Cirac–Zoller Gate

The Cirac–Zoller gate is the proposal that made quantum computing a laboratory science rather than a thought experiment. In 1995, Juan Ignacio Cirac and Peter Zoller asked a deceptively simple question: given a row of ions held in a linear Paul trap, each carrying a qubit in its electronic state, how do you make two of them interact on demand? Their answer was to route the interaction not through any direct force between the ions — the Coulomb repulsion is always on and cannot be switched — but through the collective motion of the whole chain. Push one ion and the entire Coulomb crystal rings in its normal modes; the lowest, the center-of-mass (COM) mode in which every ion oscillates in step, is a single quantized oscillator shared equally by all ions. It is a quantum bus, a wire that any ion can write to and any other can read. The gate is a three-pulse choreography — load, phase, unload. A laser pulse tuned to a motional sideband writes the control ion's excitation onto the bus as a single phonon; a second pulse asks the target to flip the sign of its amplitude, but only if it detects that phonon; a third erases the phonon, returning the bus exactly to vacuum so it carries away no which-path information. The net operation is a controlled-Z, and with single-qubit rotations that is enough for universal quantum computation. The scheme is the concrete embodiment of the DiVincenzo requirement of a universal two-qubit gate, and it opened the lineage — geometric gates, the Mølmer–Sørensen gate — that powers today's ion-trap machines. Understanding Cirac–Zoller means understanding, in its cleanest form, the idea of a phonon-mediated bus.

Key takeaways

The Cirac–Zoller gate (1995) was the first explicit proposal for a universal two-qubit gate in a scalable quantum computer, entangling two trapped ions through a shared vibrational mode — the center-of-mass phonon bus.
Three laser pulses do the work: a red-sideband π pulse loads the control excitation into the bus as one phonon, a conditional 2π pulse on the target's auxiliary transition imprints a −1 phase only when a phonon is present, and a reverse π pulse unloads the bus.
Each step is a Jaynes–Cummings sideband rotation with coupling ηΩ; the number-dependent Rabi frequency Ω_n = ηΩ√n is why the bus must be cooled near its motional ground state.
The net unitary is a controlled-Z, diag(1, 1, 1, −1), locally equivalent to CNOT; the −1 sits only on the doubly-excited input.
Its sensitivity to heating and thermal occupation drove robust successors like the Mølmer–Sørensen gate; the full two-ion Cirac–Zoller CNOT was realized in Innsbruck in 2003.
Key idea: encode qubits in ion internal states and use one ground-state-cooled phonon mode as a quantum bus — a red-sideband π pulse loads a control excitation into the bus, a conditional 2π pulse on the target imprints a −1 phase if and only if the bus holds a phonon, and the reverse π pulse unloads it, yielding a deterministic controlled-Z.

The core insight

Encode qubits in ion internal states and use one shared, ground-state-cooled phonon mode as a quantum bus: a red-sideband π pulse loads a control excitation into the bus, a conditional 2π pulse on the target imprints a −1 phase if and only if the bus holds a phonon, and the reverse π pulse unloads the bus — yielding a deterministic controlled-Z.

Where this lives today

The Cirac–Zoller gate is the ancestor every trapped-ion quantum computer descends from, though almost none run it verbatim. Its Achilles' heel — the $\sqrt n$ sensitivity to the phonon number and the demand that the bus sit in its motional ground state — pushed the field toward the Mølmer–Sørensen gate, a geometric phase gate that entangles ions through a virtual excursion of the bus and works even at finite temperature. That gate is what runs inside today's commercial machines: Quantinuum's H-series and IonQ's Forte report two-qubit fidelities above 99.9%, and NIST and Oxford demonstrations have reached 99.97% — a hundredfold better than the 70–80% of the original 2003 two-ion Cirac–Zoller CNOT. Yet the conceptual DNA is unchanged: entangle qubits by writing to and reading from a shared bosonic mode, then clean up the bus. The same picture reappears in circuit QED, where a microwave resonator plays the phonon's role for superconducting qubits. The open frontier — scaling to thousands of ions, or networking traps with photonic links — is still, at bottom, the problem Cirac and Zoller first posed: how to build and reset quantum wires.

The physical picture

Picture the ions as heavy beads on a stiff Coulomb spring. Nudge one and the whole chain rings; its slowest, most uniform ringing is the center-of-mass mode, in which every ion swings in phase like a single rigid rod. Because all ions participate equally, a vibration created by ion A can be heard and acted upon by ion B no matter how far apart they sit. Quantize that vibration and each unit of excitation is a phonon — and a single shared phonon is a single shared wire. The first figure fixes this picture and pins down the one number the gate turns on: the Lamb–Dicke parameter η, which measures how strongly the laser couples the ion's internal state to the shared motion.

Figure

A linear ion chain and its shared center-of-mass phonon mode: all ions oscillate in phase, so the mode is one quantized oscillator any ion can address. For ⁴⁰Ca⁺ on the 729 nm transition with ν/2π = 1 MHz the zero-point spread is x₀ ≈ 11 nm and the Lamb–Dicke parameter is η = k·x₀ ≈ 0.10

The gate itself is a piece of timing. Loading converts "is the control excited?" into "is there a phonon?". The phasing step asks the target to flip a sign only when it senses that phonon — and here is the subtlety Cirac and Zoller had to solve: a full 2π Rabi cycle on a two-level system returns the state to itself but multiplies it by −1, the spinor sign of SU(2). The trick is to make that cycle resonant only when a phonon is present, which requires a third, auxiliary electronic level so the −1 becomes conditional. Unloading reverses the loading step, returning the bus exactly to vacuum; if any phonon amplitude is left behind, the motion stays entangled with the qubits and quietly destroys the coherence the gate was meant to build. The second figure lays out the three pulses in time, tracks the doubly-excited input, and shows the controlled-Z that emerges — with the real pulse durations for the ⁴⁰Ca⁺ numbers above.

Figure

The three-pulse choreography and the resulting controlled-Z. Following the doubly-excited input through load, conditional phase, and unload leaves an overall −1 on it alone; every other input returns unchanged. With sideband coupling η·Ω/2π = 50 kHz the two π pulses last ≈ 10 μs and the 2π pulse ≈ 20 μs, a total near 40 μs

The core equations

The scheme rests on how a laser couples an ion's internal state to its quantized motion. The dipole interaction carries a phase factor e^{ikx}; because x is now an operator, it mixes the electronic transition with the phonon ladder. In the interaction picture with detuning δ from the carrier:

H_{\mathrm{int}} = \hbar\Omega\,\sigma_+\,e^{i\eta(a+a^\dagger)}e^{-i\delta t}+\mathrm{h.c.} \qquad (1)

where: $\Omega$ — carrier Rabi frequency, set by laser intensity and dipole matrix element; $\sigma_+=|e\rangle\langle g|$ — raising operator; $\eta$ — Lamb–Dicke parameter; $a,a^\dagger$ — phonon operators of the bus mode; $\delta$ — detuning from the carrier. Reading it: the exponential is the whole story — each power of $(a+a^\dagger)$ changes the phonon number, with amplitude set by $\eta$ . The Lamb–Dicke parameter is the dimensionless ratio of zero-point motion to the laser's reduced wavelength — the single knob setting how strongly light talks to motion.

\eta = k\,x_0 = k\sqrt{\frac{\hbar}{2 m \nu}} \qquad (2)

where: $k$ — laser wavevector projected onto the mode axis; $x_0=\sqrt{\hbar/2m\nu}$ — zero-point spread; $m$ — ion (or effective mode) mass; $\nu$ — angular trap frequency. Reading it: heavier ions and stiffer traps shrink $x_0$ and hence $\eta$ ; smaller $\eta$ means weaker but cleaner sidebands. Tuning one trap quantum below the carrier ( $\delta=-\nu$ ) selects the red sideband. Keeping only the resonant term and expanding the exponential to first order in $\eta$ collapses equation (1) to the Jaynes–Cummings form familiar from cavity QED:

H_{\mathrm{rsb}} = \hbar\,\eta\,\Omega\left(\sigma_+ a + \sigma_- a^\dagger\right) \qquad (3)

where: $\eta\Omega$ — effective sideband Rabi frequency; $\sigma_+ a$ — raises the internal state while destroying one phonon, $|g,n\rangle\to|e,n-1\rangle$ ; $\sigma_- a^\dagger$ — its Hermitian conjugate. Reading it: the ion-plus-phonon system behaves like a two-level atom coupled to one cavity mode, the trap phonon playing the photon's role. Because the coupling carries the ladder operator $a$ , its matrix element depends on how many phonons are present — the number-dependent Rabi frequency, the Achilles' heel of the gate:

\Omega_{n} = \eta\,\Omega\,\sqrt{n} \qquad (4)

where: $\Omega_n$ — red-sideband Rabi frequency between $|g,n\rangle$ and $|e,n-1\rangle$ ; $n$ — phonon number. Reading it: a pulse calibrated as a π pulse for $n=1$ over- or under-rotates for any other $n$ , so a thermal spread smears the gate — exactly why the bus must be cooled to $n=0$ . The complete gate is the ordered product of three pulses — a control red-sideband π, a target auxiliary 2π, and the inverse of the first. Following every basis state through it (Derivation 3) yields a diagonal unitary:

U_{\mathrm{CZ}}=U_{\pi}^{\mathrm{rsb},c\,\dagger}\;U_{2\pi}^{\mathrm{aux},t}\;U_{\pi}^{\mathrm{rsb},c}=\mathrm{diag}(1,1,1,-1) \qquad (5)

where: the diagonal is in the basis $\{|gg\rangle,|ge\rangle,|eg\rangle,|ee\rangle\}$ ; superscript $c$ marks a control pulse, $t$ a target pulse, $\dagger$ the time-reversed pulse. Reading it: this is the controlled-Z, CNOT-equivalent by sandwiching the target in Hadamards, and universal with single-qubit rotations. Real gates are imperfect, and the two dominant errors have a clean schematic form — residual thermal occupation, and anomalous heating of the bus during the gate:

1-\mathcal{F} \;\sim\; c_1\,\eta^2\,\bar n \;+\; \dot{\bar n}\,\tau \qquad (6)

where: $\mathcal{F}$ — gate fidelity; $\bar n$ — residual phonon number after cooling; $\dot{\bar n}$ — heating rate (quanta per second); $\tau$ — gate duration; $c_1$ — order-unity factor. Reading it: both terms vanish as the bus approaches a cold, quiet ground state — the experimental price of the Cirac–Zoller scheme.

Derivation AFrom the dipole coupling to the red-sideband Hamiltonian

The goal is to show, starting from equation (1), how the exponential $e^{i\eta(a+a^\dagger)}$ becomes the clean Jaynes–Cummings coupling $\hbar\eta\Omega(\sigma_+ a+\sigma_- a^\dagger)$ that drives one leg of the gate, and to extract the number-dependent Rabi frequency $\Omega_n=\eta\Omega\sqrt n$ that forces the bus to be cooled to its ground state. We quantize the position, expand in the small parameter $\eta$ , tune to the red sideband, and read off the matrix element between number states.

Step 1. Write the ion's position in mode operators, $x=x_0(a+a^\dagger)$ with $x_0=\sqrt{\hbar/2m\nu}$ , so the laser phase becomes $e^{ikx}=e^{i\eta(a+a^\dagger)}$ with $\eta\equiv k x_0$ . Why: this is where the Lamb–Dicke parameter is born — the wavevector measured in units of the inverse zero-point spread — and everything downstream is an expansion in it.

Step 2. In the Lamb–Dicke regime $\eta\sqrt{2\bar n+1}\ll1$ , expand to first order, $e^{i\eta(a+a^\dagger)}\approx 1+i\eta(a+a^\dagger)$ : the leading term is the carrier, the $\eta a$ term lowers the phonon number, $\eta a^\dagger$ raises it. Why: first order isolates the three clean transitions — carrier, red and blue sideband — instead of a forest of multi-phonon processes.

Step 3. Tune to the red sideband, $\delta=-\nu$ . In the interaction picture $a\to a\,e^{-i\nu t}$ and $\sigma_+\to\sigma_+ e^{i\delta t}$ , so $\sigma_+ a$ becomes stationary while $\sigma_+ a^\dagger$ and the carrier oscillate at $\pm\nu$ or $2\nu$ . Why: resonance is a matching of clocks — only the term whose time dependence cancels survives averaging over many trap periods.

Step 4. Apply the rotating-wave approximation, dropping the fast carrier and blue-sideband terms, to leave $H_{\mathrm{rsb}}=\hbar\eta\Omega(\sigma_+ a+\sigma_- a^\dagger)$ . Its matrix element follows from $a|n\rangle=\sqrt n\,|n-1\rangle$ : $\langle e,n-1|H_{\mathrm{rsb}}|g,n\rangle=\hbar\eta\Omega\sqrt n$ , hence $\Omega_n=\eta\Omega\sqrt n$ . Why: the $\sqrt n$ is the fingerprint of a bosonic bus, and here it makes the gate's calibration depend on the exact phonon count.

Go deeper · Derivation B · 2 minThe conditional 2π pulse and its minus-one phase

Derivation BThe conditional 2π pulse and its minus-one phase

The goal is to explain why a complete Rabi cycle returns a two-level state to itself but with an overall minus sign, and how coupling that cycle to an auxiliary electronic level makes the minus sign appear only when a bus phonon is present — the entire mechanism by which the gate conditions on the control. We evolve a driven two-level subsystem for a full 2π of pulse area, read off the spinor sign, then show the partner state the transition needs is absent when the bus is empty.

Step 1. On resonance the driven subsystem $\{|e,1\rangle,|\mathrm{aux},0\rangle\}$ evolves under $H=\tfrac{\hbar}{2}\eta'\Omega'\,\sigma_x$ , where $\sigma_x$ swaps the two levels and $\eta'\Omega'$ is the auxiliary sideband Rabi frequency. Why: driving the ion between $|e\rangle$ and a metastable level $|\mathrm{aux}\rangle$ on the red sideband is formally a spin-1/2 in a transverse field — the cleanest system in which to see the spinor sign.

Step 2. The propagator is $U(\theta)=e^{-i\theta\sigma_x/2}=\cos\tfrac{\theta}{2}\,\mathbb{1}-i\sin\tfrac{\theta}{2}\,\sigma_x$ for area $\theta=\eta'\Omega' t$ . At $\theta=2\pi$ this gives $U(2\pi)=-\mathbb{1}$ . Why: a spin returns to itself only after a 4π rotation, not 2π — the double-valuedness of SU(2) — so a 2π pulse does nothing observable except stamp an overall −1.

Step 3. Make the sign conditional. The auxiliary transition removes one phonon, connecting $|e,n\rangle$ to $|\mathrm{aux},n-1\rangle$ . If the bus is in $|0\rangle$ , the state $|e,0\rangle$ has no partner $|\mathrm{aux},-1\rangle$ : it is dark, untouched. Only when a phonon is present ( $|e,1\rangle$ , partner $|\mathrm{aux},0\rangle$ ) does the cycle run and the −1 appear. Why: the phonon gatekeeps the auxiliary transition, so "did a full cycle happen?" becomes a faithful proxy for "is there a phonon?" — exactly the conditioning a controlled gate needs.

Step 4. Note why the auxiliary level is not optional: a 2π pulse on the qubit transition $|g\rangle\leftrightarrow|e\rangle$ itself returns $-1$ regardless of phonon number, a trivial global phase. The extra level $|\mathrm{aux}\rangle$ is what lets the target acquire a sign only in coincidence with a bus phonon. Why: the design insight of the 1995 paper — the gate needs a transition outside the qubit space so the phase does not act on the computational states directly.

Go deeper · Derivation C · 1 minAssembling the controlled-Z from three pulses

Derivation CAssembling the controlled-Z from three pulses

The goal is to run all four computational basis states through the full load–phase–unload sequence and confirm the net operation is exactly $\mathrm{diag}(1,1,1,-1)$ , a controlled-Z, with the phonon bus returned to vacuum on every branch so the qubits are left in a pure two-qubit state. We fix the π-pulse conventions, then track the doubly-excited input in detail and the other three by the same rules.

Step 1. Fix the control loading pulse. A red-sideband π on the control acts as $|e,0\rangle_c\to -i\,|g,1\rangle_c$ and leaves $|g,0\rangle_c\to|g,0\rangle_c$ (dark, no lower phonon partner). Its inverse, used at the end, sends $|g,1\rangle_c\to i\,|e,0\rangle_c$ . Why: the factors of $\mp i$ are the ordinary phases of a resonant π rotation and must be tracked, since they combine to give the final sign.

Step 2. Track $|ee\rangle|0\rangle$ . Loading gives $|e\,e,0\rangle\to -i\,|g\,e,1\rangle$ : a phonon now exists because the control was excited. The target 2π pulse fires (target in $|e\rangle$ , one phonon present) and multiplies by $-1$ : $-i\,|g\,e,1\rangle\to +i\,|g\,e,1\rangle$ . Unloading applies the inverse π: $+i\,|g\,e,1\rangle\to +i\cdot i\,|e\,e,0\rangle=-|e\,e,0\rangle$ . Why: the two $i$ 's from loading and unloading combine with the conditional $-1$ to leave exactly one surviving minus sign on $|ee\rangle$ .

Step 3. Track $|eg\rangle|0\rangle$ : loading gives $-i\,|g\,g,1\rangle$ , the target is in $|g\rangle$ so the 2π pulse does nothing, and unloading returns $-i\cdot i\,|e\,g,0\rangle=+|e\,g,0\rangle$ . For $|ge\rangle|0\rangle$ the control is in $|g\rangle$ , so loading is dark and no phonon appears; the target's 2π pulse cannot fire, and the state returns untouched. $|gg\rangle$ is trivially fixed. Why: only the input with both control excited (to make a phonon) and target excited (to catch it) threads the conditional loop.

Step 4. Collect the results: $|gg\rangle\to|gg\rangle$ , $|ge\rangle\to|ge\rangle$ , $|eg\rangle\to|eg\rangle$ , $|ee\rangle\to-|ee\rangle$ , i.e. $U=\mathrm{diag}(1,1,1,-1)$ , with the bus in $|0\rangle$ on every branch. Why: the vacuum-restoring third pulse is essential — a surviving phonon would keep the motion entangled with the qubits, and the operation would not be unitary on the computational space alone. *The controlled-Z is locally equivalent to CNOT via Hadamards on the target.*

Limiting cases

The gate lives or dies by two numbers: the Lamb–Dicke parameter η and the bus occupation n̄.

Case	Result	Meaning
Lamb–Dicke regime	$\eta\sqrt{2\bar n+1}\ll 1$	First-order expansion holds: carrier and first-order sidebands suffice, multi-phonon processes suppressed, the clean three-pulse picture valid.
Ground-state bus	$\bar n\to 0$	Ideal initial condition: $\|g,0\rangle$ is dark on the red sideband (no $n=-1$ partner), a clean two-level structure at every step.
Resolved-sideband limit	$\eta\Omega \ll \nu$	The sideband Rabi frequency stays far below the trap frequency, so a single sideband is addressed without exciting the carrier or opposite sideband.
Thermal occupation	$\bar n>0$	The $n$ -dependent $\Omega_n=\eta\Omega\sqrt n$ smears the π and 2π calibration, leaking population — the chief reason robust successors were developed.

Common trapThe conditional 2π pulse can be driven on the qubit transition itself

The most common misreading is to imagine the 2π pulse acting on the qubit transition $|g\rangle\leftrightarrow|e\rangle$ . It cannot: a full 2π cycle there returns $-1$ regardless of phonon number, so the "gate" would apply a global phase and entangle nothing. The whole point of Cirac and Zoller's auxiliary level $|\mathrm{aux}\rangle$ is to provide a transition outside the qubit space whose 2π cycle is resonant only when a bus phonon is present — that is what makes the sign conditional. A second, equally fatal trap is forgetting to cool the bus: if the mode starts with $\bar n>0$ , the $\sqrt n$ scaling spoils every pulse, and if the third pulse fails to return the bus exactly to $|0\rangle$ , the leftover motion stays entangled with the qubits and silently decoheres them. Vacuum in, vacuum out — or the gate is not a gate.

Worked examples

1 · Lamb–Dicke parameter and gate time for a calcium ion (fully worked)

A single ⁴⁰Ca⁺ ion is driven on its 729 nm quadrupole transition in a trap with axial frequency ν/2π = 1 MHz. Find the zero-point spread x₀, the Lamb–Dicke parameter η for a laser along the trap axis, and — given a carrier Rabi frequency Ω/2π = 500 kHz — the duration of a red-sideband π pulse and of the full gate.

Step 1. Constants: $m=40\times1.6605\times10^{-27}=6.64\times10^{-26}$ kg, $\nu=2\pi\times10^6=6.28\times10^{6}$ rad/s, $k=2\pi/(729\times10^{-9})=8.62\times10^{6}$ m⁻¹.

Step 2. Zero-point spread. $x_0=\sqrt{\hbar/2m\nu}=\sqrt{1.055\times10^{-34}/(2\times6.64\times10^{-26}\times6.28\times10^{6})}=\mathbf{1.12\times10^{-8}\ m\approx 11\ nm}$ — tiny beside the 729 nm wavelength, which puts us deep in the Lamb–Dicke regime.

Step 3. Lamb–Dicke parameter. $\eta=k x_0=8.62\times10^{6}\times1.12\times10^{-8}=\mathbf{0.097\approx0.10}$ . Check with a cooled bus $\bar n\approx0.05$ : $\eta\sqrt{2\bar n+1}=0.10\times\sqrt{1.1}=0.105\ll1$ . The clean picture holds.

Step 4. Coupling and pulse times. The red-sideband Rabi frequency on the $n=1$ transition is $\eta\Omega/2\pi=0.10\times500\ \mathrm{kHz}=\mathbf{50\ kHz}$ , so a π pulse takes $t_\pi=\pi/(\eta\Omega)=1/(2\times50\times10^{3})=\mathbf{10\ \mu s}$ and the 2π pulse about $\mathbf{20\ \mu s}$ . The full sequence runs in $10+20+10=\mathbf{40\ \mu s}$ — the durations in the second figure, typical of real ion-trap gates.

2 · How cold must the bus be? A heating and thermal-error estimate

The same gate runs in τ = 40 μs on a bus with residual occupation n̄ = 0.05 and an anomalous heating rate of ṅ = 100 quanta/s. Estimate the two leading contributions to the gate error and say which dominates.

Step 1. Heating during the gate. The bus gains $\dot{\bar n}\,\tau=100\times40\times10^{-6}=\mathbf{4\times10^{-3}}$ phonons. This deposited quantum is entangled with the qubits and shows up as an infidelity of order $4\times10^{-3}$ , a fidelity ceiling near $\mathcal{F}\approx0.996$ from heating alone.

Step 2. Residual thermal spread. Because $\Omega_n=\eta\Omega\sqrt n$ , a bus not perfectly in $n=0$ mis-rotates the pulses. The error scales as $\sim\eta^2\bar n$ ; numerically $\eta^2\bar n=(0.10)^2\times0.05=5\times10^{-4}$ , a few parts in $10^4$ .

Step 3. Compare and conclude. The two terms are $\approx4\times10^{-3}$ (heating) versus $\approx5\times10^{-4}$ (thermal spread), so anomalous heating dominates here, and the total infidelity is a few parts in $10^3$ . Pushing toward the $10^{-4}$ floor demanded better cooling ( $\bar n\ll0.1$ ) and suppression of trap heating (cryogenic traps, larger ion–electrode distances) — and even then the $n$ -sensitivity motivated the robust Mølmer–Sørensen gate. (These are order-of-magnitude estimates; prefactors depend on pulse shaping and level scheme.)

Physics of the portrait

The gate carries two names, but one idea: the shared motion of trapped ions is not a nuisance to eliminate but a resource to program — decoherence source to quantum bus.

Origins · 1 minJuan Ignacio Cirac and how this was discovered

Juan Ignacio Cirac

b. 1965 · theoretical quantum optics and quantum information

Juan Ignacio Cirac is a Spanish theoretical physicist, since 2001 a director at the Max Planck Institute of Quantum Optics in Garching, whose 1995 proposal with Peter Zoller — then both at the University of Innsbruck — gave quantum computing its first concrete hardware blueprint. Before their paper, the field had Shor's algorithm and a growing theory of gates but no way to build a universal two-qubit gate in a real system meeting all the requirements at once. Cirac and Zoller's insight was to turn the trapped ions' collective vibration — the very motion spectroscopists fought to suppress — into a data bus mediating a controlled gate between ions that never touch. Their scheme was demonstrated in a single ion within months by David Wineland's NIST group and realized on two ions in 2003 by Rainer Blatt's Innsbruck team, launching trapped ions as a leading platform. Their collaboration ranged far beyond this gate — quantum repeaters and memories, cavity-QED light–matter interfaces, cold atoms in optical lattices as quantum simulators — but the gate bearing their names remains the canonical illustration of a phonon-mediated interaction.

History

1975

— Wineland and Dehmelt, and independently Hänsch and Schawlow, propose laser cooling of trapped ions

1989

— Paul and Dehmelt share the Nobel Prize for the ion-trap techniques the gate relies on

1995

— Cirac and Zoller publish the phonon-bus gate proposal in PRL 74, 4091

1995

— Monroe, Meekhof, King, Itano and Wineland demonstrate a CNOT-equivalent in a single Be⁺ ion

2003

— Schmidt-Kaler and colleagues in Innsbruck realize the full two-ion Cirac–Zoller CNOT

2012

— Wineland shares the Nobel Prize for the measurement and control of individual quantum systems

Going further

Texts. Nielsen and Chuang, Quantum Computation and Quantum Information, §7.6 (the ion-trap computer) · Preskill, Lecture Notes on Quantum Computation, Ch. 7 (physical implementations) · Wineland et al., J. Res. NIST 103, 259 (1998) (experimental issues in ion-trap gates) · Leibfried, Blatt, Monroe and Wineland, Rev. Mod. Phys. 75, 281 (2003) (quantum dynamics of single trapped ions) · Häffner, Roos and Blatt, Phys. Rep. 469, 155 (2008) (quantum computing with trapped ions). The key relations are $\eta=k\sqrt{\hbar/2m\nu}$ and $\Omega_n=\eta\Omega\sqrt n$ ; the ⁴⁰Ca⁺ numbers work out to $x_0\approx11$ nm, $\eta\approx0.10$ , and $t_\pi\approx10\ \mu$ s (see Worked Example 1). The spinor identity $U(2\pi)=-\mathbb{1}$ follows from $e^{-i\theta\sigma_x/2}$ evaluated at $\theta=2\pi$ . The fidelity estimates in Worked Example 2 are order-of-magnitude only; the modern gate fidelities above 99.9% refer to Mølmer–Sørensen-type gates, not the bare Cirac–Zoller sequence.

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

Ask the tutor: Ask me to walk through the |ee⟩ amplitude phase-by-phase, or to compare the Cirac–Zoller and Mølmer–Sørensen gates side by side.

Why route the interaction through motion instead of coupling the ions' internal states directly?

Because there is no switchable direct coupling between two ions' internal states — they interact only through the always-on Coulomb force, which you cannot turn on and off to make a gate. What the Coulomb force does give you for free is a set of shared vibrational modes. Cirac and Zoller's move was to use one as an intermediary: the laser couples an ion's internal state to the shared motion (which you can switch, by turning the laser on and off), and the motion carries the information to the other ion. The bus is the workaround for the absence of a direct, controllable qubit–qubit interaction.

What exactly is a phonon here, and why is one enough?

A phonon is a single quantum of the ion chain's collective vibration — one unit of energy $\hbar\nu$ in the center-of-mass mode, in which every ion oscillates in phase. It is the mechanical analogue of a cavity photon. The gate needs only one shared phonon because the information routed is one qubit's worth: "was the control excited?" maps onto "is there one phonon or zero?". The bus toggles between $|0\rangle$ and $|1\rangle$ , and that single bit of motional which-path information is exactly what conditions the target's phase. Higher Fock states are not used and, if populated thermally, only cause errors.

How does a controlled-Z become a CNOT?

They are locally equivalent — one turns into the other with single-qubit gates on the target. A controlled-Z applies a $-1$ only to the $|ee\rangle$ input; a CNOT flips the target when the control is excited. Sandwiching the controlled-Z between two Hadamards on the target converts the conditional sign flip into a conditional bit flip: $H\,\mathrm{CZ}\,H=\mathrm{CNOT}$ . Because Hadamards are cheap, the two gates are interchangeable in practice, and both are maximally entangling — which is why the Cirac–Zoller controlled-Z plus single-qubit rotations is a universal gate set.

Why must the ion sit in the Lamb–Dicke regime?

The Lamb–Dicke regime, $\eta\sqrt{2\bar n+1}\ll1$ , is the condition that the ion's motional wavepacket is small compared with the laser wavelength. When it holds, the laser's spatial phase $e^{ikx}$ can be expanded to first order, and the coupling splits cleanly into three well-separated transitions: the carrier (motion unchanged) and the first red and blue sidebands (one phonon removed or added). Outside it, multi-phonon processes proliferate and the clean three-pulse gate falls apart. For the ⁴⁰Ca⁺ example, $\eta\approx0.10$ and $\bar n\approx0.05$ give $\eta\sqrt{2\bar n+1}\approx0.1$ , comfortably inside the regime.

Why did the field move on to the Mølmer–Sørensen gate?

Because Cirac–Zoller is fragile in exactly the way real ion traps are noisy. Its pulses are calibrated to a specific phonon number through $\Omega_n=\eta\Omega\sqrt n$ , so any residual occupation or heating mis-rotates them — and it requires cooling the bus all the way to $n=0$ , plus an auxiliary level and individual addressing. The Mølmer–Sørensen gate sidesteps these: it drives both sidebands at once so the bus is only virtually excited and returns to its starting state automatically, making the gate insensitive to the initial phonon number and forgiving of a warm bus. That robustness, not any change in the phonon-bus idea, is why it dominates modern hardware.

Was the Cirac–Zoller gate ever actually built?

Yes, in stages. Within months of the 1995 proposal, Monroe and colleagues at NIST demonstrated the mechanism in a single trapped Be⁺ ion, using its internal state as one qubit and its motional Fock states as the second. The full two-ion version, with the phonon bus genuinely shuttling information between separate ions, was realized in 2003 by Schmidt-Kaler and coworkers in Rainer Blatt's Innsbruck group, with a gate fidelity of order 70–80%. Modest by today's standards, it was the proof of principle that turned trapped ions into a serious quantum-computing platform.