Doppler Cooling and the Doppler Limit

Point a laser at an atom and you can slow it to a crawl — but not to a standstill. Doppler cooling turns a room-temperature vapour into a micro-kelvin gas, and it is built from a single force established one article earlier: the resonant scattering force, $F=\hbar k$ per photon times the rate at which the atom cycles between ground and excited state. That force is dissipative because absorption is directional but re-emission is not — every photon comes in along the beam and leaves in a random direction (spontaneous emission). The scheme's genius is to make the rate velocity-dependent: red-detune the light, aim two beams head-on, and a moving atom scatters more from the beam it runs into than from the one chasing it. The imbalance is a friction $F=-\alpha v$ . The same spontaneous emission that makes the force dissipative also forbids the temperature from reaching zero. Each scattered photon delivers a random recoil kick $\hbar k$ , and these accumulate as a random walk in momentum space — a heating process. Cooling (smooth friction) and heating (jittery recoil) reach a truce at a nonzero equilibrium, as a Brownian particle in a warm fluid settles at the bath temperature. That balance gives a strikingly clean result: the minimum temperature depends on nothing but the natural linewidth $\Gamma$ , $k_BT_D=\hbar\Gamma/2$ . This two-level, semiclassical picture is both a triumph and a trap: it correctly predicts the friction, the capture range, and the order of magnitude of the final temperature, yet real alkali atoms routinely cool below $T_D$ because their internal structure opens gentler channels (Sisyphus cooling) the two-level model cannot see. Knowing where the Doppler picture rules and where it is superseded maps everything downstream: the magneto-optical trap, the Zeeman slower, and the road to quantum degeneracy.

Key takeaways

Doppler cooling turns resonant scattering into friction: red-detuned counter-propagating beams make a moving atom absorb preferentially from the beam opposing it, giving a linear damping $F\simeq-\alpha v$ for $kv\ll\Gamma$ .
The friction $\alpha$ is maximised at $\delta=-\Gamma/2$ and low saturation; damping requires red detuning ( $\delta<0$ ), since $\alpha\propto-\delta$ .
Random spontaneous-emission recoil heats the atom — a momentum random walk with diffusion $D_p\sim\hbar^2k^2R_{\mathrm{scatt}}$ — so cooling cannot reach absolute zero.
Balancing friction against diffusion via the Einstein relation $k_BT=D_p/\alpha$ gives the Doppler limit $k_BT_D=\hbar\Gamma/2$ , set by the linewidth alone — mass, intensity and wavevector cancel.
For alkali D2 lines $T_D$ is hundreds of micro-kelvin: $\approx146\ \mu\mathrm{K}$ ( $^{87}$ Rb), $\approx235\ \mu\mathrm{K}$ (Na), $\approx125\ \mu\mathrm{K}$ (Cs); experiments beat these via sub-Doppler mechanisms.
The true floor is the recoil limit $k_BT_r=\hbar^2k^2/m$ ( $\approx362\ \mathrm{nK}$ for $^{87}$ Rb), with $T_D/T_r=m\Gamma/(2\hbar k^2)\approx400$ , so Doppler cooling stops far above it.
Key idea: red-detuned counter-propagating light converts spontaneous scattering into a velocity-dependent friction $F=-\alpha v$ ; the random recoil of that same scattering heats at a rate set by momentum diffusion $D_p$ , and the balance $k_BT=D_p/\alpha$ fixes a fundamental minimum $k_BT_D=\hbar\Gamma/2$ .

The core insight

Red-detuned counter-propagating light turns spontaneous scattering into a velocity-dependent friction $F=-\alpha v$ ; the random recoil of that same scattering heats at a rate set by momentum diffusion $D_p$ , and their balance $k_BT=D_p/\alpha$ produces a fundamental minimum $k_BT_D=\hbar\Gamma/2$ .

Where this lives today

Every cold-atom laboratory runs on the physics of this article. The first stage of nearly all quantum-gas and atomic-clock experiments is a magneto-optical trap: three orthogonal pairs of red-detuned beams supply the Doppler friction analysed here, and a quadrupole magnetic field turns the velocity-dependent force into a position-dependent restoring force, gathering a room-temperature vapour into a millimetre cloud of a billion atoms at hundreds of micro-kelvin in under a second. The friction $\alpha$ and capture velocity $v_c\sim\Gamma/k$ set the loading rate; a Zeeman slower uses the same force to decelerate the atomic beam feeding the trap. From that cloud the modern toolchain takes over: a brief polarisation-gradient stage pushes below $T_D$ into the few-micro-kelvin range, evaporative cooling crosses the last three orders of magnitude to Bose–Einstein condensation and degenerate Fermi gases, and optical-lattice and tweezer arrays hold single atoms for quantum simulation and computing. Optical atomic clocks — now accurate to better than one part in $10^{18}$ — begin with Doppler cooling to erase the Doppler broadening that would smear the clock transition. The single formula $k_BT_D=\hbar\Gamma/2$ is the doorway to all of it.

The physical picture

Put an atom in a crossfire of two identical beams, aimed at each other and both tuned just below the atomic resonance. Standing still, it scatters equally from both and the pushes cancel. Now let it drift right. The beam it moves into is Doppler-shifted up in the atom's frame, sliding toward resonance, so the atom scatters more of those photons and takes a harder push backward; the beam chasing from behind shifts the other way and scatters less. The net force opposes the drift whichever way the atom moves — a viscous drag, the reason the configuration is nicknamed "optical molasses."

Figure

A rubidium-87 atom drifting right in two red-detuned beams at δ = −Γ/2 sees the counter-propagating beam Doppler-shifted by +kv toward resonance so it scatters more, and the co-propagating beam by −kv away so it scatters less, so the net force F = −αv opposes the motion

If photons only ever came in against the motion, the temperature would slide to zero. They do not. Every absorbed photon is re-emitted a moment later in a random direction, kicking the atom by one photon momentum $\hbar k$ . These recoils cannot be organised — they point every which way — so they act like the molecular buffeting that keeps a pollen grain jittering in water: the mean-square momentum grows by a random walk even as friction drains the drift. Equilibrium arrives when the smooth cooling and the jittery heating exactly cancel — a fluctuation–dissipation balance, the laser-cooling version of Einstein's relation between drag and diffusion. Why does the final temperature depend on nothing but $\Gamma$ ? Because $\Gamma$ is the only intrinsic frequency: it sets the width of the Lorentzian (the span of velocities friction can grip), the energy $\hbar\Gamma$ a scattering event smears out, and the optimum detuning half a linewidth to the red. When the trade-off is optimised, mass, wavevector, and intensity all cancel between heating and cooling, leaving the strikingly bare $k_BT_D=\hbar\Gamma/2$ . The second figure shows the whole story: the equilibrium temperature versus detuning, U-shaped, bottoming out at exactly $\delta=-\Gamma/2$ .

Figure

Equilibrium temperature versus red detuning at low saturation. The U-shaped curve bottoms out at δ = −Γ/2, where k_B T_D = ħΓ/2, which for the rubidium-87 D2 line with Γ/2π = 6.07 MHz gives T_D ≈ 146 μK

The core equations

The whole edifice rests on one force. A two-level atom in a near-resonant beam of wavevector $k$ scatters at the Lorentzian rate $R_{\mathrm{scatt}}=(\Gamma/2)\,s_0/[1+s_0+(2\delta/\Gamma)^2]$ , and each absorbed photon delivers $\hbar k$ ; rate times momentum is the radiation-pressure force from the preceding article.

F_{\mathrm{scatt}}=\hbar k\,\dfrac{\Gamma}{2}\,\dfrac{s_0}{1+s_0+\left(\dfrac{2\delta}{\Gamma}\right)^2} \qquad (1)

where: $\hbar k$ — photon momentum along the beam; $\Gamma$ — natural linewidth (spontaneous decay rate); $s_0=I/I_{\mathrm{sat}}$ — on-resonance saturation parameter; $\delta=\omega_L-\omega_0$ — laser detuning. The force saturates at $F_{\max}=\hbar k\,\Gamma/2$ as $s_0\to\infty$ : the atom cannot cycle faster than once per lifetime. The cooling comes entirely from making $\delta$ velocity-dependent. In the frame of an atom moving with velocity $v$ along the axis, a beam is shifted by the first-order Doppler shift $\mp kv$ , so its effective detuning differs from the laboratory value.

\delta_{\pm}=\delta\mp kv \qquad (2)

where: $\delta_{\pm}$ — effective detuning of the beam co-/counter-propagating with the velocity; $v$ — atomic velocity along the axis; $kv$ — Doppler shift. For the counter-propagating beam the shift is $+kv$ : red-detuned light slides toward resonance for an approaching atom. Add the two beam forces and expand for small velocity: the zeroth order cancels by symmetry, and the leading term is linear in $v$ — a friction.

F_{\mathrm{mol}}=F_+ + F_- \;\simeq\; -\alpha v,\qquad \alpha=\hbar k^2\,\dfrac{8\,s_0\,(|\delta|/\Gamma)}{\left[1+s_0+(2\delta/\Gamma)^2\right]^2} \qquad (3)

where: $F_{\pm}$ — forces from the two beams; $\alpha$ — velocity-damping coefficient, positive only for red detuning ( $\delta<0$ ). The bracket is dimensionless, so $\alpha$ has units $\hbar k^2$ (kg s $^{-1}$ ), a drag coefficient. Cooling would run to zero but for the recoil of spontaneous emission. Treat the momentum as a random walk: each of the $R_{\mathrm{scatt}}$ events per second imprints a kick of order $\hbar k$ , so the mean-square momentum grows diffusively as $\langle p^2\rangle=2D_pt$ .

D_p=\hbar^2k^2\,R_{\mathrm{scatt}}\,(1+\eta) \qquad (4)

where: $D_p$ — momentum-diffusion coefficient; $R_{\mathrm{scatt}}$ — total scattering rate (both beams); $\eta$ — order-unity factor splitting the fluctuation between absorption and isotropic emission ( $\eta=1$ for isotropic emission). The scaling $D_p\propto(\hbar k)^2R_{\mathrm{scatt}}$ says heating grows with recoil energy times scattering rate. The steady state equates diffusive heating to frictional cooling — the fluctuation–dissipation (Einstein) relation, the form a Brownian particle obeys in a bath at temperature $T$ .

k_BT=\dfrac{D_p}{\alpha} \qquad (5)

where: $T$ — steady-state temperature; $D_p/\alpha$ — heating-to-friction ratio, with $\langle p^2\rangle/m=k_BT$ for one degree of freedom. The balance is between the rates, not the forces. Substituting $D_p$ and $\alpha$ at low saturation, intensity, mass and wavevector cancel and a pure function of detuning survives, minimised at $\delta=-\Gamma/2$ .

k_BT=\dfrac{\hbar\Gamma}{4}\,\dfrac{1+(2\delta/\Gamma)^2}{2|\delta|/\Gamma}\;\xrightarrow{\;\delta=-\Gamma/2\;}\;\dfrac{\hbar\Gamma}{2}\equiv k_BT_D \qquad (6)

where: $T_D$ — Doppler-limit temperature; $\hbar\Gamma/2$ — the residual energy of one scattering event's frequency uncertainty. The result is species-independent except through $\Gamma$ . The Doppler limit is not the ultimate floor. That role belongs to the single-photon recoil temperature, far below $T_D$ for any strong dipole transition.

k_BT_r=\dfrac{\hbar^2k^2}{m},\qquad \dfrac{T_D}{T_r}=\dfrac{m\Gamma}{2\hbar k^2}\gg1 \qquad (7)

where: $T_r$ — recoil temperature (the true floor); $m$ — atomic mass. For $^{87}$ Rb the ratio is about $400$ : Doppler cooling halts far above the floor, leaving room for the sub-Doppler methods that follow.

Derivation AThe Doppler friction coefficient from two counter-propagating beams

The goal is to turn a pair of ordinary radiation-pressure forces into a genuine viscous friction and pin down where it is strongest. We put an atom of velocity $v$ in two counter-propagating red-detuned beams, write each force with its Doppler-shifted detuning, add them, and expand to first order in $v$ : the zeroth order cancels by symmetry, and the surviving linear term is the damping, maximised over $\delta$ .

Step 1. Give each beam its Doppler-shifted detuning: the $+x$ beam sees $\delta-kv$ and pushes toward $+x$ , the $-x$ beam sees $\delta+kv$ and pushes toward $-x$ , so

F_{\pm}=\pm\hbar k\dfrac{\Gamma}{2}\dfrac{s_0}{1+s_0+\big(2(\delta\mp kv)/\Gamma\big)^2}.

Why: the atom's motion is the only thing distinguishing the two beams; opposite Doppler shifts break the symmetry and produce a net force.

Step 2. Add the forces and Taylor-expand each Lorentzian for $kv\ll\Gamma$ . The velocity-independent parts cancel and the first derivatives add, leaving

F_{\mathrm{mol}}=F_++F_-\approx -\,\hbar k\,\Gamma s_0\,\dfrac{8(\delta/\Gamma)(kv/\Gamma)}{[1+s_0+(2\delta/\Gamma)^2]^2}.

Why: the expansion is valid only for $kv\ll\Gamma$ ; outside it the force saturates and the linear picture fails — molasses grabs only slow atoms.

Step 3. Read off the friction coefficient by writing $F_{\mathrm{mol}}=-\alpha v$ :

\alpha=\hbar k^2\dfrac{8 s_0\,(-\delta/\Gamma)}{[1+s_0+(2\delta/\Gamma)^2]^2}.

Why: with $\delta<0$ , $-\delta/\Gamma>0$ , so $\alpha>0$ and the force opposes velocity; blue detuning would flip the sign and heat.

Step 4. Optimise over detuning in the low-saturation limit $s_0\to0$ : maximising $(-\delta/\Gamma)/[1+(2\delta/\Gamma)^2]^2$ gives

\dfrac{\partial\alpha}{\partial\delta}=0\;\Rightarrow\;\delta=-\dfrac{\Gamma}{2}.

Why: half a linewidth to the red is where the Doppler shift straddles the steepest part of the Lorentzian, the biggest imbalance per unit velocity. *Verified: reproduces Foot Eqs. (9.10)–(9.13) and Metcalf and van der Straten Ch. 7; positive damping only for red detuning.*

Go deeper · Derivation B · 1 minMomentum diffusion and the Doppler-limit temperature

Derivation BMomentum diffusion and the Doppler-limit temperature

The goal is to show that the very scattering that cools the atom also heats it, and to find where the two balance. We model the momentum as a random walk fed by absorption and isotropic emission, set the diffusive growth of $\langle p^2\rangle$ equal to the frictional loss, and identify the equilibrium with $k_BT$ . Substituting the low-saturation forms, everything but linewidth and detuning cancels, and minimising delivers the celebrated $k_BT_D=\hbar\Gamma/2$ .

Step 1. Write the momentum diffusion as recoil energy per event times the scattering rate,

D_p\simeq \hbar^2 k^2\,R_{\mathrm{scatt}},\qquad R_{\mathrm{scatt}}=\dfrac{\Gamma}{2}\dfrac{2s_0}{1+2s_0+(2\delta/\Gamma)^2}.

Why: the random timing of absorption and the random direction of emission each add a kick $\sim\hbar k$ , so each event adds $\sim(\hbar k)^2$ to $\langle p^2\rangle$ .

Step 2. Balance heating against friction. The mean-square momentum obeys

\dfrac{\mathrm{d}\langle p^2\rangle}{\mathrm{d}t}=2D_p-2\alpha\dfrac{\langle p^2\rangle}{m}=0\;\Rightarrow\;\dfrac{\langle p^2\rangle}{m}=\dfrac{D_p}{\alpha}=k_BT.

Why: the first term is the random-walk growth, the second the drag-driven decay; at steady state they cancel, and $\langle p^2\rangle/m=k_BT$ .

Step 3. Insert $D_p$ and $\alpha$ in the low-saturation limit $s_0\to0$ ; the $s_0$ and $k^2$ factors cancel between numerator and denominator,

k_BT=\dfrac{\hbar^2k^2\,\Gamma s_0/[1+(2\delta/\Gamma)^2]}{\hbar k^2\,8s_0(-\delta/\Gamma)/[1+(2\delta/\Gamma)^2]^2}=\dfrac{\hbar\Gamma}{8}\,\dfrac{1+(2\delta/\Gamma)^2}{(-\delta/\Gamma)}.

Why: the cancellation is the whole point — no atomic property except $\Gamma$ survives.

Step 4. Minimise the remaining function of detuning. Setting the derivative of $[1+(2\delta/\Gamma)^2]/|\delta|$ to zero gives

|\delta|=\dfrac{\Gamma}{2}\;\Rightarrow\;k_BT_D=\dfrac{\hbar\Gamma}{2}.

Why: the same detuning that maximises friction minimises temperature — the optimal molasses is red-detuned by exactly half a linewidth. *Verified: matches the Wineland–Itano result and Foot Eq. (9.20); numerically $T_D=\hbar\Gamma/(2k_B)$ .*

Go deeper · Derivation C · 1 minDoppler limits and the recoil floor for real atoms

Derivation CDoppler limits and the recoil floor for real atoms

The goal is to convert the abstract $k_BT_D=\hbar\Gamma/2$ into laboratory temperatures and check how far above the true floor they sit. We evaluate $T_D=\hbar\Gamma/(2k_B)$ from the textbook D2 linewidths, then compute the recoil temperature $T_r=\hbar^2k^2/(mk_B)$ and form the ratio, confirming the Doppler limit lies hundreds of recoil temperatures above the floor.

Step 1. Rubidium-87 D2 ( $\lambda=780$ nm, $\Gamma/2\pi=6.07$ MHz):

T_D=\dfrac{\hbar\Gamma}{2k_B}=\dfrac{h(\Gamma/2\pi)}{2k_B}\approx 146\ \mu\mathrm{K}.

Why: the workhorse Rb line; every Rb experiment quotes $146\ \mu$ K.

Step 2. Sodium D2 ( $\Gamma/2\pi=9.79$ MHz) and cesium D2 ( $\Gamma/2\pi=5.22$ MHz):

T_D^{\mathrm{Na}}\approx235\ \mu\mathrm{K},\qquad T_D^{\mathrm{Cs}}\approx125\ \mu\mathrm{K}.

Why: $T_D\propto\Gamma$ , so sodium's broad line gives a higher ceiling and cesium's narrow line a lower one.

Step 3. Recoil floor for $^{87}$ Rb and the ratio:

k_BT_r=\dfrac{\hbar^2k^2}{m}\Rightarrow T_r\approx362\ \mathrm{nK},\qquad \dfrac{T_D}{T_r}=\dfrac{m\Gamma}{2\hbar k^2}\approx402.

Why: the Doppler limit sits about four hundred recoil temperatures above the floor, so molasses at tens of micro-kelvin beats $T_D$ yet stays far above $T_r$ — the Sisyphus regime. *Verified: linewidths from Steck data; $T_D=\hbar\Gamma/(2k_B)$ and $T_r=\hbar^2k^2/(mk_B)$ recomputed directly, giving 146/235/125 μK and $T_r=362$ nK, $T_D/T_r=402$ .*

Limiting cases

Case	Result	Meaning
Optimal detuning	$\delta=-\Gamma/2\;\Rightarrow\;k_BT_D=\dfrac{\hbar\Gamma}{2}$	Minimum temperature; the friction-to-diffusion ratio peaks here at low saturation.
General detuning (low $s_0$ )	$k_BT=\dfrac{\hbar\Gamma}{4}\,\dfrac{1+(2\delta/\Gamma)^2}{2\|\delta\|/\Gamma}$	Minimised at $\|\delta\|=\Gamma/2$ ; diverges as $\delta\to0$ (no friction) and as $\|\delta\|\to\infty$ (too little scattering).
Small velocity (viscous)	$F\simeq-\alpha v$ , $\alpha=\hbar k^2\dfrac{-8(\delta/\Gamma)s_0}{[1+s_0+(2\delta/\Gamma)^2]^2}$	Linear friction for $kv\ll\Gamma$ ; positive damping needs $\delta<0$ (blue detuning gives $\alpha<0$ , heating).
Large velocity	$\|kv\|\gtrsim\Gamma\;\Rightarrow\;F\to F_{\max}=\pm\hbar k\dfrac{\Gamma}{2}$	Force saturates; atoms faster than $v_c\sim\Gamma/k$ escape cooling.
Recoil floor	$\dfrac{T_D}{T_r}=\dfrac{m\Gamma}{2\hbar k^2}\gg1$	For strong dipole lines $T_D$ lies far above the recoil limit, so the Doppler picture never touches $T_r$ .

Common trapThe Doppler limit is the coldest a laser can cool an atom.

It is not — the discovery that it is not launched a whole subfield. The Doppler limit $k_BT_D=\hbar\Gamma/2$ is the floor for a structureless two-level atom only. In 1988 the Phillips group measured sodium molasses several times below the predicted $T_D$ , an anomaly Dalibard and Cohen-Tannoudji explained the next year: real alkali atoms have multiple ground sublevels, and the polarisation gradients of counter-propagating beams pump atoms between them so they perpetually climb potential hills and lose energy — the Sisyphus mechanism. Sub-Doppler cooling reaches a few micro-kelvin, bounded by the far lower recoil energy. And even within the two-level model, $k_BT=\hbar\Gamma/2$ holds only at low saturation and $\delta=-\Gamma/2$ — treat it as the right scale, not a five-digit prediction.

Worked examples

1 · The Doppler limit and velocity spread of rubidium-87 (confidence builder)

The $^{87}$ Rb D2 cooling line has $\Gamma/2\pi=6.07$ MHz and $\lambda=780$ nm. Find the Doppler-limit temperature, the corresponding one-dimensional rms speed, and compare it to the single-photon recoil velocity.

Step 1. Convert the linewidth to angular units: $\Gamma=2\pi\times6.07\times10^{6}=3.81\times10^{7}\ \mathrm{s^{-1}}$ . Then $k_BT_D=\hbar\Gamma/2=(1.055\times10^{-34})(3.81\times10^{7})/2=2.01\times10^{-27}$ J. Dividing by $k_B=1.381\times10^{-23}$ J/K gives $T_D=\mathbf{146\ \mu K}$ , the celebrated Doppler ceiling for rubidium.

Step 2. The rms speed along one axis follows from $\tfrac12 m v_{\mathrm{rms}}^2=\tfrac12 k_BT_D$ , so $v_{\mathrm{rms}}=\sqrt{k_BT_D/m}$ . With $m=87\times1.66\times10^{-27}=1.44\times10^{-25}$ kg, $v_{\mathrm{rms}}=\mathbf{0.12\ m/s}$ (about 12 cm/s) — from a vapour that started near 300 m/s.

Step 3. The recoil velocity is $v_r=\hbar k/m$ with $k=2\pi/\lambda=8.05\times10^{6}\ \mathrm{m^{-1}}$ , giving $v_r=(1.055\times10^{-34})(8.05\times10^{6})/(1.44\times10^{-25})=\mathbf{5.9\ mm/s}$ — the Doppler spread is about $20$ recoil velocities wide. T_D = 146 μK, v_rms ≈ 0.12 m/s, v_r ≈ 5.9 mm/s.

2 · Capture velocity and stopping power

For the same $^{87}$ Rb line, estimate the capture velocity of a molasses (the speed at which the Doppler shift equals the linewidth) and the maximum deceleration a single beam can deliver.

Step 1. The capture velocity is where the Doppler shift $kv$ reaches $\Gamma$ , i.e. the atom slides a full linewidth across resonance: $v_c\simeq\Gamma/k=(3.81\times10^{7})/(8.05\times10^{6})=\mathbf{4.7\ m/s}$ . Atoms slower than a few metres per second feel the friction; faster ones outrun it.

Step 2. The largest scattering force is the saturated $F_{\max}=\hbar k\,\Gamma/2=(1.055\times10^{-34})(8.05\times10^{6})(3.81\times10^{7})/2=1.6\times10^{-20}$ N, so the peak deceleration is $a_{\max}=F_{\max}/m=1.6\times10^{-20}/1.44\times10^{-25}=\mathbf{1.1\times10^{5}\ m/s^2}$ .

Step 3. That is about $1.1\times10^{4}$ times gravity — a stopping power of ten thousand $g$ from nothing but light. An atom entering at $v_c=4.7$ m/s is brought to rest in roughly $v_c/a_{\max}\approx40\ \mu$ s over about $0.1$ mm, which is why a table-top molasses traps a cold cloud almost instantly. The recoil floor sits far below all of this: $T_r=\hbar^2k^2/(mk_B)=362$ nK, so $T_D/T_r=m\Gamma/(2\hbar k^2)\approx400$ . v_c ≈ 4.7 m/s, a_max ≈ 1.1×10⁵ m/s² ≈ 10⁴ g.

Physics of the portrait

Doppler cooling is a rare physics idea that arrived as a proposal and, within a few years, became an experimental reality that reshaped its field.

Origins · 1 minDavid Wineland and how this was discovered

David Wineland

born 1944 · laser cooling of trapped ions and the Doppler cooling limit

Cooling atoms with light was proposed twice in 1975: Theodor Hänsch and Arthur Schawlow for neutral atoms, David Wineland and Hans Dehmelt for trapped ions. Wineland turned the proposal into fact. In 1978, at the U.S. National Bureau of Standards, he and colleagues laser-cooled trapped magnesium ions — among the first cooling of matter with light ever demonstrated — and the next year Wineland and Wayne Itano derived $k_BT_D=\hbar\Gamma/2$ from the balance of friction against recoil heating. That result is the quantitative heart of this article. Over the following decades Wineland built the field of quantum control of single trapped ions — ground-state cooling, the first quantum logic gate, record optical clocks — and for "ground-breaking experimental methods that enable measuring and manipulation of individual quantum systems" shared the 2012 Nobel Prize with Serge Haroche. The neutral-atom branch — optical molasses and sub-Doppler cooling — earned Chu, Cohen-Tannoudji, and Phillips the 1997 Nobel, but the Doppler limit itself is Wineland and Itano's.

History

1975

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

1978

— Wineland, Drullinger and Walls first laser-cool trapped magnesium ions

1979

— Wineland and Itano derive the Doppler limit k_B T_D = ħΓ/2

1985

— Chu and co-workers demonstrate three-dimensional optical molasses of sodium

1988

— Lett and co-workers measure molasses temperatures below the Doppler limit, revealing sub-Doppler cooling

Going further

Texts. Foot, Atomic Physics (2005), Ch. 9 — scattering force, optical molasses, the Doppler limit · Metcalf and van der Straten, Laser Cooling and Trapping (1999), Ch. 7 · Cohen-Tannoudji, Dupont-Roc and Grynberg, Atom–Photon Interactions · Loudon, The Quantum Theory of Light, 3rd ed. · Wineland and Itano, Phys. Rev. A 20, 1521 (1979) — the laser-cooling limits.

Provenance. Structure and equations follow Foot Ch. 9, Metcalf and van der Straten Ch. 7, and Wineland and Itano PRA 20, 1521 (1979). A verifier should recompute: (1) $\alpha$ by linearising two Lorentzian forces, optimum at $\delta=-\Gamma/2$ ; (2) the balance $k_BT=D_p/\alpha$ giving $k_BT_D=\hbar\Gamma/2$ ; (3) $T_D=\hbar\Gamma/(2k_B)$ from the D2 linewidths (Rb-87 6.07 MHz → 146 μK, Na 9.79 → 235, Cs 5.22 → 125 μK; Steck); (4) the recoil floor $T_r=362$ nK for $^{87}$ Rb and $T_D/T_r=m\Gamma/(2\hbar k^2)\approx402$ . Note the corrected factor of two: with $k_BT_r=\hbar^2k^2/m$ the ratio is $m\Gamma/(2\hbar k^2)$ , not $m\Gamma/(\hbar k^2)$ .

Prerequisites. Radiation Pressure and the Scattering Force · Optical Molasses · Dipole Force and the Optical Dipole Potential. Leads to. Sub-Doppler Cooling and the Sisyphus Mechanism · The Recoil Limit and Sub-Recoil Cooling · Magneto-Optical Trap · Zeeman Slower · Evaporative Cooling.

Ask the tutor: Ask me to derive the friction coefficient $\alpha$ line by line, or to work out the Doppler limit and molasses capture velocity for any atom you name.

Why is the counter-propagating beam the one that cools?

Because a moving atom sees it Doppler-shifted toward resonance. Suppose the atom drifts in $+x$ with both beams red-detuned. The beam it runs into (from $+x$ ) is blue-shifted in the atom's frame by $+kv$ , sliding closer to the line, so the atom scatters more of those photons and takes more backward pushes. The beam chasing from behind shifts the other way and scatters less. The imbalance is always a backward force, whatever direction the atom moves — exactly a friction. Blue detuning would reverse every sign and heat instead.

Why does the Doppler limit depend only on Γ and nothing else?

Because $\Gamma$ is the only intrinsic frequency in the two-level problem, and it enters heating and cooling in ways that cancel everything else. In the equilibrium $k_BT=D_p/\alpha$ , both $D_p$ and $\alpha$ scale as $\hbar^2k^2$ and grow linearly with $s_0$ at low intensity, so those factors divide out, leaving a pure function of $\delta/\Gamma$ times one power of $\hbar\Gamma$ . Minimising over detuning pins $\delta=-\Gamma/2$ and leaves $k_BT_D=\hbar\Gamma/2$ : mass, wavevector and intensity have vanished, and only the linewidth survives.

If light only pushes, how can spontaneous emission heat the atom?

Absorption is directional but emission is random, and randomness is heat. Each absorbed photon adds $\hbar k$ along the beam — the directed part that cools — but the atom then re-emits in a random direction, so the emission recoils average to zero drift while their mean square grows steadily. The result is diffusive growth of $\langle p^2\rangle$ at a rate $\sim\hbar^2k^2$ times the scattering rate. This recoil heating is the price of dissipation, and it is precisely what stops the temperature at $T_D$ instead of zero.

What is the difference between the Doppler limit and the recoil limit?

They are two different floors, and the recoil limit is far lower. The Doppler limit $k_BT_D=\hbar\Gamma/2$ comes from balancing friction against diffusion in the two-level model and is set by the linewidth. The recoil limit $k_BT_r=\hbar^2k^2/m$ is the kinetic-energy scale of a single photon kick — the coldest while still recoiling from one photon. For an allowed transition they differ by $T_D/T_r=m\Gamma/(2\hbar k^2)$ , a few hundred: for $^{87}$ Rb, $T_D=146\ \mu$ K but $T_r=362$ nK. Doppler cooling stops well above $T_r$ ; only sub-Doppler and sub-recoil methods reach it.

Why do real experiments measure temperatures below the Doppler limit?

Because real alkali atoms are not two-level systems. Their ground state is split into several magnetic sublevels, and the counter-propagating beams create polarisation gradients where the light's polarisation rotates through space. As an atom moves it is optically pumped among the sublevels so that it repeatedly climbs a light-shift potential hill and is then pumped to the bottom of the next one, losing energy each cycle. This Sisyphus mechanism, explained by Dalibard and Cohen-Tannoudji in 1989, cools far below $\hbar\Gamma/2$ , toward the recoil energy — which is why the 1988 sub-Doppler measurements were such a surprise.

What is the capture velocity, and why does molasses need a slower?

The capture velocity is roughly where the Doppler shift equals the linewidth, $v_c\sim\Gamma/k$ . Below it an atom sits on the steep flank of the resonance and feels linear friction $-\alpha v$ ; above it the Doppler shift throws it past resonance, the force falls, and cooling fails. For $^{87}$ Rb, $v_c\approx4.7$ m/s — but atoms from an oven move at hundreds of metres per second. A Zeeman slower (or loading from the slow tail of a vapour) first brings them below $v_c$ so the molasses can grab them, setting a trap's loading rate.