Cavity Optomechanics Radiation Pressure Dynamical Backaction Resolved Sideband Ground State Cooling

# Cavity Optomechanics: Dynamical Backaction, Ground-State Sideband Cooling, and OMIT

Cavity optomechanics controls mechanical motion with photons. A displacement shifts an optical or microwave resonance; the intracavity field in turn exerts radiation pressure. This reciprocal interaction supports displacement sensing, optical-spring tuning, amplification, coherent photon-phonon conversion, ground-state cooling, and optomechanically induced transparency (OMIT). The central engineering task is to preserve sign conventions and distinguish the single-photon coupling g_0 from the drive-enhanced coupling g=g_0sqrt(n_c).

## Radiation-pressure Hamiltonian

For one cavity mode and one mechanical mode,

The mechanical displacement is x = x_zpf(b + b†), where

Thus g_0 is the resonance shift produced by zero-point motion, expressed as an angular frequency. For a Fabry-Pérot cavity, |G| ≈ ω_c/L; nanophotonic moving-boundary and photoelastic contributions require overlap integrals.

Drive the cavity at ω_L with detuning Δ = ω_L - ω_c. Write a = α + d and neglect the nonlinear fluctuation product d†d(b + b†). Choosing the drive phase so α is real gives

The steady photon number for input power P_in is

where κ = κ_ex + κ_0 is the energy-decay linewidth. Some literature uses half-linewidth; every formula must use one convention consistently.

## Langevin equations and susceptibilities

The linearized quantum Langevin equations are

With Fourier convention exp(-iωt), the bare cavity susceptibility is

and the mechanical susceptibility near positive frequency is χ_m(ω) = [γ_m/2 - i(ω - Ω_m)]⁻¹. Radiation pressure creates a self-energy. Its real part shifts frequency; its imaginary part changes damping.

The anti-Stokes and Stokes scattering rates are

For the chosen detuning convention, red detuning Δ≈-Ω_m resonantly enhances A_-. The optical damping is

where Σ(ω)=κ/[(κ/2)²+(Δ+ω)²]. The optical spring shift is

Changing the detuning convention changes displayed signs, not the measured physics. A reliable check is that red-detuned cooling gives Γ_opt>0.

## Sideband cooling

The effective linewidth is γ_eff=γ_m+Γ_opt. Balancing thermal injection and quantum backaction gives

At Δ=-Ω_m in the resolved-sideband limit κ≪Ω_m, the quantum limit is

With cooperativity C=4g²/(κγ_m) and strong cooling, a useful approximation is

The first term is sometimes written γ_m n_th/Γ_opt; dropping n_th produces a dimensionally plausible but physically incorrect cooling claim. Ground-state cooling needs both quantum cooperativity C_q=C/n_th>1 and resolved sidebands. Absorption heating, technical laser noise, and additional mechanical modes often set a higher floor.

Blue detuning makes A_+>A_- and Γ_opt<0. When |Γ_opt|>γ_m, the linear state becomes unstable and self-oscillation begins. Close to threshold the linearized model predicts divergence; nonlinear saturation must be included.

## Beam-splitter and squeezing interactions

At red detuning and under a rotating-wave approximation,

which swaps photon and phonon excitations. At blue detuning,

which produces correlated pairs and amplification. The rotating-wave approximation requires the discarded terms to rotate rapidly compared with coupling and damping. In strong coupling g≳κ,γ_m, hybrid modes show normal-mode splitting, and rate-only cooling formulas should be replaced by the coupled susceptibility.

## Optomechanically induced transparency

Apply a strong red control tone at ω_L and a weak probe separated from it by Ω. The probe creates an intracavity sideband; the control scatters from coherent mechanical motion into the same optical frequency. Their interference opens a narrow transmission window inside the broader cavity line.

In the red-sideband rotating-wave regime, the effective cavity susceptibility is

Input-output theory, a_out = a_in - sqrt(κ_ex) a, gives probe transmission

The feature width is approximately γ_m(1 + C), and its contrast depends on cooperativity and external coupling. The rapid phase slope creates group delay τ_g = d arg(t)/dΩ. OMIT is the optomechanical analogue of electromagnetically induced transparency, but the lossy mechanical resonance limits delay-bandwidth product and adds thermal noise.

## Calibration and noise thermometry

Displacement imprints phase modulation on the output. The measured voltage spectrum is converted to S_xx with a calibration tone or known thermal motion. Integrating a correctly normalized double-sided spectrum gives ⟨x²⟩ = x_zpf²(2n̄ + 1); single-sided conventions introduce a factor of two.

Sideband asymmetry offers quantum thermometry: for an ideal detector, anti-Stokes and Stokes areas scale as n̄ and n̄+1. Classical laser noise, filtering, detuning error, and unequal gain can mimic asymmetry. A credible analysis interchanges detunings, calibrates the transfer function, and propagates uncertainty in g_0, κ, and bath temperature.

The mechanical decoherence rate is approximately Γ_th=γ_m n_th at high temperature. A large mechanical quality factor Q_m=Ω_m/γ_m is useful, but coherent exchange requires g to compete with both optical loss and thermal decoherence.

## Numerical solver

This model computes damping, optical spring, final occupation, a two-dimensional cooling map, and complex OMIT transmission. All rates use radians per second.

import numpy as np
import matplotlib.pyplot as plt

twopi = 2*np.pi
Om = twopi*5.2e9
kappa = twopi*650e6
gamma_m = twopi*42e3
g0 = twopi*1.1e6
nc = 2.5e3
g = g0*np.sqrt(nc)
nth = 18.0
kappa_ex = 0.72*kappa

def rates(detuning, coupling=g):
    a_minus = coupling**2*kappa/((kappa/2)**2+(detuning+Om)**2)
    a_plus = coupling**2*kappa/((kappa/2)**2+(detuning-Om)**2)
    return a_minus, a_plus

def spring(detuning, coupling=g):
    return coupling**2*((detuning+Om)/((kappa/2)**2+(detuning+Om)**2)
        +(detuning-Om)/((kappa/2)**2+(detuning-Om)**2))

def occupation(detuning, coupling=g, bath=nth):
    am, ap = rates(detuning, coupling)
    denominator = gamma_m + am - ap
    return np.where(denominator > 0, (gamma_m*bath+ap)/denominator, np.nan)

detuning = np.linspace(-2.2*Om, 1.2*Om, 1800)
am, ap = rates(detuning)
gopt = am-ap
nf = occupation(detuning)
dOm = spring(detuning)

# OMIT around the two-photon resonance, with control detuning -Om.
probe = np.linspace(Om-8*(gamma_m+4*g*g/kappa),
                    Om+8*(gamma_m+4*g*g/kappa), 2400)
Delta_control = -Om
chi_eff = 1/(kappa/2-1j*(Delta_control+probe)
            +g*g/(gamma_m/2-1j*(probe-Om)))
t = 1-kappa_ex*chi_eff
group_delay = np.gradient(np.unwrap(np.angle(t)), probe)

# Cooling map versus detuning and photon number.
ncs = np.geomspace(1, 2e4, 220)
D, N = np.meshgrid(detuning[::5], ncs)
G = g0*np.sqrt(N)
AM = G**2*kappa/((kappa/2)**2+(D+Om)**2)
AP = G**2*kappa/((kappa/2)**2+(D-Om)**2)
NF = (gamma_m*nth+AP)/(gamma_m+AM-AP)
NF[(gamma_m+AM-AP) <= 0] = np.nan

fig, ax = plt.subplots(2, 2, figsize=(12, 8))
ax[0,0].plot(detuning/Om, gopt/twopi/1e6, label='optical damping')
ax[0,0].plot(detuning/Om, dOm/twopi/1e6, label='spring shift')
ax[0,0].axhline(-gamma_m/twopi/1e6, color='k', ls=':')
ax[0,0].set(xlabel=r'$\Delta/\Omega_m$', ylabel='MHz'); ax[0,0].legend()
ax[0,1].semilogy(detuning/Om, nf)
ax[0,1].axhline(1, color='k', ls='--')
ax[0,1].set(xlabel=r'$\Delta/\Omega_m$', ylabel='phonon occupation')
ax[1,0].plot((probe-Om)/twopi/1e6, np.abs(t)**2, label='transmission')
ax[1,0].set(xlabel='two-photon detuning (MHz)', ylabel=r'$|S_{21}|^2$')
ax2 = ax[1,0].twinx(); ax2.plot((probe-Om)/twopi/1e6, group_delay*1e9, 'C1')
ax2.set_ylabel('group delay (ns)')
mesh = ax[1,1].pcolormesh(D/Om, N, np.log10(NF), shading='auto', cmap='viridis')
ax[1,1].set_yscale('log'); ax[1,1].set(xlabel=r'$\Delta/\Omega_m$', ylabel=r'$n_c$')
fig.colorbar(mesh, ax=ax[1,1], label=r'$\log_{10}\bar n_f$')
fig.tight_layout(); plt.show()

C = 4*g*g/(kappa*gamma_m)
print('cooperativity:', C, 'quantum cooperativity:', C/nth)
print('resolved-sideband ratio:', Om/kappa)
print('best stable occupation:', np.nanmin(nf))

The script masks the blue-detuned unstable region instead of plotting negative occupations. Parameter fitting should use complex transmission, not magnitude alone, because phase separates detuning offsets from linewidth changes.

## Device and experimental design

Optical cavities favor low absorption and small mode volume; mechanics favors low mass, high frequency, and low clamping loss. Those goals can conflict because tight optical confinement increases surface sensitivity. Phononic shields reduce anchor loss, while careful symmetry can suppress radiation without isolating the desired optical port. Microwave electromechanics replaces radiation pressure with a displacement-dependent capacitance but retains the same linearized theory.

Thermalization is often the limiting issue at millikelvin temperature. Even a weak absorbed optical power can create a local hot bath whose occupation rises with drive. Measure linewidth and noise versus power independently: increased linewidth alone does not prove cooling if the bath simultaneously heats. Pulsed protocols can obtain large instantaneous cooperativity with less average absorption, at the price of a more complex transient model.

For frequency conversion between two cavities, red-detuned pumps create two beam-splitter interactions with one mechanical mediator. High internal conversion needs matched cooperativities; low added noise additionally needs a cold mediator and low internal optical loss. Conversion bandwidth is set by the pump-broadened mechanical linewidth.

## Validation checklist

• State whether detuning is laser minus cavity or cavity minus laser.
• State whether κ is an energy linewidth or field half-linewidth.
• Quote g_0, g, and all frequencies consistently in hertz or radians per second.
• Check red detuning yields positive damping under the selected convention.
• Include Stokes backaction in the final occupation.
• Verify stability before interpreting a linear response.
• Calibrate external coupling and cable or optical-path gain before fitting S_21.
• Report absorption-induced bath heating and technical phase noise.
• Use coupled-mode poles rather than rate equations in normal-mode splitting.
• Propagate uncertainties in n_c, g_0, κ, γ_m, and thermometry.

Optomechanics is powerful because one compact susceptibility describes cooling, gain, transparency, and conversion. Its apparent simplicity is also the trap: linewidth conventions, detuning signs, uncalibrated photon number, and hidden heating can move conclusions by orders of magnitude. A defensible result closes the loop between Hamiltonian, response function, noise spectrum, stability, and thermal budget.

## Extended data analysis and system identification

A cooling result should be reconstructed from raw observables rather than inferred from a narrower peak alone. Fit the complex cavity response off mechanical resonance to obtain κ, κ_ex, cable delay, and detuning. Next fit mechanical spectra at low power to obtain Ω_m and γ_m. Calibrate intracavity photon number from attenuation and independently from an optical-spring or damping sweep. Agreement between those routes tests the inferred g_0.

At every control power, record the apparent mechanical linewidth, integrated calibrated area, and center frequency. True cold damping increases linewidth while reducing area such that the inferred occupation follows the rate equation. Absorption heating can increase both bath occupation and linewidth. Frequency jitter broadens a time-averaged spectrum without increasing energy damping; ringdown distinguishes this inhomogeneous broadening from dissipation.

For a Lorentzian displacement spectrum near positive frequency,

The two terms encode positive- and negative-frequency quantum noise. A classical spectrum analyzer folds them into a convention-dependent single-sided trace. Therefore occupation extraction must state spectral convention, imprecision subtraction, integration bandwidth, and detector gain.

## Multimode, feedback, and conversion limits

Real structures contain nearby mechanical modes. Radiation pressure then contributes a self-energy matrix rather than a scalar, and optically mediated coupling can create bright and dark mechanical superpositions. A single-mode fit fails when mode spacing is comparable to optically broadened linewidths. Fit all coupled poles and verify that residues reproduce both magnitude and phase.

Measurement-based feedback cooling applies a force proportional to estimated velocity. Ideally it adds damping without the Stokes limit, but detector imprecision is fed back as force noise. Increasing gain eventually reheats the oscillator. Coherent feedback and reservoir engineering reshape noise without digitizing a record, yet loss channels still set added noise.

For bidirectional wavelength conversion with cavities a_1 and a_2, red pumps create H/ℏ = -g_1(a_1†b + a_1b†) - g_2(a_2†b + a_2b†). On resonance, an idealized internal efficiency is

It approaches unity for large matched cooperativities. External efficiency also multiplies the input and output coupling fractions. Mechanical thermal noise is suppressed by total cooperativity but is never removed by quoting internal efficiency alone.

## Reproducible reporting table

Finally, publish enough parameters to reproduce the curves: loaded and intrinsic linewidths, effective mass definition, zero-point amplitude, bath temperature, drive detuning, photon number, and whether quoted rates are divided by 2pi. These details turn an attractive transparency window or cooling trace into a quantitative optomechanical measurement.

## 16. Driven Optomechanical Hamiltonian and Linearization

For one optical cavity mode and one mechanical mode, the radiation-pressure Hamiltonian is

The single-photon coupling g₀ is the cavity-frequency shift produced by the mechanical zero-point displacement. A coherent control drive displaces the cavity field as a = α + d. Keeping terms linear in fluctuations gives the enhanced coupling

For a red-detuned drive near Δ = −Ω_m, the rotating-wave interaction swaps cavity and mechanical excitations. A blue-detuned drive instead produces two-mode amplification and can become unstable.

## 17. Optical Susceptibility, Spring, and Damping

Define the driven-cavity susceptibility consistently as

In one common self-energy convention, the optical spring and damping can be written in the requested form

Signs depend on the Fourier transform and detuning definition. A safer experimental check is that red detuning near −Ω_m increases damping and blue detuning near +Ω_m reduces it.

The anti-Stokes and Stokes scattering rates are often denoted A_- and A_+. Then Γ_opt = A_- − A_+. Some documents reverse the subscripts. If A_- in the following occupation formula is defined as the residual heating rate rather than the anti-Stokes cooling rate, the requested expression is

Under the widespread convention where A_+ is Stokes heating, replace A_- in the numerator by A_+. Declaring the convention prevents an otherwise serious cooling-limit error.

## 18. Resolved-Sideband Cooling

The resolved-sideband condition is

It allows a red-detuned control to enhance anti-Stokes scattering while suppressing Stokes scattering. In the ideal quantum-backaction limit, the residual occupation scales approximately as (κ/4Ω_m)². Ground-state cooling also requires optical damping to dominate thermal re-heating γ_m n_th without adding excessive absorption heating or technical laser noise.

Sideband thermometry compares Stokes and anti-Stokes motional sidebands. Calibration must account for detection gain, classical laser noise, cavity filtering, and correlations introduced by measurement backaction.

## 19. Optomechanically Induced Transparency

A strong red-detuned control field and weak probe create two coherent pathways to the cavity output: direct probe excitation and anti-Stokes scattering of the control by the mechanics. Near two-photon resonance, destructive interference opens a narrow transparency window inside the broader cavity response.

In the linearized regime, a useful probe response is

The OMIT linewidth increases approximately by Γ_opt, and its depth depends on external coupling and cooperativity. An apparent narrow feature can also arise from electrical feedthrough or thermal cavity drift, so phase and control-power scaling are important controls.

## 20. Python OMIT, Backaction, and Cooling Solver

This solver calculates the control-enhanced scattering rates and damping versus detuning, the cooled occupation, and the complex OMIT transmission spectrum.

import numpy as np
import matplotlib.pyplot as plt

Omega_m = 2*np.pi*5.20e6
kappa = 2*np.pi*0.62e6
kappa_e = 0.62*kappa
gamma_m = 2*np.pi*42.0
g0 = 2*np.pi*85.0
n_c = 3.0e7
g = g0*np.sqrt(n_c)
n_th = 820.0

def scattering_rates(Delta):
    A_minus = g**2*kappa/((kappa/2)**2+(Delta+Omega_m)**2)
    A_plus = g**2*kappa/((kappa/2)**2+(Delta-Omega_m)**2)
    return A_minus, A_plus

detuning = np.linspace(-2.2*Omega_m, 1.2*Omega_m, 1200)
Aminus, Aplus = scattering_rates(detuning)
Gamma_opt = Aminus-Aplus
stable = gamma_m+Gamma_opt > 0
occupation = np.full(detuning.shape, np.nan)
occupation[stable] = (gamma_m*n_th+Aplus[stable])/(gamma_m+Gamma_opt[stable])

# OMIT probe offset around mechanical resonance.
Delta_control = -Omega_m
probe_offset = np.linspace(Omega_m-4*kappa, Omega_m+4*kappa, 2400)
denominator = (kappa/2-1j*(Delta_control+probe_offset)
               +g**2/(gamma_m/2-1j*(probe_offset-Omega_m)))
S21 = 1.0-kappa_e/denominator

# Optical spring using a self-energy expression.
chi_plus = 1.0/(kappa/2-1j*(detuning+Omega_m))
chi_minus = 1.0/(kappa/2-1j*(detuning-Omega_m))
self_energy_difference = chi_plus-chi_minus
spring = g**2*np.real(self_energy_difference)

best = np.nanargmin(occupation)
print(f"Enhanced coupling g/2pi = {g/(2*np.pi):.3e} Hz")
print(f"Minimum modeled occupation = {occupation[best]:.4f}")
print(f"Best detuning / Omega_m = {detuning[best]/Omega_m:.4f}")

fig, axes = plt.subplots(2, 2, figsize=(12, 9))
axes[0, 0].plot(detuning/Omega_m, Gamma_opt/(2*np.pi))
axes[0, 0].axhline(0, color="black", lw=0.7)
axes[0, 0].set(xlabel="Control detuning / Ωm",
               ylabel="Optical damping (Hz)",
               title="Dynamical backaction")

axes[0, 1].plot(detuning/Omega_m, spring/(2*np.pi))
axes[0, 1].set(xlabel="Control detuning / Ωm",
               ylabel="Optical spring shift (Hz)")

axes[1, 0].semilogy(detuning[stable]/Omega_m, occupation[stable])
axes[1, 0].axhline(1, color="red", ls="--", label="ground-state boundary")
axes[1, 0].set(xlabel="Control detuning / Ωm",
               ylabel="Mechanical occupation")
axes[1, 0].legend()

axes[1, 1].plot((probe_offset-Omega_m)/(2*np.pi*1e3), abs(S21)**2)
axes[1, 1].set(xlabel="Two-photon detuning (kHz)",
               ylabel="|S21|²",
               title="OMIT transmission window")
for ax in axes.flat:
    ax.grid(alpha=0.3)
plt.tight_layout()
plt.show()

The chosen n_c is illustrative. A quantitative fit should include control-induced cavity shift, absorption heating, laser phase noise, multiple mechanical modes, complex microwave or optical background, and the exact input-output port geometry.

## 21. Stability and Experimental Checks

Blue-detuned dynamical anti-damping becomes unstable when γ_m + Γ_opt crosses zero. Red-detuned experiments can also become unstable through thermo-optic effects or free-carrier dynamics not present in the ideal Hamiltonian.

Report κ and whether it is an energy or amplitude linewidth, external-coupling fraction, Ω_m, γ_m, bath temperature, g₀ calibration, intracavity photon calibration, control detuning, absorption heating, cooperativity, final occupation, and thermometry method. OMIT linewidth alone does not establish ground-state cooling.