Simple Algorithm
Quantum Computing · Excited State · 22 min

Trapped Ions

Single atoms held by electric fields and driven by lasers.

A single atom with one electron removed, floating in ultra-high vacuum about a hair's width above a chip, is one of the best qubits ever built. Every ion of a species is identical, it can stay coherent for seconds or longer, and laser light can entangle any two ions in a chain through the vibrations they share.

What you’ll learn

  • Explain why an ion trap needs an oscillating field, and derive the secular frequency of a Paul trap from the ponderomotive pseudopotential.
  • Compare hyperfine, optical and Zeeman ion qubits by coherence and control.
  • Describe Doppler and resolved-sideband cooling, and compute the Lamb–Dicke parameter.
  • Explain how shared motional modes act as a quantum bus, and build a bus-mediated entangling gate.
  • Analyze state-dependent fluorescence readout and its limits.

Holding a single atom

An ion has a net charge, so electric fields can push it. The catch is Earnshaw's theorem: in free space the electrostatic potential obeys Laplace's equation ∇2Φ=0\nabla^2\Phi = 0, so its curvatures along the three axes sum to zero. A static field can confine a charge along some directions only by expelling it along another. You get a saddle, never a bowl.

Wolfgang Paul's way out is to make the saddle flip. An electric quadrupole driven at a radio frequency Ωrf\Omega_{\rm rf} (typically tens of MHz) alternately focuses and defocuses along each direction. Because the ion jiggles in step with the fast field, it samples a slightly stronger field during the half-cycle that pushes it inward than during the half-cycle that pushes it outward. The time-averaged force points toward the field null.

DerivationThe ponderomotive pseudopotential
  1. mx¨=eE0(x)cos⁡Ωt,x(t)=X(t)+ξ(t)m\ddot x = eE_0(x)\cos\Omega t, \qquad x(t) = X(t) + \xi(t)

    Split the motion into a slow secular part XX and a small fast micromotion ξ\xi, in a field of amplitude E0(x)E_0(x) oscillating at Ω≡Ωrf\Omega \equiv \Omega_{\rm rf}.

  2. ξ(t)≈−eE0(X)mΩ2cos⁡Ωt\xi(t) \approx -\frac{eE_0(X)}{m\Omega^2}\cos\Omega t

    To leading order the fast field just shakes the ion about a fixed XX.

  3. mX¨=eE0(X+ξ)cos⁡Ωt‾≈eE0′(X) ξcos⁡Ωt‾=−e2E0E0′2mΩ2m\ddot X = \overline{eE_0(X+\xi)\cos\Omega t} \approx eE_0'(X)\,\overline{\xi\cos\Omega t} = -\frac{e^2E_0E_0'}{2m\Omega^2}

    Average over one RF period, expanding the field to first order in ξ\xi and using cos⁡2Ωt‾=12\overline{\cos^2\Omega t} = \tfrac12.

  4. mX¨=−∂Ψ∂X,Ψ(X)=e2E0(X)24mΩ2m\ddot X = -\frac{\partial\Psi}{\partial X}, \qquad \Psi(X) = \frac{e^2E_0(X)^2}{4m\Omega^2}

    The slow motion feels a pseudopotential that grows with the field amplitude, so the ion is pushed toward the point where the RF field vanishes.

  5. E0=V0r02 x  ⇒  ωsec=eV02 mΩr02=q Ω22,q=2eV0mr02Ω2E_0 = \frac{V_0}{r_0^2}\,x \;\Rightarrow\; \omega_{\rm sec} = \frac{eV_0}{\sqrt2\,m\Omega r_0^2} = \frac{q\,\Omega}{2\sqrt2}, \qquad q = \frac{2eV_0}{m r_0^2\Omega^2}

    For a quadrupole of RF amplitude V0V_0 and electrode distance r0r_0 the pseudopotential is harmonic, Ψ=12mωsec2x2\Psi = \tfrac12 m\omega_{\rm sec}^2x^2. The dimensionless qq is the Mathieu stability parameter: the averaging needs q≪1q \ll 1, and the exact motion is stable only for q≲0.9q \lesssim 0.9.

Predict before you look

A linear Paul trap is driven at Ωrf/2π=30\Omega_{\rm rf}/2\pi = 30 MHz with Mathieu parameter q=0.3q = 0.3. Estimate the radial secular frequency ωsec/2π\omega_{\rm sec}/2\pi.

Show the answer

3.2 MHz

ωsec≈q Ωrf/(22)=0.3×30/2.83≈3.2\omega_{\rm sec} \approx q\,\Omega_{\rm rf}/(2\sqrt2) = 0.3\times30/2.83 \approx 3.2 MHz. The ion oscillates about ten times more slowly than the drive, which is what justifies averaging over the fast RF cycle.

Choosing the qubit levels

A trapped-ion qubit lives in two internal levels of the ion's valence electron. Species with a single valence electron (Be⁺, Mg⁺, Ca⁺, Sr⁺, Ba⁺, Yb⁺) have simple, alkali-like level schemes, and three kinds of level pairs are in use.

TypeExampleSplittingDriven byMain limits
Hyperfine (clock)171Yb+^{171}\mathrm{Yb}^+: ∣F=0,mF=0⟩↔∣F=1,mF=0⟩\ket{F{=}0, m_F{=}0} \leftrightarrow \ket{F{=}1, m_F{=}0}≈12.6\approx 12.6 GHzMicrowaves, or two-photon Raman lasersNo significant spontaneous decay; T2T_2 of seconds or longer, set by field noise and oscillator phase noise
Optical40Ca+^{40}\mathrm{Ca}^+: S1/2↔D5/2S_{1/2} \leftrightarrow D_{5/2}≈411\approx 411 THz (729 nm)An ultra-stable, narrow-linewidth laserT1≈1T_1 \approx 1 s, the lifetime of the metastable D5/2D_{5/2} level; laser phase noise
Zeeman40Ca+^{40}\mathrm{Ca}^+: S1/2S_{1/2}, m=±12m = \pm\tfrac12≈2.8\approx 2.8 MHz per gaussRF fields or Raman lasersMagnetic-field noise enters at first order
Representative choices; many other species and level pairs are used.

Single-qubit gates are Rabi rotations. A resonant drive with Rabi frequency Ω\Omega, applied for time tt, rotates the Bloch vector by the angle Ωt\Omega t about an equatorial axis set by the drive phase. Microwaves drive hyperfine qubits directly; lasers drive optical qubits directly, or drive hyperfine qubits through a two-photon Raman transition whose frequency difference matches the qubit splitting. Try it, including what happens off resonance.

Check your understanding

Which property is chiefly responsible for the seconds-long T2T_2 of a hyperfine clock qubit such as 171Yb+^{171}\mathrm{Yb}^+?

  1. The ion is cooled to the motional ground state
  2. Both levels belong to the electronic ground state, so neither decays appreciably, and their splitting is insensitive to magnetic fields at first order
  3. The splitting is optical, so room-temperature thermal radiation cannot excite it
  4. The RF field of the Paul trap screens out magnetic noise
Show the answer

B. Both levels belong to the electronic ground state, so neither decays appreciably, and their splitting is insensitive to magnetic fields at first order

Ground-state hyperfine levels have negligible spontaneous decay, and the clock transition shifts only quadratically with the field near zero field. What remains is residual field noise and the phase noise of the control oscillator, both of which can be suppressed further by echoes.

Cooling, and motion as a bus

Laser cooling comes first. In Doppler cooling, light tuned slightly below a strong transition of linewidth Γ\Gamma is absorbed preferentially by an ion moving toward the beam, and each scattering event removes a little momentum on average. It bottoms out at the Doppler limit kBTD=ℏΓ/2k_BT_D = \hbar\Gamma/2: about 0.5 mK for Ca⁺ (Γ/2π≈22\Gamma/2\pi \approx 22 MHz). In a trap of frequency ω/2π=1\omega/2\pi = 1 MHz that still leaves nˉ≈Γ/2ω≈10\bar n \approx \Gamma/2\omega \approx 10 vibrational quanta. Resolved-sideband cooling, below, removes the rest.

Cold ions in a linear trap repel each other into a chain, a Coulomb crystal. Two 40Ca+^{40}\mathrm{Ca}^+ ions in a 1 MHz axial well sit about 5.6 μm apart: close enough to share their motion, far enough apart for focused beams to address each one. Small oscillations of the chain decompose into normal modes, the center-of-mass mode at the axial frequency ωz\omega_z, a stretch mode at 3 ωz\sqrt3\,\omega_z, and so on. Each mode is a quantum harmonic oscillator shared by every ion.

Predict before you look

Estimate η\eta for a 40Ca+^{40}\mathrm{Ca}^+ ion (m=6.6×10−26m = 6.6\times10^{-26} kg) driven at 729 nm along an axial mode of frequency ω/2π=1\omega/2\pi = 1 MHz. Use ℏ=1.05×10−34\hbar = 1.05\times10^{-34} J s.

Show the answer

0.097

x0=ℏ/2mω=1.05×10−34/(2×6.6×10−26×6.3×106)≈11x_0 = \sqrt{\hbar/2m\omega} = \sqrt{1.05\times10^{-34}/(2\times6.6\times10^{-26}\times6.3\times10^{6})} \approx 11 nm and k=2π/(729 nm)≈8.6×106 m−1k = 2\pi/(729\ \text{nm}) \approx 8.6\times10^6\ \text{m}^{-1}, so η≈0.10\eta \approx 0.10. Small but not negligible: light can still change the motion, at a rate reduced by η\eta.

DerivationCarrier and sidebands
  1. HI=ℏΩ2 σ+exp⁡ ⁣[iη(ae−iωt+a†eiωt)] e−iδt+h.c.H_I = \frac{\hbar\Omega}{2}\,\sigma_+\exp\!\big[i\eta\big(a e^{-i\omega t} + a^\dagger e^{i\omega t}\big)\big]\,e^{-i\delta t} + \text{h.c.}

    A laser detuned by δ\delta from the qubit transition, in the interaction picture of qubit and oscillator. The exponential is the photon's momentum kick eikxe^{ikx}, with kx=η(ae−iωt+a†eiωt)kx = \eta(a e^{-i\omega t} + a^\dagger e^{i\omega t}).

  2. eiη(⋯ )≈1+iη(ae−iωt+a†eiωt)e^{i\eta(\cdots)} \approx 1 + i\eta\big(a e^{-i\omega t} + a^\dagger e^{i\omega t}\big)

    In the Lamb–Dicke regime, keep the first order in η\eta.

  3. δ=0: Ωn,n=Ω,δ=−ω: Ωn,n−1=ηΩn,δ=+ω: Ωn,n+1=ηΩn+1\delta = 0:\ \Omega_{n,n} = \Omega, \qquad \delta = -\omega:\ \Omega_{n,n-1} = \eta\Omega\sqrt n, \qquad \delta = +\omega:\ \Omega_{n,n+1} = \eta\Omega\sqrt{n+1}

    Each choice of δ\delta makes one term stationary; the others average away if Ω≪ω\Omega \ll \omega. The carrier flips the qubit alone, the red sideband ∣g,n⟩→∣e,n−1⟩\ket{g,n}\to\ket{e,n-1} removes a phonon, and the blue sideband ∣g,n⟩→∣e,n+1⟩\ket{g,n}\to\ket{e,n+1} adds one.

The modes are a bus. In the Cirac–Zoller proposal (1995), a red-sideband π-pulse on ion A maps its qubit onto a shared mode that starts with zero phonons; a conditional operation on ion B depends on whether a phonon is present; a final pulse maps the phonon back onto A. Because every ion couples to the same modes, any pair in the chain can interact: all-to-all connectivity. Build the protocol below, treating the two lowest levels of the mode as a qubit.

Gates you can run

Single-qubit gates use microwaves (often applied globally, with field gradients or light shifts for addressing) or focused laser beams; microwave-driven gates have reached error rates around 10−610^{-6}. Two-qubit gates use the motional bus. The native Mølmer–Sørensen gate is an Ising-type rotation:

UMS=exp⁡ ⁣(−iπ4 X⊗X)=12(I−i X⊗X),UMS∣00⟩=∣00⟩−i∣11⟩2U_{\rm MS} = \exp\!\Big(-i\frac{\pi}{4}\,X\otimes X\Big) = \frac{1}{\sqrt2}\big(\id - i\,X\otimes X\big), \qquad U_{\rm MS}\ket{00} = \frac{\ket{00} - i\ket{11}}{\sqrt2}
Any pair in a chain can be addressed. Gate times are typically tens to hundreds of μs, and the best two-qubit fidelities are around 99.9% as of the mid-2020s.

Reading out: bright and dark

Readout uses state-dependent fluorescence. In 171Yb+^{171}\mathrm{Yb}^+, light at 369.5 nm drives a nearly closed cycling transition from the F=1F{=}1 level (∣1⟩\ket{1}), which therefore scatters millions of photons per second: it is bright. The F=0F{=}0 level (∣0⟩\ket{0}) is 12.6 GHz off resonance and stays dark. Collection optics capture a few percent of the photons; a photomultiplier or camera counts them, and a threshold decides. Optical qubits use electron shelving: population in the metastable D5/2D_{5/2} level is dark to the cooling light.

DerivationThreshold detection
  1. P(k ∣ bright)=e−μbμbkk!,P(k ∣ dark)≈δk,0P(k\,|\,\text{bright}) = e^{-\mu_b}\frac{\mu_b^k}{k!}, \qquad P(k\,|\,\text{dark}) \approx \delta_{k,0}

    Photon counts are Poissonian, with mean μb\mu_b for a bright ion and essentially zero (background only) for a dark one.

  2. ϵbright=P(0 ∣ bright)=e−μb\epsilon_{\rm bright} = P(0\,|\,\text{bright}) = e^{-\mu_b}

    Declare "bright" for at least one count. A bright ion is misread only if it yields no count at all.

  3. μb=10  ⇒  ϵbright≈4.5×10−5\mu_b = 10 \;\Rightarrow\; \epsilon_{\rm bright} \approx 4.5\times10^{-5}

    Ten detected photons already make Poisson errors small. The real limits lie elsewhere: off-resonant excitation during detection can pump a bright ion into the dark manifold (or a dark one into the bright), and background light gives false counts. Reported fidelities range from about 99.9% to above 99.99%.

Check your understanding

For a hyperfine qubit such as 171Yb+^{171}\mathrm{Yb}^+, why can't you make the readout error arbitrarily small simply by detecting for longer?

  1. Photon counts become Gaussian rather than Poissonian for long windows
  2. The detector saturates
  3. The detection light slowly pumps the ion between the bright and dark manifolds, so the state being measured changes during the measurement
  4. The dark level ∣0⟩\ket{0} spontaneously decays to ∣1⟩\ket{1} during long windows
Show the answer

C. The detection light slowly pumps the ion between the bright and dark manifolds, so the state being measured changes during the measurement

Off-resonant excitation lets a bright ion occasionally fall into F=0F{=}0 (and a dark ion into F=1F{=}1). The longer the window, the likelier such a switch, so the optimal window balances the overlap of the count distributions against optical pumping.

Key takeaways

  • Paul traps confine ions with an RF quadrupole: the time-averaged pseudopotential gives secular frequencies ωsec≈qΩrf/22\omega_{\rm sec} \approx q\Omega_{\rm rf}/2\sqrt2, typically around a MHz.
  • Hyperfine clock qubits are field-insensitive at first order and essentially never decay (T2T_2 of seconds or more); optical qubits trade some lifetime for direct laser control.
  • Doppler cooling reaches about ten quanta and sideband cooling the ground state; the Lamb–Dicke parameter η=kx0≈0.1\eta = kx_0 \approx 0.1 sets how strongly light couples to motion.
  • Shared motional modes act as a bus that connects every ion in a chain, giving all-to-all two-qubit gates with fidelities near 99.9%.
  • State-dependent fluorescence gives readout fidelities of 99.9% and better, limited by optical pumping during detection rather than photon statistics.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

Why does a Paul trap use a radio-frequency field rather than a static one?

  1. A static field would heat the ion through micromotion
  2. The RF field cancels the ion's charge on average, so it behaves like a neutral atom
  3. The RF field is needed to drive the qubit transition
  4. Laplace's equation forbids a minimum of the electrostatic potential in free space, so a static field gives a saddle; a rapidly alternating saddle yields a confining time-averaged pseudopotential
Show the answer

D. Laplace's equation forbids a minimum of the electrostatic potential in free space, so a static field gives a saddle; a rapidly alternating saddle yields a confining time-averaged pseudopotential

Earnshaw's theorem rules out static electric confinement. The oscillating quadrupole produces the pseudopotential Ψ=e2E02/4mΩ2\Psi = e^2E_0^2/4m\Omega^2, which has a true minimum at the field null.

Question 2

Two identical ions share an axial well of frequency ωz\omega_z. At what frequency does the out-of-phase (stretch) mode oscillate?

  1. ωz\omega_z
  2. 2 ωz\sqrt2\,\omega_z
  3. 3 ωz\sqrt3\,\omega_z
  4. 2 ωz2\,\omega_z
Show the answer

C. 3 ωz\sqrt3\,\omega_z

In the center-of-mass mode the spacing never changes, so the Coulomb force plays no role and the frequency is ωz\omega_z. In the stretch mode the Coulomb "spring" adds a curvature equal to mωz2m\omega_z^2 to the trap's 12mωz2\tfrac12 m\omega_z^2 for the relative coordinate (reduced mass m/2m/2), giving ω2=3ωz2\omega^2 = 3\omega_z^2, independent of charge and mass.

Question 3

An ion is in ∣g,n=0⟩\ket{g, n=0}: internal ground state and motional ground state. You apply a pulse resonant with the red sideband. What happens?

  1. Nothing: the red sideband must remove a phonon, and there is none
  2. The ion goes to ∣e,1⟩\ket{e, 1}
  3. The ion goes to ∣e,0⟩\ket{e, 0} at a rate reduced by η\eta
  4. The ion Rabi-oscillates on the carrier at the full rate Ω\Omega
Show the answer

A. Nothing: the red sideband must remove a phonon, and there is none

The red-sideband coupling is ηΩn\eta\Omega\sqrt n, which vanishes for n=0n = 0. This dark state is what makes sideband cooling converge to the ground state.

Question 4

In the Lamb–Dicke regime, how does the blue-sideband Rabi frequency for ∣g,n⟩→∣e,n+1⟩\ket{g,n}\to\ket{e,n+1} depend on nn?

  1. ηΩ\eta\Omega, independent of nn
  2. ηΩn+1\eta\Omega\sqrt{n+1}
  3. ηΩn\eta\Omega\sqrt{n}
  4. η2Ω(n+1)\eta^2\Omega(n+1)
Show the answer

B. ηΩn+1\eta\Omega\sqrt{n+1}

The coupling is ηΩ ⟨n+1∣a†∣n⟩=ηΩn+1\eta\Omega\,\bra{n+1}a^\dagger\ket{n} = \eta\Omega\sqrt{n+1}. Hotter ions Rabi-flop on the sidebands at a spread of rates, which is one reason gates that rely on a single sideband need ground-state cooling.

Question 5

Why can't you simply keep adding ions to a single chain?

  1. Each ion couples more weakly to the shared modes, the mode spectrum crowds, and the center-of-mass mode heats faster, so gates get slower and harder to make clean
  2. Ions of the same species become distinguishable in long chains
  3. The Coulomb interaction is screened beyond about ten ions
  4. Earnshaw's theorem forbids more than a few ions in one trap
Show the answer

A. Each ion couples more weakly to the shared modes, the mode spectrum crowds, and the center-of-mass mode heats faster, so gates get slower and harder to make clean

Mode participation per ion falls like 1/N1/\sqrt N, neighboring modes become hard to resolve, and uniform field noise heats the center-of-mass mode in proportion to NN. Shuttling (QCCD) and photonic links are the two main answers.

Go deeper

Further reading

  • D. Leibfried, R. Blatt, C. Monroe & D. Wineland, "Quantum dynamics of single trapped ions", Reviews of Modern Physics 75, 281 (2003).
  • C. D. Bruzewicz, J. Chiaverini, R. McConnell & J. M. Sage, "Trapped-ion quantum computing: Progress and challenges", Applied Physics Reviews 6, 021314 (2019).
  • J. I. Cirac & P. Zoller, "Quantum computations with cold trapped ions", Physical Review Letters 74, 4091 (1995).
  • D. Maslov, "Basic circuit compilation techniques for an ion-trap quantum machine", New Journal of Physics (2017).