressure, with overall compression factors \(T_0/T_{\rm eff}\) up to \(\sim 10^4\) [1202.6435]. In the parabolic-mirror trap, room-temperature motion was cooled to as little as \(T_{\rm eff}\approx 1\) mK in all three axes, with \(\Gamma_0/2\pi\simeq 2\) mHz at \(6\times 10^{-6}\) mbar and estimated \(Q>4\times 10^7\) [1603.02917]. Digital square-wave cooling achieved center-of-mass temperatures of approximately \(10\) mK at \(2\times 10^{-7}\) mbar with modulation depth up to \(0.8\%\) on the \(Y\)-axis [1904.06410].

Librational cooling has reached lower occupancies in selected modes. Using backward-scattering detection, all three libration degrees of freedom were cooled below \(16\) mK, with one mode at \(1.34\pm 0.14\) mK corresponding to \(\bar n_\alpha=84\pm 9\), and the measurement efficiency for the \(\alpha\) mode was reported as approximately \(0.5\%\) [2402.19245]. In single-atom cavity QED, parametric cooling of a radial mode at \(\omega_\rho/2\pi=4.8\) kHz increased the average storage time by a factor of \(60\) to more than \(2\) s, while a \(\sim 500\) kHz axial mode exhibited a \(\approx 30\%\) extension of trap lifetime within microseconds of cooling [1805.00241]. For pendulum-based gravity experiments, a proof-of-principle demonstration achieved a damping factor of \(5.7\), reducing the effective mode temperature from \(300\) K to approximately \(50\) K in the seismic-noise-dominated regime [2012.12158].

The limiting mechanisms are platform-dependent but structurally similar. Reported limits include reheating by gas collisions, measurement noise floor, recoil heating, finite PLL bandwidth, phase jitter, slow trap-frequency drift, cross-coupling to other modes, and actuator or electronics noise [1202.6435]; [2102.01060]; [2402.19245]. In the original nanoparticle experiment, the measurement noise estimate \(x_{\min}\approx [\hbar c/(8\pi P_{\rm det} B)]^{1/2}\approx 6.7\) pm in \(1\) Hz implied a theoretical \(T_{\rm eff,min}\approx 7\) \(\mu\)K in the absence of back-action, while photon-recoil heating was estimated to give only one event per \(\sim 10\) oscillations for \(R=70\) nm and \(\lambda=1064\) nm [1202.6435]. In the direct comparison study, two bounds were emphasized: a stability-bandwidth limit \(T_{\rm lim}^1=T_0(\gamma_0/B_{3{\rm dB}})\) and a loop-SNR limit \(T_{\rm lim}^2=(m\omega_0^2/k_B)\,2B_L S_{nn}\), with the practical minimum approximated by \(T_{\min}\simeq \max[T_{\rm lim}^1,T_{\rm lim}^2]\) [2102.01060].

## 5. Optimal, adaptive, and quantum formulations

A major development beyond fixed-phase feedback is the explicit use of optimal control and adaptive phase updates. In cavityless levitated optomechanics, parametric cooling has been cast as an LQG-type problem with performance index
\[
E\!\int_0^T \bigl[x^T(t)Qx(t)+Ru^2(t)\bigr]dt,
\]
leading through Pontryagin’s Minimum Principle and the Riccati equation to the feedback law
\[
u^*(t)=-R^{-1}B^T P(t)x(t),
\]
or \(u^*(t)=-Kx(t)\) in the infinite-horizon limit [1904.05274]. The essential implementation point is that the Kalman filter or PLL must incorporate the known modulation \(u^*(t)\) through the time-dependent matrix \(A(t)\); this allows stable tracking at modulation depths up to \(\beta\sim 0.5\), whereas the conventional double-phase scheme cannot be pushed much beyond \(\beta\sim 0.01\) without losing lock [1904.05274]. The reported consequence is up to \(10\times\) faster cooling and \(\gtrsim 10\times\) lower \(n_\infty\) than standard double-phase schemes [1904.05274].

A distinct adaptive strategy updates the modulation phase directly from measured quadratures. In phase-adaptive parametric cooling, the phase at the \(j\)-th update is set to
\[
\phi_{\rm opt}^{(j)}=\frac{\pi}{2}+2\tan^{-1}\!\Bigl[q(j\delta\tau)/p(j\delta\tau)\Bigr],
\]
with repeated stroboscopic updates yielding purely exponential energy decay,
\[
E(t)\propto e^{-\Gamma t}.
\]
The classical and quantum steady-state occupancies are then
\[
n_{\rm final}^{(\rm cl)}=\frac{\gamma n_{\rm th}}{\gamma+\Gamma}\approx \frac{\gamma}{\Gamma}n_{\rm th},
\qquad
n_{\rm final}^{(\rm qu)}=\frac{\gamma}{\Gamma}\Bigl(n_{\rm th}+\frac12\Bigr),
\]
for \(\Gamma\gg \gamma\) [2205.12791]. This result explicitly distinguishes adaptive phase control from plain fixed-phase parametric cooling, which is described there as producing a non-exponential, slow energy decay [2205.12791].

Quantum analyses impose stronger constraints. A quantum calculation in the shot-noise-dominant regime derived a stochastic master equation with measurement efficiency \(\eta\), shot-noise heating per oscillation \(\Delta n\), and parametric-feedback Hamiltonian \(H_{\rm fb}(t)=\tfrac{\chi}{2}m\omega^2 x_m(t)\dot x_m(t)\,x^2\) [1708.01203]. In that treatment, the minimum occupation under parametric feedback scales approximately as
\[
n_{\min}\approx 1.5\Bigl(\frac{\Delta n}{\eta}\Bigr)^{1/3},
\]
and ground-state cooling requires substantially higher \(\eta\) than force feedback [1708.01203]. By contrast, an optimal quantum protocol based on heterodyne measurement, resonant parametric modulation, and conditional choice of both phase and duration derived the phase relation
\[
\phi_p=\frac{\pi}{2}-2\phi,
\]
the optimal squeezing duration
\[
t_{\rm op}\approx \frac{\ln(1+4r^2)}{4\lambda},
\]
and a steady-state fixed point \(\bar n_f\approx 0.83\), even for an isolated oscillator \((\gamma=0)\) [2204.00476]. This suggests that measurement-conditioned parametric modulation can, in principle, cool below one quantum without sideband or linear-feedback cooling [2204.00476].

## 6. Comparisons, misconceptions, and scope

Parametric feedback cooling is often discussed alongside velocity damping, cold damping, and cavity sideband cooling, but the methods are not interchangeable. In a direct comparison performed on the same levitated particle and with the same detection system, velocity damping cooled the oscillator to \(T_{\min}\approx 26\) mK, almost an order of magnitude below the best parametric-feedback result of approximately \(280\) mK, and was reported to be more resilient to imperfect experimental conditions [2102.01060]. The reasons given include lower effective back-action, automatic tracking of frequency drift, and reduced sensitivity to cross-couplings and phase-estimation errors [2102.01060]. Likewise, in the shot-noise-dominant quantum calculation, force feedback reached lower occupation than parametric feedback at fixed measurement efficiency [1708.01203]. A common misconception is therefore that parametric feedback is universally the lowest-temperature route; the comparative literature does not support that claim.

Another misconception is that parametric feedback acts identically on all mechanical coordinates. In rigid-body nanodumbbells trapped in a linearly polarized laser beam, standard parametric feedback can extract energy from two of the five rotational degrees of freedom, but the dynamics after feedback are characterized by a normal mode describing precession about the laser polarization axis together with spin about the nanoparticle’s symmetry axis [1810.01797]. Full cooling of the librational coordinates requires an asymmetry in the librational frequencies and feedback modulation containing both rotational frequencies [1810.01797]. This is not a minor technicality; it shows that the modal structure of the underlying Hamiltonian can obstruct naïve extensions of center-of-mass protocols to rotational motion.

At the same time, parametric feedback remains attractive in geometries with limited optical access or restricted actuator choices. In the single-atom cavity experiment, only one optical mode was used for both measurement and actuation, and the method remained effective for a \(\sim 500\) kHz oscillation mode within microseconds [1805.00241]. In levitated optomechanics, the single-beam implementations and digital FPGA realizations show that all-optical or minimally invasive control of several motional degrees of freedom is feasible [1603.02917]; [1904.06410]. Applications identified across the literature include ultrasensitive force sensing with \(F_{\min}\sim 10^{-20}\) N/\(\sqrt{\rm Hz}\), tests of quantum mechanics with mesoscopic objects, searches for nonstandard forces, quantum-state engineering, damping of pendulum modes in gravitational-wave detectors, and preparation of rotational states relevant to matter-wave interferometry [1202.6435]; [2012.12158].

The long-term outlook is framed by the competition among damping, measurement imprecision, recoil or technical heating, and phase-tracking fidelity. For laser-trapped nanoparticles with \(\Omega_0/2\pi=120\) kHz, the condition \(\langle n\rangle<1\) implies \(T_{\rm eff}\lesssim \hbar\Omega_0/k_B\sim 6\) pK, and extrapolation of \(\Gamma_0\propto P\) to \(P\sim 10^{-11}\) mBar together with higher feedback gain was identified as a route toward \(T_{\rm eff}\sim\) pK [1202.6435]. More recent analyses instead emphasize that reaching the near-ground-state regime requires not only high \(Q\) and ultralow gas damping, but also measurement and estimator designs that remain accurate under strong modulation [1904.05274]; [2205.12791]; [2204.00476]. A plausible implication is that the future of parametric feedback cooling lies less in the elementary \(x\dot x\) loop by itself than in hybrid architectures that combine parametric modulation with state estimation, adaptive phase control, or complementary cooling channels.

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