Simple Algorithm
Quantum Computing · Rydberg State · 30 min

The Mølmer–Sørensen Gate

Entangling ions through their shared motion.

Two ions can be entangled without exchanging a photon: push on them with a force whose sign depends on their spins, steer their shared motion around a closed loop in phase space, and the area of the loop becomes a spin-dependent phase. That is the Mølmer–Sørensen gate. Because only the area matters, it works even when the ions are not motionally cold.

What you’ll learn

  • Derive the Mølmer–Sørensen Hamiltonian from a bichromatic drive in the Lamb–Dicke and rotating-wave approximations.
  • Solve its dynamics exactly with the Magnus expansion, as a spin-dependent displacement times a geometric phase.
  • Derive the gate conditions δ=2ηΩK\delta = 2\eta\Omega\sqrt K and tg=2πK/δt_g = 2\pi K/\delta.
  • Prove the gate's insensitivity to the motional state, and quantify the error of imperfect loop closure.
  • Generalize to many ions and modes, and explain how pulse shaping closes every loop.

A spin-dependent force

Take two ions sharing one motional mode of frequency ν\nu, with Lamb–Dicke parameter η\eta (including each ion's participation in the mode). Both ions are illuminated by a bichromatic field with tones at ω0±μ\omega_0 \pm \mu, symmetric about the qubit frequency ω0\omega_0, where μ=ν+δ\mu = \nu + \delta lies close to the motional sidebands: ∣δ∣≪ν\abs{\delta} \ll \nu.

DerivationFrom two tones to one force
  1. HI=∑j=1,2 ∑±ℏΩ2 σ+(j) eiη(ae−iνt+a†eiνt) e∓iμt+iϕ±+h.c.H_I = \sum_{j=1,2}\ \sum_{\pm}\frac{\hbar\Omega}{2}\,\sigma_+^{(j)}\,e^{i\eta(a e^{-i\nu t}+a^\dagger e^{i\nu t})}\,e^{\mp i\mu t + i\phi_\pm} + \text{h.c.}

    Interaction picture with respect to the qubits and the mode; each tone has Rabi frequency Ω\Omega and optical phase ϕ±\phi_\pm.

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

    Lamb–Dicke expansion, valid when η2(2nˉ+1)≪1\eta^2(2\bar n+1) \ll 1.

  3. σ+e∓iμt⏟carrier: dropped,σ+a ei(μ−ν)t,  σ+a†e−i(μ−ν)t⏟kept,σ+a e−i(μ+ν)t,  σ+a†ei(μ+ν)t⏟dropped\underbrace{\sigma_+ e^{\mp i\mu t}}_{\text{carrier: dropped}},\qquad \underbrace{\sigma_+ a\,e^{i(\mu-\nu)t},\ \ \sigma_+ a^\dagger e^{-i(\mu-\nu)t}}_{\text{kept}},\qquad \underbrace{\sigma_+ a\,e^{-i(\mu+\nu)t},\ \ \sigma_+ a^\dagger e^{i(\mu+\nu)t}}_{\text{dropped}}

    Rotating-wave approximation. The carrier terms oscillate at μ≈ν\mu \approx \nu and average out if Ω≪ν\Omega \ll \nu. The red tone times aa and the blue tone times a†a^\dagger oscillate only at ±δ\pm\delta and survive; the remaining terms oscillate near 2ν2\nu.

  4. HMS(t)=ℏF Sϕ(a eiδt+a†e−iδt),F=ηΩ2,Sϕ=∑j(σx(j)cos⁡ϕ+σy(j)sin⁡ϕ)H_{\rm MS}(t) = \hbar F\,S_\phi\big(a\,e^{i\delta t} + a^\dagger e^{-i\delta t}\big), \qquad F = \frac{\eta\Omega}{2},\quad S_\phi = \sum_j \big(\sigma_x^{(j)}\cos\phi + \sigma_y^{(j)}\sin\phi\big)

    With suitable tone phases, what remains is a force on the mode whose sign is set by the collective spin operator SϕS_\phi. The mean of ϕ±\phi_\pm fixes the spin axis ϕ\phi, their difference the phase of the force (set to zero by the choice of time origin). SϕS_\phi commutes with HMS(t)H_{\rm MS}(t) at all times, so its eigenvalues s∈{−2,0,0,2}s \in \{-2, 0, 0, 2\} label independent motional trajectories.

Exact solution: loops in phase space

DerivationProof via the Magnus expansion
  1. U(t)=exp⁡(∑k≥1Ωk(t)),Ω1=−iℏ∫0tH(t1) dt1,Ω2=−12ℏ2∫0t ⁣dt1 ⁣∫0t1 ⁣dt2 [H(t1),H(t2)]U(t) = \exp\Big(\sum_{k\ge1}\Omega_k(t)\Big), \qquad \Omega_1 = -\frac{i}{\hbar}\int_0^t H(t_1)\,dt_1,\qquad \Omega_2 = -\frac{1}{2\hbar^2}\int_0^t\!dt_1\!\int_0^{t_1}\!dt_2\,\comm{H(t_1)}{H(t_2)}

    The Magnus series writes a time-ordered exponential as a single exponential; higher terms are nested commutators of HH at different times.

  2. Ω1=−iFS∫0t(a eiδt1+a†e−iδt1)dt1=S(α(t) a†−α∗(t) a)\Omega_1 = -iFS\int_0^t\big(a\,e^{i\delta t_1} + a^\dagger e^{-i\delta t_1}\big)dt_1 = S\big(\alpha(t)\,a^\dagger - \alpha^*(t)\,a\big)

    Integrating the exponentials gives a displacement of the mode by α(t)\alpha(t), multiplied by the spin operator.

  3. [H(t1),H(t2)]=2iℏ2F2S2 sin⁡δ(t1−t2)\comm{H(t_1)}{H(t_2)} = 2i\hbar^2F^2S^2\,\sin\delta(t_1-t_2)

    From [a,a†]=1\comm{a}{a^\dagger} = 1. The commutator is a c-number times S2S^2, which commutes with every H(t)H(t), so all nested commutators vanish and the series terminates at second order.

  4. Ω2=−iF2S2∫0t1−cos⁡δt1δ dt1=−iΦ(t) S2\Omega_2 = -iF^2S^2\int_0^t\frac{1-\cos\delta t_1}{\delta}\,dt_1 = -i\Phi(t)\,S^2

    The inner integral is (1−cos⁡δt1)/δ(1-\cos\delta t_1)/\delta.

  5. U(t)=eΩ1+Ω2=eΩ1eΩ2=D^(α(t)S) e−iΦ(t)S2■U(t) = e^{\Omega_1+\Omega_2} = e^{\Omega_1}e^{\Omega_2} = \hat D\big(\alpha(t)S\big)\,e^{-i\Phi(t)S^2}\qquad\blacksquare

    Ω1\Omega_1 and Ω2\Omega_2 commute, so the exponential factorizes.

Now read the result in phase space. In the eigenspace S=sS = s the mode is displaced along s α(t)s\,\alpha(t), a circle of radius ∣s∣F/δ\abs{s}F/\delta through the origin with period 2π/δ2\pi/\delta. At tK=2πK/δt_K = 2\pi K/\delta every circle closes, α(tK)=0\alpha(t_K) = 0, and the spins are disentangled from the motion whatever the motional state was. What remains is a pure spin phase, and Φ(tK) s2=2As\Phi(t_K)\,s^2 = 2\mathcal A_s is twice the phase-space area enclosed by trajectory ss (with α\alpha dimensionless).

DerivationThe gate condition
  1. Sϕ2=(σϕ(1)+σϕ(2))2=2 I+2 σϕ(1)σϕ(2)S_\phi^2 = \big(\sigma_\phi^{(1)} + \sigma_\phi^{(2)}\big)^2 = 2\,\id + 2\,\sigma_\phi^{(1)}\sigma_\phi^{(2)}

    For two ions, using (σϕ(j))2=I(\sigma_\phi^{(j)})^2 = \id.

  2. U(tK)=e−2iΦKexp⁡ ⁣(−2iΦK σϕ(1)σϕ(2)),ΦK≡Φ(tK)=2πKF2δ2U(t_K) = e^{-2i\Phi_K}\exp\!\big(-2i\Phi_K\,\sigma_\phi^{(1)}\sigma_\phi^{(2)}\big),\qquad \Phi_K \equiv \Phi(t_K) = \frac{2\pi K F^2}{\delta^2}

    After KK closed loops, only an Ising rotation is left.

  3. 2ΦK=π4  ⟺  δ=4FK=2ηΩK,tg=2πKδ=πKηΩ2\Phi_K = \frac{\pi}{4}\;\Longleftrightarrow\;\delta = 4F\sqrt K = 2\eta\Omega\sqrt K, \qquad t_g = \frac{2\pi K}{\delta} = \frac{\pi\sqrt K}{\eta\Omega}

    Maximal entanglement, UMS=exp⁡(−iπ4σϕσϕ)U_{\rm MS} = \exp(-i\tfrac\pi4\sigma_\phi\sigma_\phi), fixes the detuning. More loops need a larger detuning and a longer gate.

  4. UMS∣00⟩=12(∣00⟩−i e2iϕ∣11⟩)U_{\rm MS}\ket{00} = \tfrac{1}{\sqrt2}\big(\ket{00} - i\,e^{2i\phi}\ket{11}\big)

    Since σϕ∣0⟩=eiϕ∣1⟩\sigma_\phi\ket0 = e^{i\phi}\ket1. Every computational basis state is mapped to a maximally entangled state.

Predict before you look

At fixed laser power (fixed ηΩ\eta\Omega) you switch from a single-loop gate (K=1K = 1) to a K=4K = 4 gate, for example to average out slow drifts. By what factor does the gate duration change?

Show the answer

2 ×

tg=πK/(ηΩ)t_g = \pi\sqrt K/(\eta\Omega), so K=4K = 4 doubles it. The detuning doubles as well (δ=2ηΩK\delta = 2\eta\Omega\sqrt K), so each loop has a quarter of the area: four loops enclose the same total area, which is what sets the phase.

Check your understanding

You stop a single-loop gate halfway, at t=π/δt = \pi/\delta. Starting from ∣00⟩\ket{00} and a thermal mode, what is the state of the two qubits?

  1. Halfway to the Bell state: a pure, partially entangled two-qubit state
  2. Exactly the Bell state, since Φ\Phi has already reached π/8\pi/8
  3. A mixed two-qubit state, because components with different ss are correlated with differently displaced motional states
  4. Still ∣00⟩\ket{00}, since the displacements of the two ions cancel
Show the answer

C. A mixed two-qubit state, because components with different ss are correlated with differently displaced motional states

At t=π/δt = \pi/\delta the displacement is α=−2F/δ\alpha = -2F/\delta times ss, so the s=±2s = \pm2 components sit on opposite sides of phase space while s=0s = 0 stays at the origin. The spins are entangled with the mode; tracing it out leaves a mixed spin state. Only at loop closure does the motion factor out.

Robustness and the error budget

At tKt_K the propagator is exp⁡(−iπ4σϕσϕ)⊗Imotion\exp(-i\tfrac\pi4\sigma_\phi\sigma_\phi)\otimes\id_{\rm motion} for any motional state: thermal, coherent or squeezed. Temperature enters only through (i) the next order of the Lamb–Dicke expansion, which makes the sideband couplings depend on nn and is suppressed when η2(2nˉ+1)≪1\eta^2(2\bar n+1) \ll 1, and (ii) imperfect loop closure, whose cost is amplified by thermal motion, as the next result quantifies.

DerivationProof sketch, and a worked consequence
  1. ρspin=∑s,s′cs cs′∗ e−iΦK(s2−s′2) χ((s−s′)β) ∣s⟩ ⁣⟨s′∣,χ(γ)=Tr⁡[ρthD^(γ)]=e−∣γ∣2(nˉ+12)\rho_{\rm spin} = \sum_{s,s'}c_s\,c_{s'}^*\,e^{-i\Phi_K(s^2-s'^2)}\,\chi\big((s-s')\beta\big)\,\ketbra{s}{s'},\qquad \chi(\gamma) = \Tr\big[\rho_{\rm th}\hat D(\gamma)\big] = e^{-\abs\gamma^2(\bar n+\frac12)}

    Trace out the mode (sum over an eigenbasis of SϕS_\phi). The coherence between components ss and s′s' is multiplied by the thermal characteristic function at the difference of their residual displacements.

  2. 1−F≈∣β∣2(nˉ+12)∑s,s′psps′(s−s′)2=2Var⁡(s) ∣β∣2(nˉ+12)=2(2nˉ+1)∣β∣21 - \mathcal F \approx \abs\beta^2\big(\bar n+\tfrac12\big)\sum_{s,s'}p_sp_{s'}(s-s')^2 = 2\operatorname{Var}(s)\,\abs\beta^2\big(\bar n+\tfrac12\big) = 2(2\bar n+1)\abs\beta^2

    Overlap with the ideal state and expand; Var⁡(s)=2\operatorname{Var}(s) = 2 for ∣00⟩\ket{00}. The factor 2nˉ+12\bar n+1 is the price of heat: a hot mode makes the same residual displacement more distinguishable.

  3. δ→δ+ε:β=α(tg)≈−iFε tgδ,∣β∣=πK2 εδ\delta\to\delta+\varepsilon:\qquad \beta = \alpha(t_g) \approx -\frac{iF\varepsilon\,t_g}{\delta},\qquad \abs\beta = \frac{\pi\sqrt K}{2}\,\frac{\varepsilon}{\delta}

    Example: a mode-frequency error. If the mode drifts by ε\varepsilon while the pulse keeps its calibrated length tg=2πK/δt_g = 2\pi K/\delta, the loops miss closure by this much (using F/δ=1/4KF/\delta = 1/4\sqrt K).

  4. 1−F≈π2K2 (2nˉ+1)(εδ)21 - \mathcal F \approx \frac{\pi^2K}{2}\,(2\bar n+1)\Big(\frac{\varepsilon}{\delta}\Big)^2

    A drift of 1% of δ\delta costs about 5×10−45\times10^{-4} for a cold single-loop gate. This quadratic but thermally amplified sensitivity is why experiments recalibrate mode frequencies and use pulse shapes designed to be robust to ε\varepsilon.

Error sourceHow it entersTypical mitigation
Mode-frequency drift, timing errorsResidual displacement; 1−F∝(2nˉ+1)∣β∣21-\mathcal F \propto (2\bar n+1)\abs\beta^2Calibration; amplitude-, frequency- or phase-modulated pulses robust to detuning
Off-resonant carrier excitationUnwanted spin rotations with amplitude of order Ω/ν\Omega/\nuΩ≪ν\Omega \ll \nu; smooth pulse ramps
Spectator modesOther modes are displaced too, and their loops need not closeLarge mode spacing; multi-mode pulse shaping
Motional heating during the gateRandom displacements accumulate, growing with nˉ˙ tg\dot{\bar n}\,t_gCryogenic traps, surface treatment, faster gates
Photon scattering (Raman gates)Spontaneous emission from the intermediate level destroys coherenceLarge detuning from the excited state
Laser intensity and phase noiseErrors in the entangling angle 2ΦK2\Phi_K and the spin axis ϕ\phiIntensity stabilization; phase-insensitive beam geometries

Many ions, many modes

In an NN-ion chain the force drives all modes mm (frequencies νm\nu_m, Lamb–Dicke parameters ηj,m\eta_{j,m}, detunings δm=μ−νm\delta_m = \mu - \nu_m), and the Rabi frequencies Ωj(t)\Omega_j(t) may be shaped in time. With gj,m(t)=ηj,mΩj(t)/2g_{j,m}(t) = \eta_{j,m}\Omega_j(t)/2, the Hamiltonian is

H(t)=ℏ∑j∑mgj,m(t) σϕ(j)(ameiδmt+am†e−iδmt)H(t) = \hbar\sum_{j}\sum_{m}g_{j,m}(t)\,\sigma_\phi^{(j)}\big(a_m e^{i\delta_m t} + a_m^\dagger e^{-i\delta_m t}\big)

The closure conditions are linear in the pulse amplitudes, while the entangling condition is quadratic. Divide the pulse into segments with amplitudes Ω(k)\Omega^{(k)}: closing all loops is a null-space problem for a linear system, and among its solutions one picks the lowest-power pulse that reaches χij=π/4\chi_{ij} = \pi/4. Amplitude modulation (Zhu, Monroe & Duan, 2006; Choi et al., 2014), frequency modulation (Leung et al., 2018) and phase modulation (Green & Biercuk, 2015) all implement this idea, and the same freedom buys robustness to drifts and timing errors.

Key takeaways

  • In the Lamb–Dicke and rotating-wave regimes, a bichromatic drive near both sidebands produces a spin-dependent force H=ℏFSϕ(aeiδt+a†e−iδt)H = \hbar F S_\phi(ae^{i\delta t} + a^\dagger e^{-i\delta t}).
  • The Magnus expansion terminates: U=D^(αS) e−iΦS2U = \hat D(\alpha S)\,e^{-i\Phi S^2}. Closed loops leave a pure geometric spin phase equal to twice the enclosed area.
  • Maximal entanglement requires δ=2ηΩK\delta = 2\eta\Omega\sqrt K and tg=2πK/δt_g = 2\pi K/\delta; the gate then works for any motional state.
  • Imperfect closure costs ≈2(2nˉ+1)∣β∣2\approx 2(2\bar n+1)\abs\beta^2; in long chains every mode must close, which pulse shaping achieves by solving linear closure constraints.

Check yourself

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

Question 1

Why does the MS gate work with a thermal motional state, unlike the original Cirac–Zoller gate?

  1. When the phase-space loops close, the motion returns to its initial state for every spin component, and the spin phase depends only on the enclosed area
  2. The bichromatic field filters out thermal phonons
  3. The gate runs so slowly that the motion averages to zero
  4. The MS gate does not involve the motion; it uses a direct spin–spin interaction
Show the answer

A. When the phase-space loops close, the motion returns to its initial state for every spin component, and the spin phase depends only on the enclosed area

U(tK)=e−iΦKS2⊗IU(t_K) = e^{-i\Phi_K S^2}\otimes\id: displacements around closed loops act trivially on any motional state, leaving only the geometric phase.

Question 2

A single-loop (K=1K = 1) MS gate with sideband Rabi frequency ηΩ\eta\Omega is maximally entangling when the detuning from the sidebands is

  1. δ=ηΩ/2\delta = \eta\Omega/2
  2. δ=ηΩ\delta = \eta\Omega
  3. δ=2ηΩ\delta = 2\eta\Omega
  4. δ=2π ηΩ\delta = 2\pi\,\eta\Omega
Show the answer

C. δ=2ηΩ\delta = 2\eta\Omega

ΦK=2πKF2/δ2\Phi_K = 2\pi KF^2/\delta^2 with F=ηΩ/2F = \eta\Omega/2 must equal π/8\pi/8, so δ2=16KF2=4K(ηΩ)2\delta^2 = 16KF^2 = 4K(\eta\Omega)^2, i.e. δ=2ηΩ\delta = 2\eta\Omega for K=1K = 1.

Question 3

At a fixed residual displacement β\beta, heating the mode from nˉ=0\bar n = 0 to nˉ=5\bar n = 5 multiplies the closure error by about

  1. 1: temperature does not matter
  2. 11\sqrt{11}
  3. 5
  4. 11
Show the answer

D. 11

1−F≈2(2nˉ+1)∣β∣21 - \mathcal F \approx 2(2\bar n+1)\abs\beta^2, so the factor grows from 1 to 11. Perfectly closed loops are temperature-independent; imperfect ones are not.

Question 4

You implement a two-qubit MS gate on ions 3 and 7 of a 20-ion chain. Which conditions must the pulse satisfy at its end?

  1. Only the center-of-mass mode must return to its initial state
  2. α3,m\alpha_{3,m} and α7,m\alpha_{7,m} must vanish for every mode mm, and χ37=π/4\chi_{37} = \pi/4
  3. All 20 ions must be illuminated with equal intensity
  4. χij=π/4\chi_{ij} = \pi/4 for every pair i,ji, j in the chain
Show the answer

B. α3,m\alpha_{3,m} and α7,m\alpha_{7,m} must vanish for every mode mm, and χ37=π/4\chi_{37} = \pi/4

Every mode the force couples to is displaced; any loop left open entangles the qubits with that mode. Unilluminated ions have gj,m=0g_{j,m} = 0 and are untouched.

Further reading

  • A. Sørensen & K. Mølmer, "Quantum computation with ions in thermal motion", Physical Review Letters 82, 1971 (1999).
  • K. Mølmer & A. Sørensen, "Multiparticle entanglement of hot trapped ions", Physical Review Letters 82, 1835 (1999).
  • A. Sørensen & K. Mølmer, "Entanglement and quantum computation with ions in thermal motion", Physical Review A 62, 022311 (2000).
  • D. Leibfried et al., "Experimental demonstration of a robust, high-fidelity geometric two ion-qubit phase gate", Nature 422, 412 (2003).
  • C. J. Ballance, T. P. Harty, N. M. Linke, M. A. Sepiol & D. M. Lucas, "High-fidelity quantum logic gates using trapped-ion hyperfine qubits", Physical Review Letters (2016).