Sub-Doppler Cooling

Sub-Doppler cooling reduces a characteristic motional energy below the two-level Doppler scale of the optical transition used

$$k_{\mathrm B}T_D = \frac{\hbar\Gamma}{2}.$$

The name describes an achieved regime, not a unique force. Several distinct pieces of physics can evade the assumptions behind the Doppler result:

• polarization gradients make degenerate ground-state sublevels experience different position-dependent light shifts;
• optical pumping can reset an atom preferentially after it has climbed a light-shift potential;
• coherent superpositions can be dark only for selected velocity classes;
• Raman transitions can address kinetic-energy classes without relying on the optical linewidth;
• confinement can resolve discrete motional sidebands and make the vibrational ground state dark to a cooling transition.

These driven systems export energy and entropy through scattered light; thermal detailed balance is not required. Recoil heating remains, but scattering depends on position, velocity, and internal state. Some subrecoil protocols produce evolving, nonthermal distributions rather than a stationary temperature.

Required background. Laser cooling defines motional observables; Doppler cooling derives the two-level benchmark; AC Stark shifts supply the conservative potentials; hyperfine structure supplies the magnetic sublevels and optical-pumping pathways.

What the Doppler model leaves out

The standard Doppler limit assumes a broad, closed, nondegenerate two-level transition and a force obtained from the velocity dependence of its scattering rate. Real alkali ground manifolds instead contain Zeeman and hyperfine sublevels. The coupling strength depends on local polarization, and repeated scattering optically pumps population among those sublevels.

Three omissions are decisive.

Internal degeneracy

For total angular momentum $F>0$, the ground manifold contains $2F+1$ states $|F,m_F\rangle$. A local spherical polarization component $q=0,\pm1$ couples them with Clebsch–Gordan amplitude

$$\Omega_{m_F,q}(\mathbf r) \propto \mathcal E_q(\mathbf r) \langle F',m_F+q|d_q|F,m_F\rangle.$$

Different magnetic sublevels therefore see different light shifts and different optical-pumping rates.

Spatial polarization structure

Counterpropagating beams can have constant total intensity while their local polarization changes on the scale of a wavelength. An atom moving through that field samples a position-dependent internal Hamiltonian even when the scalar intensity is uniform.

Coherence and quantized motion

The internal steady state need not be an incoherent population. A coherent superposition can decouple from the light, and confinement can quantize motion into states $|n\rangle$. Neither effect is represented by a two-level rate equation for a continuous velocity.

Three different ways to suppress the Doppler floor. In polarization-gradient cooling, optical pumping occurs after an atom climbs a light-shift potential. Dark-state cooling suppresses scattering in a narrow velocity class. Raman sideband cooling removes one trapped motional quantum in the coherent step; ideal repumping preserves the remaining quantum number. The two internal ladders are drawn without their energy offset. The red sideband leaves $|\downarrow,0\rangle$ dark; real repumping adds recoil.

Scale hierarchy

The relevant energy and frequency scales should be written down before a mechanism is named:

$$\begin{aligned} E_r &= \frac{(\hbar k)^2}{2m}, & T_r &= \frac{E_r}{k_{\mathrm B}}, \\ E_D &= k_{\mathrm B}T_D = \frac{\hbar\Gamma}{2}, & U_{\mathrm{ls}} &\sim \frac{\hbar|\Omega|^2}{4|\Delta|}, \\ \hbar\gamma_p &\sim \hbar\Gamma \frac{|\Omega|^2}{4\Delta^2}, & \hbar\omega_t &= \text{trap level spacing}. \end{aligned}$$

Here $E_r$ is the one-photon recoil energy, $U_{\mathrm{ls}}$ is a representative differential light-shift depth, $\gamma_p$ is an optical-pumping rate in the far-detuned scaling regime, and $\omega_t$ is a trap frequency. Numerical factors and Clebsch–Gordan coefficients are transition dependent.

For a broad optical transition,

$$E_r \ll \hbar\Gamma.$$

This separation leaves a large interval between the recoil and Doppler scales. Polarization-gradient cooling commonly occupies that interval. When the distribution approaches a few recoils, momentum changes can no longer be treated as continuous diffusion; dark-state or motional-state descriptions become essential.

Polarization gradients

Constant intensity does not imply a uniform coupling

Consider one-dimensional lin-perpendicular-lin molasses. Two equal-frequency beams propagate along $\pm z$ with orthogonal linear polarizations:

$$\mathbf E^{(+)}(z,t) = E_0 \left( \mathbf e_x e^{ikz} + \mathbf e_y e^{-ikz} \right) e^{-i\omega_L t}.$$

Because $\mathbf e_x\cdot\mathbf e_y=0$, the cycle-averaged intensity is independent of $z$:

$$I(z) \propto \mathbf E^{(-)}\cdot\mathbf E^{(+)} = 2|E_0|^2.$$

The normalized Stokes parameters, however, rotate:

$$s_2(z) = \cos(2kz), \qquad s_3(z) = \sin(2kz).$$

Equivalently, up to the convention for handedness,

$$I_{\sigma^\pm}(z) = \frac{I}{2} \left[ 1 \pm \sin(2kz) \right].$$

The polarization changes from linear to circular, back to linear, and then to the opposite circular polarization over a distance $\lambda/2$. The intensity is flat while the couplings between magnetic sublevels are not.

Position-dependent light shifts

For atom-minus-laser detuning

$$\Delta = \omega_0-\omega_L,$$

red detuning has $\Delta>0$. For an isolated excited manifold at common detuning, far from resonance and at low saturation, a diagonal shift is

$$U_m(z)\simeq-\frac{\hbar}{4\Delta}\sum_q|\Omega_{m,q}(z)|^2.$$

Several excited manifolds generally have different detunings. The effective ground-manifold matrix is then

$$(H_{\rm ls})_{mn}=-\frac{\hbar}{4} \sum_e\frac{\Omega_{em}^*(z)\Omega_{en}(z)}{\Delta_e},$$

when the ground-state splittings can be neglected in the denominators. Off-diagonal Raman couplings mean that its eigenvalues, rather than its diagonal entries alone, are the adiabatic light-shift potentials. A common, spatially constant scalar shift produces no force. Differential shifts and optical pumping provide the multilevel cooling cycle.

The same off-resonant coupling produces optical pumping. A representative far-detuned rate scales as

$$\gamma_{m\rightarrow m'} \sim \Gamma \sum_q \frac{ |\Omega_{m,q}|^2 }{ 4\Delta^2 } B_{m,q\rightarrow m'},$$

where $B_{m,q\rightarrow m'}$ is a branching factor. Near resonance, the factor $4\Delta^2$ becomes $4(\Delta^2+\Gamma^2/4)$ in an isolated weakly driven transition; interference among excited hyperfine levels may matter.

Sisyphus cooling

Energy removed by optical pumping

A minimal Sisyphus cycle has four steps:

1. the atom occupies a magnetic sublevel whose light shift rises along its trajectory;
2. kinetic energy is converted into light-shift potential energy;
3. near the crest, optical pumping transfers the atom to a sublevel with a lower local light shift;
4. spontaneous emission exports the energy difference, and the cycle repeats.

If the atom climbs through a differential depth $U_0$ before pumping, the mean mechanical energy removed in a successful cycle is of order

$$\Delta E_{\mathrm{cool}} \sim U_0.$$

This is why the achievable scale is governed by light shifts rather than directly by $\hbar\Gamma$. Recoil from the pumping photon still heats the motion. The net change per cycle is only favorable when

$$\langle \Delta E_{\mathrm{cool}}\rangle > \langle \Delta E_{\mathrm{recoil}}\rangle + \langle \Delta E_{\mathrm{technical}}\rangle.$$

The spontaneous photon is not an incidental nuisance: it is the irreversible reset that makes repeated downhill transfers possible.

Why timing matters

Let the characteristic polarization period be of order $1/k$. An atom moving at speed $v$ traverses it at rate $kv$. Optical pumping acts at rate $\gamma_p$. The largest hill-climbing energy loss per cycle often occurs at comparable scales:

$$kv \sim \gamma_p.$$

If $kv\gg\gamma_p$, the atom crosses many hills before its internal state changes. If $kv\ll\gamma_p$, it is pumped before climbing appreciably. This gives the rough capture scale

$$v_c \sim \frac{\gamma_p}{k},$$

as an order-of-magnitude width of the low-velocity force, not a sharp boundary. Finite linear friction can remain when $kv\ll\gamma_p$, even though the energy removed per individual cycle is small. An atom with $mv^2/2>U_0$ can cross and lose energy on many hills: $mv^2/2\lesssim U_0$ is not a necessary condition for cooling. Finite illumination time and the accumulated energy loss determine capture. At energies below the barriers, localized motion requires a more detailed treatment.

Friction and diffusion

At sufficiently low velocity, a polarization-gradient force can again be written

$$F(v) \simeq -\alpha_{\mathrm{PG}}v.$$

But neither $\alpha_{\mathrm{PG}}$ nor the momentum-diffusion coefficient is given by the two-level Doppler expressions. Both must be derived from the multilevel optical Bloch or master equation, including position-dependent couplings and recoil.

In a broad semiclassical Sisyphus regime, a common scaling is

$$k_{\mathrm B}T \sim C\,U_0,$$

where $C$ depends on angular momenta, detuning, polarization geometry, and dimensionality. This scaling explains the often-observed approximately linear dependence of temperature on intensity at low saturation. It should not be promoted to a universal temperature formula.

Lin-perpendicular-lin and circular configurations

The potential-hill picture is clearest for lin-perpendicular-lin light. In a counterpropagating circular-polarization geometry, the local polarization may have constant ellipticity while its orientation or internal-state basis varies. Cooling then involves motion-induced coherences and population transfer among Zeeman sublevels. It is still called polarization-gradient cooling, but a literal scalar hill-and-reset cartoon can be misleading.

The robust statement is:

Motion through a spatially varying multilevel coupling creates a lag between internal-state adaptation and the local dressed-state energy.

That lag can produce friction below the two-level Doppler scale.

Magnetic-field sensitivity

Sub-Doppler coherences and optical pumping operate on scales far smaller than the optical linewidth. A residual magnetic field introduces a Larmor frequency

$$\omega_B = \frac{|g_F|\mu_{\mathrm B}B}{\hbar}.$$

Cooling can be strongly perturbed when $\omega_B$ becomes comparable to the optical-pumping rate or differential light-shift frequency:

$$\omega_B \gtrsim \min \left( \gamma_p, \frac{U_0}{\hbar} \right).$$

The numerical threshold depends on geometry and participating states. This scale comparison explains why a field tiny compared with the field required to Zeeman-shift an optical transition by $\Gamma$ can still destroy sub-Doppler cooling.

Dark-state cooling

The Lambda-system dark state

Consider two ground states coupled to a common excited state:

$$|g_1\rangle \xleftrightarrow{\ \Omega_1\ } |e\rangle \xleftrightarrow{\ \Omega_2\ } |g_2\rangle.$$

In the rotating-wave approximation, the interaction is

$$H_{\mathrm{int}} = \frac{\hbar}{2} \left( \Omega_1|e\rangle\langle g_1| + \Omega_2|e\rangle\langle g_2| + \text{h.c.} \right).$$

At two-photon resonance, define

$$\Omega = \sqrt{ |\Omega_1|^2 + |\Omega_2|^2 }.$$

The superposition

$$|D\rangle = \frac{ \Omega_2|g_1\rangle - \Omega_1|g_2\rangle }{ \Omega }$$

obeys

$$H_{\mathrm{int}}|D\rangle = 0.$$

It is dark because the two excitation amplitudes cancel, not because either beam is weak. The orthogonal bright state couples to $|e\rangle$ and scatters.

Velocity selection

For wave vectors $\mathbf k_1$ and $\mathbf k_2$, motion changes the two-photon detuning:

$$\delta_2(\mathbf v) = \delta_2(\mathbf 0) - \Delta\mathbf k\cdot\mathbf v, \qquad \Delta\mathbf k = \mathbf k_1-\mathbf k_2.$$

Here $\delta_2$ is laser difference frequency minus the internal resonance, with any recoil and differential-light-shift corrections included in $\delta_2(\mathbf0)$. The selected velocity is its component along $\Delta\mathbf k$. Nearly copropagating beams with $|\Delta\mathbf k|\simeq0$ have little velocity selectivity. If the resonance’s effective half-width is $\gamma_D$, the corresponding velocity width scales as

$$\Delta v_D \sim \frac{\gamma_D}{|\Delta\mathbf k|}.$$

A useful schematic scattering profile is

$$R_{\mathrm{sc}}(v) \simeq R_b \frac{ \delta_2(v)^2 }{ \delta_2(v)^2+\gamma_D^2 },$$

where $R_b$ is the bright-state scattering scale. Atoms outside the dark window scatter and receive random recoil; some are transferred into the dark velocity class, where scattering is suppressed. Cooling is therefore accumulation in a low-scattering state rather than a simple linear drag.

Ground-state decoherence, differential AC Stark shifts, magnetic-field inhomogeneity, laser phase noise, and collisions all broaden $\gamma_D$ and raise the final kinetic scale.

Velocity-selective coherent population trapping

Velocity-selective coherent population trapping, or VSCPT, treats internal and translational states together. In an ideal symmetric one-dimensional model, the dark state near zero mean momentum has the form

$$|D_0\rangle \propto |g_1,-\hbar k\rangle - |g_2,+\hbar k\rangle.$$

The two components absorb into the same excited momentum state, so their amplitudes cancel. Momentum classes away from resonance are bright and are recycled by spontaneous emission.

For the ideal normalized state above, $\langle p\rangle=0$ but $\langle p^2\rangle=(\hbar k)^2$, even when each peak has zero width. The kinetic energy is then $E_r$, not zero: subrecoil peak width does not imply subrecoil total rms momentum. Internal-state-resolved or coherently transferred distributions must be distinguished from the unresolved pair.

The resulting distribution can have narrow peaks and long nonthermal tails. Near or below the recoil scale, quoting a single temperature may conceal the relevant physics. Report the momentum distribution, rms width, kinetic energy, and dark-state population.

Gray molasses

Gray molasses combines dark-state protection with polarization-dependent bright-state light shifts. Spatially varying bright and dark eigenstates make moving atoms nonadiabatically enter a bright state, climb a repulsive light-shift landscape, and optically pump back into a dark state. Slow atoms remain dark for longer and scatter less. Typical gray molasses uses blue one-photon detuning ($\Delta<0$ here), giving repulsive bright-state shifts. Darkness alone does not guarantee cooling: the spatial couplings and pumping must preferentially remove kinetic energy.

In a $\Lambda$-enhanced implementation, two optical frequencies address ground hyperfine states through a common excited manifold. Performance is sharply sensitive to the two-photon Raman detuning. The method is especially valuable when unresolved excited structure or small hyperfine splittings make ordinary polarization-gradient cooling inefficient.

Gray molasses is not “Sisyphus cooling without spontaneous emission.” Scattering remains the dissipative reset; the adjective gray means that the low-energy state is weakly coupled rather than uniformly bright.

Raman cooling of free particles

Two-photon Raman transitions can select velocity much more narrowly than an optical linewidth. For a transition that changes momentum by $\hbar\Delta\mathbf k$, energy conservation gives

$$\omega_1-\omega_2 = \omega_{21} + \Delta\mathbf k\cdot\mathbf v + \frac{ \hbar|\Delta\mathbf k|^2 }{ 2m }.$$

The last term is the two-photon recoil shift; the formula assumes negligible differential light shifts or that they have been subtracted. See Raman spectroscopy for the effective coupling. A cooling sequence uses:

1. a velocity-selective Raman pulse that transfers atoms in a chosen moving class to another internal state and changes their momentum toward zero;
2. optical pumping that returns the internal state;
3. a pulse schedule whose spectral response leaves a narrow dark interval around zero velocity.

The coherent step can remove more directed kinetic energy than the optical pumping step adds on average. By narrowing the unaddressed velocity window over successive pulses, free-particle Raman cooling can produce subrecoil, strongly non-Gaussian distributions.

This is distinct from Raman sideband cooling. Free Raman cooling selects a continuum of velocities. Raman sideband cooling selects discrete vibrational quantum numbers in a trap.

Raman sideband cooling

Resolved motional quanta

Approximate a trapped direction as a harmonic oscillator:

$$H_{\mathrm{mot}} = \hbar\omega_t \left( a^\dagger a + \frac{1}{2} \right).$$

Two long-lived internal states form $|\downarrow,n\rangle$ and $|\uparrow,n\rangle$. A Raman pair with effective wave-vector difference $\Delta\mathbf k$ has Lamb–Dicke parameter

$$\eta = |\Delta\mathbf k\cdot\mathbf e_t|x_0, \qquad x_0 = \sqrt{ \frac{\hbar}{2m\omega_t} }.$$

Here $\mathbf e_t$ is the trap-axis unit vector. In the Lamb–Dicke regime,

$$\eta\sqrt{2n+1} \ll 1,$$

the Raman coupling can be expanded to first order in displacement. Optical repumping has separate absorption and emission Lamb–Dicke parameters, set by its single-photon wave vectors. A nearly copropagating Raman pair can have small $\eta$ while its optical reset still causes appreciable recoil heating.

Cooling cycle

Tune the Raman difference frequency to the red motional sideband:

$$|\downarrow,n\rangle \longrightarrow |\uparrow,n-1\rangle.$$

For small $\eta$, its Rabi frequency scales as

$$\Omega_{n,n-1} \simeq \eta\Omega_0\sqrt n.$$

Optical pumping then resets the internal state,

$$|\uparrow,n-1\rangle \longrightarrow |\downarrow,n-1\rangle,$$

while approximately preserving $n$ in the Lamb–Dicke regime. Each ideal cycle removes $\hbar\omega_t$. Because

$$\Omega_{0,-1} = 0,$$

$|\downarrow,0\rangle$ is dark to the red sideband.

During the coherent step, the Raman spectral width, pulse bandwidth, and off-resonant carrier excitation must be small compared with $\omega_t$. A simple sufficient weak-drive condition includes $\Omega_0\ll\omega_t$, not merely $\Omega_{\rm sb}\ll\omega_t$: a small Lamb–Dicke factor can make the sideband weak while the carrier remains too strong. In a sequential protocol, a broad optical repumping line need not resolve sidebands because it is off during the Raman pulse. Its recoil must still be small. If repumping overlaps the coherent drive, its induced broadening enters the sideband-resolution condition as well.

Rate model and endpoint

Let $nA_-$ and $(n+1)A_+$ be the downward and upward transition rates. This birth–death model gives, for any distribution with finite mean,

$$\frac{d\bar n}{dt} = - \left( A_--A_+ \right) \bar n + A_+.$$

If $A_->A_+$, the stationary occupation is

$$\bar n_{\mathrm{ss}} = \frac{A_+}{A_--A_+}.$$

For constant coefficients and no additional processes, detailed balance of this effective motional chain gives the stationary geometric distribution $P_n=(1-r)r^n$, $r=A_+/A_-$. The driven atom–light system as a whole need not satisfy thermal detailed balance.

The endpoint is therefore set by off-resonant carrier and blue-sideband excitation, repumping recoil, trap-frequency noise, anharmonicity, and background heating. “Ground-state cooled” should be supported by a measured $\bar n$ or ground-state probability, not inferred merely from a narrow spatial image.

Sideband thermometry

For a thermal oscillator,

$$\bar n = \frac{1}{ \exp( \hbar\omega_t/k_{\mathrm B}T ) -1 }, \qquad P_0 = \frac{1}{1+\bar n}.$$

In the weak-probe limit, the integrated red- and blue-sideband strengths obey

$$\frac{I_{\mathrm{red}}}{I_{\mathrm{blue}}} = \frac{\bar n}{\bar n+1} = \exp \left( - \frac{\hbar\omega_t}{k_{\mathrm B}T} \right).$$

The ratio $\bar n/(1+\bar n)$ also holds for a nonthermal diagonal distribution in the weak, resolved, first-sideband limit. Its conversion to $T$ or $P_0=1/(1+\bar n)$ requires a thermal distribution. Probe saturation, line overlap, and unequal detection weights can invalidate the ratio.

Comparing rubidium scales

For $^{87}\mathrm{Rb}$ at $780\,\mathrm{nm}$, $T_D\simeq146\,\mu\mathrm K$ while $T_r=E_r/k_{\rm B}\simeq181\,\mathrm{nK}$; the calculation is given in laser cooling. A $10\,\mu\mathrm K$ sample is sub-Doppler but still about $55T_r$. A semiclassical polarization-gradient description may work there, whereas recoil-scale distributions need discrete momentum transfer. For a trapped sample, $\bar n$ additionally depends on $\omega_t$. Sub-Doppler cooling, subrecoil peak width, and motional ground-state occupation measure different properties.

Mechanism comparison

Mechanism	Selected coordinate	Dissipative reset	Characteristic endpoint	Useful diagnostic
two-level Doppler	velocity through optical detuning	spontaneous emission	$\hbar\Gamma$ scale	damping plus recoil diffusion
polarization-gradient Sisyphus	position and magnetic sublevel	optical pumping	differential light-shift scale	polarization and field dependence
gray molasses	local bright/dark dressed state	weak bright-state scattering	dark-resonance width and recoil	Raman feature in a Lambda-enhanced scheme
VSCPT	internal–momentum superposition	recycling of bright momentum classes	subrecoil width possible	non-Gaussian dark momentum peak
free Raman cooling	continuum velocity class	optical pumping between coherent pulses	pulse-defined dark window	velocity-selective Raman spectrum
Raman sideband cooling	trapped vibrational number	internal-state repumping	$\bar n\ll1$ possible	resolved red/blue sideband asymmetry

The table is a diagnostic map, not a ranking. A practical experiment often uses several rows in sequence: broad-line Doppler capture, polarization- gradient or gray-molasses cooling, trap loading, and then sideband cooling.

Distinguishing cooling mechanisms

A temperature below $T_D$ does not identify a mechanism. Vary polarization, magnetic field, intensity, and Raman detuning while measuring the full motional distribution. Near recoil, Gaussian fits can hide broad wings and multiple narrow peaks. Compare the Larmor frequency with pumping and dark linewidths, not only with the much larger optical linewidth.

Track atom number as well as width. Preferential escape of hot atoms can cool the survivors through evaporation; it is a different entropy-removal mechanism from laser damping. A narrower surviving sample alone does not establish which process occurred. In a trap, determine mode-resolved occupations rather than inferring ground-state cooling from a small image.

Exercises

1. Polarization without an intensity gradient

For

$$\mathbf E^{(+)}(z) = E_0 \left( \mathbf e_x e^{ikz} + \mathbf e_y e^{-ikz} \right),$$

show that the intensity is constant and derive the circular-component intensities. Identify positions of pure circular polarization.

Solution

Orthogonality eliminates the interference term in the scalar intensity:

$$\begin{aligned} |\mathbf E^{(+)}|^2 &= |E_0|^2 \left( |\mathbf e_x|^2 + |\mathbf e_y|^2 \right) \\ &= 2|E_0|^2. \end{aligned}$$

Using circular basis vectors, the normalized component intensities are, up to which handedness is called plus,

$$\frac{I_{\sigma^\pm}}{I} = \frac{1}{2} \left[ 1 \pm \sin(2kz) \right].$$

One component vanishes and the other carries all the intensity when $\sin(2kz)=\pm1$, namely

$$z = \frac{\lambda}{8} + n\frac{\lambda}{4}.$$

The handedness alternates between neighboring positions.

2. Energy removed in an ideal Sisyphus cycle

Model two shifted potentials as

$$U_\pm(z) = \frac{U_0}{2} \left[ 1 \pm \cos(2kz) \right].$$

An atom begins at a minimum of $U_-$, climbs to its maximum, and is pumped to $U_+$. Assume enough initial kinetic energy to reach the crest and neglect recoil. How much mechanical energy is removed?

Solution

At a minimum,

$$U_-^{\min} = 0.$$

At the corresponding crest,

$$U_-^{\max} = U_0, \qquad U_+^{\min} = 0.$$

The atom converts $U_0$ of kinetic energy into potential energy while climbing. Optical pumping changes the internal dressed state at fixed position and releases the potential difference to the light field and spontaneously emitted photon. The ideal mechanical energy loss is therefore

$$\Delta E_{\mathrm{mech}} = -U_0.$$

Real cycles pump over a distribution of positions and add recoil, so the mean magnitude is smaller.

3. Narrow peaks and total kinetic energy

An unresolved momentum distribution contains equal Gaussian peaks centered at $\pm\hbar k$, each with variance $\sigma_p^2$. Find its mean kinetic energy. Does $\sigma_p\ll\hbar k$ imply energy below $E_r$?

Solution

The mean momentum vanishes. The law of total variance gives $\langle p^2\rangle=(\hbar k)^2+\sigma_p^2$, so $\langle E_{\rm kin}\rangle=E_r+\sigma_p^2/(2m)$. The peaks can individually be arbitrarily narrow while the unresolved kinetic energy approaches $E_r$ from above. The peak separation must be included in the reported rms width.

4. Verify the dark state

For the $\Lambda$ interaction Hamiltonian in the text, verify that

$$|D\rangle = \frac{ \Omega_2|g_1\rangle-\Omega_1|g_2\rangle }{ \sqrt{|\Omega_1|^2+|\Omega_2|^2} }$$

has no excited-state coupling.

Solution

Only the excitation terms contribute:

$$\begin{aligned} H_{\mathrm{int}}|D\rangle &= \frac{\hbar}{2\Omega} |e\rangle \left( \Omega_1\Omega_2 - \Omega_2\Omega_1 \right) \\ &= 0. \end{aligned}$$

The cancellation holds for arbitrary fixed complex couplings. Changing a constant relative laser phase changes the dark superposition; it does not remove its existence. Phase fluctuations, two-photon detuning, or loss of ground-state coherence can prevent the prepared state from remaining dark.

5. Dark-resonance velocity width

Counterpropagating Raman beams have nearly equal wave-number magnitude $k$, so $|\Delta\mathbf k|\simeq2k$. If the dark resonance has angular-frequency half-width $\gamma_D/2\pi=20\,\mathrm{kHz}$ at $\lambda=780\,\mathrm{nm}$, estimate the one-dimensional velocity half-width.

Solution

The velocity width is

$$\Delta v_D \sim \frac{\gamma_D}{2k}.$$

Using

$$\gamma_D = 2\pi(20\times10^3)\,\mathrm{s^{-1}}, \qquad k = \frac{2\pi}{780\times10^{-9}\,\mathrm m},$$

gives

$$\Delta v_D \simeq 7.8\times10^{-3}\,\mathrm{m\,s^{-1}}.$$

This is comparable to a single-photon recoil velocity for rubidium, so a continuous classical diffusion model is already becoming questionable.

6. Recoil cost of resetting the internal state

A successful red-sideband pulse leaves an oscillator state with $\langle a\rangle=0$ after removing one motional quantum. Model its reset as $N$ instantaneous absorption–emission pairs in the same harmonic trap. Each absorption has the same signed axial Lamb–Dicke parameter $\eta_{\rm abs}$. The independent emission projections have zero mean and second moment $\langle\eta_{\rm em}^2\rangle$.

Compare the mean net change in occupation for (a) independent, uniformly sampled trap phases at the scattering events and (b) a burst much shorter than $1/\omega_t$. Does independence of the emitted photons alone justify adding the single-event absorption recoil energies? Explain why small Raman $\eta$ is not sufficient for cooling.

Solution

An axial momentum impulse $\hbar k_\parallel$ acts as $e^{ik_\parallel x}$. In the oscillator interaction picture, its displacement is $i\eta e^{i\omega_t t_j}$, where $\eta=k_\parallel\sqrt{\hbar/(2m\omega_t)}$ and $t_j$ is the kick time. Displacements add, up to an overall phase. Since the initial $\langle a\rangle$ vanishes, a net displacement $\beta$ adds $|\beta|^2$ quanta; see Coherent States. For fixed event times, averaging the independent, zero-mean emissions gives

$$\Delta\bar n =-1+\eta_{\rm abs}^2 \left|\sum_{j=1}^{N}e^{i\omega_t t_j}\right|^2 +N\langle\eta_{\rm em}^2\rangle.$$

In (a), the trap-phase average removes all cross terms between distinct absorption kicks, giving

$$\Delta\bar n_{\rm phase\,averaged} =-1+N\bigl(\eta_{\rm abs}^2+ \langle\eta_{\rm em}^2\rangle\bigr).$$

In (b), the phases scarcely rotate, so instead

$$\Delta\bar n_{\rm burst} \simeq-1+N^2\eta_{\rm abs}^2+ N\langle\eta_{\rm em}^2\rangle.$$

The $N^2$ contribution is coherent center-of-mass excitation; it is part of the total oscillator energy even though it is not an increase of the centered momentum variance. Independent emissions do not remove it. For example, $N=4$, $\eta_{\rm abs}=0.25$, and $\langle\eta_{\rm em}^2\rangle=0.02$ give net changes $-0.67$ and $+0.08$: the first protocol cools, while the second heats. If $N$ fluctuates, the burst uses $\mathbb E[N^2]$ for the directed term and $\mathbb E[N]$ for the emission term.

The reset must add less than one quantum on average. Reducing the Raman difference wave vector does not suppress the reset photons’ recoil or control its timing. This impulse model isolates that distinction; it does not replace a driven multilevel reset calculation. Wineland et al., Secs. 3.1 and 3.2.3, explain the separate reset and motional-displacement ingredients.

7. The stationary motional distribution

For rates $nA_-$ downward and $(n+1)A_+$ upward, derive the stationary $P_n$ and the relaxation of $\bar n$. Must the initial state be thermal?

Solution

Zero stationary flux between $n$ and $n+1$ gives $(n+1)A_+P_n=(n+1)A_-P_{n+1}$. If $r=A_+/A_-<1$, normalization gives $P_n=(1-r)r^n$ and $\bar n_{\rm ss}=r/(1-r)$. Summing the master equation weighted by $n$ gives

$$\bar n(t)=\bar n_{\rm ss}+ [\bar n(0)-\bar n_{\rm ss}]e^{-(A_--A_+)t}.$$

The mean equation closes even for an initially nonthermal state. Additional nonlinear rates or non-Markovian dynamics need not preserve this result.

8. Diagnose a mechanism claim

An alkali cloud reaches $8\,\mu\mathrm K$, below its Doppler temperature. The temperature rises sharply with a residual magnetic field, scales nearly linearly with cooling intensity, and changes little when a two-photon Raman detuning is scanned. Which mechanism is best supported, and what additional test would you perform?

Solution

The magnetic sensitivity and approximate light-shift scaling support polarization-gradient Sisyphus cooling. The absence of a narrow two-photon-detuning feature argues against a $\Lambda$ dark-state mechanism as the dominant effect.

A discriminating test is to change the relative beam polarizations while holding total intensity and one-photon detuning fixed. Suppressing the polarization gradient should weaken the cooling. One should simultaneously measure atom number to rule out velocity-selective loss.

References

• A. Aspect, E. Arimondo, R. Kaiser, N. Vansteenkiste, and C. Cohen-Tannoudji, “Laser cooling below the one-photon recoil energy by velocity-selective coherent population trapping,” Physical Review Letters 61, 826–829 (1988), doi:10.1103/PhysRevLett.61.826.
• Y. Castin, H. Wallis, and J. Dalibard, “Limit of Doppler cooling,” Journal of the Optical Society of America B 6, 2046–2057 (1989), doi:10.1364/JOSAB.6.002046.
• C. N. Cohen-Tannoudji, “Nobel Lecture: Manipulating atoms with photons,” Reviews of Modern Physics 70, 707–719 (1998), doi:10.1103/RevModPhys.70.707.
• C. Cohen-Tannoudji and D. Guéry-Odelin, Advances in Atomic Physics: An Overview (World Scientific, 2011), doi:10.1142/6631.
• J. Dalibard and C. Cohen-Tannoudji, “Laser cooling below the Doppler limit by polarization gradients: simple theoretical models,” Journal of the Optical Society of America B 6, 2023–2045 (1989), doi:10.1364/JOSAB.6.002023.
• A. T. Grier, I. Ferrier-Barbut, B. S. Rem, M. Delehaye, L. Khaykovich, F. Chevy, and C. Salomon, “$\Lambda$-enhanced sub-Doppler cooling of lithium atoms in D$_1$ gray molasses,” Physical Review A 87, 063411 (2013), doi:10.1103/PhysRevA.87.063411.
• S. E. Hamann, D. L. Haycock, G. Klose, P. H. Pax, I. H. Deutsch, and P. S. Jessen, “Resolved-sideband Raman cooling to the ground state of an optical lattice,” Physical Review Letters 80, 4149–4152 (1998), doi:10.1103/PhysRevLett.80.4149.
• M. Kasevich and S. Chu, “Laser cooling below a photon recoil with three-level atoms,” Physical Review Letters 69, 1741–1744 (1992), doi:10.1103/PhysRevLett.69.1741.
• P. D. Lett, R. N. Watts, C. I. Westbrook, W. D. Phillips, P. L. Gould, and H. J. Metcalf, “Observation of atoms laser cooled below the Doppler limit,” Physical Review Letters 61, 169–172 (1988), doi:10.1103/PhysRevLett.61.169.
• H. J. Metcalf and P. van der Straten, Laser Cooling and Trapping (Springer, 1999), doi:10.1007/978-1-4612-1470-0.
• C. Monroe, D. M. Meekhof, B. E. King, S. R. Jefferts, W. M. Itano, D. J. Wineland, and P. Gould, “Resolved-sideband Raman cooling of a bound atom to the 3D zero-point energy,” Physical Review Letters 75, 4011–4014 (1995), doi:10.1103/PhysRevLett.75.4011.
• W. D. Phillips, “Nobel Lecture: Laser cooling and trapping of neutral atoms,” Reviews of Modern Physics 70, 721–741 (1998), doi:10.1103/RevModPhys.70.721.
• J. Reichel, F. Bardou, M. Ben Dahan, E. Peik, S. Rand, C. Salomon, and C. Cohen-Tannoudji, “Raman cooling of cesium below 3 nK,” Physical Review Letters 75, 4575–4578 (1995), doi:10.1103/PhysRevLett.75.4575.
• P. J. Ungar, D. S. Weiss, E. Riis, and S. Chu, “Optical molasses and multilevel atoms: theory,” Journal of the Optical Society of America B 6, 2058–2071 (1989), doi:10.1364/JOSAB.6.002058.
• D. J. Wineland, C. Monroe, W. M. Itano, D. Leibfried, B. E. King, and D. M. Meekhof, “Experimental issues in coherent quantum-state manipulation of trapped atomic ions,” Journal of Research of the National Institute of Standards and Technology 103, 259–328 (1998), doi:10.6028/jres.103.019, open text.