Sub-Doppler Cooling and Sisyphus Mechanism

For a two-level atom the story of laser cooling ends at the Doppler limit $T_D=\hbar\Gamma/2k_B$ — roughly $235\,\mu$ K for sodium — where the viscous drag of red-detuned light balances the momentum diffusion of random photon scattering. In 1988 the NIST group released sodium optical molasses and clocked its ballistic expansion at $43\,\mu$ K — five times colder than a two-level atom is allowed to be, and impossible in Doppler theory. The missing ingredient was the internal structure quantum mechanics hands every real alkali — a manifold of ground-state Zeeman sublevels — together with two effects this article takes as given from the atom–light chapter: the light shift (AC Stark shift) a laser imposes on each sublevel, and optical pumping, the scattering that shuffles population between them. Standing waves supply the third ingredient, a polarization gradient: two counter-propagating beams with orthogonal polarizations build a field whose polarization state, not just its intensity, cycles on the scale of the wavelength. Each ground sublevel then feels a light shift keyed to the local polarization, so the sublevel energies become spatially periodic potentials — corrugated hills and valleys locked to the standing wave, with position-dependent optical pumping tying the atom's internal state to where it sits. The result is the Sisyphus mechanism: a moving atom is repeatedly caught climbing a hill, then pumped back to a well bottom, converting kinetic energy into light-shift energy that departs with a blue-shifted photon. Temperatures plunge toward a small multiple of the recoil energy, and the microkelvin regime — the doorway to Bose–Einstein condensation — swings open.

Key takeaways

Sub-Doppler cooling reaches temperatures far below $T_D=\hbar\Gamma/2k_B$ by exploiting the atom's ground-state Zeeman sublevels, which acquire spatially modulated light shifts inside a polarization gradient.
In the lin⟂lin (Sisyphus) geometry two orthogonally polarized counter-propagating beams make the polarization cycle $\sigma^+\to$ lin $\to\sigma^-$ every $\lambda/4$ , modulating the $m_g$ sublevel potentials a quarter wavelength out of phase.
A moving atom climbs a hill in one sublevel, is optically pumped near the top into a valley of the other, and climbs again — always uphill — dumping kinetic energy into blue-shifted light.
The equilibrium temperature scales as $k_BT\propto U_0\propto s_0/|\delta|$ — it falls linearly with intensity, opposite to the intensity-independent Doppler limit, and that scaling was the experimental smoking gun.
A distinct $\sigma^+\sigma^-$ mechanism cools with uniform light shifts through a velocity-induced population orientation, giving friction with no Sisyphus hills.
Key idea: when an atom has several ground sublevels, a spatially varying polarization modulates their light-shift potentials; optical pumping keeps a moving atom mostly climbing hills, continuously converting kinetic energy into light-shift potential energy that leaves with scattered photons — Sisyphus cooling.

The core insight

When an atom has several ground sublevels, a spatially varying polarization makes their light-shift potentials spatially modulated; a moving atom is optically pumped so that it spends most of its time climbing hills, continuously converting kinetic energy into light-shift potential energy that departs with scattered photons — Sisyphus cooling.

Where this lives today

Every laboratory that makes ultracold atoms uses this mechanism, usually without ceremony. In a magneto-optical trap the magnetic gradient and scattering force capture a cloud, but the final temperature is set by a brief molasses stage: the field is ramped to zero and the beams detuned further red, and polarization-gradient cooling drops the cloud to tens of microkelvin in milliseconds before it is handed to a magnetic or optical dipole trap. That low starting temperature and high phase-space density are what make the subsequent evaporative cooling to Bose–Einstein condensation feasible within a finite trap lifetime. For species where lin⟂lin cooling is inefficient — lithium and potassium, with their compressed excited states — gray molasses on the $D_1$ line uses dark states to combine Sisyphus dissipation with suppressed reabsorption, reaching a few microkelvin. The same physics sets the floor in the caesium and rubidium fountain clocks that define the SI second, and in the ultracold sources feeding atom-interferometer gravimeters and inertial sensors.

The physical picture

Picture the two ground sublevels as interleaved corrugated landscapes, shifted by half a period so every valley of one sits under a hilltop of the other. Optical pumping is a one-way trapdoor: whenever an atom drifts near the top of a hill in the landscape it occupies, the local polarization is exactly the one that pumps it out of that sublevel and into a valley of the other. Having fallen, a moving atom can only climb again. Averaged over many cycles it is almost always going uphill, paying kinetic energy at every rise — the punishment of Sisyphus, run in reverse. The energy budget is transparent: the atom absorbs a photon near a potential maximum and emits one that is, on average, blue-shifted, carrying away $U_0$ plus a recoil. The light field pockets a sliver of the atom's mechanical energy every turn.

Figure

The Sisyphus landscape the two ground sublevels m = +1/2 and −1/2 see light-shift potentials modulated a quarter wavelength out of phase (blue cos²kz, red sin²kz). A moving atom climbs a blue hill losing kinetic energy, is optically pumped near the top (green arrow) into a red valley, then climbs again — always uphill. Depth here U₀ ≈ 12 μK for rubidium at s₀ = 1, δ = −3Γ; polarization cycles σ⁺ → lin → σ⁻ every λ/4

Why so cold? Because the well depth $U_0\sim\hbar\Omega^2/(4|\delta|)$ is a knob we control — large detuning and modest intensity make it small — whereas the Doppler limit $\hbar\Gamma/2$ is fixed by nature. Cooling continues until the kinetic energy is comparable to a well depth, below which the atom can no longer climb out and the engine stalls, so $k_BT\sim U_0$ , tunable toward the recoil energy. That is a qualitatively new prediction: lower the intensity and the temperature should fall, the fingerprint the second figure shows. The $\sigma^+\sigma^-$ configuration cools by a subtler route with no wells at all, but the lin⟂lin picture is the one to hold in your head.

Figure

The experimental fingerprint sub-Doppler temperature falls linearly with laser intensity, k_B T ∝ U₀ ∝ s₀ — the opposite of the Doppler limit, which is fixed at T_D = 235 μK for sodium no matter the intensity (far above the top of this frame). Cooling floors a few times above the single-photon recoil temperature T_r ≈ 2.4 μK. Points sketch rubidium-like molasses at detuning δ = −3Γ

The core equations

Everything begins with the benchmark that sub-Doppler cooling beats: Doppler theory balances the drag of a red-detuned two-level atom against the diffusion of spontaneous emission and finds a floor set only by the linewidth.

T_D = \frac{\hbar\Gamma}{2k_B} \qquad (1)

where: $\Gamma$ — natural linewidth (spontaneous decay rate); $\hbar$ — reduced Planck constant; $k_B$ — Boltzmann constant. The limit depends on nothing tunable — one number per atom ( $235\,\mu$ K Na, $146\,\mu$ K Rb, $125\,\mu$ K Cs) that two-level theory cannot beat. The heart of Sisyphus cooling is that the sublevels acquire periodic potentials. In the $J_g=1/2$ model the two sublevels trace $\cos^2$ and $\sin^2$ along the standing wave, offset by $\lambda/4$ : $m_g=+1/2$ shifts most where the field is $\sigma^+$ , $m_g=-1/2$ where it is $\sigma^-$ .

U_+(z)=U_0\cos^2(kz),\qquad U_-(z)=U_0\sin^2(kz) \qquad (2)

where: $U_\pm$ — light-shift potential of the $m_g=\pm1/2$ sublevel; $U_0$ — peak-to-valley depth; $k=2\pi/\lambda$ . A valley of one landscape sits beneath a hilltop of the other — exactly the geometry the trapdoor of optical pumping needs. The depth $U_0$ is the ground-state AC Stark shift. Its exact form saturates logarithmically, but far-detuned and weakly saturated it collapses to a clean form proportional to intensity and inverse detuning.

U_0 = \frac{\hbar\delta}{2}\,\ln\!\left(1+\frac{s_0}{1+(2\delta/\Gamma)^2}\right)\;\xrightarrow{|\delta|\gg\Gamma,\;s_0\ll1}\;\frac{\hbar\Omega^2}{4\delta} \qquad (3)

where: $\delta=\omega_L-\omega_0$ — detuning (negative for cooling); $s_0=I/I_{\rm sat}$ — saturation parameter per beam; $\Omega=\Gamma\sqrt{s_0/2}$ — Rabi frequency. Since $U_0\propto s_0/\delta$ , the wells shrink at lower power or larger detuning — the mechanism by which temperature falls. The Sisyphus cycle is clocked by optical pumping, which far-detuned is just the photon scattering rate and falls off more steeply with detuning than $U_0$ .

\Gamma' \approx \Gamma\,\frac{s_0}{1+(2\delta/\Gamma)^2}\;\xrightarrow{|\delta|\gg\Gamma}\;\frac{\Gamma\Omega^2}{4\delta^2} \qquad (4)

where: $\Gamma'$ — pumping/scattering rate, inverse of the cycle time $\tau_p$ . The key ratio $U_0/\hbar\Gamma'\sim|\delta|/\Gamma$ means each pump event carries away a light-shift energy large compared with a recoil — so the cooling is photon-efficient. Balancing the climb against the reset gives a friction linear in velocity whose coefficient, remarkably, does not vanish as intensity is lowered.

F=-\alpha v,\qquad \alpha\simeq-\,\frac{\hbar k^2\,\delta}{\Gamma}\quad(|\delta|\gg\Gamma) \qquad (5)

where: $\alpha$ — friction coefficient; $k$ — laser wavenumber. Because $\alpha\propto\hbar k^2\delta/\Gamma$ is independent of $s_0$ , the drag stays strong even as $U_0\to0$ — unlike Doppler friction. The price is a small capture velocity: this strong drag grips only already-slow atoms. The steady-state temperature is a fluctuation–dissipation balance of recoil heating against Sisyphus drag, fixing $k_BT$ at a small multiple of the well depth.

k_BT\simeq\frac{D}{\alpha}\;\propto\;\frac{\hbar\Omega^2}{|\delta|}\;\propto\;U_0 \qquad (6)

where: $D\sim\hbar^2k^2\Gamma'$ — momentum-diffusion coefficient; $T$ — steady-state temperature. Thus $k_BT\propto U_0\propto s_0/|\delta|$ — temperature falls linearly with intensity, the plotted signature and the break from Doppler physics. Cooling cannot run forever: once the wells are shallower than one photon recoil, a single emission kicks the atom out. This sets the floor, a few times the recoil temperature.

E_r=\frac{\hbar^2k^2}{2m},\qquad T_r=\frac{\hbar^2k^2}{mk_B}=\frac{2E_r}{k_B} \qquad (7)

where: $m$ — atomic mass; $E_r$ — single-photon recoil energy. Numerically $T_r\approx2.4\,\mu$ K (Na), $0.36\,\mu$ K (Rb), $0.20\,\mu$ K (Cs) — the wall below which sub-recoil methods must take over.

Derivation AThe polarization gradient of the lin-perp-lin standing wave

The goal is to prove the geometric fact everything rests on: two counter-propagating beams with orthogonal linear polarizations build not an intensity standing wave but a field of nearly constant magnitude whose polarization state cycles $\sigma^+\to$ lin $\to\sigma^-$ with period $\lambda/2$ , its pure circular points a quarter wavelength apart — the modulation that creates the Sisyphus hills, here as a consequence of superposition alone.

Step 1. Superpose an $\hat x$ -wave in $+z$ and a $\hat y$ -wave in $-z$ of equal amplitude: the polarization is $\propto\hat x\,e^{ikz}+\hat y\,e^{-ikz}$ once $e^{-i\omega t}$ is stripped off. Why: reducing the field to one $z$ -dependent complex vector exposes the local ellipse directly.

Step 2. At $kz=0$ the vector is $\hat x+\hat y$ — equal, in-phase components — linear at $+45^\circ$ . Why: this reference axis fixes the phase convention.

Step 3. At $kz=\pi/4$ it becomes $\hat x\,e^{i\pi/4}+\hat y\,e^{-i\pi/4}\propto\hat x-i\hat y$ , a $\pi/2$ relative phase — pure circular, a $\sigma^-$ point. Why: a quarter-wave of retardation turns equal linear components into circular light.

Step 4. At $kz=\pi/2$ the vector is $\propto\hat x-\hat y$ (linear at $-45^\circ$ ); at $kz=3\pi/4$ it is $\propto\hat x+i\hat y$ , the opposite handedness — a $\sigma^+$ point. Why: two more steps complete the cycle and show the $\sigma^-$ and $\sigma^+$ points are $\lambda/4$ apart.

Step 5. So the polarization runs $\sigma^-\,(\lambda/4)\,$ lin $\,(\lambda/4)\,\sigma^+$ at constant magnitude. Since $\sigma^+$ shifts $m_g=+1/2$ most and $\sigma^-$ shifts $m_g=-1/2$ most, the two potentials are modulated $\pi/2$ out of phase. Why: this is the origin of the offset $\cos^2$ and $\sin^2$ landscapes — the geometry of the light, not any dynamics, plants the corrugation.

Go deeper · Derivation B · 1 minWhy the Sisyphus friction does not vanish at low intensity

Derivation BWhy the Sisyphus friction does not vanish at low intensity

The aim is to build the friction of equation (5) from an energy-loss argument and expose its most surprising property — it stays large as power drops, contradicting Doppler friction. The strategy: follow one atom through a pumping cycle, find the energy lost climbing before the reset, convert it to a force, then track the intensity scaling so the cancellation shows.

Step 1. A moving atom travels $\sim v\tau_p$ between pumps ( $\tau_p=1/\Gamma'$ ); optimal operation is $kv\tau_p\sim1$ , where it climbs of order a quarter well — a height $\sim U_0$ — before being reset. Why: matching travel distance to the well period guarantees the atom is caught near a hilltop, where pumping harvests its energy.

Step 2. The mechanical loss rate is energy per cycle times cycle rate, weighted by the fraction of a well climbed: $\dot E\sim U_0k^2v^2/\Gamma'$ . Why: "energy per event × events per second" is the cleanest route to a rate; the two velocity factors encode climb height and timing.

Step 3. A loss quadratic in $v$ is a force linear in $v$ : from $\dot E=Fv$ , $\alpha\sim U_0k^2/\Gamma'$ . Why: power equals force times velocity — the definition turning dissipation into a drag coefficient.

Step 4. Insert the scalings $U_0\propto\hbar\Omega^2/\delta$ and $\Gamma'\propto\Gamma\Omega^2/\delta^2$ : the intensity factor $\Omega^2$ cancels, leaving $\alpha\propto\hbar k^2\delta/\Gamma$ , independent of $s_0$ , while the capture velocity $v_c\sim\Gamma'/k$ shrinks with power. Why: that cancellation is the whole point — Sisyphus friction is strong but narrow, cooling only atoms Doppler cooling has already slowed, so the two act in series.

Go deeper · Derivation C · 1 minThe equilibrium temperature from fluctuation and dissipation

Derivation CThe equilibrium temperature from fluctuation and dissipation

This derivation delivers the central law $k_BT\propto U_0\propto s_0/|\delta|$ by treating the atom as a Brownian particle whose momentum obeys a Langevin equation. The plan: write that equation, extract the steady-state variance by equipartition, insert the diffusion and friction coefficients already built, and combine — so the temperature comes out proportional to the well depth.

Step 1. Model the momentum by $m\dot v=-\alpha v+\xi(t)$ with $\langle\xi(t)\xi(t')\rangle=2D\,\delta(t-t')$ : deterministic drag plus white-noise recoil kicks. Why: laser cooling is exactly a particle in a viscous bath, so the Langevin equation is the correct minimal description.

Step 2. Multiply by $v$ and average in steady state ( $\frac{d}{dt}\langle v^2\rangle=0$ ): $\alpha\langle v^2\rangle=D/m$ , so $\tfrac12 m\langle v^2\rangle=\tfrac12 k_BT$ with $k_BT=D/\alpha$ . Why: equipartition converts the mechanical variance into the temperature a time-of-flight actually measures.

Step 3. Estimate the diffusion: each scattered photon shifts momentum by $\sim\hbar k$ at rate $\Gamma'$ , so $D\sim\hbar^2k^2\Gamma'$ . Why: recoil diffusion is the irreducible heating any scheme must overcome; it fixes the numerator of $T$ .

Step 4. Insert $D\sim\hbar^2k^2\Gamma'$ and $\alpha\sim U_0k^2/\Gamma'$ into $k_BT=D/\alpha$ to get $\hbar^2(\Gamma')^2/U_0$ ; then use $\hbar\Gamma'\sim U_0\Gamma/|\delta|$ to eliminate $\Gamma'$ , giving $k_BT\simeq C\,U_0$ with $C=O(1)$ . Why: the ratio $U_0/\hbar\Gamma'\sim|\delta|/\Gamma$ collapses two intensity-dependent quantities into one clean result — the falsifiable prediction that halving intensity halves temperature.

Result.

k_BT=\frac{D}{\alpha}\simeq C\,\frac{\hbar\Omega^2}{4|\delta|}=C\,U_0,\qquad C=O(1) \qquad (8)

Limiting cases

The character of the cooling — and the temperature it reaches — depends on which regime of velocity, intensity, and detuning the atom sits in.

Case	Result	Meaning
Doppler regime (fast atoms, $v\gg\Gamma/k$ )	$k_BT_D=\dfrac{\hbar\Gamma}{2}$	Sub-Doppler friction has a small capture velocity, so fast atoms are cooled only by ordinary Doppler molasses, which pre-cools them to $T_D$ before Sisyphus takes over.
Sisyphus regime (slow atoms, low $s_0$ )	$k_BT\simeq C\,U_0\propto \dfrac{s_0}{\|\delta\|}$	Temperature falls linearly with intensity, lowered by reducing $s_0$ or raising $\|\delta\|$ at fixed $U_0$ ( $C$ order unity) — the sub-Doppler window.
Recoil floor	$k_BT_r=\dfrac{\hbar^2k^2}{m}=2E_r$	Cooling saturates a few times above the recoil temperature (≈ 2.4 μK Na, 0.36 μK Rb, 0.20 μK Cs); a recoil kick ejects atoms from wells shallower than $E_r$ .
High-intensity breakdown	$U_0\gtrsim\hbar\Gamma\Rightarrow T\to T_D$	When wells approach the Doppler energy scale, saturation and multi-photon effects wash out the advantage and $T$ climbs back toward $T_D$ .

Common trapSisyphus cooling is just a stronger version of Doppler cooling

It is not — it is a genuinely different mechanism. Doppler cooling needs only a two-level atom and a Doppler shift; Sisyphus cooling needs internal structure (several ground sublevels), optical pumping between them, and a polarization gradient to modulate their light shifts. A true two-level atom shows no sub-Doppler effect however the polarization varies, because there is no second sublevel to be pumped into at a well bottom. There is also a trap in "stronger friction, therefore does everything": the Sisyphus friction is large but its capture velocity is small — a well width per pumping time, far below the Doppler range. So Doppler cooling still does the initial slowing and hands slow atoms to the Sisyphus stage — the two work in series, on different velocity classes.

Worked examples

1 · The benchmark and the anomaly (confidence builder)

Sodium's cooling transition has linewidth $\Gamma/2\pi=9.79\,$ MHz and wavelength $589\,$ nm ( $m=23\,$ u). Find its Doppler limit and recoil temperature, and place the 1988 NIST measurement of $43\,\mu$ K among them.

Step 1. With $\Gamma=2\pi(9.79\times10^6)=6.15\times10^7\,$ s⁻¹, the Doppler limit is $T_D=\hbar\Gamma/2k_B=(1.055\times10^{-34})(6.15\times10^7)/[2(1.381\times10^{-23})]=\mathbf{235\,\mu K}$ .

Step 2. With $k=2\pi/\lambda=1.07\times10^7\,$ m⁻¹ and $m=3.82\times10^{-26}\,$ kg, the recoil temperature is $T_r=\hbar^2k^2/(mk_B)=\mathbf{2.4\,\mu K}$ (the single-photon recoil is $E_r/k_B=1.2\,\mu$ K).

Step 3. The measured $43\,\mu$ K sits a factor $5.5$ below $T_D$ and $\sim18$ above $T_r$ — in the forbidden zone for two-level theory yet above the recoil floor. The 1988 result is impossible in Doppler theory (43 μK vs a 235 μK limit) and about 18× above recoil — the fingerprint of a sub-Doppler mechanism.

2 · Well depth and temperature of a rubidium molasses

Rubidium-87 has $\Gamma/2\pi=6.07\,$ MHz. Run a molasses at $s_0=1$ per beam, $\delta=-3\Gamma$ . Estimate the Rabi frequency, well depth $U_0$ , pumping time, and expected temperature.

Step 1. $\Omega=\Gamma\sqrt{s_0/2}=\Gamma/\sqrt2=0.707\,\Gamma$ , with $\Gamma=3.81\times10^7\,$ s⁻¹, so $\Omega=2.70\times10^7\,$ s⁻¹.

Step 2. From equation (3), $U_0=\hbar\Omega^2/(4|\delta|)=\hbar\Omega^2/(12\Gamma)=\hbar\Gamma/24$ (using $\Omega^2=\Gamma^2/2$ ). Since $\hbar\Gamma/k_B=2T_D=291\,\mu$ K, $U_0/k_B=291/24=\mathbf{12\,\mu K}$ — the depth in the first figure.

Step 3. The pumping rate is $\Gamma'\simeq\Gamma\Omega^2/(4\delta^2)=\Gamma/72=5.3\times10^5\,$ s⁻¹, so $\tau_p=\mathbf{1.9\,\mu s}$ . The capture velocity is $v_c\sim\Gamma'/k=\mathbf{6.6\,cm/s}$ — tiny beside the Doppler capture speed $\Gamma/k\approx4.7\,$ m/s.

Step 4. With $k_BT\simeq C\,U_0$ ( $C$ of order unity), the predicted $T$ is a few times $12\,\mu$ K — tens of microkelvin, below the $146\,\mu$ K Rb Doppler limit and above the $0.36\,\mu$ K recoil floor, matching real Rb molasses. Halving the intensity would halve $U_0$ and halve $T$ — the linear fall a constant $T_D$ cannot produce. $U_0/k_B\approx12\,\mu$ K, $\tau_p\approx1.9\,\mu$ s, $v_c\approx6.6\,$ cm/s; $T\propto s_0$ .

Physics of the portrait

Sub-Doppler cooling is where a puzzle became a Nobel Prize: a 1988 anomaly met, within a year, a theory built from the humblest atomic-physics ingredients — and the person most identified with that synthesis had spent his career turning atom–photon interactions into a language precise enough to predict it.

Origins · 1 minClaude Cohen-Tannoudji and how this was discovered

Claude Cohen-Tannoudji

b. 1933 · polarization-gradient cooling and atom–photon interactions

Claude Cohen-Tannoudji, born in Constantine, Algeria, in 1933, built much of the modern language of atom–light interaction — the dressed-atom picture, the theory of light shifts, and the master-equation treatment of optical pumping developed at the École Normale Supérieure in the lineage of Alfred Kastler and Jean Brossel. When the NIST group reported sodium molasses far below the Doppler limit in 1988, two-level theory forbade the result: a genuine crisis. Within a year Cohen-Tannoudji and his student Jean Dalibard published the polarization-gradient (Sisyphus) mechanism in the Journal of the Optical Society of America B (1989), showing that spatially varying light shifts plus optical pumping among ground sublevels reproduce the observed temperatures and their $T\propto s_0/|\delta|$ scaling; Steven Chu's group reached the same understanding independently that year. For laser cooling and trapping he shared the 1997 Nobel Prize with Chu and William Phillips. His dressed-atom methods underlie the whole toolkit of ultracold physics — optical lattices, dipole traps, and the sub-recoil schemes that push below the very floor this article ends at.

History

1975

— Hänsch and Schawlow propose Doppler cooling of neutral atoms; Wineland and Dehmelt propose it for trapped ions

1985

— Chu, Hollberg, Bjorkholm, Cable and Ashkin demonstrate three-dimensional optical molasses at Bell Labs

1988

— Lett, Watts, Westbrook and Phillips at NIST measure sodium molasses at 43 μK, far below the 235 μK Doppler limit

1989

— Dalibard and Cohen-Tannoudji, and independently Ungar, Weiss, Riis and Chu, explain polarization-gradient (Sisyphus) cooling

1997

— Chu, Cohen-Tannoudji and Phillips share the Nobel Prize in Physics for laser cooling and trapping

Going further

Texts. Foot, Atomic Physics (2005), ch. 9 · Metcalf and van der Straten, Laser Cooling and Trapping (1999), ch. 8 · Cohen-Tannoudji, Dupont-Roc and Grynberg, Atom–Photon Interactions (1992).

Primary sources. J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B 6, 2023 (1989) — the polarization-gradient theory · P. J. Ungar, D. S. Weiss, E. Riis and S. Chu, J. Opt. Soc. Am. B 6, 2058 (1989) · P. D. Lett et al., Phys. Rev. Lett. 61, 169 (1988) — the sodium sub-Doppler observation.

Provenance. All equations follow the standard treatment (Dalibard and Cohen-Tannoudji 1989; Ungar, Weiss, Riis and Chu 1989; Foot ch. 9; Metcalf and van der Straten). A verifier should confirm the lin⟂lin cycle with $\sigma^\pm$ points $\lambda/4$ apart (Derivation 1), the scaling $k_BT\propto U_0\propto s_0/|\delta|$ (Derivation 3), the Doppler limits 235 μK (Na), 146 μK (Rb), 125 μK (Cs) and recoil temperatures 2.4 μK (Na), 0.36 μK (Rb), 0.20 μK (Cs) from the linewidths and wavelengths, and the well depth $U_0/k_B=\hbar\Gamma/24k_B\approx12\,\mu$ K. Order-unity constants are configuration-dependent and omitted rather than fabricated. (Some summaries mislabel sodium's Doppler limit as 146 μK, which is actually rubidium's.)

Prerequisites. Doppler Cooling and the Doppler Limit · Optical Molasses · The Dipole Force and the Optical Dipole Potential · Radiation Pressure and the Scattering Force. Leads to. The Recoil Limit and Sub-Recoil Cooling · The Magneto-Optical Trap · Evaporative Cooling · Bose–Einstein Condensation in Dilute Gases.

Ask the tutor: Ask me to walk through why the Sisyphus friction survives at vanishing laser power while Doppler friction dies, or to work the well depth and predicted temperature for your own atom, detuning, and intensity.

Why does the atom always seem to be climbing rather than descending?

That asymmetry is the whole mechanism. Optical pumping is not random in space: it is fastest where the light shift is largest, and delivers the atom to the sublevel whose well is deepest there. So an atom is preferentially pumped near a hilltop of its current potential — having climbed there — and lands in a valley of the other, whereupon it must climb again. The correlation between position, internal state, and pumping rate, engineered by the polarization gradient, biases the atom uphill on average. Remove the spatial modulation and the bias vanishes.

Where does the dissipated energy actually go?

Into the light field, as a tiny blue shift on the scattered photons. In each cycle the atom absorbs a photon near a potential maximum, where its light-shift energy is high, and emits one as it is pumped down; on average the emitted photon carries slightly more energy than the absorbed one — the difference being the $U_0$ the atom gave up climbing, plus a recoil. Summed over the cloud the fluorescence is imperceptibly blue-shifted relative to the drive: that spectral asymmetry is the energy budget balancing, mechanical energy leaving with the light.

If Sisyphus friction is so strong, why not skip Doppler cooling?

Because the strong friction grips only atoms that are already slow. Its capture velocity is one well width per pumping time — a few cm/s in the rubidium example, versus several m/s for the Doppler range. Atoms straight from an oven or a Zeeman slower move far too fast; they sail through the wells before pumping catches them near a hilltop. Doppler cooling supplies the wide velocity capture to decelerate them into the narrow window where polarization-gradient cooling then takes over — broad-but-weak, then narrow-but-strong.

The σ⁺σ⁻ configuration has uniform light shifts — how can it cool?

Through a velocity-induced orientation rather than a Sisyphus hill. Two counter-propagating $\sigma^+$ and $\sigma^-$ beams make a field of constant magnitude whose linear polarization twists helically along $z$ . The light shifts are position-independent, so a stationary atom settles into a symmetric distribution with no net force. But a moving atom sees the polarization axis rotating in its frame; if pumping cannot keep the populations adiabatically symmetric, a net orientation $\langle m\rangle\propto v$ builds up, making it scatter preferentially from the beam it moves against. The result is damping with no potential wells — a distinct mechanism needing a $J_g=1$ or higher ground state.

Why can't we reach zero temperature by lowering the intensity?

Two things break at low intensity. First, the recoil floor: as $U_0$ falls below the recoil energy $E_r$ , one spontaneous emission kicks the atom clean out of a well, so cooling stalls at a few times $T_r$ — about $0.36\,\mu$ K for rubidium. Second, the capture velocity $v_c\sim\Gamma'/k$ also shrinks, so fewer atoms are slow enough to be gripped and residual scattering heats them. The linear law $T\propto s_0$ is only the ideal trend; going lower needs sub-recoil schemes such as velocity-selective coherent population trapping or Raman sideband cooling.

How deep should the wells be for cooling to work well?

There is a comfortable window, $E_r\lesssim U_0\lesssim\hbar\Gamma$ . Shallower than the recoil energy and a single recoil ejects the atom — recoil-limited. Deeper than the Doppler scale ( $U_0\gtrsim\hbar\Gamma$ ) and saturation and multi-photon processes take over, the two-sublevel picture fails, and $T$ climbs back toward $T_D$ . In practice one sits in the middle: modest saturation and a detuning of a few linewidths, giving wells tens of times the recoil energy but well below $\hbar\Gamma$ — the $s_0\sim1$ , $\delta\sim-3\Gamma$ regime of the worked example.