Optical Tweezers for Qubits

An optical tweezer is the quietest cage in physics: a single tightly focused laser beam, detuned far below an atomic resonance, whose intensity gradient pulls one neutral atom into its brightest point and holds it there against gravity and the vacuum. The trapping force is not a push from radiation pressure but the conservative gradient of the atom's own AC Stark shift — the light polarises the atom, the induced dipole lowers its energy where the field is strongest, and a diffraction-limited focus a micrometre across becomes a bowl a few thousandths of a kelvin deep. Made that small, the bowl holds at most one atom: two atoms sharing it collide in the light and are both ejected, a collisional blockade that leaves every trap holding zero or one. That single-atom cage is the building block of a quantum computer. Because the trap is only a spot of light, its position is whatever the optics choose: acousto-optic deflectors and holographic modulators paint hundreds to thousands of tweezers at once, and moving tweezers physically carry atoms from wherever they happened to load into a target pattern, converting a coin-flip loading process into a defect-free register. Inside each atom a qubit lives in a pair of long-lived hyperfine or optical-clock states, single-qubit gates are driven by microwaves or Raman light, and two atoms are entangled by exciting them to giant Rydberg states whose dipole–dipole interaction forbids double excitation. The same hardware thus supplies identical qubits, connectivity rewired in software by moving the atoms, and parallel readout by fluorescence — the architecture that has pushed neutral-atom processors past a thousand qubits. This article builds the trap quantitatively: the dipole potential, its scattering shadow, the trap frequencies, the heating budget, and the assembly that makes the register.

Key takeaways

An optical tweezer is a far-detuned focused laser beam that traps a single neutral atom in the conservative gradient of its AC Stark shift — an optical dipole trap a few micrometres wide and about a millikelvin deep.
Red-detuned light gives an attractive potential U ∝ I/Δ pulling the atom to the intensity maximum; because the trap volume is tiny, a collisional blockade leaves each site with zero or one atom, loaded with probability near one-half.
Photon scattering — the source of heating and decoherence — scales as I/Δ², so the ratio of heating to trap depth is just Γ/Δ: run far off resonance and the trap stays deep while the scattering all but vanishes.
Moving tweezers rearrange stochastically loaded atoms into defect-free, arbitrary geometries, turning a probabilistic resource into a deterministic qubit register — the defining capability of the platform.
Qubits are encoded in hyperfine or optical-clock states and entangled through the Rydberg blockade; the trap must run at a magic wavelength or be pulsed off so its light shift does not dephase the qubit.
Key idea: a far-detuned focused beam is a conservative trap whose depth scales as I/Δ while its photon-scattering rate scales as I/Δ², so large detuning buys a deep, quiet trap — and a high-NA focus localises a single atom that can be carried anywhere to build a defect-free array.

The core insight

A far-detuned focused beam is a conservative trap whose depth scales as I/Δ and whose photon-scattering rate scales as I/Δ²; choosing large detuning buys a deep trap with negligible heating, while a high-NA focus localises a single atom that can be moved at will to build defect-free arrays.

Where this lives today

Optical-tweezer arrays have become one of the fastest-scaling routes to a quantum computer. Reconfigurable atom arrays now hold thousands of qubits: Atom Computing crossed the thousand-atom mark in 2023, QuEra's Aquila runs 256-atom programmable simulations in the cloud, and Pasqal and others are commercialising the same architecture. The signature capability — moving atoms mid-circuit — has turned out to be exactly what quantum error correction wants: in 2023 a Harvard–MIT–QuEra collaboration used coherent atom shuttling and parallel Rydberg gates to run algorithms on dozens of error-corrected logical qubits, performing transversal logical gates by physically carrying halves of code blocks into adjacency, a demonstration that would be impossible with lithographically fixed wiring. The same tweezers underpin a second revolution in metrology: arrays of alkaline-earth atoms trapped at a magic wavelength serve as tweezer optical clocks, where interrogating a hundred single-atom references in parallel improves stability while single-atom readout lets outliers be detected and discarded. The frontier now is faithful movement and gates at scale — keeping coherence while shuttling, cooling and re-cooling buses of motion, extending arrays toward the 10⁴–10⁵ physical qubits that fault tolerance will demand — all downstream of a marble held in a bowl of light.

The physical picture

Picture a marble rolling in a smooth bowl. The bowl is not made of glass but of light: where the laser is brightest, a red-detuned atom's energy is lowest, so it rolls to the focus and oscillates about it like a tiny pendulum. The trap is soft — a depth of about a millikelvin, thousands of times colder than room temperature — so the atom must first be laser-cooled to microkelvins before the bowl can hold it at all. Once captured it is a quantum particle in a near-harmonic well, its motion quantised into vibrational levels spaced by ħω. The first figure carves that bowl from a focused Gaussian beam and pins the numbers that a real ⁸⁷Rb tweezer delivers.

Figure

A ⁸⁷Rb atom trapped at the focus of a 1064 nm beam (w₀ = 1 μm, P = 12 mW). The intensity maximum is a potential minimum U₀/k_B ≈ 1.0 mK deep, giving a radial trap frequency ω_r/2π ≈ 98 kHz and an axial ω_z/2π ≈ 23 kHz. Far off resonance (Δ/2π ≈ −102 THz) photon scattering is only Γ_sc ≈ 7.6 per second, since ħΓ_sc/U₀ = Γ/Δ ≈ 6×10⁻⁸

Because each trap is only a steerable spot of light, you can drag an atom across the chip by sweeping the focus, so connectivity becomes a software choice rather than a wiring constraint — any two qubits can be brought adjacent for a gate and then returned. And single-atom occupancy is not magic but statistics plus collisions: a microtrap loaded from a cold cloud captures a Poisson-random number of atoms, but a pairwise light-assisted collision instantly ejects any two-atom occupancy, leaving zero or one with roughly equal odds. That coin-flip is the raw material; the assembly step perfects it. The second figure shows the move that makes the platform a register — imaging the random load and rearranging atoms into a defect-free target.

Figure

From random loading to a defect-free register. A 15×15 = 225-trap reservoir loads each site with probability P₁ ≈ ½, so on average ≈ 112 atoms appear at random. A camera reads the pattern, a computer plans a schedule, and moving tweezers carry atoms into a 10×10 = 100-site target — filling it defect-free with success probability ≈ 96% because 112 loaded comfortably exceeds the 100 needed

The core equations

The trap is the conservative part of the light shift of a driven atom. Treating the atom as a classical oscillator of polarisability α in a field of intensity I, the cycle-averaged interaction energy of the induced dipole with the field is the dipole potential; specialising α to a two-level Lorentz oscillator far from resonance gives the working form used throughout atom trapping.

U_{\mathrm{dip}}(\mathbf{r}) = \frac{3\pi c^2}{2\omega_0^3}\,\frac{\Gamma}{\Delta}\,I(\mathbf{r}) \qquad (1)

where: $U_{\mathrm{dip}}$ — conservative trapping potential; $\omega_0$ — atomic transition frequency; $\Gamma$ — natural linewidth of the excited state; $\Delta=\omega_L-\omega_0$ — laser detuning; $I(\mathbf{r})$ — local intensity. Reading it: for red detuning $\Delta<0$ the potential is negative, so the atom is pulled to the intensity maximum, and its depth grows linearly with intensity and inversely with detuning. Every real trap also scatters photons, and that scattering — recoil heating and loss of coherence — is the price of the trap. It comes from the absorptive (imaginary) part of the same polarisability, and it carries an extra factor of $\Gamma/\Delta$ relative to the potential, which is the single most important design relation in the field.

\Gamma_{\mathrm{sc}}(\mathbf{r}) = \frac{3\pi c^2}{2\hbar\omega_0^3}\left(\frac{\Gamma}{\Delta}\right)^2 I(\mathbf{r}),\qquad \hbar\,\Gamma_{\mathrm{sc}} = \frac{\Gamma}{\Delta}\,U_{\mathrm{dip}} \qquad (2)

where: $\Gamma_{\mathrm{sc}}$ — incoherent photon-scattering rate; $\hbar$ — reduced Planck constant. Reading it: the trap scales as $I/\Delta$ but the heating as $I/\Delta^2$ , so at fixed depth the scattering falls as $1/\Delta$ — push the detuning to hundreds of terahertz, add power to restore the depth, and the trap goes quiet. The trap's shape is the shape of the focus. A Gaussian beam of power $P$ focused to a waist $w_0$ has a peak intensity $2P/\pi w_0^2$ at the centre and spreads over the Rayleigh range $z_R$ along the axis; evaluating the potential at the peak gives the trap depth.

I(r,z)=\frac{2P}{\pi w(z)^2}\,e^{-2r^2/w(z)^2},\qquad U_0=\frac{3\pi c^2}{2\omega_0^3}\frac{\Gamma}{|\Delta|}\,\frac{2P}{\pi w_0^2} \qquad (3)

where: $w(z)=w_0\sqrt{1+(z/z_R)^2}$ — beam radius; $z_R=\pi w_0^2/\lambda$ — Rayleigh range; $U_0$ — peak depth. Reading it: depth is linear in power and scales as $1/w_0^2$ , so tight focusing is doubly rewarded — a smaller waist deepens the trap and, as the next equation shows, stiffens it. Near the focus the potential is harmonic, and its two curvatures — steep across the waist, gentle along the Rayleigh range — set the radial and axial trap frequencies. Because $z_R\gg w_0$ , the trap is always a cigar.

\omega_r=\sqrt{\frac{4U_0}{m w_0^2}},\qquad \omega_z=\sqrt{\frac{2U_0}{m z_R^2}},\qquad \frac{\omega_r}{\omega_z}=\sqrt{2}\,\frac{z_R}{w_0} \qquad (4)

where: $\omega_r,\omega_z$ — radial and axial harmonic trap frequencies; $m$ — atomic mass. Reading it: the transverse confinement is tight (tens to hundreds of kilohertz) while the axial is several times looser; the ratio is fixed by the focusing geometry alone. Loading is a coin flip enforced by collisions, and assembly is what turns many coin flips into a filled lattice. The collisional blockade fixes the single-trap occupancy; a surplus reservoir plus rearrangement fixes the register.

P_1\approx\tfrac12,\qquad P_{\mathrm{fill}}=\Pr\!\big[\mathrm{Binom}(N,P_1)\ge M\big]\xrightarrow{\,N\gtrsim 2M\,}1 \qquad (5)

where: $P_1$ — single-atom loading probability per trap (collisional-blockade value); $N$ — reservoir traps; $M$ — target sites; $P_{\mathrm{fill}}$ — probability the target can be filled defect-free. Reading it: because the number loaded is $\approx N/2\pm\sqrt N/2$ , a reservoir of about twice the target size almost always yields enough atoms to complete the pattern.

Derivation AThe dipole potential and its scattering shadow

The goal is to show, from the dressed-atom picture of a two-level atom in a classical field, that the trap depth and the photon-scattering rate are two faces of the same light shift — the real part traps and the imaginary part heats — and that their ratio collapses to the single number Γ/Δ, which is the reason optical traps run hundreds of terahertz off resonance. We diagonalise the driven two-level Hamiltonian, read off the ground-state shift, extract the excited-state admixture that leaks photons, and divide the two.

Step 1. In the rotating frame the atom driven at Rabi frequency Ω with detuning Δ has Hamiltonian $H=-\hbar\Delta\,|e\rangle\langle e|+\tfrac{\hbar\Omega}{2}(|e\rangle\langle g|+\mathrm{h.c.})$ , whose eigenvalues are $E_\pm=\tfrac{\hbar}{2}(\Delta\pm\sqrt{\Delta^2+\Omega^2})$ . Why: the light does not shift a bare level, it shifts the dressed levels — the true energies of atom-plus-field — and those shifts are what the atom actually feels as a potential.

Step 2. For $|\Delta|\gg\Omega$ expand the square root: the ground state shifts by $\delta E_g\approx\hbar\Omega^2/4\Delta$ . Since $\Omega^2\propto I$ , this is exactly the intensity-proportional dipole potential, negative and attractive for red detuning $\Delta<0$ . Why: this is equation (1) in disguise — the dressed-state shift and the classical-oscillator result must agree, and they do, fixing the sign and the $1/\Delta$ scaling.

Step 3. The same large-detuning expansion mixes a little excited state into the dressed ground state, with population $\Omega^2/4\Delta^2$ ; multiplying by the decay rate Γ gives the scattering rate $\Gamma_{\mathrm{sc}}=\Gamma\,\Omega^2/4\Delta^2$ , which falls as $1/\Delta^2$ . Why: an atom can only scatter a photon while it is momentarily excited, so the scattering rate is the excited fraction times how fast that fraction decays — one extra power of Δ in the denominator compared with the potential.

Step 4. Divide the two results: $\hbar\Gamma_{\mathrm{sc}}/U_{\mathrm{dip}}=\Gamma/\Delta$ . At fixed trap depth, doubling $|\Delta|$ (and the power, to hold the depth) halves the scattering. Why: this universal ratio is the whole engineering logic of dipole traps — it says decoherence is bought down linearly by detuning, so you go as far off resonance as your laser power allows. *The ⁸⁷Rb anchor confirms the ratio: ħΓ_sc/U₀ = Γ/|Δ| ≈ 6×10⁻⁸, consistent with the value shown in the figure.*

Go deeper · Derivation B · 2 minRecoil heating and the ground-state budget

Derivation BRecoil heating and the ground-state budget

The goal is to convert the abstract scattering rate into a heating rate a coherence budget can use, by counting the momentum kicks each scattered photon delivers, and to show that a far-detuned trap heats by only a fraction of a vibrational quantum per second — cold enough to hold a ground-state-cooled qubit through a circuit. We define the recoil energy, add the two uncorrelated kicks of a scattering event, multiply by the scattering rate, and express the result in vibrational quanta.

Step 1. A photon of wavevector $k=2\pi/\lambda$ carries momentum $\hbar k$ ; absorbing or emitting it changes the atom's kinetic energy by at most the recoil energy $E_R=(\hbar k)^2/2m=h^2/2m\lambda^2$ , which defines a temperature scale $T_R=2E_R/k_B$ . Why: the recoil energy is the irreducible granularity of light forces — every interaction with a photon is quantised in units of this kick, and for ⁸⁷Rb at 780 nm it is only E_R/k_B ≈ 0.18 μK.

Step 2. Each scattering event is two kicks: one directed kick on absorption along the beam, and one random-direction kick on spontaneous emission. Being uncorrelated, their energies add, so the mean energy gained per scattered photon is $\langle\Delta E\rangle=2E_R$ . Why: absorption and re-emission are independent momentum transfers, so the atom random-walks in momentum space — the energy grows diffusively, at $2E_R$ per step, not ballistically.

Step 3. Multiply by the scattering rate to get the heating rate $\dot E=2E_R\,\Gamma_{\mathrm{sc}}$ , and divide by ħω to express it in vibrational quanta: $\langle\dot n\rangle=(2E_R/\hbar\omega)\,\Gamma_{\mathrm{sc}}$ . Why: what a qubit cares about is not joules but whether the atom climbs out of its motional ground state, so quanta per second is the natural unit — and it must stay well below the trap frequency over a circuit.

Step 4. Insert the anchor numbers: $\Gamma_{\mathrm{sc}}\approx7.6\,\mathrm{s^{-1}}$ and $2E_R/\hbar\omega_r\approx0.077$ give $\langle\dot n\rangle\approx0.6$ quanta per second — a recoil-heating time of order a second, far longer than any gate. Why: this is the payoff of the Γ/Δ argument in physical units — the far-detuned trap is quiet enough that scattering-driven heating is negligible on gate timescales, and residual heating is dominated by technical trap noise instead. *For ⁸⁷Rb at 780 nm, inserting λ and m gives E_R/h ≈ 3.77 kHz and T_R ≈ 0.36 μK, confirming the factor-of-two energy gain per scattering event used above.*

Go deeper · Derivation C · 1 minThe magic wavelength that stops the trap from dephasing the qubit

Derivation CThe magic wavelength that stops the trap from dephasing the qubit

The goal is to explain why the very light that traps the atom threatens to dephase its qubit, and how a special choice of trap wavelength removes the threat exactly — the same trick that underlies optical-lattice clocks. We write the light shift of each qubit level, form their difference, show it maps intensity fluctuations into frequency fluctuations, and impose the condition that makes the difference vanish everywhere.

Step 1. Each qubit state feels its own dipole potential set by its own polarisability at the trap frequency: $U_i(\mathbf{r})=-\alpha_i(\omega_L)\,I(\mathbf{r})/2\epsilon_0 c$ for $i\in\{0,1\}$ . Why: the two qubit levels are different atomic states, so in general they polarise differently and are shifted by different amounts — the trap is state-dependent unless something is arranged.

Step 2. The qubit transition frequency is therefore shifted by $\hbar\,\delta(\mathbf{r})=U_1(\mathbf{r})-U_0(\mathbf{r})\propto[\alpha_1-\alpha_0]\,I(\mathbf{r})$ . A thermal atom samples a range of intensities as it oscillates, so this differential shift spreads the qubit frequency and dephases the superposition. Why: coherence is a well-defined phase between the two levels, and a position-dependent energy gap turns the atom's motion into a random clock rate — inhomogeneous dephasing.

Step 3. Impose the magic condition $\alpha_0(\omega_{\mathrm{magic}})=\alpha_1(\omega_{\mathrm{magic}})$ . At this special wavelength the two polarisabilities cross, $\delta(\mathbf{r})\equiv0$ everywhere, and the trap no longer touches the qubit frequency at all. Why: if both levels ride up and down by exactly the same amount, the gap between them is flat across the whole trap, and the atom's motion no longer scrambles its phase — decoupling coherence from position.

Step 4. Where no convenient magic wavelength exists, pulse the trap off during sensitive operations, choose a hyperfine encoding whose differential shift is intrinsically tiny, or apply a spin echo to refocus the slowly varying shift. Why: real level structures do not always offer a magic wavelength at a workable power and scattering rate, so the platform keeps a toolbox of partial fixes, each trading coherence against atom loss or complexity.

Limiting cases

The trap and its heating both hinge on the sign and size of the detuning and on the curvature of the focus; the table reads off the regimes.

Case	Result	Meaning
Far red detuning ( $\Delta<0$ , $\vert\Delta\vert\gg\Gamma$ )	$U\propto I/\Delta<0$ , $\Gamma_{\mathrm{sc}}\propto I/\Delta^2$	Attractive conservative trap with scattering suppressed as $1/\Delta^2$ — the standard storage regime; more detuning needs more power for the same depth but heats far less.
Blue detuning ( $\Delta>0$ )	$U\propto I/\Delta>0$	Repulsive: atoms are pushed toward dark regions, used in bottle beams and box traps to minimise the light shift, at the cost of trapping in an intensity minimum.
Near resonance ( $\vert\Delta\vert\sim\Gamma$ )	$\Gamma_{\mathrm{sc}}\sim\Gamma\,s/(1+s)$	Scattering saturates and dissipative radiation pressure (as in a MOT) overwhelms the conservative force — good for cooling, useless for coherent trapping.
Harmonic limit (cold atom)	$U\approx-U_0+\tfrac12 m(\omega_r^2 r^2+\omega_z^2 z^2)$	Near the bottom the trap is separable and harmonic with $\omega_r\gg\omega_z$ ; the cigar shape follows from the focus being tighter across the waist than along the Rayleigh range.
Magic wavelength	$\alpha_0(\omega_L)=\alpha_1(\omega_L)$	The differential light shift vanishes and the trap stops dephasing the qubit — the condition that lets a trapped atom keep second-scale coherence.

Common trapThe trap that holds the atom also shifts its qubit — and depth is not stiffness

The most damaging misconception is that a deep trap is automatically a good qubit host. The same light that makes the well also Stark-shifts the qubit transition, and because the two qubit states generally have different polarisabilities, a naive trap imprints a position-dependent frequency shift that dephases the qubit as the atom moves — coherence dies not from scattering but from the trap's own potential. The cure is a magic wavelength where both states shift equally, or pulsing the trap off during sensitive gates. A second, quieter trap: do not confuse trap depth with trap stiffness. Depth is set by the peak intensity ( $U_0\propto P/w_0^2$ ), but the confinement frequency is set by the intensity curvature ( $\omega\propto\sqrt{U_0}/w_0$ radially, $\sqrt{U_0}/z_R$ axially). A trap can be a millikelvin deep yet loose along its axis, where $z_R\gg w_0$ makes $\omega_z$ several times smaller than $\omega_r$ — which is exactly why axial cooling and axial heating dominate the motional error budget.

Worked examples

1 · Depth, frequencies, and quietness of a ⁸⁷Rb tweezer (fully worked)

A 1064 nm beam of power P = 12 mW is focused to a waist w₀ = 1 μm to trap ⁸⁷Rb (mass m = 1.44×10⁻²⁵ kg). Its D2 line is at λ₀ = 780 nm with linewidth Γ/2π = 6.07 MHz. Find the peak intensity, the trap depth, the radial and axial trap frequencies, and the photon-scattering rate.

Step 1. Peak intensity at the focus: $I_{\max}=2P/\pi w_0^2=2(0.012)/[\pi(10^{-6})^2]=\mathbf{7.6\times10^{9}\ W/m^2}$ , about 760 kW/cm² — a colossal intensity concentrated in a spot one wavelength across.

Step 2. Detuning and depth. The trap sits $\Delta/2\pi=(c/1064-c/780)\approx\mathbf{-102\ THz}$ below resonance, so $|\Delta|/\Gamma\approx1.7\times10^7$ — deep in the far-detuned regime. Equation (3) gives $U_0=(3\pi c^2/2\omega_0^3)(\Gamma/|\Delta|)I_{\max}$ , which evaluates to $U_0/k_B\approx\mathbf{1.0\ mK}$ (equivalently $U_0/h\approx20\ \mathrm{MHz}$ ). A millikelvin-deep well needs microkelvin atoms — hence the cooling that precedes loading.

Step 3. Trap frequencies. With $z_R=\pi w_0^2/\lambda=\pi(10^{-6})^2/(1064\times10^{-9})=\mathbf{3.0\ \mu m}$ , equation (4) gives $\omega_r/2\pi=\sqrt{4U_0/mw_0^2}/2\pi\approx\mathbf{98\ kHz}$ and $\omega_z/2\pi\approx\mathbf{23\ kHz}$ . Their ratio $\sqrt2\,(z_R/w_0)\approx4.2$ confirms the cigar shape — the trap is four times stiffer across the waist than along the beam.

Step 4. Scattering. Equation (2) gives $\Gamma_{\mathrm{sc}}=(\Gamma/\Delta)U_0/\hbar$ ; with $\Gamma/|\Delta|\approx6\times10^{-8}$ and $U_0/\hbar\approx2\pi\times20\ \mathrm{MHz}$ , the rate is only $\Gamma_{\mathrm{sc}}\approx\mathbf{7.6\ photons/s}$ — a scattering time of about 130 ms. Converting to heating (Worked Example logic of Derivation 2): $\langle\dot n\rangle\approx0.6$ quanta/s, negligible over a microsecond-to-millisecond circuit. A 1 mK trap, 98 kHz radial, that scatters under 8 photons a second.

2 · How big a reservoir fills a 100-qubit register?

You want a defect-free 10×10 = 100-atom register. Loading is a collisional-blockade coin flip, P₁ = ½. How many reservoir traps N do you need so the target can be filled with ≳ 95% probability, and what does that imply about rearrangement?

Step 1. The number of atoms actually loaded into N traps is $\mathrm{Binom}(N,\tfrac12)$ , with mean $N/2$ and standard deviation $\sqrt N/2$ . You succeed whenever at least M = 100 atoms appear, since rearrangement can then carry 100 of them into the target. So you need $\Pr[\mathrm{Binom}(N,\tfrac12)\ge100]\gtrsim0.95$ .

Step 2. Try N = 200 (mean exactly 100): success is only ≈ 53% — a bare mean leaves you short half the time. Try N = 225 (a 15×15 reservoir, mean 112.5, σ ≈ 7.5): now 100 is 1.7σ below the mean, and $\Pr[\ge100]\approx\mathbf{96\%}$ . Pushing to N = 256 (16×16, mean 128) makes the target 3.5σ down and success > 99.9%.

Step 3. The lesson is the rule of thumb $N\gtrsim2M$ : a reservoir about twice the target size buys near-certain completion, because loading fluctuations scale only as $\sqrt N$ . The cost is paid in rearrangement — the control computer must plan and execute of order M moves per shot, each a moving-tweezer transport with its own small loss probability, which is why transport fidelity (not loading) ultimately caps register size. A 15×15 reservoir fills a 10×10 target ≈ 96% of the time; 16×16 exceeds 99.9%.

Physics of the portrait

The optical tweezer is one of those rare tools whose inventor was told, early on, that it could not work — that light was far too feeble to push matter around. The number in the first figure, a colossal 760 kW/cm² squeezed into a micron, is the answer: focus hard enough and the gradient of that intensity becomes a real, usable force.

Origins · 1 minArthur Ashkin and how this was discovered

Arthur Ashkin

1922–2020 · laser physics and optical trapping

Arthur Ashkin was an American physicist at Bell Laboratories who spent decades turning the momentum of light into a laboratory tool, and who received the 2018 Nobel Prize in Physics — at ninety-six, then the oldest laureate ever — for inventing optical tweezers. In 1970 he showed that a focused laser beam could accelerate and even trap transparent micron-sized spheres, exploiting the gradient force that pulls a dielectric toward high intensity; sceptics had assumed radiation pressure would simply blow particles away, and Ashkin's insight was that a tight enough focus makes the transverse gradient win. In 1986, with Steven Chu, John Bjorkholm, and Joseph Dziedzic, he demonstrated the single-beam gradient trap — the optical tweezer proper — and Chu soon extended the same gradient force to hold laser-cooled neutral atoms, a line that led to the 1997 Nobel Prize for laser cooling and trapping. Ashkin himself turned the tweezers on the living world, trapping bacteria and viruses without harming them and measuring the piconewton forces of individual molecular motors, founding the field of single-molecule biophysics. The quantum-computing application came later and from other hands — trapping one atom per tweezer, imaging it, and rearranging arrays of them — but every neutral-atom processor rests on the gradient force Ashkin first made strong enough to hold matter still.

History

1970

— Ashkin shows a focused laser can accelerate and trap transparent micron particles by the gradient force

1986

— Ashkin, Bjorkholm, Chu and Dziedzic demonstrate the single-beam gradient trap: the optical tweezer

2001

— Schlosser, Reymond, Protsenko and Grangier trap a single ⁸⁷Rb atom, observing the collisional blockade

2016

— Endres–Lukin and Barredo–Browaeys groups assemble defect-free atom arrays by real-time rearrangement

2018

— Ashkin shares the Nobel Prize in Physics for the optical tweezers

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, Ch. 7 (physical realisations) · Preskill, Lecture Notes on Quantum Computation, Ch. 7 (physical implementations) · Grimm, Weidemüller and Ovchinnikov, "Optical Dipole Traps for Neutral Atoms", Adv. At. Mol. Opt. Phys. 42, 95 (2000) · Saffman, Walker and Mølmer, "Quantum information with Rydberg atoms", Rev. Mod. Phys. 82, 2313 (2010) · Metcalf and van der Straten, Laser Cooling and Trapping (Springer, 1999). The $U\propto I/\Delta$ , $\Gamma_{\mathrm{sc}}\propto I/\Delta^2$ scalings and their ratio $\Gamma/\Delta$ follow from the dressed two-level atom. The ⁸⁷Rb anchor numbers of Worked Example 1 — $I_{\max}\approx7.6\times10^9$ W/m², $U_0/k_B\approx1.0$ mK, $\omega_r/2\pi\approx98$ kHz, $\omega_z/2\pi\approx23$ kHz, $\Gamma_{\mathrm{sc}}\approx7.6$ s⁻¹ — follow from the stated P, w₀, λ, m, Γ. The recoil energy is $E_R=h^2/2m\lambda^2$ with a factor-of-two heating rate per scattering event. The binomial assembly probabilities are those of Worked Example 2. Species constants (λ, m, Γ) are tabulated atomic data; the two-level dipole formula is an order-of-magnitude model for a real multilevel atom, which sums contributions from both D lines.

Prerequisites. DiVincenzo Criteria · Sideband Cooling · Trapped-Ion Qubits. Leads to. Rydberg Blockade · Mølmer–Sørensen Gate · Dispersive Readout.

Ask the tutor: Ask me to size a tweezer for a different atom or trap wavelength, or to walk through why moving atoms is exactly what error correction wants, and I will take it step by step.

Is the tweezer force radiation pressure pushing the atom, or something else?

Something else — it is the conservative gradient force, not radiation pressure. Radiation pressure is the scattering force: absorbed photons push the atom along the beam, and it is dissipative and always present near resonance. The tweezer instead exploits the gradient of the atom's light shift: a far-detuned beam barely scatters, but its induced dipole has an energy that is lowest where the intensity is highest, so the atom is pulled up the intensity gradient toward the focus. Because it is conservative, this force can trap — an atom can sit at the potential minimum and oscillate — whereas radiation pressure only pushes. The whole art of dipole trapping is detuning far enough that the gradient force dominates and the scattering force nearly vanishes.

Why must the atom be laser-cooled before it can be trapped?

Because the trap is astonishingly shallow. A typical tweezer is about a millikelvin deep in temperature units, which sounds cold but is enormous compared with the atom's thermal energy after cooling: a room-temperature atom moving at hundreds of metres per second would fly straight through the micron-wide focus without noticing it. Atoms are first slowed and cooled in a magneto-optical trap to tens of microkelvins, so their kinetic energy is a small fraction of the trap depth and they stay bound. Many experiments then cool further — Raman sideband cooling to near the motional ground state — because a qubit that is not just trapped but sitting in its lowest vibrational level is far less sensitive to residual light shifts and heating.

What actually enforces one-atom-per-trap — is it a quantum exclusion rule?

No, it is collisions, not statistics. Neutral atoms are not forbidden from sharing a trap the way identical fermions are; a large trap can hold many. The single-atom limit comes from the tiny trap volume plus light-assisted collisions: when two atoms occupy the same micron-scale focus, near-resonant light drives them into a collision that converts electronic energy into kinetic energy and ejects both. The result is a parity effect — the trap ends with zero or one atom, never two — and for standard loading the two outcomes are roughly equally likely, giving P₁ ≈ ½. This "collisional blockade" is why single-atom loading is probabilistic and why a rearrangement step is needed to build filled arrays.

Why not just make the reservoir enormous so every target site is guaranteed?

Because the bottleneck moves from loading to movement. Doubling the reservoir does make it nearly certain that enough atoms are present — the shortfall probability falls exponentially once the mean load exceeds the target. But every atom that ends up in the target had to be physically carried there by a moving tweezer, and each transport has a small loss and heating probability. With more reservoir traps you also need more moves, longer schedules, and more time during which atoms can be lost to background-gas collisions. In practice a reservoir of roughly twice the target size is the sweet spot: large enough that loading almost never fails, small enough that the rearrangement stays fast and low-loss. Transport fidelity, not loading, sets the ceiling.

How are qubits stored and gated once the atoms are in place?

The qubit lives inside each atom, in a pair of long-lived internal states — typically two hyperfine ground levels of an alkali (the clock states of ⁸⁷Rb or ¹³³Cs, coherent for seconds) or an optical-clock transition in an alkaline-earth-like atom such as ¹⁷¹Yb or ⁸⁸Sr. Single-qubit rotations are driven globally by microwaves or locally by focused Raman laser beams. Two-qubit gates use the Rydberg blockade: a laser briefly excites atoms to a giant, high-n Rydberg state whose strong dipole–dipole interaction shifts the doubly-excited state out of resonance within a blockade radius, so two nearby atoms cannot both be excited — and that conditional dynamics realises controlled-phase and CZ gates. The tweezer geometry, set in software, programs which atoms are close enough to interact.

Why is the trap light itself a threat to the qubit, and how is that handled?

Because the same AC Stark shift that traps the atom also shifts the qubit's transition frequency, and it does so by different amounts for the two qubit states whenever their polarisabilities differ. As the atom oscillates through regions of different intensity, its qubit frequency wobbles, and averaging over that spread dephases the superposition — an error that grows with trap depth and atom temperature. The clean fix is a magic wavelength, where the two states have identical polarisability so the differential shift vanishes everywhere; this is borrowed straight from optical-lattice-clock technology. Where no magic wavelength is practical, experiments pulse the trap off during the most sensitive operations, pick encodings with intrinsically small differential shifts, or apply spin-echo sequences to refocus the slowly varying shift.